Update the bundled libwebp to version 0.6.1v5.11.0-beta4

Upstream changes since 0.6.0 have been merged in. Also updated version in qt_attribution.json. Change-Id: I6f46bdfe2dd7288dbe61c4d7924309bb435e3928 Reviewed-by: Liang Qi <liang.qi@qt.io>
author: Eirik Aavitsland <eirik.aavitsland@qt.io> 2018-04-12 14:25:27 +0200
committer: Eirik Aavitsland <eirik.aavitsland@qt.io> 2018-04-12 21:22:06 +0000
commit: b83a0a9460432abb82218da247710a1aaf321336 (patch)
tree: 2bd4017481cc3644034d99f1649064434478c904
parent: 4e10f35483db6d3396fe54dcf3d895d7850f8335 (diff)
161 files changed, 7243 insertions, 5553 deletions
diff --git a/src/3rdparty/libwebp.pri b/src/3rdparty/libwebp.pri
index 55d7e60..eba9212 100644
--- a/src/3rdparty/libwebp.pri
+++ b/src/3rdparty/libwebp.pri
@@ -25,9 +25,6 @@ SOURCES += \
     $$PWD/libwebp/src/dsp/alpha_processing_mips_dsp_r2.c \
     $$PWD/libwebp/src/dsp/alpha_processing_sse2.c \
     $$PWD/libwebp/src/dsp/alpha_processing_sse41.c \
-    $$PWD/libwebp/src/dsp/argb.c \
-    $$PWD/libwebp/src/dsp/argb_mips_dsp_r2.c \
-    $$PWD/libwebp/src/dsp/argb_sse2.c \
     $$PWD/libwebp/src/dsp/cost.c \
     $$PWD/libwebp/src/dsp/cost_mips32.c \
     $$PWD/libwebp/src/dsp/cost_mips_dsp_r2.c \
@@ -64,6 +61,8 @@ SOURCES += \
     $$PWD/libwebp/src/dsp/rescaler_mips_dsp_r2.c \
     $$PWD/libwebp/src/dsp/rescaler_msa.c \
     $$PWD/libwebp/src/dsp/rescaler_sse2.c \
+    $$PWD/libwebp/src/dsp/ssim.c \
+    $$PWD/libwebp/src/dsp/ssim_sse2.c \
     $$PWD/libwebp/src/dsp/upsampling.c \
     $$PWD/libwebp/src/dsp/upsampling_mips_dsp_r2.c \
     $$PWD/libwebp/src/dsp/upsampling_msa.c \
@@ -75,6 +74,7 @@ SOURCES += \
     $$PWD/libwebp/src/dsp/yuv_sse2.c \
     $$PWD/libwebp/src/enc/alpha_enc.c \
     $$PWD/libwebp/src/enc/analysis_enc.c \
+    $$PWD/libwebp/src/enc/backward_references_cost_enc.c \
     $$PWD/libwebp/src/enc/backward_references_enc.c \
     $$PWD/libwebp/src/enc/config_enc.c \
     $$PWD/libwebp/src/enc/cost_enc.c \
@@ -131,7 +131,8 @@ equals(QT_ARCH, arm)|equals(QT_ARCH, arm64) {
         $$PWD/libwebp/src/dsp/lossless_enc_neon.c \
         $$PWD/libwebp/src/dsp/lossless_neon.c \
         $$PWD/libwebp/src/dsp/rescaler_neon.c \
-        $$PWD/libwebp/src/dsp/upsampling_neon.c
+        $$PWD/libwebp/src/dsp/upsampling_neon.c \
+        $$PWD/libwebp/src/dsp/yuv_neon.c
 
     contains(QT_CPU_FEATURES.$$QT_ARCH, neon) {
         # Default compiler settings include this feature, so just add to SOURCES
diff --git a/src/3rdparty/libwebp/ChangeLog b/src/3rdparty/libwebp/ChangeLog
index 7ac7b5f..b17feb2 100644
--- a/src/3rdparty/libwebp/ChangeLog
+++ b/src/3rdparty/libwebp/ChangeLog
@@ -1,9 +1,294 @@
+f66955de WEBP_REDUCE_CSP: restrict colorspace support
+a289d8e7 update ChangeLog (tag: v0.6.1-rc2)
+c10a493c vwebp: disable double buffering on windows & mac
+0d4466c2 webp_to_sdl.c: fix file mode
+1b27bf8b WEBP_REDUCE_SIZE: disable all rescaler code
+126be109 webpinfo: add -version option
+9add62b5 bump version to 0.6.1
+d3e26144 update NEWS
+2edda639 README: add webpinfo section
+9ca568ef Merge "right-size some tables"
+31f1995c Merge "SSE2 implementation of HasAlphaXXX"
+a80c46bd SSE2 implementation of HasAlphaXXX
+083507f2 right-size some tables
+2e5785b2 anim_utils.c: remove warning when !defined(WEBP_HAVE_GIF)
+b299c47e add WEBP_REDUCE_SIZE
+f593d71a enc: disable pic->stats/extra_info w/WEBP_DISABLE_STATS
+541179a9 Merge "predictor_enc: fix build w/--disable-near-lossless"
+5755a7ec predictor_enc: fix build w/--disable-near-lossless
+eab5bab7 add WEBP_DISABLE_STATS
+8052c585 remove some petty TODOs from vwebp.
+c245343d move LOAD8x4 and STORE8x2 closer to their use location
+b9e734fd dec,cosmetics: normalize function naming style
+c188d546 dec: harmonize function suffixes
+28c5ac81 dec_sse41: harmonize function suffixes
+e65b72a3 Merge "introduce WebPHasAlpha8b and WebPHasAlpha32b"
+b94cee98 dec_sse2: remove HE8uv_SSE2
+44a0ee3f introduce WebPHasAlpha8b and WebPHasAlpha32b
+aebf59ac Merge "WebPPictureAllocARGB: align argb allocation"
+c184665e WebPPictureAllocARGB: align argb allocation
+3daf7509 WebPParseHeaders: remove obsolete animation TODO
+80285d97 cmake: avoid security warnings under msvc
+650eac55 cmake: don't set -Wall with MSVC
+c462cd00 Remove useless code.
+01a98217 Merge "remove WebPWorkerImpl declaration from the header"
+3c49fc47 Merge "thread_utils: fix potentially bad call to Execute"
+fde2782e thread_utils: fix potentially bad call to Execute
+2a270c1d remove WebPWorkerImpl declaration from the header
+f1f437cc remove mention of 'lossy-only parameters' from the doc
+3879074d Merge "WebPMemToUint32: remove ptr cast to int"
+04b029d2 WebPMemToUint32: remove ptr cast to int
+b7971d0e dsp: avoid defining _C functions w/NEON builds
+6ba98764 webpdec: correct alloc size check w/use_argb
+5cfb3b0f normalize include guards
+f433205e Merge changes Ia17c7dfc,I75423abb,Ia2f716b4,I161caa14,I4210081a, ...
+8d033b14 {dec,enc}_neon: harmonize function suffixes x2
+0295e981 upsampling_neon: harmonize function suffixes
+d572c4e5 yuv_neon: harmonize function suffixes
+ab9c2500 rescaler_neon: harmonize function suffixes
+93e0ce27 lossless_neon: harmonize function suffixes
+22fbc50e lossless_enc_neon: harmonize function suffixes
+447875b4 filters_neon,cosmetics: fix indent
+e51bdd43 remove unused VP8TokenToStats() function
+785da7ea enc_neon: harmonize function suffixes
+bc1a251f dec_neon: harmonize function suffixes
+61e535f1 dsp/lossless: workaround gcc-4.8 bug on arm
+68b2eab7 cwebp: fix alpha reporting w/lossless & metadata
+30042faa WebPDemuxGetI: add doc details around WebPFormatFeature
+0a17f471 Merge "WIP: list includes as descendants of the project dir"
+a4399721 WIP: list includes as descendants of the project dir
+08275708 Merge "Make sure we reach the full range for alpha blending."
+d361a6a7 yuv_sse2: harmonize function suffixes
+6921aa6f upsampling_sse2: harmonize function suffixes
+08c67d3e ssim_sse2: harmonize function suffixes
+582a1b57 rescaler_sse2: harmonize function suffixes
+2c1b18ba lossless_sse2: harmonize function suffixes
+0ac46e81 lossless_enc_sse2: harmonize function suffixes
+bc634d57 enc_sse2: harmonize function suffixes
+bcb7347c dec_sse2: harmonize function suffixes
+e14ad93c Make sure we reach the full range for alpha blending.
+7038ca8d demux,StoreFrame: restore hdr size check to min req
+fb3daad6 cpu: fix ssse3 check
+be590e06 Merge "Fix CMake redefinition for HAVE_CPU_FEATURES_H"
+35f736e1 Fix CMake redefinition for HAVE_CPU_FEATURES_H
+a5216efc Fix integer overflow warning.
+a9c8916b decode.h,WebPIDecGetRGB: clarify output ptr validity
+3c74c645 gif2webp: handle 1-frame case properly + fix anim_diff
+c7f295d3 Merge "gif2webp: introduce -loop_compatibility option"
+b4e04677 gif2webp: introduce -loop_compatibility option
+f78da3de add LOCAL_CLANG_PREREQ and avoid WORK_AROUND_GCC w/3.8+
+01c426f1 define WEBP_USE_INTRINSICS w/gcc-4.9+
+8635973d use sdl-config (if available) to determine the link flags
+e9459382 use CPPFLAGS before CFLAGS
+4a9d788e Merge "Android.mk,mips: fix clang build with r15"
+4fbdc9fb Android.mk,mips: fix clang build with r15
+a80fcc4a ifdef code not used by Chrome/Android.
+3993af12 Fix signed integer overflows.
+f66f94ef anim_dump: small tool to dump frames from animated WebP
+6eba857b Merge "rationalize the Makefile.am"
+c5e34fba function definition cleanup
+3822762a rationalize the Makefile.am
+501ef6e4 configure style fix: animdiff -> anim_diff
+f8bdc268 Merge "protect against NULL dump_folder[] value in ReadAnimatedImage()"
+23bfc652 protect against NULL dump_folder[] value in ReadAnimatedImage()
+8dc3d71b cosmetics,ReadAnimatedWebP: correct function comment
+5bd40066 Merge changes I66a64a0a,I4d2e520f
+7945575c cosmetics,webpinfo: remove an else after a return
+8729fa11 cosmetics,cwebp: remove an else after a return
+f324b7f9 cosmetics: normalize fn proto & decl param names
+869eb369 CMake cleanups.
+289e62a3 Remove declaration of unimplemented VP8ApplyNearLosslessPredict
+20a94186 pnmdec,PAM: validate depth before calculating bytes_per_px
+34130afe anim_encode: fix integer overflow
+42c79aa6 Merge "Encoder: harmonize function suffixes"
+b09307dc Encoder: harmonize function suffixes
+bed0456d Merge "SSIM: harmonize the function suffix"
+54f6a3cf lossless_sse2.c: fix some missed suffix changes
+088f1dcc SSIM: harmonize the function suffix
+86fc4dd9 webpdec: use ImgIoUtilCheckSizeArgumentsOverflow
+08ea9ecd imageio: add ability restrict max image size
+6f9daa4a jpegdec,ReadError: fix leaks on error
+a0f72a4f VP8LTransformColorFunc: drop an non-respected 'const' from the signature.
+8c934902 Merge "Lossess dec: harmonize the function suffixes"
+622242aa Lossess dec: harmonize the function suffixes
+1411f027 Lossless Enc: harmonize the function suffixes
+24ad2e3c add const to two variables
+46efe062 Merge "Allow the lossless cruncher to work for alpha."
+8c3f9a47 Speed-up LZ77.
+1aef4c71 Allow the lossless cruncher to work for alpha.
+b8821dbd Improve the box LZ77 speed.
+7beed280 add missing ()s to macro parameters
+6473d20b Merge "fix Android standalone toolchain build"
+dcefed95 Merge "build.gradle: fix arm64 build"
+0c83a8bc Merge "yuv: harmonize suffix naming"
+c6d1db4b fix Android standalone toolchain build
+663a6d9d unify the ALTERNATE_CODE flag usage
+73ea9f27 yuv: harmonize suffix naming
+c71b68ac build.gradle: fix arm64 build
+c4568b47 Rescaler: harmonize the suffix naming
+6cb13b05 Merge "alpha_processing: harmonize the naming suffixes to be _C()"
+83a3e69a Merge "simplify WEBP_EXTERN macro"
+7295fde2 Merge "filters: harmonize the suffixes naming to _SSE2(), _C(), etc."
+8e42ba4c simplify WEBP_EXTERN macro
+331ab34b cost*.c: harmonize the suffix namings
+b161f670 filters: harmonize the suffixes naming to _SSE2(), _C(), etc.
+dec5e4d3 alpha_processing: harmonize the naming suffixes to be _C()
+6878d427 fix memory leak in SDL_Init()
+461ae555 Merge "configure: fix warnings in sdl check"
+62486a22 configure: test for -Wundef
+92982609 dsp.h: fix -Wundef w/__mips_dsp_rev
+0265cede configure: fix warnings in sdl check
+88c73d8a backward_references_enc.h: fix WINDOW_SIZE_BITS check
+4ea49f6b rescaler_sse2.c: fix WEBP_RESCALER_FIX -> _RFIX typo
+1b526638 Clean-up some CMake
+87f57a4b Merge "cmake: fix gif lib detection when cross compiling"
+b34a9db1 cosmetics,dec_sse2: remove some redundant comments
+471c5755 cmake: fix gif lib detection when cross compiling
+c793417a cmake: disable gif2webp if gif lib isn't found
+dcbc1c88 cmake: split gif detection from IMG deps
+66ad84f0 Merge "muxread: remove unreachable code"
+50ec3ab7 muxread: remove unreachable code
+7d67a164 Lossy encoding: smoothen transparent areas to improve compression
+e50650c7 Merge "fix signature for DISABLE_TOKEN_BUFFER compilation"
+671d2567 fix signature for DISABLE_TOKEN_BUFFER compilation
+d6755580 cpu.cmake: use unique flag to test simd disable flags
+28914528 Merge "Remove the argb* files."
+8acb4942 Remove the argb* files.
+3b62347b README: correct cmake invocation note
+7ca0df13 Have the SSE2 version of PackARGB use common code.
+7b250459 Merge "Re-use the transformed image when trying several LZ77 in lossless."
+e132072f Re-use the transformed image when trying several LZ77 in lossless.
+5d7a50ef Get code to compile in C++.
+7b012987 configure: test for -Wparentheses-equality
+f0569adb Fix man pages for multi-threading.
+f1d5a397 multithread cruncher: only copy stats when picture->stats != NULL
+f8c2ac15 Multi-thread the lossless cruncher.
+a88c6522 Merge "Integrate a new LZ77 looking for matches in the neighborhood of a pixel only."
+8f6df1d0 Unroll Predictors 10, 11 and 12.
+355c3d1b Integrate a new LZ77 looking for matches in the neighborhood of a pixel only.
+a1779a01 Refactor LZ77 handling in preparation for a new method.
+67de68b5 Android.mk/build.gradle: fix mips build with clang from r14b
+f209a548 Use the plane code and not the distance when computing statistics.
+b903b80c Split cost-based backward references in its own file.
+498cad34 Cosmetic changes in backward reference.
+e4eb4587 lossless, VP8LTransformColor_C: make sure no overflow happens with colors.
+af6deaff webpinfo: handle alpha flag mismatch
+7caef29b Fix typo that creeped in.
+39e19f92 Merge "near lossless: fix unsigned int overflow warnings."
+9bbc0891 near lossless: fix unsigned int overflow warnings.
+e1118d62 Merge "cosmetics,FindClosestDiscretized: use uint in mask creation"
+186bc9b7 Merge "webpinfo: tolerate ALPH+VP8L"
+b5887297 cosmetics,FindClosestDiscretized: use uint in mask creation
+f1784aee near_lossless,FindClosestDiscretized: use unsigned ops
+0d20abb3 webpinfo: tolerate ALPH+VP8L
+972104b3 webpmux: tolerate false positive Alpha flag
+dd7e83cc tiffdec,ReadTIFF: ensure data_size is < tsize_t max
+d988eb7b tiffdec,MyRead: quiet -Wshorten-64-to-32 warning
+dabda707 webpinfo: add support to parse Alpha bitstream
+4c117643 webpinfo: correct background color output, BGRA->ARGB
+defc98d7 Doc: clarify the role of quality in WebPConfig.
+d78ff780 Merge "Fix code to compile with C++."
+c8f14093 Fix code to compile with C++.
+497dc6a7 pnmdec: sanitize invalid header output
+d78e5867 Merge "configure: test for -Wconstant-conversion"
+481e91eb Merge "pnmdec,PAM: set bytes_per_px based on depth when missing"
+93b12753 configure: test for -Wconstant-conversion
+645f0c53 pnmdec,PAM: set bytes_per_px based on depth when missing
+e9154605 Merge "vwebp: activate GLUT double-buffering"
+818d795b vwebp: activate GLUT double-buffering
+d63e6f4b Add a man page for webpinfo
+4d708435 Merge "NEON: implement ConvertRGB24ToY/BGR24/ARGB/RGBA32ToUV/ARGBToUV"
+faf42213 NEON: implement ConvertRGB24ToY/BGR24/ARGB/RGBA32ToUV/ARGBToUV
+b4d576fa Install man pages with CMake.
+cbc1b921 webpinfo: add features to parse bitstream header
+e644c556 Fix bad bit writer initialization.
+b62cdad2 Merge "Implement a cruncher for lossless at method 6."
+da3e4dfb use the exact constant for the gamma transfer function
+a9c701e0 Merge "tiffdec: fix EXTRASAMPLES check"
+adab8ce0 Implement a cruncher for lossless at method 6.
+1b92b237 Merge "Fix VP8ApplyNearLossless to respect const and stride."
+1923ff02 tiffdec: fix EXTRASAMPLES check
+97cce5ba tiffdec: only request EXTRASAMPLES w/> 3 samples/px
+0dcd85b6 Fix VP8ApplyNearLossless to respect const and stride.
+f7682189 yuv: rationalize the C/SSE2 function naming
+52245424 NEON implementation of some Sharp-YUV420 functions
+690efd82 Avoid several backward reference copies.
+4bb1f607 src/dec/vp8_dec.h, cosmetics: fix comments
+285748be cmake: build/install webpinfo
+78fd199c backward_references_enc.c: clear -Wshadow warnings
+ae836410 WebPLog2FloorC: clear -Wshadow warning
+d0b7404e Merge "WASM support"
+134e314f WASM support
+c08adb6f Merge "VP8LEnc: remove use of BitsLog2Ceiling()"
+28c37ebd VP8LEnc: remove use of BitsLog2Ceiling()
+2cb58ab2 webpinfo: output format as a human readable string
+bb175a93 Merge "rename some symbols clashing with MSVC headers"
+39eda658 Remove a duplicated pixel hash implementation.
+36b8274d rename some symbols clashing with MSVC headers
+274daf54 Add webpinfo tool.
+ec5036e4 add explicit reference to /usr/local/{lib,inc}
+18f0dfac Merge "fix TIFF encoder regarding rgbA/RGBA"
+4e2b0b50 Merge "webpdec.h: fix a doc typo"
+e2eeabff Merge "Install binaries, libraries and headers in CMake."
+836607e6 webpdec.h: fix a doc typo
+9273e441 fix TIFF encoder regarding rgbA/RGBA
+17e3c11f Add limited PAM decoding support
+5f624871 Install binaries, libraries and headers in CMake.
+976adac1 Merge "lossless incremental decoding: fix missing eos_ test"
+f8fad4fa lossless incremental decoding: fix missing eos_ test
+27415d41 Merge "vwebp_sdl: fix the makefile.unix"
+49566182 Merge "ImgIoUtilWriteFile(): use ImgIoUtilSetBinaryMode"
+6f75a51b Analyze the transform entropy on the whole image.
+a5e4e3af Use palette only if we can in entropy analysis.
+75a9c3c4 Improve compression by better entropy analysis.
+39cf6f4f vwebp_sdl: fix the makefile.unix
+699b0416 ImgIoUtilWriteFile(): use ImgIoUtilSetBinaryMode
+7d985bd1 Fix small entropy analysis bug.
+6e7caf06 Optimize the color cache size.
+833c9219 More efficient stochastic histogram merge.
+5183326b Refactor the greedy histogram merge.
+99f6f462 Merge "histogram_enc.c,MyRand: s/ul/u/ for unsigned constants"
+80a22186 ssim.c: remove dead include
+a128dfff histogram_enc.c,MyRand: s/ul/u/ for unsigned constants
+693bf74e move the SSIM calculation code in ssim.c / ssim_sse2.c
+10d791ca Merge "Fix the random generator in HistogramCombineStochastic."
+fa63a966 Fix the random generator in HistogramCombineStochastic.
+16be192f VP8LSetBitPos: remove the eos_ setting
+027151ca don't erase the surface before blitting.
+4105d565 disable WEBP_USE_XXX optimisations when EMSCRIPTEN is defined
+9ee32a75 Merge "WebP-JS: emscripten-based Javascript decoder"
+ca9f7b7d WebP-JS: emscripten-based Javascript decoder
+868aa690 Perform greedy histogram merge in a unified way.
+5b393f2d Merge "fix path typo for vwebp_sdl in Makefile.vc"
+e0012bea CMake: only use libwebpdecoder for building dwebp
+84c2a7b0 fix path typo for vwebp_sdl in Makefile.vc
+1b0e4abf Merge "Add a flag to disable SIMD optimizations."
+32263250 Add a flag to disable SIMD optimizations.
+b494fdec optimize the ARGB->ARGB Import to use memcpy
+f1536039 Merge "ReadWebP: decode directly into a pre-allocated buffer"
+e69ed291 ReadWebP: decode directly into a pre-allocated buffer
+57d8de8a Merge "vwebp_sdl: simple viewer based on SDL"
+5cfd4ebc LZ77 interval speedups. Faster, smaller, simpler.
+1e7ad88b PNM header decoder: add some basic numerical validation
+17c7890c Merge "Add a decoder only library for WebP in CMake."
+be733786 Merge "Add clang build fix for MSA"
+03cda0e4 Add a decoder only library for WebP in CMake.
+aa893914 Add clang build fix for MSA
+31a92e97 Merge "imageio: add limited PNM support for reading"
+dcf9d82a imageio: add limited PNM support for reading
+6524fcd6 vwebp_sdl: simple viewer based on SDL
+6cf24a24 get_disto: fix reference file read
+43d472aa Merge tag 'v0.6.0'
+50d1a848 update ChangeLog (tag: v0.6.0, origin/0.6.0, 0.6.0)
 20a7fea0 extras/Makefile.am: fix libwebpextras.la reference
 415f3ffe update ChangeLog (tag: v0.6.0-rc3)
 3c6d1224 update NEWS
 ee4a4141 update AUTHORS
 32ed856f Fix "all|no frames are keyframes" settings.
+1c3190b6 Merge "Fix "all|no frames are keyframes" settings."
 f4dc56fd disable GradientUnfilter_NEON
+4f3e3bbd disable GradientUnfilter_NEON
+2dc0bdca Fix "all|no frames are keyframes" settings.
 0d8e0588 img2webp: treat -loop as a no-op w/single images
 b0450139 ReadImage(): restore size reporting
 0ad3b4ef update ChangeLog (tag: v0.6.0-rc2)
diff --git a/src/3rdparty/libwebp/NEWS b/src/3rdparty/libwebp/NEWS
index 3bf4bd0..85d273e 100644
--- a/src/3rdparty/libwebp/NEWS
+++ b/src/3rdparty/libwebp/NEWS
@@ -1,3 +1,16 @@
+- 11/24/2017: version 0.6.1
+  This is a binary compatible release.
+  * lossless performance and compression improvements + a new 'cruncher' mode
+    (-m 6 -q 100)
+  * ARM performance improvements with clang (15-20% w/ndk r15c, issue #339)
+  * webp-js: emscripten/webassembly based javascript decoder
+  * miscellaneous bug & build fixes (issue #329, #332, #343, #353, #360, #361,
+    #363)
+  Tool updates / additions:
+    added webpinfo - prints file format information (issue #330)
+    gif2webp - loop behavior modified to match Chrome M63+ (crbug.com/649264);
+               '-loop_compatibility' can be used for the old behavior
+
 - 1/26/2017: version 0.6.0
   * lossless performance and compression improvements
   * miscellaneous performance improvements (SSE2, NEON, MSA)
diff --git a/src/3rdparty/libwebp/README b/src/3rdparty/libwebp/README
index 4c15c4a..e9817bf 100644
--- a/src/3rdparty/libwebp/README
+++ b/src/3rdparty/libwebp/README
@@ -4,7 +4,7 @@
           \__\__/\____/\_____/__/ ____  ___
                 / _/ /    \    \ /  _ \/ _/
                /  \_/   / /   \ \   __/  \__
-               \____/____/\_____/_____/____/v0.6.0
+               \____/____/\_____/_____/____/v0.6.1
 
 Description:
 ============
@@ -113,8 +113,8 @@ make install
 
 CMake:
 ------
-The support for CMake is minimal: it only helps you compile libwebp, cwebp and
-dwebp.
+With CMake, you can compile libwebp, cwebp, dwebp, gif2web, img2webp, webpinfo
+and the JS bindings.
 
 Prerequisites:
 A compiler (e.g., gcc with autotools) and CMake.
@@ -123,18 +123,25 @@ minimal build:
 $ sudo apt-get install build-essential cmake
 
 When building from git sources, you will need to run cmake to generate the
-configure script.
+makefiles.
 
 mkdir build && cd build && cmake ../
 make
 make install
 
-If you also want cwebp or dwebp, you will need to enable them through CMake:
+If you also want any of the executables, you will need to enable them through
+CMake, e.g.:
 
 cmake -DWEBP_BUILD_CWEBP=ON -DWEBP_BUILD_DWEBP=ON ../
 
 or through your favorite interface (like ccmake or cmake-qt-gui).
 
+Finally, once installed, you can also use WebP in your CMake project by doing:
+
+find_package(WebP)
+
+which will define the CMake variables WebP_INCLUDE_DIRS and WebP_LIBRARIES.
+
 Gradle:
 -------
 The support for Gradle is minimal: it only helps you compile libwebp, cwebp and
@@ -360,6 +367,23 @@ Use following options to convert into alternate image formats:
   -quiet ....... quiet mode, don't print anything
   -noasm ....... disable all assembly optimizations
 
+WebP file analysis tool:
+========================
+
+'webpinfo' can be used to print out the chunk level structure and bitstream
+header information of WebP files. It can also check if the files are of valid
+WebP format.
+
+Usage: webpinfo [options] in_files
+Note: there could be multiple input files;
+      options must come before input files.
+Options:
+  -version ........... Print version number and exit.
+  -quiet ............. Do not show chunk parsing information.
+  -diag .............. Show parsing error diagnosis.
+  -summary ........... Show chunk stats summary.
+  -bitstream_info .... Parse bitstream header.
+
 Visualization tool:
 ===================
 
@@ -470,6 +494,8 @@ Options:
   -metadata <string> ..... comma separated list of metadata to
                            copy from the input to the output if present
                            Valid values: all, none, icc, xmp (default)
+  -loop_compatibility .... use compatibility mode for Chrome
+                           version prior to M62 (inclusive)
   -mt .................... use multi-threading if available
 
   -version ............... print version number and exit
diff --git a/src/3rdparty/libwebp/qt_attribution.json b/src/3rdparty/libwebp/qt_attribution.json
index 09f91dd..70165be 100644
--- a/src/3rdparty/libwebp/qt_attribution.json
+++ b/src/3rdparty/libwebp/qt_attribution.json
@@ -6,7 +6,7 @@
 
     "Description": "WebP is a new image format that provides lossless and lossy compression for images on the web.",
     "Homepage": "https://developers.google.com/speed/webp/",
-    "Version": "0.6.0",
+    "Version": "0.6.1",
     "License": "BSD 3-clause \"New\" or \"Revised\" License",
     "LicenseId": "BSD-3-Clause",
     "LicenseFile": "COPYING",
diff --git a/src/3rdparty/libwebp/src/dec/alpha_dec.c b/src/3rdparty/libwebp/src/dec/alpha_dec.c
index 83ffd4b..bce735b 100644
--- a/src/3rdparty/libwebp/src/dec/alpha_dec.c
+++ b/src/3rdparty/libwebp/src/dec/alpha_dec.c
@@ -12,13 +12,13 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <stdlib.h>
-#include "./alphai_dec.h"
-#include "./vp8i_dec.h"
-#include "./vp8li_dec.h"
-#include "../dsp/dsp.h"
-#include "../utils/quant_levels_dec_utils.h"
-#include "../utils/utils.h"
-#include "../webp/format_constants.h"
+#include "src/dec/alphai_dec.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/quant_levels_dec_utils.h"
+#include "src/utils/utils.h"
+#include "src/webp/format_constants.h"
 
 //------------------------------------------------------------------------------
 // ALPHDecoder object.
diff --git a/src/3rdparty/libwebp/src/dec/alphai_dec.h b/src/3rdparty/libwebp/src/dec/alphai_dec.h
index 561e815..e0fa281 100644
--- a/src/3rdparty/libwebp/src/dec/alphai_dec.h
+++ b/src/3rdparty/libwebp/src/dec/alphai_dec.h
@@ -11,11 +11,11 @@
 //
 // Author: Urvang (urvang@google.com)
 
-#ifndef WEBP_DEC_ALPHAI_H_
-#define WEBP_DEC_ALPHAI_H_
+#ifndef WEBP_DEC_ALPHAI_DEC_H_
+#define WEBP_DEC_ALPHAI_DEC_H_
 
-#include "./webpi_dec.h"
-#include "../utils/filters_utils.h"
+#include "src/dec/webpi_dec.h"
+#include "src/utils/filters_utils.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -51,4 +51,4 @@ void WebPDeallocateAlphaMemory(VP8Decoder* const dec);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_ALPHAI_H_ */
+#endif  /* WEBP_DEC_ALPHAI_DEC_H_ */
diff --git a/src/3rdparty/libwebp/src/dec/buffer_dec.c b/src/3rdparty/libwebp/src/dec/buffer_dec.c
index c685fd5..75eb3c4 100644
--- a/src/3rdparty/libwebp/src/dec/buffer_dec.c
+++ b/src/3rdparty/libwebp/src/dec/buffer_dec.c
@@ -13,15 +13,15 @@
 
 #include <stdlib.h>
 
-#include "./vp8i_dec.h"
-#include "./webpi_dec.h"
-#include "../utils/utils.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/dec/webpi_dec.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // WebPDecBuffer
 
 // Number of bytes per pixel for the different color-spaces.
-static const int kModeBpp[MODE_LAST] = {
+static const uint8_t kModeBpp[MODE_LAST] = {
   3, 4, 3, 4, 4, 2, 2,
   4, 4, 4, 2,    // pre-multiplied modes
   1, 1 };
@@ -36,7 +36,7 @@ static int IsValidColorspace(int webp_csp_mode) {
 // strictly speaking, the very last (or first, if flipped) row
 // doesn't require padding.
 #define MIN_BUFFER_SIZE(WIDTH, HEIGHT, STRIDE)       \
-    (uint64_t)(STRIDE) * ((HEIGHT) - 1) + (WIDTH)
+    ((uint64_t)(STRIDE) * ((HEIGHT) - 1) + (WIDTH))
 
 static VP8StatusCode CheckDecBuffer(const WebPDecBuffer* const buffer) {
   int ok = 1;
@@ -98,9 +98,14 @@ static VP8StatusCode AllocateBuffer(WebPDecBuffer* const buffer) {
     uint64_t uv_size = 0, a_size = 0, total_size;
     // We need memory and it hasn't been allocated yet.
     // => initialize output buffer, now that dimensions are known.
-    const int stride = w * kModeBpp[mode];
-    const uint64_t size = (uint64_t)stride * h;
+    int stride;
+    uint64_t size;
 
+    if ((uint64_t)w * kModeBpp[mode] >= (1ull << 32)) {
+      return VP8_STATUS_INVALID_PARAM;
+    }
+    stride = w * kModeBpp[mode];
+    size = (uint64_t)stride * h;
     if (!WebPIsRGBMode(mode)) {
       uv_stride = (w + 1) / 2;
       uv_size = (uint64_t)uv_stride * ((h + 1) / 2);
@@ -169,11 +174,11 @@ VP8StatusCode WebPFlipBuffer(WebPDecBuffer* const buffer) {
   return VP8_STATUS_OK;
 }
 
-VP8StatusCode WebPAllocateDecBuffer(int w, int h,
+VP8StatusCode WebPAllocateDecBuffer(int width, int height,
                                     const WebPDecoderOptions* const options,
-                                    WebPDecBuffer* const out) {
+                                    WebPDecBuffer* const buffer) {
   VP8StatusCode status;
-  if (out == NULL || w <= 0 || h <= 0) {
+  if (buffer == NULL || width <= 0 || height <= 0) {
     return VP8_STATUS_INVALID_PARAM;
   }
   if (options != NULL) {    // First, apply options if there is any.
@@ -182,33 +187,39 @@ VP8StatusCode WebPAllocateDecBuffer(int w, int h,
       const int ch = options->crop_height;
       const int x = options->crop_left & ~1;
       const int y = options->crop_top & ~1;
-      if (x < 0 || y < 0 || cw <= 0 || ch <= 0 || x + cw > w || y + ch > h) {
+      if (x < 0 || y < 0 || cw <= 0 || ch <= 0 ||
+          x + cw > width || y + ch > height) {
         return VP8_STATUS_INVALID_PARAM;   // out of frame boundary.
       }
-      w = cw;
-      h = ch;
+      width = cw;
+      height = ch;
     }
+
     if (options->use_scaling) {
+#if !defined(WEBP_REDUCE_SIZE)
       int scaled_width = options->scaled_width;
       int scaled_height = options->scaled_height;
       if (!WebPRescalerGetScaledDimensions(
-              w, h, &scaled_width, &scaled_height)) {
+              width, height, &scaled_width, &scaled_height)) {
         return VP8_STATUS_INVALID_PARAM;
       }
-      w = scaled_width;
-      h = scaled_height;
+      width = scaled_width;
+      height = scaled_height;
+#else
+      return VP8_STATUS_INVALID_PARAM;   // rescaling not supported
+#endif
     }
   }
-  out->width = w;
-  out->height = h;
+  buffer->width = width;
+  buffer->height = height;
 
   // Then, allocate buffer for real.
-  status = AllocateBuffer(out);
+  status = AllocateBuffer(buffer);
   if (status != VP8_STATUS_OK) return status;
 
   // Use the stride trick if vertical flip is needed.
   if (options != NULL && options->flip) {
-    status = WebPFlipBuffer(out);
+    status = WebPFlipBuffer(buffer);
   }
   return status;
 }
diff --git a/src/3rdparty/libwebp/src/dec/common_dec.h b/src/3rdparty/libwebp/src/dec/common_dec.h
index 6961e22..9995f1a 100644
--- a/src/3rdparty/libwebp/src/dec/common_dec.h
+++ b/src/3rdparty/libwebp/src/dec/common_dec.h
@@ -11,8 +11,8 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_DEC_COMMON_H_
-#define WEBP_DEC_COMMON_H_
+#ifndef WEBP_DEC_COMMON_DEC_H_
+#define WEBP_DEC_COMMON_DEC_H_
 
 // intra prediction modes
 enum { B_DC_PRED = 0,   // 4x4 modes
@@ -51,4 +51,4 @@ enum { MB_FEATURE_TREE_PROBS = 3,
        NUM_PROBAS = 11
      };
 
-#endif    // WEBP_DEC_COMMON_H_
+#endif    // WEBP_DEC_COMMON_DEC_H_
diff --git a/src/3rdparty/libwebp/src/dec/frame_dec.c b/src/3rdparty/libwebp/src/dec/frame_dec.c
index f91e27f..517d0f5 100644
--- a/src/3rdparty/libwebp/src/dec/frame_dec.c
+++ b/src/3rdparty/libwebp/src/dec/frame_dec.c
@@ -12,13 +12,13 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <stdlib.h>
-#include "./vp8i_dec.h"
-#include "../utils/utils.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Main reconstruction function.
 
-static const int kScan[16] = {
+static const uint16_t kScan[16] = {
   0 +  0 * BPS,  4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS,
   0 +  4 * BPS,  4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS,
   0 +  8 * BPS,  4 +  8 * BPS, 8 +  8 * BPS, 12 +  8 * BPS,
@@ -320,7 +320,7 @@ static void PrecomputeFilterStrengths(VP8Decoder* const dec) {
 #define MIN_DITHER_AMP 4
 
 #define DITHER_AMP_TAB_SIZE 12
-static const int kQuantToDitherAmp[DITHER_AMP_TAB_SIZE] = {
+static const uint8_t kQuantToDitherAmp[DITHER_AMP_TAB_SIZE] = {
   // roughly, it's dqm->uv_mat_[1]
   8, 7, 6, 4, 4, 2, 2, 2, 1, 1, 1, 1
 };
@@ -728,7 +728,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
   }
 
   mem = (uint8_t*)dec->mem_;
-  dec->intra_t_ = (uint8_t*)mem;
+  dec->intra_t_ = mem;
   mem += intra_pred_mode_size;
 
   dec->yuv_t_ = (VP8TopSamples*)mem;
@@ -750,7 +750,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
 
   mem = (uint8_t*)WEBP_ALIGN(mem);
   assert((yuv_size & WEBP_ALIGN_CST) == 0);
-  dec->yuv_b_ = (uint8_t*)mem;
+  dec->yuv_b_ = mem;
   mem += yuv_size;
 
   dec->mb_data_ = (VP8MBData*)mem;
@@ -766,7 +766,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
     const int extra_rows = kFilterExtraRows[dec->filter_type_];
     const int extra_y = extra_rows * dec->cache_y_stride_;
     const int extra_uv = (extra_rows / 2) * dec->cache_uv_stride_;
-    dec->cache_y_ = ((uint8_t*)mem) + extra_y;
+    dec->cache_y_ = mem + extra_y;
     dec->cache_u_ = dec->cache_y_
                   + 16 * num_caches * dec->cache_y_stride_ + extra_uv;
     dec->cache_v_ = dec->cache_u_
@@ -776,7 +776,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
   mem += cache_size;
 
   // alpha plane
-  dec->alpha_plane_ = alpha_size ? (uint8_t*)mem : NULL;
+  dec->alpha_plane_ = alpha_size ? mem : NULL;
   mem += alpha_size;
   assert(mem <= (uint8_t*)dec->mem_ + dec->mem_size_);
 
diff --git a/src/3rdparty/libwebp/src/dec/idec_dec.c b/src/3rdparty/libwebp/src/dec/idec_dec.c
index 78fb2e7..a371ed7 100644
--- a/src/3rdparty/libwebp/src/dec/idec_dec.c
+++ b/src/3rdparty/libwebp/src/dec/idec_dec.c
@@ -15,10 +15,10 @@
 #include <string.h>
 #include <stdlib.h>
 
-#include "./alphai_dec.h"
-#include "./webpi_dec.h"
-#include "./vp8i_dec.h"
-#include "../utils/utils.h"
+#include "src/dec/alphai_dec.h"
+#include "src/dec/webpi_dec.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/utils/utils.h"
 
 // In append mode, buffer allocations increase as multiples of this value.
 // Needs to be a power of 2.
@@ -673,12 +673,12 @@ void WebPIDelete(WebPIDecoder* idec) {
 //------------------------------------------------------------------------------
 // Wrapper toward WebPINewDecoder
 
-WebPIDecoder* WebPINewRGB(WEBP_CSP_MODE mode, uint8_t* output_buffer,
+WebPIDecoder* WebPINewRGB(WEBP_CSP_MODE csp, uint8_t* output_buffer,
                           size_t output_buffer_size, int output_stride) {
   const int is_external_memory = (output_buffer != NULL) ? 1 : 0;
   WebPIDecoder* idec;
 
-  if (mode >= MODE_YUV) return NULL;
+  if (csp >= MODE_YUV) return NULL;
   if (is_external_memory == 0) {    // Overwrite parameters to sane values.
     output_buffer_size = 0;
     output_stride = 0;
@@ -689,7 +689,7 @@ WebPIDecoder* WebPINewRGB(WEBP_CSP_MODE mode, uint8_t* output_buffer,
   }
   idec = WebPINewDecoder(NULL);
   if (idec == NULL) return NULL;
-  idec->output_.colorspace = mode;
+  idec->output_.colorspace = csp;
   idec->output_.is_external_memory = is_external_memory;
   idec->output_.u.RGBA.rgba = output_buffer;
   idec->output_.u.RGBA.stride = output_stride;
diff --git a/src/3rdparty/libwebp/src/dec/io_dec.c b/src/3rdparty/libwebp/src/dec/io_dec.c
index 8bfab86..e603f19 100644
--- a/src/3rdparty/libwebp/src/dec/io_dec.c
+++ b/src/3rdparty/libwebp/src/dec/io_dec.c
@@ -13,11 +13,11 @@
 
 #include <assert.h>
 #include <stdlib.h>
-#include "../dec/vp8i_dec.h"
-#include "./webpi_dec.h"
-#include "../dsp/dsp.h"
-#include "../dsp/yuv.h"
-#include "../utils/utils.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/dec/webpi_dec.h"
+#include "src/dsp/dsp.h"
+#include "src/dsp/yuv.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Main YUV<->RGB conversion functions
@@ -212,7 +212,7 @@ static int EmitAlphaRGBA4444(const VP8Io* const io, WebPDecParams* const p,
     int num_rows;
     const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
     uint8_t* const base_rgba = buf->rgba + start_y * buf->stride;
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     uint8_t* alpha_dst = base_rgba;
 #else
     uint8_t* alpha_dst = base_rgba + 1;
@@ -241,6 +241,7 @@ static int EmitAlphaRGBA4444(const VP8Io* const io, WebPDecParams* const p,
 //------------------------------------------------------------------------------
 // YUV rescaling (no final RGB conversion needed)
 
+#if !defined(WEBP_REDUCE_SIZE)
 static int Rescale(const uint8_t* src, int src_stride,
                    int new_lines, WebPRescaler* const wrk) {
   int num_lines_out = 0;
@@ -431,7 +432,7 @@ static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos,
                                int max_lines_out) {
   const WebPRGBABuffer* const buf = &p->output->u.RGBA;
   uint8_t* const base_rgba = buf->rgba + y_pos * buf->stride;
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
   uint8_t* alpha_dst = base_rgba;
 #else
   uint8_t* alpha_dst = base_rgba + 1;
@@ -541,6 +542,8 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
   return 1;
 }
 
+#endif  // WEBP_REDUCE_SIZE
+
 //------------------------------------------------------------------------------
 // Default custom functions
 
@@ -561,10 +564,14 @@ static int CustomSetup(VP8Io* io) {
     WebPInitUpsamplers();
   }
   if (io->use_scaling) {
+#if !defined(WEBP_REDUCE_SIZE)
     const int ok = is_rgb ? InitRGBRescaler(io, p) : InitYUVRescaler(io, p);
     if (!ok) {
       return 0;    // memory error
     }
+#else
+    return 0;   // rescaling support not compiled
+#endif
   } else {
     if (is_rgb) {
       WebPInitSamplers();
@@ -598,9 +605,6 @@ static int CustomSetup(VP8Io* io) {
     }
   }
 
-  if (is_rgb) {
-    VP8YUVInit();
-  }
   return 1;
 }
 
diff --git a/src/3rdparty/libwebp/src/dec/quant_dec.c b/src/3rdparty/libwebp/src/dec/quant_dec.c
index 14e3198..f07212a 100644
--- a/src/3rdparty/libwebp/src/dec/quant_dec.c
+++ b/src/3rdparty/libwebp/src/dec/quant_dec.c
@@ -11,7 +11,7 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./vp8i_dec.h"
+#include "src/dec/vp8i_dec.h"
 
 static WEBP_INLINE int clip(int v, int M) {
   return v < 0 ? 0 : v > M ? M : v;
diff --git a/src/3rdparty/libwebp/src/dec/tree_dec.c b/src/3rdparty/libwebp/src/dec/tree_dec.c
index 9e805f6..3f5a957 100644
--- a/src/3rdparty/libwebp/src/dec/tree_dec.c
+++ b/src/3rdparty/libwebp/src/dec/tree_dec.c
@@ -11,15 +11,19 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./vp8i_dec.h"
-#include "../utils/bit_reader_inl_utils.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/utils/bit_reader_inl_utils.h"
 
+#if !defined(USE_GENERIC_TREE)
 #if !defined(__arm__) && !defined(_M_ARM) && !defined(__aarch64__)
 // using a table is ~1-2% slower on ARM. Prefer the coded-tree approach then.
-#define USE_GENERIC_TREE
+#define USE_GENERIC_TREE 1   // ALTERNATE_CODE
+#else
+#define USE_GENERIC_TREE 0
 #endif
+#endif  // USE_GENERIC_TREE
 
-#ifdef USE_GENERIC_TREE
+#if (USE_GENERIC_TREE == 1)
 static const int8_t kYModesIntra4[18] = {
   -B_DC_PRED, 1,
     -B_TM_PRED, 2,
@@ -317,7 +321,7 @@ static void ParseIntraMode(VP8BitReader* const br,
       int x;
       for (x = 0; x < 4; ++x) {
         const uint8_t* const prob = kBModesProba[top[x]][ymode];
-#ifdef USE_GENERIC_TREE
+#if (USE_GENERIC_TREE == 1)
         // Generic tree-parsing
         int i = kYModesIntra4[VP8GetBit(br, prob[0])];
         while (i > 0) {
@@ -335,7 +339,7 @@ static void ParseIntraMode(VP8BitReader* const br,
                         (!VP8GetBit(br, prob[6]) ? B_LD_PRED :
                           (!VP8GetBit(br, prob[7]) ? B_VL_PRED :
                             (!VP8GetBit(br, prob[8]) ? B_HD_PRED : B_HU_PRED)));
-#endif    // USE_GENERIC_TREE
+#endif  // USE_GENERIC_TREE
         top[x] = ymode;
       }
       memcpy(modes, top, 4 * sizeof(*top));
@@ -498,7 +502,7 @@ static const uint8_t
 
 // Paragraph 9.9
 
-static const int kBands[16 + 1] = {
+static const uint8_t kBands[16 + 1] = {
   0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7,
   0  // extra entry as sentinel
 };
diff --git a/src/3rdparty/libwebp/src/dec/vp8_dec.c b/src/3rdparty/libwebp/src/dec/vp8_dec.c
index fad8d9c..6212efd 100644
--- a/src/3rdparty/libwebp/src/dec/vp8_dec.c
+++ b/src/3rdparty/libwebp/src/dec/vp8_dec.c
@@ -13,12 +13,12 @@
 
 #include <stdlib.h>
 
-#include "./alphai_dec.h"
-#include "./vp8i_dec.h"
-#include "./vp8li_dec.h"
-#include "./webpi_dec.h"
-#include "../utils/bit_reader_inl_utils.h"
-#include "../utils/utils.h"
+#include "src/dec/alphai_dec.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/dec/webpi_dec.h"
+#include "src/utils/bit_reader_inl_utils.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 
diff --git a/src/3rdparty/libwebp/src/dec/vp8_dec.h b/src/3rdparty/libwebp/src/dec/vp8_dec.h
index b9337bb..ca85b34 100644
--- a/src/3rdparty/libwebp/src/dec/vp8_dec.h
+++ b/src/3rdparty/libwebp/src/dec/vp8_dec.h
@@ -11,10 +11,10 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_WEBP_DECODE_VP8_H_
-#define WEBP_WEBP_DECODE_VP8_H_
+#ifndef WEBP_DEC_VP8_DEC_H_
+#define WEBP_DEC_VP8_DEC_H_
 
-#include "../webp/decode.h"
+#include "src/webp/decode.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -33,7 +33,7 @@ extern "C" {
 //   /* customize io's functions (setup()/put()/teardown()) if needed. */
 //
 //   VP8Decoder* dec = VP8New();
-//   bool ok = VP8Decode(dec);
+//   int ok = VP8Decode(dec, &io);
 //   if (!ok) printf("Error: %s\n", VP8StatusMessage(dec));
 //   VP8Delete(dec);
 //   return ok;
@@ -157,24 +157,24 @@ void VP8Delete(VP8Decoder* const dec);
 // Miscellaneous VP8/VP8L bitstream probing functions.
 
 // Returns true if the next 3 bytes in data contain the VP8 signature.
-WEBP_EXTERN(int) VP8CheckSignature(const uint8_t* const data, size_t data_size);
+WEBP_EXTERN int VP8CheckSignature(const uint8_t* const data, size_t data_size);
 
 // Validates the VP8 data-header and retrieves basic header information viz
 // width and height. Returns 0 in case of formatting error. *width/*height
 // can be passed NULL.
-WEBP_EXTERN(int) VP8GetInfo(
+WEBP_EXTERN int VP8GetInfo(
     const uint8_t* data,
     size_t data_size,    // data available so far
     size_t chunk_size,   // total data size expected in the chunk
     int* const width, int* const height);
 
 // Returns true if the next byte(s) in data is a VP8L signature.
-WEBP_EXTERN(int) VP8LCheckSignature(const uint8_t* const data, size_t size);
+WEBP_EXTERN int VP8LCheckSignature(const uint8_t* const data, size_t size);
 
 // Validates the VP8L data-header and retrieves basic header information viz
 // width, height and alpha. Returns 0 in case of formatting error.
 // width/height/has_alpha can be passed NULL.
-WEBP_EXTERN(int) VP8LGetInfo(
+WEBP_EXTERN int VP8LGetInfo(
     const uint8_t* data, size_t data_size,  // data available so far
     int* const width, int* const height, int* const has_alpha);
 
@@ -182,4 +182,4 @@ WEBP_EXTERN(int) VP8LGetInfo(
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_WEBP_DECODE_VP8_H_ */
+#endif  /* WEBP_DEC_VP8_DEC_H_ */
diff --git a/src/3rdparty/libwebp/src/dec/vp8i_dec.h b/src/3rdparty/libwebp/src/dec/vp8i_dec.h
index 555853e..28244d9 100644
--- a/src/3rdparty/libwebp/src/dec/vp8i_dec.h
+++ b/src/3rdparty/libwebp/src/dec/vp8i_dec.h
@@ -11,16 +11,16 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_DEC_VP8I_H_
-#define WEBP_DEC_VP8I_H_
+#ifndef WEBP_DEC_VP8I_DEC_H_
+#define WEBP_DEC_VP8I_DEC_H_
 
 #include <string.h>     // for memcpy()
-#include "./common_dec.h"
-#include "./vp8li_dec.h"
-#include "../utils/bit_reader_utils.h"
-#include "../utils/random_utils.h"
-#include "../utils/thread_utils.h"
-#include "../dsp/dsp.h"
+#include "src/dec/common_dec.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/utils/bit_reader_utils.h"
+#include "src/utils/random_utils.h"
+#include "src/utils/thread_utils.h"
+#include "src/dsp/dsp.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -32,7 +32,7 @@ extern "C" {
 // version numbers
 #define DEC_MAJ_VERSION 0
 #define DEC_MIN_VERSION 6
-#define DEC_REV_VERSION 0
+#define DEC_REV_VERSION 1
 
 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
 // Constraints are: We need to store one 16x16 block of luma samples (y),
@@ -57,7 +57,6 @@ extern "C" {
 //  '|' = left sample,   '-' = top sample,    '+' = top-left sample
 //  't' = extra top-right sample for 4x4 modes
 #define YUV_SIZE (BPS * 17 + BPS * 9)
-#define Y_SIZE   (BPS * 17)
 #define Y_OFF    (BPS * 1 + 8)
 #define U_OFF    (Y_OFF + BPS * 16 + BPS)
 #define V_OFF    (U_OFF + 16)
@@ -317,4 +316,4 @@ const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_VP8I_H_ */
+#endif  /* WEBP_DEC_VP8I_DEC_H_ */
diff --git a/src/3rdparty/libwebp/src/dec/vp8l_dec.c b/src/3rdparty/libwebp/src/dec/vp8l_dec.c
index ef359a9..42ea3b5 100644
--- a/src/3rdparty/libwebp/src/dec/vp8l_dec.c
+++ b/src/3rdparty/libwebp/src/dec/vp8l_dec.c
@@ -14,22 +14,22 @@
 
 #include <stdlib.h>
 
-#include "./alphai_dec.h"
-#include "./vp8li_dec.h"
-#include "../dsp/dsp.h"
-#include "../dsp/lossless.h"
-#include "../dsp/lossless_common.h"
-#include "../dsp/yuv.h"
-#include "../utils/endian_inl_utils.h"
-#include "../utils/huffman_utils.h"
-#include "../utils/utils.h"
+#include "src/dec/alphai_dec.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/dsp/dsp.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+#include "src/dsp/yuv.h"
+#include "src/utils/endian_inl_utils.h"
+#include "src/utils/huffman_utils.h"
+#include "src/utils/utils.h"
 
 #define NUM_ARGB_CACHE_ROWS          16
 
 static const int kCodeLengthLiterals = 16;
 static const int kCodeLengthRepeatCode = 16;
-static const int kCodeLengthExtraBits[3] = { 2, 3, 7 };
-static const int kCodeLengthRepeatOffsets[3] = { 3, 3, 11 };
+static const uint8_t kCodeLengthExtraBits[3] = { 2, 3, 7 };
+static const uint8_t kCodeLengthRepeatOffsets[3] = { 3, 3, 11 };
 
 // -----------------------------------------------------------------------------
 //  Five Huffman codes are used at each meta code:
@@ -86,7 +86,7 @@ static const uint8_t kCodeToPlane[CODE_TO_PLANE_CODES] = {
 // All values computed for 8-bit first level lookup with Mark Adler's tool:
 // http://www.hdfgroup.org/ftp/lib-external/zlib/zlib-1.2.5/examples/enough.c
 #define FIXED_TABLE_SIZE (630 * 3 + 410)
-static const int kTableSize[12] = {
+static const uint16_t kTableSize[12] = {
   FIXED_TABLE_SIZE + 654,
   FIXED_TABLE_SIZE + 656,
   FIXED_TABLE_SIZE + 658,
@@ -485,6 +485,7 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
 //------------------------------------------------------------------------------
 // Scaling.
 
+#if !defined(WEBP_REDUCE_SIZE)
 static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) {
   const int num_channels = 4;
   const int in_width = io->mb_w;
@@ -516,10 +517,13 @@ static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) {
                    out_width, out_height, 0, num_channels, work);
   return 1;
 }
+#endif   // WEBP_REDUCE_SIZE
 
 //------------------------------------------------------------------------------
 // Export to ARGB
 
+#if !defined(WEBP_REDUCE_SIZE)
+
 // We have special "export" function since we need to convert from BGRA
 static int Export(WebPRescaler* const rescaler, WEBP_CSP_MODE colorspace,
                   int rgba_stride, uint8_t* const rgba) {
@@ -561,6 +565,8 @@ static int EmitRescaledRowsRGBA(const VP8LDecoder* const dec,
   return num_lines_out;
 }
 
+#endif   // WEBP_REDUCE_SIZE
+
 // Emit rows without any scaling.
 static int EmitRows(WEBP_CSP_MODE colorspace,
                     const uint8_t* row_in, int in_stride,
@@ -746,9 +752,12 @@ static void ProcessRows(VP8LDecoder* const dec, int row) {
       if (WebPIsRGBMode(output->colorspace)) {  // convert to RGBA
         const WebPRGBABuffer* const buf = &output->u.RGBA;
         uint8_t* const rgba = buf->rgba + dec->last_out_row_ * buf->stride;
-        const int num_rows_out = io->use_scaling ?
+        const int num_rows_out =
+#if !defined(WEBP_REDUCE_SIZE)
+         io->use_scaling ?
             EmitRescaledRowsRGBA(dec, rows_data, in_stride, io->mb_h,
                                  rgba, buf->stride) :
+#endif  // WEBP_REDUCE_SIZE
             EmitRows(output->colorspace, rows_data, in_stride,
                      io->mb_w, io->mb_h, rgba, buf->stride);
         // Update 'last_out_row_'.
@@ -1012,12 +1021,13 @@ static int DecodeAlphaData(VP8LDecoder* const dec, uint8_t* const data,
       ok = 0;
       goto End;
     }
-    assert(br->eos_ == VP8LIsEndOfStream(br));
+    br->eos_ = VP8LIsEndOfStream(br);
   }
   // Process the remaining rows corresponding to last row-block.
   ExtractPalettedAlphaRows(dec, row > last_row ? last_row : row);
 
  End:
+  br->eos_ = VP8LIsEndOfStream(br);
   if (!ok || (br->eos_ && pos < end)) {
     ok = 0;
     dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED
@@ -1090,11 +1100,12 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
     VP8LFillBitWindow(br);
     if (htree_group->use_packed_table) {
       code = ReadPackedSymbols(htree_group, br, src);
+      if (VP8LIsEndOfStream(br)) break;
       if (code == PACKED_NON_LITERAL_CODE) goto AdvanceByOne;
     } else {
       code = ReadSymbol(htree_group->htrees[GREEN], br);
     }
-    if (br->eos_) break;  // early out
+    if (VP8LIsEndOfStream(br)) break;
     if (code < NUM_LITERAL_CODES) {  // Literal
       if (htree_group->is_trivial_literal) {
         *src = htree_group->literal_arb | (code << 8);
@@ -1104,7 +1115,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
         VP8LFillBitWindow(br);
         blue = ReadSymbol(htree_group->htrees[BLUE], br);
         alpha = ReadSymbol(htree_group->htrees[ALPHA], br);
-        if (br->eos_) break;
+        if (VP8LIsEndOfStream(br)) break;
         *src = ((uint32_t)alpha << 24) | (red << 16) | (code << 8) | blue;
       }
     AdvanceByOne:
@@ -1132,7 +1143,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
       VP8LFillBitWindow(br);
       dist_code = GetCopyDistance(dist_symbol, br);
       dist = PlaneCodeToDistance(width, dist_code);
-      if (br->eos_) break;
+      if (VP8LIsEndOfStream(br)) break;
       if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) {
         goto Error;
       } else {
@@ -1169,9 +1180,9 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
     } else {  // Not reached
       goto Error;
     }
-    assert(br->eos_ == VP8LIsEndOfStream(br));
   }
 
+  br->eos_ = VP8LIsEndOfStream(br);
   if (dec->incremental_ && br->eos_ && src < src_end) {
     RestoreState(dec);
   } else if (!br->eos_) {
@@ -1630,12 +1641,19 @@ int VP8LDecodeImage(VP8LDecoder* const dec) {
 
     if (!AllocateInternalBuffers32b(dec, io->width)) goto Err;
 
+#if !defined(WEBP_REDUCE_SIZE)
     if (io->use_scaling && !AllocateAndInitRescaler(dec, io)) goto Err;
 
     if (io->use_scaling || WebPIsPremultipliedMode(dec->output_->colorspace)) {
       // need the alpha-multiply functions for premultiplied output or rescaling
       WebPInitAlphaProcessing();
     }
+#else
+    if (io->use_scaling) {
+      dec->status_ = VP8_STATUS_INVALID_PARAM;
+      goto Err;
+    }
+#endif
     if (!WebPIsRGBMode(dec->output_->colorspace)) {
       WebPInitConvertARGBToYUV();
       if (dec->output_->u.YUVA.a != NULL) WebPInitAlphaProcessing();
diff --git a/src/3rdparty/libwebp/src/dec/vp8li_dec.h b/src/3rdparty/libwebp/src/dec/vp8li_dec.h
index 097a9d0..8e500cf 100644
--- a/src/3rdparty/libwebp/src/dec/vp8li_dec.h
+++ b/src/3rdparty/libwebp/src/dec/vp8li_dec.h
@@ -12,14 +12,14 @@
 // Author: Skal (pascal.massimino@gmail.com)
 //         Vikas Arora(vikaas.arora@gmail.com)
 
-#ifndef WEBP_DEC_VP8LI_H_
-#define WEBP_DEC_VP8LI_H_
+#ifndef WEBP_DEC_VP8LI_DEC_H_
+#define WEBP_DEC_VP8LI_DEC_H_
 
 #include <string.h>     // for memcpy()
-#include "./webpi_dec.h"
-#include "../utils/bit_reader_utils.h"
-#include "../utils/color_cache_utils.h"
-#include "../utils/huffman_utils.h"
+#include "src/dec/webpi_dec.h"
+#include "src/utils/bit_reader_utils.h"
+#include "src/utils/color_cache_utils.h"
+#include "src/utils/huffman_utils.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -132,4 +132,4 @@ void VP8LDelete(VP8LDecoder* const dec);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_VP8LI_H_ */
+#endif  /* WEBP_DEC_VP8LI_DEC_H_ */
diff --git a/src/3rdparty/libwebp/src/dec/webp_dec.c b/src/3rdparty/libwebp/src/dec/webp_dec.c
index a8e9c2c..42d0988 100644
--- a/src/3rdparty/libwebp/src/dec/webp_dec.c
+++ b/src/3rdparty/libwebp/src/dec/webp_dec.c
@@ -13,11 +13,11 @@
 
 #include <stdlib.h>
 
-#include "./vp8i_dec.h"
-#include "./vp8li_dec.h"
-#include "./webpi_dec.h"
-#include "../utils/utils.h"
-#include "../webp/mux_types.h"  // ALPHA_FLAG
+#include "src/dec/vp8i_dec.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/dec/webpi_dec.h"
+#include "src/utils/utils.h"
+#include "src/webp/mux_types.h"  // ALPHA_FLAG
 
 //------------------------------------------------------------------------------
 // RIFF layout is:
@@ -421,7 +421,9 @@ VP8StatusCode WebPParseHeaders(WebPHeaderStructure* const headers) {
                                 NULL, NULL, NULL, &has_animation,
                                 NULL, headers);
   if (status == VP8_STATUS_OK || status == VP8_STATUS_NOT_ENOUGH_DATA) {
-    // TODO(jzern): full support of animation frames will require API additions.
+    // The WebPDemux API + libwebp can be used to decode individual
+    // uncomposited frames or the WebPAnimDecoder can be used to fully
+    // reconstruct them (see webp/demux.h).
     if (has_animation) {
       status = VP8_STATUS_UNSUPPORTED_FEATURE;
     }
diff --git a/src/3rdparty/libwebp/src/dec/webpi_dec.h b/src/3rdparty/libwebp/src/dec/webpi_dec.h
index 696abc1..c378ba6 100644
--- a/src/3rdparty/libwebp/src/dec/webpi_dec.h
+++ b/src/3rdparty/libwebp/src/dec/webpi_dec.h
@@ -11,15 +11,15 @@
 //
 // Author: somnath@google.com (Somnath Banerjee)
 
-#ifndef WEBP_DEC_WEBPI_H_
-#define WEBP_DEC_WEBPI_H_
+#ifndef WEBP_DEC_WEBPI_DEC_H_
+#define WEBP_DEC_WEBPI_DEC_H_
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-#include "../utils/rescaler_utils.h"
-#include "./vp8_dec.h"
+#include "src/utils/rescaler_utils.h"
+#include "src/dec/vp8_dec.h"
 
 //------------------------------------------------------------------------------
 // WebPDecParams: Decoding output parameters. Transient internal object.
@@ -130,4 +130,4 @@ int WebPAvoidSlowMemory(const WebPDecBuffer* const output,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_WEBPI_H_ */
+#endif  /* WEBP_DEC_WEBPI_DEC_H_ */
diff --git a/src/3rdparty/libwebp/src/demux/anim_decode.c b/src/3rdparty/libwebp/src/demux/anim_decode.c
index f1cf176..05dd707 100644
--- a/src/3rdparty/libwebp/src/demux/anim_decode.c
+++ b/src/3rdparty/libwebp/src/demux/anim_decode.c
@@ -11,15 +11,15 @@
 //
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
 #include <assert.h>
 #include <string.h>
 
-#include "../utils/utils.h"
-#include "../webp/decode.h"
-#include "../webp/demux.h"
+#include "src/utils/utils.h"
+#include "src/webp/decode.h"
+#include "src/webp/demux.h"
 
 #define NUM_CHANNELS 4
 
diff --git a/src/3rdparty/libwebp/src/demux/demux.c b/src/3rdparty/libwebp/src/demux/demux.c
index 100eab8..79c24a5 100644
--- a/src/3rdparty/libwebp/src/demux/demux.c
+++ b/src/3rdparty/libwebp/src/demux/demux.c
@@ -11,21 +11,21 @@
 //
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
 
-#include "../utils/utils.h"
-#include "../webp/decode.h"     // WebPGetFeatures
-#include "../webp/demux.h"
-#include "../webp/format_constants.h"
+#include "src/utils/utils.h"
+#include "src/webp/decode.h"     // WebPGetFeatures
+#include "src/webp/demux.h"
+#include "src/webp/format_constants.h"
 
 #define DMUX_MAJ_VERSION 0
 #define DMUX_MIN_VERSION 3
-#define DMUX_REV_VERSION 2
+#define DMUX_REV_VERSION 3
 
 typedef struct {
   size_t start_;        // start location of the data
@@ -205,12 +205,14 @@ static void SetFrameInfo(size_t start_offset, size_t size,
   frame->complete_ = complete;
 }
 
-// Store image bearing chunks to 'frame'.
+// Store image bearing chunks to 'frame'. 'min_size' is an optional size
+// requirement, it may be zero.
 static ParseStatus StoreFrame(int frame_num, uint32_t min_size,
                               MemBuffer* const mem, Frame* const frame) {
   int alpha_chunks = 0;
   int image_chunks = 0;
-  int done = (MemDataSize(mem) < min_size);
+  int done = (MemDataSize(mem) < CHUNK_HEADER_SIZE ||
+              MemDataSize(mem) < min_size);
   ParseStatus status = PARSE_OK;
 
   if (done) return PARSE_NEED_MORE_DATA;
@@ -401,9 +403,9 @@ static ParseStatus ParseSingleImage(WebPDemuxer* const dmux) {
   frame = (Frame*)WebPSafeCalloc(1ULL, sizeof(*frame));
   if (frame == NULL) return PARSE_ERROR;
 
-  // For the single image case we allow parsing of a partial frame, but we need
-  // at least CHUNK_HEADER_SIZE for parsing.
-  status = StoreFrame(1, CHUNK_HEADER_SIZE, &dmux->mem_, frame);
+  // For the single image case we allow parsing of a partial frame, so no
+  // minimum size is imposed here.
+  status = StoreFrame(1, 0, &dmux->mem_, frame);
   if (status != PARSE_ERROR) {
     const int has_alpha = !!(dmux->feature_flags_ & ALPHA_FLAG);
     // Clear any alpha when the alpha flag is missing.
diff --git a/src/3rdparty/libwebp/src/dsp/alpha_processing.c b/src/3rdparty/libwebp/src/dsp/alpha_processing.c
index 4b60e09..590e3bc 100644
--- a/src/3rdparty/libwebp/src/dsp/alpha_processing.c
+++ b/src/3rdparty/libwebp/src/dsp/alpha_processing.c
@@ -12,10 +12,13 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <assert.h>
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 // Tables can be faster on some platform but incur some extra binary size (~2k).
-// #define USE_TABLES_FOR_ALPHA_MULT
+#if !defined(USE_TABLES_FOR_ALPHA_MULT)
+#define USE_TABLES_FOR_ALPHA_MULT 0   // ALTERNATE_CODE
+#endif
+
 
 // -----------------------------------------------------------------------------
 
@@ -29,7 +32,7 @@ static uint32_t Mult(uint8_t x, uint32_t mult) {
   return v;
 }
 
-#ifdef USE_TABLES_FOR_ALPHA_MULT
+#if (USE_TABLES_FOR_ALPHA_MULT == 1)
 
 static const uint32_t kMultTables[2][256] = {
   {    // (255u << MFIX) / alpha
@@ -132,9 +135,9 @@ static WEBP_INLINE uint32_t GetScale(uint32_t a, int inverse) {
   return inverse ? (255u << MFIX) / a : a * KINV_255;
 }
 
-#endif    // USE_TABLES_FOR_ALPHA_MULT
+#endif  // USE_TABLES_FOR_ALPHA_MULT
 
-void WebPMultARGBRowC(uint32_t* const ptr, int width, int inverse) {
+void WebPMultARGBRow_C(uint32_t* const ptr, int width, int inverse) {
   int x;
   for (x = 0; x < width; ++x) {
     const uint32_t argb = ptr[x];
@@ -154,8 +157,8 @@ void WebPMultARGBRowC(uint32_t* const ptr, int width, int inverse) {
   }
 }
 
-void WebPMultRowC(uint8_t* const ptr, const uint8_t* const alpha,
-                  int width, int inverse) {
+void WebPMultRow_C(uint8_t* const ptr, const uint8_t* const alpha,
+                   int width, int inverse) {
   int x;
   for (x = 0; x < width; ++x) {
     const uint32_t a = alpha[x];
@@ -217,8 +220,9 @@ void WebPMultRows(uint8_t* ptr, int stride,
 #define PREMULTIPLY(x, m) (((x) * (m) + (1U << 23)) >> 24)
 #endif
 
-static void ApplyAlphaMultiply(uint8_t* rgba, int alpha_first,
-                               int w, int h, int stride) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void ApplyAlphaMultiply_C(uint8_t* rgba, int alpha_first,
+                                 int w, int h, int stride) {
   while (h-- > 0) {
     uint8_t* const rgb = rgba + (alpha_first ? 1 : 0);
     const uint8_t* const alpha = rgba + (alpha_first ? 0 : 3);
@@ -235,6 +239,7 @@ static void ApplyAlphaMultiply(uint8_t* rgba, int alpha_first,
     rgba += stride;
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 #undef MULTIPLIER
 #undef PREMULTIPLY
 
@@ -254,9 +259,9 @@ static WEBP_INLINE uint8_t multiply(uint8_t x, uint32_t m) {
   return (x * m) >> 16;
 }
 
-static WEBP_INLINE void ApplyAlphaMultiply4444(uint8_t* rgba4444,
-                                               int w, int h, int stride,
-                                               int rg_byte_pos /* 0 or 1 */) {
+static WEBP_INLINE void ApplyAlphaMultiply4444_C(uint8_t* rgba4444,
+                                                 int w, int h, int stride,
+                                                 int rg_byte_pos /* 0 or 1 */) {
   while (h-- > 0) {
     int i;
     for (i = 0; i < w; ++i) {
@@ -275,15 +280,16 @@ static WEBP_INLINE void ApplyAlphaMultiply4444(uint8_t* rgba4444,
 }
 #undef MULTIPLIER
 
-static void ApplyAlphaMultiply_16b(uint8_t* rgba4444,
-                                   int w, int h, int stride) {
-#ifdef WEBP_SWAP_16BIT_CSP
-  ApplyAlphaMultiply4444(rgba4444, w, h, stride, 1);
+static void ApplyAlphaMultiply_16b_C(uint8_t* rgba4444,
+                                     int w, int h, int stride) {
+#if (WEBP_SWAP_16BIT_CSP == 1)
+  ApplyAlphaMultiply4444_C(rgba4444, w, h, stride, 1);
 #else
-  ApplyAlphaMultiply4444(rgba4444, w, h, stride, 0);
+  ApplyAlphaMultiply4444_C(rgba4444, w, h, stride, 0);
 #endif
 }
 
+#if !WEBP_NEON_OMIT_C_CODE
 static int DispatchAlpha_C(const uint8_t* alpha, int alpha_stride,
                            int width, int height,
                            uint8_t* dst, int dst_stride) {
@@ -338,6 +344,36 @@ static void ExtractGreen_C(const uint32_t* argb, uint8_t* alpha, int size) {
   int i;
   for (i = 0; i < size; ++i) alpha[i] = argb[i] >> 8;
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
+
+//------------------------------------------------------------------------------
+
+static int HasAlpha8b_C(const uint8_t* src, int length) {
+  while (length-- > 0) if (*src++ != 0xff) return 1;
+  return 0;
+}
+
+static int HasAlpha32b_C(const uint8_t* src, int length) {
+  int x;
+  for (x = 0; length-- > 0; x += 4) if (src[x] != 0xff) return 1;
+  return 0;
+}
+
+//------------------------------------------------------------------------------
+// Simple channel manipulations.
+
+static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) {
+  return (((uint32_t)a << 24) | (r << 16) | (g << 8) | b);
+}
+
+static void PackRGB_C(const uint8_t* r, const uint8_t* g, const uint8_t* b,
+                      int len, int step, uint32_t* out) {
+  int i, offset = 0;
+  for (i = 0; i < len; ++i) {
+    out[i] = MakeARGB32(0xff, r[offset], g[offset], b[offset]);
+    offset += step;
+  }
+}
 
 void (*WebPApplyAlphaMultiply)(uint8_t*, int, int, int, int);
 void (*WebPApplyAlphaMultiply4444)(uint8_t*, int, int, int);
@@ -345,6 +381,11 @@ int (*WebPDispatchAlpha)(const uint8_t*, int, int, int, uint8_t*, int);
 void (*WebPDispatchAlphaToGreen)(const uint8_t*, int, int, int, uint32_t*, int);
 int (*WebPExtractAlpha)(const uint8_t*, int, int, int, uint8_t*, int);
 void (*WebPExtractGreen)(const uint32_t* argb, uint8_t* alpha, int size);
+void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
+                    int len, int step, uint32_t* out);
+
+int (*WebPHasAlpha8b)(const uint8_t* src, int length);
+int (*WebPHasAlpha32b)(const uint8_t* src, int length);
 
 //------------------------------------------------------------------------------
 // Init function
@@ -360,15 +401,21 @@ static volatile VP8CPUInfo alpha_processing_last_cpuinfo_used =
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessing(void) {
   if (alpha_processing_last_cpuinfo_used == VP8GetCPUInfo) return;
 
-  WebPMultARGBRow = WebPMultARGBRowC;
-  WebPMultRow = WebPMultRowC;
-  WebPApplyAlphaMultiply = ApplyAlphaMultiply;
-  WebPApplyAlphaMultiply4444 = ApplyAlphaMultiply_16b;
+  WebPMultARGBRow = WebPMultARGBRow_C;
+  WebPMultRow = WebPMultRow_C;
+  WebPApplyAlphaMultiply4444 = ApplyAlphaMultiply_16b_C;
 
+  WebPPackRGB = PackRGB_C;
+#if !WEBP_NEON_OMIT_C_CODE
+  WebPApplyAlphaMultiply = ApplyAlphaMultiply_C;
   WebPDispatchAlpha = DispatchAlpha_C;
   WebPDispatchAlphaToGreen = DispatchAlphaToGreen_C;
   WebPExtractAlpha = ExtractAlpha_C;
   WebPExtractGreen = ExtractGreen_C;
+#endif
+
+  WebPHasAlpha8b = HasAlpha8b_C;
+  WebPHasAlpha32b = HasAlpha32b_C;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
@@ -382,16 +429,31 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessing(void) {
 #endif
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      WebPInitAlphaProcessingNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS_DSP_R2)
     if (VP8GetCPUInfo(kMIPSdspR2)) {
       WebPInitAlphaProcessingMIPSdspR2();
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    WebPInitAlphaProcessingNEON();
+  }
+#endif
+
+  assert(WebPMultARGBRow != NULL);
+  assert(WebPMultRow != NULL);
+  assert(WebPApplyAlphaMultiply != NULL);
+  assert(WebPApplyAlphaMultiply4444 != NULL);
+  assert(WebPDispatchAlpha != NULL);
+  assert(WebPDispatchAlphaToGreen != NULL);
+  assert(WebPExtractAlpha != NULL);
+  assert(WebPExtractGreen != NULL);
+  assert(WebPPackRGB != NULL);
+  assert(WebPHasAlpha8b != NULL);
+  assert(WebPHasAlpha32b != NULL);
+
   alpha_processing_last_cpuinfo_used = VP8GetCPUInfo;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/alpha_processing_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/alpha_processing_mips_dsp_r2.c
index c631d78..e0dc91b 100644
--- a/src/3rdparty/libwebp/src/dsp/alpha_processing_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/alpha_processing_mips_dsp_r2.c
@@ -12,13 +12,13 @@
 // Author(s): Branimir Vasic (branimir.vasic@imgtec.com)
 //            Djordje Pesut  (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-static int DispatchAlpha(const uint8_t* alpha, int alpha_stride,
-                         int width, int height,
-                         uint8_t* dst, int dst_stride) {
+static int DispatchAlpha_MIPSdspR2(const uint8_t* alpha, int alpha_stride,
+                                   int width, int height,
+                                   uint8_t* dst, int dst_stride) {
   uint32_t alpha_mask = 0xffffffff;
   int i, j, temp0;
 
@@ -79,7 +79,8 @@ static int DispatchAlpha(const uint8_t* alpha, int alpha_stride,
   return (alpha_mask != 0xff);
 }
 
-static void MultARGBRow(uint32_t* const ptr, int width, int inverse) {
+static void MultARGBRow_MIPSdspR2(uint32_t* const ptr, int width,
+                                  int inverse) {
   int x;
   const uint32_t c_00ffffff = 0x00ffffffu;
   const uint32_t c_ff000000 = 0xff000000u;
@@ -124,14 +125,54 @@ static void MultARGBRow(uint32_t* const ptr, int width, int inverse) {
   }
 }
 
+static void PackRGB_MIPSdspR2(const uint8_t* r, const uint8_t* g,
+                              const uint8_t* b, int len, int step,
+                              uint32_t* out) {
+  int temp0, temp1, temp2, offset;
+  const int rest = len & 1;
+  const int a = 0xff;
+  const uint32_t* const loop_end = out + len - rest;
+  __asm__ volatile (
+    "xor          %[offset],   %[offset], %[offset]    \n\t"
+    "beq          %[loop_end], %[out],    0f           \n\t"
+  "2:                                                  \n\t"
+    "lbux         %[temp0],    %[offset](%[r])         \n\t"
+    "lbux         %[temp1],    %[offset](%[g])         \n\t"
+    "lbux         %[temp2],    %[offset](%[b])         \n\t"
+    "ins          %[temp0],    %[a],      16,     16   \n\t"
+    "ins          %[temp2],    %[temp1],  16,     16   \n\t"
+    "addiu        %[out],      %[out],    4            \n\t"
+    "precr.qb.ph  %[temp0],    %[temp0],  %[temp2]     \n\t"
+    "sw           %[temp0],    -4(%[out])              \n\t"
+    "addu         %[offset],   %[offset], %[step]      \n\t"
+    "bne          %[loop_end], %[out],    2b           \n\t"
+  "0:                                                  \n\t"
+    "beq          %[rest],     $zero,     1f           \n\t"
+    "lbux         %[temp0],    %[offset](%[r])         \n\t"
+    "lbux         %[temp1],    %[offset](%[g])         \n\t"
+    "lbux         %[temp2],    %[offset](%[b])         \n\t"
+    "ins          %[temp0],    %[a],      16,     16   \n\t"
+    "ins          %[temp2],    %[temp1],  16,     16   \n\t"
+    "precr.qb.ph  %[temp0],    %[temp0],  %[temp2]     \n\t"
+    "sw           %[temp0],    0(%[out])               \n\t"
+  "1:                                                  \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [offset]"=&r"(offset), [out]"+&r"(out)
+    : [a]"r"(a), [r]"r"(r), [g]"r"(g), [b]"r"(b), [step]"r"(step),
+      [loop_end]"r"(loop_end), [rest]"r"(rest)
+    : "memory"
+  );
+}
+
 //------------------------------------------------------------------------------
 // Entry point
 
 extern void WebPInitAlphaProcessingMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessingMIPSdspR2(void) {
-  WebPDispatchAlpha = DispatchAlpha;
-  WebPMultARGBRow = MultARGBRow;
+  WebPDispatchAlpha = DispatchAlpha_MIPSdspR2;
+  WebPMultARGBRow = MultARGBRow_MIPSdspR2;
+  WebPPackRGB = PackRGB_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/alpha_processing_neon.c b/src/3rdparty/libwebp/src/dsp/alpha_processing_neon.c
index 606a401..9d55421 100644
--- a/src/3rdparty/libwebp/src/dsp/alpha_processing_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/alpha_processing_neon.c
@@ -11,11 +11,11 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_NEON)
 
-#include "./neon.h"
+#include "src/dsp/neon.h"
 
 //------------------------------------------------------------------------------
 
diff --git a/src/3rdparty/libwebp/src/dsp/alpha_processing_sse2.c b/src/3rdparty/libwebp/src/dsp/alpha_processing_sse2.c
index 83dc559..7658700 100644
--- a/src/3rdparty/libwebp/src/dsp/alpha_processing_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/alpha_processing_sse2.c
@@ -11,16 +11,16 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 #include <emmintrin.h>
 
 //------------------------------------------------------------------------------
 
-static int DispatchAlpha(const uint8_t* alpha, int alpha_stride,
-                         int width, int height,
-                         uint8_t* dst, int dst_stride) {
+static int DispatchAlpha_SSE2(const uint8_t* alpha, int alpha_stride,
+                              int width, int height,
+                              uint8_t* dst, int dst_stride) {
   // alpha_and stores an 'and' operation of all the alpha[] values. The final
   // value is not 0xff if any of the alpha[] is not equal to 0xff.
   uint32_t alpha_and = 0xff;
@@ -72,9 +72,9 @@ static int DispatchAlpha(const uint8_t* alpha, int alpha_stride,
   return (alpha_and != 0xff);
 }
 
-static void DispatchAlphaToGreen(const uint8_t* alpha, int alpha_stride,
-                                 int width, int height,
-                                 uint32_t* dst, int dst_stride) {
+static void DispatchAlphaToGreen_SSE2(const uint8_t* alpha, int alpha_stride,
+                                      int width, int height,
+                                      uint32_t* dst, int dst_stride) {
   int i, j;
   const __m128i zero = _mm_setzero_si128();
   const int limit = width & ~15;
@@ -98,9 +98,9 @@ static void DispatchAlphaToGreen(const uint8_t* alpha, int alpha_stride,
   }
 }
 
-static int ExtractAlpha(const uint8_t* argb, int argb_stride,
-                        int width, int height,
-                        uint8_t* alpha, int alpha_stride) {
+static int ExtractAlpha_SSE2(const uint8_t* argb, int argb_stride,
+                             int width, int height,
+                             uint8_t* alpha, int alpha_stride) {
   // alpha_and stores an 'and' operation of all the alpha[] values. The final
   // value is not 0xff if any of the alpha[] is not equal to 0xff.
   uint32_t alpha_and = 0xff;
@@ -210,6 +210,61 @@ static void ApplyAlphaMultiply_SSE2(uint8_t* rgba, int alpha_first,
 #undef MULTIPLIER
 #undef PREMULTIPLY
 
+//------------------------------------------------------------------------------
+// Alpha detection
+
+static int HasAlpha8b_SSE2(const uint8_t* src, int length) {
+  const __m128i all_0xff = _mm_set1_epi8(0xff);
+  int i = 0;
+  for (; i + 16 <= length; i += 16) {
+    const __m128i v = _mm_loadu_si128((const __m128i*)(src + i));
+    const __m128i bits = _mm_cmpeq_epi8(v, all_0xff);
+    const int mask = _mm_movemask_epi8(bits);
+    if (mask != 0xffff) return 1;
+  }
+  for (; i < length; ++i) if (src[i] != 0xff) return 1;
+  return 0;
+}
+
+static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
+  const __m128i alpha_mask = _mm_set1_epi32(0xff);
+  const __m128i all_0xff = _mm_set1_epi8(0xff);
+  int i = 0;
+  // We don't know if we can access the last 3 bytes after the last alpha
+  // value 'src[4 * length - 4]' (because we don't know if alpha is the first
+  // or the last byte of the quadruplet). Hence the '-3' protection below.
+  length = length * 4 - 3;   // size in bytes
+  for (; i + 64 <= length; i += 64) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)(src + i +  0));
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)(src + i + 16));
+    const __m128i a2 = _mm_loadu_si128((const __m128i*)(src + i + 32));
+    const __m128i a3 = _mm_loadu_si128((const __m128i*)(src + i + 48));
+    const __m128i b0 = _mm_and_si128(a0, alpha_mask);
+    const __m128i b1 = _mm_and_si128(a1, alpha_mask);
+    const __m128i b2 = _mm_and_si128(a2, alpha_mask);
+    const __m128i b3 = _mm_and_si128(a3, alpha_mask);
+    const __m128i c0 = _mm_packs_epi32(b0, b1);
+    const __m128i c1 = _mm_packs_epi32(b2, b3);
+    const __m128i d  = _mm_packus_epi16(c0, c1);
+    const __m128i bits = _mm_cmpeq_epi8(d, all_0xff);
+    const int mask = _mm_movemask_epi8(bits);
+    if (mask != 0xffff) return 1;
+  }
+  for (; i + 32 <= length; i += 32) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)(src + i +  0));
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)(src + i + 16));
+    const __m128i b0 = _mm_and_si128(a0, alpha_mask);
+    const __m128i b1 = _mm_and_si128(a1, alpha_mask);
+    const __m128i c  = _mm_packs_epi32(b0, b1);
+    const __m128i d  = _mm_packus_epi16(c, c);
+    const __m128i bits = _mm_cmpeq_epi8(d, all_0xff);
+    const int mask = _mm_movemask_epi8(bits);
+    if (mask != 0xffff) return 1;
+  }
+  for (; i <= length; i += 4) if (src[i] != 0xff) return 1;
+  return 0;
+}
+
 // -----------------------------------------------------------------------------
 // Apply alpha value to rows
 
@@ -238,7 +293,7 @@ static void MultARGBRow_SSE2(uint32_t* const ptr, int width, int inverse) {
     }
   }
   width -= x;
-  if (width > 0) WebPMultARGBRowC(ptr + x, width, inverse);
+  if (width > 0) WebPMultARGBRow_C(ptr + x, width, inverse);
 }
 
 static void MultRow_SSE2(uint8_t* const ptr, const uint8_t* const alpha,
@@ -261,7 +316,7 @@ static void MultRow_SSE2(uint8_t* const ptr, const uint8_t* const alpha,
     }
   }
   width -= x;
-  if (width > 0) WebPMultRowC(ptr + x, alpha + x, width, inverse);
+  if (width > 0) WebPMultRow_C(ptr + x, alpha + x, width, inverse);
 }
 
 //------------------------------------------------------------------------------
@@ -273,9 +328,12 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessingSSE2(void) {
   WebPMultARGBRow = MultARGBRow_SSE2;
   WebPMultRow = MultRow_SSE2;
   WebPApplyAlphaMultiply = ApplyAlphaMultiply_SSE2;
-  WebPDispatchAlpha = DispatchAlpha;
-  WebPDispatchAlphaToGreen = DispatchAlphaToGreen;
-  WebPExtractAlpha = ExtractAlpha;
+  WebPDispatchAlpha = DispatchAlpha_SSE2;
+  WebPDispatchAlphaToGreen = DispatchAlphaToGreen_SSE2;
+  WebPExtractAlpha = ExtractAlpha_SSE2;
+
+  WebPHasAlpha8b = HasAlpha8b_SSE2;
+  WebPHasAlpha32b = HasAlpha32b_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/alpha_processing_sse41.c b/src/3rdparty/libwebp/src/dsp/alpha_processing_sse41.c
index 986fde9..56040f9 100644
--- a/src/3rdparty/libwebp/src/dsp/alpha_processing_sse41.c
+++ b/src/3rdparty/libwebp/src/dsp/alpha_processing_sse41.c
@@ -11,7 +11,7 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE41)
 
@@ -19,9 +19,9 @@
 
 //------------------------------------------------------------------------------
 
-static int ExtractAlpha(const uint8_t* argb, int argb_stride,
-                        int width, int height,
-                        uint8_t* alpha, int alpha_stride) {
+static int ExtractAlpha_SSE41(const uint8_t* argb, int argb_stride,
+                              int width, int height,
+                              uint8_t* alpha, int alpha_stride) {
   // alpha_and stores an 'and' operation of all the alpha[] values. The final
   // value is not 0xff if any of the alpha[] is not equal to 0xff.
   uint32_t alpha_and = 0xff;
@@ -82,7 +82,7 @@ static int ExtractAlpha(const uint8_t* argb, int argb_stride,
 extern void WebPInitAlphaProcessingSSE41(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessingSSE41(void) {
-  WebPExtractAlpha = ExtractAlpha;
+  WebPExtractAlpha = ExtractAlpha_SSE41;
 }
 
 #else  // !WEBP_USE_SSE41
diff --git a/src/3rdparty/libwebp/src/dsp/argb.c b/src/3rdparty/libwebp/src/dsp/argb.c
deleted file mode 100644
index cc1f9a9..0000000
--- a/src/3rdparty/libwebp/src/dsp/argb.c
+++ /dev/null
@@ -1,68 +0,0 @@
-// Copyright 2014 Google Inc. All Rights Reserved.
-//
-// Use of this source code is governed by a BSD-style license
-// that can be found in the COPYING file in the root of the source
-// tree. An additional intellectual property rights grant can be found
-// in the file PATENTS. All contributing project authors may
-// be found in the AUTHORS file in the root of the source tree.
-// -----------------------------------------------------------------------------
-//
-//   ARGB making functions.
-//
-// Author: Djordje Pesut (djordje.pesut@imgtec.com)
-
-#include "./dsp.h"
-
-static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) {
-  return (((uint32_t)a << 24) | (r << 16) | (g << 8) | b);
-}
-
-static void PackARGB(const uint8_t* a, const uint8_t* r, const uint8_t* g,
-                     const uint8_t* b, int len, uint32_t* out) {
-  int i;
-  for (i = 0; i < len; ++i) {
-    out[i] = MakeARGB32(a[4 * i], r[4 * i], g[4 * i], b[4 * i]);
-  }
-}
-
-static void PackRGB(const uint8_t* r, const uint8_t* g, const uint8_t* b,
-                    int len, int step, uint32_t* out) {
-  int i, offset = 0;
-  for (i = 0; i < len; ++i) {
-    out[i] = MakeARGB32(0xff, r[offset], g[offset], b[offset]);
-    offset += step;
-  }
-}
-
-void (*VP8PackARGB)(const uint8_t*, const uint8_t*, const uint8_t*,
-                    const uint8_t*, int, uint32_t*);
-void (*VP8PackRGB)(const uint8_t*, const uint8_t*, const uint8_t*,
-                   int, int, uint32_t*);
-
-extern void VP8EncDspARGBInitMIPSdspR2(void);
-extern void VP8EncDspARGBInitSSE2(void);
-
-static volatile VP8CPUInfo argb_last_cpuinfo_used =
-    (VP8CPUInfo)&argb_last_cpuinfo_used;
-
-WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInit(void) {
-  if (argb_last_cpuinfo_used == VP8GetCPUInfo) return;
-
-  VP8PackARGB = PackARGB;
-  VP8PackRGB = PackRGB;
-
-  // If defined, use CPUInfo() to overwrite some pointers with faster versions.
-  if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
-    if (VP8GetCPUInfo(kSSE2)) {
-      VP8EncDspARGBInitSSE2();
-    }
-#endif
-#if defined(WEBP_USE_MIPS_DSP_R2)
-    if (VP8GetCPUInfo(kMIPSdspR2)) {
-      VP8EncDspARGBInitMIPSdspR2();
-    }
-#endif
-  }
-  argb_last_cpuinfo_used = VP8GetCPUInfo;
-}
diff --git a/src/3rdparty/libwebp/src/dsp/argb_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/argb_mips_dsp_r2.c
deleted file mode 100644
index af65acb..0000000
--- a/src/3rdparty/libwebp/src/dsp/argb_mips_dsp_r2.c
+++ /dev/null
@@ -1,110 +0,0 @@
-// Copyright 2014 Google Inc. All Rights Reserved.
-//
-// Use of this source code is governed by a BSD-style license
-// that can be found in the COPYING file in the root of the source
-// tree. An additional intellectual property rights grant can be found
-// in the file PATENTS. All contributing project authors may
-// be found in the AUTHORS file in the root of the source tree.
-// -----------------------------------------------------------------------------
-//
-//   ARGB making functions (mips version).
-//
-// Author: Djordje Pesut (djordje.pesut@imgtec.com)
-
-#include "./dsp.h"
-
-#if defined(WEBP_USE_MIPS_DSP_R2)
-
-static void PackARGB(const uint8_t* a, const uint8_t* r, const uint8_t* g,
-                     const uint8_t* b, int len, uint32_t* out) {
-  int temp0, temp1, temp2, temp3, offset;
-  const int rest = len & 1;
-  const uint32_t* const loop_end = out + len - rest;
-  const int step = 4;
-  __asm__ volatile (
-    "xor          %[offset],   %[offset], %[offset]    \n\t"
-    "beq          %[loop_end], %[out],    0f           \n\t"
-  "2:                                                  \n\t"
-    "lbux         %[temp0],    %[offset](%[a])         \n\t"
-    "lbux         %[temp1],    %[offset](%[r])         \n\t"
-    "lbux         %[temp2],    %[offset](%[g])         \n\t"
-    "lbux         %[temp3],    %[offset](%[b])         \n\t"
-    "ins          %[temp1],    %[temp0],  16,     16   \n\t"
-    "ins          %[temp3],    %[temp2],  16,     16   \n\t"
-    "addiu        %[out],      %[out],    4            \n\t"
-    "precr.qb.ph  %[temp0],    %[temp1],  %[temp3]     \n\t"
-    "sw           %[temp0],    -4(%[out])              \n\t"
-    "addu         %[offset],   %[offset], %[step]      \n\t"
-    "bne          %[loop_end], %[out],    2b           \n\t"
-  "0:                                                  \n\t"
-    "beq          %[rest],     $zero,     1f           \n\t"
-    "lbux         %[temp0],    %[offset](%[a])         \n\t"
-    "lbux         %[temp1],    %[offset](%[r])         \n\t"
-    "lbux         %[temp2],    %[offset](%[g])         \n\t"
-    "lbux         %[temp3],    %[offset](%[b])         \n\t"
-    "ins          %[temp1],    %[temp0],  16,     16   \n\t"
-    "ins          %[temp3],    %[temp2],  16,     16   \n\t"
-    "precr.qb.ph  %[temp0],    %[temp1],  %[temp3]     \n\t"
-    "sw           %[temp0],    0(%[out])               \n\t"
-  "1:                                                  \n\t"
-    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
-      [temp3]"=&r"(temp3), [offset]"=&r"(offset), [out]"+&r"(out)
-    : [a]"r"(a), [r]"r"(r), [g]"r"(g), [b]"r"(b), [step]"r"(step),
-      [loop_end]"r"(loop_end), [rest]"r"(rest)
-    : "memory"
-  );
-}
-
-static void PackRGB(const uint8_t* r, const uint8_t* g, const uint8_t* b,
-                    int len, int step, uint32_t* out) {
-  int temp0, temp1, temp2, offset;
-  const int rest = len & 1;
-  const int a = 0xff;
-  const uint32_t* const loop_end = out + len - rest;
-  __asm__ volatile (
-    "xor          %[offset],   %[offset], %[offset]    \n\t"
-    "beq          %[loop_end], %[out],    0f           \n\t"
-  "2:                                                  \n\t"
-    "lbux         %[temp0],    %[offset](%[r])         \n\t"
-    "lbux         %[temp1],    %[offset](%[g])         \n\t"
-    "lbux         %[temp2],    %[offset](%[b])         \n\t"
-    "ins          %[temp0],    %[a],      16,     16   \n\t"
-    "ins          %[temp2],    %[temp1],  16,     16   \n\t"
-    "addiu        %[out],      %[out],    4            \n\t"
-    "precr.qb.ph  %[temp0],    %[temp0],  %[temp2]     \n\t"
-    "sw           %[temp0],    -4(%[out])              \n\t"
-    "addu         %[offset],   %[offset], %[step]      \n\t"
-    "bne          %[loop_end], %[out],    2b           \n\t"
-  "0:                                                  \n\t"
-    "beq          %[rest],     $zero,     1f           \n\t"
-    "lbux         %[temp0],    %[offset](%[r])         \n\t"
-    "lbux         %[temp1],    %[offset](%[g])         \n\t"
-    "lbux         %[temp2],    %[offset](%[b])         \n\t"
-    "ins          %[temp0],    %[a],      16,     16   \n\t"
-    "ins          %[temp2],    %[temp1],  16,     16   \n\t"
-    "precr.qb.ph  %[temp0],    %[temp0],  %[temp2]     \n\t"
-    "sw           %[temp0],    0(%[out])               \n\t"
-  "1:                                                  \n\t"
-    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
-      [offset]"=&r"(offset), [out]"+&r"(out)
-    : [a]"r"(a), [r]"r"(r), [g]"r"(g), [b]"r"(b), [step]"r"(step),
-      [loop_end]"r"(loop_end), [rest]"r"(rest)
-    : "memory"
-  );
-}
-
-//------------------------------------------------------------------------------
-// Entry point
-
-extern void VP8EncDspARGBInitMIPSdspR2(void);
-
-WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInitMIPSdspR2(void) {
-  VP8PackARGB = PackARGB;
-  VP8PackRGB = PackRGB;
-}
-
-#else  // !WEBP_USE_MIPS_DSP_R2
-
-WEBP_DSP_INIT_STUB(VP8EncDspARGBInitMIPSdspR2)
-
-#endif  // WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/argb_sse2.c b/src/3rdparty/libwebp/src/dsp/argb_sse2.c
deleted file mode 100644
index afcb195..0000000
--- a/src/3rdparty/libwebp/src/dsp/argb_sse2.c
+++ /dev/null
@@ -1,67 +0,0 @@
-// Copyright 2014 Google Inc. All Rights Reserved.
-//
-// Use of this source code is governed by a BSD-style license
-// that can be found in the COPYING file in the root of the source
-// tree. An additional intellectual property rights grant can be found
-// in the file PATENTS. All contributing project authors may
-// be found in the AUTHORS file in the root of the source tree.
-// -----------------------------------------------------------------------------
-//
-//   ARGB making functions (SSE2 version).
-//
-// Author: Skal (pascal.massimino@gmail.com)
-
-#include "./dsp.h"
-
-#if defined(WEBP_USE_SSE2)
-
-#include <assert.h>
-#include <emmintrin.h>
-#include <string.h>
-
-static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) {
-  return (((uint32_t)a << 24) | (r << 16) | (g << 8) | b);
-}
-
-static void PackARGB(const uint8_t* a, const uint8_t* r, const uint8_t* g,
-                     const uint8_t* b, int len, uint32_t* out) {
-  if (g == r + 1) {  // RGBA input order. Need to swap R and B.
-    int i = 0;
-    const int len_max = len & ~3;  // max length processed in main loop
-    const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ffu);
-    assert(b == r + 2);
-    assert(a == r + 3);
-    for (; i < len_max; i += 4) {
-      const __m128i A = _mm_loadu_si128((const __m128i*)(r + 4 * i));
-      const __m128i B = _mm_and_si128(A, red_blue_mask);     // R 0 B 0
-      const __m128i C = _mm_andnot_si128(red_blue_mask, A);  // 0 G 0 A
-      const __m128i D = _mm_shufflelo_epi16(B, _MM_SHUFFLE(2, 3, 0, 1));
-      const __m128i E = _mm_shufflehi_epi16(D, _MM_SHUFFLE(2, 3, 0, 1));
-      const __m128i F = _mm_or_si128(E, C);
-      _mm_storeu_si128((__m128i*)(out + i), F);
-    }
-    for (; i < len; ++i) {
-      out[i] = MakeARGB32(a[4 * i], r[4 * i], g[4 * i], b[4 * i]);
-    }
-  } else {
-    assert(g == b + 1);
-    assert(r == b + 2);
-    assert(a == b + 3);
-    memcpy(out, b, len * 4);
-  }
-}
-
-//------------------------------------------------------------------------------
-// Entry point
-
-extern void VP8EncDspARGBInitSSE2(void);
-
-WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInitSSE2(void) {
-  VP8PackARGB = PackARGB;
-}
-
-#else  // !WEBP_USE_SSE2
-
-WEBP_DSP_INIT_STUB(VP8EncDspARGBInitSSE2)
-
-#endif  // WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/cost.c b/src/3rdparty/libwebp/src/dsp/cost.c
index 58ddea7..a732389 100644
--- a/src/3rdparty/libwebp/src/dsp/cost.c
+++ b/src/3rdparty/libwebp/src/dsp/cost.c
@@ -9,8 +9,8 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
-#include "../enc/cost_enc.h"
+#include "src/dsp/dsp.h"
+#include "src/enc/cost_enc.h"
 
 //------------------------------------------------------------------------------
 // Boolean-cost cost table
@@ -319,7 +319,7 @@ const uint8_t VP8EncBands[16 + 1] = {
 //------------------------------------------------------------------------------
 // Mode costs
 
-static int GetResidualCost(int ctx0, const VP8Residual* const res) {
+static int GetResidualCost_C(int ctx0, const VP8Residual* const res) {
   int n = res->first;
   // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1
   const int p0 = res->prob[n][ctx0][0];
@@ -354,8 +354,8 @@ static int GetResidualCost(int ctx0, const VP8Residual* const res) {
   return cost;
 }
 
-static void SetResidualCoeffs(const int16_t* const coeffs,
-                              VP8Residual* const res) {
+static void SetResidualCoeffs_C(const int16_t* const coeffs,
+                                VP8Residual* const res) {
   int n;
   res->last = -1;
   assert(res->first == 0 || coeffs[0] == 0);
@@ -384,8 +384,8 @@ static volatile VP8CPUInfo cost_last_cpuinfo_used =
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInit(void) {
   if (cost_last_cpuinfo_used == VP8GetCPUInfo) return;
 
-  VP8GetResidualCost = GetResidualCost;
-  VP8SetResidualCoeffs = SetResidualCoeffs;
+  VP8GetResidualCost = GetResidualCost_C;
+  VP8SetResidualCoeffs = SetResidualCoeffs_C;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
diff --git a/src/3rdparty/libwebp/src/dsp/cost_mips32.c b/src/3rdparty/libwebp/src/dsp/cost_mips32.c
index 3102da8..0500f88 100644
--- a/src/3rdparty/libwebp/src/dsp/cost_mips32.c
+++ b/src/3rdparty/libwebp/src/dsp/cost_mips32.c
@@ -9,13 +9,13 @@
 //
 // Author: Djordje Pesut (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS32)
 
-#include "../enc/cost_enc.h"
+#include "src/enc/cost_enc.h"
 
-static int GetResidualCost(int ctx0, const VP8Residual* const res) {
+static int GetResidualCost_MIPS32(int ctx0, const VP8Residual* const res) {
   int temp0, temp1;
   int v_reg, ctx_reg;
   int n = res->first;
@@ -96,8 +96,8 @@ static int GetResidualCost(int ctx0, const VP8Residual* const res) {
   return cost;
 }
 
-static void SetResidualCoeffs(const int16_t* const coeffs,
-                              VP8Residual* const res) {
+static void SetResidualCoeffs_MIPS32(const int16_t* const coeffs,
+                                     VP8Residual* const res) {
   const int16_t* p_coeffs = (int16_t*)coeffs;
   int temp0, temp1, temp2, n, n1;
   assert(res->first == 0 || coeffs[0] == 0);
@@ -143,8 +143,8 @@ static void SetResidualCoeffs(const int16_t* const coeffs,
 extern void VP8EncDspCostInitMIPS32(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInitMIPS32(void) {
-  VP8GetResidualCost = GetResidualCost;
-  VP8SetResidualCoeffs = SetResidualCoeffs;
+  VP8GetResidualCost = GetResidualCost_MIPS32;
+  VP8SetResidualCoeffs = SetResidualCoeffs_MIPS32;
 }
 
 #else  // !WEBP_USE_MIPS32
diff --git a/src/3rdparty/libwebp/src/dsp/cost_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/cost_mips_dsp_r2.c
index 6ec8aeb..51248de 100644
--- a/src/3rdparty/libwebp/src/dsp/cost_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/cost_mips_dsp_r2.c
@@ -9,13 +9,13 @@
 //
 // Author: Djordje Pesut (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-#include "../enc/cost_enc.h"
+#include "src/enc/cost_enc.h"
 
-static int GetResidualCost(int ctx0, const VP8Residual* const res) {
+static int GetResidualCost_MIPSdspR2(int ctx0, const VP8Residual* const res) {
   int temp0, temp1;
   int v_reg, ctx_reg;
   int n = res->first;
@@ -97,7 +97,7 @@ static int GetResidualCost(int ctx0, const VP8Residual* const res) {
 extern void VP8EncDspCostInitMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInitMIPSdspR2(void) {
-  VP8GetResidualCost = GetResidualCost;
+  VP8GetResidualCost = GetResidualCost_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/cost_sse2.c b/src/3rdparty/libwebp/src/dsp/cost_sse2.c
index 421d51f..487a079 100644
--- a/src/3rdparty/libwebp/src/dsp/cost_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/cost_sse2.c
@@ -11,19 +11,19 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 #include <emmintrin.h>
 
-#include "../enc/cost_enc.h"
-#include "../enc/vp8i_enc.h"
-#include "../utils/utils.h"
+#include "src/enc/cost_enc.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 
-static void SetResidualCoeffsSSE2(const int16_t* const coeffs,
-                                  VP8Residual* const res) {
+static void SetResidualCoeffs_SSE2(const int16_t* const coeffs,
+                                   VP8Residual* const res) {
   const __m128i c0 = _mm_loadu_si128((const __m128i*)(coeffs + 0));
   const __m128i c1 = _mm_loadu_si128((const __m128i*)(coeffs + 8));
   // Use SSE2 to compare 16 values with a single instruction.
@@ -42,7 +42,7 @@ static void SetResidualCoeffsSSE2(const int16_t* const coeffs,
   res->coeffs = coeffs;
 }
 
-static int GetResidualCostSSE2(int ctx0, const VP8Residual* const res) {
+static int GetResidualCost_SSE2(int ctx0, const VP8Residual* const res) {
   uint8_t levels[16], ctxs[16];
   uint16_t abs_levels[16];
   int n = res->first;
@@ -108,8 +108,8 @@ static int GetResidualCostSSE2(int ctx0, const VP8Residual* const res) {
 extern void VP8EncDspCostInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInitSSE2(void) {
-  VP8SetResidualCoeffs = SetResidualCoeffsSSE2;
-  VP8GetResidualCost = GetResidualCostSSE2;
+  VP8SetResidualCoeffs = SetResidualCoeffs_SSE2;
+  VP8GetResidualCost = GetResidualCost_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/cpu.c b/src/3rdparty/libwebp/src/dsp/cpu.c
index b39184e..6f42cbb 100644
--- a/src/3rdparty/libwebp/src/dsp/cpu.c
+++ b/src/3rdparty/libwebp/src/dsp/cpu.c
@@ -11,7 +11,7 @@
 //
 // Author: Christian Duvivier (cduvivier@google.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_HAVE_NEON_RTCD)
 #include <stdio.h>
@@ -143,7 +143,7 @@ static int x86CPUInfo(CPUFeature feature) {
     return !!(cpu_info[2] & (1 << 0));
   }
   if (feature == kSlowSSSE3) {
-    if (is_intel && (cpu_info[2] & (1 << 0))) {   // SSSE3?
+    if (is_intel && (cpu_info[2] & (1 << 9))) {   // SSSE3?
       return CheckSlowModel(cpu_info[0]);
     }
     return 0;
diff --git a/src/3rdparty/libwebp/src/dsp/dec.c b/src/3rdparty/libwebp/src/dsp/dec.c
index 007e985..7e82407 100644
--- a/src/3rdparty/libwebp/src/dsp/dec.c
+++ b/src/3rdparty/libwebp/src/dsp/dec.c
@@ -11,9 +11,11 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
-#include "../dec/vp8i_dec.h"
-#include "../utils/utils.h"
+#include <assert.h>
+
+#include "src/dsp/dsp.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 
@@ -25,7 +27,7 @@ static WEBP_INLINE uint8_t clip_8b(int v) {
 // Transforms (Paragraph 14.4)
 
 #define STORE(x, y, v) \
-  dst[x + y * BPS] = clip_8b(dst[x + y * BPS] + ((v) >> 3))
+  dst[(x) + (y) * BPS] = clip_8b(dst[(x) + (y) * BPS] + ((v) >> 3))
 
 #define STORE2(y, dc, d, c) do {    \
   const int DC = (dc);              \
@@ -38,7 +40,8 @@ static WEBP_INLINE uint8_t clip_8b(int v) {
 #define MUL1(a) ((((a) * 20091) >> 16) + (a))
 #define MUL2(a) (((a) * 35468) >> 16)
 
-static void TransformOne(const int16_t* in, uint8_t* dst) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void TransformOne_C(const int16_t* in, uint8_t* dst) {
   int C[4 * 4], *tmp;
   int i;
   tmp = C;
@@ -78,7 +81,7 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
 }
 
 // Simplified transform when only in[0], in[1] and in[4] are non-zero
-static void TransformAC3(const int16_t* in, uint8_t* dst) {
+static void TransformAC3_C(const int16_t* in, uint8_t* dst) {
   const int a = in[0] + 4;
   const int c4 = MUL2(in[4]);
   const int d4 = MUL1(in[4]);
@@ -93,19 +96,21 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
 #undef MUL2
 #undef STORE2
 
-static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
-  TransformOne(in, dst);
+static void TransformTwo_C(const int16_t* in, uint8_t* dst, int do_two) {
+  TransformOne_C(in, dst);
   if (do_two) {
-    TransformOne(in + 16, dst + 4);
+    TransformOne_C(in + 16, dst + 4);
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void TransformUV(const int16_t* in, uint8_t* dst) {
+static void TransformUV_C(const int16_t* in, uint8_t* dst) {
   VP8Transform(in + 0 * 16, dst, 1);
   VP8Transform(in + 2 * 16, dst + 4 * BPS, 1);
 }
 
-static void TransformDC(const int16_t* in, uint8_t* dst) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void TransformDC_C(const int16_t* in, uint8_t* dst) {
   const int DC = in[0] + 4;
   int i, j;
   for (j = 0; j < 4; ++j) {
@@ -114,8 +119,9 @@ static void TransformDC(const int16_t* in, uint8_t* dst) {
     }
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void TransformDCUV(const int16_t* in, uint8_t* dst) {
+static void TransformDCUV_C(const int16_t* in, uint8_t* dst) {
   if (in[0 * 16]) VP8TransformDC(in + 0 * 16, dst);
   if (in[1 * 16]) VP8TransformDC(in + 1 * 16, dst + 4);
   if (in[2 * 16]) VP8TransformDC(in + 2 * 16, dst + 4 * BPS);
@@ -127,7 +133,8 @@ static void TransformDCUV(const int16_t* in, uint8_t* dst) {
 //------------------------------------------------------------------------------
 // Paragraph 14.3
 
-static void TransformWHT(const int16_t* in, int16_t* out) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void TransformWHT_C(const int16_t* in, int16_t* out) {
   int tmp[16];
   int i;
   for (i = 0; i < 4; ++i) {
@@ -153,6 +160,7 @@ static void TransformWHT(const int16_t* in, int16_t* out) {
     out += 64;
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 void (*VP8TransformWHT)(const int16_t* in, int16_t* out);
 
@@ -161,6 +169,7 @@ void (*VP8TransformWHT)(const int16_t* in, int16_t* out);
 
 #define DST(x, y) dst[(x) + (y) * BPS]
 
+#if !WEBP_NEON_OMIT_C_CODE
 static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
   const uint8_t* top = dst - BPS;
   const uint8_t* const clip0 = VP8kclip1 - top[-1];
@@ -174,21 +183,21 @@ static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
     dst += BPS;
   }
 }
-static void TM4(uint8_t* dst)   { TrueMotion(dst, 4); }
-static void TM8uv(uint8_t* dst) { TrueMotion(dst, 8); }
-static void TM16(uint8_t* dst)  { TrueMotion(dst, 16); }
+static void TM4_C(uint8_t* dst)   { TrueMotion(dst, 4); }
+static void TM8uv_C(uint8_t* dst) { TrueMotion(dst, 8); }
+static void TM16_C(uint8_t* dst)  { TrueMotion(dst, 16); }
 
 //------------------------------------------------------------------------------
 // 16x16
 
-static void VE16(uint8_t* dst) {     // vertical
+static void VE16_C(uint8_t* dst) {     // vertical
   int j;
   for (j = 0; j < 16; ++j) {
     memcpy(dst + j * BPS, dst - BPS, 16);
   }
 }
 
-static void HE16(uint8_t* dst) {     // horizontal
+static void HE16_C(uint8_t* dst) {     // horizontal
   int j;
   for (j = 16; j > 0; --j) {
     memset(dst, dst[-1], 16);
@@ -203,7 +212,7 @@ static WEBP_INLINE void Put16(int v, uint8_t* dst) {
   }
 }
 
-static void DC16(uint8_t* dst) {    // DC
+static void DC16_C(uint8_t* dst) {    // DC
   int DC = 16;
   int j;
   for (j = 0; j < 16; ++j) {
@@ -212,7 +221,7 @@ static void DC16(uint8_t* dst) {    // DC
   Put16(DC >> 5, dst);
 }
 
-static void DC16NoTop(uint8_t* dst) {   // DC with top samples not available
+static void DC16NoTop_C(uint8_t* dst) {   // DC with top samples not available
   int DC = 8;
   int j;
   for (j = 0; j < 16; ++j) {
@@ -221,7 +230,7 @@ static void DC16NoTop(uint8_t* dst) {   // DC with top samples not available
   Put16(DC >> 4, dst);
 }
 
-static void DC16NoLeft(uint8_t* dst) {  // DC with left samples not available
+static void DC16NoLeft_C(uint8_t* dst) {  // DC with left samples not available
   int DC = 8;
   int i;
   for (i = 0; i < 16; ++i) {
@@ -230,9 +239,10 @@ static void DC16NoLeft(uint8_t* dst) {  // DC with left samples not available
   Put16(DC >> 4, dst);
 }
 
-static void DC16NoTopLeft(uint8_t* dst) {  // DC with no top and left samples
+static void DC16NoTopLeft_C(uint8_t* dst) {  // DC with no top and left samples
   Put16(0x80, dst);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES];
 
@@ -242,7 +252,8 @@ VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES];
 #define AVG3(a, b, c) ((uint8_t)(((a) + 2 * (b) + (c) + 2) >> 2))
 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
 
-static void VE4(uint8_t* dst) {    // vertical
+#if !WEBP_NEON_OMIT_C_CODE
+static void VE4_C(uint8_t* dst) {    // vertical
   const uint8_t* top = dst - BPS;
   const uint8_t vals[4] = {
     AVG3(top[-1], top[0], top[1]),
@@ -255,8 +266,9 @@ static void VE4(uint8_t* dst) {    // vertical
     memcpy(dst + i * BPS, vals, sizeof(vals));
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void HE4(uint8_t* dst) {    // horizontal
+static void HE4_C(uint8_t* dst) {    // horizontal
   const int A = dst[-1 - BPS];
   const int B = dst[-1];
   const int C = dst[-1 + BPS];
@@ -268,7 +280,8 @@ static void HE4(uint8_t* dst) {    // horizontal
   WebPUint32ToMem(dst + 3 * BPS, 0x01010101U * AVG3(D, E, E));
 }
 
-static void DC4(uint8_t* dst) {   // DC
+#if !WEBP_NEON_OMIT_C_CODE
+static void DC4_C(uint8_t* dst) {   // DC
   uint32_t dc = 4;
   int i;
   for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS];
@@ -276,7 +289,7 @@ static void DC4(uint8_t* dst) {   // DC
   for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4);
 }
 
-static void RD4(uint8_t* dst) {   // Down-right
+static void RD4_C(uint8_t* dst) {   // Down-right
   const int I = dst[-1 + 0 * BPS];
   const int J = dst[-1 + 1 * BPS];
   const int K = dst[-1 + 2 * BPS];
@@ -295,7 +308,7 @@ static void RD4(uint8_t* dst) {   // Down-right
                                       DST(3, 0) = AVG3(D, C, B);
 }
 
-static void LD4(uint8_t* dst) {   // Down-Left
+static void LD4_C(uint8_t* dst) {   // Down-Left
   const int A = dst[0 - BPS];
   const int B = dst[1 - BPS];
   const int C = dst[2 - BPS];
@@ -312,8 +325,9 @@ static void LD4(uint8_t* dst) {   // Down-Left
                           DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
                                       DST(3, 3) = AVG3(G, H, H);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void VR4(uint8_t* dst) {   // Vertical-Right
+static void VR4_C(uint8_t* dst) {   // Vertical-Right
   const int I = dst[-1 + 0 * BPS];
   const int J = dst[-1 + 1 * BPS];
   const int K = dst[-1 + 2 * BPS];
@@ -335,7 +349,7 @@ static void VR4(uint8_t* dst) {   // Vertical-Right
   DST(3, 1) =             AVG3(B, C, D);
 }
 
-static void VL4(uint8_t* dst) {   // Vertical-Left
+static void VL4_C(uint8_t* dst) {   // Vertical-Left
   const int A = dst[0 - BPS];
   const int B = dst[1 - BPS];
   const int C = dst[2 - BPS];
@@ -357,7 +371,7 @@ static void VL4(uint8_t* dst) {   // Vertical-Left
               DST(3, 3) = AVG3(F, G, H);
 }
 
-static void HU4(uint8_t* dst) {   // Horizontal-Up
+static void HU4_C(uint8_t* dst) {   // Horizontal-Up
   const int I = dst[-1 + 0 * BPS];
   const int J = dst[-1 + 1 * BPS];
   const int K = dst[-1 + 2 * BPS];
@@ -372,7 +386,7 @@ static void HU4(uint8_t* dst) {   // Horizontal-Up
     DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
 }
 
-static void HD4(uint8_t* dst) {  // Horizontal-Down
+static void HD4_C(uint8_t* dst) {  // Horizontal-Down
   const int I = dst[-1 + 0 * BPS];
   const int J = dst[-1 + 1 * BPS];
   const int K = dst[-1 + 2 * BPS];
@@ -404,14 +418,15 @@ VP8PredFunc VP8PredLuma4[NUM_BMODES];
 //------------------------------------------------------------------------------
 // Chroma
 
-static void VE8uv(uint8_t* dst) {    // vertical
+#if !WEBP_NEON_OMIT_C_CODE
+static void VE8uv_C(uint8_t* dst) {    // vertical
   int j;
   for (j = 0; j < 8; ++j) {
     memcpy(dst + j * BPS, dst - BPS, 8);
   }
 }
 
-static void HE8uv(uint8_t* dst) {    // horizontal
+static void HE8uv_C(uint8_t* dst) {    // horizontal
   int j;
   for (j = 0; j < 8; ++j) {
     memset(dst, dst[-1], 8);
@@ -427,7 +442,7 @@ static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
   }
 }
 
-static void DC8uv(uint8_t* dst) {     // DC
+static void DC8uv_C(uint8_t* dst) {     // DC
   int dc0 = 8;
   int i;
   for (i = 0; i < 8; ++i) {
@@ -436,7 +451,7 @@ static void DC8uv(uint8_t* dst) {     // DC
   Put8x8uv(dc0 >> 4, dst);
 }
 
-static void DC8uvNoLeft(uint8_t* dst) {   // DC with no left samples
+static void DC8uvNoLeft_C(uint8_t* dst) {   // DC with no left samples
   int dc0 = 4;
   int i;
   for (i = 0; i < 8; ++i) {
@@ -445,7 +460,7 @@ static void DC8uvNoLeft(uint8_t* dst) {   // DC with no left samples
   Put8x8uv(dc0 >> 3, dst);
 }
 
-static void DC8uvNoTop(uint8_t* dst) {  // DC with no top samples
+static void DC8uvNoTop_C(uint8_t* dst) {  // DC with no top samples
   int dc0 = 4;
   int i;
   for (i = 0; i < 8; ++i) {
@@ -454,17 +469,19 @@ static void DC8uvNoTop(uint8_t* dst) {  // DC with no top samples
   Put8x8uv(dc0 >> 3, dst);
 }
 
-static void DC8uvNoTopLeft(uint8_t* dst) {    // DC with nothing
+static void DC8uvNoTopLeft_C(uint8_t* dst) {    // DC with nothing
   Put8x8uv(0x80, dst);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES];
 
 //------------------------------------------------------------------------------
 // Edge filtering functions
 
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 // 4 pixels in, 2 pixels out
-static WEBP_INLINE void do_filter2(uint8_t* p, int step) {
+static WEBP_INLINE void DoFilter2_C(uint8_t* p, int step) {
   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
   const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1];  // in [-893,892]
   const int a1 = VP8ksclip2[(a + 4) >> 3];            // in [-16,15]
@@ -474,7 +491,7 @@ static WEBP_INLINE void do_filter2(uint8_t* p, int step) {
 }
 
 // 4 pixels in, 4 pixels out
-static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
+static WEBP_INLINE void DoFilter4_C(uint8_t* p, int step) {
   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
   const int a = 3 * (q0 - p0);
   const int a1 = VP8ksclip2[(a + 4) >> 3];
@@ -487,7 +504,7 @@ static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
 }
 
 // 6 pixels in, 6 pixels out
-static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
+static WEBP_INLINE void DoFilter6_C(uint8_t* p, int step) {
   const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
   const int q0 = p[0], q1 = p[step], q2 = p[2*step];
   const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
@@ -503,18 +520,22 @@ static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
   p[ 2*step] = VP8kclip1[q2 - a3];
 }
 
-static WEBP_INLINE int hev(const uint8_t* p, int step, int thresh) {
+static WEBP_INLINE int Hev(const uint8_t* p, int step, int thresh) {
   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
   return (VP8kabs0[p1 - p0] > thresh) || (VP8kabs0[q1 - q0] > thresh);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 
-static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int t) {
+#if !WEBP_NEON_OMIT_C_CODE
+static WEBP_INLINE int NeedsFilter_C(const uint8_t* p, int step, int t) {
   const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
   return ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) <= t);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static WEBP_INLINE int needs_filter2(const uint8_t* p,
-                                     int step, int t, int it) {
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+static WEBP_INLINE int NeedsFilter2_C(const uint8_t* p,
+                                      int step, int t, int it) {
   const int p3 = p[-4 * step], p2 = p[-3 * step], p1 = p[-2 * step];
   const int p0 = p[-step], q0 = p[0];
   const int q1 = p[step], q2 = p[2 * step], q3 = p[3 * step];
@@ -523,140 +544,159 @@ static WEBP_INLINE int needs_filter2(const uint8_t* p,
          VP8kabs0[p1 - p0] <= it && VP8kabs0[q3 - q2] <= it &&
          VP8kabs0[q2 - q1] <= it && VP8kabs0[q1 - q0] <= it;
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 
 //------------------------------------------------------------------------------
 // Simple In-loop filtering (Paragraph 15.2)
 
-static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void SimpleVFilter16_C(uint8_t* p, int stride, int thresh) {
   int i;
   const int thresh2 = 2 * thresh + 1;
   for (i = 0; i < 16; ++i) {
-    if (needs_filter(p + i, stride, thresh2)) {
-      do_filter2(p + i, stride);
+    if (NeedsFilter_C(p + i, stride, thresh2)) {
+      DoFilter2_C(p + i, stride);
     }
   }
 }
 
-static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16_C(uint8_t* p, int stride, int thresh) {
   int i;
   const int thresh2 = 2 * thresh + 1;
   for (i = 0; i < 16; ++i) {
-    if (needs_filter(p + i * stride, 1, thresh2)) {
-      do_filter2(p + i * stride, 1);
+    if (NeedsFilter_C(p + i * stride, 1, thresh2)) {
+      DoFilter2_C(p + i * stride, 1);
     }
   }
 }
 
-static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16i_C(uint8_t* p, int stride, int thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4 * stride;
-    SimpleVFilter16(p, stride, thresh);
+    SimpleVFilter16_C(p, stride, thresh);
   }
 }
 
-static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16i_C(uint8_t* p, int stride, int thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4;
-    SimpleHFilter16(p, stride, thresh);
+    SimpleHFilter16_C(p, stride, thresh);
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 //------------------------------------------------------------------------------
 // Complex In-loop filtering (Paragraph 15.3)
 
-static WEBP_INLINE void FilterLoop26(uint8_t* p,
-                                     int hstride, int vstride, int size,
-                                     int thresh, int ithresh, int hev_thresh) {
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+static WEBP_INLINE void FilterLoop26_C(uint8_t* p,
+                                       int hstride, int vstride, int size,
+                                       int thresh, int ithresh,
+                                       int hev_thresh) {
   const int thresh2 = 2 * thresh + 1;
   while (size-- > 0) {
-    if (needs_filter2(p, hstride, thresh2, ithresh)) {
-      if (hev(p, hstride, hev_thresh)) {
-        do_filter2(p, hstride);
+    if (NeedsFilter2_C(p, hstride, thresh2, ithresh)) {
+      if (Hev(p, hstride, hev_thresh)) {
+        DoFilter2_C(p, hstride);
       } else {
-        do_filter6(p, hstride);
+        DoFilter6_C(p, hstride);
       }
     }
     p += vstride;
   }
 }
 
-static WEBP_INLINE void FilterLoop24(uint8_t* p,
-                                     int hstride, int vstride, int size,
-                                     int thresh, int ithresh, int hev_thresh) {
+static WEBP_INLINE void FilterLoop24_C(uint8_t* p,
+                                       int hstride, int vstride, int size,
+                                       int thresh, int ithresh,
+                                       int hev_thresh) {
   const int thresh2 = 2 * thresh + 1;
   while (size-- > 0) {
-    if (needs_filter2(p, hstride, thresh2, ithresh)) {
-      if (hev(p, hstride, hev_thresh)) {
-        do_filter2(p, hstride);
+    if (NeedsFilter2_C(p, hstride, thresh2, ithresh)) {
+      if (Hev(p, hstride, hev_thresh)) {
+        DoFilter2_C(p, hstride);
       } else {
-        do_filter4(p, hstride);
+        DoFilter4_C(p, hstride);
       }
     }
     p += vstride;
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 
+#if !WEBP_NEON_OMIT_C_CODE
 // on macroblock edges
-static void VFilter16(uint8_t* p, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
-  FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh);
+static void VFilter16_C(uint8_t* p, int stride,
+                        int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26_C(p, stride, 1, 16, thresh, ithresh, hev_thresh);
 }
 
-static void HFilter16(uint8_t* p, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
-  FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh);
+static void HFilter16_C(uint8_t* p, int stride,
+                        int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26_C(p, 1, stride, 16, thresh, ithresh, hev_thresh);
 }
 
 // on three inner edges
-static void VFilter16i(uint8_t* p, int stride,
-                       int thresh, int ithresh, int hev_thresh) {
+static void VFilter16i_C(uint8_t* p, int stride,
+                         int thresh, int ithresh, int hev_thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4 * stride;
-    FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh);
+    FilterLoop24_C(p, stride, 1, 16, thresh, ithresh, hev_thresh);
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void HFilter16i(uint8_t* p, int stride,
-                       int thresh, int ithresh, int hev_thresh) {
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+static void HFilter16i_C(uint8_t* p, int stride,
+                         int thresh, int ithresh, int hev_thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4;
-    FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh);
+    FilterLoop24_C(p, 1, stride, 16, thresh, ithresh, hev_thresh);
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 
+#if !WEBP_NEON_OMIT_C_CODE
 // 8-pixels wide variant, for chroma filtering
-static void VFilter8(uint8_t* u, uint8_t* v, int stride,
-                     int thresh, int ithresh, int hev_thresh) {
-  FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh);
-  FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh);
+static void VFilter8_C(uint8_t* u, uint8_t* v, int stride,
+                       int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26_C(u, stride, 1, 8, thresh, ithresh, hev_thresh);
+  FilterLoop26_C(v, stride, 1, 8, thresh, ithresh, hev_thresh);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void HFilter8(uint8_t* u, uint8_t* v, int stride,
-                     int thresh, int ithresh, int hev_thresh) {
-  FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh);
-  FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh);
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+static void HFilter8_C(uint8_t* u, uint8_t* v, int stride,
+                       int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26_C(u, 1, stride, 8, thresh, ithresh, hev_thresh);
+  FilterLoop26_C(v, 1, stride, 8, thresh, ithresh, hev_thresh);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 
-static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
-  FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
-  FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
+#if !WEBP_NEON_OMIT_C_CODE
+static void VFilter8i_C(uint8_t* u, uint8_t* v, int stride,
+                        int thresh, int ithresh, int hev_thresh) {
+  FilterLoop24_C(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
+  FilterLoop24_C(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
-  FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
-  FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+static void HFilter8i_C(uint8_t* u, uint8_t* v, int stride,
+                        int thresh, int ithresh, int hev_thresh) {
+  FilterLoop24_C(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
+  FilterLoop24_C(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 
 //------------------------------------------------------------------------------
 
-static void DitherCombine8x8(const uint8_t* dither, uint8_t* dst,
-                             int dst_stride) {
+static void DitherCombine8x8_C(const uint8_t* dither, uint8_t* dst,
+                               int dst_stride) {
   int i, j;
   for (j = 0; j < 8; ++j) {
     for (i = 0; i < 8; ++i) {
@@ -709,54 +749,66 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) {
 
   VP8InitClipTables();
 
-  VP8TransformWHT = TransformWHT;
-  VP8Transform = TransformTwo;
-  VP8TransformUV = TransformUV;
-  VP8TransformDC = TransformDC;
-  VP8TransformDCUV = TransformDCUV;
-  VP8TransformAC3 = TransformAC3;
-
-  VP8VFilter16 = VFilter16;
-  VP8HFilter16 = HFilter16;
-  VP8VFilter8 = VFilter8;
-  VP8HFilter8 = HFilter8;
-  VP8VFilter16i = VFilter16i;
-  VP8HFilter16i = HFilter16i;
-  VP8VFilter8i = VFilter8i;
-  VP8HFilter8i = HFilter8i;
-  VP8SimpleVFilter16 = SimpleVFilter16;
-  VP8SimpleHFilter16 = SimpleHFilter16;
-  VP8SimpleVFilter16i = SimpleVFilter16i;
-  VP8SimpleHFilter16i = SimpleHFilter16i;
-
-  VP8PredLuma4[0] = DC4;
-  VP8PredLuma4[1] = TM4;
-  VP8PredLuma4[2] = VE4;
-  VP8PredLuma4[3] = HE4;
-  VP8PredLuma4[4] = RD4;
-  VP8PredLuma4[5] = VR4;
-  VP8PredLuma4[6] = LD4;
-  VP8PredLuma4[7] = VL4;
-  VP8PredLuma4[8] = HD4;
-  VP8PredLuma4[9] = HU4;
-
-  VP8PredLuma16[0] = DC16;
-  VP8PredLuma16[1] = TM16;
-  VP8PredLuma16[2] = VE16;
-  VP8PredLuma16[3] = HE16;
-  VP8PredLuma16[4] = DC16NoTop;
-  VP8PredLuma16[5] = DC16NoLeft;
-  VP8PredLuma16[6] = DC16NoTopLeft;
-
-  VP8PredChroma8[0] = DC8uv;
-  VP8PredChroma8[1] = TM8uv;
-  VP8PredChroma8[2] = VE8uv;
-  VP8PredChroma8[3] = HE8uv;
-  VP8PredChroma8[4] = DC8uvNoTop;
-  VP8PredChroma8[5] = DC8uvNoLeft;
-  VP8PredChroma8[6] = DC8uvNoTopLeft;
-
-  VP8DitherCombine8x8 = DitherCombine8x8;
+#if !WEBP_NEON_OMIT_C_CODE
+  VP8TransformWHT = TransformWHT_C;
+  VP8Transform = TransformTwo_C;
+  VP8TransformDC = TransformDC_C;
+  VP8TransformAC3 = TransformAC3_C;
+#endif
+  VP8TransformUV = TransformUV_C;
+  VP8TransformDCUV = TransformDCUV_C;
+
+#if !WEBP_NEON_OMIT_C_CODE
+  VP8VFilter16 = VFilter16_C;
+  VP8VFilter16i = VFilter16i_C;
+  VP8HFilter16 = HFilter16_C;
+  VP8VFilter8 = VFilter8_C;
+  VP8VFilter8i = VFilter8i_C;
+  VP8SimpleVFilter16 = SimpleVFilter16_C;
+  VP8SimpleHFilter16 = SimpleHFilter16_C;
+  VP8SimpleVFilter16i = SimpleVFilter16i_C;
+  VP8SimpleHFilter16i = SimpleHFilter16i_C;
+#endif
+
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+  VP8HFilter16i = HFilter16i_C;
+  VP8HFilter8 = HFilter8_C;
+  VP8HFilter8i = HFilter8i_C;
+#endif
+
+#if !WEBP_NEON_OMIT_C_CODE
+  VP8PredLuma4[0] = DC4_C;
+  VP8PredLuma4[1] = TM4_C;
+  VP8PredLuma4[2] = VE4_C;
+  VP8PredLuma4[4] = RD4_C;
+  VP8PredLuma4[6] = LD4_C;
+#endif
+
+  VP8PredLuma4[3] = HE4_C;
+  VP8PredLuma4[5] = VR4_C;
+  VP8PredLuma4[7] = VL4_C;
+  VP8PredLuma4[8] = HD4_C;
+  VP8PredLuma4[9] = HU4_C;
+
+#if !WEBP_NEON_OMIT_C_CODE
+  VP8PredLuma16[0] = DC16_C;
+  VP8PredLuma16[1] = TM16_C;
+  VP8PredLuma16[2] = VE16_C;
+  VP8PredLuma16[3] = HE16_C;
+  VP8PredLuma16[4] = DC16NoTop_C;
+  VP8PredLuma16[5] = DC16NoLeft_C;
+  VP8PredLuma16[6] = DC16NoTopLeft_C;
+
+  VP8PredChroma8[0] = DC8uv_C;
+  VP8PredChroma8[1] = TM8uv_C;
+  VP8PredChroma8[2] = VE8uv_C;
+  VP8PredChroma8[3] = HE8uv_C;
+  VP8PredChroma8[4] = DC8uvNoTop_C;
+  VP8PredChroma8[5] = DC8uvNoLeft_C;
+  VP8PredChroma8[6] = DC8uvNoTopLeft_C;
+#endif
+
+  VP8DitherCombine8x8 = DitherCombine8x8_C;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
@@ -770,11 +822,6 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) {
 #endif
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      VP8DspInitNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
       VP8DspInitMIPS32();
@@ -791,5 +838,57 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) {
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    VP8DspInitNEON();
+  }
+#endif
+
+  assert(VP8TransformWHT != NULL);
+  assert(VP8Transform != NULL);
+  assert(VP8TransformDC != NULL);
+  assert(VP8TransformAC3 != NULL);
+  assert(VP8TransformUV != NULL);
+  assert(VP8TransformDCUV != NULL);
+  assert(VP8VFilter16 != NULL);
+  assert(VP8HFilter16 != NULL);
+  assert(VP8VFilter8 != NULL);
+  assert(VP8HFilter8 != NULL);
+  assert(VP8VFilter16i != NULL);
+  assert(VP8HFilter16i != NULL);
+  assert(VP8VFilter8i != NULL);
+  assert(VP8HFilter8i != NULL);
+  assert(VP8SimpleVFilter16 != NULL);
+  assert(VP8SimpleHFilter16 != NULL);
+  assert(VP8SimpleVFilter16i != NULL);
+  assert(VP8SimpleHFilter16i != NULL);
+  assert(VP8PredLuma4[0] != NULL);
+  assert(VP8PredLuma4[1] != NULL);
+  assert(VP8PredLuma4[2] != NULL);
+  assert(VP8PredLuma4[3] != NULL);
+  assert(VP8PredLuma4[4] != NULL);
+  assert(VP8PredLuma4[5] != NULL);
+  assert(VP8PredLuma4[6] != NULL);
+  assert(VP8PredLuma4[7] != NULL);
+  assert(VP8PredLuma4[8] != NULL);
+  assert(VP8PredLuma4[9] != NULL);
+  assert(VP8PredLuma16[0] != NULL);
+  assert(VP8PredLuma16[1] != NULL);
+  assert(VP8PredLuma16[2] != NULL);
+  assert(VP8PredLuma16[3] != NULL);
+  assert(VP8PredLuma16[4] != NULL);
+  assert(VP8PredLuma16[5] != NULL);
+  assert(VP8PredLuma16[6] != NULL);
+  assert(VP8PredChroma8[0] != NULL);
+  assert(VP8PredChroma8[1] != NULL);
+  assert(VP8PredChroma8[2] != NULL);
+  assert(VP8PredChroma8[3] != NULL);
+  assert(VP8PredChroma8[4] != NULL);
+  assert(VP8PredChroma8[5] != NULL);
+  assert(VP8PredChroma8[6] != NULL);
+  assert(VP8DitherCombine8x8 != NULL);
+
   dec_last_cpuinfo_used = VP8GetCPUInfo;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/dec_clip_tables.c b/src/3rdparty/libwebp/src/dsp/dec_clip_tables.c
index 74ba34c..427b74f 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_clip_tables.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_clip_tables.c
@@ -11,11 +11,14 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
-#define USE_STATIC_TABLES     // undefine to have run-time table initialization
+// define to 0 to have run-time table initialization
+#if !defined(USE_STATIC_TABLES)
+#define USE_STATIC_TABLES 1   // ALTERNATE_CODE
+#endif
 
-#ifdef USE_STATIC_TABLES
+#if (USE_STATIC_TABLES == 1)
 
 static const uint8_t abs0[255 + 255 + 1] = {
   0xff, 0xfe, 0xfd, 0xfc, 0xfb, 0xfa, 0xf9, 0xf8, 0xf7, 0xf6, 0xf5, 0xf4,
@@ -337,7 +340,7 @@ static uint8_t clip1[255 + 511 + 1];
 // and make sure it's set to true _last_ (so as to be thread-safe)
 static volatile int tables_ok = 0;
 
-#endif
+#endif    // USE_STATIC_TABLES
 
 const int8_t* const VP8ksclip1 = (const int8_t*)&sclip1[1020];
 const int8_t* const VP8ksclip2 = (const int8_t*)&sclip2[112];
@@ -345,7 +348,7 @@ const uint8_t* const VP8kclip1 = &clip1[255];
 const uint8_t* const VP8kabs0 = &abs0[255];
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8InitClipTables(void) {
-#if !defined(USE_STATIC_TABLES)
+#if (USE_STATIC_TABLES == 0)
   int i;
   if (!tables_ok) {
     for (i = -255; i <= 255; ++i) {
diff --git a/src/3rdparty/libwebp/src/dsp/dec_mips32.c b/src/3rdparty/libwebp/src/dsp/dec_mips32.c
index 4e9ef42..e4e7096 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_mips32.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_mips32.c
@@ -12,11 +12,11 @@
 // Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
 //             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS32)
 
-#include "./mips_macro.h"
+#include "src/dsp/mips_macro.h"
 
 static const int kC1 = 20091 + (1 << 16);
 static const int kC2 = 35468;
diff --git a/src/3rdparty/libwebp/src/dsp/dec_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/dec_mips_dsp_r2.c
index db5c657..b0936bc 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_mips_dsp_r2.c
@@ -12,11 +12,11 @@
 // Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
 //             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-#include "./mips_macro.h"
+#include "src/dsp/mips_macro.h"
 
 static const int kC1 = 20091 + (1 << 16);
 static const int kC2 = 35468;
diff --git a/src/3rdparty/libwebp/src/dsp/dec_msa.c b/src/3rdparty/libwebp/src/dsp/dec_msa.c
index 8d9c98c..8090622 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_msa.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_msa.c
@@ -12,11 +12,11 @@
 // Author(s):  Prashant Patil   (prashant.patil@imgtec.com)
 
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MSA)
 
-#include "./msa_macro.h"
+#include "src/dsp/msa_macro.h"
 
 //------------------------------------------------------------------------------
 // Transforms
@@ -222,6 +222,7 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
   const v16i8 cnst4b = __msa_ldi_b(4);                        \
   const v16i8 cnst3b = __msa_ldi_b(3);                        \
   const v8i16 cnst9h = __msa_ldi_h(9);                        \
+  const v8i16 cnst63h = __msa_ldi_h(63);                      \
                                                               \
   FLIP_SIGN4(p1, p0, q0, q1, p1_m, p0_m, q0_m, q1_m);         \
   filt = __msa_subs_s_b(p1_m, q1_m);                          \
@@ -241,9 +242,9 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
   ILVRL_B2_SH(filt_sign, filt, filt_r, filt_l);               \
   /* update q2/p2 */                                          \
   temp0 = filt_r * cnst9h;                                    \
-  temp1 = ADDVI_H(temp0, 63);                                 \
+  temp1 = temp0 + cnst63h;                                    \
   temp2 = filt_l * cnst9h;                                    \
-  temp3 = ADDVI_H(temp2, 63);                                 \
+  temp3 = temp2 + cnst63h;                                    \
   FILT2(q2_m, p2_m, q2, p2);                                  \
   /* update q1/p1 */                                          \
   temp1 = temp1 + temp0;                                      \
@@ -708,7 +709,7 @@ static void VE4(uint8_t* dst) {    // vertical
   const uint32_t val0 = LW(ptop + 0);
   const uint32_t val1 = LW(ptop + 4);
   uint32_t out;
-  v16u8 A, B, C, AC, B2, R;
+  v16u8 A = { 0 }, B, C, AC, B2, R;
 
   INSERT_W2_UB(val0, val1, A);
   B = SLDI_UB(A, A, 1);
@@ -725,7 +726,7 @@ static void RD4(uint8_t* dst) {   // Down-right
   uint32_t val0 = LW(ptop + 0);
   uint32_t val1 = LW(ptop + 4);
   uint32_t val2, val3;
-  v16u8 A, B, C, AC, B2, R, A1;
+  v16u8 A, B, C, AC, B2, R, A1 = { 0 };
 
   INSERT_W2_UB(val0, val1, A1);
   A = SLDI_UB(A1, A1, 12);
@@ -753,7 +754,7 @@ static void LD4(uint8_t* dst) {   // Down-Left
   uint32_t val0 = LW(ptop + 0);
   uint32_t val1 = LW(ptop + 4);
   uint32_t val2, val3;
-  v16u8 A, B, C, AC, B2, R;
+  v16u8 A = { 0 }, B, C, AC, B2, R;
 
   INSERT_W2_UB(val0, val1, A);
   B = SLDI_UB(A, A, 1);
diff --git a/src/3rdparty/libwebp/src/dsp/dec_neon.c b/src/3rdparty/libwebp/src/dsp/dec_neon.c
index 34796cf..ffa697f 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_neon.c
@@ -12,43 +12,23 @@
 // Authors: Somnath Banerjee (somnath@google.com)
 //          Johann Koenig (johannkoenig@google.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_NEON)
 
-#include "./neon.h"
-#include "../dec/vp8i_dec.h"
+#include "src/dsp/neon.h"
+#include "src/dec/vp8i_dec.h"
 
 //------------------------------------------------------------------------------
 // NxM Loading functions
 
-// Load/Store vertical edge
-#define LOAD8x4(c1, c2, c3, c4, b1, b2, stride)                                \
-  "vld4.8 {" #c1 "[0]," #c2 "[0]," #c3 "[0]," #c4 "[0]}," #b1 "," #stride "\n" \
-  "vld4.8 {" #c1 "[1]," #c2 "[1]," #c3 "[1]," #c4 "[1]}," #b2 "," #stride "\n" \
-  "vld4.8 {" #c1 "[2]," #c2 "[2]," #c3 "[2]," #c4 "[2]}," #b1 "," #stride "\n" \
-  "vld4.8 {" #c1 "[3]," #c2 "[3]," #c3 "[3]," #c4 "[3]}," #b2 "," #stride "\n" \
-  "vld4.8 {" #c1 "[4]," #c2 "[4]," #c3 "[4]," #c4 "[4]}," #b1 "," #stride "\n" \
-  "vld4.8 {" #c1 "[5]," #c2 "[5]," #c3 "[5]," #c4 "[5]}," #b2 "," #stride "\n" \
-  "vld4.8 {" #c1 "[6]," #c2 "[6]," #c3 "[6]," #c4 "[6]}," #b1 "," #stride "\n" \
-  "vld4.8 {" #c1 "[7]," #c2 "[7]," #c3 "[7]," #c4 "[7]}," #b2 "," #stride "\n"
-
-#define STORE8x2(c1, c2, p, stride)                                            \
-  "vst2.8   {" #c1 "[0], " #c2 "[0]}," #p "," #stride " \n"                    \
-  "vst2.8   {" #c1 "[1], " #c2 "[1]}," #p "," #stride " \n"                    \
-  "vst2.8   {" #c1 "[2], " #c2 "[2]}," #p "," #stride " \n"                    \
-  "vst2.8   {" #c1 "[3], " #c2 "[3]}," #p "," #stride " \n"                    \
-  "vst2.8   {" #c1 "[4], " #c2 "[4]}," #p "," #stride " \n"                    \
-  "vst2.8   {" #c1 "[5], " #c2 "[5]}," #p "," #stride " \n"                    \
-  "vst2.8   {" #c1 "[6], " #c2 "[6]}," #p "," #stride " \n"                    \
-  "vst2.8   {" #c1 "[7], " #c2 "[7]}," #p "," #stride " \n"
-
 #if !defined(WORK_AROUND_GCC)
 
 // This intrinsics version makes gcc-4.6.3 crash during Load4x??() compilation
 // (register alloc, probably). The variants somewhat mitigate the problem, but
 // not quite. HFilter16i() remains problematic.
-static WEBP_INLINE uint8x8x4_t Load4x8(const uint8_t* const src, int stride) {
+static WEBP_INLINE uint8x8x4_t Load4x8_NEON(const uint8_t* const src,
+                                            int stride) {
   const uint8x8_t zero = vdup_n_u8(0);
   uint8x8x4_t out;
   INIT_VECTOR4(out, zero, zero, zero, zero);
@@ -63,13 +43,15 @@ static WEBP_INLINE uint8x8x4_t Load4x8(const uint8_t* const src, int stride) {
   return out;
 }
 
-static WEBP_INLINE void Load4x16(const uint8_t* const src, int stride,
-                                 uint8x16_t* const p1, uint8x16_t* const p0,
-                                 uint8x16_t* const q0, uint8x16_t* const q1) {
+static WEBP_INLINE void Load4x16_NEON(const uint8_t* const src, int stride,
+                                      uint8x16_t* const p1,
+                                      uint8x16_t* const p0,
+                                      uint8x16_t* const q0,
+                                      uint8x16_t* const q1) {
   // row0 = p1[0..7]|p0[0..7]|q0[0..7]|q1[0..7]
   // row8 = p1[8..15]|p0[8..15]|q0[8..15]|q1[8..15]
-  const uint8x8x4_t row0 = Load4x8(src - 2 + 0 * stride, stride);
-  const uint8x8x4_t row8 = Load4x8(src - 2 + 8 * stride, stride);
+  const uint8x8x4_t row0 = Load4x8_NEON(src - 2 + 0 * stride, stride);
+  const uint8x8x4_t row8 = Load4x8_NEON(src - 2 + 8 * stride, stride);
   *p1 = vcombine_u8(row0.val[0], row8.val[0]);
   *p0 = vcombine_u8(row0.val[1], row8.val[1]);
   *q0 = vcombine_u8(row0.val[2], row8.val[2]);
@@ -83,9 +65,11 @@ static WEBP_INLINE void Load4x16(const uint8_t* const src, int stride,
   src += stride;                                                     \
 } while (0)
 
-static WEBP_INLINE void Load4x16(const uint8_t* src, int stride,
-                                 uint8x16_t* const p1, uint8x16_t* const p0,
-                                 uint8x16_t* const q0, uint8x16_t* const q1) {
+static WEBP_INLINE void Load4x16_NEON(const uint8_t* src, int stride,
+                                      uint8x16_t* const p1,
+                                      uint8x16_t* const p0,
+                                      uint8x16_t* const q0,
+                                      uint8x16_t* const q1) {
   const uint32x4_t zero = vdupq_n_u32(0);
   uint32x4x4_t in;
   INIT_VECTOR4(in, zero, zero, zero, zero);
@@ -126,40 +110,40 @@ static WEBP_INLINE void Load4x16(const uint8_t* src, int stride,
 
 #endif  // !WORK_AROUND_GCC
 
-static WEBP_INLINE void Load8x16(const uint8_t* const src, int stride,
-                                 uint8x16_t* const p3, uint8x16_t* const p2,
-                                 uint8x16_t* const p1, uint8x16_t* const p0,
-                                 uint8x16_t* const q0, uint8x16_t* const q1,
-                                 uint8x16_t* const q2, uint8x16_t* const q3) {
-  Load4x16(src - 2, stride, p3, p2, p1, p0);
-  Load4x16(src + 2, stride, q0, q1, q2, q3);
+static WEBP_INLINE void Load8x16_NEON(
+    const uint8_t* const src, int stride,
+    uint8x16_t* const p3, uint8x16_t* const p2, uint8x16_t* const p1,
+    uint8x16_t* const p0, uint8x16_t* const q0, uint8x16_t* const q1,
+    uint8x16_t* const q2, uint8x16_t* const q3) {
+  Load4x16_NEON(src - 2, stride, p3, p2, p1, p0);
+  Load4x16_NEON(src + 2, stride, q0, q1, q2, q3);
 }
 
-static WEBP_INLINE void Load16x4(const uint8_t* const src, int stride,
-                                 uint8x16_t* const p1, uint8x16_t* const p0,
-                                 uint8x16_t* const q0, uint8x16_t* const q1) {
+static WEBP_INLINE void Load16x4_NEON(const uint8_t* const src, int stride,
+                                      uint8x16_t* const p1,
+                                      uint8x16_t* const p0,
+                                      uint8x16_t* const q0,
+                                      uint8x16_t* const q1) {
   *p1 = vld1q_u8(src - 2 * stride);
   *p0 = vld1q_u8(src - 1 * stride);
   *q0 = vld1q_u8(src + 0 * stride);
   *q1 = vld1q_u8(src + 1 * stride);
 }
 
-static WEBP_INLINE void Load16x8(const uint8_t* const src, int stride,
-                                 uint8x16_t* const p3, uint8x16_t* const p2,
-                                 uint8x16_t* const p1, uint8x16_t* const p0,
-                                 uint8x16_t* const q0, uint8x16_t* const q1,
-                                 uint8x16_t* const q2, uint8x16_t* const q3) {
-  Load16x4(src - 2  * stride, stride, p3, p2, p1, p0);
-  Load16x4(src + 2  * stride, stride, q0, q1, q2, q3);
+static WEBP_INLINE void Load16x8_NEON(
+    const uint8_t* const src, int stride,
+    uint8x16_t* const p3, uint8x16_t* const p2, uint8x16_t* const p1,
+    uint8x16_t* const p0, uint8x16_t* const q0, uint8x16_t* const q1,
+    uint8x16_t* const q2, uint8x16_t* const q3) {
+  Load16x4_NEON(src - 2  * stride, stride, p3, p2, p1, p0);
+  Load16x4_NEON(src + 2  * stride, stride, q0, q1, q2, q3);
 }
 
-static WEBP_INLINE void Load8x8x2(const uint8_t* const u,
-                                  const uint8_t* const v,
-                                  int stride,
-                                  uint8x16_t* const p3, uint8x16_t* const p2,
-                                  uint8x16_t* const p1, uint8x16_t* const p0,
-                                  uint8x16_t* const q0, uint8x16_t* const q1,
-                                  uint8x16_t* const q2, uint8x16_t* const q3) {
+static WEBP_INLINE void Load8x8x2_NEON(
+    const uint8_t* const u, const uint8_t* const v, int stride,
+    uint8x16_t* const p3, uint8x16_t* const p2, uint8x16_t* const p1,
+    uint8x16_t* const p0, uint8x16_t* const q0, uint8x16_t* const q1,
+    uint8x16_t* const q2, uint8x16_t* const q3) {
   // We pack the 8x8 u-samples in the lower half of the uint8x16_t destination
   // and the v-samples on the higher half.
   *p3 = vcombine_u8(vld1_u8(u - 4 * stride), vld1_u8(v - 4 * stride));
@@ -177,13 +161,11 @@ static WEBP_INLINE void Load8x8x2(const uint8_t* const u,
 #define LOAD_UV_8(ROW) \
   vcombine_u8(vld1_u8(u - 4 + (ROW) * stride), vld1_u8(v - 4 + (ROW) * stride))
 
-static WEBP_INLINE void Load8x8x2T(const uint8_t* const u,
-                                   const uint8_t* const v,
-                                   int stride,
-                                   uint8x16_t* const p3, uint8x16_t* const p2,
-                                   uint8x16_t* const p1, uint8x16_t* const p0,
-                                   uint8x16_t* const q0, uint8x16_t* const q1,
-                                   uint8x16_t* const q2, uint8x16_t* const q3) {
+static WEBP_INLINE void Load8x8x2T_NEON(
+    const uint8_t* const u, const uint8_t* const v, int stride,
+    uint8x16_t* const p3, uint8x16_t* const p2, uint8x16_t* const p1,
+    uint8x16_t* const p0, uint8x16_t* const q0, uint8x16_t* const q1,
+    uint8x16_t* const q2, uint8x16_t* const q3) {
   // We pack the 8x8 u-samples in the lower half of the uint8x16_t destination
   // and the v-samples on the higher half.
   const uint8x16_t row0 = LOAD_UV_8(0);
@@ -238,8 +220,8 @@ static WEBP_INLINE void Load8x8x2T(const uint8_t* const u,
 
 #endif  // !WORK_AROUND_GCC
 
-static WEBP_INLINE void Store2x8(const uint8x8x2_t v,
-                                 uint8_t* const dst, int stride) {
+static WEBP_INLINE void Store2x8_NEON(const uint8x8x2_t v,
+                                      uint8_t* const dst, int stride) {
   vst2_lane_u8(dst + 0 * stride, v, 0);
   vst2_lane_u8(dst + 1 * stride, v, 1);
   vst2_lane_u8(dst + 2 * stride, v, 2);
@@ -250,20 +232,20 @@ static WEBP_INLINE void Store2x8(const uint8x8x2_t v,
   vst2_lane_u8(dst + 7 * stride, v, 7);
 }
 
-static WEBP_INLINE void Store2x16(const uint8x16_t p0, const uint8x16_t q0,
-                                  uint8_t* const dst, int stride) {
+static WEBP_INLINE void Store2x16_NEON(const uint8x16_t p0, const uint8x16_t q0,
+                                       uint8_t* const dst, int stride) {
   uint8x8x2_t lo, hi;
   lo.val[0] = vget_low_u8(p0);
   lo.val[1] = vget_low_u8(q0);
   hi.val[0] = vget_high_u8(p0);
   hi.val[1] = vget_high_u8(q0);
-  Store2x8(lo, dst - 1 + 0 * stride, stride);
-  Store2x8(hi, dst - 1 + 8 * stride, stride);
+  Store2x8_NEON(lo, dst - 1 + 0 * stride, stride);
+  Store2x8_NEON(hi, dst - 1 + 8 * stride, stride);
 }
 
 #if !defined(WORK_AROUND_GCC)
-static WEBP_INLINE void Store4x8(const uint8x8x4_t v,
-                                 uint8_t* const dst, int stride) {
+static WEBP_INLINE void Store4x8_NEON(const uint8x8x4_t v,
+                                      uint8_t* const dst, int stride) {
   vst4_lane_u8(dst + 0 * stride, v, 0);
   vst4_lane_u8(dst + 1 * stride, v, 1);
   vst4_lane_u8(dst + 2 * stride, v, 2);
@@ -274,9 +256,9 @@ static WEBP_INLINE void Store4x8(const uint8x8x4_t v,
   vst4_lane_u8(dst + 7 * stride, v, 7);
 }
 
-static WEBP_INLINE void Store4x16(const uint8x16_t p1, const uint8x16_t p0,
-                                  const uint8x16_t q0, const uint8x16_t q1,
-                                  uint8_t* const dst, int stride) {
+static WEBP_INLINE void Store4x16_NEON(const uint8x16_t p1, const uint8x16_t p0,
+                                       const uint8x16_t q0, const uint8x16_t q1,
+                                       uint8_t* const dst, int stride) {
   uint8x8x4_t lo, hi;
   INIT_VECTOR4(lo,
                vget_low_u8(p1), vget_low_u8(p0),
@@ -284,27 +266,28 @@ static WEBP_INLINE void Store4x16(const uint8x16_t p1, const uint8x16_t p0,
   INIT_VECTOR4(hi,
                vget_high_u8(p1), vget_high_u8(p0),
                vget_high_u8(q0), vget_high_u8(q1));
-  Store4x8(lo, dst - 2 + 0 * stride, stride);
-  Store4x8(hi, dst - 2 + 8 * stride, stride);
+  Store4x8_NEON(lo, dst - 2 + 0 * stride, stride);
+  Store4x8_NEON(hi, dst - 2 + 8 * stride, stride);
 }
 #endif  // !WORK_AROUND_GCC
 
-static WEBP_INLINE void Store16x2(const uint8x16_t p0, const uint8x16_t q0,
-                                  uint8_t* const dst, int stride) {
+static WEBP_INLINE void Store16x2_NEON(const uint8x16_t p0, const uint8x16_t q0,
+                                       uint8_t* const dst, int stride) {
   vst1q_u8(dst - stride, p0);
   vst1q_u8(dst, q0);
 }
 
-static WEBP_INLINE void Store16x4(const uint8x16_t p1, const uint8x16_t p0,
-                                  const uint8x16_t q0, const uint8x16_t q1,
-                                  uint8_t* const dst, int stride) {
-  Store16x2(p1, p0, dst - stride, stride);
-  Store16x2(q0, q1, dst + stride, stride);
+static WEBP_INLINE void Store16x4_NEON(const uint8x16_t p1, const uint8x16_t p0,
+                                       const uint8x16_t q0, const uint8x16_t q1,
+                                       uint8_t* const dst, int stride) {
+  Store16x2_NEON(p1, p0, dst - stride, stride);
+  Store16x2_NEON(q0, q1, dst + stride, stride);
 }
 
-static WEBP_INLINE void Store8x2x2(const uint8x16_t p0, const uint8x16_t q0,
-                                   uint8_t* const u, uint8_t* const v,
-                                   int stride) {
+static WEBP_INLINE void Store8x2x2_NEON(const uint8x16_t p0,
+                                        const uint8x16_t q0,
+                                        uint8_t* const u, uint8_t* const v,
+                                        int stride) {
   // p0 and q0 contain the u+v samples packed in low/high halves.
   vst1_u8(u - stride, vget_low_u8(p0));
   vst1_u8(u,          vget_low_u8(q0));
@@ -312,13 +295,15 @@ static WEBP_INLINE void Store8x2x2(const uint8x16_t p0, const uint8x16_t q0,
   vst1_u8(v,          vget_high_u8(q0));
 }
 
-static WEBP_INLINE void Store8x4x2(const uint8x16_t p1, const uint8x16_t p0,
-                                   const uint8x16_t q0, const uint8x16_t q1,
-                                   uint8_t* const u, uint8_t* const v,
-                                   int stride) {
+static WEBP_INLINE void Store8x4x2_NEON(const uint8x16_t p1,
+                                        const uint8x16_t p0,
+                                        const uint8x16_t q0,
+                                        const uint8x16_t q1,
+                                        uint8_t* const u, uint8_t* const v,
+                                        int stride) {
   // The p1...q1 registers contain the u+v samples packed in low/high halves.
-  Store8x2x2(p1, p0, u - stride, v - stride, stride);
-  Store8x2x2(q0, q1, u + stride, v + stride, stride);
+  Store8x2x2_NEON(p1, p0, u - stride, v - stride, stride);
+  Store8x2x2_NEON(q0, q1, u + stride, v + stride, stride);
 }
 
 #if !defined(WORK_AROUND_GCC)
@@ -329,11 +314,10 @@ static WEBP_INLINE void Store8x4x2(const uint8x16_t p1, const uint8x16_t p0,
   (DST) += stride;                                \
 } while (0)
 
-static WEBP_INLINE void Store6x8x2(const uint8x16_t p2, const uint8x16_t p1,
-                                   const uint8x16_t p0, const uint8x16_t q0,
-                                   const uint8x16_t q1, const uint8x16_t q2,
-                                   uint8_t* u, uint8_t* v,
-                                   int stride) {
+static WEBP_INLINE void Store6x8x2_NEON(
+    const uint8x16_t p2, const uint8x16_t p1, const uint8x16_t p0,
+    const uint8x16_t q0, const uint8x16_t q1, const uint8x16_t q2,
+    uint8_t* u, uint8_t* v, int stride) {
   uint8x8x3_t u0, u1, v0, v1;
   INIT_VECTOR3(u0, vget_low_u8(p2), vget_low_u8(p1), vget_low_u8(p0));
   INIT_VECTOR3(u1, vget_low_u8(q0), vget_low_u8(q1), vget_low_u8(q2));
@@ -358,10 +342,12 @@ static WEBP_INLINE void Store6x8x2(const uint8x16_t p2, const uint8x16_t p1,
 }
 #undef STORE6_LANE
 
-static WEBP_INLINE void Store4x8x2(const uint8x16_t p1, const uint8x16_t p0,
-                                   const uint8x16_t q0, const uint8x16_t q1,
-                                   uint8_t* const u, uint8_t* const v,
-                                   int stride) {
+static WEBP_INLINE void Store4x8x2_NEON(const uint8x16_t p1,
+                                        const uint8x16_t p0,
+                                        const uint8x16_t q0,
+                                        const uint8x16_t q1,
+                                        uint8_t* const u, uint8_t* const v,
+                                        int stride) {
   uint8x8x4_t u0, v0;
   INIT_VECTOR4(u0,
                vget_low_u8(p1), vget_low_u8(p0),
@@ -390,15 +376,15 @@ static WEBP_INLINE void Store4x8x2(const uint8x16_t p1, const uint8x16_t p0,
 #endif  // !WORK_AROUND_GCC
 
 // Zero extend 'v' to an int16x8_t.
-static WEBP_INLINE int16x8_t ConvertU8ToS16(uint8x8_t v) {
+static WEBP_INLINE int16x8_t ConvertU8ToS16_NEON(uint8x8_t v) {
   return vreinterpretq_s16_u16(vmovl_u8(v));
 }
 
 // Performs unsigned 8b saturation on 'dst01' and 'dst23' storing the result
 // to the corresponding rows of 'dst'.
-static WEBP_INLINE void SaturateAndStore4x4(uint8_t* const dst,
-                                            const int16x8_t dst01,
-                                            const int16x8_t dst23) {
+static WEBP_INLINE void SaturateAndStore4x4_NEON(uint8_t* const dst,
+                                                 const int16x8_t dst01,
+                                                 const int16x8_t dst23) {
   // Unsigned saturate to 8b.
   const uint8x8_t dst01_u8 = vqmovun_s16(dst01);
   const uint8x8_t dst23_u8 = vqmovun_s16(dst23);
@@ -410,8 +396,9 @@ static WEBP_INLINE void SaturateAndStore4x4(uint8_t* const dst,
   vst1_lane_u32((uint32_t*)(dst + 3 * BPS), vreinterpret_u32_u8(dst23_u8), 1);
 }
 
-static WEBP_INLINE void Add4x4(const int16x8_t row01, const int16x8_t row23,
-                               uint8_t* const dst) {
+static WEBP_INLINE void Add4x4_NEON(const int16x8_t row01,
+                                    const int16x8_t row23,
+                                    uint8_t* const dst) {
   uint32x2_t dst01 = vdup_n_u32(0);
   uint32x2_t dst23 = vdup_n_u32(0);
 
@@ -423,23 +410,23 @@ static WEBP_INLINE void Add4x4(const int16x8_t row01, const int16x8_t row23,
 
   {
     // Convert to 16b.
-    const int16x8_t dst01_s16 = ConvertU8ToS16(vreinterpret_u8_u32(dst01));
-    const int16x8_t dst23_s16 = ConvertU8ToS16(vreinterpret_u8_u32(dst23));
+    const int16x8_t dst01_s16 = ConvertU8ToS16_NEON(vreinterpret_u8_u32(dst01));
+    const int16x8_t dst23_s16 = ConvertU8ToS16_NEON(vreinterpret_u8_u32(dst23));
 
     // Descale with rounding.
     const int16x8_t out01 = vrsraq_n_s16(dst01_s16, row01, 3);
     const int16x8_t out23 = vrsraq_n_s16(dst23_s16, row23, 3);
     // Add the inverse transform.
-    SaturateAndStore4x4(dst, out01, out23);
+    SaturateAndStore4x4_NEON(dst, out01, out23);
   }
 }
 
 //-----------------------------------------------------------------------------
 // Simple In-loop filtering (Paragraph 15.2)
 
-static uint8x16_t NeedsFilter(const uint8x16_t p1, const uint8x16_t p0,
-                              const uint8x16_t q0, const uint8x16_t q1,
-                              int thresh) {
+static uint8x16_t NeedsFilter_NEON(const uint8x16_t p1, const uint8x16_t p0,
+                                   const uint8x16_t q0, const uint8x16_t q1,
+                                   int thresh) {
   const uint8x16_t thresh_v = vdupq_n_u8((uint8_t)thresh);
   const uint8x16_t a_p0_q0 = vabdq_u8(p0, q0);               // abs(p0-q0)
   const uint8x16_t a_p1_q1 = vabdq_u8(p1, q1);               // abs(p1-q1)
@@ -450,18 +437,18 @@ static uint8x16_t NeedsFilter(const uint8x16_t p1, const uint8x16_t p0,
   return mask;
 }
 
-static int8x16_t FlipSign(const uint8x16_t v) {
+static int8x16_t FlipSign_NEON(const uint8x16_t v) {
   const uint8x16_t sign_bit = vdupq_n_u8(0x80);
   return vreinterpretq_s8_u8(veorq_u8(v, sign_bit));
 }
 
-static uint8x16_t FlipSignBack(const int8x16_t v) {
+static uint8x16_t FlipSignBack_NEON(const int8x16_t v) {
   const int8x16_t sign_bit = vdupq_n_s8(0x80);
   return vreinterpretq_u8_s8(veorq_s8(v, sign_bit));
 }
 
-static int8x16_t GetBaseDelta(const int8x16_t p1, const int8x16_t p0,
-                              const int8x16_t q0, const int8x16_t q1) {
+static int8x16_t GetBaseDelta_NEON(const int8x16_t p1, const int8x16_t p0,
+                                   const int8x16_t q0, const int8x16_t q1) {
   const int8x16_t q0_p0 = vqsubq_s8(q0, p0);      // (q0-p0)
   const int8x16_t p1_q1 = vqsubq_s8(p1, q1);      // (p1-q1)
   const int8x16_t s1 = vqaddq_s8(p1_q1, q0_p0);   // (p1-q1) + 1 * (q0 - p0)
@@ -470,7 +457,7 @@ static int8x16_t GetBaseDelta(const int8x16_t p1, const int8x16_t p0,
   return s3;
 }
 
-static int8x16_t GetBaseDelta0(const int8x16_t p0, const int8x16_t q0) {
+static int8x16_t GetBaseDelta0_NEON(const int8x16_t p0, const int8x16_t q0) {
   const int8x16_t q0_p0 = vqsubq_s8(q0, p0);      // (q0-p0)
   const int8x16_t s1 = vqaddq_s8(q0_p0, q0_p0);   // 2 * (q0 - p0)
   const int8x16_t s2 = vqaddq_s8(q0_p0, s1);      // 3 * (q0 - p0)
@@ -479,9 +466,10 @@ static int8x16_t GetBaseDelta0(const int8x16_t p0, const int8x16_t q0) {
 
 //------------------------------------------------------------------------------
 
-static void ApplyFilter2NoFlip(const int8x16_t p0s, const int8x16_t q0s,
-                               const int8x16_t delta,
-                               int8x16_t* const op0, int8x16_t* const oq0) {
+static void ApplyFilter2NoFlip_NEON(const int8x16_t p0s, const int8x16_t q0s,
+                                    const int8x16_t delta,
+                                    int8x16_t* const op0,
+                                    int8x16_t* const oq0) {
   const int8x16_t kCst3 = vdupq_n_s8(0x03);
   const int8x16_t kCst4 = vdupq_n_s8(0x04);
   const int8x16_t delta_p3 = vqaddq_s8(delta, kCst3);
@@ -494,9 +482,9 @@ static void ApplyFilter2NoFlip(const int8x16_t p0s, const int8x16_t q0s,
 
 #if defined(WEBP_USE_INTRINSICS)
 
-static void ApplyFilter2(const int8x16_t p0s, const int8x16_t q0s,
-                         const int8x16_t delta,
-                         uint8x16_t* const op0, uint8x16_t* const oq0) {
+static void ApplyFilter2_NEON(const int8x16_t p0s, const int8x16_t q0s,
+                              const int8x16_t delta,
+                              uint8x16_t* const op0, uint8x16_t* const oq0) {
   const int8x16_t kCst3 = vdupq_n_s8(0x03);
   const int8x16_t kCst4 = vdupq_n_s8(0x04);
   const int8x16_t delta_p3 = vqaddq_s8(delta, kCst3);
@@ -505,45 +493,66 @@ static void ApplyFilter2(const int8x16_t p0s, const int8x16_t q0s,
   const int8x16_t delta4 = vshrq_n_s8(delta_p4, 3);
   const int8x16_t sp0 = vqaddq_s8(p0s, delta3);
   const int8x16_t sq0 = vqsubq_s8(q0s, delta4);
-  *op0 = FlipSignBack(sp0);
-  *oq0 = FlipSignBack(sq0);
-}
-
-static void DoFilter2(const uint8x16_t p1, const uint8x16_t p0,
-                      const uint8x16_t q0, const uint8x16_t q1,
-                      const uint8x16_t mask,
-                      uint8x16_t* const op0, uint8x16_t* const oq0) {
-  const int8x16_t p1s = FlipSign(p1);
-  const int8x16_t p0s = FlipSign(p0);
-  const int8x16_t q0s = FlipSign(q0);
-  const int8x16_t q1s = FlipSign(q1);
-  const int8x16_t delta0 = GetBaseDelta(p1s, p0s, q0s, q1s);
+  *op0 = FlipSignBack_NEON(sp0);
+  *oq0 = FlipSignBack_NEON(sq0);
+}
+
+static void DoFilter2_NEON(const uint8x16_t p1, const uint8x16_t p0,
+                           const uint8x16_t q0, const uint8x16_t q1,
+                           const uint8x16_t mask,
+                           uint8x16_t* const op0, uint8x16_t* const oq0) {
+  const int8x16_t p1s = FlipSign_NEON(p1);
+  const int8x16_t p0s = FlipSign_NEON(p0);
+  const int8x16_t q0s = FlipSign_NEON(q0);
+  const int8x16_t q1s = FlipSign_NEON(q1);
+  const int8x16_t delta0 = GetBaseDelta_NEON(p1s, p0s, q0s, q1s);
   const int8x16_t delta1 = vandq_s8(delta0, vreinterpretq_s8_u8(mask));
-  ApplyFilter2(p0s, q0s, delta1, op0, oq0);
+  ApplyFilter2_NEON(p0s, q0s, delta1, op0, oq0);
 }
 
-static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16_NEON(uint8_t* p, int stride, int thresh) {
   uint8x16_t p1, p0, q0, q1, op0, oq0;
-  Load16x4(p, stride, &p1, &p0, &q0, &q1);
+  Load16x4_NEON(p, stride, &p1, &p0, &q0, &q1);
   {
-    const uint8x16_t mask = NeedsFilter(p1, p0, q0, q1, thresh);
-    DoFilter2(p1, p0, q0, q1, mask, &op0, &oq0);
+    const uint8x16_t mask = NeedsFilter_NEON(p1, p0, q0, q1, thresh);
+    DoFilter2_NEON(p1, p0, q0, q1, mask, &op0, &oq0);
   }
-  Store16x2(op0, oq0, p, stride);
+  Store16x2_NEON(op0, oq0, p, stride);
 }
 
-static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16_NEON(uint8_t* p, int stride, int thresh) {
   uint8x16_t p1, p0, q0, q1, oq0, op0;
-  Load4x16(p, stride, &p1, &p0, &q0, &q1);
+  Load4x16_NEON(p, stride, &p1, &p0, &q0, &q1);
   {
-    const uint8x16_t mask = NeedsFilter(p1, p0, q0, q1, thresh);
-    DoFilter2(p1, p0, q0, q1, mask, &op0, &oq0);
+    const uint8x16_t mask = NeedsFilter_NEON(p1, p0, q0, q1, thresh);
+    DoFilter2_NEON(p1, p0, q0, q1, mask, &op0, &oq0);
   }
-  Store2x16(op0, oq0, p, stride);
+  Store2x16_NEON(op0, oq0, p, stride);
 }
 
 #else
 
+// Load/Store vertical edge
+#define LOAD8x4(c1, c2, c3, c4, b1, b2, stride)                                \
+  "vld4.8 {" #c1 "[0]," #c2 "[0]," #c3 "[0]," #c4 "[0]}," #b1 "," #stride "\n" \
+  "vld4.8 {" #c1 "[1]," #c2 "[1]," #c3 "[1]," #c4 "[1]}," #b2 "," #stride "\n" \
+  "vld4.8 {" #c1 "[2]," #c2 "[2]," #c3 "[2]," #c4 "[2]}," #b1 "," #stride "\n" \
+  "vld4.8 {" #c1 "[3]," #c2 "[3]," #c3 "[3]," #c4 "[3]}," #b2 "," #stride "\n" \
+  "vld4.8 {" #c1 "[4]," #c2 "[4]," #c3 "[4]," #c4 "[4]}," #b1 "," #stride "\n" \
+  "vld4.8 {" #c1 "[5]," #c2 "[5]," #c3 "[5]," #c4 "[5]}," #b2 "," #stride "\n" \
+  "vld4.8 {" #c1 "[6]," #c2 "[6]," #c3 "[6]," #c4 "[6]}," #b1 "," #stride "\n" \
+  "vld4.8 {" #c1 "[7]," #c2 "[7]," #c3 "[7]," #c4 "[7]}," #b2 "," #stride "\n"
+
+#define STORE8x2(c1, c2, p, stride)                                            \
+  "vst2.8   {" #c1 "[0], " #c2 "[0]}," #p "," #stride " \n"                    \
+  "vst2.8   {" #c1 "[1], " #c2 "[1]}," #p "," #stride " \n"                    \
+  "vst2.8   {" #c1 "[2], " #c2 "[2]}," #p "," #stride " \n"                    \
+  "vst2.8   {" #c1 "[3], " #c2 "[3]}," #p "," #stride " \n"                    \
+  "vst2.8   {" #c1 "[4], " #c2 "[4]}," #p "," #stride " \n"                    \
+  "vst2.8   {" #c1 "[5], " #c2 "[5]}," #p "," #stride " \n"                    \
+  "vst2.8   {" #c1 "[6], " #c2 "[6]}," #p "," #stride " \n"                    \
+  "vst2.8   {" #c1 "[7], " #c2 "[7]}," #p "," #stride " \n"
+
 #define QRegs "q0", "q1", "q2", "q3",                                          \
               "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
 
@@ -592,7 +601,7 @@ static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
   DO_SIMPLE_FILTER(p0, q0, q9)                 /* apply filter */              \
   FLIP_SIGN_BIT2(p0, q0, q10)
 
-static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16_NEON(uint8_t* p, int stride, int thresh) {
   __asm__ volatile (
     "sub        %[p], %[p], %[stride], lsl #1  \n"  // p -= 2 * stride
 
@@ -613,7 +622,7 @@ static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
   );
 }
 
-static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16_NEON(uint8_t* p, int stride, int thresh) {
   __asm__ volatile (
     "sub        r4, %[p], #2                   \n"  // base1 = p - 2
     "lsl        r6, %[stride], #1              \n"  // r6 = 2 * stride
@@ -639,30 +648,33 @@ static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
   );
 }
 
+#undef LOAD8x4
+#undef STORE8x2
+
 #endif    // WEBP_USE_INTRINSICS
 
-static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16i_NEON(uint8_t* p, int stride, int thresh) {
   uint32_t k;
   for (k = 3; k != 0; --k) {
     p += 4 * stride;
-    SimpleVFilter16(p, stride, thresh);
+    SimpleVFilter16_NEON(p, stride, thresh);
   }
 }
 
-static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16i_NEON(uint8_t* p, int stride, int thresh) {
   uint32_t k;
   for (k = 3; k != 0; --k) {
     p += 4;
-    SimpleHFilter16(p, stride, thresh);
+    SimpleHFilter16_NEON(p, stride, thresh);
   }
 }
 
 //------------------------------------------------------------------------------
 // Complex In-loop filtering (Paragraph 15.3)
 
-static uint8x16_t NeedsHev(const uint8x16_t p1, const uint8x16_t p0,
-                           const uint8x16_t q0, const uint8x16_t q1,
-                           int hev_thresh) {
+static uint8x16_t NeedsHev_NEON(const uint8x16_t p1, const uint8x16_t p0,
+                                const uint8x16_t q0, const uint8x16_t q1,
+                                int hev_thresh) {
   const uint8x16_t hev_thresh_v = vdupq_n_u8((uint8_t)hev_thresh);
   const uint8x16_t a_p1_p0 = vabdq_u8(p1, p0);  // abs(p1 - p0)
   const uint8x16_t a_q1_q0 = vabdq_u8(q1, q0);  // abs(q1 - q0)
@@ -671,11 +683,11 @@ static uint8x16_t NeedsHev(const uint8x16_t p1, const uint8x16_t p0,
   return mask;
 }
 
-static uint8x16_t NeedsFilter2(const uint8x16_t p3, const uint8x16_t p2,
-                               const uint8x16_t p1, const uint8x16_t p0,
-                               const uint8x16_t q0, const uint8x16_t q1,
-                               const uint8x16_t q2, const uint8x16_t q3,
-                               int ithresh, int thresh) {
+static uint8x16_t NeedsFilter2_NEON(const uint8x16_t p3, const uint8x16_t p2,
+                                    const uint8x16_t p1, const uint8x16_t p0,
+                                    const uint8x16_t q0, const uint8x16_t q1,
+                                    const uint8x16_t q2, const uint8x16_t q3,
+                                    int ithresh, int thresh) {
   const uint8x16_t ithresh_v = vdupq_n_u8((uint8_t)ithresh);
   const uint8x16_t a_p3_p2 = vabdq_u8(p3, p2);  // abs(p3 - p2)
   const uint8x16_t a_p2_p1 = vabdq_u8(p2, p1);  // abs(p2 - p1)
@@ -689,14 +701,14 @@ static uint8x16_t NeedsFilter2(const uint8x16_t p3, const uint8x16_t p2,
   const uint8x16_t max12 = vmaxq_u8(max1, max2);
   const uint8x16_t max123 = vmaxq_u8(max12, max3);
   const uint8x16_t mask2 = vcgeq_u8(ithresh_v, max123);
-  const uint8x16_t mask1 = NeedsFilter(p1, p0, q0, q1, thresh);
+  const uint8x16_t mask1 = NeedsFilter_NEON(p1, p0, q0, q1, thresh);
   const uint8x16_t mask = vandq_u8(mask1, mask2);
   return mask;
 }
 
 //  4-points filter
 
-static void ApplyFilter4(
+static void ApplyFilter4_NEON(
     const int8x16_t p1, const int8x16_t p0,
     const int8x16_t q0, const int8x16_t q1,
     const int8x16_t delta0,
@@ -709,47 +721,47 @@ static void ApplyFilter4(
   const int8x16_t a1 = vshrq_n_s8(delta1, 3);
   const int8x16_t a2 = vshrq_n_s8(delta2, 3);
   const int8x16_t a3 = vrshrq_n_s8(a1, 1);   // a3 = (a1 + 1) >> 1
-  *op0 = FlipSignBack(vqaddq_s8(p0, a2));  // clip(p0 + a2)
-  *oq0 = FlipSignBack(vqsubq_s8(q0, a1));  // clip(q0 - a1)
-  *op1 = FlipSignBack(vqaddq_s8(p1, a3));  // clip(p1 + a3)
-  *oq1 = FlipSignBack(vqsubq_s8(q1, a3));  // clip(q1 - a3)
+  *op0 = FlipSignBack_NEON(vqaddq_s8(p0, a2));  // clip(p0 + a2)
+  *oq0 = FlipSignBack_NEON(vqsubq_s8(q0, a1));  // clip(q0 - a1)
+  *op1 = FlipSignBack_NEON(vqaddq_s8(p1, a3));  // clip(p1 + a3)
+  *oq1 = FlipSignBack_NEON(vqsubq_s8(q1, a3));  // clip(q1 - a3)
 }
 
-static void DoFilter4(
+static void DoFilter4_NEON(
     const uint8x16_t p1, const uint8x16_t p0,
     const uint8x16_t q0, const uint8x16_t q1,
     const uint8x16_t mask, const uint8x16_t hev_mask,
     uint8x16_t* const op1, uint8x16_t* const op0,
     uint8x16_t* const oq0, uint8x16_t* const oq1) {
   // This is a fused version of DoFilter2() calling ApplyFilter2 directly
-  const int8x16_t p1s = FlipSign(p1);
-  int8x16_t p0s = FlipSign(p0);
-  int8x16_t q0s = FlipSign(q0);
-  const int8x16_t q1s = FlipSign(q1);
+  const int8x16_t p1s = FlipSign_NEON(p1);
+  int8x16_t p0s = FlipSign_NEON(p0);
+  int8x16_t q0s = FlipSign_NEON(q0);
+  const int8x16_t q1s = FlipSign_NEON(q1);
   const uint8x16_t simple_lf_mask = vandq_u8(mask, hev_mask);
 
   // do_filter2 part (simple loopfilter on pixels with hev)
   {
-    const int8x16_t delta = GetBaseDelta(p1s, p0s, q0s, q1s);
+    const int8x16_t delta = GetBaseDelta_NEON(p1s, p0s, q0s, q1s);
     const int8x16_t simple_lf_delta =
         vandq_s8(delta, vreinterpretq_s8_u8(simple_lf_mask));
-    ApplyFilter2NoFlip(p0s, q0s, simple_lf_delta, &p0s, &q0s);
+    ApplyFilter2NoFlip_NEON(p0s, q0s, simple_lf_delta, &p0s, &q0s);
   }
 
   // do_filter4 part (complex loopfilter on pixels without hev)
   {
-    const int8x16_t delta0 = GetBaseDelta0(p0s, q0s);
+    const int8x16_t delta0 = GetBaseDelta0_NEON(p0s, q0s);
     // we use: (mask & hev_mask) ^ mask = mask & !hev_mask
     const uint8x16_t complex_lf_mask = veorq_u8(simple_lf_mask, mask);
     const int8x16_t complex_lf_delta =
         vandq_s8(delta0, vreinterpretq_s8_u8(complex_lf_mask));
-    ApplyFilter4(p1s, p0s, q0s, q1s, complex_lf_delta, op1, op0, oq0, oq1);
+    ApplyFilter4_NEON(p1s, p0s, q0s, q1s, complex_lf_delta, op1, op0, oq0, oq1);
   }
 }
 
 //  6-points filter
 
-static void ApplyFilter6(
+static void ApplyFilter6_NEON(
     const int8x16_t p2, const int8x16_t p1, const int8x16_t p0,
     const int8x16_t q0, const int8x16_t q1, const int8x16_t q2,
     const int8x16_t delta,
@@ -778,35 +790,35 @@ static void ApplyFilter6(
   const int8x16_t a2 = vcombine_s8(a2_lo, a2_hi);
   const int8x16_t a3 = vcombine_s8(a3_lo, a3_hi);
 
-  *op0 = FlipSignBack(vqaddq_s8(p0, a1));  // clip(p0 + a1)
-  *oq0 = FlipSignBack(vqsubq_s8(q0, a1));  // clip(q0 - q1)
-  *oq1 = FlipSignBack(vqsubq_s8(q1, a2));  // clip(q1 - a2)
-  *op1 = FlipSignBack(vqaddq_s8(p1, a2));  // clip(p1 + a2)
-  *oq2 = FlipSignBack(vqsubq_s8(q2, a3));  // clip(q2 - a3)
-  *op2 = FlipSignBack(vqaddq_s8(p2, a3));  // clip(p2 + a3)
+  *op0 = FlipSignBack_NEON(vqaddq_s8(p0, a1));  // clip(p0 + a1)
+  *oq0 = FlipSignBack_NEON(vqsubq_s8(q0, a1));  // clip(q0 - q1)
+  *oq1 = FlipSignBack_NEON(vqsubq_s8(q1, a2));  // clip(q1 - a2)
+  *op1 = FlipSignBack_NEON(vqaddq_s8(p1, a2));  // clip(p1 + a2)
+  *oq2 = FlipSignBack_NEON(vqsubq_s8(q2, a3));  // clip(q2 - a3)
+  *op2 = FlipSignBack_NEON(vqaddq_s8(p2, a3));  // clip(p2 + a3)
 }
 
-static void DoFilter6(
+static void DoFilter6_NEON(
     const uint8x16_t p2, const uint8x16_t p1, const uint8x16_t p0,
     const uint8x16_t q0, const uint8x16_t q1, const uint8x16_t q2,
     const uint8x16_t mask, const uint8x16_t hev_mask,
     uint8x16_t* const op2, uint8x16_t* const op1, uint8x16_t* const op0,
     uint8x16_t* const oq0, uint8x16_t* const oq1, uint8x16_t* const oq2) {
   // This is a fused version of DoFilter2() calling ApplyFilter2 directly
-  const int8x16_t p2s = FlipSign(p2);
-  const int8x16_t p1s = FlipSign(p1);
-  int8x16_t p0s = FlipSign(p0);
-  int8x16_t q0s = FlipSign(q0);
-  const int8x16_t q1s = FlipSign(q1);
-  const int8x16_t q2s = FlipSign(q2);
+  const int8x16_t p2s = FlipSign_NEON(p2);
+  const int8x16_t p1s = FlipSign_NEON(p1);
+  int8x16_t p0s = FlipSign_NEON(p0);
+  int8x16_t q0s = FlipSign_NEON(q0);
+  const int8x16_t q1s = FlipSign_NEON(q1);
+  const int8x16_t q2s = FlipSign_NEON(q2);
   const uint8x16_t simple_lf_mask = vandq_u8(mask, hev_mask);
-  const int8x16_t delta0 = GetBaseDelta(p1s, p0s, q0s, q1s);
+  const int8x16_t delta0 = GetBaseDelta_NEON(p1s, p0s, q0s, q1s);
 
   // do_filter2 part (simple loopfilter on pixels with hev)
   {
     const int8x16_t simple_lf_delta =
         vandq_s8(delta0, vreinterpretq_s8_u8(simple_lf_mask));
-    ApplyFilter2NoFlip(p0s, q0s, simple_lf_delta, &p0s, &q0s);
+    ApplyFilter2NoFlip_NEON(p0s, q0s, simple_lf_delta, &p0s, &q0s);
   }
 
   // do_filter6 part (complex loopfilter on pixels without hev)
@@ -815,65 +827,65 @@ static void DoFilter6(
     const uint8x16_t complex_lf_mask = veorq_u8(simple_lf_mask, mask);
     const int8x16_t complex_lf_delta =
         vandq_s8(delta0, vreinterpretq_s8_u8(complex_lf_mask));
-    ApplyFilter6(p2s, p1s, p0s, q0s, q1s, q2s, complex_lf_delta,
-                 op2, op1, op0, oq0, oq1, oq2);
+    ApplyFilter6_NEON(p2s, p1s, p0s, q0s, q1s, q2s, complex_lf_delta,
+                      op2, op1, op0, oq0, oq1, oq2);
   }
 }
 
 // on macroblock edges
 
-static void VFilter16(uint8_t* p, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void VFilter16_NEON(uint8_t* p, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   uint8x16_t p3, p2, p1, p0, q0, q1, q2, q3;
-  Load16x8(p, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
+  Load16x8_NEON(p, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
   {
-    const uint8x16_t mask = NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3,
-                                         ithresh, thresh);
-    const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
+    const uint8x16_t mask = NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3,
+                                              ithresh, thresh);
+    const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
     uint8x16_t op2, op1, op0, oq0, oq1, oq2;
-    DoFilter6(p2, p1, p0, q0, q1, q2, mask, hev_mask,
-              &op2, &op1, &op0, &oq0, &oq1, &oq2);
-    Store16x2(op2, op1, p - 2 * stride, stride);
-    Store16x2(op0, oq0, p + 0 * stride, stride);
-    Store16x2(oq1, oq2, p + 2 * stride, stride);
+    DoFilter6_NEON(p2, p1, p0, q0, q1, q2, mask, hev_mask,
+                   &op2, &op1, &op0, &oq0, &oq1, &oq2);
+    Store16x2_NEON(op2, op1, p - 2 * stride, stride);
+    Store16x2_NEON(op0, oq0, p + 0 * stride, stride);
+    Store16x2_NEON(oq1, oq2, p + 2 * stride, stride);
   }
 }
 
-static void HFilter16(uint8_t* p, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void HFilter16_NEON(uint8_t* p, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   uint8x16_t p3, p2, p1, p0, q0, q1, q2, q3;
-  Load8x16(p, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
+  Load8x16_NEON(p, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
   {
-    const uint8x16_t mask = NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3,
-                                         ithresh, thresh);
-    const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
+    const uint8x16_t mask = NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3,
+                                              ithresh, thresh);
+    const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
     uint8x16_t op2, op1, op0, oq0, oq1, oq2;
-    DoFilter6(p2, p1, p0, q0, q1, q2, mask, hev_mask,
-              &op2, &op1, &op0, &oq0, &oq1, &oq2);
-    Store2x16(op2, op1, p - 2, stride);
-    Store2x16(op0, oq0, p + 0, stride);
-    Store2x16(oq1, oq2, p + 2, stride);
+    DoFilter6_NEON(p2, p1, p0, q0, q1, q2, mask, hev_mask,
+                   &op2, &op1, &op0, &oq0, &oq1, &oq2);
+    Store2x16_NEON(op2, op1, p - 2, stride);
+    Store2x16_NEON(op0, oq0, p + 0, stride);
+    Store2x16_NEON(oq1, oq2, p + 2, stride);
   }
 }
 
 // on three inner edges
-static void VFilter16i(uint8_t* p, int stride,
-                       int thresh, int ithresh, int hev_thresh) {
+static void VFilter16i_NEON(uint8_t* p, int stride,
+                            int thresh, int ithresh, int hev_thresh) {
   uint32_t k;
   uint8x16_t p3, p2, p1, p0;
-  Load16x4(p + 2  * stride, stride, &p3, &p2, &p1, &p0);
+  Load16x4_NEON(p + 2  * stride, stride, &p3, &p2, &p1, &p0);
   for (k = 3; k != 0; --k) {
     uint8x16_t q0, q1, q2, q3;
     p += 4 * stride;
-    Load16x4(p + 2  * stride, stride, &q0, &q1, &q2, &q3);
+    Load16x4_NEON(p + 2  * stride, stride, &q0, &q1, &q2, &q3);
     {
       const uint8x16_t mask =
-          NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3, ithresh, thresh);
-      const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
+          NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3, ithresh, thresh);
+      const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
       // p3 and p2 are not just temporary variables here: they will be
       // re-used for next span. And q2/q3 will become p1/p0 accordingly.
-      DoFilter4(p1, p0, q0, q1, mask, hev_mask, &p1, &p0, &p3, &p2);
-      Store16x4(p1, p0, p3, p2, p, stride);
+      DoFilter4_NEON(p1, p0, q0, q1, mask, hev_mask, &p1, &p0, &p3, &p2);
+      Store16x4_NEON(p1, p0, p3, p2, p, stride);
       p1 = q2;
       p0 = q3;
     }
@@ -881,21 +893,21 @@ static void VFilter16i(uint8_t* p, int stride,
 }
 
 #if !defined(WORK_AROUND_GCC)
-static void HFilter16i(uint8_t* p, int stride,
-                       int thresh, int ithresh, int hev_thresh) {
+static void HFilter16i_NEON(uint8_t* p, int stride,
+                            int thresh, int ithresh, int hev_thresh) {
   uint32_t k;
   uint8x16_t p3, p2, p1, p0;
-  Load4x16(p + 2, stride, &p3, &p2, &p1, &p0);
+  Load4x16_NEON(p + 2, stride, &p3, &p2, &p1, &p0);
   for (k = 3; k != 0; --k) {
     uint8x16_t q0, q1, q2, q3;
     p += 4;
-    Load4x16(p + 2, stride, &q0, &q1, &q2, &q3);
+    Load4x16_NEON(p + 2, stride, &q0, &q1, &q2, &q3);
     {
       const uint8x16_t mask =
-          NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3, ithresh, thresh);
-      const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
-      DoFilter4(p1, p0, q0, q1, mask, hev_mask, &p1, &p0, &p3, &p2);
-      Store4x16(p1, p0, p3, p2, p, stride);
+          NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3, ithresh, thresh);
+      const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
+      DoFilter4_NEON(p1, p0, q0, q1, mask, hev_mask, &p1, &p0, &p3, &p2);
+      Store4x16_NEON(p1, p0, p3, p2, p, stride);
       p1 = q2;
       p0 = q3;
     }
@@ -904,67 +916,67 @@ static void HFilter16i(uint8_t* p, int stride,
 #endif  // !WORK_AROUND_GCC
 
 // 8-pixels wide variant, for chroma filtering
-static void VFilter8(uint8_t* u, uint8_t* v, int stride,
-                     int thresh, int ithresh, int hev_thresh) {
+static void VFilter8_NEON(uint8_t* u, uint8_t* v, int stride,
+                          int thresh, int ithresh, int hev_thresh) {
   uint8x16_t p3, p2, p1, p0, q0, q1, q2, q3;
-  Load8x8x2(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
+  Load8x8x2_NEON(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
   {
-    const uint8x16_t mask = NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3,
-                                         ithresh, thresh);
-    const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
+    const uint8x16_t mask = NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3,
+                                              ithresh, thresh);
+    const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
     uint8x16_t op2, op1, op0, oq0, oq1, oq2;
-    DoFilter6(p2, p1, p0, q0, q1, q2, mask, hev_mask,
-              &op2, &op1, &op0, &oq0, &oq1, &oq2);
-    Store8x2x2(op2, op1, u - 2 * stride, v - 2 * stride, stride);
-    Store8x2x2(op0, oq0, u + 0 * stride, v + 0 * stride, stride);
-    Store8x2x2(oq1, oq2, u + 2 * stride, v + 2 * stride, stride);
+    DoFilter6_NEON(p2, p1, p0, q0, q1, q2, mask, hev_mask,
+                   &op2, &op1, &op0, &oq0, &oq1, &oq2);
+    Store8x2x2_NEON(op2, op1, u - 2 * stride, v - 2 * stride, stride);
+    Store8x2x2_NEON(op0, oq0, u + 0 * stride, v + 0 * stride, stride);
+    Store8x2x2_NEON(oq1, oq2, u + 2 * stride, v + 2 * stride, stride);
   }
 }
-static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void VFilter8i_NEON(uint8_t* u, uint8_t* v, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   uint8x16_t p3, p2, p1, p0, q0, q1, q2, q3;
   u += 4 * stride;
   v += 4 * stride;
-  Load8x8x2(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
+  Load8x8x2_NEON(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
   {
-    const uint8x16_t mask = NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3,
-                                         ithresh, thresh);
-    const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
+    const uint8x16_t mask = NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3,
+                                              ithresh, thresh);
+    const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
     uint8x16_t op1, op0, oq0, oq1;
-    DoFilter4(p1, p0, q0, q1, mask, hev_mask, &op1, &op0, &oq0, &oq1);
-    Store8x4x2(op1, op0, oq0, oq1, u, v, stride);
+    DoFilter4_NEON(p1, p0, q0, q1, mask, hev_mask, &op1, &op0, &oq0, &oq1);
+    Store8x4x2_NEON(op1, op0, oq0, oq1, u, v, stride);
   }
 }
 
 #if !defined(WORK_AROUND_GCC)
-static void HFilter8(uint8_t* u, uint8_t* v, int stride,
-                     int thresh, int ithresh, int hev_thresh) {
+static void HFilter8_NEON(uint8_t* u, uint8_t* v, int stride,
+                          int thresh, int ithresh, int hev_thresh) {
   uint8x16_t p3, p2, p1, p0, q0, q1, q2, q3;
-  Load8x8x2T(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
+  Load8x8x2T_NEON(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
   {
-    const uint8x16_t mask = NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3,
-                                         ithresh, thresh);
-    const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
+    const uint8x16_t mask = NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3,
+                                              ithresh, thresh);
+    const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
     uint8x16_t op2, op1, op0, oq0, oq1, oq2;
-    DoFilter6(p2, p1, p0, q0, q1, q2, mask, hev_mask,
-              &op2, &op1, &op0, &oq0, &oq1, &oq2);
-    Store6x8x2(op2, op1, op0, oq0, oq1, oq2, u, v, stride);
+    DoFilter6_NEON(p2, p1, p0, q0, q1, q2, mask, hev_mask,
+                   &op2, &op1, &op0, &oq0, &oq1, &oq2);
+    Store6x8x2_NEON(op2, op1, op0, oq0, oq1, oq2, u, v, stride);
   }
 }
 
-static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void HFilter8i_NEON(uint8_t* u, uint8_t* v, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   uint8x16_t p3, p2, p1, p0, q0, q1, q2, q3;
   u += 4;
   v += 4;
-  Load8x8x2T(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
+  Load8x8x2T_NEON(u, v, stride, &p3, &p2, &p1, &p0, &q0, &q1, &q2, &q3);
   {
-    const uint8x16_t mask = NeedsFilter2(p3, p2, p1, p0, q0, q1, q2, q3,
-                                         ithresh, thresh);
-    const uint8x16_t hev_mask = NeedsHev(p1, p0, q0, q1, hev_thresh);
+    const uint8x16_t mask = NeedsFilter2_NEON(p3, p2, p1, p0, q0, q1, q2, q3,
+                                              ithresh, thresh);
+    const uint8x16_t hev_mask = NeedsHev_NEON(p1, p0, q0, q1, hev_thresh);
     uint8x16_t op1, op0, oq0, oq1;
-    DoFilter4(p1, p0, q0, q1, mask, hev_mask, &op1, &op0, &oq0, &oq1);
-    Store4x8x2(op1, op0, oq0, oq1, u, v, stride);
+    DoFilter4_NEON(p1, p0, q0, q1, mask, hev_mask, &op1, &op0, &oq0, &oq1);
+    Store4x8x2_NEON(op1, op0, oq0, oq1, u, v, stride);
   }
 }
 #endif  // !WORK_AROUND_GCC
@@ -992,8 +1004,9 @@ static const int16_t kC1 = 20091;
 static const int16_t kC2 = 17734;  // half of kC2, actually. See comment above.
 
 #if defined(WEBP_USE_INTRINSICS)
-static WEBP_INLINE void Transpose8x2(const int16x8_t in0, const int16x8_t in1,
-                                     int16x8x2_t* const out) {
+static WEBP_INLINE void Transpose8x2_NEON(const int16x8_t in0,
+                                          const int16x8_t in1,
+                                          int16x8x2_t* const out) {
   // a0 a1 a2 a3 | b0 b1 b2 b3   => a0 b0 c0 d0 | a1 b1 c1 d1
   // c0 c1 c2 c3 | d0 d1 d2 d3      a2 b2 c2 d2 | a3 b3 c3 d3
   const int16x8x2_t tmp0 = vzipq_s16(in0, in1);   // a0 c0 a1 c1 a2 c2 ...
@@ -1001,7 +1014,7 @@ static WEBP_INLINE void Transpose8x2(const int16x8_t in0, const int16x8_t in1,
   *out = vzipq_s16(tmp0.val[0], tmp0.val[1]);
 }
 
-static WEBP_INLINE void TransformPass(int16x8x2_t* const rows) {
+static WEBP_INLINE void TransformPass_NEON(int16x8x2_t* const rows) {
   // {rows} = in0 | in4
   //          in8 | in12
   // B1 = in4 | in12
@@ -1024,20 +1037,20 @@ static WEBP_INLINE void TransformPass(int16x8x2_t* const rows) {
   const int16x8_t E0 = vqaddq_s16(D0, D1);      // a+d | b+c
   const int16x8_t E_tmp = vqsubq_s16(D0, D1);   // a-d | b-c
   const int16x8_t E1 = vcombine_s16(vget_high_s16(E_tmp), vget_low_s16(E_tmp));
-  Transpose8x2(E0, E1, rows);
+  Transpose8x2_NEON(E0, E1, rows);
 }
 
-static void TransformOne(const int16_t* in, uint8_t* dst) {
+static void TransformOne_NEON(const int16_t* in, uint8_t* dst) {
   int16x8x2_t rows;
   INIT_VECTOR2(rows, vld1q_s16(in + 0), vld1q_s16(in + 8));
-  TransformPass(&rows);
-  TransformPass(&rows);
-  Add4x4(rows.val[0], rows.val[1], dst);
+  TransformPass_NEON(&rows);
+  TransformPass_NEON(&rows);
+  Add4x4_NEON(rows.val[0], rows.val[1], dst);
 }
 
 #else
 
-static void TransformOne(const int16_t* in, uint8_t* dst) {
+static void TransformOne_NEON(const int16_t* in, uint8_t* dst) {
   const int kBPS = BPS;
   // kC1, kC2. Padded because vld1.16 loads 8 bytes
   const int16_t constants[4] = { kC1, kC2, 0, 0 };
@@ -1170,16 +1183,16 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
 
 #endif    // WEBP_USE_INTRINSICS
 
-static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
-  TransformOne(in, dst);
+static void TransformTwo_NEON(const int16_t* in, uint8_t* dst, int do_two) {
+  TransformOne_NEON(in, dst);
   if (do_two) {
-    TransformOne(in + 16, dst + 4);
+    TransformOne_NEON(in + 16, dst + 4);
   }
 }
 
-static void TransformDC(const int16_t* in, uint8_t* dst) {
+static void TransformDC_NEON(const int16_t* in, uint8_t* dst) {
   const int16x8_t DC = vdupq_n_s16(in[0]);
-  Add4x4(DC, DC, dst);
+  Add4x4_NEON(DC, DC, dst);
 }
 
 //------------------------------------------------------------------------------
@@ -1191,7 +1204,7 @@ static void TransformDC(const int16_t* in, uint8_t* dst) {
   *dst = vgetq_lane_s32(rows.val[3], col); (dst) += 16; \
 } while (0)
 
-static void TransformWHT(const int16_t* in, int16_t* out) {
+static void TransformWHT_NEON(const int16_t* in, int16_t* out) {
   int32x4x4_t tmp;
 
   {
@@ -1209,7 +1222,7 @@ static void TransformWHT(const int16_t* in, int16_t* out) {
     tmp.val[2] = vsubq_s32(a0, a1);
     tmp.val[3] = vsubq_s32(a3, a2);
     // Arrange the temporary results column-wise.
-    tmp = Transpose4x4(tmp);
+    tmp = Transpose4x4_NEON(tmp);
   }
 
   {
@@ -1243,7 +1256,7 @@ static void TransformWHT(const int16_t* in, int16_t* out) {
 //------------------------------------------------------------------------------
 
 #define MUL(a, b) (((a) * (b)) >> 16)
-static void TransformAC3(const int16_t* in, uint8_t* dst) {
+static void TransformAC3_NEON(const int16_t* in, uint8_t* dst) {
   static const int kC1_full = 20091 + (1 << 16);
   static const int kC2_full = 35468;
   const int16x4_t A = vld1_dup_s16(in);
@@ -1259,14 +1272,14 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
   const int16x4_t B = vqadd_s16(A, CD);
   const int16x8_t m0_m1 = vcombine_s16(vqadd_s16(B, d4), vqadd_s16(B, c4));
   const int16x8_t m2_m3 = vcombine_s16(vqsub_s16(B, c4), vqsub_s16(B, d4));
-  Add4x4(m0_m1, m2_m3, dst);
+  Add4x4_NEON(m0_m1, m2_m3, dst);
 }
 #undef MUL
 
 //------------------------------------------------------------------------------
 // 4x4
 
-static void DC4(uint8_t* dst) {    // DC
+static void DC4_NEON(uint8_t* dst) {    // DC
   const uint8x8_t A = vld1_u8(dst - BPS);  // top row
   const uint16x4_t p0 = vpaddl_u8(A);  // cascading summation of the top
   const uint16x4_t p1 = vpadd_u16(p0, p0);
@@ -1287,17 +1300,17 @@ static void DC4(uint8_t* dst) {    // DC
 }
 
 // TrueMotion (4x4 + 8x8)
-static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
+static WEBP_INLINE void TrueMotion_NEON(uint8_t* dst, int size) {
   const uint8x8_t TL = vld1_dup_u8(dst - BPS - 1);  // top-left pixel 'A[-1]'
   const uint8x8_t T = vld1_u8(dst - BPS);  // top row 'A[0..3]'
   const int16x8_t d = vreinterpretq_s16_u16(vsubl_u8(T, TL));  // A[c] - A[-1]
   int y;
   for (y = 0; y < size; y += 4) {
     // left edge
-    const int16x8_t L0 = ConvertU8ToS16(vld1_dup_u8(dst + 0 * BPS - 1));
-    const int16x8_t L1 = ConvertU8ToS16(vld1_dup_u8(dst + 1 * BPS - 1));
-    const int16x8_t L2 = ConvertU8ToS16(vld1_dup_u8(dst + 2 * BPS - 1));
-    const int16x8_t L3 = ConvertU8ToS16(vld1_dup_u8(dst + 3 * BPS - 1));
+    const int16x8_t L0 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 0 * BPS - 1));
+    const int16x8_t L1 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 1 * BPS - 1));
+    const int16x8_t L2 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 2 * BPS - 1));
+    const int16x8_t L3 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 3 * BPS - 1));
     const int16x8_t r0 = vaddq_s16(L0, d);  // L[r] + A[c] - A[-1]
     const int16x8_t r1 = vaddq_s16(L1, d);
     const int16x8_t r2 = vaddq_s16(L2, d);
@@ -1322,9 +1335,9 @@ static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
   }
 }
 
-static void TM4(uint8_t* dst) { TrueMotion(dst, 4); }
+static void TM4_NEON(uint8_t* dst) { TrueMotion_NEON(dst, 4); }
 
-static void VE4(uint8_t* dst) {    // vertical
+static void VE4_NEON(uint8_t* dst) {    // vertical
   // NB: avoid vld1_u64 here as an alignment hint may be added -> SIGBUS.
   const uint64x1_t A0 = vreinterpret_u64_u8(vld1_u8(dst - BPS - 1));  // top row
   const uint64x1_t A1 = vshr_n_u64(A0, 8);
@@ -1340,7 +1353,7 @@ static void VE4(uint8_t* dst) {    // vertical
   }
 }
 
-static void RD4(uint8_t* dst) {   // Down-right
+static void RD4_NEON(uint8_t* dst) {   // Down-right
   const uint8x8_t XABCD_u8 = vld1_u8(dst - BPS - 1);
   const uint64x1_t XABCD = vreinterpret_u64_u8(XABCD_u8);
   const uint64x1_t ____XABC = vshl_n_u64(XABCD, 32);
@@ -1368,7 +1381,7 @@ static void RD4(uint8_t* dst) {   // Down-right
   vst1_lane_u32((uint32_t*)(dst + 3 * BPS), r3, 0);
 }
 
-static void LD4(uint8_t* dst) {    // Down-left
+static void LD4_NEON(uint8_t* dst) {    // Down-left
   // Note using the same shift trick as VE4() is slower here.
   const uint8x8_t ABCDEFGH = vld1_u8(dst - BPS + 0);
   const uint8x8_t BCDEFGH0 = vld1_u8(dst - BPS + 1);
@@ -1390,7 +1403,7 @@ static void LD4(uint8_t* dst) {    // Down-left
 //------------------------------------------------------------------------------
 // Chroma
 
-static void VE8uv(uint8_t* dst) {    // vertical
+static void VE8uv_NEON(uint8_t* dst) {    // vertical
   const uint8x8_t top = vld1_u8(dst - BPS);
   int j;
   for (j = 0; j < 8; ++j) {
@@ -1398,7 +1411,7 @@ static void VE8uv(uint8_t* dst) {    // vertical
   }
 }
 
-static void HE8uv(uint8_t* dst) {    // horizontal
+static void HE8uv_NEON(uint8_t* dst) {    // horizontal
   int j;
   for (j = 0; j < 8; ++j) {
     const uint8x8_t left = vld1_dup_u8(dst - 1);
@@ -1407,7 +1420,7 @@ static void HE8uv(uint8_t* dst) {    // horizontal
   }
 }
 
-static WEBP_INLINE void DC8(uint8_t* dst, int do_top, int do_left) {
+static WEBP_INLINE void DC8_NEON(uint8_t* dst, int do_top, int do_left) {
   uint16x8_t sum_top;
   uint16x8_t sum_left;
   uint8x8_t dc0;
@@ -1458,17 +1471,17 @@ static WEBP_INLINE void DC8(uint8_t* dst, int do_top, int do_left) {
   }
 }
 
-static void DC8uv(uint8_t* dst) { DC8(dst, 1, 1); }
-static void DC8uvNoTop(uint8_t* dst) { DC8(dst, 0, 1); }
-static void DC8uvNoLeft(uint8_t* dst) { DC8(dst, 1, 0); }
-static void DC8uvNoTopLeft(uint8_t* dst) { DC8(dst, 0, 0); }
+static void DC8uv_NEON(uint8_t* dst) { DC8_NEON(dst, 1, 1); }
+static void DC8uvNoTop_NEON(uint8_t* dst) { DC8_NEON(dst, 0, 1); }
+static void DC8uvNoLeft_NEON(uint8_t* dst) { DC8_NEON(dst, 1, 0); }
+static void DC8uvNoTopLeft_NEON(uint8_t* dst) { DC8_NEON(dst, 0, 0); }
 
-static void TM8uv(uint8_t* dst) { TrueMotion(dst, 8); }
+static void TM8uv_NEON(uint8_t* dst) { TrueMotion_NEON(dst, 8); }
 
 //------------------------------------------------------------------------------
 // 16x16
 
-static void VE16(uint8_t* dst) {     // vertical
+static void VE16_NEON(uint8_t* dst) {     // vertical
   const uint8x16_t top = vld1q_u8(dst - BPS);
   int j;
   for (j = 0; j < 16; ++j) {
@@ -1476,7 +1489,7 @@ static void VE16(uint8_t* dst) {     // vertical
   }
 }
 
-static void HE16(uint8_t* dst) {     // horizontal
+static void HE16_NEON(uint8_t* dst) {     // horizontal
   int j;
   for (j = 0; j < 16; ++j) {
     const uint8x16_t left = vld1q_dup_u8(dst - 1);
@@ -1485,7 +1498,7 @@ static void HE16(uint8_t* dst) {     // horizontal
   }
 }
 
-static WEBP_INLINE void DC16(uint8_t* dst, int do_top, int do_left) {
+static WEBP_INLINE void DC16_NEON(uint8_t* dst, int do_top, int do_left) {
   uint16x8_t sum_top;
   uint16x8_t sum_left;
   uint8x8_t dc0;
@@ -1542,12 +1555,12 @@ static WEBP_INLINE void DC16(uint8_t* dst, int do_top, int do_left) {
   }
 }
 
-static void DC16TopLeft(uint8_t* dst) { DC16(dst, 1, 1); }
-static void DC16NoTop(uint8_t* dst) { DC16(dst, 0, 1); }
-static void DC16NoLeft(uint8_t* dst) { DC16(dst, 1, 0); }
-static void DC16NoTopLeft(uint8_t* dst) { DC16(dst, 0, 0); }
+static void DC16TopLeft_NEON(uint8_t* dst) { DC16_NEON(dst, 1, 1); }
+static void DC16NoTop_NEON(uint8_t* dst) { DC16_NEON(dst, 0, 1); }
+static void DC16NoLeft_NEON(uint8_t* dst) { DC16_NEON(dst, 1, 0); }
+static void DC16NoTopLeft_NEON(uint8_t* dst) { DC16_NEON(dst, 0, 0); }
 
-static void TM16(uint8_t* dst) {
+static void TM16_NEON(uint8_t* dst) {
   const uint8x8_t TL = vld1_dup_u8(dst - BPS - 1);  // top-left pixel 'A[-1]'
   const uint8x16_t T = vld1q_u8(dst - BPS);  // top row 'A[0..15]'
   // A[c] - A[-1]
@@ -1556,10 +1569,10 @@ static void TM16(uint8_t* dst) {
   int y;
   for (y = 0; y < 16; y += 4) {
     // left edge
-    const int16x8_t L0 = ConvertU8ToS16(vld1_dup_u8(dst + 0 * BPS - 1));
-    const int16x8_t L1 = ConvertU8ToS16(vld1_dup_u8(dst + 1 * BPS - 1));
-    const int16x8_t L2 = ConvertU8ToS16(vld1_dup_u8(dst + 2 * BPS - 1));
-    const int16x8_t L3 = ConvertU8ToS16(vld1_dup_u8(dst + 3 * BPS - 1));
+    const int16x8_t L0 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 0 * BPS - 1));
+    const int16x8_t L1 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 1 * BPS - 1));
+    const int16x8_t L2 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 2 * BPS - 1));
+    const int16x8_t L3 = ConvertU8ToS16_NEON(vld1_dup_u8(dst + 3 * BPS - 1));
     const int16x8_t r0_lo = vaddq_s16(L0, d_lo);  // L[r] + A[c] - A[-1]
     const int16x8_t r1_lo = vaddq_s16(L1, d_lo);
     const int16x8_t r2_lo = vaddq_s16(L2, d_lo);
@@ -1587,49 +1600,49 @@ static void TM16(uint8_t* dst) {
 extern void VP8DspInitNEON(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8DspInitNEON(void) {
-  VP8Transform = TransformTwo;
-  VP8TransformAC3 = TransformAC3;
-  VP8TransformDC = TransformDC;
-  VP8TransformWHT = TransformWHT;
-
-  VP8VFilter16 = VFilter16;
-  VP8VFilter16i = VFilter16i;
-  VP8HFilter16 = HFilter16;
+  VP8Transform = TransformTwo_NEON;
+  VP8TransformAC3 = TransformAC3_NEON;
+  VP8TransformDC = TransformDC_NEON;
+  VP8TransformWHT = TransformWHT_NEON;
+
+  VP8VFilter16 = VFilter16_NEON;
+  VP8VFilter16i = VFilter16i_NEON;
+  VP8HFilter16 = HFilter16_NEON;
 #if !defined(WORK_AROUND_GCC)
-  VP8HFilter16i = HFilter16i;
+  VP8HFilter16i = HFilter16i_NEON;
 #endif
-  VP8VFilter8 = VFilter8;
-  VP8VFilter8i = VFilter8i;
+  VP8VFilter8 = VFilter8_NEON;
+  VP8VFilter8i = VFilter8i_NEON;
 #if !defined(WORK_AROUND_GCC)
-  VP8HFilter8 = HFilter8;
-  VP8HFilter8i = HFilter8i;
+  VP8HFilter8 = HFilter8_NEON;
+  VP8HFilter8i = HFilter8i_NEON;
 #endif
-  VP8SimpleVFilter16 = SimpleVFilter16;
-  VP8SimpleHFilter16 = SimpleHFilter16;
-  VP8SimpleVFilter16i = SimpleVFilter16i;
-  VP8SimpleHFilter16i = SimpleHFilter16i;
-
-  VP8PredLuma4[0] = DC4;
-  VP8PredLuma4[1] = TM4;
-  VP8PredLuma4[2] = VE4;
-  VP8PredLuma4[4] = RD4;
-  VP8PredLuma4[6] = LD4;
-
-  VP8PredLuma16[0] = DC16TopLeft;
-  VP8PredLuma16[1] = TM16;
-  VP8PredLuma16[2] = VE16;
-  VP8PredLuma16[3] = HE16;
-  VP8PredLuma16[4] = DC16NoTop;
-  VP8PredLuma16[5] = DC16NoLeft;
-  VP8PredLuma16[6] = DC16NoTopLeft;
-
-  VP8PredChroma8[0] = DC8uv;
-  VP8PredChroma8[1] = TM8uv;
-  VP8PredChroma8[2] = VE8uv;
-  VP8PredChroma8[3] = HE8uv;
-  VP8PredChroma8[4] = DC8uvNoTop;
-  VP8PredChroma8[5] = DC8uvNoLeft;
-  VP8PredChroma8[6] = DC8uvNoTopLeft;
+  VP8SimpleVFilter16 = SimpleVFilter16_NEON;
+  VP8SimpleHFilter16 = SimpleHFilter16_NEON;
+  VP8SimpleVFilter16i = SimpleVFilter16i_NEON;
+  VP8SimpleHFilter16i = SimpleHFilter16i_NEON;
+
+  VP8PredLuma4[0] = DC4_NEON;
+  VP8PredLuma4[1] = TM4_NEON;
+  VP8PredLuma4[2] = VE4_NEON;
+  VP8PredLuma4[4] = RD4_NEON;
+  VP8PredLuma4[6] = LD4_NEON;
+
+  VP8PredLuma16[0] = DC16TopLeft_NEON;
+  VP8PredLuma16[1] = TM16_NEON;
+  VP8PredLuma16[2] = VE16_NEON;
+  VP8PredLuma16[3] = HE16_NEON;
+  VP8PredLuma16[4] = DC16NoTop_NEON;
+  VP8PredLuma16[5] = DC16NoLeft_NEON;
+  VP8PredLuma16[6] = DC16NoTopLeft_NEON;
+
+  VP8PredChroma8[0] = DC8uv_NEON;
+  VP8PredChroma8[1] = TM8uv_NEON;
+  VP8PredChroma8[2] = VE8uv_NEON;
+  VP8PredChroma8[3] = HE8uv_NEON;
+  VP8PredChroma8[4] = DC8uvNoTop_NEON;
+  VP8PredChroma8[5] = DC8uvNoLeft_NEON;
+  VP8PredChroma8[6] = DC8uvNoTopLeft_NEON;
 }
 
 #else  // !WEBP_USE_NEON
diff --git a/src/3rdparty/libwebp/src/dsp/dec_sse2.c b/src/3rdparty/libwebp/src/dsp/dec_sse2.c
index 411fb02..b3840fa 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_sse2.c
@@ -12,23 +12,25 @@
 // Author: somnath@google.com (Somnath Banerjee)
 //         cduvivier@google.com (Christian Duvivier)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 
 // The 3-coeff sparse transform in SSE2 is not really faster than the plain-C
 // one it seems => disable it by default. Uncomment the following to enable:
-// #define USE_TRANSFORM_AC3
+#if !defined(USE_TRANSFORM_AC3)
+#define USE_TRANSFORM_AC3 0   // ALTERNATE_CODE
+#endif
 
 #include <emmintrin.h>
-#include "./common_sse2.h"
-#include "../dec/vp8i_dec.h"
-#include "../utils/utils.h"
+#include "src/dsp/common_sse2.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
-static void Transform(const int16_t* in, uint8_t* dst, int do_two) {
+static void Transform_SSE2(const int16_t* in, uint8_t* dst, int do_two) {
   // This implementation makes use of 16-bit fixed point versions of two
   // multiply constants:
   //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
@@ -193,7 +195,7 @@ static void Transform(const int16_t* in, uint8_t* dst, int do_two) {
   }
 }
 
-#if defined(USE_TRANSFORM_AC3)
+#if (USE_TRANSFORM_AC3 == 1)
 #define MUL(a, b) (((a) * (b)) >> 16)
 static void TransformAC3(const int16_t* in, uint8_t* dst) {
   static const int kC1 = 20091 + (1 << 16);
@@ -248,7 +250,7 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
     _mm_subs_epu8((p), (q)))
 
 // Shift each byte of "x" by 3 bits while preserving by the sign bit.
-static WEBP_INLINE void SignedShift8b(__m128i* const x) {
+static WEBP_INLINE void SignedShift8b_SSE2(__m128i* const x) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i lo_0 = _mm_unpacklo_epi8(zero, *x);
   const __m128i hi_0 = _mm_unpackhi_epi8(zero, *x);
@@ -258,8 +260,8 @@ static WEBP_INLINE void SignedShift8b(__m128i* const x) {
 }
 
 #define FLIP_SIGN_BIT2(a, b) {                                                 \
-  a = _mm_xor_si128(a, sign_bit);                                              \
-  b = _mm_xor_si128(b, sign_bit);                                              \
+  (a) = _mm_xor_si128(a, sign_bit);                                            \
+  (b) = _mm_xor_si128(b, sign_bit);                                            \
 }
 
 #define FLIP_SIGN_BIT4(a, b, c, d) {                                           \
@@ -268,11 +270,11 @@ static WEBP_INLINE void SignedShift8b(__m128i* const x) {
 }
 
 // input/output is uint8_t
-static WEBP_INLINE void GetNotHEV(const __m128i* const p1,
-                                  const __m128i* const p0,
-                                  const __m128i* const q0,
-                                  const __m128i* const q1,
-                                  int hev_thresh, __m128i* const not_hev) {
+static WEBP_INLINE void GetNotHEV_SSE2(const __m128i* const p1,
+                                       const __m128i* const p0,
+                                       const __m128i* const q0,
+                                       const __m128i* const q1,
+                                       int hev_thresh, __m128i* const not_hev) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i t_1 = MM_ABS(*p1, *p0);
   const __m128i t_2 = MM_ABS(*q1, *q0);
@@ -285,11 +287,11 @@ static WEBP_INLINE void GetNotHEV(const __m128i* const p1,
 }
 
 // input pixels are int8_t
-static WEBP_INLINE void GetBaseDelta(const __m128i* const p1,
-                                     const __m128i* const p0,
-                                     const __m128i* const q0,
-                                     const __m128i* const q1,
-                                     __m128i* const delta) {
+static WEBP_INLINE void GetBaseDelta_SSE2(const __m128i* const p1,
+                                          const __m128i* const p0,
+                                          const __m128i* const q0,
+                                          const __m128i* const q1,
+                                          __m128i* const delta) {
   // beware of addition order, for saturation!
   const __m128i p1_q1 = _mm_subs_epi8(*p1, *q1);   // p1 - q1
   const __m128i q0_p0 = _mm_subs_epi8(*q0, *p0);   // q0 - p0
@@ -300,15 +302,16 @@ static WEBP_INLINE void GetBaseDelta(const __m128i* const p1,
 }
 
 // input and output are int8_t
-static WEBP_INLINE void DoSimpleFilter(__m128i* const p0, __m128i* const q0,
-                                       const __m128i* const fl) {
+static WEBP_INLINE void DoSimpleFilter_SSE2(__m128i* const p0,
+                                            __m128i* const q0,
+                                            const __m128i* const fl) {
   const __m128i k3 = _mm_set1_epi8(3);
   const __m128i k4 = _mm_set1_epi8(4);
   __m128i v3 = _mm_adds_epi8(*fl, k3);
   __m128i v4 = _mm_adds_epi8(*fl, k4);
 
-  SignedShift8b(&v4);                  // v4 >> 3
-  SignedShift8b(&v3);                  // v3 >> 3
+  SignedShift8b_SSE2(&v4);             // v4 >> 3
+  SignedShift8b_SSE2(&v3);             // v3 >> 3
   *q0 = _mm_subs_epi8(*q0, v4);        // q0 -= v4
   *p0 = _mm_adds_epi8(*p0, v3);        // p0 += v3
 }
@@ -317,9 +320,9 @@ static WEBP_INLINE void DoSimpleFilter(__m128i* const p0, __m128i* const q0,
 // Update operations:
 // q = q - delta and p = p + delta; where delta = [(a_hi >> 7), (a_lo >> 7)]
 // Pixels 'pi' and 'qi' are int8_t on input, uint8_t on output (sign flip).
-static WEBP_INLINE void Update2Pixels(__m128i* const pi, __m128i* const qi,
-                                      const __m128i* const a0_lo,
-                                      const __m128i* const a0_hi) {
+static WEBP_INLINE void Update2Pixels_SSE2(__m128i* const pi, __m128i* const qi,
+                                           const __m128i* const a0_lo,
+                                           const __m128i* const a0_hi) {
   const __m128i a1_lo = _mm_srai_epi16(*a0_lo, 7);
   const __m128i a1_hi = _mm_srai_epi16(*a0_hi, 7);
   const __m128i delta = _mm_packs_epi16(a1_lo, a1_hi);
@@ -330,11 +333,11 @@ static WEBP_INLINE void Update2Pixels(__m128i* const pi, __m128i* const qi,
 }
 
 // input pixels are uint8_t
-static WEBP_INLINE void NeedsFilter(const __m128i* const p1,
-                                    const __m128i* const p0,
-                                    const __m128i* const q0,
-                                    const __m128i* const q1,
-                                    int thresh, __m128i* const mask) {
+static WEBP_INLINE void NeedsFilter_SSE2(const __m128i* const p1,
+                                         const __m128i* const p0,
+                                         const __m128i* const q0,
+                                         const __m128i* const q1,
+                                         int thresh, __m128i* const mask) {
   const __m128i m_thresh = _mm_set1_epi8(thresh);
   const __m128i t1 = MM_ABS(*p1, *q1);        // abs(p1 - q1)
   const __m128i kFE = _mm_set1_epi8(0xFE);
@@ -353,28 +356,29 @@ static WEBP_INLINE void NeedsFilter(const __m128i* const p1,
 // Edge filtering functions
 
 // Applies filter on 2 pixels (p0 and q0)
-static WEBP_INLINE void DoFilter2(__m128i* const p1, __m128i* const p0,
-                                  __m128i* const q0, __m128i* const q1,
-                                  int thresh) {
+static WEBP_INLINE void DoFilter2_SSE2(__m128i* const p1, __m128i* const p0,
+                                       __m128i* const q0, __m128i* const q1,
+                                       int thresh) {
   __m128i a, mask;
   const __m128i sign_bit = _mm_set1_epi8(0x80);
-  // convert p1/q1 to int8_t (for GetBaseDelta)
+  // convert p1/q1 to int8_t (for GetBaseDelta_SSE2)
   const __m128i p1s = _mm_xor_si128(*p1, sign_bit);
   const __m128i q1s = _mm_xor_si128(*q1, sign_bit);
 
-  NeedsFilter(p1, p0, q0, q1, thresh, &mask);
+  NeedsFilter_SSE2(p1, p0, q0, q1, thresh, &mask);
 
   FLIP_SIGN_BIT2(*p0, *q0);
-  GetBaseDelta(&p1s, p0, q0, &q1s, &a);
+  GetBaseDelta_SSE2(&p1s, p0, q0, &q1s, &a);
   a = _mm_and_si128(a, mask);     // mask filter values we don't care about
-  DoSimpleFilter(p0, q0, &a);
+  DoSimpleFilter_SSE2(p0, q0, &a);
   FLIP_SIGN_BIT2(*p0, *q0);
 }
 
 // Applies filter on 4 pixels (p1, p0, q0 and q1)
-static WEBP_INLINE void DoFilter4(__m128i* const p1, __m128i* const p0,
-                                  __m128i* const q0, __m128i* const q1,
-                                  const __m128i* const mask, int hev_thresh) {
+static WEBP_INLINE void DoFilter4_SSE2(__m128i* const p1, __m128i* const p0,
+                                       __m128i* const q0, __m128i* const q1,
+                                       const __m128i* const mask,
+                                       int hev_thresh) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i sign_bit = _mm_set1_epi8(0x80);
   const __m128i k64 = _mm_set1_epi8(64);
@@ -384,7 +388,7 @@ static WEBP_INLINE void DoFilter4(__m128i* const p1, __m128i* const p0,
   __m128i t1, t2, t3;
 
   // compute hev mask
-  GetNotHEV(p1, p0, q0, q1, hev_thresh, &not_hev);
+  GetNotHEV_SSE2(p1, p0, q0, q1, hev_thresh, &not_hev);
 
   // convert to signed values
   FLIP_SIGN_BIT4(*p1, *p0, *q0, *q1);
@@ -399,8 +403,8 @@ static WEBP_INLINE void DoFilter4(__m128i* const p1, __m128i* const p0,
 
   t2 = _mm_adds_epi8(t1, k3);        // 3 * (q0 - p0) + hev(p1 - q1) + 3
   t3 = _mm_adds_epi8(t1, k4);        // 3 * (q0 - p0) + hev(p1 - q1) + 4
-  SignedShift8b(&t2);                // (3 * (q0 - p0) + hev(p1 - q1) + 3) >> 3
-  SignedShift8b(&t3);                // (3 * (q0 - p0) + hev(p1 - q1) + 4) >> 3
+  SignedShift8b_SSE2(&t2);           // (3 * (q0 - p0) + hev(p1 - q1) + 3) >> 3
+  SignedShift8b_SSE2(&t3);           // (3 * (q0 - p0) + hev(p1 - q1) + 4) >> 3
   *p0 = _mm_adds_epi8(*p0, t2);      // p0 += t2
   *q0 = _mm_subs_epi8(*q0, t3);      // q0 -= t3
   FLIP_SIGN_BIT2(*p0, *q0);
@@ -417,25 +421,26 @@ static WEBP_INLINE void DoFilter4(__m128i* const p1, __m128i* const p0,
 }
 
 // Applies filter on 6 pixels (p2, p1, p0, q0, q1 and q2)
-static WEBP_INLINE void DoFilter6(__m128i* const p2, __m128i* const p1,
-                                  __m128i* const p0, __m128i* const q0,
-                                  __m128i* const q1, __m128i* const q2,
-                                  const __m128i* const mask, int hev_thresh) {
+static WEBP_INLINE void DoFilter6_SSE2(__m128i* const p2, __m128i* const p1,
+                                       __m128i* const p0, __m128i* const q0,
+                                       __m128i* const q1, __m128i* const q2,
+                                       const __m128i* const mask,
+                                       int hev_thresh) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i sign_bit = _mm_set1_epi8(0x80);
   __m128i a, not_hev;
 
   // compute hev mask
-  GetNotHEV(p1, p0, q0, q1, hev_thresh, &not_hev);
+  GetNotHEV_SSE2(p1, p0, q0, q1, hev_thresh, &not_hev);
 
   FLIP_SIGN_BIT4(*p1, *p0, *q0, *q1);
   FLIP_SIGN_BIT2(*p2, *q2);
-  GetBaseDelta(p1, p0, q0, q1, &a);
+  GetBaseDelta_SSE2(p1, p0, q0, q1, &a);
 
   { // do simple filter on pixels with hev
     const __m128i m = _mm_andnot_si128(not_hev, *mask);
     const __m128i f = _mm_and_si128(a, m);
-    DoSimpleFilter(p0, q0, &f);
+    DoSimpleFilter_SSE2(p0, q0, &f);
   }
 
   { // do strong filter on pixels with not hev
@@ -460,15 +465,15 @@ static WEBP_INLINE void DoFilter6(__m128i* const p2, __m128i* const p1,
     const __m128i a0_lo = _mm_add_epi16(a1_lo, f9_lo);  // Filter * 27 + 63
     const __m128i a0_hi = _mm_add_epi16(a1_hi, f9_hi);  // Filter * 27 + 63
 
-    Update2Pixels(p2, q2, &a2_lo, &a2_hi);
-    Update2Pixels(p1, q1, &a1_lo, &a1_hi);
-    Update2Pixels(p0, q0, &a0_lo, &a0_hi);
+    Update2Pixels_SSE2(p2, q2, &a2_lo, &a2_hi);
+    Update2Pixels_SSE2(p1, q1, &a1_lo, &a1_hi);
+    Update2Pixels_SSE2(p0, q0, &a0_lo, &a0_hi);
   }
 }
 
 // reads 8 rows across a vertical edge.
-static WEBP_INLINE void Load8x4(const uint8_t* const b, int stride,
-                                __m128i* const p, __m128i* const q) {
+static WEBP_INLINE void Load8x4_SSE2(const uint8_t* const b, int stride,
+                                     __m128i* const p, __m128i* const q) {
   // A0 = 63 62 61 60 23 22 21 20 43 42 41 40 03 02 01 00
   // A1 = 73 72 71 70 33 32 31 30 53 52 51 50 13 12 11 10
   const __m128i A0 = _mm_set_epi32(
@@ -494,11 +499,11 @@ static WEBP_INLINE void Load8x4(const uint8_t* const b, int stride,
   *q = _mm_unpackhi_epi32(C0, C1);
 }
 
-static WEBP_INLINE void Load16x4(const uint8_t* const r0,
-                                 const uint8_t* const r8,
-                                 int stride,
-                                 __m128i* const p1, __m128i* const p0,
-                                 __m128i* const q0, __m128i* const q1) {
+static WEBP_INLINE void Load16x4_SSE2(const uint8_t* const r0,
+                                      const uint8_t* const r8,
+                                      int stride,
+                                      __m128i* const p1, __m128i* const p0,
+                                      __m128i* const q0, __m128i* const q1) {
   // Assume the pixels around the edge (|) are numbered as follows
   //                00 01 | 02 03
   //                10 11 | 12 13
@@ -514,8 +519,8 @@ static WEBP_INLINE void Load16x4(const uint8_t* const r0,
   // q0 = 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02
   // p0 = f1 e1 d1 c1 b1 a1 91 81 f0 e0 d0 c0 b0 a0 90 80
   // q1 = f3 e3 d3 c3 b3 a3 93 83 f2 e2 d2 c2 b2 a2 92 82
-  Load8x4(r0, stride, p1, q0);
-  Load8x4(r8, stride, p0, q1);
+  Load8x4_SSE2(r0, stride, p1, q0);
+  Load8x4_SSE2(r8, stride, p0, q1);
 
   {
     // p1 = f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00
@@ -531,7 +536,8 @@ static WEBP_INLINE void Load16x4(const uint8_t* const r0,
   }
 }
 
-static WEBP_INLINE void Store4x4(__m128i* const x, uint8_t* dst, int stride) {
+static WEBP_INLINE void Store4x4_SSE2(__m128i* const x,
+                                      uint8_t* dst, int stride) {
   int i;
   for (i = 0; i < 4; ++i, dst += stride) {
     WebPUint32ToMem(dst, _mm_cvtsi128_si32(*x));
@@ -540,12 +546,12 @@ static WEBP_INLINE void Store4x4(__m128i* const x, uint8_t* dst, int stride) {
 }
 
 // Transpose back and store
-static WEBP_INLINE void Store16x4(const __m128i* const p1,
-                                  const __m128i* const p0,
-                                  const __m128i* const q0,
-                                  const __m128i* const q1,
-                                  uint8_t* r0, uint8_t* r8,
-                                  int stride) {
+static WEBP_INLINE void Store16x4_SSE2(const __m128i* const p1,
+                                       const __m128i* const p0,
+                                       const __m128i* const q0,
+                                       const __m128i* const q1,
+                                       uint8_t* r0, uint8_t* r8,
+                                       int stride) {
   __m128i t1, p1_s, p0_s, q0_s, q1_s;
 
   // p0 = 71 70 61 60 51 50 41 40 31 30 21 20 11 10 01 00
@@ -572,55 +578,55 @@ static WEBP_INLINE void Store16x4(const __m128i* const p1,
   p1_s = _mm_unpacklo_epi16(t1, q1_s);
   q1_s = _mm_unpackhi_epi16(t1, q1_s);
 
-  Store4x4(&p0_s, r0, stride);
+  Store4x4_SSE2(&p0_s, r0, stride);
   r0 += 4 * stride;
-  Store4x4(&q0_s, r0, stride);
+  Store4x4_SSE2(&q0_s, r0, stride);
 
-  Store4x4(&p1_s, r8, stride);
+  Store4x4_SSE2(&p1_s, r8, stride);
   r8 += 4 * stride;
-  Store4x4(&q1_s, r8, stride);
+  Store4x4_SSE2(&q1_s, r8, stride);
 }
 
 //------------------------------------------------------------------------------
 // Simple In-loop filtering (Paragraph 15.2)
 
-static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16_SSE2(uint8_t* p, int stride, int thresh) {
   // Load
   __m128i p1 = _mm_loadu_si128((__m128i*)&p[-2 * stride]);
   __m128i p0 = _mm_loadu_si128((__m128i*)&p[-stride]);
   __m128i q0 = _mm_loadu_si128((__m128i*)&p[0]);
   __m128i q1 = _mm_loadu_si128((__m128i*)&p[stride]);
 
-  DoFilter2(&p1, &p0, &q0, &q1, thresh);
+  DoFilter2_SSE2(&p1, &p0, &q0, &q1, thresh);
 
   // Store
   _mm_storeu_si128((__m128i*)&p[-stride], p0);
   _mm_storeu_si128((__m128i*)&p[0], q0);
 }
 
-static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16_SSE2(uint8_t* p, int stride, int thresh) {
   __m128i p1, p0, q0, q1;
 
   p -= 2;  // beginning of p1
 
-  Load16x4(p, p + 8 * stride, stride, &p1, &p0, &q0, &q1);
-  DoFilter2(&p1, &p0, &q0, &q1, thresh);
-  Store16x4(&p1, &p0, &q0, &q1, p, p + 8 * stride, stride);
+  Load16x4_SSE2(p, p + 8 * stride, stride, &p1, &p0, &q0, &q1);
+  DoFilter2_SSE2(&p1, &p0, &q0, &q1, thresh);
+  Store16x4_SSE2(&p1, &p0, &q0, &q1, p, p + 8 * stride, stride);
 }
 
-static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16i_SSE2(uint8_t* p, int stride, int thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4 * stride;
-    SimpleVFilter16(p, stride, thresh);
+    SimpleVFilter16_SSE2(p, stride, thresh);
   }
 }
 
-static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16i_SSE2(uint8_t* p, int stride, int thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4;
-    SimpleHFilter16(p, stride, thresh);
+    SimpleHFilter16_SSE2(p, stride, thresh);
   }
 }
 
@@ -628,60 +634,60 @@ static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
 // Complex In-loop filtering (Paragraph 15.3)
 
 #define MAX_DIFF1(p3, p2, p1, p0, m) do {                                      \
-  m = MM_ABS(p1, p0);                                                          \
-  m = _mm_max_epu8(m, MM_ABS(p3, p2));                                         \
-  m = _mm_max_epu8(m, MM_ABS(p2, p1));                                         \
+  (m) = MM_ABS(p1, p0);                                                        \
+  (m) = _mm_max_epu8(m, MM_ABS(p3, p2));                                       \
+  (m) = _mm_max_epu8(m, MM_ABS(p2, p1));                                       \
 } while (0)
 
 #define MAX_DIFF2(p3, p2, p1, p0, m) do {                                      \
-  m = _mm_max_epu8(m, MM_ABS(p1, p0));                                         \
-  m = _mm_max_epu8(m, MM_ABS(p3, p2));                                         \
-  m = _mm_max_epu8(m, MM_ABS(p2, p1));                                         \
+  (m) = _mm_max_epu8(m, MM_ABS(p1, p0));                                       \
+  (m) = _mm_max_epu8(m, MM_ABS(p3, p2));                                       \
+  (m) = _mm_max_epu8(m, MM_ABS(p2, p1));                                       \
 } while (0)
 
 #define LOAD_H_EDGES4(p, stride, e1, e2, e3, e4) {                             \
-  e1 = _mm_loadu_si128((__m128i*)&(p)[0 * stride]);                            \
-  e2 = _mm_loadu_si128((__m128i*)&(p)[1 * stride]);                            \
-  e3 = _mm_loadu_si128((__m128i*)&(p)[2 * stride]);                            \
-  e4 = _mm_loadu_si128((__m128i*)&(p)[3 * stride]);                            \
+  (e1) = _mm_loadu_si128((__m128i*)&(p)[0 * (stride)]);                        \
+  (e2) = _mm_loadu_si128((__m128i*)&(p)[1 * (stride)]);                        \
+  (e3) = _mm_loadu_si128((__m128i*)&(p)[2 * (stride)]);                        \
+  (e4) = _mm_loadu_si128((__m128i*)&(p)[3 * (stride)]);                        \
 }
 
 #define LOADUV_H_EDGE(p, u, v, stride) do {                                    \
   const __m128i U = _mm_loadl_epi64((__m128i*)&(u)[(stride)]);                 \
   const __m128i V = _mm_loadl_epi64((__m128i*)&(v)[(stride)]);                 \
-  p = _mm_unpacklo_epi64(U, V);                                                \
+  (p) = _mm_unpacklo_epi64(U, V);                                              \
 } while (0)
 
 #define LOADUV_H_EDGES4(u, v, stride, e1, e2, e3, e4) {                        \
-  LOADUV_H_EDGE(e1, u, v, 0 * stride);                                         \
-  LOADUV_H_EDGE(e2, u, v, 1 * stride);                                         \
-  LOADUV_H_EDGE(e3, u, v, 2 * stride);                                         \
-  LOADUV_H_EDGE(e4, u, v, 3 * stride);                                         \
+  LOADUV_H_EDGE(e1, u, v, 0 * (stride));                                       \
+  LOADUV_H_EDGE(e2, u, v, 1 * (stride));                                       \
+  LOADUV_H_EDGE(e3, u, v, 2 * (stride));                                       \
+  LOADUV_H_EDGE(e4, u, v, 3 * (stride));                                       \
 }
 
 #define STOREUV(p, u, v, stride) {                                             \
-  _mm_storel_epi64((__m128i*)&u[(stride)], p);                                 \
-  p = _mm_srli_si128(p, 8);                                                    \
-  _mm_storel_epi64((__m128i*)&v[(stride)], p);                                 \
+  _mm_storel_epi64((__m128i*)&(u)[(stride)], p);                               \
+  (p) = _mm_srli_si128(p, 8);                                                  \
+  _mm_storel_epi64((__m128i*)&(v)[(stride)], p);                               \
 }
 
-static WEBP_INLINE void ComplexMask(const __m128i* const p1,
-                                    const __m128i* const p0,
-                                    const __m128i* const q0,
-                                    const __m128i* const q1,
-                                    int thresh, int ithresh,
-                                    __m128i* const mask) {
+static WEBP_INLINE void ComplexMask_SSE2(const __m128i* const p1,
+                                         const __m128i* const p0,
+                                         const __m128i* const q0,
+                                         const __m128i* const q1,
+                                         int thresh, int ithresh,
+                                         __m128i* const mask) {
   const __m128i it = _mm_set1_epi8(ithresh);
   const __m128i diff = _mm_subs_epu8(*mask, it);
   const __m128i thresh_mask = _mm_cmpeq_epi8(diff, _mm_setzero_si128());
   __m128i filter_mask;
-  NeedsFilter(p1, p0, q0, q1, thresh, &filter_mask);
+  NeedsFilter_SSE2(p1, p0, q0, q1, thresh, &filter_mask);
   *mask = _mm_and_si128(thresh_mask, filter_mask);
 }
 
 // on macroblock edges
-static void VFilter16(uint8_t* p, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void VFilter16_SSE2(uint8_t* p, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   __m128i t1;
   __m128i mask;
   __m128i p2, p1, p0, q0, q1, q2;
@@ -694,8 +700,8 @@ static void VFilter16(uint8_t* p, int stride,
   LOAD_H_EDGES4(p, stride, q0, q1, q2, t1);
   MAX_DIFF2(t1, q2, q1, q0, mask);
 
-  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
-  DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
+  ComplexMask_SSE2(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
+  DoFilter6_SSE2(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
   // Store
   _mm_storeu_si128((__m128i*)&p[-3 * stride], p2);
@@ -706,28 +712,28 @@ static void VFilter16(uint8_t* p, int stride,
   _mm_storeu_si128((__m128i*)&p[+2 * stride], q2);
 }
 
-static void HFilter16(uint8_t* p, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void HFilter16_SSE2(uint8_t* p, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i p3, p2, p1, p0, q0, q1, q2, q3;
 
   uint8_t* const b = p - 4;
-  Load16x4(b, b + 8 * stride, stride, &p3, &p2, &p1, &p0);  // p3, p2, p1, p0
+  Load16x4_SSE2(b, b + 8 * stride, stride, &p3, &p2, &p1, &p0);
   MAX_DIFF1(p3, p2, p1, p0, mask);
 
-  Load16x4(p, p + 8 * stride, stride, &q0, &q1, &q2, &q3);  // q0, q1, q2, q3
+  Load16x4_SSE2(p, p + 8 * stride, stride, &q0, &q1, &q2, &q3);
   MAX_DIFF2(q3, q2, q1, q0, mask);
 
-  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
-  DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
+  ComplexMask_SSE2(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
+  DoFilter6_SSE2(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
-  Store16x4(&p3, &p2, &p1, &p0, b, b + 8 * stride, stride);
-  Store16x4(&q0, &q1, &q2, &q3, p, p + 8 * stride, stride);
+  Store16x4_SSE2(&p3, &p2, &p1, &p0, b, b + 8 * stride, stride);
+  Store16x4_SSE2(&q0, &q1, &q2, &q3, p, p + 8 * stride, stride);
 }
 
 // on three inner edges
-static void VFilter16i(uint8_t* p, int stride,
-                       int thresh, int ithresh, int hev_thresh) {
+static void VFilter16i_SSE2(uint8_t* p, int stride,
+                            int thresh, int ithresh, int hev_thresh) {
   int k;
   __m128i p3, p2, p1, p0;   // loop invariants
 
@@ -744,8 +750,8 @@ static void VFilter16i(uint8_t* p, int stride,
 
     // p3 and p2 are not just temporary variables here: they will be
     // re-used for next span. And q2/q3 will become p1/p0 accordingly.
-    ComplexMask(&p1, &p0, &p3, &p2, thresh, ithresh, &mask);
-    DoFilter4(&p1, &p0, &p3, &p2, &mask, hev_thresh);
+    ComplexMask_SSE2(&p1, &p0, &p3, &p2, thresh, ithresh, &mask);
+    DoFilter4_SSE2(&p1, &p0, &p3, &p2, &mask, hev_thresh);
 
     // Store
     _mm_storeu_si128((__m128i*)&b[0 * stride], p1);
@@ -759,12 +765,12 @@ static void VFilter16i(uint8_t* p, int stride,
   }
 }
 
-static void HFilter16i(uint8_t* p, int stride,
-                       int thresh, int ithresh, int hev_thresh) {
+static void HFilter16i_SSE2(uint8_t* p, int stride,
+                            int thresh, int ithresh, int hev_thresh) {
   int k;
   __m128i p3, p2, p1, p0;   // loop invariants
 
-  Load16x4(p, p + 8 * stride, stride, &p3, &p2, &p1, &p0);  // prologue
+  Load16x4_SSE2(p, p + 8 * stride, stride, &p3, &p2, &p1, &p0);  // prologue
 
   for (k = 3; k > 0; --k) {
     __m128i mask, tmp1, tmp2;
@@ -773,13 +779,13 @@ static void HFilter16i(uint8_t* p, int stride,
     p += 4;  // beginning of q0 (and next span)
 
     MAX_DIFF1(p3, p2, p1, p0, mask);   // compute partial mask
-    Load16x4(p, p + 8 * stride, stride, &p3, &p2, &tmp1, &tmp2);
+    Load16x4_SSE2(p, p + 8 * stride, stride, &p3, &p2, &tmp1, &tmp2);
     MAX_DIFF2(p3, p2, tmp1, tmp2, mask);
 
-    ComplexMask(&p1, &p0, &p3, &p2, thresh, ithresh, &mask);
-    DoFilter4(&p1, &p0, &p3, &p2, &mask, hev_thresh);
+    ComplexMask_SSE2(&p1, &p0, &p3, &p2, thresh, ithresh, &mask);
+    DoFilter4_SSE2(&p1, &p0, &p3, &p2, &mask, hev_thresh);
 
-    Store16x4(&p1, &p0, &p3, &p2, b, b + 8 * stride, stride);
+    Store16x4_SSE2(&p1, &p0, &p3, &p2, b, b + 8 * stride, stride);
 
     // rotate samples
     p1 = tmp1;
@@ -788,8 +794,8 @@ static void HFilter16i(uint8_t* p, int stride,
 }
 
 // 8-pixels wide variant, for chroma filtering
-static void VFilter8(uint8_t* u, uint8_t* v, int stride,
-                     int thresh, int ithresh, int hev_thresh) {
+static void VFilter8_SSE2(uint8_t* u, uint8_t* v, int stride,
+                          int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i t1, p2, p1, p0, q0, q1, q2;
 
@@ -801,8 +807,8 @@ static void VFilter8(uint8_t* u, uint8_t* v, int stride,
   LOADUV_H_EDGES4(u, v, stride, q0, q1, q2, t1);
   MAX_DIFF2(t1, q2, q1, q0, mask);
 
-  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
-  DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
+  ComplexMask_SSE2(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
+  DoFilter6_SSE2(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
   // Store
   STOREUV(p2, u, v, -3 * stride);
@@ -813,28 +819,28 @@ static void VFilter8(uint8_t* u, uint8_t* v, int stride,
   STOREUV(q2, u, v, 2 * stride);
 }
 
-static void HFilter8(uint8_t* u, uint8_t* v, int stride,
-                     int thresh, int ithresh, int hev_thresh) {
+static void HFilter8_SSE2(uint8_t* u, uint8_t* v, int stride,
+                          int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i p3, p2, p1, p0, q0, q1, q2, q3;
 
   uint8_t* const tu = u - 4;
   uint8_t* const tv = v - 4;
-  Load16x4(tu, tv, stride, &p3, &p2, &p1, &p0);  // p3, p2, p1, p0
+  Load16x4_SSE2(tu, tv, stride, &p3, &p2, &p1, &p0);
   MAX_DIFF1(p3, p2, p1, p0, mask);
 
-  Load16x4(u, v, stride, &q0, &q1, &q2, &q3);    // q0, q1, q2, q3
+  Load16x4_SSE2(u, v, stride, &q0, &q1, &q2, &q3);
   MAX_DIFF2(q3, q2, q1, q0, mask);
 
-  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
-  DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
+  ComplexMask_SSE2(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
+  DoFilter6_SSE2(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
-  Store16x4(&p3, &p2, &p1, &p0, tu, tv, stride);
-  Store16x4(&q0, &q1, &q2, &q3, u, v, stride);
+  Store16x4_SSE2(&p3, &p2, &p1, &p0, tu, tv, stride);
+  Store16x4_SSE2(&q0, &q1, &q2, &q3, u, v, stride);
 }
 
-static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void VFilter8i_SSE2(uint8_t* u, uint8_t* v, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i t1, t2, p1, p0, q0, q1;
 
@@ -849,8 +855,8 @@ static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
   LOADUV_H_EDGES4(u, v, stride, q0, q1, t1, t2);
   MAX_DIFF2(t2, t1, q1, q0, mask);
 
-  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
-  DoFilter4(&p1, &p0, &q0, &q1, &mask, hev_thresh);
+  ComplexMask_SSE2(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
+  DoFilter4_SSE2(&p1, &p0, &q0, &q1, &mask, hev_thresh);
 
   // Store
   STOREUV(p1, u, v, -2 * stride);
@@ -859,24 +865,24 @@ static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
   STOREUV(q1, u, v, 1 * stride);
 }
 
-static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
-                      int thresh, int ithresh, int hev_thresh) {
+static void HFilter8i_SSE2(uint8_t* u, uint8_t* v, int stride,
+                           int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i t1, t2, p1, p0, q0, q1;
-  Load16x4(u, v, stride, &t2, &t1, &p1, &p0);   // p3, p2, p1, p0
+  Load16x4_SSE2(u, v, stride, &t2, &t1, &p1, &p0);   // p3, p2, p1, p0
   MAX_DIFF1(t2, t1, p1, p0, mask);
 
   u += 4;  // beginning of q0
   v += 4;
-  Load16x4(u, v, stride, &q0, &q1, &t1, &t2);  // q0, q1, q2, q3
+  Load16x4_SSE2(u, v, stride, &q0, &q1, &t1, &t2);  // q0, q1, q2, q3
   MAX_DIFF2(t2, t1, q1, q0, mask);
 
-  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
-  DoFilter4(&p1, &p0, &q0, &q1, &mask, hev_thresh);
+  ComplexMask_SSE2(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
+  DoFilter4_SSE2(&p1, &p0, &q0, &q1, &mask, hev_thresh);
 
   u -= 2;  // beginning of p1
   v -= 2;
-  Store16x4(&p1, &p0, &q0, &q1, u, v, stride);
+  Store16x4_SSE2(&p1, &p0, &q0, &q1, u, v, stride);
 }
 
 //------------------------------------------------------------------------------
@@ -893,7 +899,7 @@ static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
 //   where: AC = (a + b + 1) >> 1,   BC = (b + c + 1) >> 1
 //   and ab = a ^ b, bc = b ^ c, lsb = (AC^BC)&1
 
-static void VE4(uint8_t* dst) {    // vertical
+static void VE4_SSE2(uint8_t* dst) {    // vertical
   const __m128i one = _mm_set1_epi8(1);
   const __m128i ABCDEFGH = _mm_loadl_epi64((__m128i*)(dst - BPS - 1));
   const __m128i BCDEFGH0 = _mm_srli_si128(ABCDEFGH, 1);
@@ -909,7 +915,7 @@ static void VE4(uint8_t* dst) {    // vertical
   }
 }
 
-static void LD4(uint8_t* dst) {   // Down-Left
+static void LD4_SSE2(uint8_t* dst) {   // Down-Left
   const __m128i one = _mm_set1_epi8(1);
   const __m128i ABCDEFGH = _mm_loadl_epi64((__m128i*)(dst - BPS));
   const __m128i BCDEFGH0 = _mm_srli_si128(ABCDEFGH, 1);
@@ -925,7 +931,7 @@ static void LD4(uint8_t* dst) {   // Down-Left
   WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
 }
 
-static void VR4(uint8_t* dst) {   // Vertical-Right
+static void VR4_SSE2(uint8_t* dst) {   // Vertical-Right
   const __m128i one = _mm_set1_epi8(1);
   const int I = dst[-1 + 0 * BPS];
   const int J = dst[-1 + 1 * BPS];
@@ -950,7 +956,7 @@ static void VR4(uint8_t* dst) {   // Vertical-Right
   DST(0, 3) = AVG3(K, J, I);
 }
 
-static void VL4(uint8_t* dst) {   // Vertical-Left
+static void VL4_SSE2(uint8_t* dst) {   // Vertical-Left
   const __m128i one = _mm_set1_epi8(1);
   const __m128i ABCDEFGH = _mm_loadl_epi64((__m128i*)(dst - BPS));
   const __m128i BCDEFGH_ = _mm_srli_si128(ABCDEFGH, 1);
@@ -975,7 +981,7 @@ static void VL4(uint8_t* dst) {   // Vertical-Left
   DST(3, 3) = (extra_out >> 8) & 0xff;
 }
 
-static void RD4(uint8_t* dst) {   // Down-right
+static void RD4_SSE2(uint8_t* dst) {   // Down-right
   const __m128i one = _mm_set1_epi8(1);
   const __m128i XABCD = _mm_loadl_epi64((__m128i*)(dst - BPS - 1));
   const __m128i ____XABCD = _mm_slli_si128(XABCD, 4);
@@ -1004,7 +1010,7 @@ static void RD4(uint8_t* dst) {   // Down-right
 //------------------------------------------------------------------------------
 // Luma 16x16
 
-static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
+static WEBP_INLINE void TrueMotion_SSE2(uint8_t* dst, int size) {
   const uint8_t* top = dst - BPS;
   const __m128i zero = _mm_setzero_si128();
   int y;
@@ -1041,11 +1047,11 @@ static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
   }
 }
 
-static void TM4(uint8_t* dst)   { TrueMotion(dst, 4); }
-static void TM8uv(uint8_t* dst) { TrueMotion(dst, 8); }
-static void TM16(uint8_t* dst)  { TrueMotion(dst, 16); }
+static void TM4_SSE2(uint8_t* dst)   { TrueMotion_SSE2(dst, 4); }
+static void TM8uv_SSE2(uint8_t* dst) { TrueMotion_SSE2(dst, 8); }
+static void TM16_SSE2(uint8_t* dst)  { TrueMotion_SSE2(dst, 16); }
 
-static void VE16(uint8_t* dst) {
+static void VE16_SSE2(uint8_t* dst) {
   const __m128i top = _mm_loadu_si128((const __m128i*)(dst - BPS));
   int j;
   for (j = 0; j < 16; ++j) {
@@ -1053,7 +1059,7 @@ static void VE16(uint8_t* dst) {
   }
 }
 
-static void HE16(uint8_t* dst) {     // horizontal
+static void HE16_SSE2(uint8_t* dst) {     // horizontal
   int j;
   for (j = 16; j > 0; --j) {
     const __m128i values = _mm_set1_epi8(dst[-1]);
@@ -1062,7 +1068,7 @@ static void HE16(uint8_t* dst) {     // horizontal
   }
 }
 
-static WEBP_INLINE void Put16(uint8_t v, uint8_t* dst) {
+static WEBP_INLINE void Put16_SSE2(uint8_t v, uint8_t* dst) {
   int j;
   const __m128i values = _mm_set1_epi8(v);
   for (j = 0; j < 16; ++j) {
@@ -1070,7 +1076,7 @@ static WEBP_INLINE void Put16(uint8_t v, uint8_t* dst) {
   }
 }
 
-static void DC16(uint8_t* dst) {    // DC
+static void DC16_SSE2(uint8_t* dst) {  // DC
   const __m128i zero = _mm_setzero_si128();
   const __m128i top = _mm_loadu_si128((const __m128i*)(dst - BPS));
   const __m128i sad8x2 = _mm_sad_epu8(top, zero);
@@ -1083,37 +1089,37 @@ static void DC16(uint8_t* dst) {    // DC
   }
   {
     const int DC = _mm_cvtsi128_si32(sum) + left + 16;
-    Put16(DC >> 5, dst);
+    Put16_SSE2(DC >> 5, dst);
   }
 }
 
-static void DC16NoTop(uint8_t* dst) {   // DC with top samples not available
+static void DC16NoTop_SSE2(uint8_t* dst) {  // DC with top samples unavailable
   int DC = 8;
   int j;
   for (j = 0; j < 16; ++j) {
     DC += dst[-1 + j * BPS];
   }
-  Put16(DC >> 4, dst);
+  Put16_SSE2(DC >> 4, dst);
 }
 
-static void DC16NoLeft(uint8_t* dst) {  // DC with left samples not available
+static void DC16NoLeft_SSE2(uint8_t* dst) {  // DC with left samples unavailable
   const __m128i zero = _mm_setzero_si128();
   const __m128i top = _mm_loadu_si128((const __m128i*)(dst - BPS));
   const __m128i sad8x2 = _mm_sad_epu8(top, zero);
   // sum the two sads: sad8x2[0:1] + sad8x2[8:9]
   const __m128i sum = _mm_add_epi16(sad8x2, _mm_shuffle_epi32(sad8x2, 2));
   const int DC = _mm_cvtsi128_si32(sum) + 8;
-  Put16(DC >> 4, dst);
+  Put16_SSE2(DC >> 4, dst);
 }
 
-static void DC16NoTopLeft(uint8_t* dst) {  // DC with no top and left samples
-  Put16(0x80, dst);
+static void DC16NoTopLeft_SSE2(uint8_t* dst) {  // DC with no top & left samples
+  Put16_SSE2(0x80, dst);
 }
 
 //------------------------------------------------------------------------------
 // Chroma
 
-static void VE8uv(uint8_t* dst) {    // vertical
+static void VE8uv_SSE2(uint8_t* dst) {    // vertical
   int j;
   const __m128i top = _mm_loadl_epi64((const __m128i*)(dst - BPS));
   for (j = 0; j < 8; ++j) {
@@ -1121,17 +1127,8 @@ static void VE8uv(uint8_t* dst) {    // vertical
   }
 }
 
-static void HE8uv(uint8_t* dst) {    // horizontal
-  int j;
-  for (j = 0; j < 8; ++j) {
-    const __m128i values = _mm_set1_epi8(dst[-1]);
-    _mm_storel_epi64((__m128i*)dst, values);
-    dst += BPS;
-  }
-}
-
 // helper for chroma-DC predictions
-static WEBP_INLINE void Put8x8uv(uint8_t v, uint8_t* dst) {
+static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
   int j;
   const __m128i values = _mm_set1_epi8(v);
   for (j = 0; j < 8; ++j) {
@@ -1139,7 +1136,7 @@ static WEBP_INLINE void Put8x8uv(uint8_t v, uint8_t* dst) {
   }
 }
 
-static void DC8uv(uint8_t* dst) {     // DC
+static void DC8uv_SSE2(uint8_t* dst) {     // DC
   const __m128i zero = _mm_setzero_si128();
   const __m128i top = _mm_loadl_epi64((const __m128i*)(dst - BPS));
   const __m128i sum = _mm_sad_epu8(top, zero);
@@ -1150,29 +1147,29 @@ static void DC8uv(uint8_t* dst) {     // DC
   }
   {
     const int DC = _mm_cvtsi128_si32(sum) + left + 8;
-    Put8x8uv(DC >> 4, dst);
+    Put8x8uv_SSE2(DC >> 4, dst);
   }
 }
 
-static void DC8uvNoLeft(uint8_t* dst) {   // DC with no left samples
+static void DC8uvNoLeft_SSE2(uint8_t* dst) {   // DC with no left samples
   const __m128i zero = _mm_setzero_si128();
   const __m128i top = _mm_loadl_epi64((const __m128i*)(dst - BPS));
   const __m128i sum = _mm_sad_epu8(top, zero);
   const int DC = _mm_cvtsi128_si32(sum) + 4;
-  Put8x8uv(DC >> 3, dst);
+  Put8x8uv_SSE2(DC >> 3, dst);
 }
 
-static void DC8uvNoTop(uint8_t* dst) {  // DC with no top samples
+static void DC8uvNoTop_SSE2(uint8_t* dst) {  // DC with no top samples
   int dc0 = 4;
   int i;
   for (i = 0; i < 8; ++i) {
     dc0 += dst[-1 + i * BPS];
   }
-  Put8x8uv(dc0 >> 3, dst);
+  Put8x8uv_SSE2(dc0 >> 3, dst);
 }
 
-static void DC8uvNoTopLeft(uint8_t* dst) {    // DC with nothing
-  Put8x8uv(0x80, dst);
+static void DC8uvNoTopLeft_SSE2(uint8_t* dst) {    // DC with nothing
+  Put8x8uv_SSE2(0x80, dst);
 }
 
 //------------------------------------------------------------------------------
@@ -1181,47 +1178,46 @@ static void DC8uvNoTopLeft(uint8_t* dst) {    // DC with nothing
 extern void VP8DspInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8DspInitSSE2(void) {
-  VP8Transform = Transform;
-#if defined(USE_TRANSFORM_AC3)
-  VP8TransformAC3 = TransformAC3;
+  VP8Transform = Transform_SSE2;
+#if (USE_TRANSFORM_AC3 == 1)
+  VP8TransformAC3 = TransformAC3_SSE2;
 #endif
 
-  VP8VFilter16 = VFilter16;
-  VP8HFilter16 = HFilter16;
-  VP8VFilter8 = VFilter8;
-  VP8HFilter8 = HFilter8;
-  VP8VFilter16i = VFilter16i;
-  VP8HFilter16i = HFilter16i;
-  VP8VFilter8i = VFilter8i;
-  VP8HFilter8i = HFilter8i;
-
-  VP8SimpleVFilter16 = SimpleVFilter16;
-  VP8SimpleHFilter16 = SimpleHFilter16;
-  VP8SimpleVFilter16i = SimpleVFilter16i;
-  VP8SimpleHFilter16i = SimpleHFilter16i;
-
-  VP8PredLuma4[1] = TM4;
-  VP8PredLuma4[2] = VE4;
-  VP8PredLuma4[4] = RD4;
-  VP8PredLuma4[5] = VR4;
-  VP8PredLuma4[6] = LD4;
-  VP8PredLuma4[7] = VL4;
-
-  VP8PredLuma16[0] = DC16;
-  VP8PredLuma16[1] = TM16;
-  VP8PredLuma16[2] = VE16;
-  VP8PredLuma16[3] = HE16;
-  VP8PredLuma16[4] = DC16NoTop;
-  VP8PredLuma16[5] = DC16NoLeft;
-  VP8PredLuma16[6] = DC16NoTopLeft;
-
-  VP8PredChroma8[0] = DC8uv;
-  VP8PredChroma8[1] = TM8uv;
-  VP8PredChroma8[2] = VE8uv;
-  VP8PredChroma8[3] = HE8uv;
-  VP8PredChroma8[4] = DC8uvNoTop;
-  VP8PredChroma8[5] = DC8uvNoLeft;
-  VP8PredChroma8[6] = DC8uvNoTopLeft;
+  VP8VFilter16 = VFilter16_SSE2;
+  VP8HFilter16 = HFilter16_SSE2;
+  VP8VFilter8 = VFilter8_SSE2;
+  VP8HFilter8 = HFilter8_SSE2;
+  VP8VFilter16i = VFilter16i_SSE2;
+  VP8HFilter16i = HFilter16i_SSE2;
+  VP8VFilter8i = VFilter8i_SSE2;
+  VP8HFilter8i = HFilter8i_SSE2;
+
+  VP8SimpleVFilter16 = SimpleVFilter16_SSE2;
+  VP8SimpleHFilter16 = SimpleHFilter16_SSE2;
+  VP8SimpleVFilter16i = SimpleVFilter16i_SSE2;
+  VP8SimpleHFilter16i = SimpleHFilter16i_SSE2;
+
+  VP8PredLuma4[1] = TM4_SSE2;
+  VP8PredLuma4[2] = VE4_SSE2;
+  VP8PredLuma4[4] = RD4_SSE2;
+  VP8PredLuma4[5] = VR4_SSE2;
+  VP8PredLuma4[6] = LD4_SSE2;
+  VP8PredLuma4[7] = VL4_SSE2;
+
+  VP8PredLuma16[0] = DC16_SSE2;
+  VP8PredLuma16[1] = TM16_SSE2;
+  VP8PredLuma16[2] = VE16_SSE2;
+  VP8PredLuma16[3] = HE16_SSE2;
+  VP8PredLuma16[4] = DC16NoTop_SSE2;
+  VP8PredLuma16[5] = DC16NoLeft_SSE2;
+  VP8PredLuma16[6] = DC16NoTopLeft_SSE2;
+
+  VP8PredChroma8[0] = DC8uv_SSE2;
+  VP8PredChroma8[1] = TM8uv_SSE2;
+  VP8PredChroma8[2] = VE8uv_SSE2;
+  VP8PredChroma8[4] = DC8uvNoTop_SSE2;
+  VP8PredChroma8[5] = DC8uvNoLeft_SSE2;
+  VP8PredChroma8[6] = DC8uvNoTopLeft_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/dec_sse41.c b/src/3rdparty/libwebp/src/dsp/dec_sse41.c
index 4e81ec4..8f18506 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_sse41.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_sse41.c
@@ -11,15 +11,15 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE41)
 
 #include <smmintrin.h>
-#include "../dec/vp8i_dec.h"
-#include "../utils/utils.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/utils/utils.h"
 
-static void HE16(uint8_t* dst) {     // horizontal
+static void HE16_SSE41(uint8_t* dst) {     // horizontal
   int j;
   const __m128i kShuffle3 = _mm_set1_epi8(3);
   for (j = 16; j > 0; --j) {
@@ -36,7 +36,7 @@ static void HE16(uint8_t* dst) {     // horizontal
 extern void VP8DspInitSSE41(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8DspInitSSE41(void) {
-  VP8PredLuma16[3] = HE16;
+  VP8PredLuma16[3] = HE16_SSE41;
 }
 
 #else  // !WEBP_USE_SSE41
diff --git a/src/3rdparty/libwebp/src/dsp/dsp.h b/src/3rdparty/libwebp/src/dsp/dsp.h
index 7cf6e05..5b703a1 100644
--- a/src/3rdparty/libwebp/src/dsp/dsp.h
+++ b/src/3rdparty/libwebp/src/dsp/dsp.h
@@ -15,10 +15,10 @@
 #define WEBP_DSP_DSP_H_
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -38,10 +38,22 @@ extern "C" {
 # define LOCAL_GCC_PREREQ(maj, min) 0
 #endif
 
+#if defined(__clang__)
+# define LOCAL_CLANG_VERSION ((__clang_major__ << 8) | __clang_minor__)
+# define LOCAL_CLANG_PREREQ(maj, min) \
+    (LOCAL_CLANG_VERSION >= (((maj) << 8) | (min)))
+#else
+# define LOCAL_CLANG_VERSION 0
+# define LOCAL_CLANG_PREREQ(maj, min) 0
+#endif
+
 #ifndef __has_builtin
 # define __has_builtin(x) 0
 #endif
 
+// for now, none of the optimizations below are available in emscripten
+#if !defined(EMSCRIPTEN)
+
 #if defined(_MSC_VER) && _MSC_VER > 1310 && \
     (defined(_M_X64) || defined(_M_IX86))
 #define WEBP_MSC_SSE2  // Visual C++ SSE2 targets
@@ -69,18 +81,20 @@ extern "C" {
 #define WEBP_USE_AVX2
 #endif
 
-#if defined(__ANDROID__) && defined(__ARM_ARCH_7A__)
-#define WEBP_ANDROID_NEON  // Android targets that might support NEON
-#endif
-
 // The intrinsics currently cause compiler errors with arm-nacl-gcc and the
 // inline assembly would need to be modified for use with Native Client.
-#if (defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON) || \
+#if (defined(__ARM_NEON__) || \
      defined(__aarch64__) || defined(WEBP_HAVE_NEON)) && \
     !defined(__native_client__)
 #define WEBP_USE_NEON
 #endif
 
+#if !defined(WEBP_USE_NEON) && defined(__ANDROID__) && \
+    defined(__ARM_ARCH_7A__) && defined(HAVE_CPU_FEATURES_H)
+#define WEBP_ANDROID_NEON  // Android targets that may have NEON
+#define WEBP_USE_NEON
+#endif
+
 #if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_M_ARM)
 #define WEBP_USE_NEON
 #define WEBP_USE_INTRINSICS
@@ -91,7 +105,7 @@ extern "C" {
 #define WEBP_USE_MIPS32
 #if (__mips_isa_rev >= 2)
 #define WEBP_USE_MIPS32_R2
-#if defined(__mips_dspr2) || (__mips_dsp_rev >= 2)
+#if defined(__mips_dspr2) || (defined(__mips_dsp_rev) && __mips_dsp_rev >= 2)
 #define WEBP_USE_MIPS_DSP_R2
 #endif
 #endif
@@ -101,6 +115,24 @@ extern "C" {
 #define WEBP_USE_MSA
 #endif
 
+#endif  /* EMSCRIPTEN */
+
+#ifndef WEBP_DSP_OMIT_C_CODE
+#define WEBP_DSP_OMIT_C_CODE 1
+#endif
+
+#if (defined(__aarch64__) || defined(__ARM_NEON__)) && WEBP_DSP_OMIT_C_CODE
+#define WEBP_NEON_OMIT_C_CODE 1
+#else
+#define WEBP_NEON_OMIT_C_CODE 0
+#endif
+
+#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
+#define WEBP_NEON_WORK_AROUND_GCC 1
+#else
+#define WEBP_NEON_WORK_AROUND_GCC 0
+#endif
+
 // This macro prevents thread_sanitizer from reporting known concurrent writes.
 #define WEBP_TSAN_IGNORE_FUNCTION
 #if defined(__has_feature)
@@ -130,6 +162,11 @@ extern "C" {
 #endif
 #endif
 
+// Regularize the definition of WEBP_SWAP_16BIT_CSP (backward compatibility)
+#if !defined(WEBP_SWAP_16BIT_CSP)
+#define WEBP_SWAP_16BIT_CSP 0
+#endif
+
 typedef enum {
   kSSE2,
   kSSE3,
@@ -144,7 +181,7 @@ typedef enum {
 } CPUFeature;
 // returns true if the CPU supports the feature.
 typedef int (*VP8CPUInfo)(CPUFeature feature);
-WEBP_EXTERN(VP8CPUInfo) VP8GetCPUInfo;
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 
 //------------------------------------------------------------------------------
 // Init stub generator
@@ -272,6 +309,7 @@ typedef double (*VP8SSIMGetClippedFunc)(const uint8_t* src1, int stride1,
                                         int xo, int yo,  // center position
                                         int W, int H);   // plane dimension
 
+#if !defined(WEBP_REDUCE_SIZE)
 // This version is called with the guarantee that you can load 8 bytes and
 // 8 rows at offset src1 and src2
 typedef double (*VP8SSIMGetFunc)(const uint8_t* src1, int stride1,
@@ -279,10 +317,13 @@ typedef double (*VP8SSIMGetFunc)(const uint8_t* src1, int stride1,
 
 extern VP8SSIMGetFunc VP8SSIMGet;         // unclipped / unchecked
 extern VP8SSIMGetClippedFunc VP8SSIMGetClipped;   // with clipping
+#endif
 
+#if !defined(WEBP_DISABLE_STATS)
 typedef uint32_t (*VP8AccumulateSSEFunc)(const uint8_t* src1,
                                          const uint8_t* src2, int len);
 extern VP8AccumulateSSEFunc VP8AccumulateSSE;
+#endif
 
 // must be called before using any of the above directly
 void VP8SSIMDspInit(void);
@@ -463,12 +504,12 @@ extern WebPRescalerExportRowFunc WebPRescalerExportRowExpand;
 extern WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
 
 // Plain-C implementation, as fall-back.
-extern void WebPRescalerImportRowExpandC(struct WebPRescaler* const wrk,
-                                         const uint8_t* src);
-extern void WebPRescalerImportRowShrinkC(struct WebPRescaler* const wrk,
-                                         const uint8_t* src);
-extern void WebPRescalerExportRowExpandC(struct WebPRescaler* const wrk);
-extern void WebPRescalerExportRowShrinkC(struct WebPRescaler* const wrk);
+extern void WebPRescalerImportRowExpand_C(struct WebPRescaler* const wrk,
+                                          const uint8_t* src);
+extern void WebPRescalerImportRowShrink_C(struct WebPRescaler* const wrk,
+                                          const uint8_t* src);
+extern void WebPRescalerExportRowExpand_C(struct WebPRescaler* const wrk);
+extern void WebPRescalerExportRowShrink_C(struct WebPRescaler* const wrk);
 
 // Main entry calls:
 extern void WebPRescalerImportRow(struct WebPRescaler* const wrk,
@@ -534,24 +575,21 @@ void WebPMultRows(uint8_t* ptr, int stride,
                   int width, int num_rows, int inverse);
 
 // Plain-C versions, used as fallback by some implementations.
-void WebPMultRowC(uint8_t* const ptr, const uint8_t* const alpha,
-                  int width, int inverse);
-void WebPMultARGBRowC(uint32_t* const ptr, int width, int inverse);
-
-// To be called first before using the above.
-void WebPInitAlphaProcessing(void);
-
-// ARGB packing function: a/r/g/b input is rgba or bgra order.
-extern void (*VP8PackARGB)(const uint8_t* a, const uint8_t* r,
-                           const uint8_t* g, const uint8_t* b, int len,
-                           uint32_t* out);
+void WebPMultRow_C(uint8_t* const ptr, const uint8_t* const alpha,
+                   int width, int inverse);
+void WebPMultARGBRow_C(uint32_t* const ptr, int width, int inverse);
 
 // RGB packing function. 'step' can be 3 or 4. r/g/b input is rgb or bgr order.
-extern void (*VP8PackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
-                          int len, int step, uint32_t* out);
+extern void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
+                           int len, int step, uint32_t* out);
+
+// This function returns true if src[i] contains a value different from 0xff.
+extern int (*WebPHasAlpha8b)(const uint8_t* src, int length);
+// This function returns true if src[4*i] contains a value different from 0xff.
+extern int (*WebPHasAlpha32b)(const uint8_t* src, int length);
 
 // To be called first before using the above.
-void VP8EncDspARGBInit(void);
+void WebPInitAlphaProcessing(void);
 
 //------------------------------------------------------------------------------
 // Filter functions
diff --git a/src/3rdparty/libwebp/src/dsp/enc.c b/src/3rdparty/libwebp/src/dsp/enc.c
index f31bc6d..1c807f1 100644
--- a/src/3rdparty/libwebp/src/dsp/enc.c
+++ b/src/3rdparty/libwebp/src/dsp/enc.c
@@ -14,16 +14,18 @@
 #include <assert.h>
 #include <stdlib.h>  // for abs()
 
-#include "./dsp.h"
-#include "../enc/vp8i_enc.h"
+#include "src/dsp/dsp.h"
+#include "src/enc/vp8i_enc.h"
 
 static WEBP_INLINE uint8_t clip_8b(int v) {
   return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
 }
 
+#if !WEBP_NEON_OMIT_C_CODE
 static WEBP_INLINE int clip_max(int v, int max) {
   return (v > max) ? max : v;
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 //------------------------------------------------------------------------------
 // Compute susceptibility based on DCT-coeff histograms:
@@ -56,9 +58,10 @@ void VP8SetHistogramData(const int distribution[MAX_COEFF_THRESH + 1],
   histo->last_non_zero = last_non_zero;
 }
 
-static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
-                             int start_block, int end_block,
-                             VP8Histogram* const histo) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void CollectHistogram_C(const uint8_t* ref, const uint8_t* pred,
+                               int start_block, int end_block,
+                               VP8Histogram* const histo) {
   int j;
   int distribution[MAX_COEFF_THRESH + 1] = { 0 };
   for (j = start_block; j < end_block; ++j) {
@@ -76,6 +79,7 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
   }
   VP8SetHistogramData(distribution, histo);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 //------------------------------------------------------------------------------
 // run-time tables (~4k)
@@ -100,6 +104,8 @@ static WEBP_TSAN_IGNORE_FUNCTION void InitTables(void) {
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
+#if !WEBP_NEON_OMIT_C_CODE
+
 #define STORE(x, y, v) \
   dst[(x) + (y) * BPS] = clip_8b(ref[(x) + (y) * BPS] + ((v) >> 3))
 
@@ -140,15 +146,15 @@ static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
   }
 }
 
-static void ITransform(const uint8_t* ref, const int16_t* in, uint8_t* dst,
-                       int do_two) {
+static void ITransform_C(const uint8_t* ref, const int16_t* in, uint8_t* dst,
+                         int do_two) {
   ITransformOne(ref, in, dst);
   if (do_two) {
     ITransformOne(ref + 4, in + 16, dst + 4);
   }
 }
 
-static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+static void FTransform_C(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   int i;
   int tmp[16];
   for (i = 0; i < 4; ++i, src += BPS, ref += BPS) {
@@ -176,13 +182,16 @@ static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
     out[12+ i] = ((a3 * 2217 - a2 * 5352 + 51000) >> 16);
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void FTransform2(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+static void FTransform2_C(const uint8_t* src, const uint8_t* ref,
+                          int16_t* out) {
   VP8FTransform(src, ref, out);
   VP8FTransform(src + 4, ref + 4, out + 16);
 }
 
-static void FTransformWHT(const int16_t* in, int16_t* out) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void FTransformWHT_C(const int16_t* in, int16_t* out) {
   // input is 12b signed
   int32_t tmp[16];
   int i;
@@ -211,6 +220,7 @@ static void FTransformWHT(const int16_t* in, int16_t* out) {
     out[12 + i] = b3 >> 1;
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 #undef MUL
 #undef STORE
@@ -303,8 +313,8 @@ static WEBP_INLINE void DCMode(uint8_t* dst, const uint8_t* left,
 //------------------------------------------------------------------------------
 // Chroma 8x8 prediction (paragraph 12.2)
 
-static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
-                             const uint8_t* top) {
+static void IntraChromaPreds_C(uint8_t* dst, const uint8_t* left,
+                               const uint8_t* top) {
   // U block
   DCMode(C8DC8 + dst, left, top, 8, 8, 4);
   VerticalPred(C8VE8 + dst, top, 8);
@@ -323,8 +333,8 @@ static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
 //------------------------------------------------------------------------------
 // luma 16x16 prediction (paragraph 12.3)
 
-static void Intra16Preds(uint8_t* dst,
-                         const uint8_t* left, const uint8_t* top) {
+static void Intra16Preds_C(uint8_t* dst,
+                           const uint8_t* left, const uint8_t* top) {
   DCMode(I16DC16 + dst, left, top, 16, 16, 5);
   VerticalPred(I16VE16 + dst, top, 16);
   HorizontalPred(I16HE16 + dst, left, 16);
@@ -507,7 +517,7 @@ static void TM4(uint8_t* dst, const uint8_t* top) {
 
 // Left samples are top[-5 .. -2], top_left is top[-1], top are
 // located at top[0..3], and top right is top[4..7]
-static void Intra4Preds(uint8_t* dst, const uint8_t* top) {
+static void Intra4Preds_C(uint8_t* dst, const uint8_t* top) {
   DC4(I4DC4 + dst, top);
   TM4(I4TM4 + dst, top);
   VE4(I4VE4 + dst, top);
@@ -523,6 +533,7 @@ static void Intra4Preds(uint8_t* dst, const uint8_t* top) {
 //------------------------------------------------------------------------------
 // Metric
 
+#if !WEBP_NEON_OMIT_C_CODE
 static WEBP_INLINE int GetSSE(const uint8_t* a, const uint8_t* b,
                               int w, int h) {
   int count = 0;
@@ -538,20 +549,21 @@ static WEBP_INLINE int GetSSE(const uint8_t* a, const uint8_t* b,
   return count;
 }
 
-static int SSE16x16(const uint8_t* a, const uint8_t* b) {
+static int SSE16x16_C(const uint8_t* a, const uint8_t* b) {
   return GetSSE(a, b, 16, 16);
 }
-static int SSE16x8(const uint8_t* a, const uint8_t* b) {
+static int SSE16x8_C(const uint8_t* a, const uint8_t* b) {
   return GetSSE(a, b, 16, 8);
 }
-static int SSE8x8(const uint8_t* a, const uint8_t* b) {
+static int SSE8x8_C(const uint8_t* a, const uint8_t* b) {
   return GetSSE(a, b, 8, 8);
 }
-static int SSE4x4(const uint8_t* a, const uint8_t* b) {
+static int SSE4x4_C(const uint8_t* a, const uint8_t* b) {
   return GetSSE(a, b, 4, 4);
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void Mean16x4(const uint8_t* ref, uint32_t dc[4]) {
+static void Mean16x4_C(const uint8_t* ref, uint32_t dc[4]) {
   int k, x, y;
   for (k = 0; k < 4; ++k) {
     uint32_t avg = 0;
@@ -571,6 +583,7 @@ static void Mean16x4(const uint8_t* ref, uint32_t dc[4]) {
 // We try to match the spectral content (weighted) between source and
 // reconstructed samples.
 
+#if !WEBP_NEON_OMIT_C_CODE
 // Hadamard transform
 // Returns the weighted sum of the absolute value of transformed coefficients.
 // w[] contains a row-major 4 by 4 symmetric matrix.
@@ -608,24 +621,25 @@ static int TTransform(const uint8_t* in, const uint16_t* w) {
   return sum;
 }
 
-static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
-                    const uint16_t* const w) {
+static int Disto4x4_C(const uint8_t* const a, const uint8_t* const b,
+                      const uint16_t* const w) {
   const int sum1 = TTransform(a, w);
   const int sum2 = TTransform(b, w);
   return abs(sum2 - sum1) >> 5;
 }
 
-static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
-                      const uint16_t* const w) {
+static int Disto16x16_C(const uint8_t* const a, const uint8_t* const b,
+                        const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4(a + x + y, b + x + y, w);
+      D += Disto4x4_C(a + x + y, b + x + y, w);
     }
   }
   return D;
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 //------------------------------------------------------------------------------
 // Quantization
@@ -636,8 +650,8 @@ static const uint8_t kZigzag[16] = {
 };
 
 // Simple quantization
-static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         const VP8Matrix* const mtx) {
+static int QuantizeBlock_C(int16_t in[16], int16_t out[16],
+                           const VP8Matrix* const mtx) {
   int last = -1;
   int n;
   for (n = 0; n < 16; ++n) {
@@ -662,13 +676,15 @@ static int QuantizeBlock(int16_t in[16], int16_t out[16],
   return (last >= 0);
 }
 
-static int Quantize2Blocks(int16_t in[32], int16_t out[32],
-                           const VP8Matrix* const mtx) {
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+static int Quantize2Blocks_C(int16_t in[32], int16_t out[32],
+                             const VP8Matrix* const mtx) {
   int nz;
   nz  = VP8EncQuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
   nz |= VP8EncQuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
   return nz;
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
 
 //------------------------------------------------------------------------------
 // Block copy
@@ -682,149 +698,15 @@ static WEBP_INLINE void Copy(const uint8_t* src, uint8_t* dst, int w, int h) {
   }
 }
 
-static void Copy4x4(const uint8_t* src, uint8_t* dst) {
+static void Copy4x4_C(const uint8_t* src, uint8_t* dst) {
   Copy(src, dst, 4, 4);
 }
 
-static void Copy16x8(const uint8_t* src, uint8_t* dst) {
+static void Copy16x8_C(const uint8_t* src, uint8_t* dst) {
   Copy(src, dst, 16, 8);
 }
 
 //------------------------------------------------------------------------------
-// SSIM / PSNR
-
-// hat-shaped filter. Sum of coefficients is equal to 16.
-static const uint32_t kWeight[2 * VP8_SSIM_KERNEL + 1] = {
-  1, 2, 3, 4, 3, 2, 1
-};
-static const uint32_t kWeightSum = 16 * 16;   // sum{kWeight}^2
-
-static WEBP_INLINE double SSIMCalculation(
-    const VP8DistoStats* const stats, uint32_t N  /*num samples*/) {
-  const uint32_t w2 =  N * N;
-  const uint32_t C1 = 20 * w2;
-  const uint32_t C2 = 60 * w2;
-  const uint32_t C3 = 8 * 8 * w2;   // 'dark' limit ~= 6
-  const uint64_t xmxm = (uint64_t)stats->xm * stats->xm;
-  const uint64_t ymym = (uint64_t)stats->ym * stats->ym;
-  if (xmxm + ymym >= C3) {
-    const int64_t xmym = (int64_t)stats->xm * stats->ym;
-    const int64_t sxy = (int64_t)stats->xym * N - xmym;    // can be negative
-    const uint64_t sxx = (uint64_t)stats->xxm * N - xmxm;
-    const uint64_t syy = (uint64_t)stats->yym * N - ymym;
-    // we descale by 8 to prevent overflow during the fnum/fden multiply.
-    const uint64_t num_S = (2 * (uint64_t)(sxy < 0 ? 0 : sxy) + C2) >> 8;
-    const uint64_t den_S = (sxx + syy + C2) >> 8;
-    const uint64_t fnum = (2 * xmym + C1) * num_S;
-    const uint64_t fden = (xmxm + ymym + C1) * den_S;
-    const double r = (double)fnum / fden;
-    assert(r >= 0. && r <= 1.0);
-    return r;
-  }
-  return 1.;   // area is too dark to contribute meaningfully
-}
-
-double VP8SSIMFromStats(const VP8DistoStats* const stats) {
-  return SSIMCalculation(stats, kWeightSum);
-}
-
-double VP8SSIMFromStatsClipped(const VP8DistoStats* const stats) {
-  return SSIMCalculation(stats, stats->w);
-}
-
-static double SSIMGetClipped_C(const uint8_t* src1, int stride1,
-                               const uint8_t* src2, int stride2,
-                               int xo, int yo, int W, int H) {
-  VP8DistoStats stats = { 0, 0, 0, 0, 0, 0 };
-  const int ymin = (yo - VP8_SSIM_KERNEL < 0) ? 0 : yo - VP8_SSIM_KERNEL;
-  const int ymax = (yo + VP8_SSIM_KERNEL > H - 1) ? H - 1
-                                                  : yo + VP8_SSIM_KERNEL;
-  const int xmin = (xo - VP8_SSIM_KERNEL < 0) ? 0 : xo - VP8_SSIM_KERNEL;
-  const int xmax = (xo + VP8_SSIM_KERNEL > W - 1) ? W - 1
-                                                  : xo + VP8_SSIM_KERNEL;
-  int x, y;
-  src1 += ymin * stride1;
-  src2 += ymin * stride2;
-  for (y = ymin; y <= ymax; ++y, src1 += stride1, src2 += stride2) {
-    for (x = xmin; x <= xmax; ++x) {
-      const uint32_t w = kWeight[VP8_SSIM_KERNEL + x - xo]
-                       * kWeight[VP8_SSIM_KERNEL + y - yo];
-      const uint32_t s1 = src1[x];
-      const uint32_t s2 = src2[x];
-      stats.w   += w;
-      stats.xm  += w * s1;
-      stats.ym  += w * s2;
-      stats.xxm += w * s1 * s1;
-      stats.xym += w * s1 * s2;
-      stats.yym += w * s2 * s2;
-    }
-  }
-  return VP8SSIMFromStatsClipped(&stats);
-}
-
-static double SSIMGet_C(const uint8_t* src1, int stride1,
-                        const uint8_t* src2, int stride2) {
-  VP8DistoStats stats = { 0, 0, 0, 0, 0, 0 };
-  int x, y;
-  for (y = 0; y <= 2 * VP8_SSIM_KERNEL; ++y, src1 += stride1, src2 += stride2) {
-    for (x = 0; x <= 2 * VP8_SSIM_KERNEL; ++x) {
-      const uint32_t w = kWeight[x] * kWeight[y];
-      const uint32_t s1 = src1[x];
-      const uint32_t s2 = src2[x];
-      stats.xm  += w * s1;
-      stats.ym  += w * s2;
-      stats.xxm += w * s1 * s1;
-      stats.xym += w * s1 * s2;
-      stats.yym += w * s2 * s2;
-    }
-  }
-  return VP8SSIMFromStats(&stats);
-}
-
-//------------------------------------------------------------------------------
-
-static uint32_t AccumulateSSE(const uint8_t* src1,
-                              const uint8_t* src2, int len) {
-  int i;
-  uint32_t sse2 = 0;
-  assert(len <= 65535);  // to ensure that accumulation fits within uint32_t
-  for (i = 0; i < len; ++i) {
-    const int32_t diff = src1[i] - src2[i];
-    sse2 += diff * diff;
-  }
-  return sse2;
-}
-
-//------------------------------------------------------------------------------
-
-VP8SSIMGetFunc VP8SSIMGet;
-VP8SSIMGetClippedFunc VP8SSIMGetClipped;
-VP8AccumulateSSEFunc VP8AccumulateSSE;
-
-extern void VP8SSIMDspInitSSE2(void);
-
-static volatile VP8CPUInfo ssim_last_cpuinfo_used =
-    (VP8CPUInfo)&ssim_last_cpuinfo_used;
-
-WEBP_TSAN_IGNORE_FUNCTION void VP8SSIMDspInit(void) {
-  if (ssim_last_cpuinfo_used == VP8GetCPUInfo) return;
-
-  VP8SSIMGetClipped = SSIMGetClipped_C;
-  VP8SSIMGet = SSIMGet_C;
-
-  VP8AccumulateSSE = AccumulateSSE;
-  if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
-    if (VP8GetCPUInfo(kSSE2)) {
-      VP8SSIMDspInitSSE2();
-    }
-#endif
-  }
-
-  ssim_last_cpuinfo_used = VP8GetCPUInfo;
-}
-
-//------------------------------------------------------------------------------
 // Initialization
 
 // Speed-critical function pointers. We have to initialize them to the default
@@ -868,26 +750,32 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInit(void) {
   InitTables();
 
   // default C implementations
-  VP8CollectHistogram = CollectHistogram;
-  VP8ITransform = ITransform;
-  VP8FTransform = FTransform;
-  VP8FTransform2 = FTransform2;
-  VP8FTransformWHT = FTransformWHT;
-  VP8EncPredLuma4 = Intra4Preds;
-  VP8EncPredLuma16 = Intra16Preds;
-  VP8EncPredChroma8 = IntraChromaPreds;
-  VP8SSE16x16 = SSE16x16;
-  VP8SSE8x8 = SSE8x8;
-  VP8SSE16x8 = SSE16x8;
-  VP8SSE4x4 = SSE4x4;
-  VP8TDisto4x4 = Disto4x4;
-  VP8TDisto16x16 = Disto16x16;
-  VP8Mean16x4 = Mean16x4;
-  VP8EncQuantizeBlock = QuantizeBlock;
-  VP8EncQuantize2Blocks = Quantize2Blocks;
-  VP8EncQuantizeBlockWHT = QuantizeBlock;
-  VP8Copy4x4 = Copy4x4;
-  VP8Copy16x8 = Copy16x8;
+#if !WEBP_NEON_OMIT_C_CODE
+  VP8ITransform = ITransform_C;
+  VP8FTransform = FTransform_C;
+  VP8FTransformWHT = FTransformWHT_C;
+  VP8TDisto4x4 = Disto4x4_C;
+  VP8TDisto16x16 = Disto16x16_C;
+  VP8CollectHistogram = CollectHistogram_C;
+  VP8SSE16x16 = SSE16x16_C;
+  VP8SSE16x8 = SSE16x8_C;
+  VP8SSE8x8 = SSE8x8_C;
+  VP8SSE4x4 = SSE4x4_C;
+#endif
+
+#if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
+  VP8EncQuantizeBlock = QuantizeBlock_C;
+  VP8EncQuantize2Blocks = Quantize2Blocks_C;
+#endif
+
+  VP8FTransform2 = FTransform2_C;
+  VP8EncPredLuma4 = Intra4Preds_C;
+  VP8EncPredLuma16 = Intra16Preds_C;
+  VP8EncPredChroma8 = IntraChromaPreds_C;
+  VP8Mean16x4 = Mean16x4_C;
+  VP8EncQuantizeBlockWHT = QuantizeBlock_C;
+  VP8Copy4x4 = Copy4x4_C;
+  VP8Copy16x8 = Copy16x8_C;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
@@ -906,11 +794,6 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInit(void) {
       VP8EncDspInitAVX2();
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      VP8EncDspInitNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
       VP8EncDspInitMIPS32();
@@ -927,5 +810,34 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInit(void) {
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    VP8EncDspInitNEON();
+  }
+#endif
+
+  assert(VP8ITransform != NULL);
+  assert(VP8FTransform != NULL);
+  assert(VP8FTransformWHT != NULL);
+  assert(VP8TDisto4x4 != NULL);
+  assert(VP8TDisto16x16 != NULL);
+  assert(VP8CollectHistogram != NULL);
+  assert(VP8SSE16x16 != NULL);
+  assert(VP8SSE16x8 != NULL);
+  assert(VP8SSE8x8 != NULL);
+  assert(VP8SSE4x4 != NULL);
+  assert(VP8EncQuantizeBlock != NULL);
+  assert(VP8EncQuantize2Blocks != NULL);
+  assert(VP8FTransform2 != NULL);
+  assert(VP8EncPredLuma4 != NULL);
+  assert(VP8EncPredLuma16 != NULL);
+  assert(VP8EncPredChroma8 != NULL);
+  assert(VP8Mean16x4 != NULL);
+  assert(VP8EncQuantizeBlockWHT != NULL);
+  assert(VP8Copy4x4 != NULL);
+  assert(VP8Copy16x8 != NULL);
+
   enc_last_cpuinfo_used = VP8GetCPUInfo;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/enc_avx2.c b/src/3rdparty/libwebp/src/dsp/enc_avx2.c
index 93efb30..8bc5798 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_avx2.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_avx2.c
@@ -9,7 +9,7 @@
 //
 // AVX2 version of speed-critical encoding functions.
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_AVX2)
 
diff --git a/src/3rdparty/libwebp/src/dsp/enc_mips32.c b/src/3rdparty/libwebp/src/dsp/enc_mips32.c
index 752b14d..618f0fc 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_mips32.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_mips32.c
@@ -13,13 +13,13 @@
 //            Jovan Zelincevic (jovan.zelincevic@imgtec.com)
 //            Slobodan Prijic  (slobodan.prijic@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS32)
 
-#include "./mips_macro.h"
-#include "../enc/vp8i_enc.h"
-#include "../enc/cost_enc.h"
+#include "src/dsp/mips_macro.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/enc/cost_enc.h"
 
 static const int kC1 = 20091 + (1 << 16);
 static const int kC2 = 35468;
@@ -113,8 +113,9 @@ static const int kC2 = 35468;
   "sb      %[" #TEMP12 "],   3+" XSTR(BPS) "*" #A "(%[temp16]) \n\t"
 
 // Does one or two inverse transforms.
-static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
-                                      uint8_t* dst) {
+static WEBP_INLINE void ITransformOne_MIPS32(const uint8_t* ref,
+                                             const int16_t* in,
+                                             uint8_t* dst) {
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6;
   int temp7, temp8, temp9, temp10, temp11, temp12, temp13;
   int temp14, temp15, temp16, temp17, temp18, temp19, temp20;
@@ -144,11 +145,11 @@ static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
   );
 }
 
-static void ITransform(const uint8_t* ref, const int16_t* in,
-                       uint8_t* dst, int do_two) {
-  ITransformOne(ref, in, dst);
+static void ITransform_MIPS32(const uint8_t* ref, const int16_t* in,
+                              uint8_t* dst, int do_two) {
+  ITransformOne_MIPS32(ref, in, dst);
   if (do_two) {
-    ITransformOne(ref + 4, in + 16, dst + 4);
+    ITransformOne_MIPS32(ref + 4, in + 16, dst + 4);
   }
 }
 
@@ -187,8 +188,8 @@ static void ITransform(const uint8_t* ref, const int16_t* in,
   "sh           %[temp5],       " #J "(%[ppin])                     \n\t"   \
   "sh           %[level],       " #N "(%[pout])                     \n\t"
 
-static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         const VP8Matrix* const mtx) {
+static int QuantizeBlock_MIPS32(int16_t in[16], int16_t out[16],
+                                const VP8Matrix* const mtx) {
   int temp0, temp1, temp2, temp3, temp4, temp5;
   int sign, coeff, level, i;
   int max_level = MAX_LEVEL;
@@ -238,11 +239,11 @@ static int QuantizeBlock(int16_t in[16], int16_t out[16],
   return 0;
 }
 
-static int Quantize2Blocks(int16_t in[32], int16_t out[32],
-                           const VP8Matrix* const mtx) {
+static int Quantize2Blocks_MIPS32(int16_t in[32], int16_t out[32],
+                                  const VP8Matrix* const mtx) {
   int nz;
-  nz  = QuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
-  nz |= QuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
+  nz  = QuantizeBlock_MIPS32(in + 0 * 16, out + 0 * 16, mtx) << 0;
+  nz |= QuantizeBlock_MIPS32(in + 1 * 16, out + 1 * 16, mtx) << 1;
   return nz;
 }
 
@@ -361,8 +362,8 @@ static int Quantize2Blocks(int16_t in[32], int16_t out[32],
   "msub   %[temp6],  %[temp0]                \n\t"                \
   "msub   %[temp7],  %[temp1]                \n\t"
 
-static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
-                    const uint16_t* const w) {
+static int Disto4x4_MIPS32(const uint8_t* const a, const uint8_t* const b,
+                           const uint16_t* const w) {
   int tmp[32];
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
 
@@ -396,13 +397,13 @@ static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
 #undef VERTICAL_PASS
 #undef HORIZONTAL_PASS
 
-static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
-                      const uint16_t* const w) {
+static int Disto16x16_MIPS32(const uint8_t* const a, const uint8_t* const b,
+                             const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4(a + x + y, b + x + y, w);
+      D += Disto4x4_MIPS32(a + x + y, b + x + y, w);
     }
   }
   return D;
@@ -478,7 +479,8 @@ static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
   "sh     %[" #TEMP8 "],  " #D "(%[temp20])              \n\t"    \
   "sh     %[" #TEMP12 "], " #B "(%[temp20])              \n\t"
 
-static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+static void FTransform_MIPS32(const uint8_t* src, const uint8_t* ref,
+                              int16_t* out) {
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
   int temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
   int temp17, temp18, temp19, temp20;
@@ -539,7 +541,7 @@ static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   GET_SSE_INNER(C, C + 1, C + 2, C + 3)   \
   GET_SSE_INNER(D, D + 1, D + 2, D + 3)
 
-static int SSE16x16(const uint8_t* a, const uint8_t* b) {
+static int SSE16x16_MIPS32(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
 
@@ -573,7 +575,7 @@ static int SSE16x16(const uint8_t* a, const uint8_t* b) {
   return count;
 }
 
-static int SSE16x8(const uint8_t* a, const uint8_t* b) {
+static int SSE16x8_MIPS32(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
 
@@ -599,7 +601,7 @@ static int SSE16x8(const uint8_t* a, const uint8_t* b) {
   return count;
 }
 
-static int SSE8x8(const uint8_t* a, const uint8_t* b) {
+static int SSE8x8_MIPS32(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
 
@@ -621,7 +623,7 @@ static int SSE8x8(const uint8_t* a, const uint8_t* b) {
   return count;
 }
 
-static int SSE4x4(const uint8_t* a, const uint8_t* b) {
+static int SSE4x4_MIPS32(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
 
@@ -651,17 +653,20 @@ static int SSE4x4(const uint8_t* a, const uint8_t* b) {
 extern void VP8EncDspInitMIPS32(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitMIPS32(void) {
-  VP8ITransform = ITransform;
-  VP8FTransform = FTransform;
-  VP8EncQuantizeBlock = QuantizeBlock;
-  VP8EncQuantize2Blocks = Quantize2Blocks;
-  VP8TDisto4x4 = Disto4x4;
-  VP8TDisto16x16 = Disto16x16;
+  VP8ITransform = ITransform_MIPS32;
+  VP8FTransform = FTransform_MIPS32;
+
+  VP8EncQuantizeBlock = QuantizeBlock_MIPS32;
+  VP8EncQuantize2Blocks = Quantize2Blocks_MIPS32;
+
+  VP8TDisto4x4 = Disto4x4_MIPS32;
+  VP8TDisto16x16 = Disto16x16_MIPS32;
+
 #if !defined(WORK_AROUND_GCC)
-  VP8SSE16x16 = SSE16x16;
-  VP8SSE8x8 = SSE8x8;
-  VP8SSE16x8 = SSE16x8;
-  VP8SSE4x4 = SSE4x4;
+  VP8SSE16x16 = SSE16x16_MIPS32;
+  VP8SSE8x8 = SSE8x8_MIPS32;
+  VP8SSE16x8 = SSE16x8_MIPS32;
+  VP8SSE4x4 = SSE4x4_MIPS32;
 #endif
 }
 
diff --git a/src/3rdparty/libwebp/src/dsp/enc_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/enc_mips_dsp_r2.c
index 6c8c1c6..9ddd895 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_mips_dsp_r2.c
@@ -12,13 +12,13 @@
 // Author(s): Darko Laus (darko.laus@imgtec.com)
 //            Mirko Raus (mirko.raus@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-#include "./mips_macro.h"
-#include "../enc/cost_enc.h"
-#include "../enc/vp8i_enc.h"
+#include "src/dsp/mips_macro.h"
+#include "src/enc/cost_enc.h"
+#include "src/enc/vp8i_enc.h"
 
 static const int kC1 = 20091 + (1 << 16);
 static const int kC2 = 35468;
@@ -141,7 +141,8 @@ static const int kC2 = 35468;
   "sh              %[" #TEMP8 "],   " #D "(%[temp20])               \n\t"      \
   "sh              %[" #TEMP12 "],  " #B "(%[temp20])               \n\t"
 
-static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+static void FTransform_MIPSdspR2(const uint8_t* src, const uint8_t* ref,
+                                 int16_t* out) {
   const int c2217 = 2217;
   const int c5352 = 5352;
   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
@@ -238,16 +239,16 @@ static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
   );
 }
 
-static void ITransform(const uint8_t* ref, const int16_t* in, uint8_t* dst,
-                       int do_two) {
+static void ITransform_MIPSdspR2(const uint8_t* ref, const int16_t* in,
+                                 uint8_t* dst, int do_two) {
   ITransformOne(ref, in, dst);
   if (do_two) {
     ITransformOne(ref + 4, in + 16, dst + 4);
   }
 }
 
-static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
-                    const uint16_t* const w) {
+static int Disto4x4_MIPSdspR2(const uint8_t* const a, const uint8_t* const b,
+                              const uint16_t* const w) {
   int temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9;
   int temp10, temp11, temp12, temp13, temp14, temp15, temp16, temp17;
 
@@ -313,13 +314,14 @@ static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
   return abs(temp3 - temp17) >> 5;
 }
 
-static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
-                      const uint16_t* const w) {
+static int Disto16x16_MIPSdspR2(const uint8_t* const a,
+                                const uint8_t* const b,
+                                const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4(a + x + y, b + x + y, w);
+      D += Disto4x4_MIPSdspR2(a + x + y, b + x + y, w);
     }
   }
   return D;
@@ -1011,8 +1013,8 @@ static void HU4(uint8_t* dst, const uint8_t* top) {
 //------------------------------------------------------------------------------
 // Chroma 8x8 prediction (paragraph 12.2)
 
-static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
-                             const uint8_t* top) {
+static void IntraChromaPreds_MIPSdspR2(uint8_t* dst, const uint8_t* left,
+                                       const uint8_t* top) {
   // U block
   DCMode8(C8DC8 + dst, left, top);
   VerticalPred8(C8VE8 + dst, top);
@@ -1031,8 +1033,8 @@ static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
 //------------------------------------------------------------------------------
 // luma 16x16 prediction (paragraph 12.3)
 
-static void Intra16Preds(uint8_t* dst,
-                         const uint8_t* left, const uint8_t* top) {
+static void Intra16Preds_MIPSdspR2(uint8_t* dst,
+                                   const uint8_t* left, const uint8_t* top) {
   DCMode16(I16DC16 + dst, left, top);
   VerticalPred16(I16VE16 + dst, top);
   HorizontalPred16(I16HE16 + dst, left);
@@ -1041,7 +1043,7 @@ static void Intra16Preds(uint8_t* dst,
 
 // Left samples are top[-5 .. -2], top_left is top[-1], top are
 // located at top[0..3], and top right is top[4..7]
-static void Intra4Preds(uint8_t* dst, const uint8_t* top) {
+static void Intra4Preds_MIPSdspR2(uint8_t* dst, const uint8_t* top) {
   DC4(I4DC4 + dst, top);
   TM4(I4TM4 + dst, top);
   VE4(I4VE4 + dst, top);
@@ -1077,7 +1079,7 @@ static void Intra4Preds(uint8_t* dst, const uint8_t* top) {
   GET_SSE_INNER(C)                        \
   GET_SSE_INNER(D)
 
-static int SSE16x16(const uint8_t* a, const uint8_t* b) {
+static int SSE16x16_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3;
   __asm__ volatile (
@@ -1107,7 +1109,7 @@ static int SSE16x16(const uint8_t* a, const uint8_t* b) {
   return count;
 }
 
-static int SSE16x8(const uint8_t* a, const uint8_t* b) {
+static int SSE16x8_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3;
   __asm__ volatile (
@@ -1129,7 +1131,7 @@ static int SSE16x8(const uint8_t* a, const uint8_t* b) {
   return count;
 }
 
-static int SSE8x8(const uint8_t* a, const uint8_t* b) {
+static int SSE8x8_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3;
   __asm__ volatile (
@@ -1147,7 +1149,7 @@ static int SSE8x8(const uint8_t* a, const uint8_t* b) {
   return count;
 }
 
-static int SSE4x4(const uint8_t* a, const uint8_t* b) {
+static int SSE4x4_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
   int count;
   int temp0, temp1, temp2, temp3;
   __asm__ volatile (
@@ -1270,8 +1272,8 @@ static int SSE4x4(const uint8_t* a, const uint8_t* b) {
   "usw         $0,           " #J "(%[ppin])                 \n\t"        \
 "3:                                                          \n\t"
 
-static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         const VP8Matrix* const mtx) {
+static int QuantizeBlock_MIPSdspR2(int16_t in[16], int16_t out[16],
+                                   const VP8Matrix* const mtx) {
   int temp0, temp1, temp2, temp3, temp4, temp5,temp6;
   int sign, coeff, level;
   int max_level = MAX_LEVEL;
@@ -1311,11 +1313,11 @@ static int QuantizeBlock(int16_t in[16], int16_t out[16],
   return (ret != 0);
 }
 
-static int Quantize2Blocks(int16_t in[32], int16_t out[32],
-                           const VP8Matrix* const mtx) {
+static int Quantize2Blocks_MIPSdspR2(int16_t in[32], int16_t out[32],
+                                     const VP8Matrix* const mtx) {
   int nz;
-  nz  = QuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
-  nz |= QuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
+  nz  = QuantizeBlock_MIPSdspR2(in + 0 * 16, out + 0 * 16, mtx) << 0;
+  nz |= QuantizeBlock_MIPSdspR2(in + 1 * 16, out + 1 * 16, mtx) << 1;
   return nz;
 }
 
@@ -1358,7 +1360,7 @@ static int Quantize2Blocks(int16_t in[32], int16_t out[32],
   "usw             %[" #TEMP4 "],  " #C "(%[out])                 \n\t"        \
   "usw             %[" #TEMP6 "],  " #D "(%[out])                 \n\t"
 
-static void FTransformWHT(const int16_t* in, int16_t* out) {
+static void FTransformWHT_MIPSdspR2(const int16_t* in, int16_t* out) {
   int temp0, temp1, temp2, temp3, temp4;
   int temp5, temp6, temp7, temp8, temp9;
 
@@ -1450,9 +1452,9 @@ static void FTransformWHT(const int16_t* in, int16_t* out) {
   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
   "sw         %[temp8],  0(%[temp3])                   \n\t"
 
-static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
-                             int start_block, int end_block,
-                             VP8Histogram* const histo) {
+static void CollectHistogram_MIPSdspR2(const uint8_t* ref, const uint8_t* pred,
+                                       int start_block, int end_block,
+                                       VP8Histogram* const histo) {
   int j;
   int distribution[MAX_COEFF_THRESH + 1] = { 0 };
   const int max_coeff = (MAX_COEFF_THRESH << 16) + MAX_COEFF_THRESH;
@@ -1484,23 +1486,28 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
 extern void VP8EncDspInitMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitMIPSdspR2(void) {
-  VP8FTransform = FTransform;
-  VP8ITransform = ITransform;
-  VP8TDisto4x4 = Disto4x4;
-  VP8TDisto16x16 = Disto16x16;
-  VP8EncPredLuma16 = Intra16Preds;
-  VP8EncPredChroma8 = IntraChromaPreds;
-  VP8EncPredLuma4 = Intra4Preds;
+  VP8FTransform = FTransform_MIPSdspR2;
+  VP8FTransformWHT = FTransformWHT_MIPSdspR2;
+  VP8ITransform = ITransform_MIPSdspR2;
+
+  VP8TDisto4x4 = Disto4x4_MIPSdspR2;
+  VP8TDisto16x16 = Disto16x16_MIPSdspR2;
+
+  VP8EncPredLuma16 = Intra16Preds_MIPSdspR2;
+  VP8EncPredChroma8 = IntraChromaPreds_MIPSdspR2;
+  VP8EncPredLuma4 = Intra4Preds_MIPSdspR2;
+
 #if !defined(WORK_AROUND_GCC)
-  VP8SSE16x16 = SSE16x16;
-  VP8SSE8x8 = SSE8x8;
-  VP8SSE16x8 = SSE16x8;
-  VP8SSE4x4 = SSE4x4;
+  VP8SSE16x16 = SSE16x16_MIPSdspR2;
+  VP8SSE8x8 = SSE8x8_MIPSdspR2;
+  VP8SSE16x8 = SSE16x8_MIPSdspR2;
+  VP8SSE4x4 = SSE4x4_MIPSdspR2;
 #endif
-  VP8EncQuantizeBlock = QuantizeBlock;
-  VP8EncQuantize2Blocks = Quantize2Blocks;
-  VP8FTransformWHT = FTransformWHT;
-  VP8CollectHistogram = CollectHistogram;
+
+  VP8EncQuantizeBlock = QuantizeBlock_MIPSdspR2;
+  VP8EncQuantize2Blocks = Quantize2Blocks_MIPSdspR2;
+
+  VP8CollectHistogram = CollectHistogram_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/enc_msa.c b/src/3rdparty/libwebp/src/dsp/enc_msa.c
index 909b46d..6f85add 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_msa.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_msa.c
@@ -11,13 +11,13 @@
 //
 // Author:  Prashant Patil   (prashant.patil@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MSA)
 
 #include <stdlib.h>
-#include "./msa_macro.h"
-#include "../enc/vp8i_enc.h"
+#include "src/dsp/msa_macro.h"
+#include "src/enc/vp8i_enc.h"
 
 //------------------------------------------------------------------------------
 // Transforms
@@ -69,20 +69,21 @@ static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
   ST4x4_UB(res0, res0, 3, 2, 1, 0, dst, BPS);
 }
 
-static void ITransform(const uint8_t* ref, const int16_t* in, uint8_t* dst,
-                       int do_two) {
+static void ITransform_MSA(const uint8_t* ref, const int16_t* in, uint8_t* dst,
+                           int do_two) {
   ITransformOne(ref, in, dst);
   if (do_two) {
     ITransformOne(ref + 4, in + 16, dst + 4);
   }
 }
 
-static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+static void FTransform_MSA(const uint8_t* src, const uint8_t* ref,
+                           int16_t* out) {
   uint64_t out0, out1, out2, out3;
   uint32_t in0, in1, in2, in3;
   v4i32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
   v8i16 t0, t1, t2, t3;
-  v16u8 srcl0, srcl1, src0, src1;
+  v16u8 srcl0, srcl1, src0 = { 0 }, src1 = { 0 };
   const v8i16 mask0 = { 0, 4, 8, 12, 1, 5, 9, 13 };
   const v8i16 mask1 = { 3, 7, 11, 15, 2, 6, 10, 14 };
   const v8i16 mask2 = { 4, 0, 5, 1, 6, 2, 7, 3 };
@@ -130,7 +131,7 @@ static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   SD4(out0, out1, out2, out3, out, 8);
 }
 
-static void FTransformWHT(const int16_t* in, int16_t* out) {
+static void FTransformWHT_MSA(const int16_t* in, int16_t* out) {
   v8i16 in0 = { 0 };
   v8i16 in1 = { 0 };
   v8i16 tmp0, tmp1, tmp2, tmp3;
@@ -167,10 +168,10 @@ static void FTransformWHT(const int16_t* in, int16_t* out) {
   ST_SH2(out0, out1, out, 8);
 }
 
-static int TTransform(const uint8_t* in, const uint16_t* w) {
+static int TTransform_MSA(const uint8_t* in, const uint16_t* w) {
   int sum;
   uint32_t in0_m, in1_m, in2_m, in3_m;
-  v16i8 src0;
+  v16i8 src0 = { 0 };
   v8i16 in0, in1, tmp0, tmp1, tmp2, tmp3;
   v4i32 dst0, dst1;
   const v16i8 zero = { 0 };
@@ -199,20 +200,20 @@ static int TTransform(const uint8_t* in, const uint16_t* w) {
   return sum;
 }
 
-static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
-                    const uint16_t* const w) {
-  const int sum1 = TTransform(a, w);
-  const int sum2 = TTransform(b, w);
+static int Disto4x4_MSA(const uint8_t* const a, const uint8_t* const b,
+                        const uint16_t* const w) {
+  const int sum1 = TTransform_MSA(a, w);
+  const int sum2 = TTransform_MSA(b, w);
   return abs(sum2 - sum1) >> 5;
 }
 
-static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
-                      const uint16_t* const w) {
+static int Disto16x16_MSA(const uint8_t* const a, const uint8_t* const b,
+                          const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4(a + x + y, b + x + y, w);
+      D += Disto4x4_MSA(a + x + y, b + x + y, w);
     }
   }
   return D;
@@ -221,9 +222,9 @@ static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
 //------------------------------------------------------------------------------
 // Histogram
 
-static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
-                             int start_block, int end_block,
-                             VP8Histogram* const histo) {
+static void CollectHistogram_MSA(const uint8_t* ref, const uint8_t* pred,
+                                 int start_block, int end_block,
+                                 VP8Histogram* const histo) {
   int j;
   int distribution[MAX_COEFF_THRESH + 1] = { 0 };
   for (j = start_block; j < end_block; ++j) {
@@ -259,8 +260,9 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
 
 static WEBP_INLINE void VE4(uint8_t* dst, const uint8_t* top) {    // vertical
+  const v16u8 A1 = { 0 };
   const uint64_t val_m = LD(top - 1);
-  const v16u8 A = (v16u8)__msa_insert_d((v2i64)A, 0, val_m);
+  const v16u8 A = (v16u8)__msa_insert_d((v2i64)A1, 0, val_m);
   const v16u8 B = SLDI_UB(A, A, 1);
   const v16u8 C = SLDI_UB(A, A, 2);
   const v16u8 AC = __msa_ave_u_b(A, C);
@@ -292,8 +294,9 @@ static WEBP_INLINE void DC4(uint8_t* dst, const uint8_t* top) {
 }
 
 static WEBP_INLINE void RD4(uint8_t* dst, const uint8_t* top) {
+  const v16u8 A2 = { 0 };
   const uint64_t val_m = LD(top - 5);
-  const v16u8 A1 = (v16u8)__msa_insert_d((v2i64)A1, 0, val_m);
+  const v16u8 A1 = (v16u8)__msa_insert_d((v2i64)A2, 0, val_m);
   const v16u8 A = (v16u8)__msa_insert_b((v16i8)A1, 8, top[3]);
   const v16u8 B = SLDI_UB(A, A, 1);
   const v16u8 C = SLDI_UB(A, A, 2);
@@ -311,8 +314,9 @@ static WEBP_INLINE void RD4(uint8_t* dst, const uint8_t* top) {
 }
 
 static WEBP_INLINE void LD4(uint8_t* dst, const uint8_t* top) {
+  const v16u8 A1 = { 0 };
   const uint64_t val_m = LD(top);
-  const v16u8 A = (v16u8)__msa_insert_d((v2i64)A, 0, val_m);
+  const v16u8 A = (v16u8)__msa_insert_d((v2i64)A1, 0, val_m);
   const v16u8 B = SLDI_UB(A, A, 1);
   const v16u8 C1 = SLDI_UB(A, A, 2);
   const v16u8 C = (v16u8)__msa_insert_b((v16i8)C1, 6, top[7]);
@@ -427,7 +431,7 @@ static WEBP_INLINE void TM4(uint8_t* dst, const uint8_t* top) {
 #undef AVG3
 #undef AVG2
 
-static void Intra4Preds(uint8_t* dst, const uint8_t* top) {
+static void Intra4Preds_MSA(uint8_t* dst, const uint8_t* top) {
   DC4(I4DC4 + dst, top);
   TM4(I4TM4 + dst, top);
   VE4(I4VE4 + dst, top);
@@ -544,8 +548,8 @@ static WEBP_INLINE void DCMode16x16(uint8_t* dst, const uint8_t* left,
   STORE16x16(out, dst);
 }
 
-static void Intra16Preds(uint8_t* dst,
-                         const uint8_t* left, const uint8_t* top) {
+static void Intra16Preds_MSA(uint8_t* dst,
+                             const uint8_t* left, const uint8_t* top) {
   DCMode16x16(I16DC16 + dst, left, top);
   VerticalPred16x16(I16VE16 + dst, top);
   HorizontalPred16x16(I16HE16 + dst, left);
@@ -645,7 +649,7 @@ static WEBP_INLINE void TrueMotion8x8(uint8_t* dst, const uint8_t* left,
 static WEBP_INLINE void DCMode8x8(uint8_t* dst, const uint8_t* left,
                                   const uint8_t* top) {
   uint64_t out;
-  v16u8 src;
+  v16u8 src = { 0 };
   if (top != NULL && left != NULL) {
     const uint64_t left_m = LD(left);
     const uint64_t top_m = LD(top);
@@ -666,8 +670,8 @@ static WEBP_INLINE void DCMode8x8(uint8_t* dst, const uint8_t* left,
   STORE8x8(out, dst);
 }
 
-static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
-                             const uint8_t* top) {
+static void IntraChromaPreds_MSA(uint8_t* dst, const uint8_t* left,
+                                 const uint8_t* top) {
   // U block
   DCMode8x8(C8DC8 + dst, left, top);
   VerticalPred8x8(C8VE8 + dst, top);
@@ -708,7 +712,7 @@ static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
   DPADD_SH2_SW(tmp2, tmp3, tmp2, tmp3, out2, out3);                         \
 } while (0)
 
-static int SSE16x16(const uint8_t* a, const uint8_t* b) {
+static int SSE16x16_MSA(const uint8_t* a, const uint8_t* b) {
   uint32_t sum;
   v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
   v16u8 ref0, ref1, ref2, ref3, ref4, ref5, ref6, ref7;
@@ -735,7 +739,7 @@ static int SSE16x16(const uint8_t* a, const uint8_t* b) {
   return sum;
 }
 
-static int SSE16x8(const uint8_t* a, const uint8_t* b) {
+static int SSE16x8_MSA(const uint8_t* a, const uint8_t* b) {
   uint32_t sum;
   v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
   v16u8 ref0, ref1, ref2, ref3, ref4, ref5, ref6, ref7;
@@ -754,7 +758,7 @@ static int SSE16x8(const uint8_t* a, const uint8_t* b) {
   return sum;
 }
 
-static int SSE8x8(const uint8_t* a, const uint8_t* b) {
+static int SSE8x8_MSA(const uint8_t* a, const uint8_t* b) {
   uint32_t sum;
   v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
   v16u8 ref0, ref1, ref2, ref3, ref4, ref5, ref6, ref7;
@@ -774,10 +778,10 @@ static int SSE8x8(const uint8_t* a, const uint8_t* b) {
   return sum;
 }
 
-static int SSE4x4(const uint8_t* a, const uint8_t* b) {
+static int SSE4x4_MSA(const uint8_t* a, const uint8_t* b) {
   uint32_t sum = 0;
   uint32_t src0, src1, src2, src3, ref0, ref1, ref2, ref3;
-  v16u8 src, ref, tmp0, tmp1;
+  v16u8 src = { 0 }, ref = { 0 }, tmp0, tmp1;
   v8i16 diff0, diff1;
   v4i32 out0, out1;
 
@@ -796,8 +800,8 @@ static int SSE4x4(const uint8_t* a, const uint8_t* b) {
 //------------------------------------------------------------------------------
 // Quantization
 
-static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         const VP8Matrix* const mtx) {
+static int QuantizeBlock_MSA(int16_t in[16], int16_t out[16],
+                             const VP8Matrix* const mtx) {
   int sum;
   v8i16 in0, in1, sh0, sh1, out0, out1;
   v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, sign0, sign1;
@@ -828,7 +832,7 @@ static int QuantizeBlock(int16_t in[16], int16_t out[16],
   tmp1 = (tmp3 > maxlevel);
   tmp2 = (v8i16)__msa_bmnz_v((v16u8)tmp2, (v16u8)maxlevel, (v16u8)tmp0);
   tmp3 = (v8i16)__msa_bmnz_v((v16u8)tmp3, (v16u8)maxlevel, (v16u8)tmp1);
-  SUB2(0, tmp2, 0, tmp3, tmp0, tmp1);
+  SUB2(zero, tmp2, zero, tmp3, tmp0, tmp1);
   tmp2 = (v8i16)__msa_bmnz_v((v16u8)tmp2, (v16u8)tmp0, (v16u8)sign0);
   tmp3 = (v8i16)__msa_bmnz_v((v16u8)tmp3, (v16u8)tmp1, (v16u8)sign1);
   LD_SW4(&mtx->zthresh_[0], 4, t0, t1, t2, t3);   // zthresh
@@ -849,8 +853,8 @@ static int QuantizeBlock(int16_t in[16], int16_t out[16],
   return (sum > 0);
 }
 
-static int Quantize2Blocks(int16_t in[32], int16_t out[32],
-                           const VP8Matrix* const mtx) {
+static int Quantize2Blocks_MSA(int16_t in[32], int16_t out[32],
+                               const VP8Matrix* const mtx) {
   int nz;
   nz  = VP8EncQuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
   nz |= VP8EncQuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
@@ -863,26 +867,26 @@ static int Quantize2Blocks(int16_t in[32], int16_t out[32],
 extern void VP8EncDspInitMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitMSA(void) {
-  VP8ITransform = ITransform;
-  VP8FTransform = FTransform;
-  VP8FTransformWHT = FTransformWHT;
-
-  VP8TDisto4x4 = Disto4x4;
-  VP8TDisto16x16 = Disto16x16;
-  VP8CollectHistogram = CollectHistogram;
-
-  VP8EncPredLuma4 = Intra4Preds;
-  VP8EncPredLuma16 = Intra16Preds;
-  VP8EncPredChroma8 = IntraChromaPreds;
-
-  VP8SSE16x16 = SSE16x16;
-  VP8SSE16x8 = SSE16x8;
-  VP8SSE8x8 = SSE8x8;
-  VP8SSE4x4 = SSE4x4;
-
-  VP8EncQuantizeBlock = QuantizeBlock;
-  VP8EncQuantize2Blocks = Quantize2Blocks;
-  VP8EncQuantizeBlockWHT = QuantizeBlock;
+  VP8ITransform = ITransform_MSA;
+  VP8FTransform = FTransform_MSA;
+  VP8FTransformWHT = FTransformWHT_MSA;
+
+  VP8TDisto4x4 = Disto4x4_MSA;
+  VP8TDisto16x16 = Disto16x16_MSA;
+  VP8CollectHistogram = CollectHistogram_MSA;
+
+  VP8EncPredLuma4 = Intra4Preds_MSA;
+  VP8EncPredLuma16 = Intra16Preds_MSA;
+  VP8EncPredChroma8 = IntraChromaPreds_MSA;
+
+  VP8SSE16x16 = SSE16x16_MSA;
+  VP8SSE16x8 = SSE16x8_MSA;
+  VP8SSE8x8 = SSE8x8_MSA;
+  VP8SSE4x4 = SSE4x4_MSA;
+
+  VP8EncQuantizeBlock = QuantizeBlock_MSA;
+  VP8EncQuantize2Blocks = Quantize2Blocks_MSA;
+  VP8EncQuantizeBlockWHT = QuantizeBlock_MSA;
 }
 
 #else  // !WEBP_USE_MSA
diff --git a/src/3rdparty/libwebp/src/dsp/enc_neon.c b/src/3rdparty/libwebp/src/dsp/enc_neon.c
index 6a078d6..43bf124 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_neon.c
@@ -11,14 +11,14 @@
 //
 // adapted from libvpx (http://www.webmproject.org/code/)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_NEON)
 
 #include <assert.h>
 
-#include "./neon.h"
-#include "../enc/vp8i_enc.h"
+#include "src/dsp/neon.h"
+#include "src/enc/vp8i_enc.h"
 
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
@@ -37,15 +37,15 @@ static const int16_t kC2 = 17734;  // half of kC2, actually. See comment above.
 #if defined(WEBP_USE_INTRINSICS)
 
 // Treats 'v' as an uint8x8_t and zero extends to an int16x8_t.
-static WEBP_INLINE int16x8_t ConvertU8ToS16(uint32x2_t v) {
+static WEBP_INLINE int16x8_t ConvertU8ToS16_NEON(uint32x2_t v) {
   return vreinterpretq_s16_u16(vmovl_u8(vreinterpret_u8_u32(v)));
 }
 
 // Performs unsigned 8b saturation on 'dst01' and 'dst23' storing the result
 // to the corresponding rows of 'dst'.
-static WEBP_INLINE void SaturateAndStore4x4(uint8_t* const dst,
-                                            const int16x8_t dst01,
-                                            const int16x8_t dst23) {
+static WEBP_INLINE void SaturateAndStore4x4_NEON(uint8_t* const dst,
+                                                 const int16x8_t dst01,
+                                                 const int16x8_t dst23) {
   // Unsigned saturate to 8b.
   const uint8x8_t dst01_u8 = vqmovun_s16(dst01);
   const uint8x8_t dst23_u8 = vqmovun_s16(dst23);
@@ -57,8 +57,10 @@ static WEBP_INLINE void SaturateAndStore4x4(uint8_t* const dst,
   vst1_lane_u32((uint32_t*)(dst + 3 * BPS), vreinterpret_u32_u8(dst23_u8), 1);
 }
 
-static WEBP_INLINE void Add4x4(const int16x8_t row01, const int16x8_t row23,
-                               const uint8_t* const ref, uint8_t* const dst) {
+static WEBP_INLINE void Add4x4_NEON(const int16x8_t row01,
+                                    const int16x8_t row23,
+                                    const uint8_t* const ref,
+                                    uint8_t* const dst) {
   uint32x2_t dst01 = vdup_n_u32(0);
   uint32x2_t dst23 = vdup_n_u32(0);
 
@@ -70,19 +72,20 @@ static WEBP_INLINE void Add4x4(const int16x8_t row01, const int16x8_t row23,
 
   {
     // Convert to 16b.
-    const int16x8_t dst01_s16 = ConvertU8ToS16(dst01);
-    const int16x8_t dst23_s16 = ConvertU8ToS16(dst23);
+    const int16x8_t dst01_s16 = ConvertU8ToS16_NEON(dst01);
+    const int16x8_t dst23_s16 = ConvertU8ToS16_NEON(dst23);
 
     // Descale with rounding.
     const int16x8_t out01 = vrsraq_n_s16(dst01_s16, row01, 3);
     const int16x8_t out23 = vrsraq_n_s16(dst23_s16, row23, 3);
     // Add the inverse transform.
-    SaturateAndStore4x4(dst, out01, out23);
+    SaturateAndStore4x4_NEON(dst, out01, out23);
   }
 }
 
-static WEBP_INLINE void Transpose8x2(const int16x8_t in0, const int16x8_t in1,
-                                     int16x8x2_t* const out) {
+static WEBP_INLINE void Transpose8x2_NEON(const int16x8_t in0,
+                                          const int16x8_t in1,
+                                          int16x8x2_t* const out) {
   // a0 a1 a2 a3 | b0 b1 b2 b3   => a0 b0 c0 d0 | a1 b1 c1 d1
   // c0 c1 c2 c3 | d0 d1 d2 d3      a2 b2 c2 d2 | a3 b3 c3 d3
   const int16x8x2_t tmp0 = vzipq_s16(in0, in1);   // a0 c0 a1 c1 a2 c2 ...
@@ -90,7 +93,7 @@ static WEBP_INLINE void Transpose8x2(const int16x8_t in0, const int16x8_t in1,
   *out = vzipq_s16(tmp0.val[0], tmp0.val[1]);
 }
 
-static WEBP_INLINE void TransformPass(int16x8x2_t* const rows) {
+static WEBP_INLINE void TransformPass_NEON(int16x8x2_t* const rows) {
   // {rows} = in0 | in4
   //          in8 | in12
   // B1 = in4 | in12
@@ -113,22 +116,22 @@ static WEBP_INLINE void TransformPass(int16x8x2_t* const rows) {
   const int16x8_t E0 = vqaddq_s16(D0, D1);      // a+d | b+c
   const int16x8_t E_tmp = vqsubq_s16(D0, D1);   // a-d | b-c
   const int16x8_t E1 = vcombine_s16(vget_high_s16(E_tmp), vget_low_s16(E_tmp));
-  Transpose8x2(E0, E1, rows);
+  Transpose8x2_NEON(E0, E1, rows);
 }
 
-static void ITransformOne(const uint8_t* ref,
-                          const int16_t* in, uint8_t* dst) {
+static void ITransformOne_NEON(const uint8_t* ref,
+                               const int16_t* in, uint8_t* dst) {
   int16x8x2_t rows;
   INIT_VECTOR2(rows, vld1q_s16(in + 0), vld1q_s16(in + 8));
-  TransformPass(&rows);
-  TransformPass(&rows);
-  Add4x4(rows.val[0], rows.val[1], ref, dst);
+  TransformPass_NEON(&rows);
+  TransformPass_NEON(&rows);
+  Add4x4_NEON(rows.val[0], rows.val[1], ref, dst);
 }
 
 #else
 
-static void ITransformOne(const uint8_t* ref,
-                          const int16_t* in, uint8_t* dst) {
+static void ITransformOne_NEON(const uint8_t* ref,
+                               const int16_t* in, uint8_t* dst) {
   const int kBPS = BPS;
   const int16_t kC1C2[] = { kC1, kC2, 0, 0 };
 
@@ -243,16 +246,16 @@ static void ITransformOne(const uint8_t* ref,
 
 #endif    // WEBP_USE_INTRINSICS
 
-static void ITransform(const uint8_t* ref,
-                       const int16_t* in, uint8_t* dst, int do_two) {
-  ITransformOne(ref, in, dst);
+static void ITransform_NEON(const uint8_t* ref,
+                            const int16_t* in, uint8_t* dst, int do_two) {
+  ITransformOne_NEON(ref, in, dst);
   if (do_two) {
-    ITransformOne(ref + 4, in + 16, dst + 4);
+    ITransformOne_NEON(ref + 4, in + 16, dst + 4);
   }
 }
 
 // Load all 4x4 pixels into a single uint8x16_t variable.
-static uint8x16_t Load4x4(const uint8_t* src) {
+static uint8x16_t Load4x4_NEON(const uint8_t* src) {
   uint32x4_t out = vdupq_n_u32(0);
   out = vld1q_lane_u32((const uint32_t*)(src + 0 * BPS), out, 0);
   out = vld1q_lane_u32((const uint32_t*)(src + 1 * BPS), out, 1);
@@ -265,10 +268,12 @@ static uint8x16_t Load4x4(const uint8_t* src) {
 
 #if defined(WEBP_USE_INTRINSICS)
 
-static WEBP_INLINE void Transpose4x4_S16(const int16x4_t A, const int16x4_t B,
-                                         const int16x4_t C, const int16x4_t D,
-                                         int16x8_t* const out01,
-                                         int16x8_t* const out32) {
+static WEBP_INLINE void Transpose4x4_S16_NEON(const int16x4_t A,
+                                              const int16x4_t B,
+                                              const int16x4_t C,
+                                              const int16x4_t D,
+                                              int16x8_t* const out01,
+                                              int16x8_t* const out32) {
   const int16x4x2_t AB = vtrn_s16(A, B);
   const int16x4x2_t CD = vtrn_s16(C, D);
   const int32x2x2_t tmp02 = vtrn_s32(vreinterpret_s32_s16(AB.val[0]),
@@ -283,24 +288,24 @@ static WEBP_INLINE void Transpose4x4_S16(const int16x4_t A, const int16x4_t B,
                    vreinterpret_s64_s32(tmp02.val[1])));
 }
 
-static WEBP_INLINE int16x8_t DiffU8ToS16(const uint8x8_t a,
-                                         const uint8x8_t b) {
+static WEBP_INLINE int16x8_t DiffU8ToS16_NEON(const uint8x8_t a,
+                                              const uint8x8_t b) {
   return vreinterpretq_s16_u16(vsubl_u8(a, b));
 }
 
-static void FTransform(const uint8_t* src, const uint8_t* ref,
-                       int16_t* out) {
+static void FTransform_NEON(const uint8_t* src, const uint8_t* ref,
+                            int16_t* out) {
   int16x8_t d0d1, d3d2;   // working 4x4 int16 variables
   {
-    const uint8x16_t S0 = Load4x4(src);
-    const uint8x16_t R0 = Load4x4(ref);
-    const int16x8_t D0D1 = DiffU8ToS16(vget_low_u8(S0), vget_low_u8(R0));
-    const int16x8_t D2D3 = DiffU8ToS16(vget_high_u8(S0), vget_high_u8(R0));
+    const uint8x16_t S0 = Load4x4_NEON(src);
+    const uint8x16_t R0 = Load4x4_NEON(ref);
+    const int16x8_t D0D1 = DiffU8ToS16_NEON(vget_low_u8(S0), vget_low_u8(R0));
+    const int16x8_t D2D3 = DiffU8ToS16_NEON(vget_high_u8(S0), vget_high_u8(R0));
     const int16x4_t D0 = vget_low_s16(D0D1);
     const int16x4_t D1 = vget_high_s16(D0D1);
     const int16x4_t D2 = vget_low_s16(D2D3);
     const int16x4_t D3 = vget_high_s16(D2D3);
-    Transpose4x4_S16(D0, D1, D2, D3, &d0d1, &d3d2);
+    Transpose4x4_S16_NEON(D0, D1, D2, D3, &d0d1, &d3d2);
   }
   {    // 1rst pass
     const int32x4_t kCst937 = vdupq_n_s32(937);
@@ -318,7 +323,7 @@ static void FTransform(const uint8_t* src, const uint8_t* ref,
     const int32x4_t a3_m_a2 = vmlsl_n_s16(a3_2217, vget_high_s16(a3a2), 5352);
     const int16x4_t tmp1 = vshrn_n_s32(vaddq_s32(a2_p_a3, kCst1812), 9);
     const int16x4_t tmp3 = vshrn_n_s32(vaddq_s32(a3_m_a2, kCst937), 9);
-    Transpose4x4_S16(tmp0, tmp1, tmp2, tmp3, &d0d1, &d3d2);
+    Transpose4x4_S16_NEON(tmp0, tmp1, tmp2, tmp3, &d0d1, &d3d2);
   }
   {    // 2nd pass
     // the (1<<16) addition is for the replacement: a3!=0  <-> 1-(a3==0)
@@ -358,8 +363,8 @@ static const int32_t kCoeff32[] = {
   51000, 51000, 51000, 51000
 };
 
-static void FTransform(const uint8_t* src, const uint8_t* ref,
-                       int16_t* out) {
+static void FTransform_NEON(const uint8_t* src, const uint8_t* ref,
+                            int16_t* out) {
   const int kBPS = BPS;
   const uint8_t* src_ptr = src;
   const uint8_t* ref_ptr = ref;
@@ -478,7 +483,7 @@ static void FTransform(const uint8_t* src, const uint8_t* ref,
   src += stride;                                    \
 } while (0)
 
-static void FTransformWHT(const int16_t* src, int16_t* out) {
+static void FTransformWHT_NEON(const int16_t* src, int16_t* out) {
   const int stride = 16;
   const int16x4_t zero = vdup_n_s16(0);
   int32x4x4_t tmp0;
@@ -516,7 +521,7 @@ static void FTransformWHT(const int16_t* src, int16_t* out) {
     tmp0.val[3] = vsubq_s32(a0, a1);
   }
   {
-    const int32x4x4_t tmp1 = Transpose4x4(tmp0);
+    const int32x4x4_t tmp1 = Transpose4x4_NEON(tmp0);
     // a0 = tmp[0 + i] + tmp[ 8 + i]
     // a1 = tmp[4 + i] + tmp[12 + i]
     // a2 = tmp[4 + i] - tmp[12 + i]
@@ -560,7 +565,7 @@ static void FTransformWHT(const int16_t* src, int16_t* out) {
 // a 26ae, b 26ae
 // a 37bf, b 37bf
 //
-static WEBP_INLINE int16x8x4_t DistoTranspose4x4S16(int16x8x4_t q4_in) {
+static WEBP_INLINE int16x8x4_t DistoTranspose4x4S16_NEON(int16x8x4_t q4_in) {
   const int16x8x2_t q2_tmp0 = vtrnq_s16(q4_in.val[0], q4_in.val[1]);
   const int16x8x2_t q2_tmp1 = vtrnq_s16(q4_in.val[2], q4_in.val[3]);
   const int32x4x2_t q2_tmp2 = vtrnq_s32(vreinterpretq_s32_s16(q2_tmp0.val[0]),
@@ -574,7 +579,8 @@ static WEBP_INLINE int16x8x4_t DistoTranspose4x4S16(int16x8x4_t q4_in) {
   return q4_in;
 }
 
-static WEBP_INLINE int16x8x4_t DistoHorizontalPass(const int16x8x4_t q4_in) {
+static WEBP_INLINE int16x8x4_t DistoHorizontalPass_NEON(
+    const int16x8x4_t q4_in) {
   // {a0, a1} = {in[0] + in[2], in[1] + in[3]}
   // {a3, a2} = {in[0] - in[2], in[1] - in[3]}
   const int16x8_t q_a0 = vaddq_s16(q4_in.val[0], q4_in.val[2]);
@@ -593,7 +599,7 @@ static WEBP_INLINE int16x8x4_t DistoHorizontalPass(const int16x8x4_t q4_in) {
   return q4_out;
 }
 
-static WEBP_INLINE int16x8x4_t DistoVerticalPass(const uint8x8x4_t q4_in) {
+static WEBP_INLINE int16x8x4_t DistoVerticalPass_NEON(const uint8x8x4_t q4_in) {
   const int16x8_t q_a0 = vreinterpretq_s16_u16(vaddl_u8(q4_in.val[0],
                                                         q4_in.val[2]));
   const int16x8_t q_a1 = vreinterpretq_s16_u16(vaddl_u8(q4_in.val[1],
@@ -610,7 +616,7 @@ static WEBP_INLINE int16x8x4_t DistoVerticalPass(const uint8x8x4_t q4_in) {
   return q4_out;
 }
 
-static WEBP_INLINE int16x4x4_t DistoLoadW(const uint16_t* w) {
+static WEBP_INLINE int16x4x4_t DistoLoadW_NEON(const uint16_t* w) {
   const uint16x8_t q_w07 = vld1q_u16(&w[0]);
   const uint16x8_t q_w8f = vld1q_u16(&w[8]);
   int16x4x4_t d4_w;
@@ -622,8 +628,8 @@ static WEBP_INLINE int16x4x4_t DistoLoadW(const uint16_t* w) {
   return d4_w;
 }
 
-static WEBP_INLINE int32x2_t DistoSum(const int16x8x4_t q4_in,
-                                      const int16x4x4_t d4_w) {
+static WEBP_INLINE int32x2_t DistoSum_NEON(const int16x8x4_t q4_in,
+                                           const int16x4x4_t d4_w) {
   int32x2_t d_sum;
   // sum += w[ 0] * abs(b0);
   // sum += w[ 4] * abs(b1);
@@ -652,8 +658,8 @@ static WEBP_INLINE int32x2_t DistoSum(const int16x8x4_t q4_in,
 // Hadamard transform
 // Returns the weighted sum of the absolute value of transformed coefficients.
 // w[] contains a row-major 4 by 4 symmetric matrix.
-static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
-                    const uint16_t* const w) {
+static int Disto4x4_NEON(const uint8_t* const a, const uint8_t* const b,
+                         const uint16_t* const w) {
   uint32x2_t d_in_ab_0123 = vdup_n_u32(0);
   uint32x2_t d_in_ab_4567 = vdup_n_u32(0);
   uint32x2_t d_in_ab_89ab = vdup_n_u32(0);
@@ -679,12 +685,12 @@ static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
     // Vertical pass first to avoid a transpose (vertical and horizontal passes
     // are commutative because w/kWeightY is symmetric) and subsequent
     // transpose.
-    const int16x8x4_t q4_v = DistoVerticalPass(d4_in);
-    const int16x4x4_t d4_w = DistoLoadW(w);
+    const int16x8x4_t q4_v = DistoVerticalPass_NEON(d4_in);
+    const int16x4x4_t d4_w = DistoLoadW_NEON(w);
     // horizontal pass
-    const int16x8x4_t q4_t = DistoTranspose4x4S16(q4_v);
-    const int16x8x4_t q4_h = DistoHorizontalPass(q4_t);
-    int32x2_t d_sum = DistoSum(q4_h, d4_w);
+    const int16x8x4_t q4_t = DistoTranspose4x4S16_NEON(q4_v);
+    const int16x8x4_t q4_h = DistoHorizontalPass_NEON(q4_t);
+    int32x2_t d_sum = DistoSum_NEON(q4_h, d4_w);
 
     // abs(sum2 - sum1) >> 5
     d_sum = vabs_s32(d_sum);
@@ -694,13 +700,13 @@ static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
 }
 #undef LOAD_LANE_32b
 
-static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
-                      const uint16_t* const w) {
+static int Disto16x16_NEON(const uint8_t* const a, const uint8_t* const b,
+                           const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4(a + x + y, b + x + y, w);
+      D += Disto4x4_NEON(a + x + y, b + x + y, w);
     }
   }
   return D;
@@ -708,15 +714,15 @@ static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
 
 //------------------------------------------------------------------------------
 
-static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
-                             int start_block, int end_block,
-                             VP8Histogram* const histo) {
+static void CollectHistogram_NEON(const uint8_t* ref, const uint8_t* pred,
+                                  int start_block, int end_block,
+                                  VP8Histogram* const histo) {
   const uint16x8_t max_coeff_thresh = vdupq_n_u16(MAX_COEFF_THRESH);
   int j;
   int distribution[MAX_COEFF_THRESH + 1] = { 0 };
   for (j = start_block; j < end_block; ++j) {
     int16_t out[16];
-    FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
+    FTransform_NEON(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
     {
       int k;
       const int16x8_t a0 = vld1q_s16(out + 0);
@@ -740,9 +746,9 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
 
 //------------------------------------------------------------------------------
 
-static WEBP_INLINE void AccumulateSSE16(const uint8_t* const a,
-                                        const uint8_t* const b,
-                                        uint32x4_t* const sum) {
+static WEBP_INLINE void AccumulateSSE16_NEON(const uint8_t* const a,
+                                             const uint8_t* const b,
+                                             uint32x4_t* const sum) {
   const uint8x16_t a0 = vld1q_u8(a);
   const uint8x16_t b0 = vld1q_u8(b);
   const uint8x16_t abs_diff = vabdq_u8(a0, b0);
@@ -757,7 +763,7 @@ static WEBP_INLINE void AccumulateSSE16(const uint8_t* const a,
 }
 
 // Horizontal sum of all four uint32_t values in 'sum'.
-static int SumToInt(uint32x4_t sum) {
+static int SumToInt_NEON(uint32x4_t sum) {
   const uint64x2_t sum2 = vpaddlq_u32(sum);
   const uint64_t sum3 = vgetq_lane_u64(sum2, 0) + vgetq_lane_u64(sum2, 1);
   return (int)sum3;
@@ -767,18 +773,18 @@ static int SSE16x16_NEON(const uint8_t* a, const uint8_t* b) {
   uint32x4_t sum = vdupq_n_u32(0);
   int y;
   for (y = 0; y < 16; ++y) {
-    AccumulateSSE16(a + y * BPS, b + y * BPS, &sum);
+    AccumulateSSE16_NEON(a + y * BPS, b + y * BPS, &sum);
   }
-  return SumToInt(sum);
+  return SumToInt_NEON(sum);
 }
 
 static int SSE16x8_NEON(const uint8_t* a, const uint8_t* b) {
   uint32x4_t sum = vdupq_n_u32(0);
   int y;
   for (y = 0; y < 8; ++y) {
-    AccumulateSSE16(a + y * BPS, b + y * BPS, &sum);
+    AccumulateSSE16_NEON(a + y * BPS, b + y * BPS, &sum);
   }
-  return SumToInt(sum);
+  return SumToInt_NEON(sum);
 }
 
 static int SSE8x8_NEON(const uint8_t* a, const uint8_t* b) {
@@ -791,12 +797,12 @@ static int SSE8x8_NEON(const uint8_t* a, const uint8_t* b) {
     const uint16x8_t prod = vmull_u8(abs_diff, abs_diff);
     sum = vpadalq_u16(sum, prod);
   }
-  return SumToInt(sum);
+  return SumToInt_NEON(sum);
 }
 
 static int SSE4x4_NEON(const uint8_t* a, const uint8_t* b) {
-  const uint8x16_t a0 = Load4x4(a);
-  const uint8x16_t b0 = Load4x4(b);
+  const uint8x16_t a0 = Load4x4_NEON(a);
+  const uint8x16_t b0 = Load4x4_NEON(b);
   const uint8x16_t abs_diff = vabdq_u8(a0, b0);
   const uint16x8_t prod1 = vmull_u8(vget_low_u8(abs_diff),
                                     vget_low_u8(abs_diff));
@@ -805,7 +811,7 @@ static int SSE4x4_NEON(const uint8_t* a, const uint8_t* b) {
   /* pair-wise adds and widen */
   const uint32x4_t sum1 = vpaddlq_u16(prod1);
   const uint32x4_t sum2 = vpaddlq_u16(prod2);
-  return SumToInt(vaddq_u32(sum1, sum2));
+  return SumToInt_NEON(vaddq_u32(sum1, sum2));
 }
 
 //------------------------------------------------------------------------------
@@ -813,8 +819,8 @@ static int SSE4x4_NEON(const uint8_t* a, const uint8_t* b) {
 // Compilation with gcc-4.6.x is problematic for now.
 #if !defined(WORK_AROUND_GCC)
 
-static int16x8_t Quantize(int16_t* const in,
-                          const VP8Matrix* const mtx, int offset) {
+static int16x8_t Quantize_NEON(int16_t* const in,
+                               const VP8Matrix* const mtx, int offset) {
   const uint16x8_t sharp = vld1q_u16(&mtx->sharpen_[offset]);
   const uint16x8_t q = vld1q_u16(&mtx->q_[offset]);
   const uint16x8_t iq = vld1q_u16(&mtx->iq_[offset]);
@@ -847,10 +853,10 @@ static const uint8_t kShuffles[4][8] = {
   { 14, 15, 22, 23, 28, 29, 30, 31 }
 };
 
-static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         const VP8Matrix* const mtx) {
-  const int16x8_t out0 = Quantize(in, mtx, 0);
-  const int16x8_t out1 = Quantize(in, mtx, 8);
+static int QuantizeBlock_NEON(int16_t in[16], int16_t out[16],
+                              const VP8Matrix* const mtx) {
+  const int16x8_t out0 = Quantize_NEON(in, mtx, 0);
+  const int16x8_t out1 = Quantize_NEON(in, mtx, 8);
   uint8x8x4_t shuffles;
   // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
   // non-standard versions there.
@@ -889,11 +895,11 @@ static int QuantizeBlock(int16_t in[16], int16_t out[16],
   return 0;
 }
 
-static int Quantize2Blocks(int16_t in[32], int16_t out[32],
-                           const VP8Matrix* const mtx) {
+static int Quantize2Blocks_NEON(int16_t in[32], int16_t out[32],
+                                const VP8Matrix* const mtx) {
   int nz;
-  nz  = QuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
-  nz |= QuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
+  nz  = QuantizeBlock_NEON(in + 0 * 16, out + 0 * 16, mtx) << 0;
+  nz |= QuantizeBlock_NEON(in + 1 * 16, out + 1 * 16, mtx) << 1;
   return nz;
 }
 
@@ -905,14 +911,14 @@ static int Quantize2Blocks(int16_t in[32], int16_t out[32],
 extern void VP8EncDspInitNEON(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitNEON(void) {
-  VP8ITransform = ITransform;
-  VP8FTransform = FTransform;
+  VP8ITransform = ITransform_NEON;
+  VP8FTransform = FTransform_NEON;
 
-  VP8FTransformWHT = FTransformWHT;
+  VP8FTransformWHT = FTransformWHT_NEON;
 
-  VP8TDisto4x4 = Disto4x4;
-  VP8TDisto16x16 = Disto16x16;
-  VP8CollectHistogram = CollectHistogram;
+  VP8TDisto4x4 = Disto4x4_NEON;
+  VP8TDisto16x16 = Disto16x16_NEON;
+  VP8CollectHistogram = CollectHistogram_NEON;
 
   VP8SSE16x16 = SSE16x16_NEON;
   VP8SSE16x8 = SSE16x8_NEON;
@@ -920,8 +926,8 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitNEON(void) {
   VP8SSE4x4 = SSE4x4_NEON;
 
 #if !defined(WORK_AROUND_GCC)
-  VP8EncQuantizeBlock = QuantizeBlock;
-  VP8EncQuantize2Blocks = Quantize2Blocks;
+  VP8EncQuantizeBlock = QuantizeBlock_NEON;
+  VP8EncQuantize2Blocks = Quantize2Blocks_NEON;
 #endif
 }
 
diff --git a/src/3rdparty/libwebp/src/dsp/enc_sse2.c b/src/3rdparty/libwebp/src/dsp/enc_sse2.c
index 2026a74..7b3f142 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_sse2.c
@@ -11,23 +11,23 @@
 //
 // Author: Christian Duvivier (cduvivier@google.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 #include <assert.h>
 #include <stdlib.h>  // for abs()
 #include <emmintrin.h>
 
-#include "./common_sse2.h"
-#include "../enc/cost_enc.h"
-#include "../enc/vp8i_enc.h"
+#include "src/dsp/common_sse2.h"
+#include "src/enc/cost_enc.h"
+#include "src/enc/vp8i_enc.h"
 
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
 // Does one or two inverse transforms.
-static void ITransform(const uint8_t* ref, const int16_t* in, uint8_t* dst,
-                       int do_two) {
+static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
+                            int do_two) {
   // This implementation makes use of 16-bit fixed point versions of two
   // multiply constants:
   //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
@@ -193,10 +193,10 @@ static void ITransform(const uint8_t* ref, const int16_t* in, uint8_t* dst,
   }
 }
 
-static void FTransformPass1(const __m128i* const in01,
-                            const __m128i* const in23,
-                            __m128i* const out01,
-                            __m128i* const out32) {
+static void FTransformPass1_SSE2(const __m128i* const in01,
+                                 const __m128i* const in23,
+                                 __m128i* const out01,
+                                 __m128i* const out32) {
   const __m128i k937 = _mm_set1_epi32(937);
   const __m128i k1812 = _mm_set1_epi32(1812);
 
@@ -239,8 +239,9 @@ static void FTransformPass1(const __m128i* const in01,
   *out32 = _mm_shuffle_epi32(v23, _MM_SHUFFLE(1, 0, 3, 2));  // 3 2 3 2 3 2..
 }
 
-static void FTransformPass2(const __m128i* const v01, const __m128i* const v32,
-                            int16_t* out) {
+static void FTransformPass2_SSE2(const __m128i* const v01,
+                                 const __m128i* const v32,
+                                 int16_t* out) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i seven = _mm_set1_epi16(7);
   const __m128i k5352_2217 = _mm_set_epi16(5352,  2217, 5352,  2217,
@@ -291,7 +292,8 @@ static void FTransformPass2(const __m128i* const v01, const __m128i* const v32,
   _mm_storeu_si128((__m128i*)&out[8], d2_f3);
 }
 
-static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+static void FTransform_SSE2(const uint8_t* src, const uint8_t* ref,
+                            int16_t* out) {
   const __m128i zero = _mm_setzero_si128();
   // Load src.
   const __m128i src0 = _mm_loadl_epi64((const __m128i*)&src[0 * BPS]);
@@ -328,13 +330,14 @@ static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   __m128i v01, v32;
 
   // First pass
-  FTransformPass1(&row01, &row23, &v01, &v32);
+  FTransformPass1_SSE2(&row01, &row23, &v01, &v32);
 
   // Second pass
-  FTransformPass2(&v01, &v32, out);
+  FTransformPass2_SSE2(&v01, &v32, out);
 }
 
-static void FTransform2(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+static void FTransform2_SSE2(const uint8_t* src, const uint8_t* ref,
+                             int16_t* out) {
   const __m128i zero = _mm_setzero_si128();
 
   // Load src and convert to 16b.
@@ -374,15 +377,15 @@ static void FTransform2(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   __m128i v01h, v32h;
 
   // First pass
-  FTransformPass1(&shuf01l, &shuf23l, &v01l, &v32l);
-  FTransformPass1(&shuf01h, &shuf23h, &v01h, &v32h);
+  FTransformPass1_SSE2(&shuf01l, &shuf23l, &v01l, &v32l);
+  FTransformPass1_SSE2(&shuf01h, &shuf23h, &v01h, &v32h);
 
   // Second pass
-  FTransformPass2(&v01l, &v32l, out + 0);
-  FTransformPass2(&v01h, &v32h, out + 16);
+  FTransformPass2_SSE2(&v01l, &v32l, out + 0);
+  FTransformPass2_SSE2(&v01h, &v32h, out + 16);
 }
 
-static void FTransformWHTRow(const int16_t* const in, __m128i* const out) {
+static void FTransformWHTRow_SSE2(const int16_t* const in, __m128i* const out) {
   const __m128i kMult = _mm_set_epi16(-1, 1, -1, 1, 1, 1, 1, 1);
   const __m128i src0 = _mm_loadl_epi64((__m128i*)&in[0 * 16]);
   const __m128i src1 = _mm_loadl_epi64((__m128i*)&in[1 * 16]);
@@ -398,14 +401,14 @@ static void FTransformWHTRow(const int16_t* const in, __m128i* const out) {
   *out = _mm_madd_epi16(D, kMult);
 }
 
-static void FTransformWHT(const int16_t* in, int16_t* out) {
+static void FTransformWHT_SSE2(const int16_t* in, int16_t* out) {
   // Input is 12b signed.
   __m128i row0, row1, row2, row3;
   // Rows are 14b signed.
-  FTransformWHTRow(in + 0 * 64, &row0);
-  FTransformWHTRow(in + 1 * 64, &row1);
-  FTransformWHTRow(in + 2 * 64, &row2);
-  FTransformWHTRow(in + 3 * 64, &row3);
+  FTransformWHTRow_SSE2(in + 0 * 64, &row0);
+  FTransformWHTRow_SSE2(in + 1 * 64, &row1);
+  FTransformWHTRow_SSE2(in + 2 * 64, &row2);
+  FTransformWHTRow_SSE2(in + 3 * 64, &row3);
 
   {
     // The a* are 15b signed.
@@ -431,9 +434,9 @@ static void FTransformWHT(const int16_t* in, int16_t* out) {
 // Compute susceptibility based on DCT-coeff histograms:
 // the higher, the "easier" the macroblock is to compress.
 
-static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
-                             int start_block, int end_block,
-                             VP8Histogram* const histo) {
+static void CollectHistogram_SSE2(const uint8_t* ref, const uint8_t* pred,
+                                  int start_block, int end_block,
+                                  VP8Histogram* const histo) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i max_coeff_thresh = _mm_set1_epi16(MAX_COEFF_THRESH);
   int j;
@@ -442,7 +445,7 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
     int16_t out[16];
     int k;
 
-    FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
+    FTransform_SSE2(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
 
     // Convert coefficients to bin (within out[]).
     {
@@ -476,7 +479,7 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
 // Intra predictions
 
 // helper for chroma-DC predictions
-static WEBP_INLINE void Put8x8uv(uint8_t v, uint8_t* dst) {
+static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
   int j;
   const __m128i values = _mm_set1_epi8(v);
   for (j = 0; j < 8; ++j) {
@@ -484,7 +487,7 @@ static WEBP_INLINE void Put8x8uv(uint8_t v, uint8_t* dst) {
   }
 }
 
-static WEBP_INLINE void Put16(uint8_t v, uint8_t* dst) {
+static WEBP_INLINE void Put16_SSE2(uint8_t v, uint8_t* dst) {
   int j;
   const __m128i values = _mm_set1_epi8(v);
   for (j = 0; j < 16; ++j) {
@@ -492,20 +495,20 @@ static WEBP_INLINE void Put16(uint8_t v, uint8_t* dst) {
   }
 }
 
-static WEBP_INLINE void Fill(uint8_t* dst, int value, int size) {
+static WEBP_INLINE void Fill_SSE2(uint8_t* dst, int value, int size) {
   if (size == 4) {
     int j;
     for (j = 0; j < 4; ++j) {
       memset(dst + j * BPS, value, 4);
     }
   } else if (size == 8) {
-    Put8x8uv(value, dst);
+    Put8x8uv_SSE2(value, dst);
   } else {
-    Put16(value, dst);
+    Put16_SSE2(value, dst);
   }
 }
 
-static WEBP_INLINE void VE8uv(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void VE8uv_SSE2(uint8_t* dst, const uint8_t* top) {
   int j;
   const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
   for (j = 0; j < 8; ++j) {
@@ -513,7 +516,7 @@ static WEBP_INLINE void VE8uv(uint8_t* dst, const uint8_t* top) {
   }
 }
 
-static WEBP_INLINE void VE16(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void VE16_SSE2(uint8_t* dst, const uint8_t* top) {
   const __m128i top_values = _mm_load_si128((const __m128i*)top);
   int j;
   for (j = 0; j < 16; ++j) {
@@ -521,20 +524,20 @@ static WEBP_INLINE void VE16(uint8_t* dst, const uint8_t* top) {
   }
 }
 
-static WEBP_INLINE void VerticalPred(uint8_t* dst,
-                                     const uint8_t* top, int size) {
+static WEBP_INLINE void VerticalPred_SSE2(uint8_t* dst,
+                                          const uint8_t* top, int size) {
   if (top != NULL) {
     if (size == 8) {
-      VE8uv(dst, top);
+      VE8uv_SSE2(dst, top);
     } else {
-      VE16(dst, top);
+      VE16_SSE2(dst, top);
     }
   } else {
-    Fill(dst, 127, size);
+    Fill_SSE2(dst, 127, size);
   }
 }
 
-static WEBP_INLINE void HE8uv(uint8_t* dst, const uint8_t* left) {
+static WEBP_INLINE void HE8uv_SSE2(uint8_t* dst, const uint8_t* left) {
   int j;
   for (j = 0; j < 8; ++j) {
     const __m128i values = _mm_set1_epi8(left[j]);
@@ -543,7 +546,7 @@ static WEBP_INLINE void HE8uv(uint8_t* dst, const uint8_t* left) {
   }
 }
 
-static WEBP_INLINE void HE16(uint8_t* dst, const uint8_t* left) {
+static WEBP_INLINE void HE16_SSE2(uint8_t* dst, const uint8_t* left) {
   int j;
   for (j = 0; j < 16; ++j) {
     const __m128i values = _mm_set1_epi8(left[j]);
@@ -552,21 +555,21 @@ static WEBP_INLINE void HE16(uint8_t* dst, const uint8_t* left) {
   }
 }
 
-static WEBP_INLINE void HorizontalPred(uint8_t* dst,
-                                       const uint8_t* left, int size) {
+static WEBP_INLINE void HorizontalPred_SSE2(uint8_t* dst,
+                                            const uint8_t* left, int size) {
   if (left != NULL) {
     if (size == 8) {
-      HE8uv(dst, left);
+      HE8uv_SSE2(dst, left);
     } else {
-      HE16(dst, left);
+      HE16_SSE2(dst, left);
     }
   } else {
-    Fill(dst, 129, size);
+    Fill_SSE2(dst, 129, size);
   }
 }
 
-static WEBP_INLINE void TM(uint8_t* dst, const uint8_t* left,
-                           const uint8_t* top, int size) {
+static WEBP_INLINE void TM_SSE2(uint8_t* dst, const uint8_t* left,
+                                const uint8_t* top, int size) {
   const __m128i zero = _mm_setzero_si128();
   int y;
   if (size == 8) {
@@ -593,13 +596,13 @@ static WEBP_INLINE void TM(uint8_t* dst, const uint8_t* left,
   }
 }
 
-static WEBP_INLINE void TrueMotion(uint8_t* dst, const uint8_t* left,
-                                   const uint8_t* top, int size) {
+static WEBP_INLINE void TrueMotion_SSE2(uint8_t* dst, const uint8_t* left,
+                                        const uint8_t* top, int size) {
   if (left != NULL) {
     if (top != NULL) {
-      TM(dst, left, top, size);
+      TM_SSE2(dst, left, top, size);
     } else {
-      HorizontalPred(dst, left, size);
+      HorizontalPred_SSE2(dst, left, size);
     }
   } else {
     // true motion without left samples (hence: with default 129 value)
@@ -607,90 +610,90 @@ static WEBP_INLINE void TrueMotion(uint8_t* dst, const uint8_t* left,
     // Note that if top samples are not available, the default value is
     // then 129, and not 127 as in the VerticalPred case.
     if (top != NULL) {
-      VerticalPred(dst, top, size);
+      VerticalPred_SSE2(dst, top, size);
     } else {
-      Fill(dst, 129, size);
+      Fill_SSE2(dst, 129, size);
     }
   }
 }
 
-static WEBP_INLINE void DC8uv(uint8_t* dst, const uint8_t* left,
-                              const uint8_t* top) {
+static WEBP_INLINE void DC8uv_SSE2(uint8_t* dst, const uint8_t* left,
+                                   const uint8_t* top) {
   const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
   const __m128i left_values = _mm_loadl_epi64((const __m128i*)left);
   const __m128i combined = _mm_unpacklo_epi64(top_values, left_values);
   const int DC = VP8HorizontalAdd8b(&combined) + 8;
-  Put8x8uv(DC >> 4, dst);
+  Put8x8uv_SSE2(DC >> 4, dst);
 }
 
-static WEBP_INLINE void DC8uvNoLeft(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void DC8uvNoLeft_SSE2(uint8_t* dst, const uint8_t* top) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
   const __m128i sum = _mm_sad_epu8(top_values, zero);
   const int DC = _mm_cvtsi128_si32(sum) + 4;
-  Put8x8uv(DC >> 3, dst);
+  Put8x8uv_SSE2(DC >> 3, dst);
 }
 
-static WEBP_INLINE void DC8uvNoTop(uint8_t* dst, const uint8_t* left) {
+static WEBP_INLINE void DC8uvNoTop_SSE2(uint8_t* dst, const uint8_t* left) {
   // 'left' is contiguous so we can reuse the top summation.
-  DC8uvNoLeft(dst, left);
+  DC8uvNoLeft_SSE2(dst, left);
 }
 
-static WEBP_INLINE void DC8uvNoTopLeft(uint8_t* dst) {
-  Put8x8uv(0x80, dst);
+static WEBP_INLINE void DC8uvNoTopLeft_SSE2(uint8_t* dst) {
+  Put8x8uv_SSE2(0x80, dst);
 }
 
-static WEBP_INLINE void DC8uvMode(uint8_t* dst, const uint8_t* left,
-                                  const uint8_t* top) {
+static WEBP_INLINE void DC8uvMode_SSE2(uint8_t* dst, const uint8_t* left,
+                                       const uint8_t* top) {
   if (top != NULL) {
     if (left != NULL) {  // top and left present
-      DC8uv(dst, left, top);
+      DC8uv_SSE2(dst, left, top);
     } else {  // top, but no left
-      DC8uvNoLeft(dst, top);
+      DC8uvNoLeft_SSE2(dst, top);
     }
   } else if (left != NULL) {  // left but no top
-    DC8uvNoTop(dst, left);
+    DC8uvNoTop_SSE2(dst, left);
   } else {  // no top, no left, nothing.
-    DC8uvNoTopLeft(dst);
+    DC8uvNoTopLeft_SSE2(dst);
   }
 }
 
-static WEBP_INLINE void DC16(uint8_t* dst, const uint8_t* left,
-                             const uint8_t* top) {
+static WEBP_INLINE void DC16_SSE2(uint8_t* dst, const uint8_t* left,
+                                  const uint8_t* top) {
   const __m128i top_row = _mm_load_si128((const __m128i*)top);
   const __m128i left_row = _mm_load_si128((const __m128i*)left);
   const int DC =
       VP8HorizontalAdd8b(&top_row) + VP8HorizontalAdd8b(&left_row) + 16;
-  Put16(DC >> 5, dst);
+  Put16_SSE2(DC >> 5, dst);
 }
 
-static WEBP_INLINE void DC16NoLeft(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void DC16NoLeft_SSE2(uint8_t* dst, const uint8_t* top) {
   const __m128i top_row = _mm_load_si128((const __m128i*)top);
   const int DC = VP8HorizontalAdd8b(&top_row) + 8;
-  Put16(DC >> 4, dst);
+  Put16_SSE2(DC >> 4, dst);
 }
 
-static WEBP_INLINE void DC16NoTop(uint8_t* dst, const uint8_t* left) {
+static WEBP_INLINE void DC16NoTop_SSE2(uint8_t* dst, const uint8_t* left) {
   // 'left' is contiguous so we can reuse the top summation.
-  DC16NoLeft(dst, left);
+  DC16NoLeft_SSE2(dst, left);
 }
 
-static WEBP_INLINE void DC16NoTopLeft(uint8_t* dst) {
-  Put16(0x80, dst);
+static WEBP_INLINE void DC16NoTopLeft_SSE2(uint8_t* dst) {
+  Put16_SSE2(0x80, dst);
 }
 
-static WEBP_INLINE void DC16Mode(uint8_t* dst, const uint8_t* left,
-                                 const uint8_t* top) {
+static WEBP_INLINE void DC16Mode_SSE2(uint8_t* dst, const uint8_t* left,
+                                      const uint8_t* top) {
   if (top != NULL) {
     if (left != NULL) {  // top and left present
-      DC16(dst, left, top);
+      DC16_SSE2(dst, left, top);
     } else {  // top, but no left
-      DC16NoLeft(dst, top);
+      DC16NoLeft_SSE2(dst, top);
     }
   } else if (left != NULL) {  // left but no top
-    DC16NoTop(dst, left);
+    DC16NoTop_SSE2(dst, left);
   } else {  // no top, no left, nothing.
-    DC16NoTopLeft(dst);
+    DC16NoTopLeft_SSE2(dst);
   }
 }
 
@@ -709,7 +712,8 @@ static WEBP_INLINE void DC16Mode(uint8_t* dst, const uint8_t* left,
 //   where: AC = (a + b + 1) >> 1,   BC = (b + c + 1) >> 1
 //   and ab = a ^ b, bc = b ^ c, lsb = (AC^BC)&1
 
-static WEBP_INLINE void VE4(uint8_t* dst, const uint8_t* top) {  // vertical
+static WEBP_INLINE void VE4_SSE2(uint8_t* dst,
+                                 const uint8_t* top) {  // vertical
   const __m128i one = _mm_set1_epi8(1);
   const __m128i ABCDEFGH = _mm_loadl_epi64((__m128i*)(top - 1));
   const __m128i BCDEFGH0 = _mm_srli_si128(ABCDEFGH, 1);
@@ -725,7 +729,8 @@ static WEBP_INLINE void VE4(uint8_t* dst, const uint8_t* top) {  // vertical
   }
 }
 
-static WEBP_INLINE void HE4(uint8_t* dst, const uint8_t* top) {  // horizontal
+static WEBP_INLINE void HE4_SSE2(uint8_t* dst,
+                                 const uint8_t* top) {  // horizontal
   const int X = top[-1];
   const int I = top[-2];
   const int J = top[-3];
@@ -737,14 +742,15 @@ static WEBP_INLINE void HE4(uint8_t* dst, const uint8_t* top) {  // horizontal
   WebPUint32ToMem(dst + 3 * BPS, 0x01010101U * AVG3(K, L, L));
 }
 
-static WEBP_INLINE void DC4(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void DC4_SSE2(uint8_t* dst, const uint8_t* top) {
   uint32_t dc = 4;
   int i;
   for (i = 0; i < 4; ++i) dc += top[i] + top[-5 + i];
-  Fill(dst, dc >> 3, 4);
+  Fill_SSE2(dst, dc >> 3, 4);
 }
 
-static WEBP_INLINE void LD4(uint8_t* dst, const uint8_t* top) {  // Down-Left
+static WEBP_INLINE void LD4_SSE2(uint8_t* dst,
+                                 const uint8_t* top) {  // Down-Left
   const __m128i one = _mm_set1_epi8(1);
   const __m128i ABCDEFGH = _mm_loadl_epi64((const __m128i*)top);
   const __m128i BCDEFGH0 = _mm_srli_si128(ABCDEFGH, 1);
@@ -760,8 +766,8 @@ static WEBP_INLINE void LD4(uint8_t* dst, const uint8_t* top) {  // Down-Left
   WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
 }
 
-static WEBP_INLINE void VR4(uint8_t* dst,
-                            const uint8_t* top) {  // Vertical-Right
+static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
+                                 const uint8_t* top) {  // Vertical-Right
   const __m128i one = _mm_set1_epi8(1);
   const int I = top[-2];
   const int J = top[-3];
@@ -786,8 +792,8 @@ static WEBP_INLINE void VR4(uint8_t* dst,
   DST(0, 3) = AVG3(K, J, I);
 }
 
-static WEBP_INLINE void VL4(uint8_t* dst,
-                            const uint8_t* top) {  // Vertical-Left
+static WEBP_INLINE void VL4_SSE2(uint8_t* dst,
+                                 const uint8_t* top) {  // Vertical-Left
   const __m128i one = _mm_set1_epi8(1);
   const __m128i ABCDEFGH = _mm_loadl_epi64((const __m128i*)top);
   const __m128i BCDEFGH_ = _mm_srli_si128(ABCDEFGH, 1);
@@ -812,7 +818,8 @@ static WEBP_INLINE void VL4(uint8_t* dst,
   DST(3, 3) = (extra_out >> 8) & 0xff;
 }
 
-static WEBP_INLINE void RD4(uint8_t* dst, const uint8_t* top) {  // Down-right
+static WEBP_INLINE void RD4_SSE2(uint8_t* dst,
+                                 const uint8_t* top) {  // Down-right
   const __m128i one = _mm_set1_epi8(1);
   const __m128i LKJIXABC = _mm_loadl_epi64((const __m128i*)(top - 5));
   const __m128i LKJIXABCD = _mm_insert_epi16(LKJIXABC, top[3], 4);
@@ -828,7 +835,7 @@ static WEBP_INLINE void RD4(uint8_t* dst, const uint8_t* top) {  // Down-right
   WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
 }
 
-static WEBP_INLINE void HU4(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void HU4_SSE2(uint8_t* dst, const uint8_t* top) {
   const int I = top[-2];
   const int J = top[-3];
   const int K = top[-4];
@@ -843,7 +850,7 @@ static WEBP_INLINE void HU4(uint8_t* dst, const uint8_t* top) {
   DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
 }
 
-static WEBP_INLINE void HD4(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void HD4_SSE2(uint8_t* dst, const uint8_t* top) {
   const int X = top[-1];
   const int I = top[-2];
   const int J = top[-3];
@@ -866,7 +873,7 @@ static WEBP_INLINE void HD4(uint8_t* dst, const uint8_t* top) {
   DST(1, 3)             = AVG3(L, K, J);
 }
 
-static WEBP_INLINE void TM4(uint8_t* dst, const uint8_t* top) {
+static WEBP_INLINE void TM4_SSE2(uint8_t* dst, const uint8_t* top) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i top_values = _mm_cvtsi32_si128(WebPMemToUint32(top));
   const __m128i top_base = _mm_unpacklo_epi8(top_values, zero);
@@ -888,55 +895,56 @@ static WEBP_INLINE void TM4(uint8_t* dst, const uint8_t* top) {
 
 // Left samples are top[-5 .. -2], top_left is top[-1], top are
 // located at top[0..3], and top right is top[4..7]
-static void Intra4Preds(uint8_t* dst, const uint8_t* top) {
-  DC4(I4DC4 + dst, top);
-  TM4(I4TM4 + dst, top);
-  VE4(I4VE4 + dst, top);
-  HE4(I4HE4 + dst, top);
-  RD4(I4RD4 + dst, top);
-  VR4(I4VR4 + dst, top);
-  LD4(I4LD4 + dst, top);
-  VL4(I4VL4 + dst, top);
-  HD4(I4HD4 + dst, top);
-  HU4(I4HU4 + dst, top);
+static void Intra4Preds_SSE2(uint8_t* dst, const uint8_t* top) {
+  DC4_SSE2(I4DC4 + dst, top);
+  TM4_SSE2(I4TM4 + dst, top);
+  VE4_SSE2(I4VE4 + dst, top);
+  HE4_SSE2(I4HE4 + dst, top);
+  RD4_SSE2(I4RD4 + dst, top);
+  VR4_SSE2(I4VR4 + dst, top);
+  LD4_SSE2(I4LD4 + dst, top);
+  VL4_SSE2(I4VL4 + dst, top);
+  HD4_SSE2(I4HD4 + dst, top);
+  HU4_SSE2(I4HU4 + dst, top);
 }
 
 //------------------------------------------------------------------------------
 // Chroma 8x8 prediction (paragraph 12.2)
 
-static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
-                             const uint8_t* top) {
+static void IntraChromaPreds_SSE2(uint8_t* dst, const uint8_t* left,
+                                  const uint8_t* top) {
   // U block
-  DC8uvMode(C8DC8 + dst, left, top);
-  VerticalPred(C8VE8 + dst, top, 8);
-  HorizontalPred(C8HE8 + dst, left, 8);
-  TrueMotion(C8TM8 + dst, left, top, 8);
+  DC8uvMode_SSE2(C8DC8 + dst, left, top);
+  VerticalPred_SSE2(C8VE8 + dst, top, 8);
+  HorizontalPred_SSE2(C8HE8 + dst, left, 8);
+  TrueMotion_SSE2(C8TM8 + dst, left, top, 8);
   // V block
   dst += 8;
   if (top != NULL) top += 8;
   if (left != NULL) left += 16;
-  DC8uvMode(C8DC8 + dst, left, top);
-  VerticalPred(C8VE8 + dst, top, 8);
-  HorizontalPred(C8HE8 + dst, left, 8);
-  TrueMotion(C8TM8 + dst, left, top, 8);
+  DC8uvMode_SSE2(C8DC8 + dst, left, top);
+  VerticalPred_SSE2(C8VE8 + dst, top, 8);
+  HorizontalPred_SSE2(C8HE8 + dst, left, 8);
+  TrueMotion_SSE2(C8TM8 + dst, left, top, 8);
 }
 
 //------------------------------------------------------------------------------
 // luma 16x16 prediction (paragraph 12.3)
 
-static void Intra16Preds(uint8_t* dst,
-                         const uint8_t* left, const uint8_t* top) {
-  DC16Mode(I16DC16 + dst, left, top);
-  VerticalPred(I16VE16 + dst, top, 16);
-  HorizontalPred(I16HE16 + dst, left, 16);
-  TrueMotion(I16TM16 + dst, left, top, 16);
+static void Intra16Preds_SSE2(uint8_t* dst,
+                              const uint8_t* left, const uint8_t* top) {
+  DC16Mode_SSE2(I16DC16 + dst, left, top);
+  VerticalPred_SSE2(I16VE16 + dst, top, 16);
+  HorizontalPred_SSE2(I16HE16 + dst, left, 16);
+  TrueMotion_SSE2(I16TM16 + dst, left, top, 16);
 }
 
 //------------------------------------------------------------------------------
 // Metric
 
-static WEBP_INLINE void SubtractAndAccumulate(const __m128i a, const __m128i b,
-                                              __m128i* const sum) {
+static WEBP_INLINE void SubtractAndAccumulate_SSE2(const __m128i a,
+                                                   const __m128i b,
+                                                   __m128i* const sum) {
   // take abs(a-b) in 8b
   const __m128i a_b = _mm_subs_epu8(a, b);
   const __m128i b_a = _mm_subs_epu8(b, a);
@@ -951,8 +959,8 @@ static WEBP_INLINE void SubtractAndAccumulate(const __m128i a, const __m128i b,
   *sum = _mm_add_epi32(sum1, sum2);
 }
 
-static WEBP_INLINE int SSE_16xN(const uint8_t* a, const uint8_t* b,
-                                int num_pairs) {
+static WEBP_INLINE int SSE_16xN_SSE2(const uint8_t* a, const uint8_t* b,
+                                     int num_pairs) {
   __m128i sum = _mm_setzero_si128();
   int32_t tmp[4];
   int i;
@@ -963,8 +971,8 @@ static WEBP_INLINE int SSE_16xN(const uint8_t* a, const uint8_t* b,
     const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[BPS * 1]);
     const __m128i b1 = _mm_loadu_si128((const __m128i*)&b[BPS * 1]);
     __m128i sum1, sum2;
-    SubtractAndAccumulate(a0, b0, &sum1);
-    SubtractAndAccumulate(a1, b1, &sum2);
+    SubtractAndAccumulate_SSE2(a0, b0, &sum1);
+    SubtractAndAccumulate_SSE2(a1, b1, &sum2);
     sum = _mm_add_epi32(sum, _mm_add_epi32(sum1, sum2));
     a += 2 * BPS;
     b += 2 * BPS;
@@ -973,18 +981,18 @@ static WEBP_INLINE int SSE_16xN(const uint8_t* a, const uint8_t* b,
   return (tmp[3] + tmp[2] + tmp[1] + tmp[0]);
 }
 
-static int SSE16x16(const uint8_t* a, const uint8_t* b) {
-  return SSE_16xN(a, b, 8);
+static int SSE16x16_SSE2(const uint8_t* a, const uint8_t* b) {
+  return SSE_16xN_SSE2(a, b, 8);
 }
 
-static int SSE16x8(const uint8_t* a, const uint8_t* b) {
-  return SSE_16xN(a, b, 4);
+static int SSE16x8_SSE2(const uint8_t* a, const uint8_t* b) {
+  return SSE_16xN_SSE2(a, b, 4);
 }
 
 #define LOAD_8x16b(ptr) \
   _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i*)(ptr)), zero)
 
-static int SSE8x8(const uint8_t* a, const uint8_t* b) {
+static int SSE8x8_SSE2(const uint8_t* a, const uint8_t* b) {
   const __m128i zero = _mm_setzero_si128();
   int num_pairs = 4;
   __m128i sum = zero;
@@ -1011,7 +1019,7 @@ static int SSE8x8(const uint8_t* a, const uint8_t* b) {
 }
 #undef LOAD_8x16b
 
-static int SSE4x4(const uint8_t* a, const uint8_t* b) {
+static int SSE4x4_SSE2(const uint8_t* a, const uint8_t* b) {
   const __m128i zero = _mm_setzero_si128();
 
   // Load values. Note that we read 8 pixels instead of 4,
@@ -1048,7 +1056,7 @@ static int SSE4x4(const uint8_t* a, const uint8_t* b) {
 
 //------------------------------------------------------------------------------
 
-static void Mean16x4(const uint8_t* ref, uint32_t dc[4]) {
+static void Mean16x4_SSE2(const uint8_t* ref, uint32_t dc[4]) {
   const __m128i mask = _mm_set1_epi16(0x00ff);
   const __m128i a0 = _mm_loadu_si128((const __m128i*)&ref[BPS * 0]);
   const __m128i a1 = _mm_loadu_si128((const __m128i*)&ref[BPS * 1]);
@@ -1086,8 +1094,8 @@ static void Mean16x4(const uint8_t* ref, uint32_t dc[4]) {
 // Hadamard transform
 // Returns the weighted sum of the absolute value of transformed coefficients.
 // w[] contains a row-major 4 by 4 symmetric matrix.
-static int TTransform(const uint8_t* inA, const uint8_t* inB,
-                      const uint16_t* const w) {
+static int TTransform_SSE2(const uint8_t* inA, const uint8_t* inB,
+                           const uint16_t* const w) {
   int32_t sum[4];
   __m128i tmp_0, tmp_1, tmp_2, tmp_3;
   const __m128i zero = _mm_setzero_si128();
@@ -1187,19 +1195,19 @@ static int TTransform(const uint8_t* inA, const uint8_t* inB,
   return sum[0] + sum[1] + sum[2] + sum[3];
 }
 
-static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
-                    const uint16_t* const w) {
-  const int diff_sum = TTransform(a, b, w);
+static int Disto4x4_SSE2(const uint8_t* const a, const uint8_t* const b,
+                         const uint16_t* const w) {
+  const int diff_sum = TTransform_SSE2(a, b, w);
   return abs(diff_sum) >> 5;
 }
 
-static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
-                      const uint16_t* const w) {
+static int Disto16x16_SSE2(const uint8_t* const a, const uint8_t* const b,
+                           const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4(a + x + y, b + x + y, w);
+      D += Disto4x4_SSE2(a + x + y, b + x + y, w);
     }
   }
   return D;
@@ -1209,9 +1217,9 @@ static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
 // Quantization
 //
 
-static WEBP_INLINE int DoQuantizeBlock(int16_t in[16], int16_t out[16],
-                                       const uint16_t* const sharpen,
-                                       const VP8Matrix* const mtx) {
+static WEBP_INLINE int DoQuantizeBlock_SSE2(int16_t in[16], int16_t out[16],
+                                            const uint16_t* const sharpen,
+                                            const VP8Matrix* const mtx) {
   const __m128i max_coeff_2047 = _mm_set1_epi16(MAX_LEVEL);
   const __m128i zero = _mm_setzero_si128();
   __m128i coeff0, coeff8;
@@ -1321,22 +1329,22 @@ static WEBP_INLINE int DoQuantizeBlock(int16_t in[16], int16_t out[16],
   return (_mm_movemask_epi8(_mm_cmpeq_epi8(packed_out, zero)) != 0xffff);
 }
 
-static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         const VP8Matrix* const mtx) {
-  return DoQuantizeBlock(in, out, &mtx->sharpen_[0], mtx);
+static int QuantizeBlock_SSE2(int16_t in[16], int16_t out[16],
+                              const VP8Matrix* const mtx) {
+  return DoQuantizeBlock_SSE2(in, out, &mtx->sharpen_[0], mtx);
 }
 
-static int QuantizeBlockWHT(int16_t in[16], int16_t out[16],
-                            const VP8Matrix* const mtx) {
-  return DoQuantizeBlock(in, out, NULL, mtx);
+static int QuantizeBlockWHT_SSE2(int16_t in[16], int16_t out[16],
+                                 const VP8Matrix* const mtx) {
+  return DoQuantizeBlock_SSE2(in, out, NULL, mtx);
 }
 
-static int Quantize2Blocks(int16_t in[32], int16_t out[32],
-                           const VP8Matrix* const mtx) {
+static int Quantize2Blocks_SSE2(int16_t in[32], int16_t out[32],
+                                const VP8Matrix* const mtx) {
   int nz;
   const uint16_t* const sharpen = &mtx->sharpen_[0];
-  nz  = DoQuantizeBlock(in + 0 * 16, out + 0 * 16, sharpen, mtx) << 0;
-  nz |= DoQuantizeBlock(in + 1 * 16, out + 1 * 16, sharpen, mtx) << 1;
+  nz  = DoQuantizeBlock_SSE2(in + 0 * 16, out + 0 * 16, sharpen, mtx) << 0;
+  nz |= DoQuantizeBlock_SSE2(in + 1 * 16, out + 1 * 16, sharpen, mtx) << 1;
   return nz;
 }
 
@@ -1346,139 +1354,28 @@ static int Quantize2Blocks(int16_t in[32], int16_t out[32],
 extern void VP8EncDspInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitSSE2(void) {
-  VP8CollectHistogram = CollectHistogram;
-  VP8EncPredLuma16 = Intra16Preds;
-  VP8EncPredChroma8 = IntraChromaPreds;
-  VP8EncPredLuma4 = Intra4Preds;
-  VP8EncQuantizeBlock = QuantizeBlock;
-  VP8EncQuantize2Blocks = Quantize2Blocks;
-  VP8EncQuantizeBlockWHT = QuantizeBlockWHT;
-  VP8ITransform = ITransform;
-  VP8FTransform = FTransform;
-  VP8FTransform2 = FTransform2;
-  VP8FTransformWHT = FTransformWHT;
-  VP8SSE16x16 = SSE16x16;
-  VP8SSE16x8 = SSE16x8;
-  VP8SSE8x8 = SSE8x8;
-  VP8SSE4x4 = SSE4x4;
-  VP8TDisto4x4 = Disto4x4;
-  VP8TDisto16x16 = Disto16x16;
-  VP8Mean16x4 = Mean16x4;
-}
-
-//------------------------------------------------------------------------------
-// SSIM / PSNR entry point (TODO(skal): move to its own file later)
-
-static uint32_t AccumulateSSE_SSE2(const uint8_t* src1,
-                                   const uint8_t* src2, int len) {
-  int i = 0;
-  uint32_t sse2 = 0;
-  if (len >= 16) {
-    const int limit = len - 32;
-    int32_t tmp[4];
-    __m128i sum1;
-    __m128i sum = _mm_setzero_si128();
-    __m128i a0 = _mm_loadu_si128((const __m128i*)&src1[i]);
-    __m128i b0 = _mm_loadu_si128((const __m128i*)&src2[i]);
-    i += 16;
-    while (i <= limit) {
-      const __m128i a1 = _mm_loadu_si128((const __m128i*)&src1[i]);
-      const __m128i b1 = _mm_loadu_si128((const __m128i*)&src2[i]);
-      __m128i sum2;
-      i += 16;
-      SubtractAndAccumulate(a0, b0, &sum1);
-      sum = _mm_add_epi32(sum, sum1);
-      a0 = _mm_loadu_si128((const __m128i*)&src1[i]);
-      b0 = _mm_loadu_si128((const __m128i*)&src2[i]);
-      i += 16;
-      SubtractAndAccumulate(a1, b1, &sum2);
-      sum = _mm_add_epi32(sum, sum2);
-    }
-    SubtractAndAccumulate(a0, b0, &sum1);
-    sum = _mm_add_epi32(sum, sum1);
-    _mm_storeu_si128((__m128i*)tmp, sum);
-    sse2 += (tmp[3] + tmp[2] + tmp[1] + tmp[0]);
-  }
-
-  for (; i < len; ++i) {
-    const int32_t diff = src1[i] - src2[i];
-    sse2 += diff * diff;
-  }
-  return sse2;
-}
-
-static uint32_t HorizontalAdd16b(const __m128i* const m) {
-  uint16_t tmp[8];
-  const __m128i a = _mm_srli_si128(*m, 8);
-  const __m128i b = _mm_add_epi16(*m, a);
-  _mm_storeu_si128((__m128i*)tmp, b);
-  return (uint32_t)tmp[3] + tmp[2] + tmp[1] + tmp[0];
-}
-
-static uint32_t HorizontalAdd32b(const __m128i* const m) {
-  const __m128i a = _mm_srli_si128(*m, 8);
-  const __m128i b = _mm_add_epi32(*m, a);
-  const __m128i c = _mm_add_epi32(b, _mm_srli_si128(b, 4));
-  return (uint32_t)_mm_cvtsi128_si32(c);
-}
-
-static const uint16_t kWeight[] = { 1, 2, 3, 4, 3, 2, 1, 0 };
-
-#define ACCUMULATE_ROW(WEIGHT) do {                         \
-  /* compute row weight (Wx * Wy) */                        \
-  const __m128i Wy = _mm_set1_epi16((WEIGHT));              \
-  const __m128i W = _mm_mullo_epi16(Wx, Wy);                \
-  /* process 8 bytes at a time (7 bytes, actually) */       \
-  const __m128i a0 = _mm_loadl_epi64((const __m128i*)src1); \
-  const __m128i b0 = _mm_loadl_epi64((const __m128i*)src2); \
-  /* convert to 16b and multiply by weight */               \
-  const __m128i a1 = _mm_unpacklo_epi8(a0, zero);           \
-  const __m128i b1 = _mm_unpacklo_epi8(b0, zero);           \
-  const __m128i wa1 = _mm_mullo_epi16(a1, W);               \
-  const __m128i wb1 = _mm_mullo_epi16(b1, W);               \
-  /* accumulate */                                          \
-  xm  = _mm_add_epi16(xm, wa1);                             \
-  ym  = _mm_add_epi16(ym, wb1);                             \
-  xxm = _mm_add_epi32(xxm, _mm_madd_epi16(a1, wa1));        \
-  xym = _mm_add_epi32(xym, _mm_madd_epi16(a1, wb1));        \
-  yym = _mm_add_epi32(yym, _mm_madd_epi16(b1, wb1));        \
-  src1 += stride1;                                          \
-  src2 += stride2;                                          \
-} while (0)
-
-static double SSIMGet_SSE2(const uint8_t* src1, int stride1,
-                           const uint8_t* src2, int stride2) {
-  VP8DistoStats stats;
-  const __m128i zero = _mm_setzero_si128();
-  __m128i xm = zero, ym = zero;                // 16b accums
-  __m128i xxm = zero, yym = zero, xym = zero;  // 32b accum
-  const __m128i Wx = _mm_loadu_si128((const __m128i*)kWeight);
-  assert(2 * VP8_SSIM_KERNEL + 1 == 7);
-  ACCUMULATE_ROW(1);
-  ACCUMULATE_ROW(2);
-  ACCUMULATE_ROW(3);
-  ACCUMULATE_ROW(4);
-  ACCUMULATE_ROW(3);
-  ACCUMULATE_ROW(2);
-  ACCUMULATE_ROW(1);
-  stats.xm  = HorizontalAdd16b(&xm);
-  stats.ym  = HorizontalAdd16b(&ym);
-  stats.xxm = HorizontalAdd32b(&xxm);
-  stats.xym = HorizontalAdd32b(&xym);
-  stats.yym = HorizontalAdd32b(&yym);
-  return VP8SSIMFromStats(&stats);
-}
-
-extern void VP8SSIMDspInitSSE2(void);
-
-WEBP_TSAN_IGNORE_FUNCTION void VP8SSIMDspInitSSE2(void) {
-  VP8AccumulateSSE = AccumulateSSE_SSE2;
-  VP8SSIMGet = SSIMGet_SSE2;
+  VP8CollectHistogram = CollectHistogram_SSE2;
+  VP8EncPredLuma16 = Intra16Preds_SSE2;
+  VP8EncPredChroma8 = IntraChromaPreds_SSE2;
+  VP8EncPredLuma4 = Intra4Preds_SSE2;
+  VP8EncQuantizeBlock = QuantizeBlock_SSE2;
+  VP8EncQuantize2Blocks = Quantize2Blocks_SSE2;
+  VP8EncQuantizeBlockWHT = QuantizeBlockWHT_SSE2;
+  VP8ITransform = ITransform_SSE2;
+  VP8FTransform = FTransform_SSE2;
+  VP8FTransform2 = FTransform2_SSE2;
+  VP8FTransformWHT = FTransformWHT_SSE2;
+  VP8SSE16x16 = SSE16x16_SSE2;
+  VP8SSE16x8 = SSE16x8_SSE2;
+  VP8SSE8x8 = SSE8x8_SSE2;
+  VP8SSE4x4 = SSE4x4_SSE2;
+  VP8TDisto4x4 = Disto4x4_SSE2;
+  VP8TDisto16x16 = Disto16x16_SSE2;
+  VP8Mean16x4 = Mean16x4_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2
 
 WEBP_DSP_INIT_STUB(VP8EncDspInitSSE2)
-WEBP_DSP_INIT_STUB(VP8SSIMDspInitSSE2)
 
 #endif  // WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/enc_sse41.c b/src/3rdparty/libwebp/src/dsp/enc_sse41.c
index e32086d..924035a 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_sse41.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_sse41.c
@@ -11,21 +11,21 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE41)
 #include <smmintrin.h>
 #include <stdlib.h>  // for abs()
 
-#include "./common_sse2.h"
-#include "../enc/vp8i_enc.h"
+#include "src/dsp/common_sse2.h"
+#include "src/enc/vp8i_enc.h"
 
 //------------------------------------------------------------------------------
 // Compute susceptibility based on DCT-coeff histograms.
 
-static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
-                             int start_block, int end_block,
-                             VP8Histogram* const histo) {
+static void CollectHistogram_SSE41(const uint8_t* ref, const uint8_t* pred,
+                                   int start_block, int end_block,
+                                   VP8Histogram* const histo) {
   const __m128i max_coeff_thresh = _mm_set1_epi16(MAX_COEFF_THRESH);
   int j;
   int distribution[MAX_COEFF_THRESH + 1] = { 0 };
@@ -70,8 +70,8 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
 // Hadamard transform
 // Returns the weighted sum of the absolute value of transformed coefficients.
 // w[] contains a row-major 4 by 4 symmetric matrix.
-static int TTransform(const uint8_t* inA, const uint8_t* inB,
-                      const uint16_t* const w) {
+static int TTransform_SSE41(const uint8_t* inA, const uint8_t* inB,
+                            const uint16_t* const w) {
   int32_t sum[4];
   __m128i tmp_0, tmp_1, tmp_2, tmp_3;
 
@@ -168,19 +168,19 @@ static int TTransform(const uint8_t* inA, const uint8_t* inB,
   return sum[0] + sum[1] + sum[2] + sum[3];
 }
 
-static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
-                    const uint16_t* const w) {
-  const int diff_sum = TTransform(a, b, w);
+static int Disto4x4_SSE41(const uint8_t* const a, const uint8_t* const b,
+                          const uint16_t* const w) {
+  const int diff_sum = TTransform_SSE41(a, b, w);
   return abs(diff_sum) >> 5;
 }
 
-static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
-                      const uint16_t* const w) {
+static int Disto16x16_SSE41(const uint8_t* const a, const uint8_t* const b,
+                            const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4(a + x + y, b + x + y, w);
+      D += Disto4x4_SSE41(a + x + y, b + x + y, w);
     }
   }
   return D;
@@ -197,9 +197,9 @@ static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
                2 * (D) + 1, 2 * (D) + 0, 2 * (C) + 1, 2 * (C) + 0, \
                2 * (B) + 1, 2 * (B) + 0, 2 * (A) + 1, 2 * (A) + 0)
 
-static WEBP_INLINE int DoQuantizeBlock(int16_t in[16], int16_t out[16],
-                                       const uint16_t* const sharpen,
-                                       const VP8Matrix* const mtx) {
+static WEBP_INLINE int DoQuantizeBlock_SSE41(int16_t in[16], int16_t out[16],
+                                             const uint16_t* const sharpen,
+                                             const VP8Matrix* const mtx) {
   const __m128i max_coeff_2047 = _mm_set1_epi16(MAX_LEVEL);
   const __m128i zero = _mm_setzero_si128();
   __m128i out0, out8;
@@ -300,22 +300,22 @@ static WEBP_INLINE int DoQuantizeBlock(int16_t in[16], int16_t out[16],
 
 #undef PSHUFB_CST
 
-static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         const VP8Matrix* const mtx) {
-  return DoQuantizeBlock(in, out, &mtx->sharpen_[0], mtx);
+static int QuantizeBlock_SSE41(int16_t in[16], int16_t out[16],
+                               const VP8Matrix* const mtx) {
+  return DoQuantizeBlock_SSE41(in, out, &mtx->sharpen_[0], mtx);
 }
 
-static int QuantizeBlockWHT(int16_t in[16], int16_t out[16],
-                            const VP8Matrix* const mtx) {
-  return DoQuantizeBlock(in, out, NULL, mtx);
+static int QuantizeBlockWHT_SSE41(int16_t in[16], int16_t out[16],
+                                  const VP8Matrix* const mtx) {
+  return DoQuantizeBlock_SSE41(in, out, NULL, mtx);
 }
 
-static int Quantize2Blocks(int16_t in[32], int16_t out[32],
-                           const VP8Matrix* const mtx) {
+static int Quantize2Blocks_SSE41(int16_t in[32], int16_t out[32],
+                                 const VP8Matrix* const mtx) {
   int nz;
   const uint16_t* const sharpen = &mtx->sharpen_[0];
-  nz  = DoQuantizeBlock(in + 0 * 16, out + 0 * 16, sharpen, mtx) << 0;
-  nz |= DoQuantizeBlock(in + 1 * 16, out + 1 * 16, sharpen, mtx) << 1;
+  nz  = DoQuantizeBlock_SSE41(in + 0 * 16, out + 0 * 16, sharpen, mtx) << 0;
+  nz |= DoQuantizeBlock_SSE41(in + 1 * 16, out + 1 * 16, sharpen, mtx) << 1;
   return nz;
 }
 
@@ -324,12 +324,12 @@ static int Quantize2Blocks(int16_t in[32], int16_t out[32],
 
 extern void VP8EncDspInitSSE41(void);
 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitSSE41(void) {
-  VP8CollectHistogram = CollectHistogram;
-  VP8EncQuantizeBlock = QuantizeBlock;
-  VP8EncQuantize2Blocks = Quantize2Blocks;
-  VP8EncQuantizeBlockWHT = QuantizeBlockWHT;
-  VP8TDisto4x4 = Disto4x4;
-  VP8TDisto16x16 = Disto16x16;
+  VP8CollectHistogram = CollectHistogram_SSE41;
+  VP8EncQuantizeBlock = QuantizeBlock_SSE41;
+  VP8EncQuantize2Blocks = Quantize2Blocks_SSE41;
+  VP8EncQuantizeBlockWHT = QuantizeBlockWHT_SSE41;
+  VP8TDisto4x4 = Disto4x4_SSE41;
+  VP8TDisto16x16 = Disto16x16_SSE41;
 }
 
 #else  // !WEBP_USE_SSE41
diff --git a/src/3rdparty/libwebp/src/dsp/filters.c b/src/3rdparty/libwebp/src/dsp/filters.c
index 65f34aa..ca5f877 100644
--- a/src/3rdparty/libwebp/src/dsp/filters.c
+++ b/src/3rdparty/libwebp/src/dsp/filters.c
@@ -11,7 +11,7 @@
 //
 // Author: Urvang (urvang@google.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
@@ -20,16 +20,17 @@
 // Helpful macro.
 
 # define SANITY_CHECK(in, out)                                                 \
-  assert(in != NULL);                                                          \
-  assert(out != NULL);                                                         \
+  assert((in) != NULL);                                                        \
+  assert((out) != NULL);                                                       \
   assert(width > 0);                                                           \
   assert(height > 0);                                                          \
   assert(stride >= width);                                                     \
   assert(row >= 0 && num_rows > 0 && row + num_rows <= height);                \
   (void)height;  // Silence unused warning.
 
-static WEBP_INLINE void PredictLine(const uint8_t* src, const uint8_t* pred,
-                                    uint8_t* dst, int length, int inverse) {
+#if !WEBP_NEON_OMIT_C_CODE
+static WEBP_INLINE void PredictLine_C(const uint8_t* src, const uint8_t* pred,
+                                      uint8_t* dst, int length, int inverse) {
   int i;
   if (inverse) {
     for (i = 0; i < length; ++i) dst[i] = src[i] + pred[i];
@@ -41,10 +42,10 @@ static WEBP_INLINE void PredictLine(const uint8_t* src, const uint8_t* pred,
 //------------------------------------------------------------------------------
 // Horizontal filter.
 
-static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
-                                           int width, int height, int stride,
-                                           int row, int num_rows,
-                                           int inverse, uint8_t* out) {
+static WEBP_INLINE void DoHorizontalFilter_C(const uint8_t* in,
+                                             int width, int height, int stride,
+                                             int row, int num_rows,
+                                             int inverse, uint8_t* out) {
   const uint8_t* preds;
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
@@ -56,7 +57,7 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
   if (row == 0) {
     // Leftmost pixel is the same as input for topmost scanline.
     out[0] = in[0];
-    PredictLine(in + 1, preds, out + 1, width - 1, inverse);
+    PredictLine_C(in + 1, preds, out + 1, width - 1, inverse);
     row = 1;
     preds += stride;
     in += stride;
@@ -66,8 +67,8 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
   // Filter line-by-line.
   while (row < last_row) {
     // Leftmost pixel is predicted from above.
-    PredictLine(in, preds - stride, out, 1, inverse);
-    PredictLine(in + 1, preds, out + 1, width - 1, inverse);
+    PredictLine_C(in, preds - stride, out, 1, inverse);
+    PredictLine_C(in + 1, preds, out + 1, width - 1, inverse);
     ++row;
     preds += stride;
     in += stride;
@@ -78,10 +79,10 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
 //------------------------------------------------------------------------------
 // Vertical filter.
 
-static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
-                                         int width, int height, int stride,
-                                         int row, int num_rows,
-                                         int inverse, uint8_t* out) {
+static WEBP_INLINE void DoVerticalFilter_C(const uint8_t* in,
+                                           int width, int height, int stride,
+                                           int row, int num_rows,
+                                           int inverse, uint8_t* out) {
   const uint8_t* preds;
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
@@ -94,7 +95,7 @@ static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
     // Very first top-left pixel is copied.
     out[0] = in[0];
     // Rest of top scan-line is left-predicted.
-    PredictLine(in + 1, preds, out + 1, width - 1, inverse);
+    PredictLine_C(in + 1, preds, out + 1, width - 1, inverse);
     row = 1;
     in += stride;
     out += stride;
@@ -105,26 +106,28 @@ static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
 
   // Filter line-by-line.
   while (row < last_row) {
-    PredictLine(in, preds, out, width, inverse);
+    PredictLine_C(in, preds, out, width, inverse);
     ++row;
     preds += stride;
     in += stride;
     out += stride;
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 //------------------------------------------------------------------------------
 // Gradient filter.
 
-static WEBP_INLINE int GradientPredictor(uint8_t a, uint8_t b, uint8_t c) {
+static WEBP_INLINE int GradientPredictor_C(uint8_t a, uint8_t b, uint8_t c) {
   const int g = a + b - c;
   return ((g & ~0xff) == 0) ? g : (g < 0) ? 0 : 255;  // clip to 8bit
 }
 
-static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
-                                         int width, int height, int stride,
-                                         int row, int num_rows,
-                                         int inverse, uint8_t* out) {
+#if !WEBP_NEON_OMIT_C_CODE
+static WEBP_INLINE void DoGradientFilter_C(const uint8_t* in,
+                                           int width, int height, int stride,
+                                           int row, int num_rows,
+                                           int inverse, uint8_t* out) {
   const uint8_t* preds;
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
@@ -136,7 +139,7 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
   // left prediction for top scan-line
   if (row == 0) {
     out[0] = in[0];
-    PredictLine(in + 1, preds, out + 1, width - 1, inverse);
+    PredictLine_C(in + 1, preds, out + 1, width - 1, inverse);
     row = 1;
     preds += stride;
     in += stride;
@@ -147,11 +150,11 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
   while (row < last_row) {
     int w;
     // leftmost pixel: predict from above.
-    PredictLine(in, preds - stride, out, 1, inverse);
+    PredictLine_C(in, preds - stride, out, 1, inverse);
     for (w = 1; w < width; ++w) {
-      const int pred = GradientPredictor(preds[w - 1],
-                                         preds[w - stride],
-                                         preds[w - stride - 1]);
+      const int pred = GradientPredictor_C(preds[w - 1],
+                                           preds[w - stride],
+                                           preds[w - stride - 1]);
       out[w] = in[w] + (inverse ? pred : -pred);
     }
     ++row;
@@ -160,32 +163,34 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
     out += stride;
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 #undef SANITY_CHECK
 
 //------------------------------------------------------------------------------
 
-static void HorizontalFilter(const uint8_t* data, int width, int height,
-                             int stride, uint8_t* filtered_data) {
-  DoHorizontalFilter(data, width, height, stride, 0, height, 0, filtered_data);
+#if !WEBP_NEON_OMIT_C_CODE
+static void HorizontalFilter_C(const uint8_t* data, int width, int height,
+                               int stride, uint8_t* filtered_data) {
+  DoHorizontalFilter_C(data, width, height, stride, 0, height, 0,
+                       filtered_data);
 }
 
-static void VerticalFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
-  DoVerticalFilter(data, width, height, stride, 0, height, 0, filtered_data);
+static void VerticalFilter_C(const uint8_t* data, int width, int height,
+                             int stride, uint8_t* filtered_data) {
+  DoVerticalFilter_C(data, width, height, stride, 0, height, 0, filtered_data);
 }
 
-
-static void GradientFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
-  DoGradientFilter(data, width, height, stride, 0, height, 0, filtered_data);
+static void GradientFilter_C(const uint8_t* data, int width, int height,
+                             int stride, uint8_t* filtered_data) {
+  DoGradientFilter_C(data, width, height, stride, 0, height, 0, filtered_data);
 }
-
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 //------------------------------------------------------------------------------
 
-static void HorizontalUnfilter(const uint8_t* prev, const uint8_t* in,
-                               uint8_t* out, int width) {
+static void HorizontalUnfilter_C(const uint8_t* prev, const uint8_t* in,
+                                 uint8_t* out, int width) {
   uint8_t pred = (prev == NULL) ? 0 : prev[0];
   int i;
   for (i = 0; i < width; ++i) {
@@ -194,26 +199,28 @@ static void HorizontalUnfilter(const uint8_t* prev, const uint8_t* in,
   }
 }
 
-static void VerticalUnfilter(const uint8_t* prev, const uint8_t* in,
-                             uint8_t* out, int width) {
+#if !WEBP_NEON_OMIT_C_CODE
+static void VerticalUnfilter_C(const uint8_t* prev, const uint8_t* in,
+                               uint8_t* out, int width) {
   if (prev == NULL) {
-    HorizontalUnfilter(NULL, in, out, width);
+    HorizontalUnfilter_C(NULL, in, out, width);
   } else {
     int i;
     for (i = 0; i < width; ++i) out[i] = prev[i] + in[i];
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
-static void GradientUnfilter(const uint8_t* prev, const uint8_t* in,
-                             uint8_t* out, int width) {
+static void GradientUnfilter_C(const uint8_t* prev, const uint8_t* in,
+                               uint8_t* out, int width) {
   if (prev == NULL) {
-    HorizontalUnfilter(NULL, in, out, width);
+    HorizontalUnfilter_C(NULL, in, out, width);
   } else {
     uint8_t top = prev[0], top_left = top, left = top;
     int i;
     for (i = 0; i < width; ++i) {
       top = prev[i];  // need to read this first, in case prev==out
-      left = in[i] + GradientPredictor(left, top, top_left);
+      left = in[i] + GradientPredictor_C(left, top, top_left);
       top_left = top;
       out[i] = left;
     }
@@ -238,14 +245,18 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInit(void) {
   if (filters_last_cpuinfo_used == VP8GetCPUInfo) return;
 
   WebPUnfilters[WEBP_FILTER_NONE] = NULL;
-  WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter;
-  WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter;
-  WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter;
+#if !WEBP_NEON_OMIT_C_CODE
+  WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter_C;
+  WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter_C;
+#endif
+  WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter_C;
 
   WebPFilters[WEBP_FILTER_NONE] = NULL;
-  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter;
-  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter;
-  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter;
+#if !WEBP_NEON_OMIT_C_CODE
+  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter_C;
+  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter_C;
+  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter_C;
+#endif
 
   if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
@@ -253,11 +264,6 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInit(void) {
       VP8FiltersInitSSE2();
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      VP8FiltersInitNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS_DSP_R2)
     if (VP8GetCPUInfo(kMIPSdspR2)) {
       VP8FiltersInitMIPSdspR2();
@@ -269,5 +275,20 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInit(void) {
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    VP8FiltersInitNEON();
+  }
+#endif
+
+  assert(WebPUnfilters[WEBP_FILTER_HORIZONTAL] != NULL);
+  assert(WebPUnfilters[WEBP_FILTER_VERTICAL] != NULL);
+  assert(WebPUnfilters[WEBP_FILTER_GRADIENT] != NULL);
+  assert(WebPFilters[WEBP_FILTER_HORIZONTAL] != NULL);
+  assert(WebPFilters[WEBP_FILTER_VERTICAL] != NULL);
+  assert(WebPFilters[WEBP_FILTER_GRADIENT] != NULL);
+
   filters_last_cpuinfo_used = VP8GetCPUInfo;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/filters_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/filters_mips_dsp_r2.c
index 1d82e3c..9382b12 100644
--- a/src/3rdparty/libwebp/src/dsp/filters_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/filters_mips_dsp_r2.c
@@ -12,11 +12,11 @@
 // Author(s): Branimir Vasic (branimir.vasic@imgtec.com)
 //            Djordje Pesut (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-#include "../dsp/dsp.h"
+#include "src/dsp/dsp.h"
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
@@ -101,8 +101,8 @@
     );                                                                         \
   } while (0)
 
-static WEBP_INLINE void PredictLine(const uint8_t* src, uint8_t* dst,
-                                    int length) {
+static WEBP_INLINE void PredictLine_MIPSdspR2(const uint8_t* src, uint8_t* dst,
+                                              int length) {
   DO_PREDICT_LINE(src, dst, length, 0);
 }
 
@@ -192,10 +192,11 @@ static WEBP_INLINE void PredictLine(const uint8_t* src, uint8_t* dst,
     }                                                                          \
   } while (0)
 
-static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
-                                           int width, int height, int stride,
-                                           int row, int num_rows,
-                                           uint8_t* out) {
+static WEBP_INLINE void DoHorizontalFilter_MIPSdspR2(const uint8_t* in,
+                                                     int width, int height,
+                                                     int stride,
+                                                     int row, int num_rows,
+                                                     uint8_t* out) {
   const uint8_t* preds;
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
@@ -207,7 +208,7 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
   if (row == 0) {
     // Leftmost pixel is the same as input for topmost scanline.
     out[0] = in[0];
-    PredictLine(in + 1, out + 1, width - 1);
+    PredictLine_MIPSdspR2(in + 1, out + 1, width - 1);
     row = 1;
     preds += stride;
     in += stride;
@@ -219,9 +220,11 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
 }
 #undef FILTER_LINE_BY_LINE
 
-static void HorizontalFilter(const uint8_t* data, int width, int height,
-                             int stride, uint8_t* filtered_data) {
-  DoHorizontalFilter(data, width, height, stride, 0, height, filtered_data);
+static void HorizontalFilter_MIPSdspR2(const uint8_t* data,
+                                       int width, int height,
+                                       int stride, uint8_t* filtered_data) {
+  DoHorizontalFilter_MIPSdspR2(data, width, height, stride, 0, height,
+                               filtered_data);
 }
 
 //------------------------------------------------------------------------------
@@ -237,9 +240,11 @@ static void HorizontalFilter(const uint8_t* data, int width, int height,
     }                                                                          \
   } while (0)
 
-static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
-                                         int width, int height, int stride,
-                                         int row, int num_rows, uint8_t* out) {
+static WEBP_INLINE void DoVerticalFilter_MIPSdspR2(const uint8_t* in,
+                                                   int width, int height,
+                                                   int stride,
+                                                   int row, int num_rows,
+                                                   uint8_t* out) {
   const uint8_t* preds;
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
@@ -252,7 +257,7 @@ static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
     // Very first top-left pixel is copied.
     out[0] = in[0];
     // Rest of top scan-line is left-predicted.
-    PredictLine(in + 1, out + 1, width - 1);
+    PredictLine_MIPSdspR2(in + 1, out + 1, width - 1);
     row = 1;
     in += stride;
     out += stride;
@@ -266,15 +271,16 @@ static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
 }
 #undef FILTER_LINE_BY_LINE
 
-static void VerticalFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
-  DoVerticalFilter(data, width, height, stride, 0, height, filtered_data);
+static void VerticalFilter_MIPSdspR2(const uint8_t* data, int width, int height,
+                                     int stride, uint8_t* filtered_data) {
+  DoVerticalFilter_MIPSdspR2(data, width, height, stride, 0, height,
+                             filtered_data);
 }
 
 //------------------------------------------------------------------------------
 // Gradient filter.
 
-static WEBP_INLINE int GradientPredictor(uint8_t a, uint8_t b, uint8_t c) {
+static int GradientPredictor_MIPSdspR2(uint8_t a, uint8_t b, uint8_t c) {
   int temp0;
   __asm__ volatile (
     "addu             %[temp0],   %[a],       %[b]        \n\t"
@@ -293,9 +299,9 @@ static WEBP_INLINE int GradientPredictor(uint8_t a, uint8_t b, uint8_t c) {
       int w;                                                                   \
       PREDICT_LINE_ONE_PASS(in, PREDS - stride, out);                          \
       for (w = 1; w < width; ++w) {                                            \
-        const int pred = GradientPredictor(PREDS[w - 1],                       \
-                                           PREDS[w - stride],                  \
-                                           PREDS[w - stride - 1]);             \
+        const int pred = GradientPredictor_MIPSdspR2(PREDS[w - 1],             \
+                                                     PREDS[w - stride],        \
+                                                     PREDS[w - stride - 1]);   \
         out[w] = in[w] OPERATION pred;                                         \
       }                                                                        \
       ++row;                                                                   \
@@ -304,9 +310,9 @@ static WEBP_INLINE int GradientPredictor(uint8_t a, uint8_t b, uint8_t c) {
     }                                                                          \
   } while (0)
 
-static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
-                                         int width, int height, int stride,
-                                         int row, int num_rows, uint8_t* out) {
+static void DoGradientFilter_MIPSdspR2(const uint8_t* in,
+                                       int width, int height, int stride,
+                                       int row, int num_rows, uint8_t* out) {
   const uint8_t* preds;
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
@@ -318,7 +324,7 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
   // left prediction for top scan-line
   if (row == 0) {
     out[0] = in[0];
-    PredictLine(in + 1, out + 1, width - 1);
+    PredictLine_MIPSdspR2(in + 1, out + 1, width - 1);
     row = 1;
     preds += stride;
     in += stride;
@@ -330,38 +336,39 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
 }
 #undef FILTER_LINE_BY_LINE
 
-static void GradientFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
-  DoGradientFilter(data, width, height, stride, 0, height, filtered_data);
+static void GradientFilter_MIPSdspR2(const uint8_t* data, int width, int height,
+                                     int stride, uint8_t* filtered_data) {
+  DoGradientFilter_MIPSdspR2(data, width, height, stride, 0, height,
+                             filtered_data);
 }
 
 //------------------------------------------------------------------------------
 
-static void HorizontalUnfilter(const uint8_t* prev, const uint8_t* in,
-                               uint8_t* out, int width) {
+static void HorizontalUnfilter_MIPSdspR2(const uint8_t* prev, const uint8_t* in,
+                                         uint8_t* out, int width) {
  out[0] = in[0] + (prev == NULL ? 0 : prev[0]);
  DO_PREDICT_LINE(in + 1, out + 1, width - 1, 1);
 }
 
-static void VerticalUnfilter(const uint8_t* prev, const uint8_t* in,
-                             uint8_t* out, int width) {
+static void VerticalUnfilter_MIPSdspR2(const uint8_t* prev, const uint8_t* in,
+                                       uint8_t* out, int width) {
   if (prev == NULL) {
-    HorizontalUnfilter(NULL, in, out, width);
+    HorizontalUnfilter_MIPSdspR2(NULL, in, out, width);
   } else {
     DO_PREDICT_LINE_VERTICAL(in, prev, out, width, 1);
   }
 }
 
-static void GradientUnfilter(const uint8_t* prev, const uint8_t* in,
-                             uint8_t* out, int width) {
+static void GradientUnfilter_MIPSdspR2(const uint8_t* prev, const uint8_t* in,
+                                       uint8_t* out, int width) {
   if (prev == NULL) {
-    HorizontalUnfilter(NULL, in, out, width);
+    HorizontalUnfilter_MIPSdspR2(NULL, in, out, width);
   } else {
     uint8_t top = prev[0], top_left = top, left = top;
     int i;
     for (i = 0; i < width; ++i) {
       top = prev[i];  // need to read this first, in case prev==dst
-      left = in[i] + GradientPredictor(left, top, top_left);
+      left = in[i] + GradientPredictor_MIPSdspR2(left, top, top_left);
       top_left = top;
       out[i] = left;
     }
@@ -379,13 +386,13 @@ static void GradientUnfilter(const uint8_t* prev, const uint8_t* in,
 extern void VP8FiltersInitMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInitMIPSdspR2(void) {
-  WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter;
-  WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter;
-  WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter;
+  WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter_MIPSdspR2;
+  WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter_MIPSdspR2;
+  WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter_MIPSdspR2;
 
-  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter;
-  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter;
-  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter;
+  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter_MIPSdspR2;
+  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter_MIPSdspR2;
+  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/filters_msa.c b/src/3rdparty/libwebp/src/dsp/filters_msa.c
index 4b8922d..14c437d 100644
--- a/src/3rdparty/libwebp/src/dsp/filters_msa.c
+++ b/src/3rdparty/libwebp/src/dsp/filters_msa.c
@@ -11,11 +11,11 @@
 //
 // Author: Prashant Patil (prashant.patil@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MSA)
 
-#include "./msa_macro.h"
+#include "src/dsp/msa_macro.h"
 
 #include <assert.h>
 
@@ -66,8 +66,8 @@ static WEBP_INLINE void PredictLineInverse0(const uint8_t* src,
 //------------------------------------------------------------------------------
 // Horrizontal filter
 
-static void HorizontalFilter(const uint8_t* data, int width, int height,
-                             int stride, uint8_t* filtered_data) {
+static void HorizontalFilter_MSA(const uint8_t* data, int width, int height,
+                                 int stride, uint8_t* filtered_data) {
   const uint8_t* preds = data;
   const uint8_t* in = data;
   uint8_t* out = filtered_data;
@@ -129,8 +129,8 @@ static WEBP_INLINE void PredictLineGradient(const uint8_t* pinput,
 }
 
 
-static void GradientFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
+static void GradientFilter_MSA(const uint8_t* data, int width, int height,
+                               int stride, uint8_t* filtered_data) {
   const uint8_t* in = data;
   const uint8_t* preds = data;
   uint8_t* out = filtered_data;
@@ -157,8 +157,8 @@ static void GradientFilter(const uint8_t* data, int width, int height,
 //------------------------------------------------------------------------------
 // Vertical filter
 
-static void VerticalFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
+static void VerticalFilter_MSA(const uint8_t* data, int width, int height,
+                               int stride, uint8_t* filtered_data) {
   const uint8_t* in = data;
   const uint8_t* preds = data;
   uint8_t* out = filtered_data;
@@ -190,9 +190,9 @@ static void VerticalFilter(const uint8_t* data, int width, int height,
 extern void VP8FiltersInitMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInitMSA(void) {
-  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter;
-  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter;
-  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter;
+  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter_MSA;
+  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter_MSA;
+  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter_MSA;
 }
 
 #else  // !WEBP_USE_MSA
diff --git a/src/3rdparty/libwebp/src/dsp/filters_neon.c b/src/3rdparty/libwebp/src/dsp/filters_neon.c
index 4d6e50c..3e6a578 100644
--- a/src/3rdparty/libwebp/src/dsp/filters_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/filters_neon.c
@@ -11,12 +11,12 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_NEON)
 
 #include <assert.h>
-#include "./neon.h"
+#include "src/dsp/neon.h"
 
 //------------------------------------------------------------------------------
 // Helpful macros.
@@ -134,7 +134,7 @@ static WEBP_INLINE void DoVerticalFilter_NEON(const uint8_t* in,
 }
 
 static void VerticalFilter_NEON(const uint8_t* data, int width, int height,
-                               int stride, uint8_t* filtered_data) {
+                                int stride, uint8_t* filtered_data) {
   DoVerticalFilter_NEON(data, width, height, stride, 0, height,
                         filtered_data);
 }
@@ -196,7 +196,7 @@ static WEBP_INLINE void DoGradientFilter_NEON(const uint8_t* in,
 }
 
 static void GradientFilter_NEON(const uint8_t* data, int width, int height,
-                               int stride, uint8_t* filtered_data) {
+                                int stride, uint8_t* filtered_data) {
   DoGradientFilter_NEON(data, width, height, stride, 0, height,
                         filtered_data);
 }
@@ -251,9 +251,11 @@ static void VerticalUnfilter_NEON(const uint8_t* prev, const uint8_t* in,
 // GradientUnfilter_NEON is correct but slower than the C-version,
 // at least on ARM64. For armv7, it's a wash.
 // So best is to disable it for now, but keep the idea around...
-// #define USE_GRADIENT_UNFILTER
+#if !defined(USE_GRADIENT_UNFILTER)
+#define USE_GRADIENT_UNFILTER 0   // ALTERNATE_CODE
+#endif
 
-#if defined(USE_GRADIENT_UNFILTER)
+#if (USE_GRADIENT_UNFILTER == 1)
 #define GRAD_PROCESS_LANE(L)  do {                                             \
   const uint8x8_t tmp1 = ROTATE_RIGHT_N(pred, 1);  /* rotate predictor in */   \
   const int16x8_t tmp2 = vaddq_s16(BC, U8_TO_S16(tmp1));                       \
@@ -292,7 +294,7 @@ static void GradientPredictInverse_NEON(const uint8_t* const in,
 #undef GRAD_PROCESS_LANE
 
 static void GradientUnfilter_NEON(const uint8_t* prev, const uint8_t* in,
-                                 uint8_t* out, int width) {
+                                  uint8_t* out, int width) {
   if (prev == NULL) {
     HorizontalUnfilter_NEON(NULL, in, out, width);
   } else {
@@ -311,7 +313,7 @@ extern void VP8FiltersInitNEON(void);
 WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInitNEON(void) {
   WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter_NEON;
   WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter_NEON;
-#if defined(USE_GRADIENT_UNFILTER)
+#if (USE_GRADIENT_UNFILTER == 1)
   WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter_NEON;
 #endif
 
diff --git a/src/3rdparty/libwebp/src/dsp/filters_sse2.c b/src/3rdparty/libwebp/src/dsp/filters_sse2.c
index 67f7799..5a18895 100644
--- a/src/3rdparty/libwebp/src/dsp/filters_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/filters_sse2.c
@@ -11,7 +11,7 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 
@@ -24,16 +24,16 @@
 // Helpful macro.
 
 # define SANITY_CHECK(in, out)                                                 \
-  assert(in != NULL);                                                          \
-  assert(out != NULL);                                                         \
+  assert((in) != NULL);                                                        \
+  assert((out) != NULL);                                                       \
   assert(width > 0);                                                           \
   assert(height > 0);                                                          \
   assert(stride >= width);                                                     \
   assert(row >= 0 && num_rows > 0 && row + num_rows <= height);                \
   (void)height;  // Silence unused warning.
 
-static void PredictLineTop(const uint8_t* src, const uint8_t* pred,
-                           uint8_t* dst, int length) {
+static void PredictLineTop_SSE2(const uint8_t* src, const uint8_t* pred,
+                                uint8_t* dst, int length) {
   int i;
   const int max_pos = length & ~31;
   assert(length >= 0);
@@ -51,7 +51,7 @@ static void PredictLineTop(const uint8_t* src, const uint8_t* pred,
 }
 
 // Special case for left-based prediction (when preds==dst-1 or preds==src-1).
-static void PredictLineLeft(const uint8_t* src, uint8_t* dst, int length) {
+static void PredictLineLeft_SSE2(const uint8_t* src, uint8_t* dst, int length) {
   int i;
   const int max_pos = length & ~31;
   assert(length >= 0);
@@ -71,10 +71,11 @@ static void PredictLineLeft(const uint8_t* src, uint8_t* dst, int length) {
 //------------------------------------------------------------------------------
 // Horizontal filter.
 
-static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
-                                           int width, int height, int stride,
-                                           int row, int num_rows,
-                                           uint8_t* out) {
+static WEBP_INLINE void DoHorizontalFilter_SSE2(const uint8_t* in,
+                                                int width, int height,
+                                                int stride,
+                                                int row, int num_rows,
+                                                uint8_t* out) {
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
   SANITY_CHECK(in, out);
@@ -84,7 +85,7 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
   if (row == 0) {
     // Leftmost pixel is the same as input for topmost scanline.
     out[0] = in[0];
-    PredictLineLeft(in + 1, out + 1, width - 1);
+    PredictLineLeft_SSE2(in + 1, out + 1, width - 1);
     row = 1;
     in += stride;
     out += stride;
@@ -94,7 +95,7 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
   while (row < last_row) {
     // Leftmost pixel is predicted from above.
     out[0] = in[0] - in[-stride];
-    PredictLineLeft(in + 1, out + 1, width - 1);
+    PredictLineLeft_SSE2(in + 1, out + 1, width - 1);
     ++row;
     in += stride;
     out += stride;
@@ -104,9 +105,10 @@ static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in,
 //------------------------------------------------------------------------------
 // Vertical filter.
 
-static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
-                                         int width, int height, int stride,
-                                         int row, int num_rows, uint8_t* out) {
+static WEBP_INLINE void DoVerticalFilter_SSE2(const uint8_t* in,
+                                              int width, int height, int stride,
+                                              int row, int num_rows,
+                                              uint8_t* out) {
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
   SANITY_CHECK(in, out);
@@ -117,7 +119,7 @@ static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
     // Very first top-left pixel is copied.
     out[0] = in[0];
     // Rest of top scan-line is left-predicted.
-    PredictLineLeft(in + 1, out + 1, width - 1);
+    PredictLineLeft_SSE2(in + 1, out + 1, width - 1);
     row = 1;
     in += stride;
     out += stride;
@@ -125,7 +127,7 @@ static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
 
   // Filter line-by-line.
   while (row < last_row) {
-    PredictLineTop(in, in - stride, out, width);
+    PredictLineTop_SSE2(in, in - stride, out, width);
     ++row;
     in += stride;
     out += stride;
@@ -135,14 +137,14 @@ static WEBP_INLINE void DoVerticalFilter(const uint8_t* in,
 //------------------------------------------------------------------------------
 // Gradient filter.
 
-static WEBP_INLINE int GradientPredictorC(uint8_t a, uint8_t b, uint8_t c) {
+static WEBP_INLINE int GradientPredictor_SSE2(uint8_t a, uint8_t b, uint8_t c) {
   const int g = a + b - c;
   return ((g & ~0xff) == 0) ? g : (g < 0) ? 0 : 255;  // clip to 8bit
 }
 
-static void GradientPredictDirect(const uint8_t* const row,
-                                  const uint8_t* const top,
-                                  uint8_t* const out, int length) {
+static void GradientPredictDirect_SSE2(const uint8_t* const row,
+                                       const uint8_t* const top,
+                                       uint8_t* const out, int length) {
   const int max_pos = length & ~7;
   int i;
   const __m128i zero = _mm_setzero_si128();
@@ -161,14 +163,14 @@ static void GradientPredictDirect(const uint8_t* const row,
     _mm_storel_epi64((__m128i*)(out + i), H);
   }
   for (; i < length; ++i) {
-    out[i] = row[i] - GradientPredictorC(row[i - 1], top[i], top[i - 1]);
+    out[i] = row[i] - GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
   }
 }
 
-static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
-                                         int width, int height, int stride,
-                                         int row, int num_rows,
-                                         uint8_t* out) {
+static WEBP_INLINE void DoGradientFilter_SSE2(const uint8_t* in,
+                                              int width, int height, int stride,
+                                              int row, int num_rows,
+                                              uint8_t* out) {
   const size_t start_offset = row * stride;
   const int last_row = row + num_rows;
   SANITY_CHECK(in, out);
@@ -178,7 +180,7 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
   // left prediction for top scan-line
   if (row == 0) {
     out[0] = in[0];
-    PredictLineLeft(in + 1, out + 1, width - 1);
+    PredictLineLeft_SSE2(in + 1, out + 1, width - 1);
     row = 1;
     in += stride;
     out += stride;
@@ -187,7 +189,7 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
   // Filter line-by-line.
   while (row < last_row) {
     out[0] = in[0] - in[-stride];
-    GradientPredictDirect(in + 1, in + 1 - stride, out + 1, width - 1);
+    GradientPredictDirect_SSE2(in + 1, in + 1 - stride, out + 1, width - 1);
     ++row;
     in += stride;
     out += stride;
@@ -198,26 +200,27 @@ static WEBP_INLINE void DoGradientFilter(const uint8_t* in,
 
 //------------------------------------------------------------------------------
 
-static void HorizontalFilter(const uint8_t* data, int width, int height,
-                             int stride, uint8_t* filtered_data) {
-  DoHorizontalFilter(data, width, height, stride, 0, height, filtered_data);
+static void HorizontalFilter_SSE2(const uint8_t* data, int width, int height,
+                                  int stride, uint8_t* filtered_data) {
+  DoHorizontalFilter_SSE2(data, width, height, stride, 0, height,
+                          filtered_data);
 }
 
-static void VerticalFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
-  DoVerticalFilter(data, width, height, stride, 0, height, filtered_data);
+static void VerticalFilter_SSE2(const uint8_t* data, int width, int height,
+                                int stride, uint8_t* filtered_data) {
+  DoVerticalFilter_SSE2(data, width, height, stride, 0, height, filtered_data);
 }
 
-static void GradientFilter(const uint8_t* data, int width, int height,
-                           int stride, uint8_t* filtered_data) {
-  DoGradientFilter(data, width, height, stride, 0, height, filtered_data);
+static void GradientFilter_SSE2(const uint8_t* data, int width, int height,
+                                int stride, uint8_t* filtered_data) {
+  DoGradientFilter_SSE2(data, width, height, stride, 0, height, filtered_data);
 }
 
 //------------------------------------------------------------------------------
 // Inverse transforms
 
-static void HorizontalUnfilter(const uint8_t* prev, const uint8_t* in,
-                               uint8_t* out, int width) {
+static void HorizontalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
+                                    uint8_t* out, int width) {
   int i;
   __m128i last;
   out[0] = in[0] + (prev == NULL ? 0 : prev[0]);
@@ -238,10 +241,10 @@ static void HorizontalUnfilter(const uint8_t* prev, const uint8_t* in,
   for (; i < width; ++i) out[i] = in[i] + out[i - 1];
 }
 
-static void VerticalUnfilter(const uint8_t* prev, const uint8_t* in,
-                             uint8_t* out, int width) {
+static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
+                                  uint8_t* out, int width) {
   if (prev == NULL) {
-    HorizontalUnfilter(NULL, in, out, width);
+    HorizontalUnfilter_SSE2(NULL, in, out, width);
   } else {
     int i;
     const int max_pos = width & ~31;
@@ -260,9 +263,9 @@ static void VerticalUnfilter(const uint8_t* prev, const uint8_t* in,
   }
 }
 
-static void GradientPredictInverse(const uint8_t* const in,
-                                   const uint8_t* const top,
-                                   uint8_t* const row, int length) {
+static void GradientPredictInverse_SSE2(const uint8_t* const in,
+                                        const uint8_t* const top,
+                                        uint8_t* const row, int length) {
   if (length > 0) {
     int i;
     const int max_pos = length & ~7;
@@ -293,18 +296,18 @@ static void GradientPredictInverse(const uint8_t* const in,
       _mm_storel_epi64((__m128i*)&row[i], out);
     }
     for (; i < length; ++i) {
-      row[i] = in[i] + GradientPredictorC(row[i - 1], top[i], top[i - 1]);
+      row[i] = in[i] + GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
     }
   }
 }
 
-static void GradientUnfilter(const uint8_t* prev, const uint8_t* in,
-                             uint8_t* out, int width) {
+static void GradientUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
+                                  uint8_t* out, int width) {
   if (prev == NULL) {
-    HorizontalUnfilter(NULL, in, out, width);
+    HorizontalUnfilter_SSE2(NULL, in, out, width);
   } else {
     out[0] = in[0] + prev[0];  // predict from above
-    GradientPredictInverse(in + 1, prev + 1, out + 1, width - 1);
+    GradientPredictInverse_SSE2(in + 1, prev + 1, out + 1, width - 1);
   }
 }
 
@@ -314,13 +317,13 @@ static void GradientUnfilter(const uint8_t* prev, const uint8_t* in,
 extern void VP8FiltersInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInitSSE2(void) {
-  WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter;
-  WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter;
-  WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter;
+  WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter_SSE2;
+  WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter_SSE2;
+  WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter_SSE2;
 
-  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter;
-  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter;
-  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter;
+  WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter_SSE2;
+  WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter_SSE2;
+  WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/lossless.c b/src/3rdparty/libwebp/src/dsp/lossless.c
index 20d18f6..83f553d 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless.c
@@ -13,14 +13,15 @@
 //          Jyrki Alakuijala (jyrki@google.com)
 //          Urvang Joshi (urvang@google.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
+#include <assert.h>
 #include <math.h>
 #include <stdlib.h>
-#include "../dec/vp8li_dec.h"
-#include "../utils/endian_inl_utils.h"
-#include "./lossless.h"
-#include "./lossless_common.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/utils/endian_inl_utils.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
 
 #define MAX_DIFF_COST (1e30f)
 
@@ -80,8 +81,9 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
   return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b;
 }
 
-// gcc-4.9 on ARM generates incorrect code in Select() when Sub3() is inlined.
-#if defined(__arm__) && LOCAL_GCC_VERSION == 0x409
+// gcc <= 4.9 on ARM generates incorrect code in Select() when Sub3() is
+// inlined.
+#if defined(__arm__) && LOCAL_GCC_VERSION <= 0x409
 # define LOCAL_INLINE __attribute__ ((noinline))
 #else
 # define LOCAL_INLINE WEBP_INLINE
@@ -107,69 +109,69 @@ static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
 //------------------------------------------------------------------------------
 // Predictors
 
-static uint32_t Predictor0(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor0_C(uint32_t left, const uint32_t* const top) {
   (void)top;
   (void)left;
   return ARGB_BLACK;
 }
-static uint32_t Predictor1(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor1_C(uint32_t left, const uint32_t* const top) {
   (void)top;
   return left;
 }
-static uint32_t Predictor2(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor2_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[0];
 }
-static uint32_t Predictor3(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor3_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[1];
 }
-static uint32_t Predictor4(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor4_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[-1];
 }
-static uint32_t Predictor5(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor5_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average3(left, top[0], top[1]);
   return pred;
 }
-static uint32_t Predictor6(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor6_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(left, top[-1]);
   return pred;
 }
-static uint32_t Predictor7(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor7_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(left, top[0]);
   return pred;
 }
-static uint32_t Predictor8(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor8_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(top[-1], top[0]);
   (void)left;
   return pred;
 }
-static uint32_t Predictor9(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor9_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(top[0], top[1]);
   (void)left;
   return pred;
 }
-static uint32_t Predictor10(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor10_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average4(left, top[-1], top[0], top[1]);
   return pred;
 }
-static uint32_t Predictor11(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor11_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Select(top[0], left, top[-1]);
   return pred;
 }
-static uint32_t Predictor12(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor12_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]);
   return pred;
 }
-static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor13_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]);
   return pred;
 }
 
-GENERATE_PREDICTOR_ADD(Predictor0, PredictorAdd0)
-static void PredictorAdd1(const uint32_t* in, const uint32_t* upper,
-                          int num_pixels, uint32_t* out) {
+GENERATE_PREDICTOR_ADD(Predictor0_C, PredictorAdd0_C)
+static void PredictorAdd1_C(const uint32_t* in, const uint32_t* upper,
+                            int num_pixels, uint32_t* out) {
   int i;
   uint32_t left = out[-1];
   for (i = 0; i < num_pixels; ++i) {
@@ -177,29 +179,29 @@ static void PredictorAdd1(const uint32_t* in, const uint32_t* upper,
   }
   (void)upper;
 }
-GENERATE_PREDICTOR_ADD(Predictor2, PredictorAdd2)
-GENERATE_PREDICTOR_ADD(Predictor3, PredictorAdd3)
-GENERATE_PREDICTOR_ADD(Predictor4, PredictorAdd4)
-GENERATE_PREDICTOR_ADD(Predictor5, PredictorAdd5)
-GENERATE_PREDICTOR_ADD(Predictor6, PredictorAdd6)
-GENERATE_PREDICTOR_ADD(Predictor7, PredictorAdd7)
-GENERATE_PREDICTOR_ADD(Predictor8, PredictorAdd8)
-GENERATE_PREDICTOR_ADD(Predictor9, PredictorAdd9)
-GENERATE_PREDICTOR_ADD(Predictor10, PredictorAdd10)
-GENERATE_PREDICTOR_ADD(Predictor11, PredictorAdd11)
-GENERATE_PREDICTOR_ADD(Predictor12, PredictorAdd12)
-GENERATE_PREDICTOR_ADD(Predictor13, PredictorAdd13)
+GENERATE_PREDICTOR_ADD(Predictor2_C, PredictorAdd2_C)
+GENERATE_PREDICTOR_ADD(Predictor3_C, PredictorAdd3_C)
+GENERATE_PREDICTOR_ADD(Predictor4_C, PredictorAdd4_C)
+GENERATE_PREDICTOR_ADD(Predictor5_C, PredictorAdd5_C)
+GENERATE_PREDICTOR_ADD(Predictor6_C, PredictorAdd6_C)
+GENERATE_PREDICTOR_ADD(Predictor7_C, PredictorAdd7_C)
+GENERATE_PREDICTOR_ADD(Predictor8_C, PredictorAdd8_C)
+GENERATE_PREDICTOR_ADD(Predictor9_C, PredictorAdd9_C)
+GENERATE_PREDICTOR_ADD(Predictor10_C, PredictorAdd10_C)
+GENERATE_PREDICTOR_ADD(Predictor11_C, PredictorAdd11_C)
+GENERATE_PREDICTOR_ADD(Predictor12_C, PredictorAdd12_C)
+GENERATE_PREDICTOR_ADD(Predictor13_C, PredictorAdd13_C)
 
 //------------------------------------------------------------------------------
 
 // Inverse prediction.
-static void PredictorInverseTransform(const VP8LTransform* const transform,
-                                      int y_start, int y_end,
-                                      const uint32_t* in, uint32_t* out) {
+static void PredictorInverseTransform_C(const VP8LTransform* const transform,
+                                        int y_start, int y_end,
+                                        const uint32_t* in, uint32_t* out) {
   const int width = transform->xsize_;
   if (y_start == 0) {  // First Row follows the L (mode=1) mode.
-    PredictorAdd0(in, NULL, 1, out);
-    PredictorAdd1(in + 1, NULL, width - 1, out + 1);
+    PredictorAdd0_C(in, NULL, 1, out);
+    PredictorAdd1_C(in + 1, NULL, width - 1, out + 1);
     in += width;
     out += width;
     ++y_start;
@@ -217,7 +219,7 @@ static void PredictorInverseTransform(const VP8LTransform* const transform,
       const uint32_t* pred_mode_src = pred_mode_base;
       int x = 1;
       // First pixel follows the T (mode=2) mode.
-      PredictorAdd2(in, out - width, 1, out);
+      PredictorAdd2_C(in, out - width, 1, out);
       // .. the rest:
       while (x < width) {
         const VP8LPredictorAddSubFunc pred_func =
@@ -272,8 +274,8 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
     const uint32_t argb = src[i];
     const uint32_t green = argb >> 8;
     const uint32_t red = argb >> 16;
-    int new_red = red;
-    int new_blue = argb;
+    int new_red = red & 0xff;
+    int new_blue = argb & 0xff;
     new_red += ColorTransformDelta(m->green_to_red_, green);
     new_red &= 0xff;
     new_blue += ColorTransformDelta(m->green_to_blue_, green);
@@ -284,9 +286,9 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
 }
 
 // Color space inverse transform.
-static void ColorSpaceInverseTransform(const VP8LTransform* const transform,
-                                       int y_start, int y_end,
-                                       const uint32_t* src, uint32_t* dst) {
+static void ColorSpaceInverseTransform_C(const VP8LTransform* const transform,
+                                         int y_start, int y_end,
+                                         const uint32_t* src, uint32_t* dst) {
   const int width = transform->xsize_;
   const int tile_width = 1 << transform->bits_;
   const int mask = tile_width - 1;
@@ -362,10 +364,10 @@ STATIC_DECL void FUNC_NAME(const VP8LTransform* const transform,               \
   }                                                                            \
 }
 
-COLOR_INDEX_INVERSE(ColorIndexInverseTransform, MapARGB, static, uint32_t, 32b,
-                    VP8GetARGBIndex, VP8GetARGBValue)
-COLOR_INDEX_INVERSE(VP8LColorIndexInverseTransformAlpha, MapAlpha, , uint8_t,
-                    8b, VP8GetAlphaIndex, VP8GetAlphaValue)
+COLOR_INDEX_INVERSE(ColorIndexInverseTransform_C, MapARGB_C, static,
+                    uint32_t, 32b, VP8GetARGBIndex, VP8GetARGBValue)
+COLOR_INDEX_INVERSE(VP8LColorIndexInverseTransformAlpha, MapAlpha_C, ,
+                    uint8_t, 8b, VP8GetAlphaIndex, VP8GetAlphaValue)
 
 #undef COLOR_INDEX_INVERSE
 
@@ -380,7 +382,7 @@ void VP8LInverseTransform(const VP8LTransform* const transform,
       VP8LAddGreenToBlueAndRed(in, (row_end - row_start) * width, out);
       break;
     case PREDICTOR_TRANSFORM:
-      PredictorInverseTransform(transform, row_start, row_end, in, out);
+      PredictorInverseTransform_C(transform, row_start, row_end, in, out);
       if (row_end != transform->ysize_) {
         // The last predicted row in this iteration will be the top-pred row
         // for the first row in next iteration.
@@ -389,7 +391,7 @@ void VP8LInverseTransform(const VP8LTransform* const transform,
       }
       break;
     case CROSS_COLOR_TRANSFORM:
-      ColorSpaceInverseTransform(transform, row_start, row_end, in, out);
+      ColorSpaceInverseTransform_C(transform, row_start, row_end, in, out);
       break;
     case COLOR_INDEXING_TRANSFORM:
       if (in == out && transform->bits_ > 0) {
@@ -403,9 +405,9 @@ void VP8LInverseTransform(const VP8LTransform* const transform,
             VP8LSubSampleSize(transform->xsize_, transform->bits_);
         uint32_t* const src = out + out_stride - in_stride;
         memmove(src, out, in_stride * sizeof(*src));
-        ColorIndexInverseTransform(transform, row_start, row_end, src, out);
+        ColorIndexInverseTransform_C(transform, row_start, row_end, src, out);
       } else {
-        ColorIndexInverseTransform(transform, row_start, row_end, in, out);
+        ColorIndexInverseTransform_C(transform, row_start, row_end, in, out);
       }
       break;
   }
@@ -452,7 +454,7 @@ void VP8LConvertBGRAToRGBA4444_C(const uint32_t* src,
     const uint32_t argb = *src++;
     const uint8_t rg = ((argb >> 16) & 0xf0) | ((argb >> 12) & 0xf);
     const uint8_t ba = ((argb >>  0) & 0xf0) | ((argb >> 28) & 0xf);
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     *dst++ = ba;
     *dst++ = rg;
 #else
@@ -469,7 +471,7 @@ void VP8LConvertBGRAToRGB565_C(const uint32_t* src,
     const uint32_t argb = *src++;
     const uint8_t rg = ((argb >> 16) & 0xf8) | ((argb >> 13) & 0x7);
     const uint8_t gb = ((argb >>  5) & 0xe0) | ((argb >>  3) & 0x1f);
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     *dst++ = gb;
     *dst++ = rg;
 #else
@@ -496,22 +498,7 @@ static void CopyOrSwap(const uint32_t* src, int num_pixels, uint8_t* dst,
     const uint32_t* const src_end = src + num_pixels;
     while (src < src_end) {
       const uint32_t argb = *src++;
-
-#if !defined(WORDS_BIGENDIAN)
-#if !defined(WEBP_REFERENCE_IMPLEMENTATION)
       WebPUint32ToMem(dst, BSwap32(argb));
-#else  // WEBP_REFERENCE_IMPLEMENTATION
-      dst[0] = (argb >> 24) & 0xff;
-      dst[1] = (argb >> 16) & 0xff;
-      dst[2] = (argb >>  8) & 0xff;
-      dst[3] = (argb >>  0) & 0xff;
-#endif
-#else  // WORDS_BIGENDIAN
-      dst[0] = (argb >>  0) & 0xff;
-      dst[1] = (argb >>  8) & 0xff;
-      dst[2] = (argb >> 16) & 0xff;
-      dst[3] = (argb >> 24) & 0xff;
-#endif
       dst += sizeof(argb);
     }
   } else {
@@ -593,23 +580,23 @@ extern void VP8LDspInitMSA(void);
 static volatile VP8CPUInfo lossless_last_cpuinfo_used =
     (VP8CPUInfo)&lossless_last_cpuinfo_used;
 
-#define COPY_PREDICTOR_ARRAY(IN, OUT) do {              \
-  (OUT)[0] = IN##0;                                     \
-  (OUT)[1] = IN##1;                                     \
-  (OUT)[2] = IN##2;                                     \
-  (OUT)[3] = IN##3;                                     \
-  (OUT)[4] = IN##4;                                     \
-  (OUT)[5] = IN##5;                                     \
-  (OUT)[6] = IN##6;                                     \
-  (OUT)[7] = IN##7;                                     \
-  (OUT)[8] = IN##8;                                     \
-  (OUT)[9] = IN##9;                                     \
-  (OUT)[10] = IN##10;                                   \
-  (OUT)[11] = IN##11;                                   \
-  (OUT)[12] = IN##12;                                   \
-  (OUT)[13] = IN##13;                                   \
-  (OUT)[14] = IN##0; /* <- padding security sentinels*/ \
-  (OUT)[15] = IN##0;                                    \
+#define COPY_PREDICTOR_ARRAY(IN, OUT) do {                \
+  (OUT)[0] = IN##0_C;                                     \
+  (OUT)[1] = IN##1_C;                                     \
+  (OUT)[2] = IN##2_C;                                     \
+  (OUT)[3] = IN##3_C;                                     \
+  (OUT)[4] = IN##4_C;                                     \
+  (OUT)[5] = IN##5_C;                                     \
+  (OUT)[6] = IN##6_C;                                     \
+  (OUT)[7] = IN##7_C;                                     \
+  (OUT)[8] = IN##8_C;                                     \
+  (OUT)[9] = IN##9_C;                                     \
+  (OUT)[10] = IN##10_C;                                   \
+  (OUT)[11] = IN##11_C;                                   \
+  (OUT)[12] = IN##12_C;                                   \
+  (OUT)[13] = IN##13_C;                                   \
+  (OUT)[14] = IN##0_C; /* <- padding security sentinels*/ \
+  (OUT)[15] = IN##0_C;                                    \
 } while (0);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInit(void) {
@@ -620,18 +607,21 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInit(void) {
   COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd)
   COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd_C)
 
+#if !WEBP_NEON_OMIT_C_CODE
   VP8LAddGreenToBlueAndRed = VP8LAddGreenToBlueAndRed_C;
 
   VP8LTransformColorInverse = VP8LTransformColorInverse_C;
 
-  VP8LConvertBGRAToRGB = VP8LConvertBGRAToRGB_C;
   VP8LConvertBGRAToRGBA = VP8LConvertBGRAToRGBA_C;
+  VP8LConvertBGRAToRGB = VP8LConvertBGRAToRGB_C;
+  VP8LConvertBGRAToBGR = VP8LConvertBGRAToBGR_C;
+#endif
+
   VP8LConvertBGRAToRGBA4444 = VP8LConvertBGRAToRGBA4444_C;
   VP8LConvertBGRAToRGB565 = VP8LConvertBGRAToRGB565_C;
-  VP8LConvertBGRAToBGR = VP8LConvertBGRAToBGR_C;
 
-  VP8LMapColor32b = MapARGB;
-  VP8LMapColor8b = MapAlpha;
+  VP8LMapColor32b = MapARGB_C;
+  VP8LMapColor8b = MapAlpha_C;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
@@ -640,11 +630,6 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInit(void) {
       VP8LDspInitSSE2();
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      VP8LDspInitNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS_DSP_R2)
     if (VP8GetCPUInfo(kMIPSdspR2)) {
       VP8LDspInitMIPSdspR2();
@@ -656,6 +641,24 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInit(void) {
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    VP8LDspInitNEON();
+  }
+#endif
+
+  assert(VP8LAddGreenToBlueAndRed != NULL);
+  assert(VP8LTransformColorInverse != NULL);
+  assert(VP8LConvertBGRAToRGBA != NULL);
+  assert(VP8LConvertBGRAToRGB != NULL);
+  assert(VP8LConvertBGRAToBGR != NULL);
+  assert(VP8LConvertBGRAToRGBA4444 != NULL);
+  assert(VP8LConvertBGRAToRGB565 != NULL);
+  assert(VP8LMapColor32b != NULL);
+  assert(VP8LMapColor8b != NULL);
+
   lossless_last_cpuinfo_used = VP8GetCPUInfo;
 }
 #undef COPY_PREDICTOR_ARRAY
diff --git a/src/3rdparty/libwebp/src/dsp/lossless.h b/src/3rdparty/libwebp/src/dsp/lossless.h
index 352a54e..a99dbda 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless.h
+++ b/src/3rdparty/libwebp/src/dsp/lossless.h
@@ -15,18 +15,18 @@
 #ifndef WEBP_DSP_LOSSLESS_H_
 #define WEBP_DSP_LOSSLESS_H_
 
-#include "../webp/types.h"
-#include "../webp/decode.h"
+#include "src/webp/types.h"
+#include "src/webp/decode.h"
 
-#include "../enc/histogram_enc.h"
-#include "../utils/utils.h"
+#include "src/enc/histogram_enc.h"
+#include "src/utils/utils.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
 #ifdef WEBP_EXPERIMENTAL_FEATURES
-#include "../enc/delta_palettization_enc.h"
+#include "src/enc/delta_palettization_enc.h"
 #endif  // WEBP_EXPERIMENTAL_FEATURES
 
 //------------------------------------------------------------------------------
@@ -124,7 +124,7 @@ void VP8LDspInit(void);
 typedef void (*VP8LProcessEncBlueAndRedFunc)(uint32_t* dst, int num_pixels);
 extern VP8LProcessEncBlueAndRedFunc VP8LSubtractGreenFromBlueAndRed;
 typedef void (*VP8LTransformColorFunc)(const VP8LMultipliers* const m,
-                                       uint32_t* const dst, int num_pixels);
+                                       uint32_t* dst, int num_pixels);
 extern VP8LTransformColorFunc VP8LTransformColor;
 typedef void (*VP8LCollectColorBlueTransformsFunc)(
     const uint32_t* argb, int stride,
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_common.h b/src/3rdparty/libwebp/src/dsp/lossless_common.h
index c40f711..a2648d1 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_common.h
+++ b/src/3rdparty/libwebp/src/dsp/lossless_common.h
@@ -16,9 +16,9 @@
 #ifndef WEBP_DSP_LOSSLESS_COMMON_H_
 #define WEBP_DSP_LOSSLESS_COMMON_H_
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
-#include "../utils/utils.h"
+#include "src/utils/utils.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -93,14 +93,6 @@ static WEBP_INLINE float VP8LFastSLog2(uint32_t v) {
 // -----------------------------------------------------------------------------
 // PrefixEncode()
 
-static WEBP_INLINE int VP8LBitsLog2Ceiling(uint32_t n) {
-  const int log_floor = BitsLog2Floor(n);
-  if (n == (n & ~(n - 1))) {  // zero or a power of two.
-    return log_floor;
-  }
-  return log_floor + 1;
-}
-
 // Splitting of distance and length codes into prefixes and
 // extra bits. The prefixes are encoded with an entropy code
 // while the extra bits are stored just as normal bits.
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc.c b/src/3rdparty/libwebp/src/dsp/lossless_enc.c
index 4e46fba..92ca3c0 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc.c
@@ -13,15 +13,16 @@
 //          Jyrki Alakuijala (jyrki@google.com)
 //          Urvang Joshi (urvang@google.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
+#include <assert.h>
 #include <math.h>
 #include <stdlib.h>
-#include "../dec/vp8li_dec.h"
-#include "../utils/endian_inl_utils.h"
-#include "./lossless.h"
-#include "./lossless_common.h"
-#include "./yuv.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/utils/endian_inl_utils.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+#include "src/dsp/yuv.h"
 
 // lookup table for small values of log2(int)
 const float kLog2Table[LOG_LOOKUP_IDX_MAX] = {
@@ -325,7 +326,7 @@ const uint8_t kPrefixEncodeExtraBitsValue[PREFIX_LOOKUP_IDX_MAX] = {
   112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126
 };
 
-static float FastSLog2Slow(uint32_t v) {
+static float FastSLog2Slow_C(uint32_t v) {
   assert(v >= LOG_LOOKUP_IDX_MAX);
   if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
     int log_cnt = 0;
@@ -351,7 +352,7 @@ static float FastSLog2Slow(uint32_t v) {
   }
 }
 
-static float FastLog2Slow(uint32_t v) {
+static float FastLog2Slow_C(uint32_t v) {
   assert(v >= LOG_LOOKUP_IDX_MAX);
   if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
     int log_cnt = 0;
@@ -380,7 +381,7 @@ static float FastLog2Slow(uint32_t v) {
 // Methods to calculate Entropy (Shannon).
 
 // Compute the combined Shanon's entropy for distribution {X} and {X+Y}
-static float CombinedShannonEntropy(const int X[256], const int Y[256]) {
+static float CombinedShannonEntropy_C(const int X[256], const int Y[256]) {
   int i;
   double retval = 0.;
   int sumX = 0, sumXY = 0;
@@ -453,9 +454,9 @@ static WEBP_INLINE void GetEntropyUnrefinedHelper(
   *i_prev = i;
 }
 
-static void GetEntropyUnrefined(const uint32_t X[], int length,
-                                VP8LBitEntropy* const bit_entropy,
-                                VP8LStreaks* const stats) {
+static void GetEntropyUnrefined_C(const uint32_t X[], int length,
+                                  VP8LBitEntropy* const bit_entropy,
+                                  VP8LStreaks* const stats) {
   int i;
   int i_prev = 0;
   uint32_t x_prev = X[0];
@@ -474,10 +475,11 @@ static void GetEntropyUnrefined(const uint32_t X[], int length,
   bit_entropy->entropy += VP8LFastSLog2(bit_entropy->sum);
 }
 
-static void GetCombinedEntropyUnrefined(const uint32_t X[], const uint32_t Y[],
-                                        int length,
-                                        VP8LBitEntropy* const bit_entropy,
-                                        VP8LStreaks* const stats) {
+static void GetCombinedEntropyUnrefined_C(const uint32_t X[],
+                                          const uint32_t Y[],
+                                          int length,
+                                          VP8LBitEntropy* const bit_entropy,
+                                          VP8LStreaks* const stats) {
   int i = 1;
   int i_prev = 0;
   uint32_t xy_prev = X[0] + Y[0];
@@ -520,8 +522,8 @@ void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
     const uint32_t argb = data[i];
     const uint32_t green = argb >> 8;
     const uint32_t red = argb >> 16;
-    int new_red = red;
-    int new_blue = argb;
+    int new_red = red & 0xff;
+    int new_blue = argb & 0xff;
     new_red -= ColorTransformDelta(m->green_to_red_, green);
     new_red &= 0xff;
     new_blue -= ColorTransformDelta(m->green_to_blue_, green);
@@ -577,8 +579,8 @@ void VP8LCollectColorBlueTransforms_C(const uint32_t* argb, int stride,
 
 //------------------------------------------------------------------------------
 
-static int VectorMismatch(const uint32_t* const array1,
-                          const uint32_t* const array2, int length) {
+static int VectorMismatch_C(const uint32_t* const array1,
+                            const uint32_t* const array2, int length) {
   int match_len = 0;
 
   while (match_len < length && array1[match_len] == array2[match_len]) {
@@ -610,15 +612,15 @@ void VP8LBundleColorMap_C(const uint8_t* const row, int width, int xbits,
 
 //------------------------------------------------------------------------------
 
-static double ExtraCost(const uint32_t* population, int length) {
+static double ExtraCost_C(const uint32_t* population, int length) {
   int i;
   double cost = 0.;
   for (i = 2; i < length - 2; ++i) cost += (i >> 1) * population[i + 2];
   return cost;
 }
 
-static double ExtraCostCombined(const uint32_t* X, const uint32_t* Y,
-                                int length) {
+static double ExtraCostCombined_C(const uint32_t* X, const uint32_t* Y,
+                                  int length) {
   int i;
   double cost = 0.;
   for (i = 2; i < length - 2; ++i) {
@@ -630,9 +632,9 @@ static double ExtraCostCombined(const uint32_t* X, const uint32_t* Y,
 
 //------------------------------------------------------------------------------
 
-static void HistogramAdd(const VP8LHistogram* const a,
-                         const VP8LHistogram* const b,
-                         VP8LHistogram* const out) {
+static void HistogramAdd_C(const VP8LHistogram* const a,
+                           const VP8LHistogram* const b,
+                           VP8LHistogram* const out) {
   int i;
   const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
   assert(a->palette_code_bits_ == b->palette_code_bits_);
@@ -869,26 +871,28 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInit(void) {
 
   VP8LDspInit();
 
+#if !WEBP_NEON_OMIT_C_CODE
   VP8LSubtractGreenFromBlueAndRed = VP8LSubtractGreenFromBlueAndRed_C;
 
   VP8LTransformColor = VP8LTransformColor_C;
+#endif
 
   VP8LCollectColorBlueTransforms = VP8LCollectColorBlueTransforms_C;
   VP8LCollectColorRedTransforms = VP8LCollectColorRedTransforms_C;
 
-  VP8LFastLog2Slow = FastLog2Slow;
-  VP8LFastSLog2Slow = FastSLog2Slow;
+  VP8LFastLog2Slow = FastLog2Slow_C;
+  VP8LFastSLog2Slow = FastSLog2Slow_C;
 
-  VP8LExtraCost = ExtraCost;
-  VP8LExtraCostCombined = ExtraCostCombined;
-  VP8LCombinedShannonEntropy = CombinedShannonEntropy;
+  VP8LExtraCost = ExtraCost_C;
+  VP8LExtraCostCombined = ExtraCostCombined_C;
+  VP8LCombinedShannonEntropy = CombinedShannonEntropy_C;
 
-  VP8LGetEntropyUnrefined = GetEntropyUnrefined;
-  VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined;
+  VP8LGetEntropyUnrefined = GetEntropyUnrefined_C;
+  VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined_C;
 
-  VP8LHistogramAdd = HistogramAdd;
+  VP8LHistogramAdd = HistogramAdd_C;
 
-  VP8LVectorMismatch = VectorMismatch;
+  VP8LVectorMismatch = VectorMismatch_C;
   VP8LBundleColorMap = VP8LBundleColorMap_C;
 
   VP8LPredictorsSub[0] = PredictorSub0_C;
@@ -937,11 +941,6 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInit(void) {
 #endif
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      VP8LEncDspInitNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
       VP8LEncDspInitMIPS32();
@@ -958,6 +957,61 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInit(void) {
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    VP8LEncDspInitNEON();
+  }
+#endif
+
+  assert(VP8LSubtractGreenFromBlueAndRed != NULL);
+  assert(VP8LTransformColor != NULL);
+  assert(VP8LCollectColorBlueTransforms != NULL);
+  assert(VP8LCollectColorRedTransforms != NULL);
+  assert(VP8LFastLog2Slow != NULL);
+  assert(VP8LFastSLog2Slow != NULL);
+  assert(VP8LExtraCost != NULL);
+  assert(VP8LExtraCostCombined != NULL);
+  assert(VP8LCombinedShannonEntropy != NULL);
+  assert(VP8LGetEntropyUnrefined != NULL);
+  assert(VP8LGetCombinedEntropyUnrefined != NULL);
+  assert(VP8LHistogramAdd != NULL);
+  assert(VP8LVectorMismatch != NULL);
+  assert(VP8LBundleColorMap != NULL);
+  assert(VP8LPredictorsSub[0] != NULL);
+  assert(VP8LPredictorsSub[1] != NULL);
+  assert(VP8LPredictorsSub[2] != NULL);
+  assert(VP8LPredictorsSub[3] != NULL);
+  assert(VP8LPredictorsSub[4] != NULL);
+  assert(VP8LPredictorsSub[5] != NULL);
+  assert(VP8LPredictorsSub[6] != NULL);
+  assert(VP8LPredictorsSub[7] != NULL);
+  assert(VP8LPredictorsSub[8] != NULL);
+  assert(VP8LPredictorsSub[9] != NULL);
+  assert(VP8LPredictorsSub[10] != NULL);
+  assert(VP8LPredictorsSub[11] != NULL);
+  assert(VP8LPredictorsSub[12] != NULL);
+  assert(VP8LPredictorsSub[13] != NULL);
+  assert(VP8LPredictorsSub[14] != NULL);
+  assert(VP8LPredictorsSub[15] != NULL);
+  assert(VP8LPredictorsSub_C[0] != NULL);
+  assert(VP8LPredictorsSub_C[1] != NULL);
+  assert(VP8LPredictorsSub_C[2] != NULL);
+  assert(VP8LPredictorsSub_C[3] != NULL);
+  assert(VP8LPredictorsSub_C[4] != NULL);
+  assert(VP8LPredictorsSub_C[5] != NULL);
+  assert(VP8LPredictorsSub_C[6] != NULL);
+  assert(VP8LPredictorsSub_C[7] != NULL);
+  assert(VP8LPredictorsSub_C[8] != NULL);
+  assert(VP8LPredictorsSub_C[9] != NULL);
+  assert(VP8LPredictorsSub_C[10] != NULL);
+  assert(VP8LPredictorsSub_C[11] != NULL);
+  assert(VP8LPredictorsSub_C[12] != NULL);
+  assert(VP8LPredictorsSub_C[13] != NULL);
+  assert(VP8LPredictorsSub_C[14] != NULL);
+  assert(VP8LPredictorsSub_C[15] != NULL);
+
   lossless_enc_last_cpuinfo_used = VP8GetCPUInfo;
 }
 
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc_mips32.c b/src/3rdparty/libwebp/src/dsp/lossless_enc_mips32.c
index 4186b9f..e7b58f4 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc_mips32.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc_mips32.c
@@ -12,9 +12,9 @@
 // Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
 //             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
 
-#include "./dsp.h"
-#include "./lossless.h"
-#include "./lossless_common.h"
+#include "src/dsp/dsp.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
 
 #if defined(WEBP_USE_MIPS32)
 
@@ -23,7 +23,7 @@
 #include <stdlib.h>
 #include <string.h>
 
-static float FastSLog2Slow(uint32_t v) {
+static float FastSLog2Slow_MIPS32(uint32_t v) {
   assert(v >= LOG_LOOKUP_IDX_MAX);
   if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
     uint32_t log_cnt, y, correction;
@@ -59,7 +59,7 @@ static float FastSLog2Slow(uint32_t v) {
   }
 }
 
-static float FastLog2Slow(uint32_t v) {
+static float FastLog2Slow_MIPS32(uint32_t v) {
   assert(v >= LOG_LOOKUP_IDX_MAX);
   if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
     uint32_t log_cnt, y;
@@ -104,7 +104,7 @@ static float FastLog2Slow(uint32_t v) {
 //     pop += 2;
 //   }
 //   return (double)cost;
-static double ExtraCost(const uint32_t* const population, int length) {
+static double ExtraCost_MIPS32(const uint32_t* const population, int length) {
   int i, temp0, temp1;
   const uint32_t* pop = &population[4];
   const uint32_t* const LoopEnd = &population[length];
@@ -149,8 +149,8 @@ static double ExtraCost(const uint32_t* const population, int length) {
 //     pY += 2;
 //   }
 //   return (double)cost;
-static double ExtraCostCombined(const uint32_t* const X,
-                                const uint32_t* const Y, int length) {
+static double ExtraCostCombined_MIPS32(const uint32_t* const X,
+                                       const uint32_t* const Y, int length) {
   int i, temp0, temp1, temp2, temp3;
   const uint32_t* pX = &X[4];
   const uint32_t* pY = &Y[4];
@@ -241,9 +241,9 @@ static WEBP_INLINE void GetEntropyUnrefinedHelper(
   *i_prev = i;
 }
 
-static void GetEntropyUnrefined(const uint32_t X[], int length,
-                                VP8LBitEntropy* const bit_entropy,
-                                VP8LStreaks* const stats) {
+static void GetEntropyUnrefined_MIPS32(const uint32_t X[], int length,
+                                       VP8LBitEntropy* const bit_entropy,
+                                       VP8LStreaks* const stats) {
   int i;
   int i_prev = 0;
   uint32_t x_prev = X[0];
@@ -262,26 +262,27 @@ static void GetEntropyUnrefined(const uint32_t X[], int length,
   bit_entropy->entropy += VP8LFastSLog2(bit_entropy->sum);
 }
 
-static void GetCombinedEntropyUnrefined(const uint32_t X[], const uint32_t Y[],
-                                        int length,
-                                        VP8LBitEntropy* const bit_entropy,
-                                        VP8LStreaks* const stats) {
+static void GetCombinedEntropyUnrefined_MIPS32(const uint32_t X[],
+                                               const uint32_t Y[],
+                                               int length,
+                                               VP8LBitEntropy* const entropy,
+                                               VP8LStreaks* const stats) {
   int i = 1;
   int i_prev = 0;
   uint32_t xy_prev = X[0] + Y[0];
 
   memset(stats, 0, sizeof(*stats));
-  VP8LBitEntropyInit(bit_entropy);
+  VP8LBitEntropyInit(entropy);
 
   for (i = 1; i < length; ++i) {
     const uint32_t xy = X[i] + Y[i];
     if (xy != xy_prev) {
-      GetEntropyUnrefinedHelper(xy, i, &xy_prev, &i_prev, bit_entropy, stats);
+      GetEntropyUnrefinedHelper(xy, i, &xy_prev, &i_prev, entropy, stats);
     }
   }
-  GetEntropyUnrefinedHelper(0, i, &xy_prev, &i_prev, bit_entropy, stats);
+  GetEntropyUnrefinedHelper(0, i, &xy_prev, &i_prev, entropy, stats);
 
-  bit_entropy->entropy += VP8LFastSLog2(bit_entropy->sum);
+  entropy->entropy += VP8LFastSLog2(entropy->sum);
 }
 
 #define ASM_START                                       \
@@ -374,9 +375,9 @@ static void GetCombinedEntropyUnrefined(const uint32_t X[], const uint32_t Y[],
   }                                                     \
 } while (0)
 
-static void HistogramAdd(const VP8LHistogram* const a,
-                         const VP8LHistogram* const b,
-                         VP8LHistogram* const out) {
+static void HistogramAdd_MIPS32(const VP8LHistogram* const a,
+                                const VP8LHistogram* const b,
+                                VP8LHistogram* const out) {
   uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
   const int extra_cache_size = VP8LHistogramNumCodes(a->palette_code_bits_)
                              - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
@@ -415,13 +416,13 @@ static void HistogramAdd(const VP8LHistogram* const a,
 extern void VP8LEncDspInitMIPS32(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitMIPS32(void) {
-  VP8LFastSLog2Slow = FastSLog2Slow;
-  VP8LFastLog2Slow = FastLog2Slow;
-  VP8LExtraCost = ExtraCost;
-  VP8LExtraCostCombined = ExtraCostCombined;
-  VP8LGetEntropyUnrefined = GetEntropyUnrefined;
-  VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined;
-  VP8LHistogramAdd = HistogramAdd;
+  VP8LFastSLog2Slow = FastSLog2Slow_MIPS32;
+  VP8LFastLog2Slow = FastLog2Slow_MIPS32;
+  VP8LExtraCost = ExtraCost_MIPS32;
+  VP8LExtraCostCombined = ExtraCostCombined_MIPS32;
+  VP8LGetEntropyUnrefined = GetEntropyUnrefined_MIPS32;
+  VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined_MIPS32;
+  VP8LHistogramAdd = HistogramAdd_MIPS32;
 }
 
 #else  // !WEBP_USE_MIPS32
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/lossless_enc_mips_dsp_r2.c
index 0abf3c4..5855e6a 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc_mips_dsp_r2.c
@@ -12,14 +12,14 @@
 // Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
 //             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-#include "./lossless.h"
+#include "src/dsp/lossless.h"
 
-static void SubtractGreenFromBlueAndRed(uint32_t* argb_data,
-                                        int num_pixels) {
+static void SubtractGreenFromBlueAndRed_MIPSdspR2(uint32_t* argb_data,
+                                                  int num_pixels) {
   uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
   uint32_t* const p_loop1_end = argb_data + (num_pixels & ~3);
   uint32_t* const p_loop2_end = p_loop1_end + (num_pixels & 3);
@@ -78,8 +78,8 @@ static WEBP_INLINE uint32_t ColorTransformDelta(int8_t color_pred,
   return (uint32_t)((int)(color_pred) * color) >> 5;
 }
 
-static void TransformColor(const VP8LMultipliers* const m, uint32_t* data,
-                           int num_pixels) {
+static void TransformColor_MIPSdspR2(const VP8LMultipliers* const m,
+                                     uint32_t* data, int num_pixels) {
   int temp0, temp1, temp2, temp3, temp4, temp5;
   uint32_t argb, argb1, new_red, new_red1;
   const uint32_t G_to_R = m->green_to_red_;
@@ -171,10 +171,13 @@ static WEBP_INLINE uint8_t TransformColorBlue(uint8_t green_to_blue,
   return (new_blue & 0xff);
 }
 
-static void CollectColorBlueTransforms(const uint32_t* argb, int stride,
-                                       int tile_width, int tile_height,
-                                       int green_to_blue, int red_to_blue,
-                                       int histo[]) {
+static void CollectColorBlueTransforms_MIPSdspR2(const uint32_t* argb,
+                                                 int stride,
+                                                 int tile_width,
+                                                 int tile_height,
+                                                 int green_to_blue,
+                                                 int red_to_blue,
+                                                 int histo[]) {
   const int rtb = (red_to_blue << 16) | (red_to_blue & 0xffff);
   const int gtb = (green_to_blue << 16) | (green_to_blue & 0xffff);
   const uint32_t mask = 0xff00ffu;
@@ -222,9 +225,12 @@ static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
   return (new_red & 0xff);
 }
 
-static void CollectColorRedTransforms(const uint32_t* argb, int stride,
-                                      int tile_width, int tile_height,
-                                      int green_to_red, int histo[]) {
+static void CollectColorRedTransforms_MIPSdspR2(const uint32_t* argb,
+                                                int stride,
+                                                int tile_width,
+                                                int tile_height,
+                                                int green_to_red,
+                                                int histo[]) {
   const int gtr = (green_to_red << 16) | (green_to_red & 0xffff);
   while (tile_height-- > 0) {
     int x;
@@ -262,10 +268,10 @@ static void CollectColorRedTransforms(const uint32_t* argb, int stride,
 extern void VP8LEncDspInitMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitMIPSdspR2(void) {
-  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
-  VP8LTransformColor = TransformColor;
-  VP8LCollectColorBlueTransforms = CollectColorBlueTransforms;
-  VP8LCollectColorRedTransforms = CollectColorRedTransforms;
+  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed_MIPSdspR2;
+  VP8LTransformColor = TransformColor_MIPSdspR2;
+  VP8LCollectColorBlueTransforms = CollectColorBlueTransforms_MIPSdspR2;
+  VP8LCollectColorRedTransforms = CollectColorRedTransforms_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc_msa.c b/src/3rdparty/libwebp/src/dsp/lossless_enc_msa.c
index 2f69ba3..600dddf 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc_msa.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc_msa.c
@@ -11,12 +11,12 @@
 //
 // Authors: Prashant Patil (Prashant.Patil@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MSA)
 
-#include "./lossless.h"
-#include "./msa_macro.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/msa_macro.h"
 
 #define TRANSFORM_COLOR_8(src0, src1, dst0, dst1, c0, c1, mask0, mask1) do {  \
   v8i16 g0, g1, t0, t1, t2, t3;                                               \
@@ -48,8 +48,8 @@
   dst = VSHF_UB(src, t0, mask1);                                \
 } while (0)
 
-static void TransformColor(const VP8LMultipliers* const m, uint32_t* data,
-                           int num_pixels) {
+static void TransformColor_MSA(const VP8LMultipliers* const m, uint32_t* data,
+                               int num_pixels) {
   v16u8 src0, dst0;
   const v16i8 g2br = (v16i8)__msa_fill_w(m->green_to_blue_ |
                                          (m->green_to_red_ << 16));
@@ -94,7 +94,8 @@ static void TransformColor(const VP8LMultipliers* const m, uint32_t* data,
   }
 }
 
-static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
+static void SubtractGreenFromBlueAndRed_MSA(uint32_t* argb_data,
+                                            int num_pixels) {
   int i;
   uint8_t* ptemp_data = (uint8_t*)argb_data;
   v16u8 src0, dst0, tmp0;
@@ -136,8 +137,8 @@ static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
 extern void VP8LEncDspInitMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitMSA(void) {
-  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
-  VP8LTransformColor = TransformColor;
+  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed_MSA;
+  VP8LTransformColor = TransformColor_MSA;
 }
 
 #else  // !WEBP_USE_MSA
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c b/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c
index 4c56f25..7c7b73f 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c
@@ -11,14 +11,14 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_NEON)
 
 #include <arm_neon.h>
 
-#include "./lossless.h"
-#include "./neon.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/neon.h"
 
 //------------------------------------------------------------------------------
 // Subtract-Green Transform
@@ -36,8 +36,8 @@ static const uint8_t kGreenShuffle[16] = {
   1, 255, 1, 255, 5, 255, 5, 255, 9, 255, 9, 255, 13, 255, 13, 255
 };
 
-static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
-                                             const uint8x16_t shuffle) {
+static WEBP_INLINE uint8x16_t DoGreenShuffle_NEON(const uint8x16_t argb,
+                                                  const uint8x16_t shuffle) {
   return vcombine_u8(vtbl1q_u8(argb, vget_low_u8(shuffle)),
                      vtbl1q_u8(argb, vget_high_u8(shuffle)));
 }
@@ -45,14 +45,15 @@ static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
 // 255 = byte will be zeroed
 static const uint8_t kGreenShuffle[8] = { 1, 255, 1, 255, 5, 255, 5, 255  };
 
-static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
-                                             const uint8x8_t shuffle) {
+static WEBP_INLINE uint8x16_t DoGreenShuffle_NEON(const uint8x16_t argb,
+                                                  const uint8x8_t shuffle) {
   return vcombine_u8(vtbl1_u8(vget_low_u8(argb), shuffle),
                      vtbl1_u8(vget_high_u8(argb), shuffle));
 }
 #endif  // USE_VTBLQ
 
-static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
+static void SubtractGreenFromBlueAndRed_NEON(uint32_t* argb_data,
+                                             int num_pixels) {
   const uint32_t* const end = argb_data + (num_pixels & ~3);
 #ifdef USE_VTBLQ
   const uint8x16_t shuffle = vld1q_u8(kGreenShuffle);
@@ -61,7 +62,7 @@ static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
 #endif
   for (; argb_data < end; argb_data += 4) {
     const uint8x16_t argb = vld1q_u8((uint8_t*)argb_data);
-    const uint8x16_t greens = DoGreenShuffle(argb, shuffle);
+    const uint8x16_t greens = DoGreenShuffle_NEON(argb, shuffle);
     vst1q_u8((uint8_t*)argb_data, vsubq_u8(argb, greens));
   }
   // fallthrough and finish off with plain-C
@@ -71,8 +72,8 @@ static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
 //------------------------------------------------------------------------------
 // Color Transform
 
-static void TransformColor(const VP8LMultipliers* const m,
-                           uint32_t* argb_data, int num_pixels) {
+static void TransformColor_NEON(const VP8LMultipliers* const m,
+                                uint32_t* argb_data, int num_pixels) {
   // sign-extended multiplying constants, pre-shifted by 6.
 #define CST(X)  (((int16_t)(m->X << 8)) >> 6)
   const int16_t rb[8] = {
@@ -102,7 +103,7 @@ static void TransformColor(const VP8LMultipliers* const m,
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     const uint8x16_t in = vld1q_u8((uint8_t*)(argb_data + i));
     // 0 g 0 g
-    const uint8x16_t greens = DoGreenShuffle(in, shuffle);
+    const uint8x16_t greens = DoGreenShuffle_NEON(in, shuffle);
     // x dr  x db1
     const int16x8_t A = vqdmulhq_s16(vreinterpretq_s16_u8(greens), mults_rb);
     // r 0   b   0
@@ -132,8 +133,8 @@ static void TransformColor(const VP8LMultipliers* const m,
 extern void VP8LEncDspInitNEON(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitNEON(void) {
-  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
-  VP8LTransformColor = TransformColor;
+  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed_NEON;
+  VP8LTransformColor = TransformColor_NEON;
 }
 
 #else  // !WEBP_USE_NEON
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc_sse2.c b/src/3rdparty/libwebp/src/dsp/lossless_enc_sse2.c
index 8ad85d9..1eaf35c 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc_sse2.c
@@ -11,22 +11,23 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 #include <assert.h>
 #include <emmintrin.h>
-#include "./lossless.h"
-#include "./common_sse2.h"
-#include "./lossless_common.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/common_sse2.h"
+#include "src/dsp/lossless_common.h"
 
 // For sign-extended multiplying constants, pre-shifted by 5:
-#define CST_5b(X)  (((int16_t)((uint16_t)X << 8)) >> 5)
+#define CST_5b(X)  (((int16_t)((uint16_t)(X) << 8)) >> 5)
 
 //------------------------------------------------------------------------------
 // Subtract-Green Transform
 
-static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
+static void SubtractGreenFromBlueAndRed_SSE2(uint32_t* argb_data,
+                                             int num_pixels) {
   int i;
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); // argb
@@ -45,8 +46,8 @@ static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
 //------------------------------------------------------------------------------
 // Color Transform
 
-static void TransformColor(const VP8LMultipliers* const m,
-                           uint32_t* argb_data, int num_pixels) {
+static void TransformColor_SSE2(const VP8LMultipliers* const m,
+                                uint32_t* argb_data, int num_pixels) {
   const __m128i mults_rb = _mm_set_epi16(
       CST_5b(m->green_to_red_), CST_5b(m->green_to_blue_),
       CST_5b(m->green_to_red_), CST_5b(m->green_to_blue_),
@@ -80,10 +81,10 @@ static void TransformColor(const VP8LMultipliers* const m,
 
 //------------------------------------------------------------------------------
 #define SPAN 8
-static void CollectColorBlueTransforms(const uint32_t* argb, int stride,
-                                       int tile_width, int tile_height,
-                                       int green_to_blue, int red_to_blue,
-                                       int histo[]) {
+static void CollectColorBlueTransforms_SSE2(const uint32_t* argb, int stride,
+                                            int tile_width, int tile_height,
+                                            int green_to_blue, int red_to_blue,
+                                            int histo[]) {
   const __m128i mults_r = _mm_set_epi16(
       CST_5b(red_to_blue), 0, CST_5b(red_to_blue), 0,
       CST_5b(red_to_blue), 0, CST_5b(red_to_blue), 0);
@@ -131,9 +132,9 @@ static void CollectColorBlueTransforms(const uint32_t* argb, int stride,
   }
 }
 
-static void CollectColorRedTransforms(const uint32_t* argb, int stride,
-                                      int tile_width, int tile_height,
-                                      int green_to_red, int histo[]) {
+static void CollectColorRedTransforms_SSE2(const uint32_t* argb, int stride,
+                                           int tile_width, int tile_height,
+                                           int green_to_red, int histo[]) {
   const __m128i mults_g = _mm_set_epi16(
       0, CST_5b(green_to_red), 0, CST_5b(green_to_red),
       0, CST_5b(green_to_red), 0, CST_5b(green_to_red));
@@ -177,8 +178,8 @@ static void CollectColorRedTransforms(const uint32_t* argb, int stride,
 //------------------------------------------------------------------------------
 
 #define LINE_SIZE 16    // 8 or 16
-static void AddVector(const uint32_t* a, const uint32_t* b, uint32_t* out,
-                      int size) {
+static void AddVector_SSE2(const uint32_t* a, const uint32_t* b, uint32_t* out,
+                           int size) {
   int i;
   assert(size % LINE_SIZE == 0);
   for (i = 0; i < size; i += LINE_SIZE) {
@@ -203,7 +204,7 @@ static void AddVector(const uint32_t* a, const uint32_t* b, uint32_t* out,
   }
 }
 
-static void AddVectorEq(const uint32_t* a, uint32_t* out, int size) {
+static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
   int i;
   assert(size % LINE_SIZE == 0);
   for (i = 0; i < size; i += LINE_SIZE) {
@@ -231,22 +232,22 @@ static void AddVectorEq(const uint32_t* a, uint32_t* out, int size) {
 
 // Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
 // that's ok since the histogram values are less than 1<<28 (max picture size).
-static void HistogramAdd(const VP8LHistogram* const a,
-                         const VP8LHistogram* const b,
-                         VP8LHistogram* const out) {
+static void HistogramAdd_SSE2(const VP8LHistogram* const a,
+                              const VP8LHistogram* const b,
+                              VP8LHistogram* const out) {
   int i;
   const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
   assert(a->palette_code_bits_ == b->palette_code_bits_);
   if (b != out) {
-    AddVector(a->literal_, b->literal_, out->literal_, NUM_LITERAL_CODES);
-    AddVector(a->red_, b->red_, out->red_, NUM_LITERAL_CODES);
-    AddVector(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES);
-    AddVector(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES);
+    AddVector_SSE2(a->literal_, b->literal_, out->literal_, NUM_LITERAL_CODES);
+    AddVector_SSE2(a->red_, b->red_, out->red_, NUM_LITERAL_CODES);
+    AddVector_SSE2(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES);
+    AddVector_SSE2(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES);
   } else {
-    AddVectorEq(a->literal_, out->literal_, NUM_LITERAL_CODES);
-    AddVectorEq(a->red_, out->red_, NUM_LITERAL_CODES);
-    AddVectorEq(a->blue_, out->blue_, NUM_LITERAL_CODES);
-    AddVectorEq(a->alpha_, out->alpha_, NUM_LITERAL_CODES);
+    AddVectorEq_SSE2(a->literal_, out->literal_, NUM_LITERAL_CODES);
+    AddVectorEq_SSE2(a->red_, out->red_, NUM_LITERAL_CODES);
+    AddVectorEq_SSE2(a->blue_, out->blue_, NUM_LITERAL_CODES);
+    AddVectorEq_SSE2(a->alpha_, out->alpha_, NUM_LITERAL_CODES);
   }
   for (i = NUM_LITERAL_CODES; i < literal_size; ++i) {
     out->literal_[i] = a->literal_[i] + b->literal_[i];
@@ -261,9 +262,9 @@ static void HistogramAdd(const VP8LHistogram* const a,
 
 // Checks whether the X or Y contribution is worth computing and adding.
 // Used in loop unrolling.
-#define ANALYZE_X_OR_Y(x_or_y, j)                                   \
-  do {                                                              \
-    if (x_or_y[i + j] != 0) retval -= VP8LFastSLog2(x_or_y[i + j]); \
+#define ANALYZE_X_OR_Y(x_or_y, j)                                           \
+  do {                                                                      \
+    if ((x_or_y)[i + (j)] != 0) retval -= VP8LFastSLog2((x_or_y)[i + (j)]); \
   } while (0)
 
 // Checks whether the X + Y contribution is worth computing and adding.
@@ -276,7 +277,7 @@ static void HistogramAdd(const VP8LHistogram* const a,
     }                                  \
   } while (0)
 
-static float CombinedShannonEntropy(const int X[256], const int Y[256]) {
+static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
   int i;
   double retval = 0.;
   int sumX, sumXY;
@@ -332,8 +333,8 @@ static float CombinedShannonEntropy(const int X[256], const int Y[256]) {
 
 //------------------------------------------------------------------------------
 
-static int VectorMismatch(const uint32_t* const array1,
-                          const uint32_t* const array2, int length) {
+static int VectorMismatch_SSE2(const uint32_t* const array1,
+                               const uint32_t* const array2, int length) {
   int match_len;
 
   if (length >= 12) {
@@ -574,8 +575,8 @@ static void PredictorSub10_SSE2(const uint32_t* in, const uint32_t* upper,
 }
 
 // Predictor11: select.
-static void GetSumAbsDiff32(const __m128i* const A, const __m128i* const B,
-                            __m128i* const out) {
+static void GetSumAbsDiff32_SSE2(const __m128i* const A, const __m128i* const B,
+                                 __m128i* const out) {
   // We can unpack with any value on the upper 32 bits, provided it's the same
   // on both operands (to that their sum of abs diff is zero). Here we use *A.
   const __m128i A_lo = _mm_unpacklo_epi32(*A, *A);
@@ -596,8 +597,8 @@ static void PredictorSub11_SSE2(const uint32_t* in, const uint32_t* upper,
     const __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
     const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
     __m128i pa, pb;
-    GetSumAbsDiff32(&T, &TL, &pa);   // pa = sum |T-TL|
-    GetSumAbsDiff32(&L, &TL, &pb);   // pb = sum |L-TL|
+    GetSumAbsDiff32_SSE2(&T, &TL, &pa);   // pa = sum |T-TL|
+    GetSumAbsDiff32_SSE2(&L, &TL, &pb);   // pb = sum |L-TL|
     {
       const __m128i mask = _mm_cmpgt_epi32(pb, pa);
       const __m128i A = _mm_and_si128(mask, L);
@@ -677,13 +678,13 @@ static void PredictorSub13_SSE2(const uint32_t* in, const uint32_t* upper,
 extern void VP8LEncDspInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE2(void) {
-  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
-  VP8LTransformColor = TransformColor;
-  VP8LCollectColorBlueTransforms = CollectColorBlueTransforms;
-  VP8LCollectColorRedTransforms = CollectColorRedTransforms;
-  VP8LHistogramAdd = HistogramAdd;
-  VP8LCombinedShannonEntropy = CombinedShannonEntropy;
-  VP8LVectorMismatch = VectorMismatch;
+  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed_SSE2;
+  VP8LTransformColor = TransformColor_SSE2;
+  VP8LCollectColorBlueTransforms = CollectColorBlueTransforms_SSE2;
+  VP8LCollectColorRedTransforms = CollectColorRedTransforms_SSE2;
+  VP8LHistogramAdd = HistogramAdd_SSE2;
+  VP8LCombinedShannonEntropy = CombinedShannonEntropy_SSE2;
+  VP8LVectorMismatch = VectorMismatch_SSE2;
   VP8LBundleColorMap = BundleColorMap_SSE2;
 
   VP8LPredictorsSub[0] = PredictorSub0_SSE2;
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc_sse41.c b/src/3rdparty/libwebp/src/dsp/lossless_enc_sse41.c
index 821057c..3526a34 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc_sse41.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc_sse41.c
@@ -11,17 +11,18 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE41)
 #include <assert.h>
 #include <smmintrin.h>
-#include "./lossless.h"
+#include "src/dsp/lossless.h"
 
 //------------------------------------------------------------------------------
 // Subtract-Green Transform
 
-static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
+static void SubtractGreenFromBlueAndRed_SSE41(uint32_t* argb_data,
+                                              int num_pixels) {
   int i;
   const __m128i kCstShuffle = _mm_set_epi8(-1, 13, -1, 13, -1, 9, -1, 9,
                                            -1,  5, -1,  5, -1, 1, -1, 1);
@@ -43,7 +44,7 @@ static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
 extern void VP8LEncDspInitSSE41(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE41(void) {
-  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
+  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed_SSE41;
 }
 
 #else  // !WEBP_USE_SSE41
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/lossless_mips_dsp_r2.c
index 2984ce8..9888854 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_mips_dsp_r2.c
@@ -12,12 +12,12 @@
 // Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
 //             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-#include "./lossless.h"
-#include "./lossless_common.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
 
 #define MAP_COLOR_FUNCS(FUNC_NAME, TYPE, GET_INDEX, GET_VALUE)                 \
 static void FUNC_NAME(const TYPE* src,                                         \
@@ -86,8 +86,8 @@ static void FUNC_NAME(const TYPE* src,                                         \
   }                                                                            \
 }
 
-MAP_COLOR_FUNCS(MapARGB, uint32_t, VP8GetARGBIndex, VP8GetARGBValue)
-MAP_COLOR_FUNCS(MapAlpha, uint8_t, VP8GetAlphaIndex, VP8GetAlphaValue)
+MAP_COLOR_FUNCS(MapARGB_MIPSdspR2, uint32_t, VP8GetARGBIndex, VP8GetARGBValue)
+MAP_COLOR_FUNCS(MapAlpha_MIPSdspR2, uint8_t, VP8GetAlphaIndex, VP8GetAlphaValue)
 
 #undef MAP_COLOR_FUNCS
 
@@ -188,48 +188,52 @@ static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
   return Average2(Average2(a0, a1), Average2(a2, a3));
 }
 
-static uint32_t Predictor5(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor5_MIPSdspR2(uint32_t left, const uint32_t* const top) {
   return Average3(left, top[0], top[1]);
 }
 
-static uint32_t Predictor6(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor6_MIPSdspR2(uint32_t left, const uint32_t* const top) {
   return Average2(left, top[-1]);
 }
 
-static uint32_t Predictor7(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor7_MIPSdspR2(uint32_t left, const uint32_t* const top) {
   return Average2(left, top[0]);
 }
 
-static uint32_t Predictor8(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor8_MIPSdspR2(uint32_t left, const uint32_t* const top) {
   (void)left;
   return Average2(top[-1], top[0]);
 }
 
-static uint32_t Predictor9(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor9_MIPSdspR2(uint32_t left, const uint32_t* const top) {
   (void)left;
   return Average2(top[0], top[1]);
 }
 
-static uint32_t Predictor10(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor10_MIPSdspR2(uint32_t left,
+                                      const uint32_t* const top) {
   return Average4(left, top[-1], top[0], top[1]);
 }
 
-static uint32_t Predictor11(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor11_MIPSdspR2(uint32_t left,
+                                      const uint32_t* const top) {
   return Select(top[0], left, top[-1]);
 }
 
-static uint32_t Predictor12(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor12_MIPSdspR2(uint32_t left,
+                                      const uint32_t* const top) {
   return ClampedAddSubtractFull(left, top[0], top[-1]);
 }
 
-static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor13_MIPSdspR2(uint32_t left,
+                                      const uint32_t* const top) {
   return ClampedAddSubtractHalf(left, top[0], top[-1]);
 }
 
 // Add green to blue and red channels (i.e. perform the inverse transform of
 // 'subtract green').
-static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels,
-                                 uint32_t* dst) {
+static void AddGreenToBlueAndRed_MIPSdspR2(const uint32_t* src, int num_pixels,
+                                           uint32_t* dst) {
   uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
   const uint32_t* const p_loop2_end = src + num_pixels;
@@ -285,9 +289,9 @@ static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels,
   );
 }
 
-static void TransformColorInverse(const VP8LMultipliers* const m,
-                                  const uint32_t* src, int num_pixels,
-                                  uint32_t* dst) {
+static void TransformColorInverse_MIPSdspR2(const VP8LMultipliers* const m,
+                                            const uint32_t* src, int num_pixels,
+                                            uint32_t* dst) {
   int temp0, temp1, temp2, temp3, temp4, temp5;
   uint32_t argb, argb1, new_red;
   const uint32_t G_to_R = m->green_to_red_;
@@ -356,8 +360,8 @@ static void TransformColorInverse(const VP8LMultipliers* const m,
   if (num_pixels & 1) VP8LTransformColorInverse_C(m, src, 1, dst);
 }
 
-static void ConvertBGRAToRGB(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGB_MIPSdspR2(const uint32_t* src,
+                                       int num_pixels, uint8_t* dst) {
   int temp0, temp1, temp2, temp3;
   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
   const uint32_t* const p_loop2_end = src + num_pixels;
@@ -408,8 +412,8 @@ static void ConvertBGRAToRGB(const uint32_t* src,
   );
 }
 
-static void ConvertBGRAToRGBA(const uint32_t* src,
-                              int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGBA_MIPSdspR2(const uint32_t* src,
+                                        int num_pixels, uint8_t* dst) {
   int temp0, temp1, temp2, temp3;
   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
   const uint32_t* const p_loop2_end = src + num_pixels;
@@ -458,8 +462,8 @@ static void ConvertBGRAToRGBA(const uint32_t* src,
   );
 }
 
-static void ConvertBGRAToRGBA4444(const uint32_t* src,
-                                  int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGBA4444_MIPSdspR2(const uint32_t* src,
+                                            int num_pixels, uint8_t* dst) {
   int temp0, temp1, temp2, temp3, temp4, temp5;
   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
   const uint32_t* const p_loop2_end = src + num_pixels;
@@ -492,7 +496,7 @@ static void ConvertBGRAToRGBA4444(const uint32_t* src,
     "ins            %[temp3],    %[temp5],          16,   4    \n\t"
     "addiu          %[src],      %[src],            16         \n\t"
     "precr.qb.ph    %[temp3],    %[temp3],          %[temp2]   \n\t"
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     "usw            %[temp1],    0(%[dst])                     \n\t"
     "usw            %[temp3],    4(%[dst])                     \n\t"
 #else
@@ -514,7 +518,7 @@ static void ConvertBGRAToRGBA4444(const uint32_t* src,
     "ins            %[temp0],    %[temp5],          16,   4    \n\t"
     "addiu          %[src],      %[src],            4          \n\t"
     "precr.qb.ph    %[temp0],    %[temp0],          %[temp0]   \n\t"
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     "ush            %[temp0],    0(%[dst])                     \n\t"
 #else
     "wsbh           %[temp0],    %[temp0]                      \n\t"
@@ -532,8 +536,8 @@ static void ConvertBGRAToRGBA4444(const uint32_t* src,
   );
 }
 
-static void ConvertBGRAToRGB565(const uint32_t* src,
-                                int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGB565_MIPSdspR2(const uint32_t* src,
+                                          int num_pixels, uint8_t* dst) {
   int temp0, temp1, temp2, temp3, temp4, temp5;
   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
   const uint32_t* const p_loop2_end = src + num_pixels;
@@ -570,7 +574,7 @@ static void ConvertBGRAToRGB565(const uint32_t* src,
     "ins            %[temp2],    %[temp3],          0,    5    \n\t"
     "addiu          %[src],      %[src],            16         \n\t"
     "append         %[temp2],    %[temp1],          16         \n\t"
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     "usw            %[temp0],    0(%[dst])                     \n\t"
     "usw            %[temp2],    4(%[dst])                     \n\t"
 #else
@@ -592,7 +596,7 @@ static void ConvertBGRAToRGB565(const uint32_t* src,
     "ins            %[temp4],    %[temp5],          0,    11   \n\t"
     "addiu          %[src],      %[src],            4          \n\t"
     "ins            %[temp4],    %[temp0],          0,    5    \n\t"
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     "ush            %[temp4],    0(%[dst])                     \n\t"
 #else
     "wsbh           %[temp4],    %[temp4]                      \n\t"
@@ -610,8 +614,8 @@ static void ConvertBGRAToRGB565(const uint32_t* src,
   );
 }
 
-static void ConvertBGRAToBGR(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToBGR_MIPSdspR2(const uint32_t* src,
+                                       int num_pixels, uint8_t* dst) {
   int temp0, temp1, temp2, temp3;
   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
   const uint32_t* const p_loop2_end = src + num_pixels;
@@ -662,24 +666,27 @@ static void ConvertBGRAToBGR(const uint32_t* src,
 extern void VP8LDspInitMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitMIPSdspR2(void) {
-  VP8LMapColor32b = MapARGB;
-  VP8LMapColor8b = MapAlpha;
-  VP8LPredictors[5] = Predictor5;
-  VP8LPredictors[6] = Predictor6;
-  VP8LPredictors[7] = Predictor7;
-  VP8LPredictors[8] = Predictor8;
-  VP8LPredictors[9] = Predictor9;
-  VP8LPredictors[10] = Predictor10;
-  VP8LPredictors[11] = Predictor11;
-  VP8LPredictors[12] = Predictor12;
-  VP8LPredictors[13] = Predictor13;
-  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
-  VP8LTransformColorInverse = TransformColorInverse;
-  VP8LConvertBGRAToRGB = ConvertBGRAToRGB;
-  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
-  VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444;
-  VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565;
-  VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
+  VP8LMapColor32b = MapARGB_MIPSdspR2;
+  VP8LMapColor8b = MapAlpha_MIPSdspR2;
+
+  VP8LPredictors[5] = Predictor5_MIPSdspR2;
+  VP8LPredictors[6] = Predictor6_MIPSdspR2;
+  VP8LPredictors[7] = Predictor7_MIPSdspR2;
+  VP8LPredictors[8] = Predictor8_MIPSdspR2;
+  VP8LPredictors[9] = Predictor9_MIPSdspR2;
+  VP8LPredictors[10] = Predictor10_MIPSdspR2;
+  VP8LPredictors[11] = Predictor11_MIPSdspR2;
+  VP8LPredictors[12] = Predictor12_MIPSdspR2;
+  VP8LPredictors[13] = Predictor13_MIPSdspR2;
+
+  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed_MIPSdspR2;
+  VP8LTransformColorInverse = TransformColorInverse_MIPSdspR2;
+
+  VP8LConvertBGRAToRGB = ConvertBGRAToRGB_MIPSdspR2;
+  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA_MIPSdspR2;
+  VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444_MIPSdspR2;
+  VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565_MIPSdspR2;
+  VP8LConvertBGRAToBGR = ConvertBGRAToBGR_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_msa.c b/src/3rdparty/libwebp/src/dsp/lossless_msa.c
index f6dd564..9f54720 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_msa.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_msa.c
@@ -11,12 +11,12 @@
 //
 // Author: Prashant Patil (prashant.patil@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MSA)
 
-#include "./lossless.h"
-#include "./msa_macro.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/msa_macro.h"
 
 //------------------------------------------------------------------------------
 // Colorspace conversion functions
@@ -43,7 +43,7 @@
 
 #define CONVERT8_BGRA_XXX(psrc, pdst, m0, m1) do {         \
   uint64_t pix_d;                                          \
-  v16u8 src0, src1, src2, dst0, dst1;                      \
+  v16u8 src0, src1, src2 = { 0 }, dst0, dst1;              \
   LD_UB2(psrc, 16, src0, src1);                            \
   VSHF_B2_UB(src0, src1, src1, src2, m0, m1, dst0, dst1);  \
   ST_UB(dst0, pdst);                                       \
@@ -109,8 +109,8 @@
   dst = VSHF_UB(src, t0, mask1);                                        \
 } while (0)
 
-static void ConvertBGRAToRGBA(const uint32_t* src,
-                              int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGBA_MSA(const uint32_t* src,
+                                  int num_pixels, uint8_t* dst) {
   int i;
   const uint8_t* ptemp_src = (const uint8_t*)src;
   uint8_t* ptemp_dst = (uint8_t*)dst;
@@ -150,8 +150,8 @@ static void ConvertBGRAToRGBA(const uint32_t* src,
   }
 }
 
-static void ConvertBGRAToBGR(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToBGR_MSA(const uint32_t* src,
+                                 int num_pixels, uint8_t* dst) {
   const uint8_t* ptemp_src = (const uint8_t*)src;
   uint8_t* ptemp_dst = (uint8_t*)dst;
   const v16u8 mask0 = { 0, 1, 2, 4, 5, 6, 8, 9, 10, 12, 13, 14,
@@ -197,8 +197,8 @@ static void ConvertBGRAToBGR(const uint32_t* src,
   }
 }
 
-static void ConvertBGRAToRGB(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGB_MSA(const uint32_t* src,
+                                 int num_pixels, uint8_t* dst) {
   const uint8_t* ptemp_src = (const uint8_t*)src;
   uint8_t* ptemp_dst = (uint8_t*)dst;
   const v16u8 mask0 = { 2, 1, 0, 6, 5, 4, 10, 9, 8, 14, 13, 12,
@@ -244,8 +244,8 @@ static void ConvertBGRAToRGB(const uint32_t* src,
   }
 }
 
-static void AddGreenToBlueAndRed(const uint32_t* const src, int num_pixels,
-                                 uint32_t* dst) {
+static void AddGreenToBlueAndRed_MSA(const uint32_t* const src, int num_pixels,
+                                     uint32_t* dst) {
   int i;
   const uint8_t* in = (const uint8_t*)src;
   uint8_t* out = (uint8_t*)dst;
@@ -286,9 +286,9 @@ static void AddGreenToBlueAndRed(const uint32_t* const src, int num_pixels,
   }
 }
 
-static void TransformColorInverse(const VP8LMultipliers* const m,
-                                  const uint32_t* src, int num_pixels,
-                                  uint32_t* dst) {
+static void TransformColorInverse_MSA(const VP8LMultipliers* const m,
+                                      const uint32_t* src, int num_pixels,
+                                      uint32_t* dst) {
   v16u8 src0, dst0;
   const v16i8 g2br = (v16i8)__msa_fill_w(m->green_to_blue_ |
                                          (m->green_to_red_ << 16));
@@ -341,11 +341,12 @@ static void TransformColorInverse(const VP8LMultipliers* const m,
 extern void VP8LDspInitMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitMSA(void) {
-  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
-  VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
-  VP8LConvertBGRAToRGB = ConvertBGRAToRGB;
-  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
-  VP8LTransformColorInverse = TransformColorInverse;
+  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA_MSA;
+  VP8LConvertBGRAToBGR = ConvertBGRAToBGR_MSA;
+  VP8LConvertBGRAToRGB = ConvertBGRAToRGB_MSA;
+
+  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed_MSA;
+  VP8LTransformColorInverse = TransformColorInverse_MSA;
 }
 
 #else  // !WEBP_USE_MSA
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_neon.c b/src/3rdparty/libwebp/src/dsp/lossless_neon.c
index 1145d5f..76a1b6f 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_neon.c
@@ -11,14 +11,14 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_NEON)
 
 #include <arm_neon.h>
 
-#include "./lossless.h"
-#include "./neon.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/neon.h"
 
 //------------------------------------------------------------------------------
 // Colorspace conversion functions
@@ -26,8 +26,8 @@
 #if !defined(WORK_AROUND_GCC)
 // gcc 4.6.0 had some trouble (NDK-r9) with this code. We only use it for
 // gcc-4.8.x at least.
-static void ConvertBGRAToRGBA(const uint32_t* src,
-                              int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGBA_NEON(const uint32_t* src,
+                                   int num_pixels, uint8_t* dst) {
   const uint32_t* const end = src + (num_pixels & ~15);
   for (; src < end; src += 16) {
     uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
@@ -41,8 +41,8 @@ static void ConvertBGRAToRGBA(const uint32_t* src,
   VP8LConvertBGRAToRGBA_C(src, num_pixels & 15, dst);  // left-overs
 }
 
-static void ConvertBGRAToBGR(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToBGR_NEON(const uint32_t* src,
+                                  int num_pixels, uint8_t* dst) {
   const uint32_t* const end = src + (num_pixels & ~15);
   for (; src < end; src += 16) {
     const uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
@@ -53,8 +53,8 @@ static void ConvertBGRAToBGR(const uint32_t* src,
   VP8LConvertBGRAToBGR_C(src, num_pixels & 15, dst);  // left-overs
 }
 
-static void ConvertBGRAToRGB(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGB_NEON(const uint32_t* src,
+                                  int num_pixels, uint8_t* dst) {
   const uint32_t* const end = src + (num_pixels & ~15);
   for (; src < end; src += 16) {
     const uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
@@ -71,8 +71,8 @@ static void ConvertBGRAToRGB(const uint32_t* src,
 
 static const uint8_t kRGBAShuffle[8] = { 2, 1, 0, 3, 6, 5, 4, 7 };
 
-static void ConvertBGRAToRGBA(const uint32_t* src,
-                              int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGBA_NEON(const uint32_t* src,
+                                   int num_pixels, uint8_t* dst) {
   const uint32_t* const end = src + (num_pixels & ~1);
   const uint8x8_t shuffle = vld1_u8(kRGBAShuffle);
   for (; src < end; src += 2) {
@@ -89,8 +89,8 @@ static const uint8_t kBGRShuffle[3][8] = {
   { 21, 22, 24, 25, 26, 28, 29, 30 }
 };
 
-static void ConvertBGRAToBGR(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToBGR_NEON(const uint32_t* src,
+                                  int num_pixels, uint8_t* dst) {
   const uint32_t* const end = src + (num_pixels & ~7);
   const uint8x8_t shuffle0 = vld1_u8(kBGRShuffle[0]);
   const uint8x8_t shuffle1 = vld1_u8(kBGRShuffle[1]);
@@ -116,8 +116,8 @@ static const uint8_t kRGBShuffle[3][8] = {
   { 21, 20, 26, 25, 24, 30, 29, 28 }
 };
 
-static void ConvertBGRAToRGB(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGB_NEON(const uint32_t* src,
+                                  int num_pixels, uint8_t* dst) {
   const uint32_t* const end = src + (num_pixels & ~7);
   const uint8x8_t shuffle0 = vld1_u8(kRGBShuffle[0]);
   const uint8x8_t shuffle1 = vld1_u8(kRGBShuffle[1]);
@@ -139,7 +139,6 @@ static void ConvertBGRAToRGB(const uint32_t* src,
 
 #endif   // !WORK_AROUND_GCC
 
-
 //------------------------------------------------------------------------------
 // Predictor Transform
 
@@ -506,8 +505,8 @@ static const uint8_t kGreenShuffle[16] = {
   1, 255, 1, 255, 5, 255, 5, 255, 9, 255, 9, 255, 13, 255, 13, 255
 };
 
-static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
-                                             const uint8x16_t shuffle) {
+static WEBP_INLINE uint8x16_t DoGreenShuffle_NEON(const uint8x16_t argb,
+                                                  const uint8x16_t shuffle) {
   return vcombine_u8(vtbl1q_u8(argb, vget_low_u8(shuffle)),
                      vtbl1q_u8(argb, vget_high_u8(shuffle)));
 }
@@ -515,15 +514,15 @@ static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
 // 255 = byte will be zeroed
 static const uint8_t kGreenShuffle[8] = { 1, 255, 1, 255, 5, 255, 5, 255  };
 
-static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb,
-                                             const uint8x8_t shuffle) {
+static WEBP_INLINE uint8x16_t DoGreenShuffle_NEON(const uint8x16_t argb,
+                                                  const uint8x8_t shuffle) {
   return vcombine_u8(vtbl1_u8(vget_low_u8(argb), shuffle),
                      vtbl1_u8(vget_high_u8(argb), shuffle));
 }
 #endif  // USE_VTBLQ
 
-static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels,
-                                 uint32_t* dst) {
+static void AddGreenToBlueAndRed_NEON(const uint32_t* src, int num_pixels,
+                                      uint32_t* dst) {
   const uint32_t* const end = src + (num_pixels & ~3);
 #ifdef USE_VTBLQ
   const uint8x16_t shuffle = vld1q_u8(kGreenShuffle);
@@ -532,7 +531,7 @@ static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels,
 #endif
   for (; src < end; src += 4, dst += 4) {
     const uint8x16_t argb = vld1q_u8((const uint8_t*)src);
-    const uint8x16_t greens = DoGreenShuffle(argb, shuffle);
+    const uint8x16_t greens = DoGreenShuffle_NEON(argb, shuffle);
     vst1q_u8((uint8_t*)dst, vaddq_u8(argb, greens));
   }
   // fallthrough and finish off with plain-C
@@ -542,9 +541,9 @@ static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels,
 //------------------------------------------------------------------------------
 // Color Transform
 
-static void TransformColorInverse(const VP8LMultipliers* const m,
-                                  const uint32_t* const src, int num_pixels,
-                                  uint32_t* dst) {
+static void TransformColorInverse_NEON(const VP8LMultipliers* const m,
+                                       const uint32_t* const src,
+                                       int num_pixels, uint32_t* dst) {
 // sign-extended multiplying constants, pre-shifted by 6.
 #define CST(X)  (((int16_t)(m->X << 8)) >> 6)
   const int16_t rb[8] = {
@@ -575,7 +574,7 @@ static void TransformColorInverse(const VP8LMultipliers* const m,
     const uint8x16_t in = vld1q_u8((const uint8_t*)(src + i));
     const uint32x4_t a0g0 = vandq_u32(vreinterpretq_u32_u8(in), mask_ag);
     // 0 g 0 g
-    const uint8x16_t greens = DoGreenShuffle(in, shuffle);
+    const uint8x16_t greens = DoGreenShuffle_NEON(in, shuffle);
     // x dr  x db1
     const int16x8_t A = vqdmulhq_s16(vreinterpretq_s16_u8(greens), mults_rb);
     // x r'  x   b'
@@ -627,12 +626,12 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitNEON(void) {
   VP8LPredictorsAdd[12] = PredictorAdd12_NEON;
   VP8LPredictorsAdd[13] = PredictorAdd13_NEON;
 
-  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
-  VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
-  VP8LConvertBGRAToRGB = ConvertBGRAToRGB;
+  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA_NEON;
+  VP8LConvertBGRAToBGR = ConvertBGRAToBGR_NEON;
+  VP8LConvertBGRAToRGB = ConvertBGRAToRGB_NEON;
 
-  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
-  VP8LTransformColorInverse = TransformColorInverse;
+  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed_NEON;
+  VP8LTransformColorInverse = TransformColorInverse_NEON;
 }
 
 #else  // !WEBP_USE_NEON
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_sse2.c b/src/3rdparty/libwebp/src/dsp/lossless_sse2.c
index 15aae93..653b466 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_sse2.c
@@ -11,21 +11,22 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 
-#include "./common_sse2.h"
-#include "./lossless.h"
-#include "./lossless_common.h"
+#include "src/dsp/common_sse2.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
 #include <assert.h>
 #include <emmintrin.h>
 
 //------------------------------------------------------------------------------
 // Predictor Transform
 
-static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
-                                                   uint32_t c2) {
+static WEBP_INLINE uint32_t ClampedAddSubtractFull_SSE2(uint32_t c0,
+                                                        uint32_t c1,
+                                                        uint32_t c2) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
   const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
@@ -37,8 +38,9 @@ static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
   return output;
 }
 
-static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
-                                                   uint32_t c2) {
+static WEBP_INLINE uint32_t ClampedAddSubtractHalf_SSE2(uint32_t c0,
+                                                        uint32_t c1,
+                                                        uint32_t c2) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
   const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
@@ -55,7 +57,7 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
   return output;
 }
 
-static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
+static WEBP_INLINE uint32_t Select_SSE2(uint32_t a, uint32_t b, uint32_t c) {
   int pa_minus_pb;
   const __m128i zero = _mm_setzero_si128();
   const __m128i A0 = _mm_cvtsi32_si128(a);
@@ -88,8 +90,9 @@ static WEBP_INLINE void Average2_m128i(const __m128i* const a0,
   *avg = _mm_sub_epi8(avg1, one);
 }
 
-static WEBP_INLINE void Average2_uint32(const uint32_t a0, const uint32_t a1,
-                                        __m128i* const avg) {
+static WEBP_INLINE void Average2_uint32_SSE2(const uint32_t a0,
+                                             const uint32_t a1,
+                                             __m128i* const avg) {
   // (a + b) >> 1 = ((a + b + 1) >> 1) - ((a ^ b) & 1)
   const __m128i ones = _mm_set1_epi8(1);
   const __m128i A0 = _mm_cvtsi32_si128(a0);
@@ -99,7 +102,7 @@ static WEBP_INLINE void Average2_uint32(const uint32_t a0, const uint32_t a1,
   *avg = _mm_sub_epi8(avg1, one);
 }
 
-static WEBP_INLINE __m128i Average2_uint32_16(uint32_t a0, uint32_t a1) {
+static WEBP_INLINE __m128i Average2_uint32_16_SSE2(uint32_t a0, uint32_t a1) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a0), zero);
   const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
@@ -107,15 +110,16 @@ static WEBP_INLINE __m128i Average2_uint32_16(uint32_t a0, uint32_t a1) {
   return _mm_srli_epi16(sum, 1);
 }
 
-static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) {
+static WEBP_INLINE uint32_t Average2_SSE2(uint32_t a0, uint32_t a1) {
   __m128i output;
-  Average2_uint32(a0, a1, &output);
+  Average2_uint32_SSE2(a0, a1, &output);
   return _mm_cvtsi128_si32(output);
 }
 
-static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) {
+static WEBP_INLINE uint32_t Average3_SSE2(uint32_t a0, uint32_t a1,
+                                          uint32_t a2) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i avg1 = Average2_uint32_16(a0, a2);
+  const __m128i avg1 = Average2_uint32_16_SSE2(a0, a2);
   const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
   const __m128i sum = _mm_add_epi16(avg1, A1);
   const __m128i avg2 = _mm_srli_epi16(sum, 1);
@@ -124,10 +128,10 @@ static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) {
   return output;
 }
 
-static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
-                                     uint32_t a2, uint32_t a3) {
-  const __m128i avg1 = Average2_uint32_16(a0, a1);
-  const __m128i avg2 = Average2_uint32_16(a2, a3);
+static WEBP_INLINE uint32_t Average4_SSE2(uint32_t a0, uint32_t a1,
+                                          uint32_t a2, uint32_t a3) {
+  const __m128i avg1 = Average2_uint32_16_SSE2(a0, a1);
+  const __m128i avg2 = Average2_uint32_16_SSE2(a2, a3);
   const __m128i sum = _mm_add_epi16(avg2, avg1);
   const __m128i avg3 = _mm_srli_epi16(sum, 1);
   const __m128i A0 = _mm_packus_epi16(avg3, avg3);
@@ -136,41 +140,41 @@ static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
 }
 
 static uint32_t Predictor5_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average3(left, top[0], top[1]);
+  const uint32_t pred = Average3_SSE2(left, top[0], top[1]);
   return pred;
 }
 static uint32_t Predictor6_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(left, top[-1]);
+  const uint32_t pred = Average2_SSE2(left, top[-1]);
   return pred;
 }
 static uint32_t Predictor7_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(left, top[0]);
+  const uint32_t pred = Average2_SSE2(left, top[0]);
   return pred;
 }
 static uint32_t Predictor8_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(top[-1], top[0]);
+  const uint32_t pred = Average2_SSE2(top[-1], top[0]);
   (void)left;
   return pred;
 }
 static uint32_t Predictor9_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(top[0], top[1]);
+  const uint32_t pred = Average2_SSE2(top[0], top[1]);
   (void)left;
   return pred;
 }
 static uint32_t Predictor10_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average4(left, top[-1], top[0], top[1]);
+  const uint32_t pred = Average4_SSE2(left, top[-1], top[0], top[1]);
   return pred;
 }
 static uint32_t Predictor11_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Select(top[0], left, top[-1]);
+  const uint32_t pred = Select_SSE2(top[0], left, top[-1]);
   return pred;
 }
 static uint32_t Predictor12_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]);
+  const uint32_t pred = ClampedAddSubtractFull_SSE2(left, top[0], top[-1]);
   return pred;
 }
 static uint32_t Predictor13_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]);
+  const uint32_t pred = ClampedAddSubtractHalf_SSE2(left, top[0], top[-1]);
   return pred;
 }
 
@@ -272,9 +276,24 @@ GENERATE_PREDICTOR_2(9, upper[i + 1])
 #undef GENERATE_PREDICTOR_2
 
 // Predictor10: average of (average of (L,TL), average of (T, TR)).
+#define DO_PRED10(OUT) do {               \
+  __m128i avgLTL, avg;                    \
+  Average2_m128i(&L, &TL, &avgLTL);       \
+  Average2_m128i(&avgTTR, &avgLTL, &avg); \
+  L = _mm_add_epi8(avg, src);             \
+  out[i + (OUT)] = _mm_cvtsi128_si32(L);  \
+} while (0)
+
+#define DO_PRED10_SHIFT do {                                  \
+  /* Rotate the pre-computed values for the next iteration.*/ \
+  avgTTR = _mm_srli_si128(avgTTR, 4);                         \
+  TL = _mm_srli_si128(TL, 4);                                 \
+  src = _mm_srli_si128(src, 4);                               \
+} while (0)
+
 static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper,
                                 int num_pixels, uint32_t* out) {
-  int i, j;
+  int i;
   __m128i L = _mm_cvtsi32_si128(out[-1]);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
@@ -283,79 +302,90 @@ static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper,
     const __m128i TR = _mm_loadu_si128((const __m128i*)&upper[i + 1]);
     __m128i avgTTR;
     Average2_m128i(&T, &TR, &avgTTR);
-    for (j = 0; j < 4; ++j) {
-      __m128i avgLTL, avg;
-      Average2_m128i(&L, &TL, &avgLTL);
-      Average2_m128i(&avgTTR, &avgLTL, &avg);
-      L = _mm_add_epi8(avg, src);
-      out[i + j] = _mm_cvtsi128_si32(L);
-      // Rotate the pre-computed values for the next iteration.
-      avgTTR = _mm_srli_si128(avgTTR, 4);
-      TL = _mm_srli_si128(TL, 4);
-      src = _mm_srli_si128(src, 4);
-    }
+    DO_PRED10(0);
+    DO_PRED10_SHIFT;
+    DO_PRED10(1);
+    DO_PRED10_SHIFT;
+    DO_PRED10(2);
+    DO_PRED10_SHIFT;
+    DO_PRED10(3);
   }
   if (i != num_pixels) {
     VP8LPredictorsAdd_C[10](in + i, upper + i, num_pixels - i, out + i);
   }
 }
+#undef DO_PRED10
+#undef DO_PRED10_SHIFT
 
 // Predictor11: select.
-static void GetSumAbsDiff32(const __m128i* const A, const __m128i* const B,
-                            __m128i* const out) {
-  // We can unpack with any value on the upper 32 bits, provided it's the same
-  // on both operands (to that their sum of abs diff is zero). Here we use *A.
-  const __m128i A_lo = _mm_unpacklo_epi32(*A, *A);
-  const __m128i B_lo = _mm_unpacklo_epi32(*B, *A);
-  const __m128i A_hi = _mm_unpackhi_epi32(*A, *A);
-  const __m128i B_hi = _mm_unpackhi_epi32(*B, *A);
-  const __m128i s_lo = _mm_sad_epu8(A_lo, B_lo);
-  const __m128i s_hi = _mm_sad_epu8(A_hi, B_hi);
-  *out = _mm_packs_epi32(s_lo, s_hi);
-}
+#define DO_PRED11(OUT) do {                                            \
+  const __m128i L_lo = _mm_unpacklo_epi32(L, T);                       \
+  const __m128i TL_lo = _mm_unpacklo_epi32(TL, T);                     \
+  const __m128i pb = _mm_sad_epu8(L_lo, TL_lo); /* pb = sum |L-TL|*/   \
+  const __m128i mask = _mm_cmpgt_epi32(pb, pa);                        \
+  const __m128i A = _mm_and_si128(mask, L);                            \
+  const __m128i B = _mm_andnot_si128(mask, T);                         \
+  const __m128i pred = _mm_or_si128(A, B); /* pred = (pa > b)? L : T*/ \
+  L = _mm_add_epi8(src, pred);                                         \
+  out[i + (OUT)] = _mm_cvtsi128_si32(L);                               \
+} while (0)
+
+#define DO_PRED11_SHIFT do {                                \
+  /* Shift the pre-computed value for the next iteration.*/ \
+  T = _mm_srli_si128(T, 4);                                 \
+  TL = _mm_srli_si128(TL, 4);                               \
+  src = _mm_srli_si128(src, 4);                             \
+  pa = _mm_srli_si128(pa, 4);                               \
+} while (0)
 
 static void PredictorAdd11_SSE2(const uint32_t* in, const uint32_t* upper,
                                 int num_pixels, uint32_t* out) {
-  int i, j;
+  int i;
+  __m128i pa;
   __m128i L = _mm_cvtsi32_si128(out[-1]);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     __m128i T = _mm_loadu_si128((const __m128i*)&upper[i]);
     __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
     __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
-    __m128i pa;
-    GetSumAbsDiff32(&T, &TL, &pa);   // pa = sum |T-TL|
-    for (j = 0; j < 4; ++j) {
-      const __m128i L_lo = _mm_unpacklo_epi32(L, L);
-      const __m128i TL_lo = _mm_unpacklo_epi32(TL, L);
-      const __m128i pb = _mm_sad_epu8(L_lo, TL_lo);  // pb = sum |L-TL|
-      const __m128i mask = _mm_cmpgt_epi32(pb, pa);
-      const __m128i A = _mm_and_si128(mask, L);
-      const __m128i B = _mm_andnot_si128(mask, T);
-      const __m128i pred = _mm_or_si128(A, B);    // pred = (L > T)? L : T
-      L = _mm_add_epi8(src, pred);
-      out[i + j] = _mm_cvtsi128_si32(L);
-      // Shift the pre-computed value for the next iteration.
-      T = _mm_srli_si128(T, 4);
-      TL = _mm_srli_si128(TL, 4);
-      src = _mm_srli_si128(src, 4);
-      pa = _mm_srli_si128(pa, 4);
+    {
+      // We can unpack with any value on the upper 32 bits, provided it's the
+      // same on both operands (so that their sum of abs diff is zero). Here we
+      // use T.
+      const __m128i T_lo = _mm_unpacklo_epi32(T, T);
+      const __m128i TL_lo = _mm_unpacklo_epi32(TL, T);
+      const __m128i T_hi = _mm_unpackhi_epi32(T, T);
+      const __m128i TL_hi = _mm_unpackhi_epi32(TL, T);
+      const __m128i s_lo = _mm_sad_epu8(T_lo, TL_lo);
+      const __m128i s_hi = _mm_sad_epu8(T_hi, TL_hi);
+      pa = _mm_packs_epi32(s_lo, s_hi);  // pa = sum |T-TL|
     }
+    DO_PRED11(0);
+    DO_PRED11_SHIFT;
+    DO_PRED11(1);
+    DO_PRED11_SHIFT;
+    DO_PRED11(2);
+    DO_PRED11_SHIFT;
+    DO_PRED11(3);
   }
   if (i != num_pixels) {
     VP8LPredictorsAdd_C[11](in + i, upper + i, num_pixels - i, out + i);
   }
 }
+#undef DO_PRED11
+#undef DO_PRED11_SHIFT
 
 // Predictor12: ClampedAddSubtractFull.
-#define DO_PRED12(DIFF, LANE, OUT)                          \
-do {                                                        \
-  const __m128i all = _mm_add_epi16(L, (DIFF));             \
-  const __m128i alls = _mm_packus_epi16(all, all);          \
-  const __m128i res = _mm_add_epi8(src, alls);              \
-  out[i + (OUT)] = _mm_cvtsi128_si32(res);                  \
-  L = _mm_unpacklo_epi8(res, zero);                         \
+#define DO_PRED12(DIFF, LANE, OUT) do {            \
+  const __m128i all = _mm_add_epi16(L, (DIFF));    \
+  const __m128i alls = _mm_packus_epi16(all, all); \
+  const __m128i res = _mm_add_epi8(src, alls);     \
+  out[i + (OUT)] = _mm_cvtsi128_si32(res);         \
+  L = _mm_unpacklo_epi8(res, zero);                \
+} while (0)
+
+#define DO_PRED12_SHIFT(DIFF, LANE) do {                    \
   /* Shift the pre-computed value for the next iteration.*/ \
-  if (LANE == 0) (DIFF) = _mm_srli_si128((DIFF), 8);        \
+  if ((LANE) == 0) (DIFF) = _mm_srli_si128((DIFF), 8);      \
   src = _mm_srli_si128(src, 4);                             \
 } while (0)
 
@@ -377,8 +407,11 @@ static void PredictorAdd12_SSE2(const uint32_t* in, const uint32_t* upper,
     __m128i diff_lo = _mm_sub_epi16(T_lo, TL_lo);
     __m128i diff_hi = _mm_sub_epi16(T_hi, TL_hi);
     DO_PRED12(diff_lo, 0, 0);
+    DO_PRED12_SHIFT(diff_lo, 0);
     DO_PRED12(diff_lo, 1, 1);
+    DO_PRED12_SHIFT(diff_lo, 1);
     DO_PRED12(diff_hi, 0, 2);
+    DO_PRED12_SHIFT(diff_hi, 0);
     DO_PRED12(diff_hi, 1, 3);
   }
   if (i != num_pixels) {
@@ -386,6 +419,7 @@ static void PredictorAdd12_SSE2(const uint32_t* in, const uint32_t* upper,
   }
 }
 #undef DO_PRED12
+#undef DO_PRED12_SHIFT
 
 // Due to averages with integers, values cannot be accumulated in parallel for
 // predictors 13.
@@ -394,8 +428,8 @@ GENERATE_PREDICTOR_ADD(Predictor13_SSE2, PredictorAdd13_SSE2)
 //------------------------------------------------------------------------------
 // Subtract-Green Transform
 
-static void AddGreenToBlueAndRed(const uint32_t* const src, int num_pixels,
-                                 uint32_t* dst) {
+static void AddGreenToBlueAndRed_SSE2(const uint32_t* const src, int num_pixels,
+                                      uint32_t* dst) {
   int i;
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb
@@ -414,9 +448,9 @@ static void AddGreenToBlueAndRed(const uint32_t* const src, int num_pixels,
 //------------------------------------------------------------------------------
 // Color Transform
 
-static void TransformColorInverse(const VP8LMultipliers* const m,
-                                  const uint32_t* const src, int num_pixels,
-                                  uint32_t* dst) {
+static void TransformColorInverse_SSE2(const VP8LMultipliers* const m,
+                                       const uint32_t* const src,
+                                       int num_pixels, uint32_t* dst) {
 // sign-extended multiplying constants, pre-shifted by 5.
 #define CST(X)  (((int16_t)(m->X << 8)) >> 5)   // sign-extend
   const __m128i mults_rb = _mm_set_epi16(
@@ -454,8 +488,8 @@ static void TransformColorInverse(const VP8LMultipliers* const m,
 //------------------------------------------------------------------------------
 // Color-space conversion functions
 
-static void ConvertBGRAToRGB(const uint32_t* src, int num_pixels,
-                             uint8_t* dst) {
+static void ConvertBGRAToRGB_SSE2(const uint32_t* src, int num_pixels,
+                                  uint8_t* dst) {
   const __m128i* in = (const __m128i*)src;
   __m128i* out = (__m128i*)dst;
 
@@ -490,27 +524,26 @@ static void ConvertBGRAToRGB(const uint32_t* src, int num_pixels,
   }
 }
 
-static void ConvertBGRAToRGBA(const uint32_t* src,
-                              int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGBA_SSE2(const uint32_t* src,
+                                   int num_pixels, uint8_t* dst) {
+  const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ffu);
   const __m128i* in = (const __m128i*)src;
   __m128i* out = (__m128i*)dst;
   while (num_pixels >= 8) {
-    const __m128i bgra0 = _mm_loadu_si128(in++);     // bgra0|bgra1|bgra2|bgra3
-    const __m128i bgra4 = _mm_loadu_si128(in++);     // bgra4|bgra5|bgra6|bgra7
-    const __m128i v0l = _mm_unpacklo_epi8(bgra0, bgra4);  // b0b4g0g4r0r4a0a4...
-    const __m128i v0h = _mm_unpackhi_epi8(bgra0, bgra4);  // b2b6g2g6r2r6a2a6...
-    const __m128i v1l = _mm_unpacklo_epi8(v0l, v0h);   // b0b2b4b6g0g2g4g6...
-    const __m128i v1h = _mm_unpackhi_epi8(v0l, v0h);   // b1b3b5b7g1g3g5g7...
-    const __m128i v2l = _mm_unpacklo_epi8(v1l, v1h);   // b0...b7 | g0...g7
-    const __m128i v2h = _mm_unpackhi_epi8(v1l, v1h);   // r0...r7 | a0...a7
-    const __m128i ga0 = _mm_unpackhi_epi64(v2l, v2h);  // g0...g7 | a0...a7
-    const __m128i rb0 = _mm_unpacklo_epi64(v2h, v2l);  // r0...r7 | b0...b7
-    const __m128i rg0 = _mm_unpacklo_epi8(rb0, ga0);   // r0g0r1g1 ... r6g6r7g7
-    const __m128i ba0 = _mm_unpackhi_epi8(rb0, ga0);   // b0a0b1a1 ... b6a6b7a7
-    const __m128i rgba0 = _mm_unpacklo_epi16(rg0, ba0);  // rgba0|rgba1...
-    const __m128i rgba4 = _mm_unpackhi_epi16(rg0, ba0);  // rgba4|rgba5...
-    _mm_storeu_si128(out++, rgba0);
-    _mm_storeu_si128(out++, rgba4);
+    const __m128i A1 = _mm_loadu_si128(in++);
+    const __m128i A2 = _mm_loadu_si128(in++);
+    const __m128i B1 = _mm_and_si128(A1, red_blue_mask);     // R 0 B 0
+    const __m128i B2 = _mm_and_si128(A2, red_blue_mask);     // R 0 B 0
+    const __m128i C1 = _mm_andnot_si128(red_blue_mask, A1);  // 0 G 0 A
+    const __m128i C2 = _mm_andnot_si128(red_blue_mask, A2);  // 0 G 0 A
+    const __m128i D1 = _mm_shufflelo_epi16(B1, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128i D2 = _mm_shufflelo_epi16(B2, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128i E1 = _mm_shufflehi_epi16(D1, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128i E2 = _mm_shufflehi_epi16(D2, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128i F1 = _mm_or_si128(E1, C1);
+    const __m128i F2 = _mm_or_si128(E2, C2);
+    _mm_storeu_si128(out++, F1);
+    _mm_storeu_si128(out++, F2);
     num_pixels -= 8;
   }
   // left-overs
@@ -519,8 +552,8 @@ static void ConvertBGRAToRGBA(const uint32_t* src,
   }
 }
 
-static void ConvertBGRAToRGBA4444(const uint32_t* src,
-                                  int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGBA4444_SSE2(const uint32_t* src,
+                                       int num_pixels, uint8_t* dst) {
   const __m128i mask_0x0f = _mm_set1_epi8(0x0f);
   const __m128i mask_0xf0 = _mm_set1_epi8(0xf0);
   const __m128i* in = (const __m128i*)src;
@@ -541,7 +574,7 @@ static void ConvertBGRAToRGBA4444(const uint32_t* src,
     const __m128i ga2 = _mm_and_si128(ga1, mask_0x0f);  // g0-|g1-|...|a6-|a7-
     const __m128i rgba0 = _mm_or_si128(ga2, rb1);       // rg0..rg7 | ba0..ba7
     const __m128i rgba1 = _mm_srli_si128(rgba0, 8);     // ba0..ba7 | 0
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     const __m128i rgba = _mm_unpacklo_epi8(rgba1, rgba0);  // barg0...barg7
 #else
     const __m128i rgba = _mm_unpacklo_epi8(rgba0, rgba1);  // rgba0...rgba7
@@ -555,8 +588,8 @@ static void ConvertBGRAToRGBA4444(const uint32_t* src,
   }
 }
 
-static void ConvertBGRAToRGB565(const uint32_t* src,
-                                int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToRGB565_SSE2(const uint32_t* src,
+                                     int num_pixels, uint8_t* dst) {
   const __m128i mask_0xe0 = _mm_set1_epi8(0xe0);
   const __m128i mask_0xf8 = _mm_set1_epi8(0xf8);
   const __m128i mask_0x07 = _mm_set1_epi8(0x07);
@@ -582,7 +615,7 @@ static void ConvertBGRAToRGB565(const uint32_t* src,
     const __m128i rg1 = _mm_or_si128(rb1, g_lo2);           // gr0...gr7|xx
     const __m128i b1 = _mm_srli_epi16(b0, 3);
     const __m128i gb1 = _mm_or_si128(b1, g_hi2);            // bg0...bg7|xx
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     const __m128i rgba = _mm_unpacklo_epi8(gb1, rg1);     // rggb0...rggb7
 #else
     const __m128i rgba = _mm_unpacklo_epi8(rg1, gb1);     // bgrb0...bgrb7
@@ -596,8 +629,8 @@ static void ConvertBGRAToRGB565(const uint32_t* src,
   }
 }
 
-static void ConvertBGRAToBGR(const uint32_t* src,
-                             int num_pixels, uint8_t* dst) {
+static void ConvertBGRAToBGR_SSE2(const uint32_t* src,
+                                  int num_pixels, uint8_t* dst) {
   const __m128i mask_l = _mm_set_epi32(0, 0x00ffffff, 0, 0x00ffffff);
   const __m128i mask_h = _mm_set_epi32(0x00ffffff, 0, 0x00ffffff, 0);
   const __m128i* in = (const __m128i*)src;
@@ -660,14 +693,14 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitSSE2(void) {
   VP8LPredictorsAdd[12] = PredictorAdd12_SSE2;
   VP8LPredictorsAdd[13] = PredictorAdd13_SSE2;
 
-  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
-  VP8LTransformColorInverse = TransformColorInverse;
+  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed_SSE2;
+  VP8LTransformColorInverse = TransformColorInverse_SSE2;
 
-  VP8LConvertBGRAToRGB = ConvertBGRAToRGB;
-  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
-  VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444;
-  VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565;
-  VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
+  VP8LConvertBGRAToRGB = ConvertBGRAToRGB_SSE2;
+  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA_SSE2;
+  VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444_SSE2;
+  VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565_SSE2;
+  VP8LConvertBGRAToBGR = ConvertBGRAToBGR_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/msa_macro.h b/src/3rdparty/libwebp/src/dsp/msa_macro.h
index d0e5f45..dfacda6 100644
--- a/src/3rdparty/libwebp/src/dsp/msa_macro.h
+++ b/src/3rdparty/libwebp/src/dsp/msa_macro.h
@@ -22,6 +22,7 @@
 #endif
 
 #ifdef CLANG_BUILD
+  #define ALPHAVAL  (-1)
   #define ADDVI_H(a, b)  __msa_addvi_h((v8i16)a, b)
   #define ADDVI_W(a, b)  __msa_addvi_w((v4i32)a, b)
   #define SRAI_B(a, b)  __msa_srai_b((v16i8)a, b)
@@ -32,6 +33,7 @@
   #define ANDI_B(a, b)  __msa_andi_b((v16u8)a, b)
   #define ORI_B(a, b)   __msa_ori_b((v16u8)a, b)
 #else
+  #define ALPHAVAL  (0xff)
   #define ADDVI_H(a, b)  (a + b)
   #define ADDVI_W(a, b)  (a + b)
   #define SRAI_B(a, b)  (a >> b)
diff --git a/src/3rdparty/libwebp/src/dsp/neon.h b/src/3rdparty/libwebp/src/dsp/neon.h
index 3b548a6..aa1dea1 100644
--- a/src/3rdparty/libwebp/src/dsp/neon.h
+++ b/src/3rdparty/libwebp/src/dsp/neon.h
@@ -14,11 +14,12 @@
 
 #include <arm_neon.h>
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 // Right now, some intrinsics functions seem slower, so we disable them
-// everywhere except aarch64 where the inline assembly is incompatible.
-#if defined(__aarch64__)
+// everywhere except newer clang/gcc or aarch64 where the inline assembly is
+// incompatible.
+#if LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,9) || defined(__aarch64__)
 #define WEBP_USE_INTRINSICS   // use intrinsics when possible
 #endif
 
@@ -43,11 +44,11 @@
 // if using intrinsics, this flag avoids some functions that make gcc-4.6.3
 // crash ("internal compiler error: in immed_double_const, at emit-rtl.").
 // (probably similar to gcc.gnu.org/bugzilla/show_bug.cgi?id=48183)
-#if !(LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
+#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
 #define WORK_AROUND_GCC
 #endif
 
-static WEBP_INLINE int32x4x4_t Transpose4x4(const int32x4x4_t rows) {
+static WEBP_INLINE int32x4x4_t Transpose4x4_NEON(const int32x4x4_t rows) {
   uint64x2x2_t row01, row23;
 
   row01.val[0] = vreinterpretq_u64_s32(rows.val[0]);
diff --git a/src/3rdparty/libwebp/src/dsp/rescaler.c b/src/3rdparty/libwebp/src/dsp/rescaler.c
index 0f54502..4b6b783 100644
--- a/src/3rdparty/libwebp/src/dsp/rescaler.c
+++ b/src/3rdparty/libwebp/src/dsp/rescaler.c
@@ -13,8 +13,8 @@
 
 #include <assert.h>
 
-#include "./dsp.h"
-#include "../utils/rescaler_utils.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/rescaler_utils.h"
 
 //------------------------------------------------------------------------------
 // Implementations of critical functions ImportRow / ExportRow
@@ -25,7 +25,8 @@
 //------------------------------------------------------------------------------
 // Row import
 
-void WebPRescalerImportRowExpandC(WebPRescaler* const wrk, const uint8_t* src) {
+void WebPRescalerImportRowExpand_C(WebPRescaler* const wrk,
+                                   const uint8_t* src) {
   const int x_stride = wrk->num_channels;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   int channel;
@@ -56,7 +57,8 @@ void WebPRescalerImportRowExpandC(WebPRescaler* const wrk, const uint8_t* src) {
   }
 }
 
-void WebPRescalerImportRowShrinkC(WebPRescaler* const wrk, const uint8_t* src) {
+void WebPRescalerImportRowShrink_C(WebPRescaler* const wrk,
+                                   const uint8_t* src) {
   const int x_stride = wrk->num_channels;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   int channel;
@@ -92,7 +94,7 @@ void WebPRescalerImportRowShrinkC(WebPRescaler* const wrk, const uint8_t* src) {
 //------------------------------------------------------------------------------
 // Row export
 
-void WebPRescalerExportRowExpandC(WebPRescaler* const wrk) {
+void WebPRescalerExportRowExpand_C(WebPRescaler* const wrk) {
   int x_out;
   uint8_t* const dst = wrk->dst;
   rescaler_t* const irow = wrk->irow;
@@ -123,7 +125,7 @@ void WebPRescalerExportRowExpandC(WebPRescaler* const wrk) {
   }
 }
 
-void WebPRescalerExportRowShrinkC(WebPRescaler* const wrk) {
+void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
   int x_out;
   uint8_t* const dst = wrk->dst;
   rescaler_t* const irow = wrk->irow;
@@ -207,11 +209,14 @@ static volatile VP8CPUInfo rescaler_last_cpuinfo_used =
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void) {
   if (rescaler_last_cpuinfo_used == VP8GetCPUInfo) return;
+#if !defined(WEBP_REDUCE_SIZE)
+#if !WEBP_NEON_OMIT_C_CODE
+  WebPRescalerExportRowExpand = WebPRescalerExportRowExpand_C;
+  WebPRescalerExportRowShrink = WebPRescalerExportRowShrink_C;
+#endif
 
-  WebPRescalerImportRowExpand = WebPRescalerImportRowExpandC;
-  WebPRescalerImportRowShrink = WebPRescalerImportRowShrinkC;
-  WebPRescalerExportRowExpand = WebPRescalerExportRowExpandC;
-  WebPRescalerExportRowShrink = WebPRescalerExportRowShrinkC;
+  WebPRescalerImportRowExpand = WebPRescalerImportRowExpand_C;
+  WebPRescalerImportRowShrink = WebPRescalerImportRowShrink_C;
 
   if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
@@ -219,11 +224,6 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void) {
       WebPRescalerDspInitSSE2();
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      WebPRescalerDspInitNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
       WebPRescalerDspInitMIPS32();
@@ -240,5 +240,18 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void) {
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    WebPRescalerDspInitNEON();
+  }
+#endif
+
+  assert(WebPRescalerExportRowExpand != NULL);
+  assert(WebPRescalerExportRowShrink != NULL);
+  assert(WebPRescalerImportRowExpand != NULL);
+  assert(WebPRescalerImportRowShrink != NULL);
+#endif   // WEBP_REDUCE_SIZE
   rescaler_last_cpuinfo_used = VP8GetCPUInfo;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/rescaler_mips32.c b/src/3rdparty/libwebp/src/dsp/rescaler_mips32.c
index e09ad5d..542f7e5 100644
--- a/src/3rdparty/libwebp/src/dsp/rescaler_mips32.c
+++ b/src/3rdparty/libwebp/src/dsp/rescaler_mips32.c
@@ -11,17 +11,18 @@
 //
 // Author(s): Djordje Pesut (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
-#if defined(WEBP_USE_MIPS32)
+#if defined(WEBP_USE_MIPS32) && !defined(WEBP_REDUCE_SIZE)
 
 #include <assert.h>
-#include "../utils/rescaler_utils.h"
+#include "src/utils/rescaler_utils.h"
 
 //------------------------------------------------------------------------------
 // Row import
 
-static void ImportRowShrink(WebPRescaler* const wrk, const uint8_t* src) {
+static void ImportRowShrink_MIPS32(WebPRescaler* const wrk,
+                                   const uint8_t* src) {
   const int x_stride = wrk->num_channels;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   const int fx_scale = wrk->fx_scale;
@@ -80,7 +81,8 @@ static void ImportRowShrink(WebPRescaler* const wrk, const uint8_t* src) {
   }
 }
 
-static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
+static void ImportRowExpand_MIPS32(WebPRescaler* const wrk,
+                                   const uint8_t* src) {
   const int x_stride = wrk->num_channels;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   const int x_add = wrk->x_add;
@@ -144,7 +146,7 @@ static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
 //------------------------------------------------------------------------------
 // Row export
 
-static void ExportRowExpand(WebPRescaler* const wrk) {
+static void ExportRowExpand_MIPS32(WebPRescaler* const wrk) {
   uint8_t* dst = wrk->dst;
   rescaler_t* irow = wrk->irow;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
@@ -207,7 +209,7 @@ static void ExportRowExpand(WebPRescaler* const wrk) {
   }
 }
 
-static void ExportRowShrink(WebPRescaler* const wrk) {
+static void ExportRowShrink_MIPS32(WebPRescaler* const wrk) {
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   uint8_t* dst = wrk->dst;
   rescaler_t* irow = wrk->irow;
@@ -278,10 +280,10 @@ static void ExportRowShrink(WebPRescaler* const wrk) {
 extern void WebPRescalerDspInitMIPS32(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPS32(void) {
-  WebPRescalerImportRowExpand = ImportRowExpand;
-  WebPRescalerImportRowShrink = ImportRowShrink;
-  WebPRescalerExportRowExpand = ExportRowExpand;
-  WebPRescalerExportRowShrink = ExportRowShrink;
+  WebPRescalerImportRowExpand = ImportRowExpand_MIPS32;
+  WebPRescalerImportRowShrink = ImportRowShrink_MIPS32;
+  WebPRescalerExportRowExpand = ExportRowExpand_MIPS32;
+  WebPRescalerExportRowShrink = ExportRowShrink_MIPS32;
 }
 
 #else  // !WEBP_USE_MIPS32
diff --git a/src/3rdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c
index 2308d64..b78aac1 100644
--- a/src/3rdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c
@@ -11,12 +11,12 @@
 //
 // Author(s): Djordje Pesut (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
-#if defined(WEBP_USE_MIPS_DSP_R2)
+#if defined(WEBP_USE_MIPS_DSP_R2) && !defined(WEBP_REDUCE_SIZE)
 
 #include <assert.h>
-#include "../utils/rescaler_utils.h"
+#include "src/utils/rescaler_utils.h"
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
@@ -24,7 +24,7 @@
 //------------------------------------------------------------------------------
 // Row export
 
-static void ExportRowShrink(WebPRescaler* const wrk) {
+static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
   int i;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   uint8_t* dst = wrk->dst;
@@ -162,7 +162,7 @@ static void ExportRowShrink(WebPRescaler* const wrk) {
   }
 }
 
-static void ExportRowExpand(WebPRescaler* const wrk) {
+static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
   int i;
   uint8_t* dst = wrk->dst;
   rescaler_t* irow = wrk->irow;
@@ -303,8 +303,8 @@ static void ExportRowExpand(WebPRescaler* const wrk) {
 extern void WebPRescalerDspInitMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPSdspR2(void) {
-  WebPRescalerExportRowExpand = ExportRowExpand;
-  WebPRescalerExportRowShrink = ExportRowShrink;
+  WebPRescalerExportRowExpand = ExportRowExpand_MIPSdspR2;
+  WebPRescalerExportRowShrink = ExportRowShrink_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/rescaler_msa.c b/src/3rdparty/libwebp/src/dsp/rescaler_msa.c
index 2c10e55..f3bc99f 100644
--- a/src/3rdparty/libwebp/src/dsp/rescaler_msa.c
+++ b/src/3rdparty/libwebp/src/dsp/rescaler_msa.c
@@ -11,14 +11,14 @@
 //
 // Author: Prashant Patil (prashant.patil@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
-#if defined(WEBP_USE_MSA)
+#if defined(WEBP_USE_MSA) && !defined(WEBP_REDUCE_SIZE)
 
 #include <assert.h>
 
-#include "../utils/rescaler_utils.h"
-#include "./msa_macro.h"
+#include "src/utils/rescaler_utils.h"
+#include "src/dsp/msa_macro.h"
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
@@ -246,7 +246,7 @@ static WEBP_INLINE void ExportRowExpand_1(const uint32_t* frow, uint32_t* irow,
   }
 }
 
-static void RescalerExportRowExpand(WebPRescaler* const wrk) {
+static void RescalerExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
   uint8_t* dst = wrk->dst;
   rescaler_t* irow = wrk->irow;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
@@ -411,7 +411,7 @@ static WEBP_INLINE void ExportRowShrink_1(uint32_t* irow, uint8_t* dst,
   }
 }
 
-static void RescalerExportRowShrink(WebPRescaler* const wrk) {
+static void RescalerExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
   uint8_t* dst = wrk->dst;
   rescaler_t* irow = wrk->irow;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
@@ -433,8 +433,8 @@ static void RescalerExportRowShrink(WebPRescaler* const wrk) {
 extern void WebPRescalerDspInitMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMSA(void) {
-  WebPRescalerExportRowExpand = RescalerExportRowExpand;
-  WebPRescalerExportRowShrink = RescalerExportRowShrink;
+  WebPRescalerExportRowExpand = RescalerExportRowExpand_MIPSdspR2;
+  WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
 }
 
 #else     // !WEBP_USE_MSA
diff --git a/src/3rdparty/libwebp/src/dsp/rescaler_neon.c b/src/3rdparty/libwebp/src/dsp/rescaler_neon.c
index b2dd8f3..3eff9fb 100644
--- a/src/3rdparty/libwebp/src/dsp/rescaler_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/rescaler_neon.c
@@ -11,14 +11,14 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_USE_NEON) && !defined(WEBP_REDUCE_SIZE)
 
 #include <arm_neon.h>
 #include <assert.h>
-#include "./neon.h"
-#include "../utils/rescaler_utils.h"
+#include "src/dsp/neon.h"
+#include "src/utils/rescaler_utils.h"
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX_C(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
@@ -41,9 +41,9 @@
 #error "MULT_FIX/WEBP_RESCALER_RFIX need some more work"
 #endif
 
-static uint32x4_t Interpolate(const rescaler_t* const frow,
-                              const rescaler_t* const irow,
-                              uint32_t A, uint32_t B) {
+static uint32x4_t Interpolate_NEON(const rescaler_t* const frow,
+                                   const rescaler_t* const irow,
+                                   uint32_t A, uint32_t B) {
   LOAD_32x4(frow, A0);
   LOAD_32x4(irow, B0);
   const uint64x2_t C0 = vmull_n_u32(vget_low_u32(A0), A);
@@ -56,7 +56,7 @@ static uint32x4_t Interpolate(const rescaler_t* const frow,
   return E;
 }
 
-static void RescalerExportRowExpand(WebPRescaler* const wrk) {
+static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
   int x_out;
   uint8_t* const dst = wrk->dst;
   rescaler_t* const irow = wrk->irow;
@@ -91,9 +91,9 @@ static void RescalerExportRowExpand(WebPRescaler* const wrk) {
     const uint32_t A = (uint32_t)(WEBP_RESCALER_ONE - B);
     for (x_out = 0; x_out < max_span; x_out += 8) {
       const uint32x4_t C0 =
-          Interpolate(frow + x_out + 0, irow + x_out + 0, A, B);
+          Interpolate_NEON(frow + x_out + 0, irow + x_out + 0, A, B);
       const uint32x4_t C1 =
-          Interpolate(frow + x_out + 4, irow + x_out + 4, A, B);
+          Interpolate_NEON(frow + x_out + 4, irow + x_out + 4, A, B);
       const uint32x4_t D0 = MULT_FIX(C0, fy_scale_half);
       const uint32x4_t D1 = MULT_FIX(C1, fy_scale_half);
       const uint16x4_t E0 = vmovn_u32(D0);
@@ -112,7 +112,7 @@ static void RescalerExportRowExpand(WebPRescaler* const wrk) {
   }
 }
 
-static void RescalerExportRowShrink(WebPRescaler* const wrk) {
+static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
   int x_out;
   uint8_t* const dst = wrk->dst;
   rescaler_t* const irow = wrk->irow;
@@ -175,8 +175,8 @@ static void RescalerExportRowShrink(WebPRescaler* const wrk) {
 extern void WebPRescalerDspInitNEON(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitNEON(void) {
-  WebPRescalerExportRowExpand = RescalerExportRowExpand;
-  WebPRescalerExportRowShrink = RescalerExportRowShrink;
+  WebPRescalerExportRowExpand = RescalerExportRowExpand_NEON;
+  WebPRescalerExportRowShrink = RescalerExportRowShrink_NEON;
 }
 
 #else     // !WEBP_USE_NEON
diff --git a/src/3rdparty/libwebp/src/dsp/rescaler_sse2.c b/src/3rdparty/libwebp/src/dsp/rescaler_sse2.c
index 8271c22..f93b204 100644
--- a/src/3rdparty/libwebp/src/dsp/rescaler_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/rescaler_sse2.c
@@ -11,14 +11,14 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_USE_SSE2) && !defined(WEBP_REDUCE_SIZE)
 #include <emmintrin.h>
 
 #include <assert.h>
-#include "../utils/rescaler_utils.h"
-#include "../utils/utils.h"
+#include "src/utils/rescaler_utils.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Implementations of critical functions ImportRow / ExportRow
@@ -27,7 +27,7 @@
 #define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
 
 // input: 8 bytes ABCDEFGH -> output: A0E0B0F0C0G0D0H0
-static void LoadTwoPixels(const uint8_t* const src, __m128i* out) {
+static void LoadTwoPixels_SSE2(const uint8_t* const src, __m128i* out) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i A = _mm_loadl_epi64((const __m128i*)(src));  // ABCDEFGH
   const __m128i B = _mm_unpacklo_epi8(A, zero);              // A0B0C0D0E0F0G0H0
@@ -36,14 +36,14 @@ static void LoadTwoPixels(const uint8_t* const src, __m128i* out) {
 }
 
 // input: 8 bytes ABCDEFGH -> output: A0B0C0D0E0F0G0H0
-static void LoadHeightPixels(const uint8_t* const src, __m128i* out) {
+static void LoadHeightPixels_SSE2(const uint8_t* const src, __m128i* out) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i A = _mm_loadl_epi64((const __m128i*)(src));  // ABCDEFGH
   *out = _mm_unpacklo_epi8(A, zero);
 }
 
-static void RescalerImportRowExpandSSE2(WebPRescaler* const wrk,
-                                        const uint8_t* src) {
+static void RescalerImportRowExpand_SSE2(WebPRescaler* const wrk,
+                                         const uint8_t* src) {
   rescaler_t* frow = wrk->frow;
   const rescaler_t* const frow_end = frow + wrk->dst_width * wrk->num_channels;
   const int x_add = wrk->x_add;
@@ -54,10 +54,10 @@ static void RescalerImportRowExpandSSE2(WebPRescaler* const wrk,
   assert(wrk->x_expand);
   if (wrk->num_channels == 4) {
     if (wrk->src_width < 2) {
-      WebPRescalerImportRowExpandC(wrk, src);
+      WebPRescalerImportRowExpand_C(wrk, src);
       return;
     }
-    LoadTwoPixels(src, &cur_pixels);
+    LoadTwoPixels_SSE2(src, &cur_pixels);
     src += 4;
     while (1) {
       const __m128i mult = _mm_set1_epi32(((x_add - accum) << 16) | accum);
@@ -67,7 +67,7 @@ static void RescalerImportRowExpandSSE2(WebPRescaler* const wrk,
       if (frow >= frow_end) break;
       accum -= wrk->x_sub;
       if (accum < 0) {
-        LoadTwoPixels(src, &cur_pixels);
+        LoadTwoPixels_SSE2(src, &cur_pixels);
         src += 4;
         accum += x_add;
       }
@@ -76,10 +76,10 @@ static void RescalerImportRowExpandSSE2(WebPRescaler* const wrk,
     int left;
     const uint8_t* const src_limit = src + wrk->src_width - 8;
     if (wrk->src_width < 8) {
-      WebPRescalerImportRowExpandC(wrk, src);
+      WebPRescalerImportRowExpand_C(wrk, src);
       return;
     }
-    LoadHeightPixels(src, &cur_pixels);
+    LoadHeightPixels_SSE2(src, &cur_pixels);
     src += 7;
     left = 7;
     while (1) {
@@ -94,7 +94,7 @@ static void RescalerImportRowExpandSSE2(WebPRescaler* const wrk,
         if (--left) {
           cur_pixels = _mm_srli_si128(cur_pixels, 2);
         } else if (src <= src_limit) {
-          LoadHeightPixels(src, &cur_pixels);
+          LoadHeightPixels_SSE2(src, &cur_pixels);
           src += 7;
           left = 7;
         } else {   // tail
@@ -110,8 +110,8 @@ static void RescalerImportRowExpandSSE2(WebPRescaler* const wrk,
   assert(accum == 0);
 }
 
-static void RescalerImportRowShrinkSSE2(WebPRescaler* const wrk,
-                                        const uint8_t* src) {
+static void RescalerImportRowShrink_SSE2(WebPRescaler* const wrk,
+                                         const uint8_t* src) {
   const int x_sub = wrk->x_sub;
   int accum = 0;
   const __m128i zero = _mm_setzero_si128();
@@ -123,7 +123,7 @@ static void RescalerImportRowShrinkSSE2(WebPRescaler* const wrk,
   const rescaler_t* const frow_end = wrk->frow + 4 * wrk->dst_width;
 
   if (wrk->num_channels != 4 || wrk->x_add > (x_sub << 7)) {
-    WebPRescalerImportRowShrinkC(wrk, src);
+    WebPRescalerImportRowShrink_C(wrk, src);
     return;
   }
   assert(!WebPRescalerInputDone(wrk));
@@ -169,12 +169,12 @@ static void RescalerImportRowShrinkSSE2(WebPRescaler* const wrk,
 // Row export
 
 // load *src as epi64, multiply by mult and store result in [out0 ... out3]
-static WEBP_INLINE void LoadDispatchAndMult(const rescaler_t* const src,
-                                            const __m128i* const mult,
-                                            __m128i* const out0,
-                                            __m128i* const out1,
-                                            __m128i* const out2,
-                                            __m128i* const out3) {
+static WEBP_INLINE void LoadDispatchAndMult_SSE2(const rescaler_t* const src,
+                                                 const __m128i* const mult,
+                                                 __m128i* const out0,
+                                                 __m128i* const out1,
+                                                 __m128i* const out2,
+                                                 __m128i* const out3) {
   const __m128i A0 = _mm_loadu_si128((const __m128i*)(src + 0));
   const __m128i A1 = _mm_loadu_si128((const __m128i*)(src + 4));
   const __m128i A2 = _mm_srli_epi64(A0, 32);
@@ -192,12 +192,12 @@ static WEBP_INLINE void LoadDispatchAndMult(const rescaler_t* const src,
   }
 }
 
-static WEBP_INLINE void ProcessRow(const __m128i* const A0,
-                                   const __m128i* const A1,
-                                   const __m128i* const A2,
-                                   const __m128i* const A3,
-                                   const __m128i* const mult,
-                                   uint8_t* const dst) {
+static WEBP_INLINE void ProcessRow_SSE2(const __m128i* const A0,
+                                        const __m128i* const A1,
+                                        const __m128i* const A2,
+                                        const __m128i* const A3,
+                                        const __m128i* const mult,
+                                        uint8_t* const dst) {
   const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
   const __m128i mask = _mm_set_epi32(0xffffffffu, 0, 0xffffffffu, 0);
   const __m128i B0 = _mm_mul_epu32(*A0, *mult);
@@ -210,7 +210,7 @@ static WEBP_INLINE void ProcessRow(const __m128i* const A0,
   const __m128i C3 = _mm_add_epi64(B3, rounder);
   const __m128i D0 = _mm_srli_epi64(C0, WEBP_RESCALER_RFIX);
   const __m128i D1 = _mm_srli_epi64(C1, WEBP_RESCALER_RFIX);
-#if (WEBP_RESCALER_FIX < 32)
+#if (WEBP_RESCALER_RFIX < 32)
   const __m128i D2 =
       _mm_and_si128(_mm_slli_epi64(C2, 32 - WEBP_RESCALER_RFIX), mask);
   const __m128i D3 =
@@ -226,7 +226,7 @@ static WEBP_INLINE void ProcessRow(const __m128i* const A0,
   _mm_storel_epi64((__m128i*)dst, G);
 }
 
-static void RescalerExportRowExpandSSE2(WebPRescaler* const wrk) {
+static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
   int x_out;
   uint8_t* const dst = wrk->dst;
   rescaler_t* const irow = wrk->irow;
@@ -240,8 +240,8 @@ static void RescalerExportRowExpandSSE2(WebPRescaler* const wrk) {
   if (wrk->y_accum == 0) {
     for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
       __m128i A0, A1, A2, A3;
-      LoadDispatchAndMult(frow + x_out, NULL, &A0, &A1, &A2, &A3);
-      ProcessRow(&A0, &A1, &A2, &A3, &mult, dst + x_out);
+      LoadDispatchAndMult_SSE2(frow + x_out, NULL, &A0, &A1, &A2, &A3);
+      ProcessRow_SSE2(&A0, &A1, &A2, &A3, &mult, dst + x_out);
     }
     for (; x_out < x_out_max; ++x_out) {
       const uint32_t J = frow[x_out];
@@ -257,8 +257,8 @@ static void RescalerExportRowExpandSSE2(WebPRescaler* const wrk) {
     const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
     for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
       __m128i A0, A1, A2, A3, B0, B1, B2, B3;
-      LoadDispatchAndMult(frow + x_out, &mA, &A0, &A1, &A2, &A3);
-      LoadDispatchAndMult(irow + x_out, &mB, &B0, &B1, &B2, &B3);
+      LoadDispatchAndMult_SSE2(frow + x_out, &mA, &A0, &A1, &A2, &A3);
+      LoadDispatchAndMult_SSE2(irow + x_out, &mB, &B0, &B1, &B2, &B3);
       {
         const __m128i C0 = _mm_add_epi64(A0, B0);
         const __m128i C1 = _mm_add_epi64(A1, B1);
@@ -272,7 +272,7 @@ static void RescalerExportRowExpandSSE2(WebPRescaler* const wrk) {
         const __m128i E1 = _mm_srli_epi64(D1, WEBP_RESCALER_RFIX);
         const __m128i E2 = _mm_srli_epi64(D2, WEBP_RESCALER_RFIX);
         const __m128i E3 = _mm_srli_epi64(D3, WEBP_RESCALER_RFIX);
-        ProcessRow(&E0, &E1, &E2, &E3, &mult, dst + x_out);
+        ProcessRow_SSE2(&E0, &E1, &E2, &E3, &mult, dst + x_out);
       }
     }
     for (; x_out < x_out_max; ++x_out) {
@@ -286,7 +286,7 @@ static void RescalerExportRowExpandSSE2(WebPRescaler* const wrk) {
   }
 }
 
-static void RescalerExportRowShrinkSSE2(WebPRescaler* const wrk) {
+static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
   int x_out;
   uint8_t* const dst = wrk->dst;
   rescaler_t* const irow = wrk->irow;
@@ -303,8 +303,8 @@ static void RescalerExportRowShrinkSSE2(WebPRescaler* const wrk) {
     const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
     for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
       __m128i A0, A1, A2, A3, B0, B1, B2, B3;
-      LoadDispatchAndMult(irow + x_out, NULL, &A0, &A1, &A2, &A3);
-      LoadDispatchAndMult(frow + x_out, &mult_y, &B0, &B1, &B2, &B3);
+      LoadDispatchAndMult_SSE2(irow + x_out, NULL, &A0, &A1, &A2, &A3);
+      LoadDispatchAndMult_SSE2(frow + x_out, &mult_y, &B0, &B1, &B2, &B3);
       {
         const __m128i C0 = _mm_add_epi64(B0, rounder);
         const __m128i C1 = _mm_add_epi64(B1, rounder);
@@ -324,7 +324,7 @@ static void RescalerExportRowShrinkSSE2(WebPRescaler* const wrk) {
         const __m128i G1 = _mm_or_si128(D1, F3);
         _mm_storeu_si128((__m128i*)(irow + x_out + 0), G0);
         _mm_storeu_si128((__m128i*)(irow + x_out + 4), G1);
-        ProcessRow(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
+        ProcessRow_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
       }
     }
     for (; x_out < x_out_max; ++x_out) {
@@ -340,10 +340,10 @@ static void RescalerExportRowShrinkSSE2(WebPRescaler* const wrk) {
     const __m128i zero = _mm_setzero_si128();
     for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
       __m128i A0, A1, A2, A3;
-      LoadDispatchAndMult(irow + x_out, NULL, &A0, &A1, &A2, &A3);
+      LoadDispatchAndMult_SSE2(irow + x_out, NULL, &A0, &A1, &A2, &A3);
       _mm_storeu_si128((__m128i*)(irow + x_out + 0), zero);
       _mm_storeu_si128((__m128i*)(irow + x_out + 4), zero);
-      ProcessRow(&A0, &A1, &A2, &A3, &mult, dst + x_out);
+      ProcessRow_SSE2(&A0, &A1, &A2, &A3, &mult, dst + x_out);
     }
     for (; x_out < x_out_max; ++x_out) {
       const int v = (int)MULT_FIX(irow[x_out], scale);
@@ -362,10 +362,10 @@ static void RescalerExportRowShrinkSSE2(WebPRescaler* const wrk) {
 extern void WebPRescalerDspInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitSSE2(void) {
-  WebPRescalerImportRowExpand = RescalerImportRowExpandSSE2;
-  WebPRescalerImportRowShrink = RescalerImportRowShrinkSSE2;
-  WebPRescalerExportRowExpand = RescalerExportRowExpandSSE2;
-  WebPRescalerExportRowShrink = RescalerExportRowShrinkSSE2;
+  WebPRescalerImportRowExpand = RescalerImportRowExpand_SSE2;
+  WebPRescalerImportRowShrink = RescalerImportRowShrink_SSE2;
+  WebPRescalerExportRowExpand = RescalerExportRowExpand_SSE2;
+  WebPRescalerExportRowShrink = RescalerExportRowShrink_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/ssim.c b/src/3rdparty/libwebp/src/dsp/ssim.c
new file mode 100644
index 0000000..dc1b518
--- /dev/null
+++ b/src/3rdparty/libwebp/src/dsp/ssim.c
@@ -0,0 +1,166 @@
+// Copyright 2017 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// distortion calculation
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include <assert.h>
+#include <stdlib.h>  // for abs()
+
+#include "src/dsp/dsp.h"
+
+#if !defined(WEBP_REDUCE_SIZE)
+
+//------------------------------------------------------------------------------
+// SSIM / PSNR
+
+// hat-shaped filter. Sum of coefficients is equal to 16.
+static const uint32_t kWeight[2 * VP8_SSIM_KERNEL + 1] = {
+  1, 2, 3, 4, 3, 2, 1
+};
+static const uint32_t kWeightSum = 16 * 16;   // sum{kWeight}^2
+
+static WEBP_INLINE double SSIMCalculation(
+    const VP8DistoStats* const stats, uint32_t N  /*num samples*/) {
+  const uint32_t w2 =  N * N;
+  const uint32_t C1 = 20 * w2;
+  const uint32_t C2 = 60 * w2;
+  const uint32_t C3 = 8 * 8 * w2;   // 'dark' limit ~= 6
+  const uint64_t xmxm = (uint64_t)stats->xm * stats->xm;
+  const uint64_t ymym = (uint64_t)stats->ym * stats->ym;
+  if (xmxm + ymym >= C3) {
+    const int64_t xmym = (int64_t)stats->xm * stats->ym;
+    const int64_t sxy = (int64_t)stats->xym * N - xmym;    // can be negative
+    const uint64_t sxx = (uint64_t)stats->xxm * N - xmxm;
+    const uint64_t syy = (uint64_t)stats->yym * N - ymym;
+    // we descale by 8 to prevent overflow during the fnum/fden multiply.
+    const uint64_t num_S = (2 * (uint64_t)(sxy < 0 ? 0 : sxy) + C2) >> 8;
+    const uint64_t den_S = (sxx + syy + C2) >> 8;
+    const uint64_t fnum = (2 * xmym + C1) * num_S;
+    const uint64_t fden = (xmxm + ymym + C1) * den_S;
+    const double r = (double)fnum / fden;
+    assert(r >= 0. && r <= 1.0);
+    return r;
+  }
+  return 1.;   // area is too dark to contribute meaningfully
+}
+
+double VP8SSIMFromStats(const VP8DistoStats* const stats) {
+  return SSIMCalculation(stats, kWeightSum);
+}
+
+double VP8SSIMFromStatsClipped(const VP8DistoStats* const stats) {
+  return SSIMCalculation(stats, stats->w);
+}
+
+static double SSIMGetClipped_C(const uint8_t* src1, int stride1,
+                               const uint8_t* src2, int stride2,
+                               int xo, int yo, int W, int H) {
+  VP8DistoStats stats = { 0, 0, 0, 0, 0, 0 };
+  const int ymin = (yo - VP8_SSIM_KERNEL < 0) ? 0 : yo - VP8_SSIM_KERNEL;
+  const int ymax = (yo + VP8_SSIM_KERNEL > H - 1) ? H - 1
+                                                  : yo + VP8_SSIM_KERNEL;
+  const int xmin = (xo - VP8_SSIM_KERNEL < 0) ? 0 : xo - VP8_SSIM_KERNEL;
+  const int xmax = (xo + VP8_SSIM_KERNEL > W - 1) ? W - 1
+                                                  : xo + VP8_SSIM_KERNEL;
+  int x, y;
+  src1 += ymin * stride1;
+  src2 += ymin * stride2;
+  for (y = ymin; y <= ymax; ++y, src1 += stride1, src2 += stride2) {
+    for (x = xmin; x <= xmax; ++x) {
+      const uint32_t w = kWeight[VP8_SSIM_KERNEL + x - xo]
+                       * kWeight[VP8_SSIM_KERNEL + y - yo];
+      const uint32_t s1 = src1[x];
+      const uint32_t s2 = src2[x];
+      stats.w   += w;
+      stats.xm  += w * s1;
+      stats.ym  += w * s2;
+      stats.xxm += w * s1 * s1;
+      stats.xym += w * s1 * s2;
+      stats.yym += w * s2 * s2;
+    }
+  }
+  return VP8SSIMFromStatsClipped(&stats);
+}
+
+static double SSIMGet_C(const uint8_t* src1, int stride1,
+                        const uint8_t* src2, int stride2) {
+  VP8DistoStats stats = { 0, 0, 0, 0, 0, 0 };
+  int x, y;
+  for (y = 0; y <= 2 * VP8_SSIM_KERNEL; ++y, src1 += stride1, src2 += stride2) {
+    for (x = 0; x <= 2 * VP8_SSIM_KERNEL; ++x) {
+      const uint32_t w = kWeight[x] * kWeight[y];
+      const uint32_t s1 = src1[x];
+      const uint32_t s2 = src2[x];
+      stats.xm  += w * s1;
+      stats.ym  += w * s2;
+      stats.xxm += w * s1 * s1;
+      stats.xym += w * s1 * s2;
+      stats.yym += w * s2 * s2;
+    }
+  }
+  return VP8SSIMFromStats(&stats);
+}
+
+#endif  // !defined(WEBP_REDUCE_SIZE)
+
+//------------------------------------------------------------------------------
+
+#if !defined(WEBP_DISABLE_STATS)
+static uint32_t AccumulateSSE_C(const uint8_t* src1,
+                                const uint8_t* src2, int len) {
+  int i;
+  uint32_t sse2 = 0;
+  assert(len <= 65535);  // to ensure that accumulation fits within uint32_t
+  for (i = 0; i < len; ++i) {
+    const int32_t diff = src1[i] - src2[i];
+    sse2 += diff * diff;
+  }
+  return sse2;
+}
+#endif
+
+//------------------------------------------------------------------------------
+
+#if !defined(WEBP_REDUCE_SIZE)
+VP8SSIMGetFunc VP8SSIMGet;
+VP8SSIMGetClippedFunc VP8SSIMGetClipped;
+#endif
+#if !defined(WEBP_DISABLE_STATS)
+VP8AccumulateSSEFunc VP8AccumulateSSE;
+#endif
+
+extern void VP8SSIMDspInitSSE2(void);
+
+static volatile VP8CPUInfo ssim_last_cpuinfo_used =
+    (VP8CPUInfo)&ssim_last_cpuinfo_used;
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8SSIMDspInit(void) {
+  if (ssim_last_cpuinfo_used == VP8GetCPUInfo) return;
+
+#if !defined(WEBP_REDUCE_SIZE)
+  VP8SSIMGetClipped = SSIMGetClipped_C;
+  VP8SSIMGet = SSIMGet_C;
+#endif
+
+#if !defined(WEBP_DISABLE_STATS)
+  VP8AccumulateSSE = AccumulateSSE_C;
+#endif
+
+  if (VP8GetCPUInfo != NULL) {
+#if defined(WEBP_USE_SSE2)
+    if (VP8GetCPUInfo(kSSE2)) {
+      VP8SSIMDspInitSSE2();
+    }
+#endif
+  }
+
+  ssim_last_cpuinfo_used = VP8GetCPUInfo;
+}
diff --git a/src/3rdparty/libwebp/src/dsp/ssim_sse2.c b/src/3rdparty/libwebp/src/dsp/ssim_sse2.c
new file mode 100644
index 0000000..1dcb0eb
--- /dev/null
+++ b/src/3rdparty/libwebp/src/dsp/ssim_sse2.c
@@ -0,0 +1,165 @@
+// Copyright 2017 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// SSE2 version of distortion calculation
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "src/dsp/dsp.h"
+
+#if defined(WEBP_USE_SSE2)
+
+#include <assert.h>
+#include <emmintrin.h>
+
+#include "src/dsp/common_sse2.h"
+
+#if !defined(WEBP_DISABLE_STATS)
+
+// Helper function
+static WEBP_INLINE void SubtractAndSquare_SSE2(const __m128i a, const __m128i b,
+                                               __m128i* const sum) {
+  // take abs(a-b) in 8b
+  const __m128i a_b = _mm_subs_epu8(a, b);
+  const __m128i b_a = _mm_subs_epu8(b, a);
+  const __m128i abs_a_b = _mm_or_si128(a_b, b_a);
+  // zero-extend to 16b
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i C0 = _mm_unpacklo_epi8(abs_a_b, zero);
+  const __m128i C1 = _mm_unpackhi_epi8(abs_a_b, zero);
+  // multiply with self
+  const __m128i sum1 = _mm_madd_epi16(C0, C0);
+  const __m128i sum2 = _mm_madd_epi16(C1, C1);
+  *sum = _mm_add_epi32(sum1, sum2);
+}
+
+//------------------------------------------------------------------------------
+// SSIM / PSNR entry point
+
+static uint32_t AccumulateSSE_SSE2(const uint8_t* src1,
+                                   const uint8_t* src2, int len) {
+  int i = 0;
+  uint32_t sse2 = 0;
+  if (len >= 16) {
+    const int limit = len - 32;
+    int32_t tmp[4];
+    __m128i sum1;
+    __m128i sum = _mm_setzero_si128();
+    __m128i a0 = _mm_loadu_si128((const __m128i*)&src1[i]);
+    __m128i b0 = _mm_loadu_si128((const __m128i*)&src2[i]);
+    i += 16;
+    while (i <= limit) {
+      const __m128i a1 = _mm_loadu_si128((const __m128i*)&src1[i]);
+      const __m128i b1 = _mm_loadu_si128((const __m128i*)&src2[i]);
+      __m128i sum2;
+      i += 16;
+      SubtractAndSquare_SSE2(a0, b0, &sum1);
+      sum = _mm_add_epi32(sum, sum1);
+      a0 = _mm_loadu_si128((const __m128i*)&src1[i]);
+      b0 = _mm_loadu_si128((const __m128i*)&src2[i]);
+      i += 16;
+      SubtractAndSquare_SSE2(a1, b1, &sum2);
+      sum = _mm_add_epi32(sum, sum2);
+    }
+    SubtractAndSquare_SSE2(a0, b0, &sum1);
+    sum = _mm_add_epi32(sum, sum1);
+    _mm_storeu_si128((__m128i*)tmp, sum);
+    sse2 += (tmp[3] + tmp[2] + tmp[1] + tmp[0]);
+  }
+
+  for (; i < len; ++i) {
+    const int32_t diff = src1[i] - src2[i];
+    sse2 += diff * diff;
+  }
+  return sse2;
+}
+#endif  // !defined(WEBP_DISABLE_STATS)
+
+#if !defined(WEBP_REDUCE_SIZE)
+
+static uint32_t HorizontalAdd16b_SSE2(const __m128i* const m) {
+  uint16_t tmp[8];
+  const __m128i a = _mm_srli_si128(*m, 8);
+  const __m128i b = _mm_add_epi16(*m, a);
+  _mm_storeu_si128((__m128i*)tmp, b);
+  return (uint32_t)tmp[3] + tmp[2] + tmp[1] + tmp[0];
+}
+
+static uint32_t HorizontalAdd32b_SSE2(const __m128i* const m) {
+  const __m128i a = _mm_srli_si128(*m, 8);
+  const __m128i b = _mm_add_epi32(*m, a);
+  const __m128i c = _mm_add_epi32(b, _mm_srli_si128(b, 4));
+  return (uint32_t)_mm_cvtsi128_si32(c);
+}
+
+static const uint16_t kWeight[] = { 1, 2, 3, 4, 3, 2, 1, 0 };
+
+#define ACCUMULATE_ROW(WEIGHT) do {                         \
+  /* compute row weight (Wx * Wy) */                        \
+  const __m128i Wy = _mm_set1_epi16((WEIGHT));              \
+  const __m128i W = _mm_mullo_epi16(Wx, Wy);                \
+  /* process 8 bytes at a time (7 bytes, actually) */       \
+  const __m128i a0 = _mm_loadl_epi64((const __m128i*)src1); \
+  const __m128i b0 = _mm_loadl_epi64((const __m128i*)src2); \
+  /* convert to 16b and multiply by weight */               \
+  const __m128i a1 = _mm_unpacklo_epi8(a0, zero);           \
+  const __m128i b1 = _mm_unpacklo_epi8(b0, zero);           \
+  const __m128i wa1 = _mm_mullo_epi16(a1, W);               \
+  const __m128i wb1 = _mm_mullo_epi16(b1, W);               \
+  /* accumulate */                                          \
+  xm  = _mm_add_epi16(xm, wa1);                             \
+  ym  = _mm_add_epi16(ym, wb1);                             \
+  xxm = _mm_add_epi32(xxm, _mm_madd_epi16(a1, wa1));        \
+  xym = _mm_add_epi32(xym, _mm_madd_epi16(a1, wb1));        \
+  yym = _mm_add_epi32(yym, _mm_madd_epi16(b1, wb1));        \
+  src1 += stride1;                                          \
+  src2 += stride2;                                          \
+} while (0)
+
+static double SSIMGet_SSE2(const uint8_t* src1, int stride1,
+                           const uint8_t* src2, int stride2) {
+  VP8DistoStats stats;
+  const __m128i zero = _mm_setzero_si128();
+  __m128i xm = zero, ym = zero;                // 16b accums
+  __m128i xxm = zero, yym = zero, xym = zero;  // 32b accum
+  const __m128i Wx = _mm_loadu_si128((const __m128i*)kWeight);
+  assert(2 * VP8_SSIM_KERNEL + 1 == 7);
+  ACCUMULATE_ROW(1);
+  ACCUMULATE_ROW(2);
+  ACCUMULATE_ROW(3);
+  ACCUMULATE_ROW(4);
+  ACCUMULATE_ROW(3);
+  ACCUMULATE_ROW(2);
+  ACCUMULATE_ROW(1);
+  stats.xm  = HorizontalAdd16b_SSE2(&xm);
+  stats.ym  = HorizontalAdd16b_SSE2(&ym);
+  stats.xxm = HorizontalAdd32b_SSE2(&xxm);
+  stats.xym = HorizontalAdd32b_SSE2(&xym);
+  stats.yym = HorizontalAdd32b_SSE2(&yym);
+  return VP8SSIMFromStats(&stats);
+}
+
+#endif  // !defined(WEBP_REDUCE_SIZE)
+
+extern void VP8SSIMDspInitSSE2(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8SSIMDspInitSSE2(void) {
+#if !defined(WEBP_DISABLE_STATS)
+  VP8AccumulateSSE = AccumulateSSE_SSE2;
+#endif
+#if !defined(WEBP_REDUCE_SIZE)
+  VP8SSIMGet = SSIMGet_SSE2;
+#endif
+}
+
+#else  // !WEBP_USE_SSE2
+
+WEBP_DSP_INIT_STUB(VP8SSIMDspInitSSE2)
+
+#endif  // WEBP_USE_SSE2
diff --git a/src/3rdparty/libwebp/src/dsp/upsampling.c b/src/3rdparty/libwebp/src/dsp/upsampling.c
index 265e722..e72626a 100644
--- a/src/3rdparty/libwebp/src/dsp/upsampling.c
+++ b/src/3rdparty/libwebp/src/dsp/upsampling.c
@@ -11,8 +11,8 @@
 //
 // Author: somnath@google.com (Somnath Banerjee)
 
-#include "./dsp.h"
-#include "./yuv.h"
+#include "src/dsp/dsp.h"
+#include "src/dsp/yuv.h"
 
 #include <assert.h>
 
@@ -63,17 +63,17 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
       const uint32_t uv0 = (diag_12 + tl_uv) >> 1;                             \
       const uint32_t uv1 = (diag_03 + t_uv) >> 1;                              \
       FUNC(top_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16),                          \
-           top_dst + (2 * x - 1) * XSTEP);                                     \
+           top_dst + (2 * x - 1) * (XSTEP));                                   \
       FUNC(top_y[2 * x - 0], uv1 & 0xff, (uv1 >> 16),                          \
-           top_dst + (2 * x - 0) * XSTEP);                                     \
+           top_dst + (2 * x - 0) * (XSTEP));                                   \
     }                                                                          \
     if (bottom_y != NULL) {                                                    \
       const uint32_t uv0 = (diag_03 + l_uv) >> 1;                              \
       const uint32_t uv1 = (diag_12 + uv) >> 1;                                \
       FUNC(bottom_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16),                       \
-           bottom_dst + (2 * x - 1) * XSTEP);                                  \
+           bottom_dst + (2 * x - 1) * (XSTEP));                                \
       FUNC(bottom_y[2 * x + 0], uv1 & 0xff, (uv1 >> 16),                       \
-           bottom_dst + (2 * x + 0) * XSTEP);                                  \
+           bottom_dst + (2 * x + 0) * (XSTEP));                                \
     }                                                                          \
     tl_uv = t_uv;                                                              \
     l_uv = uv;                                                                 \
@@ -82,24 +82,50 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
     {                                                                          \
       const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2;              \
       FUNC(top_y[len - 1], uv0 & 0xff, (uv0 >> 16),                            \
-           top_dst + (len - 1) * XSTEP);                                       \
+           top_dst + (len - 1) * (XSTEP));                                     \
     }                                                                          \
     if (bottom_y != NULL) {                                                    \
       const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2;              \
       FUNC(bottom_y[len - 1], uv0 & 0xff, (uv0 >> 16),                         \
-           bottom_dst + (len - 1) * XSTEP);                                    \
+           bottom_dst + (len - 1) * (XSTEP));                                  \
     }                                                                          \
   }                                                                            \
 }
 
 // All variants implemented.
-UPSAMPLE_FUNC(UpsampleRgbLinePair,  VP8YuvToRgb,  3)
-UPSAMPLE_FUNC(UpsampleBgrLinePair,  VP8YuvToBgr,  3)
-UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
-UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
-UPSAMPLE_FUNC(UpsampleArgbLinePair, VP8YuvToArgb, 4)
-UPSAMPLE_FUNC(UpsampleRgba4444LinePair, VP8YuvToRgba4444, 2)
-UPSAMPLE_FUNC(UpsampleRgb565LinePair,  VP8YuvToRgb565,  2)
+#if !WEBP_NEON_OMIT_C_CODE
+UPSAMPLE_FUNC(UpsampleRgbaLinePair_C, VP8YuvToRgba, 4)
+UPSAMPLE_FUNC(UpsampleBgraLinePair_C, VP8YuvToBgra, 4)
+#if !defined(WEBP_REDUCE_CSP)
+UPSAMPLE_FUNC(UpsampleArgbLinePair_C, VP8YuvToArgb, 4)
+UPSAMPLE_FUNC(UpsampleRgbLinePair_C,  VP8YuvToRgb,  3)
+UPSAMPLE_FUNC(UpsampleBgrLinePair_C,  VP8YuvToBgr,  3)
+UPSAMPLE_FUNC(UpsampleRgba4444LinePair_C, VP8YuvToRgba4444, 2)
+UPSAMPLE_FUNC(UpsampleRgb565LinePair_C,  VP8YuvToRgb565,  2)
+#else
+static void EmptyUpsampleFunc(const uint8_t* top_y, const uint8_t* bottom_y,
+                              const uint8_t* top_u, const uint8_t* top_v,
+                              const uint8_t* cur_u, const uint8_t* cur_v,
+                              uint8_t* top_dst, uint8_t* bottom_dst, int len) {
+  (void)top_y;
+  (void)bottom_y;
+  (void)top_u;
+  (void)top_v;
+  (void)cur_u;
+  (void)cur_v;
+  (void)top_dst;
+  (void)bottom_dst;
+  (void)len;
+  assert(0);   // COLORSPACE SUPPORT NOT COMPILED
+}
+#define UpsampleArgbLinePair_C EmptyUpsampleFunc
+#define UpsampleRgbLinePair_C EmptyUpsampleFunc
+#define UpsampleBgrLinePair_C EmptyUpsampleFunc
+#define UpsampleRgba4444LinePair_C EmptyUpsampleFunc
+#define UpsampleRgb565LinePair_C EmptyUpsampleFunc
+#endif   // WEBP_REDUCE_CSP
+
+#endif
 
 #undef LOAD_UV
 #undef UPSAMPLE_FUNC
@@ -141,7 +167,6 @@ DUAL_SAMPLE_FUNC(DualLineSamplerARGB, VP8YuvToArgb)
 
 WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last) {
   WebPInitUpsamplers();
-  VP8YUVInit();
 #ifdef FANCY_UPSAMPLING
   return WebPUpsamplers[alpha_is_last ? MODE_BGRA : MODE_ARGB];
 #else
@@ -158,16 +183,33 @@ extern void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v,    \
 void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v,           \
                uint8_t* dst, int len) {                                        \
   int i;                                                                       \
-  for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * XSTEP]);           \
+  for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * (XSTEP)]);         \
 }
 
-YUV444_FUNC(WebPYuv444ToRgbC,      VP8YuvToRgb,  3)
-YUV444_FUNC(WebPYuv444ToBgrC,      VP8YuvToBgr,  3)
-YUV444_FUNC(WebPYuv444ToRgbaC,     VP8YuvToRgba, 4)
-YUV444_FUNC(WebPYuv444ToBgraC,     VP8YuvToBgra, 4)
-YUV444_FUNC(WebPYuv444ToArgbC,     VP8YuvToArgb, 4)
-YUV444_FUNC(WebPYuv444ToRgba4444C, VP8YuvToRgba4444, 2)
-YUV444_FUNC(WebPYuv444ToRgb565C,   VP8YuvToRgb565, 2)
+YUV444_FUNC(WebPYuv444ToRgba_C,     VP8YuvToRgba, 4)
+YUV444_FUNC(WebPYuv444ToBgra_C,     VP8YuvToBgra, 4)
+#if !defined(WEBP_REDUCE_CSP)
+YUV444_FUNC(WebPYuv444ToRgb_C,      VP8YuvToRgb,  3)
+YUV444_FUNC(WebPYuv444ToBgr_C,      VP8YuvToBgr,  3)
+YUV444_FUNC(WebPYuv444ToArgb_C,     VP8YuvToArgb, 4)
+YUV444_FUNC(WebPYuv444ToRgba4444_C, VP8YuvToRgba4444, 2)
+YUV444_FUNC(WebPYuv444ToRgb565_C,   VP8YuvToRgb565, 2)
+#else
+static void EmptyYuv444Func(const uint8_t* y,
+                            const uint8_t* u, const uint8_t* v,
+                            uint8_t* dst, int len) {
+  (void)y;
+  (void)u;
+  (void)v;
+  (void)dst;
+  (void)len;
+}
+#define WebPYuv444ToRgb_C EmptyYuv444Func
+#define WebPYuv444ToBgr_C EmptyYuv444Func
+#define WebPYuv444ToArgb_C EmptyYuv444Func
+#define WebPYuv444ToRgba4444_C EmptyYuv444Func
+#define WebPYuv444ToRgb565_C EmptyYuv444Func
+#endif   // WEBP_REDUCE_CSP
 
 #undef YUV444_FUNC
 
@@ -182,17 +224,17 @@ static volatile VP8CPUInfo upsampling_last_cpuinfo_used1 =
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444Converters(void) {
   if (upsampling_last_cpuinfo_used1 == VP8GetCPUInfo) return;
 
-  WebPYUV444Converters[MODE_RGB]       = WebPYuv444ToRgbC;
-  WebPYUV444Converters[MODE_RGBA]      = WebPYuv444ToRgbaC;
-  WebPYUV444Converters[MODE_BGR]       = WebPYuv444ToBgrC;
-  WebPYUV444Converters[MODE_BGRA]      = WebPYuv444ToBgraC;
-  WebPYUV444Converters[MODE_ARGB]      = WebPYuv444ToArgbC;
-  WebPYUV444Converters[MODE_RGBA_4444] = WebPYuv444ToRgba4444C;
-  WebPYUV444Converters[MODE_RGB_565]   = WebPYuv444ToRgb565C;
-  WebPYUV444Converters[MODE_rgbA]      = WebPYuv444ToRgbaC;
-  WebPYUV444Converters[MODE_bgrA]      = WebPYuv444ToBgraC;
-  WebPYUV444Converters[MODE_Argb]      = WebPYuv444ToArgbC;
-  WebPYUV444Converters[MODE_rgbA_4444] = WebPYuv444ToRgba4444C;
+  WebPYUV444Converters[MODE_RGBA]      = WebPYuv444ToRgba_C;
+  WebPYUV444Converters[MODE_BGRA]      = WebPYuv444ToBgra_C;
+  WebPYUV444Converters[MODE_RGB]       = WebPYuv444ToRgb_C;
+  WebPYUV444Converters[MODE_BGR]       = WebPYuv444ToBgr_C;
+  WebPYUV444Converters[MODE_ARGB]      = WebPYuv444ToArgb_C;
+  WebPYUV444Converters[MODE_RGBA_4444] = WebPYuv444ToRgba4444_C;
+  WebPYUV444Converters[MODE_RGB_565]   = WebPYuv444ToRgb565_C;
+  WebPYUV444Converters[MODE_rgbA]      = WebPYuv444ToRgba_C;
+  WebPYUV444Converters[MODE_bgrA]      = WebPYuv444ToBgra_C;
+  WebPYUV444Converters[MODE_Argb]      = WebPYuv444ToArgb_C;
+  WebPYUV444Converters[MODE_rgbA_4444] = WebPYuv444ToRgba4444_C;
 
   if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
@@ -224,17 +266,19 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplers(void) {
   if (upsampling_last_cpuinfo_used2 == VP8GetCPUInfo) return;
 
 #ifdef FANCY_UPSAMPLING
-  WebPUpsamplers[MODE_RGB]       = UpsampleRgbLinePair;
-  WebPUpsamplers[MODE_RGBA]      = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_BGR]       = UpsampleBgrLinePair;
-  WebPUpsamplers[MODE_BGRA]      = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_ARGB]      = UpsampleArgbLinePair;
-  WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
-  WebPUpsamplers[MODE_RGB_565]   = UpsampleRgb565LinePair;
-  WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_Argb]      = UpsampleArgbLinePair;
-  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
+#if !WEBP_NEON_OMIT_C_CODE
+  WebPUpsamplers[MODE_RGBA]      = UpsampleRgbaLinePair_C;
+  WebPUpsamplers[MODE_BGRA]      = UpsampleBgraLinePair_C;
+  WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair_C;
+  WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair_C;
+  WebPUpsamplers[MODE_RGB]       = UpsampleRgbLinePair_C;
+  WebPUpsamplers[MODE_BGR]       = UpsampleBgrLinePair_C;
+  WebPUpsamplers[MODE_ARGB]      = UpsampleArgbLinePair_C;
+  WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair_C;
+  WebPUpsamplers[MODE_RGB_565]   = UpsampleRgb565LinePair_C;
+  WebPUpsamplers[MODE_Argb]      = UpsampleArgbLinePair_C;
+  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair_C;
+#endif
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
@@ -243,11 +287,6 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplers(void) {
       WebPInitUpsamplersSSE2();
     }
 #endif
-#if defined(WEBP_USE_NEON)
-    if (VP8GetCPUInfo(kNEON)) {
-      WebPInitUpsamplersNEON();
-    }
-#endif
 #if defined(WEBP_USE_MIPS_DSP_R2)
     if (VP8GetCPUInfo(kMIPSdspR2)) {
       WebPInitUpsamplersMIPSdspR2();
@@ -259,6 +298,26 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplers(void) {
     }
 #endif
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    WebPInitUpsamplersNEON();
+  }
+#endif
+
+  assert(WebPUpsamplers[MODE_RGBA] != NULL);
+  assert(WebPUpsamplers[MODE_BGRA] != NULL);
+  assert(WebPUpsamplers[MODE_rgbA] != NULL);
+  assert(WebPUpsamplers[MODE_bgrA] != NULL);
+  assert(WebPUpsamplers[MODE_RGB] != NULL);
+  assert(WebPUpsamplers[MODE_BGR] != NULL);
+  assert(WebPUpsamplers[MODE_ARGB] != NULL);
+  assert(WebPUpsamplers[MODE_RGBA_4444] != NULL);
+  assert(WebPUpsamplers[MODE_RGB_565] != NULL);
+  assert(WebPUpsamplers[MODE_Argb] != NULL);
+  assert(WebPUpsamplers[MODE_rgbA_4444] != NULL);
+
 #endif  // FANCY_UPSAMPLING
   upsampling_last_cpuinfo_used2 = VP8GetCPUInfo;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/upsampling_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/upsampling_mips_dsp_r2.c
index ed2eb74..10d499d 100644
--- a/src/3rdparty/libwebp/src/dsp/upsampling_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/upsampling_mips_dsp_r2.c
@@ -12,14 +12,12 @@
 // Author(s): Branimir Vasic (branimir.vasic@imgtec.com)
 //            Djordje Pesut  (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
 #include <assert.h>
-#include "./yuv.h"
-
-#if !defined(WEBP_YUV_USE_TABLE)
+#include "src/dsp/yuv.h"
 
 #define YUV_TO_RGB(Y, U, V, R, G, B) do {                                      \
     const int t1 = MultHi(Y, 19077);                                           \
@@ -48,6 +46,7 @@
     );                                                                         \
   } while (0)
 
+#if !defined(WEBP_REDUCE_CSP)
 static WEBP_INLINE void YuvToRgb(int y, int u, int v, uint8_t* const rgb) {
   int r, g, b;
   YUV_TO_RGB(y, u, v, r, g, b);
@@ -68,7 +67,7 @@ static WEBP_INLINE void YuvToRgb565(int y, int u, int v, uint8_t* const rgb) {
   {
     const int rg = (r & 0xf8) | (g >> 5);
     const int gb = ((g << 3) & 0xe0) | (b >> 3);
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     rgb[0] = gb;
     rgb[1] = rg;
 #else
@@ -84,7 +83,7 @@ static WEBP_INLINE void YuvToRgba4444(int y, int u, int v,
   {
     const int rg = (r & 0xf0) | (g >> 4);
     const int ba = (b & 0xf0) | 0x0f;     // overwrite the lower 4 bits
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     argb[0] = ba;
     argb[1] = rg;
 #else
@@ -93,11 +92,12 @@ static WEBP_INLINE void YuvToRgba4444(int y, int u, int v,
 #endif
    }
 }
-#endif  // WEBP_YUV_USE_TABLE
+#endif   // WEBP_REDUCE_CSP
 
 //-----------------------------------------------------------------------------
 // Alpha handling variants
 
+#if !defined(WEBP_REDUCE_CSP)
 static WEBP_INLINE void YuvToArgb(uint8_t y, uint8_t u, uint8_t v,
                                   uint8_t* const argb) {
   int r, g, b;
@@ -107,6 +107,7 @@ static WEBP_INLINE void YuvToArgb(uint8_t y, uint8_t u, uint8_t v,
   argb[2] = g;
   argb[3] = b;
 }
+#endif   // WEBP_REDUCE_CSP
 static WEBP_INLINE void YuvToBgra(uint8_t y, uint8_t u, uint8_t v,
                                   uint8_t* const bgra) {
   int r, g, b;
@@ -200,13 +201,15 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
 }
 
 // All variants implemented.
-UPSAMPLE_FUNC(UpsampleRgbLinePair,      YuvToRgb,      3)
-UPSAMPLE_FUNC(UpsampleBgrLinePair,      YuvToBgr,      3)
 UPSAMPLE_FUNC(UpsampleRgbaLinePair,     YuvToRgba,     4)
 UPSAMPLE_FUNC(UpsampleBgraLinePair,     YuvToBgra,     4)
+#if !defined(WEBP_REDUCE_CSP)
+UPSAMPLE_FUNC(UpsampleRgbLinePair,      YuvToRgb,      3)
+UPSAMPLE_FUNC(UpsampleBgrLinePair,      YuvToBgr,      3)
 UPSAMPLE_FUNC(UpsampleArgbLinePair,     YuvToArgb,     4)
 UPSAMPLE_FUNC(UpsampleRgba4444LinePair, YuvToRgba4444, 2)
 UPSAMPLE_FUNC(UpsampleRgb565LinePair,   YuvToRgb565,   2)
+#endif   // WEBP_REDUCE_CSP
 
 #undef LOAD_UV
 #undef UPSAMPLE_FUNC
@@ -217,17 +220,19 @@ UPSAMPLE_FUNC(UpsampleRgb565LinePair,   YuvToRgb565,   2)
 extern void WebPInitUpsamplersMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersMIPSdspR2(void) {
-  WebPUpsamplers[MODE_RGB]       = UpsampleRgbLinePair;
   WebPUpsamplers[MODE_RGBA]      = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_BGR]       = UpsampleBgrLinePair;
   WebPUpsamplers[MODE_BGRA]      = UpsampleBgraLinePair;
+  WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair;
+#if !defined(WEBP_REDUCE_CSP)
+  WebPUpsamplers[MODE_RGB]       = UpsampleRgbLinePair;
+  WebPUpsamplers[MODE_BGR]       = UpsampleBgrLinePair;
   WebPUpsamplers[MODE_ARGB]      = UpsampleArgbLinePair;
   WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
   WebPUpsamplers[MODE_RGB_565]   = UpsampleRgb565LinePair;
-  WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair;
   WebPUpsamplers[MODE_Argb]      = UpsampleArgbLinePair;
   WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
+#endif   // WEBP_REDUCE_CSP
 }
 
 #endif  // FANCY_UPSAMPLING
@@ -242,13 +247,15 @@ static void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v,    \
   for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * XSTEP]);           \
 }
 
-YUV444_FUNC(Yuv444ToRgb,      YuvToRgb,      3)
-YUV444_FUNC(Yuv444ToBgr,      YuvToBgr,      3)
 YUV444_FUNC(Yuv444ToRgba,     YuvToRgba,     4)
 YUV444_FUNC(Yuv444ToBgra,     YuvToBgra,     4)
+#if !defined(WEBP_REDUCE_CSP)
+YUV444_FUNC(Yuv444ToRgb,      YuvToRgb,      3)
+YUV444_FUNC(Yuv444ToBgr,      YuvToBgr,      3)
 YUV444_FUNC(Yuv444ToArgb,     YuvToArgb,     4)
 YUV444_FUNC(Yuv444ToRgba4444, YuvToRgba4444, 2)
 YUV444_FUNC(Yuv444ToRgb565,   YuvToRgb565,   2)
+#endif   // WEBP_REDUCE_CSP
 
 #undef YUV444_FUNC
 
@@ -258,17 +265,19 @@ YUV444_FUNC(Yuv444ToRgb565,   YuvToRgb565,   2)
 extern void WebPInitYUV444ConvertersMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444ConvertersMIPSdspR2(void) {
-  WebPYUV444Converters[MODE_RGB]       = Yuv444ToRgb;
   WebPYUV444Converters[MODE_RGBA]      = Yuv444ToRgba;
-  WebPYUV444Converters[MODE_BGR]       = Yuv444ToBgr;
   WebPYUV444Converters[MODE_BGRA]      = Yuv444ToBgra;
+  WebPYUV444Converters[MODE_rgbA]      = Yuv444ToRgba;
+  WebPYUV444Converters[MODE_bgrA]      = Yuv444ToBgra;
+#if !defined(WEBP_REDUCE_CSP)
+  WebPYUV444Converters[MODE_RGB]       = Yuv444ToRgb;
+  WebPYUV444Converters[MODE_BGR]       = Yuv444ToBgr;
   WebPYUV444Converters[MODE_ARGB]      = Yuv444ToArgb;
   WebPYUV444Converters[MODE_RGBA_4444] = Yuv444ToRgba4444;
   WebPYUV444Converters[MODE_RGB_565]   = Yuv444ToRgb565;
-  WebPYUV444Converters[MODE_rgbA]      = Yuv444ToRgba;
-  WebPYUV444Converters[MODE_bgrA]      = Yuv444ToBgra;
   WebPYUV444Converters[MODE_Argb]      = Yuv444ToArgb;
   WebPYUV444Converters[MODE_rgbA_4444] = Yuv444ToRgba4444;
+#endif   // WEBP_REDUCE_CSP
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/upsampling_msa.c b/src/3rdparty/libwebp/src/dsp/upsampling_msa.c
index f24926f..535ffb7 100644
--- a/src/3rdparty/libwebp/src/dsp/upsampling_msa.c
+++ b/src/3rdparty/libwebp/src/dsp/upsampling_msa.c
@@ -12,12 +12,12 @@
 // Author: Prashant Patil (prashant.patil@imgtec.com)
 
 #include <string.h>
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MSA)
 
-#include "./msa_macro.h"
-#include "./yuv.h"
+#include "src/dsp/msa_macro.h"
+#include "src/dsp/yuv.h"
 
 #ifdef FANCY_UPSAMPLING
 
@@ -274,7 +274,7 @@ static void YuvToRgb565(int y, int u, int v, uint8_t* const rgb) {
   const int b = Clip8(b1 >> 6);
   const int rg = (r & 0xf8) | (g >> 5);
   const int gb = ((g << 3) & 0xe0) | (b >> 3);
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
   rgb[0] = gb;
   rgb[1] = rg;
 #else
@@ -293,7 +293,7 @@ static void YuvToRgba4444(int y, int u, int v, uint8_t* const argb) {
   const int b = Clip8(b1 >> 6);
   const int rg = (r & 0xf0) | (g >> 4);
   const int ba = (b & 0xf0) | 0x0f;     // overwrite the lower 4 bits
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
   argb[0] = ba;
   argb[1] = rg;
 #else
@@ -374,7 +374,7 @@ static void YuvToBgrLine(const uint8_t* y, const uint8_t* u,
 static void YuvToRgbaLine(const uint8_t* y, const uint8_t* u,
                           const uint8_t* v, uint8_t* dst, int length) {
   v16u8 R, G, B;
-  const v16u8 A = (v16u8)__msa_ldi_b(0xff);
+  const v16u8 A = (v16u8)__msa_ldi_b(ALPHAVAL);
   while (length >= 16) {
     CALC_RGB16(y, u, v, R, G, B);
     STORE16_4(R, G, B, A, dst);
@@ -402,7 +402,7 @@ static void YuvToRgbaLine(const uint8_t* y, const uint8_t* u,
 static void YuvToBgraLine(const uint8_t* y, const uint8_t* u,
                           const uint8_t* v, uint8_t* dst, int length) {
   v16u8 R, G, B;
-  const v16u8 A = (v16u8)__msa_ldi_b(0xff);
+  const v16u8 A = (v16u8)__msa_ldi_b(ALPHAVAL);
   while (length >= 16) {
     CALC_RGB16(y, u, v, R, G, B);
     STORE16_4(B, G, R, A, dst);
@@ -430,7 +430,7 @@ static void YuvToBgraLine(const uint8_t* y, const uint8_t* u,
 static void YuvToArgbLine(const uint8_t* y, const uint8_t* u,
                           const uint8_t* v, uint8_t* dst, int length) {
   v16u8 R, G, B;
-  const v16u8 A = (v16u8)__msa_ldi_b(0xff);
+  const v16u8 A = (v16u8)__msa_ldi_b(ALPHAVAL);
   while (length >= 16) {
     CALC_RGB16(y, u, v, R, G, B);
     STORE16_4(A, R, G, B, dst);
@@ -459,11 +459,11 @@ static void YuvToRgba4444Line(const uint8_t* y, const uint8_t* u,
                               const uint8_t* v, uint8_t* dst, int length) {
   v16u8 R, G, B, RG, BA, tmp0, tmp1;
   while (length >= 16) {
-  #ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     CALC_RGBA4444(y, u, v, BA, RG, 16, dst);
-  #else
+#else
     CALC_RGBA4444(y, u, v, RG, BA, 16, dst);
-  #endif
+#endif
     y      += 16;
     u      += 16;
     v      += 16;
@@ -473,7 +473,7 @@ static void YuvToRgba4444Line(const uint8_t* y, const uint8_t* u,
   if (length > 8) {
     uint8_t temp[2 * 16] = { 0 };
     memcpy(temp, y, length * sizeof(*temp));
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     CALC_RGBA4444(temp, u, v, BA, RG, 16, temp);
 #else
     CALC_RGBA4444(temp, u, v, RG, BA, 16, temp);
@@ -482,7 +482,7 @@ static void YuvToRgba4444Line(const uint8_t* y, const uint8_t* u,
   } else if (length > 0) {
     uint8_t temp[2 * 8] = { 0 };
     memcpy(temp, y, length * sizeof(*temp));
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     CALC_RGBA4444(temp, u, v, BA, RG, 8, temp);
 #else
     CALC_RGBA4444(temp, u, v, RG, BA, 8, temp);
@@ -495,11 +495,11 @@ static void YuvToRgb565Line(const uint8_t* y, const uint8_t* u,
                             const uint8_t* v, uint8_t* dst, int length) {
   v16u8 R, G, B, RG, GB, tmp0, tmp1;
   while (length >= 16) {
-  #ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     CALC_RGB565(y, u, v, GB, RG, 16, dst);
-  #else
+#else
     CALC_RGB565(y, u, v, RG, GB, 16, dst);
-  #endif
+#endif
     y      += 16;
     u      += 16;
     v      += 16;
@@ -509,7 +509,7 @@ static void YuvToRgb565Line(const uint8_t* y, const uint8_t* u,
   if (length > 8) {
     uint8_t temp[2 * 16] = { 0 };
     memcpy(temp, y, length * sizeof(*temp));
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     CALC_RGB565(temp, u, v, GB, RG, 16, temp);
 #else
     CALC_RGB565(temp, u, v, RG, GB, 16, temp);
@@ -518,7 +518,7 @@ static void YuvToRgb565Line(const uint8_t* y, const uint8_t* u,
   } else if (length > 0) {
     uint8_t temp[2 * 8] = { 0 };
     memcpy(temp, y, length * sizeof(*temp));
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
     CALC_RGB565(temp, u, v, GB, RG, 8, temp);
 #else
     CALC_RGB565(temp, u, v, RG, GB, 8, temp);
@@ -640,13 +640,15 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y,        \
   }                                                                      \
 }
 
-UPSAMPLE_FUNC(UpsampleRgbLinePair,      YuvToRgb,      3)
-UPSAMPLE_FUNC(UpsampleBgrLinePair,      YuvToBgr,      3)
 UPSAMPLE_FUNC(UpsampleRgbaLinePair,     YuvToRgba,     4)
 UPSAMPLE_FUNC(UpsampleBgraLinePair,     YuvToBgra,     4)
+#if !defined(WEBP_REDUCE_CSP)
+UPSAMPLE_FUNC(UpsampleRgbLinePair,      YuvToRgb,      3)
+UPSAMPLE_FUNC(UpsampleBgrLinePair,      YuvToBgr,      3)
 UPSAMPLE_FUNC(UpsampleArgbLinePair,     YuvToArgb,     4)
 UPSAMPLE_FUNC(UpsampleRgba4444LinePair, YuvToRgba4444, 2)
 UPSAMPLE_FUNC(UpsampleRgb565LinePair,   YuvToRgb565,   2)
+#endif   // WEBP_REDUCE_CSP
 
 //------------------------------------------------------------------------------
 // Entry point
@@ -656,17 +658,19 @@ extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
 extern void WebPInitUpsamplersMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersMSA(void) {
-  WebPUpsamplers[MODE_RGB]       = UpsampleRgbLinePair;
   WebPUpsamplers[MODE_RGBA]      = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_BGR]       = UpsampleBgrLinePair;
   WebPUpsamplers[MODE_BGRA]      = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_ARGB]      = UpsampleArgbLinePair;
   WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair;
   WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair;
+#if !defined(WEBP_REDUCE_CSP)
+  WebPUpsamplers[MODE_RGB]       = UpsampleRgbLinePair;
+  WebPUpsamplers[MODE_BGR]       = UpsampleBgrLinePair;
+  WebPUpsamplers[MODE_ARGB]      = UpsampleArgbLinePair;
   WebPUpsamplers[MODE_Argb]      = UpsampleArgbLinePair;
   WebPUpsamplers[MODE_RGB_565]   = UpsampleRgb565LinePair;
   WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
   WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
+#endif   // WEBP_REDUCE_CSP
 }
 
 #endif  // FANCY_UPSAMPLING
diff --git a/src/3rdparty/libwebp/src/dsp/upsampling_neon.c b/src/3rdparty/libwebp/src/dsp/upsampling_neon.c
index d371a83..17cbc9f 100644
--- a/src/3rdparty/libwebp/src/dsp/upsampling_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/upsampling_neon.c
@@ -12,15 +12,15 @@
 // Author: mans@mansr.com (Mans Rullgard)
 // Based on SSE code by: somnath@google.com (Somnath Banerjee)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_NEON)
 
 #include <assert.h>
 #include <arm_neon.h>
 #include <string.h>
-#include "./neon.h"
-#include "./yuv.h"
+#include "src/dsp/neon.h"
+#include "src/dsp/yuv.h"
 
 #ifdef FANCY_UPSAMPLING
 
@@ -58,8 +58,8 @@
 } while (0)
 
 // Turn the macro into a function for reducing code-size when non-critical
-static void Upsample16Pixels(const uint8_t *r1, const uint8_t *r2,
-                             uint8_t *out) {
+static void Upsample16Pixels_NEON(const uint8_t *r1, const uint8_t *r2,
+                                  uint8_t *out) {
   UPSAMPLE_16PIXELS(r1, r2, out);
 }
 
@@ -70,7 +70,7 @@ static void Upsample16Pixels(const uint8_t *r1, const uint8_t *r2,
   /* replicate last byte */                                             \
   memset(r1 + (num_pixels), r1[(num_pixels) - 1], 9 - (num_pixels));    \
   memset(r2 + (num_pixels), r2[(num_pixels) - 1], 9 - (num_pixels));    \
-  Upsample16Pixels(r1, r2, out);                                        \
+  Upsample16Pixels_NEON(r1, r2, out);                                   \
 }
 
 //-----------------------------------------------------------------------------
@@ -243,13 +243,15 @@ static void FUNC_NAME(const uint8_t *top_y, const uint8_t *bottom_y,    \
 }
 
 // NEON variants of the fancy upsampler.
-NEON_UPSAMPLE_FUNC(UpsampleRgbLinePair,  Rgb,  3)
-NEON_UPSAMPLE_FUNC(UpsampleBgrLinePair,  Bgr,  3)
-NEON_UPSAMPLE_FUNC(UpsampleRgbaLinePair, Rgba, 4)
-NEON_UPSAMPLE_FUNC(UpsampleBgraLinePair, Bgra, 4)
-NEON_UPSAMPLE_FUNC(UpsampleArgbLinePair, Argb, 4)
-NEON_UPSAMPLE_FUNC(UpsampleRgba4444LinePair, Rgba4444, 2)
-NEON_UPSAMPLE_FUNC(UpsampleRgb565LinePair, Rgb565, 2)
+NEON_UPSAMPLE_FUNC(UpsampleRgbaLinePair_NEON, Rgba, 4)
+NEON_UPSAMPLE_FUNC(UpsampleBgraLinePair_NEON, Bgra, 4)
+#if !defined(WEBP_REDUCE_CSP)
+NEON_UPSAMPLE_FUNC(UpsampleRgbLinePair_NEON,  Rgb,  3)
+NEON_UPSAMPLE_FUNC(UpsampleBgrLinePair_NEON,  Bgr,  3)
+NEON_UPSAMPLE_FUNC(UpsampleArgbLinePair_NEON, Argb, 4)
+NEON_UPSAMPLE_FUNC(UpsampleRgba4444LinePair_NEON, Rgba4444, 2)
+NEON_UPSAMPLE_FUNC(UpsampleRgb565LinePair_NEON, Rgb565, 2)
+#endif   // WEBP_REDUCE_CSP
 
 //------------------------------------------------------------------------------
 // Entry point
@@ -259,17 +261,19 @@ extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
 extern void WebPInitUpsamplersNEON(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersNEON(void) {
-  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair;
-  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair;
-  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair;
-  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair;
-  WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair;
-  WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
-  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
+  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair_NEON;
+  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair_NEON;
+  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair_NEON;
+  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair_NEON;
+#if !defined(WEBP_REDUCE_CSP)
+  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair_NEON;
+  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair_NEON;
+  WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair_NEON;
+  WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair_NEON;
+  WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair_NEON;
+  WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair_NEON;
+  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair_NEON;
+#endif   // WEBP_REDUCE_CSP
 }
 
 #endif  // FANCY_UPSAMPLING
diff --git a/src/3rdparty/libwebp/src/dsp/upsampling_sse2.c b/src/3rdparty/libwebp/src/dsp/upsampling_sse2.c
index b5b6689..fd5d303 100644
--- a/src/3rdparty/libwebp/src/dsp/upsampling_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/upsampling_sse2.c
@@ -11,14 +11,14 @@
 //
 // Author: somnath@google.com (Somnath Banerjee)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 
 #include <assert.h>
 #include <emmintrin.h>
 #include <string.h>
-#include "./yuv.h"
+#include "src/dsp/yuv.h"
 
 #ifdef FANCY_UPSAMPLING
 
@@ -83,13 +83,13 @@
   GET_M(ad, s, diag2);                  /* diag2 = (3a + b + c + 3d) / 8 */    \
                                                                                \
   /* pack the alternate pixels */                                              \
-  PACK_AND_STORE(a, b, diag1, diag2, out +      0);  /* store top */           \
-  PACK_AND_STORE(c, d, diag2, diag1, out + 2 * 32);  /* store bottom */        \
+  PACK_AND_STORE(a, b, diag1, diag2, (out) +      0);  /* store top */         \
+  PACK_AND_STORE(c, d, diag2, diag1, (out) + 2 * 32);  /* store bottom */      \
 }
 
 // Turn the macro into a function for reducing code-size when non-critical
-static void Upsample32Pixels(const uint8_t r1[], const uint8_t r2[],
-                             uint8_t* const out) {
+static void Upsample32Pixels_SSE2(const uint8_t r1[], const uint8_t r2[],
+                                  uint8_t* const out) {
   UPSAMPLE_32PIXELS(r1, r2, out);
 }
 
@@ -101,30 +101,30 @@ static void Upsample32Pixels(const uint8_t r1[], const uint8_t r2[],
   memset(r1 + (num_pixels), r1[(num_pixels) - 1], 17 - (num_pixels));          \
   memset(r2 + (num_pixels), r2[(num_pixels) - 1], 17 - (num_pixels));          \
   /* using the shared function instead of the macro saves ~3k code size */     \
-  Upsample32Pixels(r1, r2, out);                                               \
+  Upsample32Pixels_SSE2(r1, r2, out);                                          \
 }
 
 #define CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y,                              \
                     top_dst, bottom_dst, cur_x, num_pixels) {                  \
   int n;                                                                       \
   for (n = 0; n < (num_pixels); ++n) {                                         \
-    FUNC(top_y[(cur_x) + n], r_u[n], r_v[n],                                   \
-         top_dst + ((cur_x) + n) * XSTEP);                                     \
+    FUNC((top_y)[(cur_x) + n], r_u[n], r_v[n],                                 \
+         (top_dst) + ((cur_x) + n) * (XSTEP));                                 \
   }                                                                            \
-  if (bottom_y != NULL) {                                                      \
+  if ((bottom_y) != NULL) {                                                    \
     for (n = 0; n < (num_pixels); ++n) {                                       \
-      FUNC(bottom_y[(cur_x) + n], r_u[64 + n], r_v[64 + n],                    \
-           bottom_dst + ((cur_x) + n) * XSTEP);                                \
+      FUNC((bottom_y)[(cur_x) + n], r_u[64 + n], r_v[64 + n],                  \
+           (bottom_dst) + ((cur_x) + n) * (XSTEP));                            \
     }                                                                          \
   }                                                                            \
 }
 
 #define CONVERT2RGB_32(FUNC, XSTEP, top_y, bottom_y,                           \
                        top_dst, bottom_dst, cur_x) do {                        \
-  FUNC##32(top_y + (cur_x), r_u, r_v, top_dst + (cur_x) * XSTEP);              \
-  if (bottom_y != NULL) {                                                      \
-    FUNC##32(bottom_y + (cur_x), r_u + 64, r_v + 64,                           \
-             bottom_dst + (cur_x) * XSTEP);                                    \
+  FUNC##32_SSE2((top_y) + (cur_x), r_u, r_v, (top_dst) + (cur_x) * (XSTEP));   \
+  if ((bottom_y) != NULL) {                                                    \
+    FUNC##32_SSE2((bottom_y) + (cur_x), r_u + 64, r_v + 64,                    \
+                  (bottom_dst) + (cur_x) * (XSTEP));                           \
   }                                                                            \
 } while (0)
 
@@ -169,13 +169,16 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
 }
 
 // SSE2 variants of the fancy upsampler.
-SSE2_UPSAMPLE_FUNC(UpsampleRgbLinePair,  VP8YuvToRgb,  3)
-SSE2_UPSAMPLE_FUNC(UpsampleBgrLinePair,  VP8YuvToBgr,  3)
-SSE2_UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
-SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
-SSE2_UPSAMPLE_FUNC(UpsampleArgbLinePair, VP8YuvToArgb, 4)
-SSE2_UPSAMPLE_FUNC(UpsampleRgba4444LinePair, VP8YuvToRgba4444, 2)
-SSE2_UPSAMPLE_FUNC(UpsampleRgb565LinePair, VP8YuvToRgb565, 2)
+SSE2_UPSAMPLE_FUNC(UpsampleRgbaLinePair_SSE2, VP8YuvToRgba, 4)
+SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePair_SSE2, VP8YuvToBgra, 4)
+
+#if !defined(WEBP_REDUCE_CSP)
+SSE2_UPSAMPLE_FUNC(UpsampleRgbLinePair_SSE2,  VP8YuvToRgb,  3)
+SSE2_UPSAMPLE_FUNC(UpsampleBgrLinePair_SSE2,  VP8YuvToBgr,  3)
+SSE2_UPSAMPLE_FUNC(UpsampleArgbLinePair_SSE2, VP8YuvToArgb, 4)
+SSE2_UPSAMPLE_FUNC(UpsampleRgba4444LinePair_SSE2, VP8YuvToRgba4444, 2)
+SSE2_UPSAMPLE_FUNC(UpsampleRgb565LinePair_SSE2, VP8YuvToRgb565, 2)
+#endif   // WEBP_REDUCE_CSP
 
 #undef GET_M
 #undef PACK_AND_STORE
@@ -193,17 +196,19 @@ extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
 extern void WebPInitUpsamplersSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersSSE2(void) {
-  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair;
-  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair;
-  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair;
-  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair;
-  WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair;
-  WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
-  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
+  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair_SSE2;
+  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair_SSE2;
+  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair_SSE2;
+  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair_SSE2;
+#if !defined(WEBP_REDUCE_CSP)
+  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair_SSE2;
+  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair_SSE2;
+  WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair_SSE2;
+  WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair_SSE2;
+  WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair_SSE2;
+  WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair_SSE2;
+  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair_SSE2;
+#endif   // WEBP_REDUCE_CSP
 }
 
 #endif  // FANCY_UPSAMPLING
@@ -213,29 +218,46 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersSSE2(void) {
 extern WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */];
 extern void WebPInitYUV444ConvertersSSE2(void);
 
-#define YUV444_FUNC(FUNC_NAME, CALL, XSTEP) \
-extern void WebP##FUNC_NAME##C(const uint8_t* y, const uint8_t* u,             \
-                               const uint8_t* v, uint8_t* dst, int len);       \
+#define YUV444_FUNC(FUNC_NAME, CALL, CALL_C, XSTEP)                            \
+extern void CALL_C(const uint8_t* y, const uint8_t* u, const uint8_t* v,       \
+                   uint8_t* dst, int len);                                     \
 static void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v,    \
                       uint8_t* dst, int len) {                                 \
   int i;                                                                       \
   const int max_len = len & ~31;                                               \
-  for (i = 0; i < max_len; i += 32) CALL(y + i, u + i, v + i, dst + i * XSTEP);\
+  for (i = 0; i < max_len; i += 32) {                                          \
+    CALL(y + i, u + i, v + i, dst + i * (XSTEP));                              \
+  }                                                                            \
   if (i < len) {  /* C-fallback */                                             \
-    WebP##FUNC_NAME##C(y + i, u + i, v + i, dst + i * XSTEP, len - i);         \
+    CALL_C(y + i, u + i, v + i, dst + i * (XSTEP), len - i);                   \
   }                                                                            \
 }
 
-YUV444_FUNC(Yuv444ToRgba, VP8YuvToRgba32, 4);
-YUV444_FUNC(Yuv444ToBgra, VP8YuvToBgra32, 4);
-YUV444_FUNC(Yuv444ToRgb, VP8YuvToRgb32, 3);
-YUV444_FUNC(Yuv444ToBgr, VP8YuvToBgr32, 3);
+YUV444_FUNC(Yuv444ToRgba_SSE2, VP8YuvToRgba32_SSE2, WebPYuv444ToRgba_C, 4);
+YUV444_FUNC(Yuv444ToBgra_SSE2, VP8YuvToBgra32_SSE2, WebPYuv444ToBgra_C, 4);
+#if !defined(WEBP_REDUCE_CSP)
+YUV444_FUNC(Yuv444ToRgb_SSE2, VP8YuvToRgb32_SSE2, WebPYuv444ToRgb_C, 3);
+YUV444_FUNC(Yuv444ToBgr_SSE2, VP8YuvToBgr32_SSE2, WebPYuv444ToBgr_C, 3);
+YUV444_FUNC(Yuv444ToArgb_SSE2, VP8YuvToArgb32_SSE2, WebPYuv444ToArgb_C, 4)
+YUV444_FUNC(Yuv444ToRgba4444_SSE2, VP8YuvToRgba444432_SSE2, \
+            WebPYuv444ToRgba4444_C, 2)
+YUV444_FUNC(Yuv444ToRgb565_SSE2, VP8YuvToRgb56532_SSE2, WebPYuv444ToRgb565_C, 2)
+#endif   // WEBP_REDUCE_CSP
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444ConvertersSSE2(void) {
-  WebPYUV444Converters[MODE_RGBA] = Yuv444ToRgba;
-  WebPYUV444Converters[MODE_BGRA] = Yuv444ToBgra;
-  WebPYUV444Converters[MODE_RGB]  = Yuv444ToRgb;
-  WebPYUV444Converters[MODE_BGR]  = Yuv444ToBgr;
+  WebPYUV444Converters[MODE_RGBA]      = Yuv444ToRgba_SSE2;
+  WebPYUV444Converters[MODE_BGRA]      = Yuv444ToBgra_SSE2;
+  WebPYUV444Converters[MODE_rgbA]      = Yuv444ToRgba_SSE2;
+  WebPYUV444Converters[MODE_bgrA]      = Yuv444ToBgra_SSE2;
+#if !defined(WEBP_REDUCE_CSP)
+  WebPYUV444Converters[MODE_RGB]       = Yuv444ToRgb_SSE2;
+  WebPYUV444Converters[MODE_BGR]       = Yuv444ToBgr_SSE2;
+  WebPYUV444Converters[MODE_ARGB]      = Yuv444ToArgb_SSE2;
+  WebPYUV444Converters[MODE_RGBA_4444] = Yuv444ToRgba4444_SSE2;
+  WebPYUV444Converters[MODE_RGB_565]   = Yuv444ToRgb565_SSE2;
+  WebPYUV444Converters[MODE_Argb]      = Yuv444ToArgb_SSE2;
+  WebPYUV444Converters[MODE_rgbA_4444] = Yuv444ToRgba4444_SSE2;
+#endif   // WEBP_REDUCE_CSP
 }
 
 #else
diff --git a/src/3rdparty/libwebp/src/dsp/yuv.c b/src/3rdparty/libwebp/src/dsp/yuv.c
index dd7d9de..bddf81f 100644
--- a/src/3rdparty/libwebp/src/dsp/yuv.c
+++ b/src/3rdparty/libwebp/src/dsp/yuv.c
@@ -11,63 +11,11 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./yuv.h"
+#include "src/dsp/yuv.h"
 
+#include <assert.h>
 #include <stdlib.h>
 
-#if defined(WEBP_YUV_USE_TABLE)
-
-static int done = 0;
-
-static WEBP_INLINE uint8_t clip(int v, int max_value) {
-  return v < 0 ? 0 : v > max_value ? max_value : v;
-}
-
-int16_t VP8kVToR[256], VP8kUToB[256];
-int32_t VP8kVToG[256], VP8kUToG[256];
-uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
-uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
-
-WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInit(void) {
-  int i;
-  if (done) {
-    return;
-  }
-#ifndef USE_YUVj
-  for (i = 0; i < 256; ++i) {
-    VP8kVToR[i] = (89858 * (i - 128) + YUV_HALF) >> YUV_FIX;
-    VP8kUToG[i] = -22014 * (i - 128) + YUV_HALF;
-    VP8kVToG[i] = -45773 * (i - 128);
-    VP8kUToB[i] = (113618 * (i - 128) + YUV_HALF) >> YUV_FIX;
-  }
-  for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) {
-    const int k = ((i - 16) * 76283 + YUV_HALF) >> YUV_FIX;
-    VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
-    VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
-  }
-#else
-  for (i = 0; i < 256; ++i) {
-    VP8kVToR[i] = (91881 * (i - 128) + YUV_HALF) >> YUV_FIX;
-    VP8kUToG[i] = -22554 * (i - 128) + YUV_HALF;
-    VP8kVToG[i] = -46802 * (i - 128);
-    VP8kUToB[i] = (116130 * (i - 128) + YUV_HALF) >> YUV_FIX;
-  }
-  for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) {
-    const int k = i;
-    VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
-    VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
-  }
-#endif
-
-  done = 1;
-}
-
-#else
-
-WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInit(void) {}
-
-#endif  // WEBP_YUV_USE_TABLE
-
 //-----------------------------------------------------------------------------
 // Plain-C version
 
@@ -75,14 +23,14 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInit(void) {}
 static void FUNC_NAME(const uint8_t* y,                                        \
                       const uint8_t* u, const uint8_t* v,                      \
                       uint8_t* dst, int len) {                                 \
-  const uint8_t* const end = dst + (len & ~1) * XSTEP;                         \
+  const uint8_t* const end = dst + (len & ~1) * (XSTEP);                       \
   while (dst != end) {                                                         \
     FUNC(y[0], u[0], v[0], dst);                                               \
-    FUNC(y[1], u[0], v[0], dst + XSTEP);                                       \
+    FUNC(y[1], u[0], v[0], dst + (XSTEP));                                     \
     y += 2;                                                                    \
     ++u;                                                                       \
     ++v;                                                                       \
-    dst += 2 * XSTEP;                                                          \
+    dst += 2 * (XSTEP);                                                        \
   }                                                                            \
   if (len & 1) {                                                               \
     FUNC(y[0], u[0], v[0], dst);                                               \
@@ -168,7 +116,7 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplers(void) {
 //-----------------------------------------------------------------------------
 // ARGB -> YUV converters
 
-static void ConvertARGBToY(const uint32_t* argb, uint8_t* y, int width) {
+static void ConvertARGBToY_C(const uint32_t* argb, uint8_t* y, int width) {
   int i;
   for (i = 0; i < width; ++i) {
     const uint32_t p = argb[i];
@@ -220,14 +168,14 @@ void WebPConvertARGBToUV_C(const uint32_t* argb, uint8_t* u, uint8_t* v,
 
 //-----------------------------------------------------------------------------
 
-static void ConvertRGB24ToY(const uint8_t* rgb, uint8_t* y, int width) {
+static void ConvertRGB24ToY_C(const uint8_t* rgb, uint8_t* y, int width) {
   int i;
   for (i = 0; i < width; ++i, rgb += 3) {
     y[i] = VP8RGBToY(rgb[0], rgb[1], rgb[2], YUV_HALF);
   }
 }
 
-static void ConvertBGR24ToY(const uint8_t* bgr, uint8_t* y, int width) {
+static void ConvertBGR24ToY_C(const uint8_t* bgr, uint8_t* y, int width) {
   int i;
   for (i = 0; i < width; ++i, bgr += 3) {
     y[i] = VP8RGBToY(bgr[2], bgr[1], bgr[0], YUV_HALF);
@@ -246,6 +194,7 @@ void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
 
 //-----------------------------------------------------------------------------
 
+#if !WEBP_NEON_OMIT_C_CODE
 #define MAX_Y ((1 << 10) - 1)    // 10b precision over 16b-arithmetic
 static uint16_t clip_y(int v) {
   return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
@@ -283,6 +232,7 @@ static void SharpYUVFilterRow_C(const int16_t* A, const int16_t* B, int len,
     out[2 * i + 1] = clip_y(best_y[2 * i + 1] + v1);
   }
 }
+#endif  // !WEBP_NEON_OMIT_C_CODE
 
 #undef MAX_Y
 
@@ -308,22 +258,26 @@ static volatile VP8CPUInfo rgba_to_yuv_last_cpuinfo_used =
     (VP8CPUInfo)&rgba_to_yuv_last_cpuinfo_used;
 
 extern void WebPInitConvertARGBToYUVSSE2(void);
+extern void WebPInitConvertARGBToYUVNEON(void);
 extern void WebPInitSharpYUVSSE2(void);
+extern void WebPInitSharpYUVNEON(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUV(void) {
   if (rgba_to_yuv_last_cpuinfo_used == VP8GetCPUInfo) return;
 
-  WebPConvertARGBToY = ConvertARGBToY;
+  WebPConvertARGBToY = ConvertARGBToY_C;
   WebPConvertARGBToUV = WebPConvertARGBToUV_C;
 
-  WebPConvertRGB24ToY = ConvertRGB24ToY;
-  WebPConvertBGR24ToY = ConvertBGR24ToY;
+  WebPConvertRGB24ToY = ConvertRGB24ToY_C;
+  WebPConvertBGR24ToY = ConvertBGR24ToY_C;
 
   WebPConvertRGBA32ToUV = WebPConvertRGBA32ToUV_C;
 
+#if !WEBP_NEON_OMIT_C_CODE
   WebPSharpYUVUpdateY = SharpYUVUpdateY_C;
   WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_C;
   WebPSharpYUVFilterRow = SharpYUVFilterRow_C;
+#endif
 
   if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
@@ -333,5 +287,23 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUV(void) {
     }
 #endif  // WEBP_USE_SSE2
   }
+
+#if defined(WEBP_USE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
+    WebPInitConvertARGBToYUVNEON();
+    WebPInitSharpYUVNEON();
+  }
+#endif  // WEBP_USE_NEON
+
+  assert(WebPConvertARGBToY != NULL);
+  assert(WebPConvertARGBToUV != NULL);
+  assert(WebPConvertRGB24ToY != NULL);
+  assert(WebPConvertBGR24ToY != NULL);
+  assert(WebPConvertRGBA32ToUV != NULL);
+  assert(WebPSharpYUVUpdateY != NULL);
+  assert(WebPSharpYUVUpdateRGB != NULL);
+  assert(WebPSharpYUVFilterRow != NULL);
+
   rgba_to_yuv_last_cpuinfo_used = VP8GetCPUInfo;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/yuv.h b/src/3rdparty/libwebp/src/dsp/yuv.h
index 1d33b58..c8a5583 100644
--- a/src/3rdparty/libwebp/src/dsp/yuv.h
+++ b/src/3rdparty/libwebp/src/dsp/yuv.h
@@ -35,18 +35,8 @@
 #ifndef WEBP_DSP_YUV_H_
 #define WEBP_DSP_YUV_H_
 
-#include "./dsp.h"
-#include "../dec/vp8_dec.h"
-
-#if defined(WEBP_EXPERIMENTAL_FEATURES)
-// Do NOT activate this feature for real compression. This is only experimental!
-// This flag is for comparison purpose against JPEG's "YUVj" natural colorspace.
-// This colorspace is close to Rec.601's Y'CbCr model with the notable
-// difference of allowing larger range for luma/chroma.
-// See http://en.wikipedia.org/wiki/YCbCr#JPEG_conversion paragraph, and its
-// difference with http://en.wikipedia.org/wiki/YCbCr#ITU-R_BT.601_conversion
-// #define USE_YUVj
-#endif
+#include "src/dsp/dsp.h"
+#include "src/dec/vp8_dec.h"
 
 //------------------------------------------------------------------------------
 // YUV -> RGB conversion
@@ -58,12 +48,8 @@ extern "C" {
 enum {
   YUV_FIX = 16,                    // fixed-point precision for RGB->YUV
   YUV_HALF = 1 << (YUV_FIX - 1),
-  YUV_MASK = (256 << YUV_FIX) - 1,
-  YUV_RANGE_MIN = -227,            // min value of r/g/b output
-  YUV_RANGE_MAX = 256 + 226,       // max value of r/g/b output
 
   YUV_FIX2 = 6,                   // fixed-point precision for YUV->RGB
-  YUV_HALF2 = 1 << YUV_FIX2 >> 1,
   YUV_MASK2 = (256 << YUV_FIX2) - 1
 };
 
@@ -111,7 +97,7 @@ static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
   const int b = VP8YUVToB(y, u);      // 5 usable bits
   const int rg = (r & 0xf8) | (g >> 5);
   const int gb = ((g << 3) & 0xe0) | (b >> 3);
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
   rgb[0] = gb;
   rgb[1] = rg;
 #else
@@ -127,7 +113,7 @@ static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
   const int b = VP8YUVToB(y, u);        // 4 usable bits
   const int rg = (r & 0xf0) | (g >> 4);
   const int ba = (b & 0xf0) | 0x0f;     // overwrite the lower 4 bits
-#ifdef WEBP_SWAP_16BIT_CSP
+#if (WEBP_SWAP_16BIT_CSP == 1)
   argb[0] = ba;
   argb[1] = rg;
 #else
@@ -157,29 +143,26 @@ static WEBP_INLINE void VP8YuvToRgba(uint8_t y, uint8_t u, uint8_t v,
   rgba[3] = 0xff;
 }
 
-// Must be called before everything, to initialize the tables.
-void VP8YUVInit(void);
-
 //-----------------------------------------------------------------------------
 // SSE2 extra functions (mostly for upsampling_sse2.c)
 
 #if defined(WEBP_USE_SSE2)
 
 // Process 32 pixels and store the result (16b, 24b or 32b per pixel) in *dst.
-void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                    uint8_t* dst);
-void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                   uint8_t* dst);
-void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                    uint8_t* dst);
-void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                   uint8_t* dst);
-void VP8YuvToArgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                    uint8_t* dst);
-void VP8YuvToRgba444432(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+void VP8YuvToRgba32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                         uint8_t* dst);
+void VP8YuvToRgb32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                        uint8_t* dst);
+void VP8YuvToBgra32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                         uint8_t* dst);
+void VP8YuvToBgr32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
                         uint8_t* dst);
-void VP8YuvToRgb56532(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                      uint8_t* dst);
+void VP8YuvToArgb32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                         uint8_t* dst);
+void VP8YuvToRgba444432_SSE2(const uint8_t* y, const uint8_t* u,
+                             const uint8_t* v, uint8_t* dst);
+void VP8YuvToRgb56532_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                           uint8_t* dst);
 
 #endif    // WEBP_USE_SSE2
 
@@ -192,8 +175,6 @@ static WEBP_INLINE int VP8ClipUV(int uv, int rounding) {
   return ((uv & ~0xff) == 0) ? uv : (uv < 0) ? 0 : 255;
 }
 
-#ifndef USE_YUVj
-
 static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
   const int luma = 16839 * r + 33059 * g + 6420 * b;
   return (luma + rounding + (16 << YUV_FIX)) >> YUV_FIX;  // no need to clip
@@ -209,28 +190,6 @@ static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
   return VP8ClipUV(v, rounding);
 }
 
-#else
-
-// This JPEG-YUV colorspace, only for comparison!
-// These are also 16bit precision coefficients from Rec.601, but with full
-// [0..255] output range.
-static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
-  const int luma = 19595 * r + 38470 * g + 7471 * b;
-  return (luma + rounding) >> YUV_FIX;  // no need to clip
-}
-
-static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
-  const int u = -11058 * r - 21710 * g + 32768 * b;
-  return VP8ClipUV(u, rounding);
-}
-
-static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
-  const int v = 32768 * r - 27439 * g - 5329 * b;
-  return VP8ClipUV(v, rounding);
-}
-
-#endif    // USE_YUVj
-
 #ifdef __cplusplus
 }    // extern "C"
 #endif
diff --git a/src/3rdparty/libwebp/src/dsp/yuv_mips32.c b/src/3rdparty/libwebp/src/dsp/yuv_mips32.c
index e61aac5..9d0a887 100644
--- a/src/3rdparty/libwebp/src/dsp/yuv_mips32.c
+++ b/src/3rdparty/libwebp/src/dsp/yuv_mips32.c
@@ -12,11 +12,11 @@
 // Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
 //             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS32)
 
-#include "./yuv.h"
+#include "src/dsp/yuv.h"
 
 //------------------------------------------------------------------------------
 // simple point-sampling
@@ -77,10 +77,10 @@ static void FUNC_NAME(const uint8_t* y,                                        \
   }                                                                            \
 }
 
-ROW_FUNC(YuvToRgbRow,      3, 0, 1, 2, 0)
-ROW_FUNC(YuvToRgbaRow,     4, 0, 1, 2, 3)
-ROW_FUNC(YuvToBgrRow,      3, 2, 1, 0, 0)
-ROW_FUNC(YuvToBgraRow,     4, 2, 1, 0, 3)
+ROW_FUNC(YuvToRgbRow_MIPS32,      3, 0, 1, 2, 0)
+ROW_FUNC(YuvToRgbaRow_MIPS32,     4, 0, 1, 2, 3)
+ROW_FUNC(YuvToBgrRow_MIPS32,      3, 2, 1, 0, 0)
+ROW_FUNC(YuvToBgraRow_MIPS32,     4, 2, 1, 0, 3)
 
 #undef ROW_FUNC
 
@@ -90,10 +90,10 @@ ROW_FUNC(YuvToBgraRow,     4, 2, 1, 0, 3)
 extern void WebPInitSamplersMIPS32(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplersMIPS32(void) {
-  WebPSamplers[MODE_RGB]  = YuvToRgbRow;
-  WebPSamplers[MODE_RGBA] = YuvToRgbaRow;
-  WebPSamplers[MODE_BGR]  = YuvToBgrRow;
-  WebPSamplers[MODE_BGRA] = YuvToBgraRow;
+  WebPSamplers[MODE_RGB]  = YuvToRgbRow_MIPS32;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRow_MIPS32;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRow_MIPS32;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRow_MIPS32;
 }
 
 #else  // !WEBP_USE_MIPS32
diff --git a/src/3rdparty/libwebp/src/dsp/yuv_mips_dsp_r2.c b/src/3rdparty/libwebp/src/dsp/yuv_mips_dsp_r2.c
index 1720d41..cc8afcc 100644
--- a/src/3rdparty/libwebp/src/dsp/yuv_mips_dsp_r2.c
+++ b/src/3rdparty/libwebp/src/dsp/yuv_mips_dsp_r2.c
@@ -12,11 +12,11 @@
 // Author(s):  Branimir Vasic (branimir.vasic@imgtec.com)
 //             Djordje Pesut  (djordje.pesut@imgtec.com)
 
-#include "./dsp.h"
+#include "src/dsp/dsp.h"
 
 #if defined(WEBP_USE_MIPS_DSP_R2)
 
-#include "./yuv.h"
+#include "src/dsp/yuv.h"
 
 //------------------------------------------------------------------------------
 // simple point-sampling
@@ -105,10 +105,10 @@ static void FUNC_NAME(const uint8_t* y,                                        \
   }                                                                            \
 }
 
-ROW_FUNC(YuvToRgbRow,      3, 0, 1, 2, 0)
-ROW_FUNC(YuvToRgbaRow,     4, 0, 1, 2, 3)
-ROW_FUNC(YuvToBgrRow,      3, 2, 1, 0, 0)
-ROW_FUNC(YuvToBgraRow,     4, 2, 1, 0, 3)
+ROW_FUNC(YuvToRgbRow_MIPSdspR2,      3, 0, 1, 2, 0)
+ROW_FUNC(YuvToRgbaRow_MIPSdspR2,     4, 0, 1, 2, 3)
+ROW_FUNC(YuvToBgrRow_MIPSdspR2,      3, 2, 1, 0, 0)
+ROW_FUNC(YuvToBgraRow_MIPSdspR2,     4, 2, 1, 0, 3)
 
 #undef ROW_FUNC
 #undef ASM_CLOBBER_LIST
@@ -121,10 +121,10 @@ ROW_FUNC(YuvToBgraRow,     4, 2, 1, 0, 3)
 extern void WebPInitSamplersMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplersMIPSdspR2(void) {
-  WebPSamplers[MODE_RGB]  = YuvToRgbRow;
-  WebPSamplers[MODE_RGBA] = YuvToRgbaRow;
-  WebPSamplers[MODE_BGR]  = YuvToBgrRow;
-  WebPSamplers[MODE_BGRA] = YuvToBgraRow;
+  WebPSamplers[MODE_RGB]  = YuvToRgbRow_MIPSdspR2;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRow_MIPSdspR2;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRow_MIPSdspR2;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRow_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2
diff --git a/src/3rdparty/libwebp/src/dsp/yuv_neon.c b/src/3rdparty/libwebp/src/dsp/yuv_neon.c
new file mode 100644
index 0000000..a34d602
--- /dev/null
+++ b/src/3rdparty/libwebp/src/dsp/yuv_neon.c
@@ -0,0 +1,288 @@
+// Copyright 2017 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// YUV->RGB conversion functions
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "src/dsp/yuv.h"
+
+#if defined(WEBP_USE_NEON)
+
+#include <assert.h>
+#include <stdlib.h>
+
+#include "src/dsp/neon.h"
+
+//-----------------------------------------------------------------------------
+
+static uint8x8_t ConvertRGBToY_NEON(const uint8x8_t R,
+                                    const uint8x8_t G,
+                                    const uint8x8_t B) {
+  const uint16x8_t r = vmovl_u8(R);
+  const uint16x8_t g = vmovl_u8(G);
+  const uint16x8_t b = vmovl_u8(B);
+  const uint16x4_t r_lo = vget_low_u16(r);
+  const uint16x4_t r_hi = vget_high_u16(r);
+  const uint16x4_t g_lo = vget_low_u16(g);
+  const uint16x4_t g_hi = vget_high_u16(g);
+  const uint16x4_t b_lo = vget_low_u16(b);
+  const uint16x4_t b_hi = vget_high_u16(b);
+  const uint32x4_t tmp0_lo = vmull_n_u16(         r_lo, 16839u);
+  const uint32x4_t tmp0_hi = vmull_n_u16(         r_hi, 16839u);
+  const uint32x4_t tmp1_lo = vmlal_n_u16(tmp0_lo, g_lo, 33059u);
+  const uint32x4_t tmp1_hi = vmlal_n_u16(tmp0_hi, g_hi, 33059u);
+  const uint32x4_t tmp2_lo = vmlal_n_u16(tmp1_lo, b_lo, 6420u);
+  const uint32x4_t tmp2_hi = vmlal_n_u16(tmp1_hi, b_hi, 6420u);
+  const uint16x8_t Y1 = vcombine_u16(vrshrn_n_u32(tmp2_lo, 16),
+                                     vrshrn_n_u32(tmp2_hi, 16));
+  const uint16x8_t Y2 = vaddq_u16(Y1, vdupq_n_u16(16));
+  return vqmovn_u16(Y2);
+}
+
+static void ConvertRGB24ToY_NEON(const uint8_t* rgb, uint8_t* y, int width) {
+  int i;
+  for (i = 0; i + 8 <= width; i += 8, rgb += 3 * 8) {
+    const uint8x8x3_t RGB = vld3_u8(rgb);
+    const uint8x8_t Y = ConvertRGBToY_NEON(RGB.val[0], RGB.val[1], RGB.val[2]);
+    vst1_u8(y + i, Y);
+  }
+  for (; i < width; ++i, rgb += 3) {   // left-over
+    y[i] = VP8RGBToY(rgb[0], rgb[1], rgb[2], YUV_HALF);
+  }
+}
+
+static void ConvertBGR24ToY_NEON(const uint8_t* bgr, uint8_t* y, int width) {
+  int i;
+  for (i = 0; i + 8 <= width; i += 8, bgr += 3 * 8) {
+    const uint8x8x3_t BGR = vld3_u8(bgr);
+    const uint8x8_t Y = ConvertRGBToY_NEON(BGR.val[2], BGR.val[1], BGR.val[0]);
+    vst1_u8(y + i, Y);
+  }
+  for (; i < width; ++i, bgr += 3) {  // left-over
+    y[i] = VP8RGBToY(bgr[2], bgr[1], bgr[0], YUV_HALF);
+  }
+}
+
+static void ConvertARGBToY_NEON(const uint32_t* argb, uint8_t* y, int width) {
+  int i;
+  for (i = 0; i + 8 <= width; i += 8) {
+    const uint8x8x4_t RGB = vld4_u8((const uint8_t*)&argb[i]);
+    const uint8x8_t Y = ConvertRGBToY_NEON(RGB.val[2], RGB.val[1], RGB.val[0]);
+    vst1_u8(y + i, Y);
+  }
+  for (; i < width; ++i) {   // left-over
+    const uint32_t p = argb[i];
+    y[i] = VP8RGBToY((p >> 16) & 0xff, (p >> 8) & 0xff, (p >>  0) & 0xff,
+                     YUV_HALF);
+  }
+}
+
+//-----------------------------------------------------------------------------
+
+// computes: DST_s16 = [(C0 * r + C1 * g + C2 * b) >> 16] + CST
+#define MULTIPLY_16b_PREAMBLE(r, g, b)                           \
+  const int16x4_t r_lo = vreinterpret_s16_u16(vget_low_u16(r));  \
+  const int16x4_t r_hi = vreinterpret_s16_u16(vget_high_u16(r)); \
+  const int16x4_t g_lo = vreinterpret_s16_u16(vget_low_u16(g));  \
+  const int16x4_t g_hi = vreinterpret_s16_u16(vget_high_u16(g)); \
+  const int16x4_t b_lo = vreinterpret_s16_u16(vget_low_u16(b));  \
+  const int16x4_t b_hi = vreinterpret_s16_u16(vget_high_u16(b))
+
+#define MULTIPLY_16b(C0, C1, C2, CST, DST_s16) do {              \
+  const int32x4_t tmp0_lo = vmull_n_s16(         r_lo, C0);      \
+  const int32x4_t tmp0_hi = vmull_n_s16(         r_hi, C0);      \
+  const int32x4_t tmp1_lo = vmlal_n_s16(tmp0_lo, g_lo, C1);      \
+  const int32x4_t tmp1_hi = vmlal_n_s16(tmp0_hi, g_hi, C1);      \
+  const int32x4_t tmp2_lo = vmlal_n_s16(tmp1_lo, b_lo, C2);      \
+  const int32x4_t tmp2_hi = vmlal_n_s16(tmp1_hi, b_hi, C2);      \
+  const int16x8_t tmp3 = vcombine_s16(vshrn_n_s32(tmp2_lo, 16),  \
+                                      vshrn_n_s32(tmp2_hi, 16)); \
+  DST_s16 = vaddq_s16(tmp3, vdupq_n_s16(CST));                   \
+} while (0)
+
+// This needs to be a macro, since (128 << SHIFT) needs to be an immediate.
+#define CONVERT_RGB_TO_UV(r, g, b, SHIFT, U_DST, V_DST) do {     \
+  MULTIPLY_16b_PREAMBLE(r, g, b);                                \
+  MULTIPLY_16b(-9719, -19081, 28800, 128 << SHIFT, U_DST);       \
+  MULTIPLY_16b(28800, -24116, -4684, 128 << SHIFT, V_DST);       \
+} while (0)
+
+static void ConvertRGBA32ToUV_NEON(const uint16_t* rgb,
+                                   uint8_t* u, uint8_t* v, int width) {
+  int i;
+  for (i = 0; i + 8 <= width; i += 8, rgb += 4 * 8) {
+    const uint16x8x4_t RGB = vld4q_u16((const uint16_t*)rgb);
+    int16x8_t U, V;
+    CONVERT_RGB_TO_UV(RGB.val[0], RGB.val[1], RGB.val[2], 2, U, V);
+    vst1_u8(u + i, vqrshrun_n_s16(U, 2));
+    vst1_u8(v + i, vqrshrun_n_s16(V, 2));
+  }
+  for (; i < width; i += 1, rgb += 4) {
+    const int r = rgb[0], g = rgb[1], b = rgb[2];
+    u[i] = VP8RGBToU(r, g, b, YUV_HALF << 2);
+    v[i] = VP8RGBToV(r, g, b, YUV_HALF << 2);
+  }
+}
+
+static void ConvertARGBToUV_NEON(const uint32_t* argb, uint8_t* u, uint8_t* v,
+                                 int src_width, int do_store) {
+  int i;
+  for (i = 0; i + 16 <= src_width; i += 16, u += 8, v += 8) {
+    const uint8x16x4_t RGB = vld4q_u8((const uint8_t*)&argb[i]);
+    const uint16x8_t R = vpaddlq_u8(RGB.val[2]);  // pair-wise adds
+    const uint16x8_t G = vpaddlq_u8(RGB.val[1]);
+    const uint16x8_t B = vpaddlq_u8(RGB.val[0]);
+    int16x8_t U_tmp, V_tmp;
+    CONVERT_RGB_TO_UV(R, G, B, 1, U_tmp, V_tmp);
+    {
+      const uint8x8_t U = vqrshrun_n_s16(U_tmp, 1);
+      const uint8x8_t V = vqrshrun_n_s16(V_tmp, 1);
+      if (do_store) {
+        vst1_u8(u, U);
+        vst1_u8(v, V);
+      } else {
+        const uint8x8_t prev_u = vld1_u8(u);
+        const uint8x8_t prev_v = vld1_u8(v);
+        vst1_u8(u, vrhadd_u8(U, prev_u));
+        vst1_u8(v, vrhadd_u8(V, prev_v));
+      }
+    }
+  }
+  if (i < src_width) {  // left-over
+    WebPConvertARGBToUV_C(argb + i, u, v, src_width - i, do_store);
+  }
+}
+
+
+//------------------------------------------------------------------------------
+
+extern void WebPInitConvertARGBToYUVNEON(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUVNEON(void) {
+  WebPConvertRGB24ToY = ConvertRGB24ToY_NEON;
+  WebPConvertBGR24ToY = ConvertBGR24ToY_NEON;
+  WebPConvertARGBToY = ConvertARGBToY_NEON;
+  WebPConvertARGBToUV = ConvertARGBToUV_NEON;
+  WebPConvertRGBA32ToUV = ConvertRGBA32ToUV_NEON;
+}
+
+//------------------------------------------------------------------------------
+
+#define MAX_Y ((1 << 10) - 1)    // 10b precision over 16b-arithmetic
+static uint16_t clip_y_NEON(int v) {
+  return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
+}
+
+static uint64_t SharpYUVUpdateY_NEON(const uint16_t* ref, const uint16_t* src,
+                                     uint16_t* dst, int len) {
+  int i;
+  const int16x8_t zero = vdupq_n_s16(0);
+  const int16x8_t max = vdupq_n_s16(MAX_Y);
+  uint64x2_t sum = vdupq_n_u64(0);
+  uint64_t diff;
+
+  for (i = 0; i + 8 <= len; i += 8) {
+    const int16x8_t A = vreinterpretq_s16_u16(vld1q_u16(ref + i));
+    const int16x8_t B = vreinterpretq_s16_u16(vld1q_u16(src + i));
+    const int16x8_t C = vreinterpretq_s16_u16(vld1q_u16(dst + i));
+    const int16x8_t D = vsubq_s16(A, B);       // diff_y
+    const int16x8_t F = vaddq_s16(C, D);       // new_y
+    const uint16x8_t H =
+        vreinterpretq_u16_s16(vmaxq_s16(vminq_s16(F, max), zero));
+    const int16x8_t I = vabsq_s16(D);          // abs(diff_y)
+    vst1q_u16(dst + i, H);
+    sum = vpadalq_u32(sum, vpaddlq_u16(vreinterpretq_u16_s16(I)));
+  }
+  diff = vgetq_lane_u64(sum, 0) + vgetq_lane_u64(sum, 1);
+  for (; i < len; ++i) {
+    const int diff_y = ref[i] - src[i];
+    const int new_y = (int)(dst[i]) + diff_y;
+    dst[i] = clip_y_NEON(new_y);
+    diff += (uint64_t)(abs(diff_y));
+  }
+  return diff;
+}
+
+static void SharpYUVUpdateRGB_NEON(const int16_t* ref, const int16_t* src,
+                                   int16_t* dst, int len) {
+  int i;
+  for (i = 0; i + 8 <= len; i += 8) {
+    const int16x8_t A = vld1q_s16(ref + i);
+    const int16x8_t B = vld1q_s16(src + i);
+    const int16x8_t C = vld1q_s16(dst + i);
+    const int16x8_t D = vsubq_s16(A, B);   // diff_uv
+    const int16x8_t E = vaddq_s16(C, D);   // new_uv
+    vst1q_s16(dst + i, E);
+  }
+  for (; i < len; ++i) {
+    const int diff_uv = ref[i] - src[i];
+    dst[i] += diff_uv;
+  }
+}
+
+static void SharpYUVFilterRow_NEON(const int16_t* A, const int16_t* B, int len,
+                                   const uint16_t* best_y, uint16_t* out) {
+  int i;
+  const int16x8_t max = vdupq_n_s16(MAX_Y);
+  const int16x8_t zero = vdupq_n_s16(0);
+  for (i = 0; i + 8 <= len; i += 8) {
+    const int16x8_t a0 = vld1q_s16(A + i + 0);
+    const int16x8_t a1 = vld1q_s16(A + i + 1);
+    const int16x8_t b0 = vld1q_s16(B + i + 0);
+    const int16x8_t b1 = vld1q_s16(B + i + 1);
+    const int16x8_t a0b1 = vaddq_s16(a0, b1);
+    const int16x8_t a1b0 = vaddq_s16(a1, b0);
+    const int16x8_t a0a1b0b1 = vaddq_s16(a0b1, a1b0);  // A0+A1+B0+B1
+    const int16x8_t a0b1_2 = vaddq_s16(a0b1, a0b1);    // 2*(A0+B1)
+    const int16x8_t a1b0_2 = vaddq_s16(a1b0, a1b0);    // 2*(A1+B0)
+    const int16x8_t c0 = vshrq_n_s16(vaddq_s16(a0b1_2, a0a1b0b1), 3);
+    const int16x8_t c1 = vshrq_n_s16(vaddq_s16(a1b0_2, a0a1b0b1), 3);
+    const int16x8_t d0 = vaddq_s16(c1, a0);
+    const int16x8_t d1 = vaddq_s16(c0, a1);
+    const int16x8_t e0 = vrshrq_n_s16(d0, 1);
+    const int16x8_t e1 = vrshrq_n_s16(d1, 1);
+    const int16x8x2_t f = vzipq_s16(e0, e1);
+    const int16x8_t g0 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 0));
+    const int16x8_t g1 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 8));
+    const int16x8_t h0 = vaddq_s16(g0, f.val[0]);
+    const int16x8_t h1 = vaddq_s16(g1, f.val[1]);
+    const int16x8_t i0 = vmaxq_s16(vminq_s16(h0, max), zero);
+    const int16x8_t i1 = vmaxq_s16(vminq_s16(h1, max), zero);
+    vst1q_u16(out + 2 * i + 0, vreinterpretq_u16_s16(i0));
+    vst1q_u16(out + 2 * i + 8, vreinterpretq_u16_s16(i1));
+  }
+  for (; i < len; ++i) {
+    const int a0b1 = A[i + 0] + B[i + 1];
+    const int a1b0 = A[i + 1] + B[i + 0];
+    const int a0a1b0b1 = a0b1 + a1b0 + 8;
+    const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
+    const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
+    out[2 * i + 0] = clip_y_NEON(best_y[2 * i + 0] + v0);
+    out[2 * i + 1] = clip_y_NEON(best_y[2 * i + 1] + v1);
+  }
+}
+#undef MAX_Y
+
+//------------------------------------------------------------------------------
+
+extern void WebPInitSharpYUVNEON(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitSharpYUVNEON(void) {
+  WebPSharpYUVUpdateY = SharpYUVUpdateY_NEON;
+  WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_NEON;
+  WebPSharpYUVFilterRow = SharpYUVFilterRow_NEON;
+}
+
+#else  // !WEBP_USE_NEON
+
+WEBP_DSP_INIT_STUB(WebPInitConvertARGBToYUVNEON)
+WEBP_DSP_INIT_STUB(WebPInitSharpYUVNEON)
+
+#endif  // WEBP_USE_NEON
diff --git a/src/3rdparty/libwebp/src/dsp/yuv_sse2.c b/src/3rdparty/libwebp/src/dsp/yuv_sse2.c
index e33c2bb..6810bf8 100644
--- a/src/3rdparty/libwebp/src/dsp/yuv_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/yuv_sse2.c
@@ -11,11 +11,11 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./yuv.h"
+#include "src/dsp/yuv.h"
 
 #if defined(WEBP_USE_SSE2)
 
-#include "./common_sse2.h"
+#include "src/dsp/common_sse2.h"
 #include <stdlib.h>
 #include <emmintrin.h>
 
@@ -26,12 +26,12 @@
 // R = (19077 * y             + 26149 * v - 14234) >> 6
 // G = (19077 * y -  6419 * u - 13320 * v +  8708) >> 6
 // B = (19077 * y + 33050 * u             - 17685) >> 6
-static void ConvertYUV444ToRGB(const __m128i* const Y0,
-                               const __m128i* const U0,
-                               const __m128i* const V0,
-                               __m128i* const R,
-                               __m128i* const G,
-                               __m128i* const B) {
+static void ConvertYUV444ToRGB_SSE2(const __m128i* const Y0,
+                                    const __m128i* const U0,
+                                    const __m128i* const V0,
+                                    __m128i* const R,
+                                    __m128i* const G,
+                                    __m128i* const B) {
   const __m128i k19077 = _mm_set1_epi16(19077);
   const __m128i k26149 = _mm_set1_epi16(26149);
   const __m128i k14234 = _mm_set1_epi16(14234);
@@ -66,13 +66,13 @@ static void ConvertYUV444ToRGB(const __m128i* const Y0,
 }
 
 // Load the bytes into the *upper* part of 16b words. That's "<< 8", basically.
-static WEBP_INLINE __m128i Load_HI_16(const uint8_t* src) {
+static WEBP_INLINE __m128i Load_HI_16_SSE2(const uint8_t* src) {
   const __m128i zero = _mm_setzero_si128();
   return _mm_unpacklo_epi8(zero, _mm_loadl_epi64((const __m128i*)src));
 }
 
 // Load and replicate the U/V samples
-static WEBP_INLINE __m128i Load_UV_HI_8(const uint8_t* src) {
+static WEBP_INLINE __m128i Load_UV_HI_8_SSE2(const uint8_t* src) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i tmp0 = _mm_cvtsi32_si128(*(const uint32_t*)src);
   const __m128i tmp1 = _mm_unpacklo_epi8(zero, tmp0);
@@ -80,29 +80,33 @@ static WEBP_INLINE __m128i Load_UV_HI_8(const uint8_t* src) {
 }
 
 // Convert 32 samples of YUV444 to R/G/B
-static void YUV444ToRGB(const uint8_t* const y,
-                        const uint8_t* const u,
-                        const uint8_t* const v,
-                        __m128i* const R, __m128i* const G, __m128i* const B) {
-  const __m128i Y0 = Load_HI_16(y), U0 = Load_HI_16(u), V0 = Load_HI_16(v);
-  ConvertYUV444ToRGB(&Y0, &U0, &V0, R, G, B);
+static void YUV444ToRGB_SSE2(const uint8_t* const y,
+                             const uint8_t* const u,
+                             const uint8_t* const v,
+                             __m128i* const R, __m128i* const G,
+                             __m128i* const B) {
+  const __m128i Y0 = Load_HI_16_SSE2(y), U0 = Load_HI_16_SSE2(u),
+                V0 = Load_HI_16_SSE2(v);
+  ConvertYUV444ToRGB_SSE2(&Y0, &U0, &V0, R, G, B);
 }
 
 // Convert 32 samples of YUV420 to R/G/B
-static void YUV420ToRGB(const uint8_t* const y,
-                        const uint8_t* const u,
-                        const uint8_t* const v,
-                        __m128i* const R, __m128i* const G, __m128i* const B) {
-  const __m128i Y0 = Load_HI_16(y), U0 = Load_UV_HI_8(u), V0 = Load_UV_HI_8(v);
-  ConvertYUV444ToRGB(&Y0, &U0, &V0, R, G, B);
+static void YUV420ToRGB_SSE2(const uint8_t* const y,
+                             const uint8_t* const u,
+                             const uint8_t* const v,
+                             __m128i* const R, __m128i* const G,
+                             __m128i* const B) {
+  const __m128i Y0 = Load_HI_16_SSE2(y), U0 = Load_UV_HI_8_SSE2(u),
+                V0 = Load_UV_HI_8_SSE2(v);
+  ConvertYUV444ToRGB_SSE2(&Y0, &U0, &V0, R, G, B);
 }
 
 // Pack R/G/B/A results into 32b output.
-static WEBP_INLINE void PackAndStore4(const __m128i* const R,
-                                      const __m128i* const G,
-                                      const __m128i* const B,
-                                      const __m128i* const A,
-                                      uint8_t* const dst) {
+static WEBP_INLINE void PackAndStore4_SSE2(const __m128i* const R,
+                                           const __m128i* const G,
+                                           const __m128i* const B,
+                                           const __m128i* const A,
+                                           uint8_t* const dst) {
   const __m128i rb = _mm_packus_epi16(*R, *B);
   const __m128i ga = _mm_packus_epi16(*G, *A);
   const __m128i rg = _mm_unpacklo_epi8(rb, ga);
@@ -114,12 +118,12 @@ static WEBP_INLINE void PackAndStore4(const __m128i* const R,
 }
 
 // Pack R/G/B/A results into 16b output.
-static WEBP_INLINE void PackAndStore4444(const __m128i* const R,
-                                         const __m128i* const G,
-                                         const __m128i* const B,
-                                         const __m128i* const A,
-                                         uint8_t* const dst) {
-#if !defined(WEBP_SWAP_16BIT_CSP)
+static WEBP_INLINE void PackAndStore4444_SSE2(const __m128i* const R,
+                                              const __m128i* const G,
+                                              const __m128i* const B,
+                                              const __m128i* const A,
+                                              uint8_t* const dst) {
+#if (WEBP_SWAP_16BIT_CSP == 0)
   const __m128i rg0 = _mm_packus_epi16(*R, *G);
   const __m128i ba0 = _mm_packus_epi16(*B, *A);
 #else
@@ -136,10 +140,10 @@ static WEBP_INLINE void PackAndStore4444(const __m128i* const R,
 }
 
 // Pack R/G/B results into 16b output.
-static WEBP_INLINE void PackAndStore565(const __m128i* const R,
-                                        const __m128i* const G,
-                                        const __m128i* const B,
-                                        uint8_t* const dst) {
+static WEBP_INLINE void PackAndStore565_SSE2(const __m128i* const R,
+                                             const __m128i* const G,
+                                             const __m128i* const B,
+                                             uint8_t* const dst) {
   const __m128i r0 = _mm_packus_epi16(*R, *R);
   const __m128i g0 = _mm_packus_epi16(*G, *G);
   const __m128i b0 = _mm_packus_epi16(*B, *B);
@@ -149,7 +153,7 @@ static WEBP_INLINE void PackAndStore565(const __m128i* const R,
   const __m128i g2 = _mm_slli_epi16(_mm_and_si128(g0, _mm_set1_epi8(0x1c)), 3);
   const __m128i rg = _mm_or_si128(r1, g1);
   const __m128i gb = _mm_or_si128(g2, b1);
-#if !defined(WEBP_SWAP_16BIT_CSP)
+#if (WEBP_SWAP_16BIT_CSP == 0)
   const __m128i rgb565 = _mm_unpacklo_epi8(rg, gb);
 #else
   const __m128i rgb565 = _mm_unpacklo_epi8(gb, rg);
@@ -160,10 +164,10 @@ static WEBP_INLINE void PackAndStore565(const __m128i* const R,
 // Pack the planar buffers
 // rrrr... rrrr... gggg... gggg... bbbb... bbbb....
 // triplet by triplet in the output buffer rgb as rgbrgbrgbrgb ...
-static WEBP_INLINE void PlanarTo24b(__m128i* const in0, __m128i* const in1,
-                                    __m128i* const in2, __m128i* const in3,
-                                    __m128i* const in4, __m128i* const in5,
-                                    uint8_t* const rgb) {
+static WEBP_INLINE void PlanarTo24b_SSE2(__m128i* const in0, __m128i* const in1,
+                                         __m128i* const in2, __m128i* const in3,
+                                         __m128i* const in4, __m128i* const in5,
+                                         uint8_t* const rgb) {
   // The input is 6 registers of sixteen 8b but for the sake of explanation,
   // let's take 6 registers of four 8b values.
   // To pack, we will keep taking one every two 8b integer and move it
@@ -186,69 +190,69 @@ static WEBP_INLINE void PlanarTo24b(__m128i* const in0, __m128i* const in1,
   _mm_storeu_si128((__m128i*)(rgb + 80), *in5);
 }
 
-void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                    uint8_t* dst) {
+void VP8YuvToRgba32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                         uint8_t* dst) {
   const __m128i kAlpha = _mm_set1_epi16(255);
   int n;
   for (n = 0; n < 32; n += 8, dst += 32) {
     __m128i R, G, B;
-    YUV444ToRGB(y + n, u + n, v + n, &R, &G, &B);
-    PackAndStore4(&R, &G, &B, &kAlpha, dst);
+    YUV444ToRGB_SSE2(y + n, u + n, v + n, &R, &G, &B);
+    PackAndStore4_SSE2(&R, &G, &B, &kAlpha, dst);
   }
 }
 
-void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                    uint8_t* dst) {
+void VP8YuvToBgra32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                         uint8_t* dst) {
   const __m128i kAlpha = _mm_set1_epi16(255);
   int n;
   for (n = 0; n < 32; n += 8, dst += 32) {
     __m128i R, G, B;
-    YUV444ToRGB(y + n, u + n, v + n, &R, &G, &B);
-    PackAndStore4(&B, &G, &R, &kAlpha, dst);
+    YUV444ToRGB_SSE2(y + n, u + n, v + n, &R, &G, &B);
+    PackAndStore4_SSE2(&B, &G, &R, &kAlpha, dst);
   }
 }
 
-void VP8YuvToArgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                    uint8_t* dst) {
+void VP8YuvToArgb32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                         uint8_t* dst) {
   const __m128i kAlpha = _mm_set1_epi16(255);
   int n;
   for (n = 0; n < 32; n += 8, dst += 32) {
     __m128i R, G, B;
-    YUV444ToRGB(y + n, u + n, v + n, &R, &G, &B);
-    PackAndStore4(&kAlpha, &R, &G, &B, dst);
+    YUV444ToRGB_SSE2(y + n, u + n, v + n, &R, &G, &B);
+    PackAndStore4_SSE2(&kAlpha, &R, &G, &B, dst);
   }
 }
 
-void VP8YuvToRgba444432(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                        uint8_t* dst) {
+void VP8YuvToRgba444432_SSE2(const uint8_t* y, const uint8_t* u,
+                             const uint8_t* v, uint8_t* dst) {
   const __m128i kAlpha = _mm_set1_epi16(255);
   int n;
   for (n = 0; n < 32; n += 8, dst += 16) {
     __m128i R, G, B;
-    YUV444ToRGB(y + n, u + n, v + n, &R, &G, &B);
-    PackAndStore4444(&R, &G, &B, &kAlpha, dst);
+    YUV444ToRGB_SSE2(y + n, u + n, v + n, &R, &G, &B);
+    PackAndStore4444_SSE2(&R, &G, &B, &kAlpha, dst);
   }
 }
 
-void VP8YuvToRgb56532(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                      uint8_t* dst) {
+void VP8YuvToRgb56532_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                           uint8_t* dst) {
   int n;
   for (n = 0; n < 32; n += 8, dst += 16) {
     __m128i R, G, B;
-    YUV444ToRGB(y + n, u + n, v + n, &R, &G, &B);
-    PackAndStore565(&R, &G, &B, dst);
+    YUV444ToRGB_SSE2(y + n, u + n, v + n, &R, &G, &B);
+    PackAndStore565_SSE2(&R, &G, &B, dst);
   }
 }
 
-void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                   uint8_t* dst) {
+void VP8YuvToRgb32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                        uint8_t* dst) {
   __m128i R0, R1, R2, R3, G0, G1, G2, G3, B0, B1, B2, B3;
   __m128i rgb0, rgb1, rgb2, rgb3, rgb4, rgb5;
 
-  YUV444ToRGB(y + 0, u + 0, v + 0, &R0, &G0, &B0);
-  YUV444ToRGB(y + 8, u + 8, v + 8, &R1, &G1, &B1);
-  YUV444ToRGB(y + 16, u + 16, v + 16, &R2, &G2, &B2);
-  YUV444ToRGB(y + 24, u + 24, v + 24, &R3, &G3, &B3);
+  YUV444ToRGB_SSE2(y + 0, u + 0, v + 0, &R0, &G0, &B0);
+  YUV444ToRGB_SSE2(y + 8, u + 8, v + 8, &R1, &G1, &B1);
+  YUV444ToRGB_SSE2(y + 16, u + 16, v + 16, &R2, &G2, &B2);
+  YUV444ToRGB_SSE2(y + 24, u + 24, v + 24, &R3, &G3, &B3);
 
   // Cast to 8b and store as RRRRGGGGBBBB.
   rgb0 = _mm_packus_epi16(R0, R1);
@@ -259,18 +263,18 @@ void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
   rgb5 = _mm_packus_epi16(B2, B3);
 
   // Pack as RGBRGBRGBRGB.
-  PlanarTo24b(&rgb0, &rgb1, &rgb2, &rgb3, &rgb4, &rgb5, dst);
+  PlanarTo24b_SSE2(&rgb0, &rgb1, &rgb2, &rgb3, &rgb4, &rgb5, dst);
 }
 
-void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                   uint8_t* dst) {
+void VP8YuvToBgr32_SSE2(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                        uint8_t* dst) {
   __m128i R0, R1, R2, R3, G0, G1, G2, G3, B0, B1, B2, B3;
   __m128i bgr0, bgr1, bgr2, bgr3, bgr4, bgr5;
 
-  YUV444ToRGB(y +  0, u +  0, v +  0, &R0, &G0, &B0);
-  YUV444ToRGB(y +  8, u +  8, v +  8, &R1, &G1, &B1);
-  YUV444ToRGB(y + 16, u + 16, v + 16, &R2, &G2, &B2);
-  YUV444ToRGB(y + 24, u + 24, v + 24, &R3, &G3, &B3);
+  YUV444ToRGB_SSE2(y +  0, u +  0, v +  0, &R0, &G0, &B0);
+  YUV444ToRGB_SSE2(y +  8, u +  8, v +  8, &R1, &G1, &B1);
+  YUV444ToRGB_SSE2(y + 16, u + 16, v + 16, &R2, &G2, &B2);
+  YUV444ToRGB_SSE2(y + 24, u + 24, v + 24, &R3, &G3, &B3);
 
   // Cast to 8b and store as BBBBGGGGRRRR.
   bgr0 = _mm_packus_epi16(B0, B1);
@@ -281,20 +285,21 @@ void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
   bgr5= _mm_packus_epi16(R2, R3);
 
   // Pack as BGRBGRBGRBGR.
-  PlanarTo24b(&bgr0, &bgr1, &bgr2, &bgr3, &bgr4, &bgr5, dst);
+  PlanarTo24b_SSE2(&bgr0, &bgr1, &bgr2, &bgr3, &bgr4, &bgr5, dst);
 }
 
 //-----------------------------------------------------------------------------
 // Arbitrary-length row conversion functions
 
-static void YuvToRgbaRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                         uint8_t* dst, int len) {
+static void YuvToRgbaRow_SSE2(const uint8_t* y,
+                              const uint8_t* u, const uint8_t* v,
+                              uint8_t* dst, int len) {
   const __m128i kAlpha = _mm_set1_epi16(255);
   int n;
   for (n = 0; n + 8 <= len; n += 8, dst += 32) {
     __m128i R, G, B;
-    YUV420ToRGB(y, u, v, &R, &G, &B);
-    PackAndStore4(&R, &G, &B, &kAlpha, dst);
+    YUV420ToRGB_SSE2(y, u, v, &R, &G, &B);
+    PackAndStore4_SSE2(&R, &G, &B, &kAlpha, dst);
     y += 8;
     u += 4;
     v += 4;
@@ -308,14 +313,15 @@ static void YuvToRgbaRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
   }
 }
 
-static void YuvToBgraRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                         uint8_t* dst, int len) {
+static void YuvToBgraRow_SSE2(const uint8_t* y,
+                              const uint8_t* u, const uint8_t* v,
+                              uint8_t* dst, int len) {
   const __m128i kAlpha = _mm_set1_epi16(255);
   int n;
   for (n = 0; n + 8 <= len; n += 8, dst += 32) {
     __m128i R, G, B;
-    YUV420ToRGB(y, u, v, &R, &G, &B);
-    PackAndStore4(&B, &G, &R, &kAlpha, dst);
+    YUV420ToRGB_SSE2(y, u, v, &R, &G, &B);
+    PackAndStore4_SSE2(&B, &G, &R, &kAlpha, dst);
     y += 8;
     u += 4;
     v += 4;
@@ -329,14 +335,15 @@ static void YuvToBgraRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
   }
 }
 
-static void YuvToArgbRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                         uint8_t* dst, int len) {
+static void YuvToArgbRow_SSE2(const uint8_t* y,
+                              const uint8_t* u, const uint8_t* v,
+                              uint8_t* dst, int len) {
   const __m128i kAlpha = _mm_set1_epi16(255);
   int n;
   for (n = 0; n + 8 <= len; n += 8, dst += 32) {
     __m128i R, G, B;
-    YUV420ToRGB(y, u, v, &R, &G, &B);
-    PackAndStore4(&kAlpha, &R, &G, &B, dst);
+    YUV420ToRGB_SSE2(y, u, v, &R, &G, &B);
+    PackAndStore4_SSE2(&kAlpha, &R, &G, &B, dst);
     y += 8;
     u += 4;
     v += 4;
@@ -350,17 +357,18 @@ static void YuvToArgbRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
   }
 }
 
-static void YuvToRgbRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                        uint8_t* dst, int len) {
+static void YuvToRgbRow_SSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
   int n;
   for (n = 0; n + 32 <= len; n += 32, dst += 32 * 3) {
     __m128i R0, R1, R2, R3, G0, G1, G2, G3, B0, B1, B2, B3;
     __m128i rgb0, rgb1, rgb2, rgb3, rgb4, rgb5;
 
-    YUV420ToRGB(y +  0, u +  0, v +  0, &R0, &G0, &B0);
-    YUV420ToRGB(y +  8, u +  4, v +  4, &R1, &G1, &B1);
-    YUV420ToRGB(y + 16, u +  8, v +  8, &R2, &G2, &B2);
-    YUV420ToRGB(y + 24, u + 12, v + 12, &R3, &G3, &B3);
+    YUV420ToRGB_SSE2(y +  0, u +  0, v +  0, &R0, &G0, &B0);
+    YUV420ToRGB_SSE2(y +  8, u +  4, v +  4, &R1, &G1, &B1);
+    YUV420ToRGB_SSE2(y + 16, u +  8, v +  8, &R2, &G2, &B2);
+    YUV420ToRGB_SSE2(y + 24, u + 12, v + 12, &R3, &G3, &B3);
 
     // Cast to 8b and store as RRRRGGGGBBBB.
     rgb0 = _mm_packus_epi16(R0, R1);
@@ -371,7 +379,7 @@ static void YuvToRgbRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
     rgb5 = _mm_packus_epi16(B2, B3);
 
     // Pack as RGBRGBRGBRGB.
-    PlanarTo24b(&rgb0, &rgb1, &rgb2, &rgb3, &rgb4, &rgb5, dst);
+    PlanarTo24b_SSE2(&rgb0, &rgb1, &rgb2, &rgb3, &rgb4, &rgb5, dst);
 
     y += 32;
     u += 16;
@@ -386,17 +394,18 @@ static void YuvToRgbRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
   }
 }
 
-static void YuvToBgrRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
-                        uint8_t* dst, int len) {
+static void YuvToBgrRow_SSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
   int n;
   for (n = 0; n + 32 <= len; n += 32, dst += 32 * 3) {
     __m128i R0, R1, R2, R3, G0, G1, G2, G3, B0, B1, B2, B3;
     __m128i bgr0, bgr1, bgr2, bgr3, bgr4, bgr5;
 
-    YUV420ToRGB(y +  0, u +  0, v +  0, &R0, &G0, &B0);
-    YUV420ToRGB(y +  8, u +  4, v +  4, &R1, &G1, &B1);
-    YUV420ToRGB(y + 16, u +  8, v +  8, &R2, &G2, &B2);
-    YUV420ToRGB(y + 24, u + 12, v + 12, &R3, &G3, &B3);
+    YUV420ToRGB_SSE2(y +  0, u +  0, v +  0, &R0, &G0, &B0);
+    YUV420ToRGB_SSE2(y +  8, u +  4, v +  4, &R1, &G1, &B1);
+    YUV420ToRGB_SSE2(y + 16, u +  8, v +  8, &R2, &G2, &B2);
+    YUV420ToRGB_SSE2(y + 24, u + 12, v + 12, &R3, &G3, &B3);
 
     // Cast to 8b and store as BBBBGGGGRRRR.
     bgr0 = _mm_packus_epi16(B0, B1);
@@ -407,7 +416,7 @@ static void YuvToBgrRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
     bgr5 = _mm_packus_epi16(R2, R3);
 
     // Pack as BGRBGRBGRBGR.
-    PlanarTo24b(&bgr0, &bgr1, &bgr2, &bgr3, &bgr4, &bgr5, dst);
+    PlanarTo24b_SSE2(&bgr0, &bgr1, &bgr2, &bgr3, &bgr4, &bgr5, dst);
 
     y += 32;
     u += 16;
@@ -428,11 +437,11 @@ static void YuvToBgrRow(const uint8_t* y, const uint8_t* u, const uint8_t* v,
 extern void WebPInitSamplersSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplersSSE2(void) {
-  WebPSamplers[MODE_RGB]  = YuvToRgbRow;
-  WebPSamplers[MODE_RGBA] = YuvToRgbaRow;
-  WebPSamplers[MODE_BGR]  = YuvToBgrRow;
-  WebPSamplers[MODE_BGRA] = YuvToBgraRow;
-  WebPSamplers[MODE_ARGB] = YuvToArgbRow;
+  WebPSamplers[MODE_RGB]  = YuvToRgbRow_SSE2;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRow_SSE2;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRow_SSE2;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRow_SSE2;
+  WebPSamplers[MODE_ARGB] = YuvToArgbRow_SSE2;
 }
 
 //------------------------------------------------------------------------------
@@ -445,7 +454,7 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplersSSE2(void) {
 
 // Function that inserts a value of the second half of the in buffer in between
 // every two char of the first half.
-static WEBP_INLINE void RGB24PackedToPlanarHelper(
+static WEBP_INLINE void RGB24PackedToPlanarHelper_SSE2(
     const __m128i* const in /*in[6]*/, __m128i* const out /*out[6]*/) {
   out[0] = _mm_unpacklo_epi8(in[0], in[3]);
   out[1] = _mm_unpackhi_epi8(in[0], in[3]);
@@ -458,8 +467,8 @@ static WEBP_INLINE void RGB24PackedToPlanarHelper(
 // Unpack the 8b input rgbrgbrgbrgb ... as contiguous registers:
 // rrrr... rrrr... gggg... gggg... bbbb... bbbb....
 // Similar to PlanarTo24bHelper(), but in reverse order.
-static WEBP_INLINE void RGB24PackedToPlanar(const uint8_t* const rgb,
-                                            __m128i* const out /*out[6]*/) {
+static WEBP_INLINE void RGB24PackedToPlanar_SSE2(
+    const uint8_t* const rgb, __m128i* const out /*out[6]*/) {
   __m128i tmp[6];
   tmp[0] = _mm_loadu_si128((const __m128i*)(rgb +  0));
   tmp[1] = _mm_loadu_si128((const __m128i*)(rgb + 16));
@@ -468,16 +477,16 @@ static WEBP_INLINE void RGB24PackedToPlanar(const uint8_t* const rgb,
   tmp[4] = _mm_loadu_si128((const __m128i*)(rgb + 64));
   tmp[5] = _mm_loadu_si128((const __m128i*)(rgb + 80));
 
-  RGB24PackedToPlanarHelper(tmp, out);
-  RGB24PackedToPlanarHelper(out, tmp);
-  RGB24PackedToPlanarHelper(tmp, out);
-  RGB24PackedToPlanarHelper(out, tmp);
-  RGB24PackedToPlanarHelper(tmp, out);
+  RGB24PackedToPlanarHelper_SSE2(tmp, out);
+  RGB24PackedToPlanarHelper_SSE2(out, tmp);
+  RGB24PackedToPlanarHelper_SSE2(tmp, out);
+  RGB24PackedToPlanarHelper_SSE2(out, tmp);
+  RGB24PackedToPlanarHelper_SSE2(tmp, out);
 }
 
 // Convert 8 packed ARGB to r[], g[], b[]
-static WEBP_INLINE void RGB32PackedToPlanar(const uint32_t* const argb,
-                                            __m128i* const rgb /*in[6]*/) {
+static WEBP_INLINE void RGB32PackedToPlanar_SSE2(const uint32_t* const argb,
+                                                 __m128i* const rgb /*in[6]*/) {
   const __m128i zero = _mm_setzero_si128();
   __m128i a0 = LOAD_16(argb + 0);
   __m128i a1 = LOAD_16(argb + 4);
@@ -511,10 +520,10 @@ static WEBP_INLINE void RGB32PackedToPlanar(const uint32_t* const argb,
 } while (0)
 
 #define MK_CST_16(A, B) _mm_set_epi16((B), (A), (B), (A), (B), (A), (B), (A))
-static WEBP_INLINE void ConvertRGBToY(const __m128i* const R,
-                                      const __m128i* const G,
-                                      const __m128i* const B,
-                                      __m128i* const Y) {
+static WEBP_INLINE void ConvertRGBToY_SSE2(const __m128i* const R,
+                                           const __m128i* const G,
+                                           const __m128i* const B,
+                                           __m128i* const Y) {
   const __m128i kRG_y = MK_CST_16(16839, 33059 - 16384);
   const __m128i kGB_y = MK_CST_16(16384, 6420);
   const __m128i kHALF_Y = _mm_set1_epi32((16 << YUV_FIX) + YUV_HALF);
@@ -526,10 +535,11 @@ static WEBP_INLINE void ConvertRGBToY(const __m128i* const R,
   TRANSFORM(RG_lo, RG_hi, GB_lo, GB_hi, kRG_y, kGB_y, kHALF_Y, YUV_FIX, *Y);
 }
 
-static WEBP_INLINE void ConvertRGBToUV(const __m128i* const R,
-                                       const __m128i* const G,
-                                       const __m128i* const B,
-                                       __m128i* const U, __m128i* const V) {
+static WEBP_INLINE void ConvertRGBToUV_SSE2(const __m128i* const R,
+                                            const __m128i* const G,
+                                            const __m128i* const B,
+                                            __m128i* const U,
+                                            __m128i* const V) {
   const __m128i kRG_u = MK_CST_16(-9719, -19081);
   const __m128i kGB_u = MK_CST_16(0, 28800);
   const __m128i kRG_v = MK_CST_16(28800, 0);
@@ -549,14 +559,14 @@ static WEBP_INLINE void ConvertRGBToUV(const __m128i* const R,
 #undef MK_CST_16
 #undef TRANSFORM
 
-static void ConvertRGB24ToY(const uint8_t* rgb, uint8_t* y, int width) {
+static void ConvertRGB24ToY_SSE2(const uint8_t* rgb, uint8_t* y, int width) {
   const int max_width = width & ~31;
   int i;
   for (i = 0; i < max_width; rgb += 3 * 16 * 2) {
     __m128i rgb_plane[6];
     int j;
 
-    RGB24PackedToPlanar(rgb, rgb_plane);
+    RGB24PackedToPlanar_SSE2(rgb, rgb_plane);
 
     for (j = 0; j < 2; ++j, i += 16) {
       const __m128i zero = _mm_setzero_si128();
@@ -566,13 +576,13 @@ static void ConvertRGB24ToY(const uint8_t* rgb, uint8_t* y, int width) {
       r = _mm_unpacklo_epi8(rgb_plane[0 + j], zero);
       g = _mm_unpacklo_epi8(rgb_plane[2 + j], zero);
       b = _mm_unpacklo_epi8(rgb_plane[4 + j], zero);
-      ConvertRGBToY(&r, &g, &b, &Y0);
+      ConvertRGBToY_SSE2(&r, &g, &b, &Y0);
 
       // Convert to 16-bit Y.
       r = _mm_unpackhi_epi8(rgb_plane[0 + j], zero);
       g = _mm_unpackhi_epi8(rgb_plane[2 + j], zero);
       b = _mm_unpackhi_epi8(rgb_plane[4 + j], zero);
-      ConvertRGBToY(&r, &g, &b, &Y1);
+      ConvertRGBToY_SSE2(&r, &g, &b, &Y1);
 
       // Cast to 8-bit and store.
       STORE_16(_mm_packus_epi16(Y0, Y1), y + i);
@@ -583,14 +593,14 @@ static void ConvertRGB24ToY(const uint8_t* rgb, uint8_t* y, int width) {
   }
 }
 
-static void ConvertBGR24ToY(const uint8_t* bgr, uint8_t* y, int width) {
+static void ConvertBGR24ToY_SSE2(const uint8_t* bgr, uint8_t* y, int width) {
   const int max_width = width & ~31;
   int i;
   for (i = 0; i < max_width; bgr += 3 * 16 * 2) {
     __m128i bgr_plane[6];
     int j;
 
-    RGB24PackedToPlanar(bgr, bgr_plane);
+    RGB24PackedToPlanar_SSE2(bgr, bgr_plane);
 
     for (j = 0; j < 2; ++j, i += 16) {
       const __m128i zero = _mm_setzero_si128();
@@ -600,13 +610,13 @@ static void ConvertBGR24ToY(const uint8_t* bgr, uint8_t* y, int width) {
       b = _mm_unpacklo_epi8(bgr_plane[0 + j], zero);
       g = _mm_unpacklo_epi8(bgr_plane[2 + j], zero);
       r = _mm_unpacklo_epi8(bgr_plane[4 + j], zero);
-      ConvertRGBToY(&r, &g, &b, &Y0);
+      ConvertRGBToY_SSE2(&r, &g, &b, &Y0);
 
       // Convert to 16-bit Y.
       b = _mm_unpackhi_epi8(bgr_plane[0 + j], zero);
       g = _mm_unpackhi_epi8(bgr_plane[2 + j], zero);
       r = _mm_unpackhi_epi8(bgr_plane[4 + j], zero);
-      ConvertRGBToY(&r, &g, &b, &Y1);
+      ConvertRGBToY_SSE2(&r, &g, &b, &Y1);
 
       // Cast to 8-bit and store.
       STORE_16(_mm_packus_epi16(Y0, Y1), y + i);
@@ -617,14 +627,14 @@ static void ConvertBGR24ToY(const uint8_t* bgr, uint8_t* y, int width) {
   }
 }
 
-static void ConvertARGBToY(const uint32_t* argb, uint8_t* y, int width) {
+static void ConvertARGBToY_SSE2(const uint32_t* argb, uint8_t* y, int width) {
   const int max_width = width & ~15;
   int i;
   for (i = 0; i < max_width; i += 16) {
     __m128i Y0, Y1, rgb[6];
-    RGB32PackedToPlanar(&argb[i], rgb);
-    ConvertRGBToY(&rgb[0], &rgb[2], &rgb[4], &Y0);
-    ConvertRGBToY(&rgb[1], &rgb[3], &rgb[5], &Y1);
+    RGB32PackedToPlanar_SSE2(&argb[i], rgb);
+    ConvertRGBToY_SSE2(&rgb[0], &rgb[2], &rgb[4], &Y0);
+    ConvertRGBToY_SSE2(&rgb[1], &rgb[3], &rgb[5], &Y1);
     STORE_16(_mm_packus_epi16(Y0, Y1), y + i);
   }
   for (; i < width; ++i) {   // left-over
@@ -636,31 +646,33 @@ static void ConvertARGBToY(const uint32_t* argb, uint8_t* y, int width) {
 
 // Horizontal add (doubled) of two 16b values, result is 16b.
 // in: A | B | C | D | ... -> out: 2*(A+B) | 2*(C+D) | ...
-static void HorizontalAddPack(const __m128i* const A, const __m128i* const B,
-                              __m128i* const out) {
+static void HorizontalAddPack_SSE2(const __m128i* const A,
+                                   const __m128i* const B,
+                                   __m128i* const out) {
   const __m128i k2 = _mm_set1_epi16(2);
   const __m128i C = _mm_madd_epi16(*A, k2);
   const __m128i D = _mm_madd_epi16(*B, k2);
   *out = _mm_packs_epi32(C, D);
 }
 
-static void ConvertARGBToUV(const uint32_t* argb, uint8_t* u, uint8_t* v,
-                            int src_width, int do_store) {
+static void ConvertARGBToUV_SSE2(const uint32_t* argb,
+                                 uint8_t* u, uint8_t* v,
+                                 int src_width, int do_store) {
   const int max_width = src_width & ~31;
   int i;
   for (i = 0; i < max_width; i += 32, u += 16, v += 16) {
     __m128i rgb[6], U0, V0, U1, V1;
-    RGB32PackedToPlanar(&argb[i], rgb);
-    HorizontalAddPack(&rgb[0], &rgb[1], &rgb[0]);
-    HorizontalAddPack(&rgb[2], &rgb[3], &rgb[2]);
-    HorizontalAddPack(&rgb[4], &rgb[5], &rgb[4]);
-    ConvertRGBToUV(&rgb[0], &rgb[2], &rgb[4], &U0, &V0);
-
-    RGB32PackedToPlanar(&argb[i + 16], rgb);
-    HorizontalAddPack(&rgb[0], &rgb[1], &rgb[0]);
-    HorizontalAddPack(&rgb[2], &rgb[3], &rgb[2]);
-    HorizontalAddPack(&rgb[4], &rgb[5], &rgb[4]);
-    ConvertRGBToUV(&rgb[0], &rgb[2], &rgb[4], &U1, &V1);
+    RGB32PackedToPlanar_SSE2(&argb[i], rgb);
+    HorizontalAddPack_SSE2(&rgb[0], &rgb[1], &rgb[0]);
+    HorizontalAddPack_SSE2(&rgb[2], &rgb[3], &rgb[2]);
+    HorizontalAddPack_SSE2(&rgb[4], &rgb[5], &rgb[4]);
+    ConvertRGBToUV_SSE2(&rgb[0], &rgb[2], &rgb[4], &U0, &V0);
+
+    RGB32PackedToPlanar_SSE2(&argb[i + 16], rgb);
+    HorizontalAddPack_SSE2(&rgb[0], &rgb[1], &rgb[0]);
+    HorizontalAddPack_SSE2(&rgb[2], &rgb[3], &rgb[2]);
+    HorizontalAddPack_SSE2(&rgb[4], &rgb[5], &rgb[4]);
+    ConvertRGBToUV_SSE2(&rgb[0], &rgb[2], &rgb[4], &U1, &V1);
 
     U0 = _mm_packus_epi16(U0, U1);
     V0 = _mm_packus_epi16(V0, V1);
@@ -679,10 +691,9 @@ static void ConvertARGBToUV(const uint32_t* argb, uint8_t* u, uint8_t* v,
 }
 
 // Convert 16 packed ARGB 16b-values to r[], g[], b[]
-static WEBP_INLINE void RGBA32PackedToPlanar_16b(const uint16_t* const rgbx,
-                                                 __m128i* const r,
-                                                 __m128i* const g,
-                                                 __m128i* const b) {
+static WEBP_INLINE void RGBA32PackedToPlanar_16b_SSE2(
+    const uint16_t* const rgbx,
+    __m128i* const r, __m128i* const g, __m128i* const b) {
   const __m128i in0 = LOAD_16(rgbx +  0);  // r0 | g0 | b0 |x| r1 | g1 | b1 |x
   const __m128i in1 = LOAD_16(rgbx +  8);  // r2 | g2 | b2 |x| r3 | g3 | b3 |x
   const __m128i in2 = LOAD_16(rgbx + 16);  // r4 | ...
@@ -701,16 +712,16 @@ static WEBP_INLINE void RGBA32PackedToPlanar_16b(const uint16_t* const rgbx,
   *b = _mm_unpacklo_epi64(B1, B3);
 }
 
-static void ConvertRGBA32ToUV(const uint16_t* rgb,
-                              uint8_t* u, uint8_t* v, int width) {
+static void ConvertRGBA32ToUV_SSE2(const uint16_t* rgb,
+                                   uint8_t* u, uint8_t* v, int width) {
   const int max_width = width & ~15;
   const uint16_t* const last_rgb = rgb + 4 * max_width;
   while (rgb < last_rgb) {
     __m128i r, g, b, U0, V0, U1, V1;
-    RGBA32PackedToPlanar_16b(rgb +  0, &r, &g, &b);
-    ConvertRGBToUV(&r, &g, &b, &U0, &V0);
-    RGBA32PackedToPlanar_16b(rgb + 32, &r, &g, &b);
-    ConvertRGBToUV(&r, &g, &b, &U1, &V1);
+    RGBA32PackedToPlanar_16b_SSE2(rgb +  0, &r, &g, &b);
+    ConvertRGBToUV_SSE2(&r, &g, &b, &U0, &V0);
+    RGBA32PackedToPlanar_16b_SSE2(rgb + 32, &r, &g, &b);
+    ConvertRGBToUV_SSE2(&r, &g, &b, &U1, &V1);
     STORE_16(_mm_packus_epi16(U0, U1), u);
     STORE_16(_mm_packus_epi16(V0, V1), v);
     u += 16;
@@ -727,13 +738,13 @@ static void ConvertRGBA32ToUV(const uint16_t* rgb,
 extern void WebPInitConvertARGBToYUVSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUVSSE2(void) {
-  WebPConvertARGBToY = ConvertARGBToY;
-  WebPConvertARGBToUV = ConvertARGBToUV;
+  WebPConvertARGBToY = ConvertARGBToY_SSE2;
+  WebPConvertARGBToUV = ConvertARGBToUV_SSE2;
 
-  WebPConvertRGB24ToY = ConvertRGB24ToY;
-  WebPConvertBGR24ToY = ConvertBGR24ToY;
+  WebPConvertRGB24ToY = ConvertRGB24ToY_SSE2;
+  WebPConvertBGR24ToY = ConvertBGR24ToY_SSE2;
 
-  WebPConvertRGBA32ToUV = ConvertRGBA32ToUV;
+  WebPConvertRGBA32ToUV = ConvertRGBA32ToUV_SSE2;
 }
 
 //------------------------------------------------------------------------------
diff --git a/src/3rdparty/libwebp/src/enc/alpha_enc.c b/src/3rdparty/libwebp/src/enc/alpha_enc.c
index 5a2c931..7e8d87f 100644
--- a/src/3rdparty/libwebp/src/enc/alpha_enc.c
+++ b/src/3rdparty/libwebp/src/enc/alpha_enc.c
@@ -14,12 +14,12 @@
 #include <assert.h>
 #include <stdlib.h>
 
-#include "./vp8i_enc.h"
-#include "../dsp/dsp.h"
-#include "../utils/filters_utils.h"
-#include "../utils/quant_levels_utils.h"
-#include "../utils/utils.h"
-#include "../webp/format_constants.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/filters_utils.h"
+#include "src/utils/quant_levels_utils.h"
+#include "src/utils/utils.h"
+#include "src/webp/format_constants.h"
 
 // -----------------------------------------------------------------------------
 // Encodes the given alpha data via specified compression method 'method'.
@@ -44,11 +44,11 @@
 //           invalid quality or method, or
 //           memory allocation for the compressed data fails.
 
-#include "../enc/vp8li_enc.h"
+#include "src/enc/vp8li_enc.h"
 
 static int EncodeLossless(const uint8_t* const data, int width, int height,
                           int effort_level,  // in [0..6] range
-                          VP8LBitWriter* const bw,
+                          int use_quality_100, VP8LBitWriter* const bw,
                           WebPAuxStats* const stats) {
   int ok = 0;
   WebPConfig config;
@@ -76,7 +76,10 @@ static int EncodeLossless(const uint8_t* const data, int width, int height,
   // Set a low default quality for encoding alpha. Ensure that Alpha quality at
   // lower methods (3 and below) is less than the threshold for triggering
   // costly 'BackwardReferencesTraceBackwards'.
-  config.quality = 8.f * effort_level;
+  // If the alpha quality is set to 100 and the method to 6, allow for a high
+  // lossless quality to trigger the cruncher.
+  config.quality =
+      (use_quality_100 && effort_level == 6) ? 100 : 8.f * effort_level;
   assert(config.quality >= 0 && config.quality <= 100.f);
 
   // TODO(urvang): Temporary fix to avoid generating images that trigger
@@ -134,7 +137,7 @@ static int EncodeAlphaInternal(const uint8_t* const data, int width, int height,
   if (method != ALPHA_NO_COMPRESSION) {
     ok = VP8LBitWriterInit(&tmp_bw, data_size >> 3);
     ok = ok && EncodeLossless(alpha_src, width, height, effort_level,
-                              &tmp_bw, &result->stats);
+                              !reduce_levels, &tmp_bw, &result->stats);
     if (ok) {
       output = VP8LBitWriterFinish(&tmp_bw);
       output_size = VP8LBitWriterNumBytes(&tmp_bw);
@@ -264,6 +267,7 @@ static int ApplyFiltersAndEncode(const uint8_t* alpha, int width, int height,
                              reduce_levels, effort_level, NULL, &best);
   }
   if (ok) {
+#if !defined(WEBP_DISABLE_STATS)
     if (stats != NULL) {
       stats->lossless_features = best.stats.lossless_features;
       stats->histogram_bits = best.stats.histogram_bits;
@@ -274,6 +278,9 @@ static int ApplyFiltersAndEncode(const uint8_t* alpha, int width, int height,
       stats->lossless_hdr_size = best.stats.lossless_hdr_size;
       stats->lossless_data_size = best.stats.lossless_data_size;
     }
+#else
+    (void)stats;
+#endif
     *output_size = VP8BitWriterSize(&best.bw);
     *output = VP8BitWriterBuf(&best.bw);
   } else {
@@ -339,10 +346,12 @@ static int EncodeAlpha(VP8Encoder* const enc,
     ok = ApplyFiltersAndEncode(quant_alpha, width, height, data_size, method,
                                filter, reduce_levels, effort_level, output,
                                output_size, pic->stats);
+#if !defined(WEBP_DISABLE_STATS)
     if (pic->stats != NULL) {  // need stats?
       pic->stats->coded_size += (int)(*output_size);
       enc->sse_[3] = sse;
     }
+#endif
   }
 
   WebPSafeFree(quant_alpha);
diff --git a/src/3rdparty/libwebp/src/enc/analysis_enc.c b/src/3rdparty/libwebp/src/enc/analysis_enc.c
index dce159b..08f471f 100644
--- a/src/3rdparty/libwebp/src/enc/analysis_enc.c
+++ b/src/3rdparty/libwebp/src/enc/analysis_enc.c
@@ -15,9 +15,9 @@
 #include <string.h>
 #include <assert.h>
 
-#include "./vp8i_enc.h"
-#include "./cost_enc.h"
-#include "../utils/utils.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/enc/cost_enc.h"
+#include "src/utils/utils.h"
 
 #define MAX_ITERS_K_MEANS  6
 
diff --git a/src/3rdparty/libwebp/src/enc/backward_references_cost_enc.c b/src/3rdparty/libwebp/src/enc/backward_references_cost_enc.c
new file mode 100644
index 0000000..7175496
--- /dev/null
+++ b/src/3rdparty/libwebp/src/enc/backward_references_cost_enc.c
@@ -0,0 +1,790 @@
+// Copyright 2017 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Improves a given set of backward references by analyzing its bit cost.
+// The algorithm is similar to the Zopfli compression algorithm but tailored to
+// images.
+//
+// Author: Vincent Rabaud (vrabaud@google.com)
+//
+
+#include <assert.h>
+
+#include "src/enc/backward_references_enc.h"
+#include "src/enc/histogram_enc.h"
+#include "src/dsp/lossless_common.h"
+#include "src/utils/color_cache_utils.h"
+#include "src/utils/utils.h"
+
+#define VALUES_IN_BYTE 256
+
+extern void VP8LClearBackwardRefs(VP8LBackwardRefs* const refs);
+extern int VP8LDistanceToPlaneCode(int xsize, int dist);
+extern void VP8LBackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
+                                      const PixOrCopy v);
+
+typedef struct {
+  double alpha_[VALUES_IN_BYTE];
+  double red_[VALUES_IN_BYTE];
+  double blue_[VALUES_IN_BYTE];
+  double distance_[NUM_DISTANCE_CODES];
+  double* literal_;
+} CostModel;
+
+static void ConvertPopulationCountTableToBitEstimates(
+    int num_symbols, const uint32_t population_counts[], double output[]) {
+  uint32_t sum = 0;
+  int nonzeros = 0;
+  int i;
+  for (i = 0; i < num_symbols; ++i) {
+    sum += population_counts[i];
+    if (population_counts[i] > 0) {
+      ++nonzeros;
+    }
+  }
+  if (nonzeros <= 1) {
+    memset(output, 0, num_symbols * sizeof(*output));
+  } else {
+    const double logsum = VP8LFastLog2(sum);
+    for (i = 0; i < num_symbols; ++i) {
+      output[i] = logsum - VP8LFastLog2(population_counts[i]);
+    }
+  }
+}
+
+static int CostModelBuild(CostModel* const m, int xsize, int cache_bits,
+                          const VP8LBackwardRefs* const refs) {
+  int ok = 0;
+  VP8LRefsCursor c = VP8LRefsCursorInit(refs);
+  VP8LHistogram* const histo = VP8LAllocateHistogram(cache_bits);
+  if (histo == NULL) goto Error;
+
+  // The following code is similar to VP8LHistogramCreate but converts the
+  // distance to plane code.
+  VP8LHistogramInit(histo, cache_bits);
+  while (VP8LRefsCursorOk(&c)) {
+    VP8LHistogramAddSinglePixOrCopy(histo, c.cur_pos, VP8LDistanceToPlaneCode,
+                                    xsize);
+    VP8LRefsCursorNext(&c);
+  }
+
+  ConvertPopulationCountTableToBitEstimates(
+      VP8LHistogramNumCodes(histo->palette_code_bits_),
+      histo->literal_, m->literal_);
+  ConvertPopulationCountTableToBitEstimates(
+      VALUES_IN_BYTE, histo->red_, m->red_);
+  ConvertPopulationCountTableToBitEstimates(
+      VALUES_IN_BYTE, histo->blue_, m->blue_);
+  ConvertPopulationCountTableToBitEstimates(
+      VALUES_IN_BYTE, histo->alpha_, m->alpha_);
+  ConvertPopulationCountTableToBitEstimates(
+      NUM_DISTANCE_CODES, histo->distance_, m->distance_);
+  ok = 1;
+
+ Error:
+  VP8LFreeHistogram(histo);
+  return ok;
+}
+
+static WEBP_INLINE double GetLiteralCost(const CostModel* const m, uint32_t v) {
+  return m->alpha_[v >> 24] +
+         m->red_[(v >> 16) & 0xff] +
+         m->literal_[(v >> 8) & 0xff] +
+         m->blue_[v & 0xff];
+}
+
+static WEBP_INLINE double GetCacheCost(const CostModel* const m, uint32_t idx) {
+  const int literal_idx = VALUES_IN_BYTE + NUM_LENGTH_CODES + idx;
+  return m->literal_[literal_idx];
+}
+
+static WEBP_INLINE double GetLengthCost(const CostModel* const m,
+                                        uint32_t length) {
+  int code, extra_bits;
+  VP8LPrefixEncodeBits(length, &code, &extra_bits);
+  return m->literal_[VALUES_IN_BYTE + code] + extra_bits;
+}
+
+static WEBP_INLINE double GetDistanceCost(const CostModel* const m,
+                                          uint32_t distance) {
+  int code, extra_bits;
+  VP8LPrefixEncodeBits(distance, &code, &extra_bits);
+  return m->distance_[code] + extra_bits;
+}
+
+static WEBP_INLINE void AddSingleLiteralWithCostModel(
+    const uint32_t* const argb, VP8LColorCache* const hashers,
+    const CostModel* const cost_model, int idx, int use_color_cache,
+    float prev_cost, float* const cost, uint16_t* const dist_array) {
+  double cost_val = prev_cost;
+  const uint32_t color = argb[idx];
+  const int ix = use_color_cache ? VP8LColorCacheContains(hashers, color) : -1;
+  if (ix >= 0) {
+    // use_color_cache is true and hashers contains color
+    const double mul0 = 0.68;
+    cost_val += GetCacheCost(cost_model, ix) * mul0;
+  } else {
+    const double mul1 = 0.82;
+    if (use_color_cache) VP8LColorCacheInsert(hashers, color);
+    cost_val += GetLiteralCost(cost_model, color) * mul1;
+  }
+  if (cost[idx] > cost_val) {
+    cost[idx] = (float)cost_val;
+    dist_array[idx] = 1;  // only one is inserted.
+  }
+}
+
+// -----------------------------------------------------------------------------
+// CostManager and interval handling
+
+// Empirical value to avoid high memory consumption but good for performance.
+#define COST_CACHE_INTERVAL_SIZE_MAX 500
+
+// To perform backward reference every pixel at index index_ is considered and
+// the cost for the MAX_LENGTH following pixels computed. Those following pixels
+// at index index_ + k (k from 0 to MAX_LENGTH) have a cost of:
+//     cost_ = distance cost at index + GetLengthCost(cost_model, k)
+// and the minimum value is kept. GetLengthCost(cost_model, k) is cached in an
+// array of size MAX_LENGTH.
+// Instead of performing MAX_LENGTH comparisons per pixel, we keep track of the
+// minimal values using intervals of constant cost.
+// An interval is defined by the index_ of the pixel that generated it and
+// is only useful in a range of indices from start_ to end_ (exclusive), i.e.
+// it contains the minimum value for pixels between start_ and end_.
+// Intervals are stored in a linked list and ordered by start_. When a new
+// interval has a better value, old intervals are split or removed. There are
+// therefore no overlapping intervals.
+typedef struct CostInterval CostInterval;
+struct CostInterval {
+  float cost_;
+  int start_;
+  int end_;
+  int index_;
+  CostInterval* previous_;
+  CostInterval* next_;
+};
+
+// The GetLengthCost(cost_model, k) are cached in a CostCacheInterval.
+typedef struct {
+  double cost_;
+  int start_;
+  int end_;       // Exclusive.
+} CostCacheInterval;
+
+// This structure is in charge of managing intervals and costs.
+// It caches the different CostCacheInterval, caches the different
+// GetLengthCost(cost_model, k) in cost_cache_ and the CostInterval's (whose
+// count_ is limited by COST_CACHE_INTERVAL_SIZE_MAX).
+#define COST_MANAGER_MAX_FREE_LIST 10
+typedef struct {
+  CostInterval* head_;
+  int count_;  // The number of stored intervals.
+  CostCacheInterval* cache_intervals_;
+  size_t cache_intervals_size_;
+  double cost_cache_[MAX_LENGTH];  // Contains the GetLengthCost(cost_model, k).
+  float* costs_;
+  uint16_t* dist_array_;
+  // Most of the time, we only need few intervals -> use a free-list, to avoid
+  // fragmentation with small allocs in most common cases.
+  CostInterval intervals_[COST_MANAGER_MAX_FREE_LIST];
+  CostInterval* free_intervals_;
+  // These are regularly malloc'd remains. This list can't grow larger than than
+  // size COST_CACHE_INTERVAL_SIZE_MAX - COST_MANAGER_MAX_FREE_LIST, note.
+  CostInterval* recycled_intervals_;
+} CostManager;
+
+static void CostIntervalAddToFreeList(CostManager* const manager,
+                                      CostInterval* const interval) {
+  interval->next_ = manager->free_intervals_;
+  manager->free_intervals_ = interval;
+}
+
+static int CostIntervalIsInFreeList(const CostManager* const manager,
+                                    const CostInterval* const interval) {
+  return (interval >= &manager->intervals_[0] &&
+          interval <= &manager->intervals_[COST_MANAGER_MAX_FREE_LIST - 1]);
+}
+
+static void CostManagerInitFreeList(CostManager* const manager) {
+  int i;
+  manager->free_intervals_ = NULL;
+  for (i = 0; i < COST_MANAGER_MAX_FREE_LIST; ++i) {
+    CostIntervalAddToFreeList(manager, &manager->intervals_[i]);
+  }
+}
+
+static void DeleteIntervalList(CostManager* const manager,
+                               const CostInterval* interval) {
+  while (interval != NULL) {
+    const CostInterval* const next = interval->next_;
+    if (!CostIntervalIsInFreeList(manager, interval)) {
+      WebPSafeFree((void*)interval);
+    }  // else: do nothing
+    interval = next;
+  }
+}
+
+static void CostManagerClear(CostManager* const manager) {
+  if (manager == NULL) return;
+
+  WebPSafeFree(manager->costs_);
+  WebPSafeFree(manager->cache_intervals_);
+
+  // Clear the interval lists.
+  DeleteIntervalList(manager, manager->head_);
+  manager->head_ = NULL;
+  DeleteIntervalList(manager, manager->recycled_intervals_);
+  manager->recycled_intervals_ = NULL;
+
+  // Reset pointers, count_ and cache_intervals_size_.
+  memset(manager, 0, sizeof(*manager));
+  CostManagerInitFreeList(manager);
+}
+
+static int CostManagerInit(CostManager* const manager,
+                           uint16_t* const dist_array, int pix_count,
+                           const CostModel* const cost_model) {
+  int i;
+  const int cost_cache_size = (pix_count > MAX_LENGTH) ? MAX_LENGTH : pix_count;
+
+  manager->costs_ = NULL;
+  manager->cache_intervals_ = NULL;
+  manager->head_ = NULL;
+  manager->recycled_intervals_ = NULL;
+  manager->count_ = 0;
+  manager->dist_array_ = dist_array;
+  CostManagerInitFreeList(manager);
+
+  // Fill in the cost_cache_.
+  manager->cache_intervals_size_ = 1;
+  manager->cost_cache_[0] = GetLengthCost(cost_model, 0);
+  for (i = 1; i < cost_cache_size; ++i) {
+    manager->cost_cache_[i] = GetLengthCost(cost_model, i);
+    // Get the number of bound intervals.
+    if (manager->cost_cache_[i] != manager->cost_cache_[i - 1]) {
+      ++manager->cache_intervals_size_;
+    }
+  }
+
+  // With the current cost model, we usually have below 20 intervals.
+  // The worst case scenario with a cost model would be if every length has a
+  // different cost, hence MAX_LENGTH but that is impossible with the current
+  // implementation that spirals around a pixel.
+  assert(manager->cache_intervals_size_ <= MAX_LENGTH);
+  manager->cache_intervals_ = (CostCacheInterval*)WebPSafeMalloc(
+      manager->cache_intervals_size_, sizeof(*manager->cache_intervals_));
+  if (manager->cache_intervals_ == NULL) {
+    CostManagerClear(manager);
+    return 0;
+  }
+
+  // Fill in the cache_intervals_.
+  {
+    CostCacheInterval* cur = manager->cache_intervals_;
+
+    // Consecutive values in cost_cache_ are compared and if a big enough
+    // difference is found, a new interval is created and bounded.
+    cur->start_ = 0;
+    cur->end_ = 1;
+    cur->cost_ = manager->cost_cache_[0];
+    for (i = 1; i < cost_cache_size; ++i) {
+      const double cost_val = manager->cost_cache_[i];
+      if (cost_val != cur->cost_) {
+        ++cur;
+        // Initialize an interval.
+        cur->start_ = i;
+        cur->cost_ = cost_val;
+      }
+      cur->end_ = i + 1;
+    }
+  }
+
+  manager->costs_ = (float*)WebPSafeMalloc(pix_count, sizeof(*manager->costs_));
+  if (manager->costs_ == NULL) {
+    CostManagerClear(manager);
+    return 0;
+  }
+  // Set the initial costs_ high for every pixel as we will keep the minimum.
+  for (i = 0; i < pix_count; ++i) manager->costs_[i] = 1e38f;
+
+  return 1;
+}
+
+// Given the cost and the position that define an interval, update the cost at
+// pixel 'i' if it is smaller than the previously computed value.
+static WEBP_INLINE void UpdateCost(CostManager* const manager, int i,
+                                   int position, float cost) {
+  const int k = i - position;
+  assert(k >= 0 && k < MAX_LENGTH);
+
+  if (manager->costs_[i] > cost) {
+    manager->costs_[i] = cost;
+    manager->dist_array_[i] = k + 1;
+  }
+}
+
+// Given the cost and the position that define an interval, update the cost for
+// all the pixels between 'start' and 'end' excluded.
+static WEBP_INLINE void UpdateCostPerInterval(CostManager* const manager,
+                                              int start, int end, int position,
+                                              float cost) {
+  int i;
+  for (i = start; i < end; ++i) UpdateCost(manager, i, position, cost);
+}
+
+// Given two intervals, make 'prev' be the previous one of 'next' in 'manager'.
+static WEBP_INLINE void ConnectIntervals(CostManager* const manager,
+                                         CostInterval* const prev,
+                                         CostInterval* const next) {
+  if (prev != NULL) {
+    prev->next_ = next;
+  } else {
+    manager->head_ = next;
+  }
+
+  if (next != NULL) next->previous_ = prev;
+}
+
+// Pop an interval in the manager.
+static WEBP_INLINE void PopInterval(CostManager* const manager,
+                                    CostInterval* const interval) {
+  if (interval == NULL) return;
+
+  ConnectIntervals(manager, interval->previous_, interval->next_);
+  if (CostIntervalIsInFreeList(manager, interval)) {
+    CostIntervalAddToFreeList(manager, interval);
+  } else {  // recycle regularly malloc'd intervals too
+    interval->next_ = manager->recycled_intervals_;
+    manager->recycled_intervals_ = interval;
+  }
+  --manager->count_;
+  assert(manager->count_ >= 0);
+}
+
+// Update the cost at index i by going over all the stored intervals that
+// overlap with i.
+// If 'do_clean_intervals' is set to something different than 0, intervals that
+// end before 'i' will be popped.
+static WEBP_INLINE void UpdateCostAtIndex(CostManager* const manager, int i,
+                                          int do_clean_intervals) {
+  CostInterval* current = manager->head_;
+
+  while (current != NULL && current->start_ <= i) {
+    CostInterval* const next = current->next_;
+    if (current->end_ <= i) {
+      if (do_clean_intervals) {
+        // We have an outdated interval, remove it.
+        PopInterval(manager, current);
+      }
+    } else {
+      UpdateCost(manager, i, current->index_, current->cost_);
+    }
+    current = next;
+  }
+}
+
+// Given a current orphan interval and its previous interval, before
+// it was orphaned (which can be NULL), set it at the right place in the list
+// of intervals using the start_ ordering and the previous interval as a hint.
+static WEBP_INLINE void PositionOrphanInterval(CostManager* const manager,
+                                               CostInterval* const current,
+                                               CostInterval* previous) {
+  assert(current != NULL);
+
+  if (previous == NULL) previous = manager->head_;
+  while (previous != NULL && current->start_ < previous->start_) {
+    previous = previous->previous_;
+  }
+  while (previous != NULL && previous->next_ != NULL &&
+         previous->next_->start_ < current->start_) {
+    previous = previous->next_;
+  }
+
+  if (previous != NULL) {
+    ConnectIntervals(manager, current, previous->next_);
+  } else {
+    ConnectIntervals(manager, current, manager->head_);
+  }
+  ConnectIntervals(manager, previous, current);
+}
+
+// Insert an interval in the list contained in the manager by starting at
+// interval_in as a hint. The intervals are sorted by start_ value.
+static WEBP_INLINE void InsertInterval(CostManager* const manager,
+                                       CostInterval* const interval_in,
+                                       float cost, int position, int start,
+                                       int end) {
+  CostInterval* interval_new;
+
+  if (start >= end) return;
+  if (manager->count_ >= COST_CACHE_INTERVAL_SIZE_MAX) {
+    // Serialize the interval if we cannot store it.
+    UpdateCostPerInterval(manager, start, end, position, cost);
+    return;
+  }
+  if (manager->free_intervals_ != NULL) {
+    interval_new = manager->free_intervals_;
+    manager->free_intervals_ = interval_new->next_;
+  } else if (manager->recycled_intervals_ != NULL) {
+    interval_new = manager->recycled_intervals_;
+    manager->recycled_intervals_ = interval_new->next_;
+  } else {  // malloc for good
+    interval_new = (CostInterval*)WebPSafeMalloc(1, sizeof(*interval_new));
+    if (interval_new == NULL) {
+      // Write down the interval if we cannot create it.
+      UpdateCostPerInterval(manager, start, end, position, cost);
+      return;
+    }
+  }
+
+  interval_new->cost_ = cost;
+  interval_new->index_ = position;
+  interval_new->start_ = start;
+  interval_new->end_ = end;
+  PositionOrphanInterval(manager, interval_new, interval_in);
+
+  ++manager->count_;
+}
+
+// Given a new cost interval defined by its start at position, its length value
+// and distance_cost, add its contributions to the previous intervals and costs.
+// If handling the interval or one of its subintervals becomes to heavy, its
+// contribution is added to the costs right away.
+static WEBP_INLINE void PushInterval(CostManager* const manager,
+                                     double distance_cost, int position,
+                                     int len) {
+  size_t i;
+  CostInterval* interval = manager->head_;
+  CostInterval* interval_next;
+  const CostCacheInterval* const cost_cache_intervals =
+      manager->cache_intervals_;
+  // If the interval is small enough, no need to deal with the heavy
+  // interval logic, just serialize it right away. This constant is empirical.
+  const int kSkipDistance = 10;
+
+  if (len < kSkipDistance) {
+    int j;
+    for (j = position; j < position + len; ++j) {
+      const int k = j - position;
+      float cost_tmp;
+      assert(k >= 0 && k < MAX_LENGTH);
+      cost_tmp = (float)(distance_cost + manager->cost_cache_[k]);
+
+      if (manager->costs_[j] > cost_tmp) {
+        manager->costs_[j] = cost_tmp;
+        manager->dist_array_[j] = k + 1;
+      }
+    }
+    return;
+  }
+
+  for (i = 0; i < manager->cache_intervals_size_ &&
+              cost_cache_intervals[i].start_ < len;
+       ++i) {
+    // Define the intersection of the ith interval with the new one.
+    int start = position + cost_cache_intervals[i].start_;
+    const int end = position + (cost_cache_intervals[i].end_ > len
+                                 ? len
+                                 : cost_cache_intervals[i].end_);
+    const float cost = (float)(distance_cost + cost_cache_intervals[i].cost_);
+
+    for (; interval != NULL && interval->start_ < end;
+         interval = interval_next) {
+      interval_next = interval->next_;
+
+      // Make sure we have some overlap
+      if (start >= interval->end_) continue;
+
+      if (cost >= interval->cost_) {
+        // When intervals are represented, the lower, the better.
+        // [**********************************************************[
+        // start                                                    end
+        //                   [----------------------------------[
+        //                   interval->start_       interval->end_
+        // If we are worse than what we already have, add whatever we have so
+        // far up to interval.
+        const int start_new = interval->end_;
+        InsertInterval(manager, interval, cost, position, start,
+                       interval->start_);
+        start = start_new;
+        if (start >= end) break;
+        continue;
+      }
+
+      if (start <= interval->start_) {
+        if (interval->end_ <= end) {
+          //                   [----------------------------------[
+          //                   interval->start_       interval->end_
+          // [**************************************************************[
+          // start                                                        end
+          // We can safely remove the old interval as it is fully included.
+          PopInterval(manager, interval);
+        } else {
+          //              [------------------------------------[
+          //              interval->start_        interval->end_
+          // [*****************************[
+          // start                       end
+          interval->start_ = end;
+          break;
+        }
+      } else {
+        if (end < interval->end_) {
+          // [--------------------------------------------------------------[
+          // interval->start_                                  interval->end_
+          //                     [*****************************[
+          //                     start                       end
+          // We have to split the old interval as it fully contains the new one.
+          const int end_original = interval->end_;
+          interval->end_ = start;
+          InsertInterval(manager, interval, interval->cost_, interval->index_,
+                         end, end_original);
+          interval = interval->next_;
+          break;
+        } else {
+          // [------------------------------------[
+          // interval->start_        interval->end_
+          //                     [*****************************[
+          //                     start                       end
+          interval->end_ = start;
+        }
+      }
+    }
+    // Insert the remaining interval from start to end.
+    InsertInterval(manager, interval, cost, position, start, end);
+  }
+}
+
+static int BackwardReferencesHashChainDistanceOnly(
+    int xsize, int ysize, const uint32_t* const argb, int cache_bits,
+    const VP8LHashChain* const hash_chain, const VP8LBackwardRefs* const refs,
+    uint16_t* const dist_array) {
+  int i;
+  int ok = 0;
+  int cc_init = 0;
+  const int pix_count = xsize * ysize;
+  const int use_color_cache = (cache_bits > 0);
+  const size_t literal_array_size =
+      sizeof(double) * (NUM_LITERAL_CODES + NUM_LENGTH_CODES +
+                        ((cache_bits > 0) ? (1 << cache_bits) : 0));
+  const size_t cost_model_size = sizeof(CostModel) + literal_array_size;
+  CostModel* const cost_model =
+      (CostModel*)WebPSafeCalloc(1ULL, cost_model_size);
+  VP8LColorCache hashers;
+  CostManager* cost_manager =
+      (CostManager*)WebPSafeMalloc(1ULL, sizeof(*cost_manager));
+  int offset_prev = -1, len_prev = -1;
+  double offset_cost = -1;
+  int first_offset_is_constant = -1;  // initialized with 'impossible' value
+  int reach = 0;
+
+  if (cost_model == NULL || cost_manager == NULL) goto Error;
+
+  cost_model->literal_ = (double*)(cost_model + 1);
+  if (use_color_cache) {
+    cc_init = VP8LColorCacheInit(&hashers, cache_bits);
+    if (!cc_init) goto Error;
+  }
+
+  if (!CostModelBuild(cost_model, xsize, cache_bits, refs)) {
+    goto Error;
+  }
+
+  if (!CostManagerInit(cost_manager, dist_array, pix_count, cost_model)) {
+    goto Error;
+  }
+
+  // We loop one pixel at a time, but store all currently best points to
+  // non-processed locations from this point.
+  dist_array[0] = 0;
+  // Add first pixel as literal.
+  AddSingleLiteralWithCostModel(argb, &hashers, cost_model, 0, use_color_cache,
+                                0.f, cost_manager->costs_, dist_array);
+
+  for (i = 1; i < pix_count; ++i) {
+    const float prev_cost = cost_manager->costs_[i - 1];
+    int offset, len;
+    VP8LHashChainFindCopy(hash_chain, i, &offset, &len);
+
+    // Try adding the pixel as a literal.
+    AddSingleLiteralWithCostModel(argb, &hashers, cost_model, i,
+                                  use_color_cache, prev_cost,
+                                  cost_manager->costs_, dist_array);
+
+    // If we are dealing with a non-literal.
+    if (len >= 2) {
+      if (offset != offset_prev) {
+        const int code = VP8LDistanceToPlaneCode(xsize, offset);
+        offset_cost = GetDistanceCost(cost_model, code);
+        first_offset_is_constant = 1;
+        PushInterval(cost_manager, prev_cost + offset_cost, i, len);
+      } else {
+        assert(offset_cost >= 0);
+        assert(len_prev >= 0);
+        assert(first_offset_is_constant == 0 || first_offset_is_constant == 1);
+        // Instead of considering all contributions from a pixel i by calling:
+        //         PushInterval(cost_manager, prev_cost + offset_cost, i, len);
+        // we optimize these contributions in case offset_cost stays the same
+        // for consecutive pixels. This describes a set of pixels similar to a
+        // previous set (e.g. constant color regions).
+        if (first_offset_is_constant) {
+          reach = i - 1 + len_prev - 1;
+          first_offset_is_constant = 0;
+        }
+
+        if (i + len - 1 > reach) {
+          // We can only be go further with the same offset if the previous
+          // length was maxed, hence len_prev == len == MAX_LENGTH.
+          // TODO(vrabaud), bump i to the end right away (insert cache and
+          // update cost).
+          // TODO(vrabaud), check if one of the points in between does not have
+          // a lower cost.
+          // Already consider the pixel at "reach" to add intervals that are
+          // better than whatever we add.
+          int offset_j, len_j = 0;
+          int j;
+          assert(len == MAX_LENGTH || len == pix_count - i);
+          // Figure out the last consecutive pixel within [i, reach + 1] with
+          // the same offset.
+          for (j = i; j <= reach; ++j) {
+            VP8LHashChainFindCopy(hash_chain, j + 1, &offset_j, &len_j);
+            if (offset_j != offset) {
+              VP8LHashChainFindCopy(hash_chain, j, &offset_j, &len_j);
+              break;
+            }
+          }
+          // Update the cost at j - 1 and j.
+          UpdateCostAtIndex(cost_manager, j - 1, 0);
+          UpdateCostAtIndex(cost_manager, j, 0);
+
+          PushInterval(cost_manager, cost_manager->costs_[j - 1] + offset_cost,
+                       j, len_j);
+          reach = j + len_j - 1;
+        }
+      }
+    }
+
+    UpdateCostAtIndex(cost_manager, i, 1);
+    offset_prev = offset;
+    len_prev = len;
+  }
+
+  ok = !refs->error_;
+Error:
+  if (cc_init) VP8LColorCacheClear(&hashers);
+  CostManagerClear(cost_manager);
+  WebPSafeFree(cost_model);
+  WebPSafeFree(cost_manager);
+  return ok;
+}
+
+// We pack the path at the end of *dist_array and return
+// a pointer to this part of the array. Example:
+// dist_array = [1x2xx3x2] => packed [1x2x1232], chosen_path = [1232]
+static void TraceBackwards(uint16_t* const dist_array,
+                           int dist_array_size,
+                           uint16_t** const chosen_path,
+                           int* const chosen_path_size) {
+  uint16_t* path = dist_array + dist_array_size;
+  uint16_t* cur = dist_array + dist_array_size - 1;
+  while (cur >= dist_array) {
+    const int k = *cur;
+    --path;
+    *path = k;
+    cur -= k;
+  }
+  *chosen_path = path;
+  *chosen_path_size = (int)(dist_array + dist_array_size - path);
+}
+
+static int BackwardReferencesHashChainFollowChosenPath(
+    const uint32_t* const argb, int cache_bits,
+    const uint16_t* const chosen_path, int chosen_path_size,
+    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs) {
+  const int use_color_cache = (cache_bits > 0);
+  int ix;
+  int i = 0;
+  int ok = 0;
+  int cc_init = 0;
+  VP8LColorCache hashers;
+
+  if (use_color_cache) {
+    cc_init = VP8LColorCacheInit(&hashers, cache_bits);
+    if (!cc_init) goto Error;
+  }
+
+  VP8LClearBackwardRefs(refs);
+  for (ix = 0; ix < chosen_path_size; ++ix) {
+    const int len = chosen_path[ix];
+    if (len != 1) {
+      int k;
+      const int offset = VP8LHashChainFindOffset(hash_chain, i);
+      VP8LBackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(offset, len));
+      if (use_color_cache) {
+        for (k = 0; k < len; ++k) {
+          VP8LColorCacheInsert(&hashers, argb[i + k]);
+        }
+      }
+      i += len;
+    } else {
+      PixOrCopy v;
+      const int idx =
+          use_color_cache ? VP8LColorCacheContains(&hashers, argb[i]) : -1;
+      if (idx >= 0) {
+        // use_color_cache is true and hashers contains argb[i]
+        // push pixel as a color cache index
+        v = PixOrCopyCreateCacheIdx(idx);
+      } else {
+        if (use_color_cache) VP8LColorCacheInsert(&hashers, argb[i]);
+        v = PixOrCopyCreateLiteral(argb[i]);
+      }
+      VP8LBackwardRefsCursorAdd(refs, v);
+      ++i;
+    }
+  }
+  ok = !refs->error_;
+ Error:
+  if (cc_init) VP8LColorCacheClear(&hashers);
+  return ok;
+}
+
+// Returns 1 on success.
+extern int VP8LBackwardReferencesTraceBackwards(
+    int xsize, int ysize, const uint32_t* const argb, int cache_bits,
+    const VP8LHashChain* const hash_chain,
+    const VP8LBackwardRefs* const refs_src, VP8LBackwardRefs* const refs_dst);
+int VP8LBackwardReferencesTraceBackwards(int xsize, int ysize,
+                                         const uint32_t* const argb,
+                                         int cache_bits,
+                                         const VP8LHashChain* const hash_chain,
+                                         const VP8LBackwardRefs* const refs_src,
+                                         VP8LBackwardRefs* const refs_dst) {
+  int ok = 0;
+  const int dist_array_size = xsize * ysize;
+  uint16_t* chosen_path = NULL;
+  int chosen_path_size = 0;
+  uint16_t* dist_array =
+      (uint16_t*)WebPSafeMalloc(dist_array_size, sizeof(*dist_array));
+
+  if (dist_array == NULL) goto Error;
+
+  if (!BackwardReferencesHashChainDistanceOnly(
+          xsize, ysize, argb, cache_bits, hash_chain, refs_src, dist_array)) {
+    goto Error;
+  }
+  TraceBackwards(dist_array, dist_array_size, &chosen_path, &chosen_path_size);
+  if (!BackwardReferencesHashChainFollowChosenPath(
+          argb, cache_bits, chosen_path, chosen_path_size, hash_chain,
+          refs_dst)) {
+    goto Error;
+  }
+  ok = 1;
+ Error:
+  WebPSafeFree(dist_array);
+  return ok;
+}
diff --git a/src/3rdparty/libwebp/src/enc/backward_references_enc.c b/src/3rdparty/libwebp/src/enc/backward_references_enc.c
index 7c0559f..3923018 100644
--- a/src/3rdparty/libwebp/src/enc/backward_references_enc.c
+++ b/src/3rdparty/libwebp/src/enc/backward_references_enc.c
@@ -13,35 +13,24 @@
 #include <assert.h>
 #include <math.h>
 
-#include "./backward_references_enc.h"
-#include "./histogram_enc.h"
-#include "../dsp/lossless.h"
-#include "../dsp/lossless_common.h"
-#include "../dsp/dsp.h"
-#include "../utils/color_cache_utils.h"
-#include "../utils/utils.h"
-
-#define VALUES_IN_BYTE 256
+#include "src/enc/backward_references_enc.h"
+#include "src/enc/histogram_enc.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/color_cache_utils.h"
+#include "src/utils/utils.h"
 
 #define MIN_BLOCK_SIZE 256  // minimum block size for backward references
 
 #define MAX_ENTROPY    (1e30f)
 
 // 1M window (4M bytes) minus 120 special codes for short distances.
-#define WINDOW_SIZE_BITS 20
 #define WINDOW_SIZE ((1 << WINDOW_SIZE_BITS) - 120)
 
 // Minimum number of pixels for which it is cheaper to encode a
 // distance + length instead of each pixel as a literal.
 #define MIN_LENGTH 4
-// If you change this, you need MAX_LENGTH_BITS + WINDOW_SIZE_BITS <= 32 as it
-// is used in VP8LHashChain.
-#define MAX_LENGTH_BITS 12
-// We want the max value to be attainable and stored in MAX_LENGTH_BITS bits.
-#define MAX_LENGTH ((1 << MAX_LENGTH_BITS) - 1)
-#if MAX_LENGTH_BITS + WINDOW_SIZE_BITS > 32
-#error "MAX_LENGTH_BITS + WINDOW_SIZE_BITS > 32"
-#endif
 
 // -----------------------------------------------------------------------------
 
@@ -56,7 +45,8 @@ static const uint8_t plane_to_code_lut[128] = {
  119, 116, 111, 106,  97,  88,  84,  74,  72,  75,  85,  89,  98, 107, 112, 117
 };
 
-static int DistanceToPlaneCode(int xsize, int dist) {
+extern int VP8LDistanceToPlaneCode(int xsize, int dist);
+int VP8LDistanceToPlaneCode(int xsize, int dist) {
   const int yoffset = dist / xsize;
   const int xoffset = dist - yoffset * xsize;
   if (xoffset <= 8 && yoffset < 8) {
@@ -91,7 +81,8 @@ struct PixOrCopyBlock {
   int size_;               // currently used size
 };
 
-static void ClearBackwardRefs(VP8LBackwardRefs* const refs) {
+extern void VP8LClearBackwardRefs(VP8LBackwardRefs* const refs);
+void VP8LClearBackwardRefs(VP8LBackwardRefs* const refs) {
   assert(refs != NULL);
   if (refs->tail_ != NULL) {
     *refs->tail_ = refs->free_blocks_;  // recycle all blocks at once
@@ -104,7 +95,7 @@ static void ClearBackwardRefs(VP8LBackwardRefs* const refs) {
 
 void VP8LBackwardRefsClear(VP8LBackwardRefs* const refs) {
   assert(refs != NULL);
-  ClearBackwardRefs(refs);
+  VP8LClearBackwardRefs(refs);
   while (refs->free_blocks_ != NULL) {
     PixOrCopyBlock* const next = refs->free_blocks_->next_;
     WebPSafeFree(refs->free_blocks_);
@@ -163,8 +154,10 @@ static PixOrCopyBlock* BackwardRefsNewBlock(VP8LBackwardRefs* const refs) {
   return b;
 }
 
-static WEBP_INLINE void BackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
-                                              const PixOrCopy v) {
+extern void VP8LBackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
+                                      const PixOrCopy v);
+void VP8LBackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
+                               const PixOrCopy v) {
   PixOrCopyBlock* b = refs->last_block_;
   if (b == NULL || b->size_ == refs->block_size_) {
     b = BackwardRefsNewBlock(refs);
@@ -173,21 +166,6 @@ static WEBP_INLINE void BackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
   b->start_[b->size_++] = v;
 }
 
-int VP8LBackwardRefsCopy(const VP8LBackwardRefs* const src,
-                         VP8LBackwardRefs* const dst) {
-  const PixOrCopyBlock* b = src->refs_;
-  ClearBackwardRefs(dst);
-  assert(src->block_size_ == dst->block_size_);
-  while (b != NULL) {
-    PixOrCopyBlock* const new_b = BackwardRefsNewBlock(dst);
-    if (new_b == NULL) return 0;   // dst->error_ is set
-    memcpy(new_b->start_, b->start_, b->size_ * sizeof(*b->start_));
-    new_b->size_ = b->size_;
-    b = b->next_;
-  }
-  return 1;
-}
-
 // -----------------------------------------------------------------------------
 // Hash chains
 
@@ -411,24 +389,6 @@ int VP8LHashChainFill(VP8LHashChain* const p, int quality,
   return 1;
 }
 
-static WEBP_INLINE int HashChainFindOffset(const VP8LHashChain* const p,
-                                           const int base_position) {
-  return p->offset_length_[base_position] >> MAX_LENGTH_BITS;
-}
-
-static WEBP_INLINE int HashChainFindLength(const VP8LHashChain* const p,
-                                           const int base_position) {
-  return p->offset_length_[base_position] & ((1U << MAX_LENGTH_BITS) - 1);
-}
-
-static WEBP_INLINE void HashChainFindCopy(const VP8LHashChain* const p,
-                                          int base_position,
-                                          int* const offset_ptr,
-                                          int* const length_ptr) {
-  *offset_ptr = HashChainFindOffset(p, base_position);
-  *length_ptr = HashChainFindLength(p, base_position);
-}
-
 static WEBP_INLINE void AddSingleLiteral(uint32_t pixel, int use_color_cache,
                                          VP8LColorCache* const hashers,
                                          VP8LBackwardRefs* const refs) {
@@ -444,7 +404,7 @@ static WEBP_INLINE void AddSingleLiteral(uint32_t pixel, int use_color_cache,
   } else {
     v = PixOrCopyCreateLiteral(pixel);
   }
-  BackwardRefsCursorAdd(refs, v);
+  VP8LBackwardRefsCursorAdd(refs, v);
 }
 
 static int BackwardReferencesRle(int xsize, int ysize,
@@ -458,7 +418,7 @@ static int BackwardReferencesRle(int xsize, int ysize,
   if (use_color_cache && !VP8LColorCacheInit(&hashers, cache_bits)) {
     return 0;
   }
-  ClearBackwardRefs(refs);
+  VP8LClearBackwardRefs(refs);
   // Add first pixel as literal.
   AddSingleLiteral(argb[0], use_color_cache, &hashers, refs);
   i = 1;
@@ -468,13 +428,13 @@ static int BackwardReferencesRle(int xsize, int ysize,
     const int prev_row_len = (i < xsize) ? 0 :
         FindMatchLength(argb + i, argb + i - xsize, 0, max_len);
     if (rle_len >= prev_row_len && rle_len >= MIN_LENGTH) {
-      BackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(1, rle_len));
+      VP8LBackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(1, rle_len));
       // We don't need to update the color cache here since it is always the
       // same pixel being copied, and that does not change the color cache
       // state.
       i += rle_len;
     } else if (prev_row_len >= MIN_LENGTH) {
-      BackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(xsize, prev_row_len));
+      VP8LBackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(xsize, prev_row_len));
       if (use_color_cache) {
         for (k = 0; k < prev_row_len; ++k) {
           VP8LColorCacheInsert(&hashers, argb[i + k]);
@@ -506,17 +466,18 @@ static int BackwardReferencesLz77(int xsize, int ysize,
     cc_init = VP8LColorCacheInit(&hashers, cache_bits);
     if (!cc_init) goto Error;
   }
-  ClearBackwardRefs(refs);
+  VP8LClearBackwardRefs(refs);
   for (i = 0; i < pix_count;) {
     // Alternative#1: Code the pixels starting at 'i' using backward reference.
     int offset = 0;
     int len = 0;
     int j;
-    HashChainFindCopy(hash_chain, i, &offset, &len);
+    VP8LHashChainFindCopy(hash_chain, i, &offset, &len);
     if (len >= MIN_LENGTH) {
       const int len_ini = len;
       int max_reach = 0;
-      assert(i + len < pix_count);
+      const int j_max =
+          (i + len_ini >= pix_count) ? pix_count - 1 : i + len_ini;
       // Only start from what we have not checked already.
       i_last_check = (i > i_last_check) ? i : i_last_check;
       // We know the best match for the current pixel but we try to find the
@@ -525,13 +486,14 @@ static int BackwardReferencesLz77(int xsize, int ysize,
       // [i,i+len) + [i+len, length of best match at i+len)
       // while we check if we can use:
       // [i,j) (where j<=i+len) + [j, length of best match at j)
-      for (j = i_last_check + 1; j <= i + len_ini; ++j) {
-        const int len_j = HashChainFindLength(hash_chain, j);
+      for (j = i_last_check + 1; j <= j_max; ++j) {
+        const int len_j = VP8LHashChainFindLength(hash_chain, j);
         const int reach =
             j + (len_j >= MIN_LENGTH ? len_j : 1);  // 1 for single literal.
         if (reach > max_reach) {
           len = j - i;
           max_reach = reach;
+          if (max_reach >= pix_count) break;
         }
       }
     } else {
@@ -542,7 +504,7 @@ static int BackwardReferencesLz77(int xsize, int ysize,
     if (len == 1) {
       AddSingleLiteral(argb[i], use_color_cache, &hashers, refs);
     } else {
-      BackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(offset, len));
+      VP8LBackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(offset, len));
       if (use_color_cache) {
         for (j = i; j < i + len; ++j) VP8LColorCacheInsert(&hashers, argb[j]);
       }
@@ -556,989 +518,200 @@ static int BackwardReferencesLz77(int xsize, int ysize,
   return ok;
 }
 
-// -----------------------------------------------------------------------------
-
-typedef struct {
-  double alpha_[VALUES_IN_BYTE];
-  double red_[VALUES_IN_BYTE];
-  double blue_[VALUES_IN_BYTE];
-  double distance_[NUM_DISTANCE_CODES];
-  double* literal_;
-} CostModel;
-
-static int BackwardReferencesTraceBackwards(
-    int xsize, int ysize, const uint32_t* const argb, int quality,
-    int cache_bits, const VP8LHashChain* const hash_chain,
-    VP8LBackwardRefs* const refs);
-
-static void ConvertPopulationCountTableToBitEstimates(
-    int num_symbols, const uint32_t population_counts[], double output[]) {
-  uint32_t sum = 0;
-  int nonzeros = 0;
-  int i;
-  for (i = 0; i < num_symbols; ++i) {
-    sum += population_counts[i];
-    if (population_counts[i] > 0) {
-      ++nonzeros;
-    }
-  }
-  if (nonzeros <= 1) {
-    memset(output, 0, num_symbols * sizeof(*output));
-  } else {
-    const double logsum = VP8LFastLog2(sum);
-    for (i = 0; i < num_symbols; ++i) {
-      output[i] = logsum - VP8LFastLog2(population_counts[i]);
-    }
-  }
-}
-
-static int CostModelBuild(CostModel* const m, int cache_bits,
-                          VP8LBackwardRefs* const refs) {
-  int ok = 0;
-  VP8LHistogram* const histo = VP8LAllocateHistogram(cache_bits);
-  if (histo == NULL) goto Error;
-
-  VP8LHistogramCreate(histo, refs, cache_bits);
-
-  ConvertPopulationCountTableToBitEstimates(
-      VP8LHistogramNumCodes(histo->palette_code_bits_),
-      histo->literal_, m->literal_);
-  ConvertPopulationCountTableToBitEstimates(
-      VALUES_IN_BYTE, histo->red_, m->red_);
-  ConvertPopulationCountTableToBitEstimates(
-      VALUES_IN_BYTE, histo->blue_, m->blue_);
-  ConvertPopulationCountTableToBitEstimates(
-      VALUES_IN_BYTE, histo->alpha_, m->alpha_);
-  ConvertPopulationCountTableToBitEstimates(
-      NUM_DISTANCE_CODES, histo->distance_, m->distance_);
-  ok = 1;
-
- Error:
-  VP8LFreeHistogram(histo);
-  return ok;
-}
-
-static WEBP_INLINE double GetLiteralCost(const CostModel* const m, uint32_t v) {
-  return m->alpha_[v >> 24] +
-         m->red_[(v >> 16) & 0xff] +
-         m->literal_[(v >> 8) & 0xff] +
-         m->blue_[v & 0xff];
-}
-
-static WEBP_INLINE double GetCacheCost(const CostModel* const m, uint32_t idx) {
-  const int literal_idx = VALUES_IN_BYTE + NUM_LENGTH_CODES + idx;
-  return m->literal_[literal_idx];
-}
-
-static WEBP_INLINE double GetLengthCost(const CostModel* const m,
-                                        uint32_t length) {
-  int code, extra_bits;
-  VP8LPrefixEncodeBits(length, &code, &extra_bits);
-  return m->literal_[VALUES_IN_BYTE + code] + extra_bits;
-}
-
-static WEBP_INLINE double GetDistanceCost(const CostModel* const m,
-                                          uint32_t distance) {
-  int code, extra_bits;
-  VP8LPrefixEncodeBits(distance, &code, &extra_bits);
-  return m->distance_[code] + extra_bits;
-}
-
-static void AddSingleLiteralWithCostModel(const uint32_t* const argb,
-                                          VP8LColorCache* const hashers,
-                                          const CostModel* const cost_model,
-                                          int idx, int use_color_cache,
-                                          double prev_cost, float* const cost,
-                                          uint16_t* const dist_array) {
-  double cost_val = prev_cost;
-  const uint32_t color = argb[0];
-  const int ix = use_color_cache ? VP8LColorCacheContains(hashers, color) : -1;
-  if (ix >= 0) {
-    // use_color_cache is true and hashers contains color
-    const double mul0 = 0.68;
-    cost_val += GetCacheCost(cost_model, ix) * mul0;
-  } else {
-    const double mul1 = 0.82;
-    if (use_color_cache) VP8LColorCacheInsert(hashers, color);
-    cost_val += GetLiteralCost(cost_model, color) * mul1;
-  }
-  if (cost[idx] > cost_val) {
-    cost[idx] = (float)cost_val;
-    dist_array[idx] = 1;  // only one is inserted.
-  }
-}
-
-// -----------------------------------------------------------------------------
-// CostManager and interval handling
-
-// Empirical value to avoid high memory consumption but good for performance.
-#define COST_CACHE_INTERVAL_SIZE_MAX 100
-
-// To perform backward reference every pixel at index index_ is considered and
-// the cost for the MAX_LENGTH following pixels computed. Those following pixels
-// at index index_ + k (k from 0 to MAX_LENGTH) have a cost of:
-//     distance_cost_ at index_ + GetLengthCost(cost_model, k)
-//            (named cost)            (named cached cost)
-// and the minimum value is kept. GetLengthCost(cost_model, k) is cached in an
-// array of size MAX_LENGTH.
-// Instead of performing MAX_LENGTH comparisons per pixel, we keep track of the
-// minimal values using intervals, for which lower_ and upper_ bounds are kept.
-// An interval is defined by the index_ of the pixel that generated it and
-// is only useful in a range of indices from start_ to end_ (exclusive), i.e.
-// it contains the minimum value for pixels between start_ and end_.
-// Intervals are stored in a linked list and ordered by start_. When a new
-// interval has a better minimum, old intervals are split or removed.
-typedef struct CostInterval CostInterval;
-struct CostInterval {
-  double lower_;
-  double upper_;
-  int start_;
-  int end_;
-  double distance_cost_;
-  int index_;
-  CostInterval* previous_;
-  CostInterval* next_;
-};
-
-// The GetLengthCost(cost_model, k) part of the costs is also bounded for
-// efficiency in a set of intervals of a different type.
-// If those intervals are small enough, they are not used for comparison and
-// written into the costs right away.
-typedef struct {
-  double lower_;  // Lower bound of the interval.
-  double upper_;  // Upper bound of the interval.
-  int start_;
-  int end_;       // Exclusive.
-  int do_write_;  // If !=0, the interval is saved to cost instead of being kept
-                  // for comparison.
-} CostCacheInterval;
-
-// This structure is in charge of managing intervals and costs.
-// It caches the different CostCacheInterval, caches the different
-// GetLengthCost(cost_model, k) in cost_cache_ and the CostInterval's (whose
-// count_ is limited by COST_CACHE_INTERVAL_SIZE_MAX).
-#define COST_MANAGER_MAX_FREE_LIST 10
-typedef struct {
-  CostInterval* head_;
-  int count_;  // The number of stored intervals.
-  CostCacheInterval* cache_intervals_;
-  size_t cache_intervals_size_;
-  double cost_cache_[MAX_LENGTH];  // Contains the GetLengthCost(cost_model, k).
-  double min_cost_cache_;          // The minimum value in cost_cache_[1:].
-  double max_cost_cache_;          // The maximum value in cost_cache_[1:].
-  float* costs_;
-  uint16_t* dist_array_;
-  // Most of the time, we only need few intervals -> use a free-list, to avoid
-  // fragmentation with small allocs in most common cases.
-  CostInterval intervals_[COST_MANAGER_MAX_FREE_LIST];
-  CostInterval* free_intervals_;
-  // These are regularly malloc'd remains. This list can't grow larger than than
-  // size COST_CACHE_INTERVAL_SIZE_MAX - COST_MANAGER_MAX_FREE_LIST, note.
-  CostInterval* recycled_intervals_;
-  // Buffer used in BackwardReferencesHashChainDistanceOnly to store the ends
-  // of the intervals that can have impacted the cost at a pixel.
-  int* interval_ends_;
-  int interval_ends_size_;
-} CostManager;
-
-static int IsCostCacheIntervalWritable(int start, int end) {
-  // 100 is the length for which we consider an interval for comparison, and not
-  // for writing.
-  // The first intervals are very small and go in increasing size. This constant
-  // helps merging them into one big interval (up to index 150/200 usually from
-  // which intervals start getting much bigger).
-  // This value is empirical.
-  return (end - start + 1 < 100);
-}
-
-static void CostIntervalAddToFreeList(CostManager* const manager,
-                                      CostInterval* const interval) {
-  interval->next_ = manager->free_intervals_;
-  manager->free_intervals_ = interval;
-}
-
-static int CostIntervalIsInFreeList(const CostManager* const manager,
-                                    const CostInterval* const interval) {
-  return (interval >= &manager->intervals_[0] &&
-          interval <= &manager->intervals_[COST_MANAGER_MAX_FREE_LIST - 1]);
-}
-
-static void CostManagerInitFreeList(CostManager* const manager) {
-  int i;
-  manager->free_intervals_ = NULL;
-  for (i = 0; i < COST_MANAGER_MAX_FREE_LIST; ++i) {
-    CostIntervalAddToFreeList(manager, &manager->intervals_[i]);
-  }
-}
-
-static void DeleteIntervalList(CostManager* const manager,
-                               const CostInterval* interval) {
-  while (interval != NULL) {
-    const CostInterval* const next = interval->next_;
-    if (!CostIntervalIsInFreeList(manager, interval)) {
-      WebPSafeFree((void*)interval);
-    }  // else: do nothing
-    interval = next;
-  }
-}
-
-static void CostManagerClear(CostManager* const manager) {
-  if (manager == NULL) return;
-
-  WebPSafeFree(manager->costs_);
-  WebPSafeFree(manager->cache_intervals_);
-  WebPSafeFree(manager->interval_ends_);
-
-  // Clear the interval lists.
-  DeleteIntervalList(manager, manager->head_);
-  manager->head_ = NULL;
-  DeleteIntervalList(manager, manager->recycled_intervals_);
-  manager->recycled_intervals_ = NULL;
-
-  // Reset pointers, count_ and cache_intervals_size_.
-  memset(manager, 0, sizeof(*manager));
-  CostManagerInitFreeList(manager);
-}
-
-static int CostManagerInit(CostManager* const manager,
-                           uint16_t* const dist_array, int pix_count,
-                           const CostModel* const cost_model) {
+// Compute an LZ77 by forcing matches to happen within a given distance cost.
+// We therefore limit the algorithm to the lowest 32 values in the PlaneCode
+// definition.
+#define WINDOW_OFFSETS_SIZE_MAX 32
+static int BackwardReferencesLz77Box(int xsize, int ysize,
+                                     const uint32_t* const argb, int cache_bits,
+                                     const VP8LHashChain* const hash_chain_best,
+                                     VP8LHashChain* hash_chain,
+                                     VP8LBackwardRefs* const refs) {
   int i;
-  const int cost_cache_size = (pix_count > MAX_LENGTH) ? MAX_LENGTH : pix_count;
-  // This constant is tied to the cost_model we use.
-  // Empirically, differences between intervals is usually of more than 1.
-  const double min_cost_diff = 0.1;
-
-  manager->costs_ = NULL;
-  manager->cache_intervals_ = NULL;
-  manager->interval_ends_ = NULL;
-  manager->head_ = NULL;
-  manager->recycled_intervals_ = NULL;
-  manager->count_ = 0;
-  manager->dist_array_ = dist_array;
-  CostManagerInitFreeList(manager);
-
-  // Fill in the cost_cache_.
-  manager->cache_intervals_size_ = 1;
-  manager->cost_cache_[0] = 0;
-  for (i = 1; i < cost_cache_size; ++i) {
-    manager->cost_cache_[i] = GetLengthCost(cost_model, i);
-    // Get an approximation of the number of bound intervals.
-    if (fabs(manager->cost_cache_[i] - manager->cost_cache_[i - 1]) >
-        min_cost_diff) {
-      ++manager->cache_intervals_size_;
-    }
-    // Compute the minimum of cost_cache_.
-    if (i == 1) {
-      manager->min_cost_cache_ = manager->cost_cache_[1];
-      manager->max_cost_cache_ = manager->cost_cache_[1];
-    } else if (manager->cost_cache_[i] < manager->min_cost_cache_) {
-      manager->min_cost_cache_ = manager->cost_cache_[i];
-    } else if (manager->cost_cache_[i] > manager->max_cost_cache_) {
-      manager->max_cost_cache_ = manager->cost_cache_[i];
-    }
-  }
-
-  // With the current cost models, we have 15 intervals, so we are safe by
-  // setting a maximum of COST_CACHE_INTERVAL_SIZE_MAX.
-  if (manager->cache_intervals_size_ > COST_CACHE_INTERVAL_SIZE_MAX) {
-    manager->cache_intervals_size_ = COST_CACHE_INTERVAL_SIZE_MAX;
-  }
-  manager->cache_intervals_ = (CostCacheInterval*)WebPSafeMalloc(
-      manager->cache_intervals_size_, sizeof(*manager->cache_intervals_));
-  if (manager->cache_intervals_ == NULL) {
-    CostManagerClear(manager);
-    return 0;
-  }
-
-  // Fill in the cache_intervals_.
-  {
-    double cost_prev = -1e38f;  // unprobably low initial value
-    CostCacheInterval* prev = NULL;
-    CostCacheInterval* cur = manager->cache_intervals_;
-    const CostCacheInterval* const end =
-        manager->cache_intervals_ + manager->cache_intervals_size_;
-
-    // Consecutive values in cost_cache_ are compared and if a big enough
-    // difference is found, a new interval is created and bounded.
-    for (i = 0; i < cost_cache_size; ++i) {
-      const double cost_val = manager->cost_cache_[i];
-      if (i == 0 ||
-          (fabs(cost_val - cost_prev) > min_cost_diff && cur + 1 < end)) {
-        if (i > 1) {
-          const int is_writable =
-              IsCostCacheIntervalWritable(cur->start_, cur->end_);
-          // Merge with the previous interval if both are writable.
-          if (is_writable && cur != manager->cache_intervals_ &&
-              prev->do_write_) {
-            // Update the previous interval.
-            prev->end_ = cur->end_;
-            if (cur->lower_ < prev->lower_) {
-              prev->lower_ = cur->lower_;
-            } else if (cur->upper_ > prev->upper_) {
-              prev->upper_ = cur->upper_;
-            }
-          } else {
-            cur->do_write_ = is_writable;
-            prev = cur;
-            ++cur;
-          }
-        }
-        // Initialize an interval.
-        cur->start_ = i;
-        cur->do_write_ = 0;
-        cur->lower_ = cost_val;
-        cur->upper_ = cost_val;
-      } else {
-        // Update the current interval bounds.
-        if (cost_val < cur->lower_) {
-          cur->lower_ = cost_val;
-        } else if (cost_val > cur->upper_) {
-          cur->upper_ = cost_val;
-        }
-      }
-      cur->end_ = i + 1;
-      cost_prev = cost_val;
+  const int pix_count = xsize * ysize;
+  uint16_t* counts;
+  int window_offsets[WINDOW_OFFSETS_SIZE_MAX] = {0};
+  int window_offsets_new[WINDOW_OFFSETS_SIZE_MAX] = {0};
+  int window_offsets_size = 0;
+  int window_offsets_new_size = 0;
+  uint16_t* const counts_ini =
+      (uint16_t*)WebPSafeMalloc(xsize * ysize, sizeof(*counts_ini));
+  int best_offset_prev = -1, best_length_prev = -1;
+  if (counts_ini == NULL) return 0;
+
+  // counts[i] counts how many times a pixel is repeated starting at position i.
+  i = pix_count - 2;
+  counts = counts_ini + i;
+  counts[1] = 1;
+  for (; i >= 0; --i, --counts) {
+    if (argb[i] == argb[i + 1]) {
+      // Max out the counts to MAX_LENGTH.
+      counts[0] = counts[1] + (counts[1] != MAX_LENGTH);
+    } else {
+      counts[0] = 1;
     }
-    manager->cache_intervals_size_ = cur + 1 - manager->cache_intervals_;
   }
 
-  manager->costs_ = (float*)WebPSafeMalloc(pix_count, sizeof(*manager->costs_));
-  if (manager->costs_ == NULL) {
-    CostManagerClear(manager);
-    return 0;
-  }
-  // Set the initial costs_ high for every pixel as we will keep the minimum.
-  for (i = 0; i < pix_count; ++i) manager->costs_[i] = 1e38f;
-
-  // The cost at pixel is influenced by the cost intervals from previous pixels.
-  // Let us take the specific case where the offset is the same (which actually
-  // happens a lot in case of uniform regions).
-  // pixel i contributes to j>i a cost of: offset cost + cost_cache_[j-i]
-  // pixel i+1 contributes to j>i a cost of: 2*offset cost + cost_cache_[j-i-1]
-  // pixel i+2 contributes to j>i a cost of: 3*offset cost + cost_cache_[j-i-2]
-  // and so on.
-  // A pixel i influences the following length(j) < MAX_LENGTH pixels. What is
-  // the value of j such that pixel i + j cannot influence any of those pixels?
-  // This value is such that:
-  //               max of cost_cache_ < j*offset cost + min of cost_cache_
-  // (pixel i + j 's cost cannot beat the worst cost given by pixel i).
-  // This value will be used to optimize the cost computation in
-  // BackwardReferencesHashChainDistanceOnly.
+  // Figure out the window offsets around a pixel. They are stored in a
+  // spiraling order around the pixel as defined by VP8LDistanceToPlaneCode.
   {
-    // The offset cost is computed in GetDistanceCost and has a minimum value of
-    // the minimum in cost_model->distance_. The case where the offset cost is 0
-    // will be dealt with differently later so we are only interested in the
-    // minimum non-zero offset cost.
-    double offset_cost_min = 0.;
-    int size;
-    for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
-      if (cost_model->distance_[i] != 0) {
-        if (offset_cost_min == 0.) {
-          offset_cost_min = cost_model->distance_[i];
-        } else if (cost_model->distance_[i] < offset_cost_min) {
-          offset_cost_min = cost_model->distance_[i];
-        }
+    int x, y;
+    for (y = 0; y <= 6; ++y) {
+      for (x = -6; x <= 6; ++x) {
+        const int offset = y * xsize + x;
+        int plane_code;
+        // Ignore offsets that bring us after the pixel.
+        if (offset <= 0) continue;
+        plane_code = VP8LDistanceToPlaneCode(xsize, offset) - 1;
+        if (plane_code >= WINDOW_OFFSETS_SIZE_MAX) continue;
+        window_offsets[plane_code] = offset;
       }
     }
-    // In case all the cost_model->distance_ is 0, the next non-zero cost we
-    // can have is from the extra bit in GetDistanceCost, hence 1.
-    if (offset_cost_min < 1.) offset_cost_min = 1.;
-
-    size = 1 + (int)ceil((manager->max_cost_cache_ - manager->min_cost_cache_) /
-                         offset_cost_min);
-    // Empirically, we usually end up with a value below 100.
-    if (size > MAX_LENGTH) size = MAX_LENGTH;
-
-    manager->interval_ends_ =
-        (int*)WebPSafeMalloc(size, sizeof(*manager->interval_ends_));
-    if (manager->interval_ends_ == NULL) {
-      CostManagerClear(manager);
-      return 0;
-    }
-    manager->interval_ends_size_ = size;
-  }
-
-  return 1;
-}
-
-// Given the distance_cost for pixel 'index', update the cost at pixel 'i' if it
-// is smaller than the previously computed value.
-static WEBP_INLINE void UpdateCost(CostManager* const manager, int i, int index,
-                                   double distance_cost) {
-  int k = i - index;
-  double cost_tmp;
-  assert(k >= 0 && k < MAX_LENGTH);
-  cost_tmp = distance_cost + manager->cost_cache_[k];
-
-  if (manager->costs_[i] > cost_tmp) {
-    manager->costs_[i] = (float)cost_tmp;
-    manager->dist_array_[i] = k + 1;
-  }
-}
-
-// Given the distance_cost for pixel 'index', update the cost for all the pixels
-// between 'start' and 'end' excluded.
-static WEBP_INLINE void UpdateCostPerInterval(CostManager* const manager,
-                                              int start, int end, int index,
-                                              double distance_cost) {
-  int i;
-  for (i = start; i < end; ++i) UpdateCost(manager, i, index, distance_cost);
-}
-
-// Given two intervals, make 'prev' be the previous one of 'next' in 'manager'.
-static WEBP_INLINE void ConnectIntervals(CostManager* const manager,
-                                         CostInterval* const prev,
-                                         CostInterval* const next) {
-  if (prev != NULL) {
-    prev->next_ = next;
-  } else {
-    manager->head_ = next;
-  }
-
-  if (next != NULL) next->previous_ = prev;
-}
-
-// Pop an interval in the manager.
-static WEBP_INLINE void PopInterval(CostManager* const manager,
-                                    CostInterval* const interval) {
-  CostInterval* const next = interval->next_;
-
-  if (interval == NULL) return;
-
-  ConnectIntervals(manager, interval->previous_, next);
-  if (CostIntervalIsInFreeList(manager, interval)) {
-    CostIntervalAddToFreeList(manager, interval);
-  } else {  // recycle regularly malloc'd intervals too
-    interval->next_ = manager->recycled_intervals_;
-    manager->recycled_intervals_ = interval;
-  }
-  --manager->count_;
-  assert(manager->count_ >= 0);
-}
-
-// Update the cost at index i by going over all the stored intervals that
-// overlap with i.
-static WEBP_INLINE void UpdateCostPerIndex(CostManager* const manager, int i) {
-  CostInterval* current = manager->head_;
-
-  while (current != NULL && current->start_ <= i) {
-    if (current->end_ <= i) {
-      // We have an outdated interval, remove it.
-      CostInterval* next = current->next_;
-      PopInterval(manager, current);
-      current = next;
-    } else {
-      UpdateCost(manager, i, current->index_, current->distance_cost_);
-      current = current->next_;
-    }
-  }
-}
-
-// Given a current orphan interval and its previous interval, before
-// it was orphaned (which can be NULL), set it at the right place in the list
-// of intervals using the start_ ordering and the previous interval as a hint.
-static WEBP_INLINE void PositionOrphanInterval(CostManager* const manager,
-                                               CostInterval* const current,
-                                               CostInterval* previous) {
-  assert(current != NULL);
-
-  if (previous == NULL) previous = manager->head_;
-  while (previous != NULL && current->start_ < previous->start_) {
-    previous = previous->previous_;
-  }
-  while (previous != NULL && previous->next_ != NULL &&
-         previous->next_->start_ < current->start_) {
-    previous = previous->next_;
-  }
-
-  if (previous != NULL) {
-    ConnectIntervals(manager, current, previous->next_);
-  } else {
-    ConnectIntervals(manager, current, manager->head_);
-  }
-  ConnectIntervals(manager, previous, current);
-}
-
-// Insert an interval in the list contained in the manager by starting at
-// interval_in as a hint. The intervals are sorted by start_ value.
-static WEBP_INLINE void InsertInterval(CostManager* const manager,
-                                       CostInterval* const interval_in,
-                                       double distance_cost, double lower,
-                                       double upper, int index, int start,
-                                       int end) {
-  CostInterval* interval_new;
-
-  if (IsCostCacheIntervalWritable(start, end) ||
-      manager->count_ >= COST_CACHE_INTERVAL_SIZE_MAX) {
-    // Write down the interval if it is too small.
-    UpdateCostPerInterval(manager, start, end, index, distance_cost);
-    return;
-  }
-  if (manager->free_intervals_ != NULL) {
-    interval_new = manager->free_intervals_;
-    manager->free_intervals_ = interval_new->next_;
-  } else if (manager->recycled_intervals_ != NULL) {
-    interval_new = manager->recycled_intervals_;
-    manager->recycled_intervals_ = interval_new->next_;
-  } else {   // malloc for good
-    interval_new = (CostInterval*)WebPSafeMalloc(1, sizeof(*interval_new));
-    if (interval_new == NULL) {
-      // Write down the interval if we cannot create it.
-      UpdateCostPerInterval(manager, start, end, index, distance_cost);
-      return;
-    }
-  }
-
-  interval_new->distance_cost_ = distance_cost;
-  interval_new->lower_ = lower;
-  interval_new->upper_ = upper;
-  interval_new->index_ = index;
-  interval_new->start_ = start;
-  interval_new->end_ = end;
-  PositionOrphanInterval(manager, interval_new, interval_in);
-
-  ++manager->count_;
-}
-
-// When an interval has its start_ or end_ modified, it needs to be
-// repositioned in the linked list.
-static WEBP_INLINE void RepositionInterval(CostManager* const manager,
-                                           CostInterval* const interval) {
-  if (IsCostCacheIntervalWritable(interval->start_, interval->end_)) {
-    // Maybe interval has been resized and is small enough to be removed.
-    UpdateCostPerInterval(manager, interval->start_, interval->end_,
-                          interval->index_, interval->distance_cost_);
-    PopInterval(manager, interval);
-    return;
-  }
-
-  // Early exit if interval is at the right spot.
-  if ((interval->previous_ == NULL ||
-       interval->previous_->start_ <= interval->start_) &&
-      (interval->next_ == NULL ||
-       interval->start_ <= interval->next_->start_)) {
-    return;
-  }
-
-  ConnectIntervals(manager, interval->previous_, interval->next_);
-  PositionOrphanInterval(manager, interval, interval->previous_);
-}
-
-// Given a new cost interval defined by its start at index, its last value and
-// distance_cost, add its contributions to the previous intervals and costs.
-// If handling the interval or one of its subintervals becomes to heavy, its
-// contribution is added to the costs right away.
-static WEBP_INLINE void PushInterval(CostManager* const manager,
-                                     double distance_cost, int index,
-                                     int last) {
-  size_t i;
-  CostInterval* interval = manager->head_;
-  CostInterval* interval_next;
-  const CostCacheInterval* const cost_cache_intervals =
-      manager->cache_intervals_;
-
-  for (i = 0; i < manager->cache_intervals_size_ &&
-              cost_cache_intervals[i].start_ < last;
-       ++i) {
-    // Define the intersection of the ith interval with the new one.
-    int start = index + cost_cache_intervals[i].start_;
-    const int end = index + (cost_cache_intervals[i].end_ > last
-                                 ? last
-                                 : cost_cache_intervals[i].end_);
-    const double lower_in = cost_cache_intervals[i].lower_;
-    const double upper_in = cost_cache_intervals[i].upper_;
-    const double lower_full_in = distance_cost + lower_in;
-    const double upper_full_in = distance_cost + upper_in;
-
-    if (cost_cache_intervals[i].do_write_) {
-      UpdateCostPerInterval(manager, start, end, index, distance_cost);
-      continue;
+    // For narrow images, not all plane codes are reached, so remove those.
+    for (i = 0; i < WINDOW_OFFSETS_SIZE_MAX; ++i) {
+      if (window_offsets[i] == 0) continue;
+      window_offsets[window_offsets_size++] = window_offsets[i];
     }
-
-    for (; interval != NULL && interval->start_ < end && start < end;
-         interval = interval_next) {
-      const double lower_full_interval =
-          interval->distance_cost_ + interval->lower_;
-      const double upper_full_interval =
-          interval->distance_cost_ + interval->upper_;
-
-      interval_next = interval->next_;
-
-      // Make sure we have some overlap
-      if (start >= interval->end_) continue;
-
-      if (lower_full_in >= upper_full_interval) {
-        // When intervals are represented, the lower, the better.
-        // [**********************************************************]
-        // start                                                    end
-        //                   [----------------------------------]
-        //                   interval->start_       interval->end_
-        // If we are worse than what we already have, add whatever we have so
-        // far up to interval.
-        const int start_new = interval->end_;
-        InsertInterval(manager, interval, distance_cost, lower_in, upper_in,
-                       index, start, interval->start_);
-        start = start_new;
-        continue;
+    // Given a pixel P, find the offsets that reach pixels unreachable from P-1
+    // with any of the offsets in window_offsets[].
+    for (i = 0; i < window_offsets_size; ++i) {
+      int j;
+      int is_reachable = 0;
+      for (j = 0; j < window_offsets_size && !is_reachable; ++j) {
+        is_reachable |= (window_offsets[i] == window_offsets[j] + 1);
       }
-
-      // We know the two intervals intersect.
-      if (upper_full_in >= lower_full_interval) {
-        // There is no clear cut on which is best, so let's keep both.
-        // [*********[*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*]***********]
-        // start     interval->start_     interval->end_         end
-        // OR
-        // [*********[*-*-*-*-*-*-*-*-*-*-*-]----------------------]
-        // start     interval->start_     end          interval->end_
-        const int end_new = (interval->end_ <= end) ? interval->end_ : end;
-        InsertInterval(manager, interval, distance_cost, lower_in, upper_in,
-                       index, start, end_new);
-        start = end_new;
-      } else if (start <= interval->start_ && interval->end_ <= end) {
-        //                   [----------------------------------]
-        //                   interval->start_       interval->end_
-        // [**************************************************************]
-        // start                                                        end
-        // We can safely remove the old interval as it is fully included.
-        PopInterval(manager, interval);
-      } else {
-        if (interval->start_ <= start && end <= interval->end_) {
-          // [--------------------------------------------------------------]
-          // interval->start_                                  interval->end_
-          //                     [*****************************]
-          //                     start                       end
-          // We have to split the old interval as it fully contains the new one.
-          const int end_original = interval->end_;
-          interval->end_ = start;
-          InsertInterval(manager, interval, interval->distance_cost_,
-                         interval->lower_, interval->upper_, interval->index_,
-                         end, end_original);
-        } else if (interval->start_ < start) {
-          // [------------------------------------]
-          // interval->start_        interval->end_
-          //                     [*****************************]
-          //                     start                       end
-          interval->end_ = start;
-        } else {
-          //              [------------------------------------]
-          //              interval->start_        interval->end_
-          // [*****************************]
-          // start                       end
-          interval->start_ = end;
-        }
-
-        // The interval has been modified, we need to reposition it or write it.
-        RepositionInterval(manager, interval);
+      if (!is_reachable) {
+        window_offsets_new[window_offsets_new_size] = window_offsets[i];
+        ++window_offsets_new_size;
       }
     }
-    // Insert the remaining interval from start to end.
-    InsertInterval(manager, interval, distance_cost, lower_in, upper_in, index,
-                   start, end);
-  }
-}
-
-static int BackwardReferencesHashChainDistanceOnly(
-    int xsize, int ysize, const uint32_t* const argb, int quality,
-    int cache_bits, const VP8LHashChain* const hash_chain,
-    VP8LBackwardRefs* const refs, uint16_t* const dist_array) {
-  int i;
-  int ok = 0;
-  int cc_init = 0;
-  const int pix_count = xsize * ysize;
-  const int use_color_cache = (cache_bits > 0);
-  const size_t literal_array_size = sizeof(double) *
-      (NUM_LITERAL_CODES + NUM_LENGTH_CODES +
-       ((cache_bits > 0) ? (1 << cache_bits) : 0));
-  const size_t cost_model_size = sizeof(CostModel) + literal_array_size;
-  CostModel* const cost_model =
-      (CostModel*)WebPSafeCalloc(1ULL, cost_model_size);
-  VP8LColorCache hashers;
-  const int skip_length = 32 + quality;
-  const int skip_min_distance_code = 2;
-  CostManager* cost_manager =
-      (CostManager*)WebPSafeMalloc(1ULL, sizeof(*cost_manager));
-
-  if (cost_model == NULL || cost_manager == NULL) goto Error;
-
-  cost_model->literal_ = (double*)(cost_model + 1);
-  if (use_color_cache) {
-    cc_init = VP8LColorCacheInit(&hashers, cache_bits);
-    if (!cc_init) goto Error;
   }
 
-  if (!CostModelBuild(cost_model, cache_bits, refs)) {
-    goto Error;
-  }
-
-  if (!CostManagerInit(cost_manager, dist_array, pix_count, cost_model)) {
-    goto Error;
-  }
-
-  // We loop one pixel at a time, but store all currently best points to
-  // non-processed locations from this point.
-  dist_array[0] = 0;
-  // Add first pixel as literal.
-  AddSingleLiteralWithCostModel(argb + 0, &hashers, cost_model, 0,
-                                use_color_cache, 0.0, cost_manager->costs_,
-                                dist_array);
-
-  for (i = 1; i < pix_count - 1; ++i) {
-    int offset = 0, len = 0;
-    double prev_cost = cost_manager->costs_[i - 1];
-    HashChainFindCopy(hash_chain, i, &offset, &len);
-    if (len >= 2) {
-      // If we are dealing with a non-literal.
-      const int code = DistanceToPlaneCode(xsize, offset);
-      const double offset_cost = GetDistanceCost(cost_model, code);
-      const int first_i = i;
-      int j_max = 0, interval_ends_index = 0;
-      const int is_offset_zero = (offset_cost == 0.);
-
-      if (!is_offset_zero) {
-        j_max = (int)ceil(
-            (cost_manager->max_cost_cache_ - cost_manager->min_cost_cache_) /
-            offset_cost);
-        if (j_max < 1) {
-          j_max = 1;
-        } else if (j_max > cost_manager->interval_ends_size_ - 1) {
-          // This could only happen in the case of MAX_LENGTH.
-          j_max = cost_manager->interval_ends_size_ - 1;
+  hash_chain->offset_length_[0] = 0;
+  for (i = 1; i < pix_count; ++i) {
+    int ind;
+    int best_length = VP8LHashChainFindLength(hash_chain_best, i);
+    int best_offset;
+    int do_compute = 1;
+
+    if (best_length >= MAX_LENGTH) {
+      // Do not recompute the best match if we already have a maximal one in the
+      // window.
+      best_offset = VP8LHashChainFindOffset(hash_chain_best, i);
+      for (ind = 0; ind < window_offsets_size; ++ind) {
+        if (best_offset == window_offsets[ind]) {
+          do_compute = 0;
+          break;
         }
-      }  // else j_max is unused anyway.
-
-      // Instead of considering all contributions from a pixel i by calling:
-      //         PushInterval(cost_manager, prev_cost + offset_cost, i, len);
-      // we optimize these contributions in case offset_cost stays the same for
-      // consecutive pixels. This describes a set of pixels similar to a
-      // previous set (e.g. constant color regions).
-      for (; i < pix_count - 1; ++i) {
-        int offset_next, len_next;
-        prev_cost = cost_manager->costs_[i - 1];
-
-        if (is_offset_zero) {
-          // No optimization can be made so we just push all of the
-          // contributions from i.
-          PushInterval(cost_manager, prev_cost, i, len);
-        } else {
-          // j_max is chosen as the smallest j such that:
-          //       max of cost_cache_ < j*offset cost + min of cost_cache_
-          // Therefore, the pixel influenced by i-j_max, cannot be influenced
-          // by i. Only the costs after the end of what i contributed need to be
-          // updated. cost_manager->interval_ends_ is a circular buffer that
-          // stores those ends.
-          const double distance_cost = prev_cost + offset_cost;
-          int j = cost_manager->interval_ends_[interval_ends_index];
-          if (i - first_i <= j_max ||
-              !IsCostCacheIntervalWritable(j, i + len)) {
-            PushInterval(cost_manager, distance_cost, i, len);
-          } else {
-            for (; j < i + len; ++j) {
-              UpdateCost(cost_manager, j, i, distance_cost);
-            }
-          }
-          // Store the new end in the circular buffer.
-          assert(interval_ends_index < cost_manager->interval_ends_size_);
-          cost_manager->interval_ends_[interval_ends_index] = i + len;
-          if (++interval_ends_index > j_max) interval_ends_index = 0;
-        }
-
-        // Check whether i is the last pixel to consider, as it is handled
-        // differently.
-        if (i + 1 >= pix_count - 1) break;
-        HashChainFindCopy(hash_chain, i + 1, &offset_next, &len_next);
-        if (offset_next != offset) break;
-        len = len_next;
-        UpdateCostPerIndex(cost_manager, i);
-        AddSingleLiteralWithCostModel(argb + i, &hashers, cost_model, i,
-                                      use_color_cache, prev_cost,
-                                      cost_manager->costs_, dist_array);
       }
-      // Submit the last pixel.
-      UpdateCostPerIndex(cost_manager, i + 1);
-
-      // This if is for speedup only. It roughly doubles the speed, and
-      // makes compression worse by .1 %.
-      if (len >= skip_length && code <= skip_min_distance_code) {
-        // Long copy for short distances, let's skip the middle
-        // lookups for better copies.
-        // 1) insert the hashes.
-        if (use_color_cache) {
-          int k;
-          for (k = 0; k < len; ++k) {
-            VP8LColorCacheInsert(&hashers, argb[i + k]);
+    }
+    if (do_compute) {
+      // Figure out if we should use the offset/length from the previous pixel
+      // as an initial guess and therefore only inspect the offsets in
+      // window_offsets_new[].
+      const int use_prev =
+          (best_length_prev > 1) && (best_length_prev < MAX_LENGTH);
+      const int num_ind =
+          use_prev ? window_offsets_new_size : window_offsets_size;
+      best_length = use_prev ? best_length_prev - 1 : 0;
+      best_offset = use_prev ? best_offset_prev : 0;
+      // Find the longest match in a window around the pixel.
+      for (ind = 0; ind < num_ind; ++ind) {
+        int curr_length = 0;
+        int j = i;
+        int j_offset =
+            use_prev ? i - window_offsets_new[ind] : i - window_offsets[ind];
+        if (j_offset < 0 || argb[j_offset] != argb[i]) continue;
+        // The longest match is the sum of how many times each pixel is
+        // repeated.
+        do {
+          const int counts_j_offset = counts_ini[j_offset];
+          const int counts_j = counts_ini[j];
+          if (counts_j_offset != counts_j) {
+            curr_length +=
+                (counts_j_offset < counts_j) ? counts_j_offset : counts_j;
+            break;
+          }
+          // The same color is repeated counts_pos times at j_offset and j.
+          curr_length += counts_j_offset;
+          j_offset += counts_j_offset;
+          j += counts_j_offset;
+        } while (curr_length <= MAX_LENGTH && j < pix_count &&
+                 argb[j_offset] == argb[j]);
+        if (best_length < curr_length) {
+          best_offset =
+              use_prev ? window_offsets_new[ind] : window_offsets[ind];
+          if (curr_length >= MAX_LENGTH) {
+            best_length = MAX_LENGTH;
+            break;
+          } else {
+            best_length = curr_length;
           }
-        }
-        // 2) jump.
-        {
-          const int i_next = i + len - 1;  // for loop does ++i, thus -1 here.
-          for (; i <= i_next; ++i) UpdateCostPerIndex(cost_manager, i + 1);
-          i = i_next;
-        }
-        goto next_symbol;
-      }
-      if (len > 2) {
-        // Also try the smallest interval possible (size 2).
-        double cost_total =
-            prev_cost + offset_cost + GetLengthCost(cost_model, 1);
-        if (cost_manager->costs_[i + 1] > cost_total) {
-          cost_manager->costs_[i + 1] = (float)cost_total;
-          dist_array[i + 1] = 2;
         }
       }
-    } else {
-      // The pixel is added as a single literal so just update the costs.
-      UpdateCostPerIndex(cost_manager, i + 1);
     }
 
-    AddSingleLiteralWithCostModel(argb + i, &hashers, cost_model, i,
-                                  use_color_cache, prev_cost,
-                                  cost_manager->costs_, dist_array);
-
- next_symbol: ;
-  }
-  // Handle the last pixel.
-  if (i == (pix_count - 1)) {
-    AddSingleLiteralWithCostModel(
-        argb + i, &hashers, cost_model, i, use_color_cache,
-        cost_manager->costs_[pix_count - 2], cost_manager->costs_, dist_array);
-  }
-
-  ok = !refs->error_;
- Error:
-  if (cc_init) VP8LColorCacheClear(&hashers);
-  CostManagerClear(cost_manager);
-  WebPSafeFree(cost_model);
-  WebPSafeFree(cost_manager);
-  return ok;
-}
-
-// We pack the path at the end of *dist_array and return
-// a pointer to this part of the array. Example:
-// dist_array = [1x2xx3x2] => packed [1x2x1232], chosen_path = [1232]
-static void TraceBackwards(uint16_t* const dist_array,
-                           int dist_array_size,
-                           uint16_t** const chosen_path,
-                           int* const chosen_path_size) {
-  uint16_t* path = dist_array + dist_array_size;
-  uint16_t* cur = dist_array + dist_array_size - 1;
-  while (cur >= dist_array) {
-    const int k = *cur;
-    --path;
-    *path = k;
-    cur -= k;
-  }
-  *chosen_path = path;
-  *chosen_path_size = (int)(dist_array + dist_array_size - path);
-}
-
-static int BackwardReferencesHashChainFollowChosenPath(
-    const uint32_t* const argb, int cache_bits,
-    const uint16_t* const chosen_path, int chosen_path_size,
-    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs) {
-  const int use_color_cache = (cache_bits > 0);
-  int ix;
-  int i = 0;
-  int ok = 0;
-  int cc_init = 0;
-  VP8LColorCache hashers;
-
-  if (use_color_cache) {
-    cc_init = VP8LColorCacheInit(&hashers, cache_bits);
-    if (!cc_init) goto Error;
-  }
-
-  ClearBackwardRefs(refs);
-  for (ix = 0; ix < chosen_path_size; ++ix) {
-    const int len = chosen_path[ix];
-    if (len != 1) {
-      int k;
-      const int offset = HashChainFindOffset(hash_chain, i);
-      BackwardRefsCursorAdd(refs, PixOrCopyCreateCopy(offset, len));
-      if (use_color_cache) {
-        for (k = 0; k < len; ++k) {
-          VP8LColorCacheInsert(&hashers, argb[i + k]);
-        }
-      }
-      i += len;
+    assert(i + best_length <= pix_count);
+    assert(best_length <= MAX_LENGTH);
+    if (best_length <= MIN_LENGTH) {
+      hash_chain->offset_length_[i] = 0;
+      best_offset_prev = 0;
+      best_length_prev = 0;
     } else {
-      PixOrCopy v;
-      const int idx =
-          use_color_cache ? VP8LColorCacheContains(&hashers, argb[i]) : -1;
-      if (idx >= 0) {
-        // use_color_cache is true and hashers contains argb[i]
-        // push pixel as a color cache index
-        v = PixOrCopyCreateCacheIdx(idx);
-      } else {
-        if (use_color_cache) VP8LColorCacheInsert(&hashers, argb[i]);
-        v = PixOrCopyCreateLiteral(argb[i]);
-      }
-      BackwardRefsCursorAdd(refs, v);
-      ++i;
+      hash_chain->offset_length_[i] =
+          (best_offset << MAX_LENGTH_BITS) | (uint32_t)best_length;
+      best_offset_prev = best_offset;
+      best_length_prev = best_length;
     }
   }
-  ok = !refs->error_;
- Error:
-  if (cc_init) VP8LColorCacheClear(&hashers);
-  return ok;
-}
+  hash_chain->offset_length_[0] = 0;
+  WebPSafeFree(counts_ini);
 
-// Returns 1 on success.
-static int BackwardReferencesTraceBackwards(
-    int xsize, int ysize, const uint32_t* const argb, int quality,
-    int cache_bits, const VP8LHashChain* const hash_chain,
-    VP8LBackwardRefs* const refs) {
-  int ok = 0;
-  const int dist_array_size = xsize * ysize;
-  uint16_t* chosen_path = NULL;
-  int chosen_path_size = 0;
-  uint16_t* dist_array =
-      (uint16_t*)WebPSafeMalloc(dist_array_size, sizeof(*dist_array));
-
-  if (dist_array == NULL) goto Error;
-
-  if (!BackwardReferencesHashChainDistanceOnly(
-      xsize, ysize, argb, quality, cache_bits, hash_chain,
-      refs, dist_array)) {
-    goto Error;
-  }
-  TraceBackwards(dist_array, dist_array_size, &chosen_path, &chosen_path_size);
-  if (!BackwardReferencesHashChainFollowChosenPath(
-          argb, cache_bits, chosen_path, chosen_path_size, hash_chain, refs)) {
-    goto Error;
-  }
-  ok = 1;
- Error:
-  WebPSafeFree(dist_array);
-  return ok;
+  return BackwardReferencesLz77(xsize, ysize, argb, cache_bits, hash_chain,
+                                refs);
 }
 
+// -----------------------------------------------------------------------------
+
 static void BackwardReferences2DLocality(int xsize,
                                          const VP8LBackwardRefs* const refs) {
   VP8LRefsCursor c = VP8LRefsCursorInit(refs);
   while (VP8LRefsCursorOk(&c)) {
     if (PixOrCopyIsCopy(c.cur_pos)) {
       const int dist = c.cur_pos->argb_or_distance;
-      const int transformed_dist = DistanceToPlaneCode(xsize, dist);
+      const int transformed_dist = VP8LDistanceToPlaneCode(xsize, dist);
       c.cur_pos->argb_or_distance = transformed_dist;
     }
     VP8LRefsCursorNext(&c);
   }
 }
 
-// Computes the entropies for a color cache size (in bits) between 0 (unused)
-// and cache_bits_max (inclusive).
-// Returns 1 on success, 0 in case of allocation error.
-static int ComputeCacheEntropies(const uint32_t* argb,
-                                 const VP8LBackwardRefs* const refs,
-                                 int cache_bits_max, double entropies[]) {
+// Evaluate optimal cache bits for the local color cache.
+// The input *best_cache_bits sets the maximum cache bits to use (passing 0
+// implies disabling the local color cache). The local color cache is also
+// disabled for the lower (<= 25) quality.
+// Returns 0 in case of memory error.
+static int CalculateBestCacheSize(const uint32_t* argb, int quality,
+                                  const VP8LBackwardRefs* const refs,
+                                  int* const best_cache_bits) {
+  int i;
+  const int cache_bits_max = (quality <= 25) ? 0 : *best_cache_bits;
+  double entropy_min = MAX_ENTROPY;
   int cc_init[MAX_COLOR_CACHE_BITS + 1] = { 0 };
   VP8LColorCache hashers[MAX_COLOR_CACHE_BITS + 1];
   VP8LRefsCursor c = VP8LRefsCursorInit(refs);
   VP8LHistogram* histos[MAX_COLOR_CACHE_BITS + 1] = { NULL };
   int ok = 0;
-  int i;
 
+  assert(cache_bits_max >= 0 && cache_bits_max <= MAX_COLOR_CACHE_BITS);
+
+  if (cache_bits_max == 0) {
+    *best_cache_bits = 0;
+    // Local color cache is disabled.
+    return 1;
+  }
+
+  // Allocate data.
   for (i = 0; i <= cache_bits_max; ++i) {
     histos[i] = VP8LAllocateHistogram(i);
     if (histos[i] == NULL) goto Error;
@@ -1547,60 +720,65 @@ static int ComputeCacheEntropies(const uint32_t* argb,
     if (!cc_init[i]) goto Error;
   }
 
-  assert(cache_bits_max >= 0);
-  // Do not use the color cache for cache_bits=0.
+  // Find the cache_bits giving the lowest entropy. The search is done in a
+  // brute-force way as the function (entropy w.r.t cache_bits) can be
+  // anything in practice.
   while (VP8LRefsCursorOk(&c)) {
-    VP8LHistogramAddSinglePixOrCopy(histos[0], c.cur_pos);
-    VP8LRefsCursorNext(&c);
-  }
-  if (cache_bits_max > 0) {
-    c = VP8LRefsCursorInit(refs);
-    while (VP8LRefsCursorOk(&c)) {
-      const PixOrCopy* const v = c.cur_pos;
-      if (PixOrCopyIsLiteral(v)) {
-        const uint32_t pix = *argb++;
-        // The keys of the caches can be derived from the longest one.
-        int key = HashPix(pix, 32 - cache_bits_max);
-        for (i = cache_bits_max; i >= 1; --i, key >>= 1) {
-          if (VP8LColorCacheLookup(&hashers[i], key) == pix) {
-            ++histos[i]->literal_[NUM_LITERAL_CODES + NUM_LENGTH_CODES + key];
-          } else {
-            VP8LColorCacheSet(&hashers[i], key, pix);
-            ++histos[i]->blue_[pix & 0xff];
-            ++histos[i]->literal_[(pix >> 8) & 0xff];
-            ++histos[i]->red_[(pix >> 16) & 0xff];
-            ++histos[i]->alpha_[pix >> 24];
-          }
-        }
-      } else {
-        // Update the histograms for distance/length.
-        int len = PixOrCopyLength(v);
-        int code_dist, code_len, extra_bits;
-        uint32_t argb_prev = *argb ^ 0xffffffffu;
-        VP8LPrefixEncodeBits(len, &code_len, &extra_bits);
-        VP8LPrefixEncodeBits(PixOrCopyDistance(v), &code_dist, &extra_bits);
-        for (i = 1; i <= cache_bits_max; ++i) {
-          ++histos[i]->literal_[NUM_LITERAL_CODES + code_len];
-          ++histos[i]->distance_[code_dist];
+    const PixOrCopy* const v = c.cur_pos;
+    if (PixOrCopyIsLiteral(v)) {
+      const uint32_t pix = *argb++;
+      const uint32_t a = (pix >> 24) & 0xff;
+      const uint32_t r = (pix >> 16) & 0xff;
+      const uint32_t g = (pix >>  8) & 0xff;
+      const uint32_t b = (pix >>  0) & 0xff;
+      // The keys of the caches can be derived from the longest one.
+      int key = VP8LHashPix(pix, 32 - cache_bits_max);
+      // Do not use the color cache for cache_bits = 0.
+      ++histos[0]->blue_[b];
+      ++histos[0]->literal_[g];
+      ++histos[0]->red_[r];
+      ++histos[0]->alpha_[a];
+      // Deal with cache_bits > 0.
+      for (i = cache_bits_max; i >= 1; --i, key >>= 1) {
+        if (VP8LColorCacheLookup(&hashers[i], key) == pix) {
+          ++histos[i]->literal_[NUM_LITERAL_CODES + NUM_LENGTH_CODES + key];
+        } else {
+          VP8LColorCacheSet(&hashers[i], key, pix);
+          ++histos[i]->blue_[b];
+          ++histos[i]->literal_[g];
+          ++histos[i]->red_[r];
+          ++histos[i]->alpha_[a];
         }
-        // Update the colors caches.
-        do {
-          if (*argb != argb_prev) {
-            // Efficiency: insert only if the color changes.
-            int key = HashPix(*argb, 32 - cache_bits_max);
-            for (i = cache_bits_max; i >= 1; --i, key >>= 1) {
-              hashers[i].colors_[key] = *argb;
-            }
-            argb_prev = *argb;
-          }
-          argb++;
-        } while (--len != 0);
       }
-      VP8LRefsCursorNext(&c);
+    } else {
+      // We should compute the contribution of the (distance,length)
+      // histograms but those are the same independently from the cache size.
+      // As those constant contributions are in the end added to the other
+      // histogram contributions, we can safely ignore them.
+      int len = PixOrCopyLength(v);
+      uint32_t argb_prev = *argb ^ 0xffffffffu;
+      // Update the color caches.
+      do {
+        if (*argb != argb_prev) {
+          // Efficiency: insert only if the color changes.
+          int key = VP8LHashPix(*argb, 32 - cache_bits_max);
+          for (i = cache_bits_max; i >= 1; --i, key >>= 1) {
+            hashers[i].colors_[key] = *argb;
+          }
+          argb_prev = *argb;
+        }
+        argb++;
+      } while (--len != 0);
     }
+    VP8LRefsCursorNext(&c);
   }
+
   for (i = 0; i <= cache_bits_max; ++i) {
-    entropies[i] = VP8LHistogramEstimateBits(histos[i]);
+    const double entropy = VP8LHistogramEstimateBits(histos[i]);
+    if (i == 0 || entropy < entropy_min) {
+      entropy_min = entropy;
+      *best_cache_bits = i;
+    }
   }
   ok = 1;
 Error:
@@ -1611,50 +789,6 @@ Error:
   return ok;
 }
 
-// Evaluate optimal cache bits for the local color cache.
-// The input *best_cache_bits sets the maximum cache bits to use (passing 0
-// implies disabling the local color cache). The local color cache is also
-// disabled for the lower (<= 25) quality.
-// Returns 0 in case of memory error.
-static int CalculateBestCacheSize(const uint32_t* const argb,
-                                  int xsize, int ysize, int quality,
-                                  const VP8LHashChain* const hash_chain,
-                                  VP8LBackwardRefs* const refs,
-                                  int* const lz77_computed,
-                                  int* const best_cache_bits) {
-  int i;
-  int cache_bits_high = (quality <= 25) ? 0 : *best_cache_bits;
-  double entropy_min = MAX_ENTROPY;
-  double entropies[MAX_COLOR_CACHE_BITS + 1];
-
-  assert(cache_bits_high <= MAX_COLOR_CACHE_BITS);
-
-  *lz77_computed = 0;
-  if (cache_bits_high == 0) {
-    *best_cache_bits = 0;
-    // Local color cache is disabled.
-    return 1;
-  }
-  // Compute LZ77 with no cache (0 bits), as the ideal LZ77 with a color cache
-  // is not that different in practice.
-  if (!BackwardReferencesLz77(xsize, ysize, argb, 0, hash_chain, refs)) {
-    return 0;
-  }
-  // Find the cache_bits giving the lowest entropy. The search is done in a
-  // brute-force way as the function (entropy w.r.t cache_bits) can be
-  // anything in practice.
-  if (!ComputeCacheEntropies(argb, refs, cache_bits_high, entropies)) {
-    return 0;
-  }
-  for (i = 0; i <= cache_bits_high; ++i) {
-    if (i == 0 || entropies[i] < entropy_min) {
-      entropy_min = entropies[i];
-      *best_cache_bits = i;
-    }
-  }
-  return 1;
-}
-
 // Update (in-place) backward references for specified cache_bits.
 static int BackwardRefsWithLocalCache(const uint32_t* const argb,
                                       int cache_bits,
@@ -1693,8 +827,7 @@ static int BackwardRefsWithLocalCache(const uint32_t* const argb,
 static VP8LBackwardRefs* GetBackwardReferencesLowEffort(
     int width, int height, const uint32_t* const argb,
     int* const cache_bits, const VP8LHashChain* const hash_chain,
-    VP8LBackwardRefs refs_array[2]) {
-  VP8LBackwardRefs* refs_lz77 = &refs_array[0];
+    VP8LBackwardRefs* const refs_lz77) {
   *cache_bits = 0;
   if (!BackwardReferencesLz77(width, height, argb, 0, hash_chain, refs_lz77)) {
     return NULL;
@@ -1703,98 +836,108 @@ static VP8LBackwardRefs* GetBackwardReferencesLowEffort(
   return refs_lz77;
 }
 
+extern int VP8LBackwardReferencesTraceBackwards(
+    int xsize, int ysize, const uint32_t* const argb, int cache_bits,
+    const VP8LHashChain* const hash_chain,
+    const VP8LBackwardRefs* const refs_src, VP8LBackwardRefs* const refs_dst);
 static VP8LBackwardRefs* GetBackwardReferences(
     int width, int height, const uint32_t* const argb, int quality,
-    int* const cache_bits, const VP8LHashChain* const hash_chain,
-    VP8LBackwardRefs refs_array[2]) {
-  int lz77_is_useful;
-  int lz77_computed;
-  double bit_cost_lz77, bit_cost_rle;
-  VP8LBackwardRefs* best = NULL;
-  VP8LBackwardRefs* refs_lz77 = &refs_array[0];
-  VP8LBackwardRefs* refs_rle = &refs_array[1];
+    int lz77_types_to_try, int* const cache_bits,
+    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* best,
+    VP8LBackwardRefs* worst) {
+  const int cache_bits_initial = *cache_bits;
+  double bit_cost_best = -1;
   VP8LHistogram* histo = NULL;
+  int lz77_type, lz77_type_best = 0;
+  VP8LHashChain hash_chain_box;
+  memset(&hash_chain_box, 0, sizeof(hash_chain_box));
 
-  if (!CalculateBestCacheSize(argb, width, height, quality, hash_chain,
-                              refs_lz77, &lz77_computed, cache_bits)) {
-    goto Error;
-  }
+  histo = VP8LAllocateHistogram(MAX_COLOR_CACHE_BITS);
+  if (histo == NULL) goto Error;
 
-  if (lz77_computed) {
-    // Transform refs_lz77 for the optimized cache_bits.
-    if (*cache_bits > 0) {
-      if (!BackwardRefsWithLocalCache(argb, *cache_bits, refs_lz77)) {
-        goto Error;
-      }
+  for (lz77_type = 1; lz77_types_to_try;
+       lz77_types_to_try &= ~lz77_type, lz77_type <<= 1) {
+    int res = 0;
+    double bit_cost;
+    int cache_bits_tmp = cache_bits_initial;
+    if ((lz77_types_to_try & lz77_type) == 0) continue;
+    switch (lz77_type) {
+      case kLZ77RLE:
+        res = BackwardReferencesRle(width, height, argb, 0, worst);
+        break;
+      case kLZ77Standard:
+        // Compute LZ77 with no cache (0 bits), as the ideal LZ77 with a color
+        // cache is not that different in practice.
+        res = BackwardReferencesLz77(width, height, argb, 0, hash_chain, worst);
+        break;
+      case kLZ77Box:
+        if (!VP8LHashChainInit(&hash_chain_box, width * height)) goto Error;
+        res = BackwardReferencesLz77Box(width, height, argb, 0, hash_chain,
+                                        &hash_chain_box, worst);
+        break;
+      default:
+        assert(0);
     }
-  } else {
-    if (!BackwardReferencesLz77(width, height, argb, *cache_bits, hash_chain,
-                                refs_lz77)) {
+    if (!res) goto Error;
+
+    // Next, try with a color cache and update the references.
+    if (!CalculateBestCacheSize(argb, quality, worst, &cache_bits_tmp)) {
       goto Error;
     }
-  }
-
-  if (!BackwardReferencesRle(width, height, argb, *cache_bits, refs_rle)) {
-    goto Error;
-  }
-
-  histo = VP8LAllocateHistogram(*cache_bits);
-  if (histo == NULL) goto Error;
-
-  {
-    // Evaluate LZ77 coding.
-    VP8LHistogramCreate(histo, refs_lz77, *cache_bits);
-    bit_cost_lz77 = VP8LHistogramEstimateBits(histo);
-    // Evaluate RLE coding.
-    VP8LHistogramCreate(histo, refs_rle, *cache_bits);
-    bit_cost_rle = VP8LHistogramEstimateBits(histo);
-    // Decide if LZ77 is useful.
-    lz77_is_useful = (bit_cost_lz77 < bit_cost_rle);
-  }
-
-  // Choose appropriate backward reference.
-  if (lz77_is_useful) {
-    // TraceBackwards is costly. Don't execute it at lower quality.
-    const int try_lz77_trace_backwards = (quality >= 25);
-    best = refs_lz77;   // default guess: lz77 is better
-    if (try_lz77_trace_backwards) {
-      VP8LBackwardRefs* const refs_trace = refs_rle;
-      if (!VP8LBackwardRefsCopy(refs_lz77, refs_trace)) {
-        best = NULL;
+    if (cache_bits_tmp > 0) {
+      if (!BackwardRefsWithLocalCache(argb, cache_bits_tmp, worst)) {
         goto Error;
       }
-      if (BackwardReferencesTraceBackwards(width, height, argb, quality,
-                                           *cache_bits, hash_chain,
-                                           refs_trace)) {
-        double bit_cost_trace;
-        // Evaluate LZ77 coding.
-        VP8LHistogramCreate(histo, refs_trace, *cache_bits);
-        bit_cost_trace = VP8LHistogramEstimateBits(histo);
-        if (bit_cost_trace < bit_cost_lz77) {
-          best = refs_trace;
-        }
-      }
     }
-  } else {
-    best = refs_rle;
+
+    // Keep the best backward references.
+    VP8LHistogramCreate(histo, worst, cache_bits_tmp);
+    bit_cost = VP8LHistogramEstimateBits(histo);
+    if (lz77_type_best == 0 || bit_cost < bit_cost_best) {
+      VP8LBackwardRefs* const tmp = worst;
+      worst = best;
+      best = tmp;
+      bit_cost_best = bit_cost;
+      *cache_bits = cache_bits_tmp;
+      lz77_type_best = lz77_type;
+    }
+  }
+  assert(lz77_type_best > 0);
+
+  // Improve on simple LZ77 but only for high quality (TraceBackwards is
+  // costly).
+  if ((lz77_type_best == kLZ77Standard || lz77_type_best == kLZ77Box) &&
+      quality >= 25) {
+    const VP8LHashChain* const hash_chain_tmp =
+        (lz77_type_best == kLZ77Standard) ? hash_chain : &hash_chain_box;
+    if (VP8LBackwardReferencesTraceBackwards(width, height, argb, *cache_bits,
+                                             hash_chain_tmp, best, worst)) {
+      double bit_cost_trace;
+      VP8LHistogramCreate(histo, worst, *cache_bits);
+      bit_cost_trace = VP8LHistogramEstimateBits(histo);
+      if (bit_cost_trace < bit_cost_best) best = worst;
+    }
   }
 
   BackwardReferences2DLocality(width, best);
 
- Error:
+Error:
+  VP8LHashChainClear(&hash_chain_box);
   VP8LFreeHistogram(histo);
   return best;
 }
 
 VP8LBackwardRefs* VP8LGetBackwardReferences(
     int width, int height, const uint32_t* const argb, int quality,
-    int low_effort, int* const cache_bits,
-    const VP8LHashChain* const hash_chain, VP8LBackwardRefs refs_array[2]) {
+    int low_effort, int lz77_types_to_try, int* const cache_bits,
+    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs_tmp1,
+    VP8LBackwardRefs* const refs_tmp2) {
   if (low_effort) {
     return GetBackwardReferencesLowEffort(width, height, argb, cache_bits,
-                                          hash_chain, refs_array);
+                                          hash_chain, refs_tmp1);
   } else {
-    return GetBackwardReferences(width, height, argb, quality, cache_bits,
-                                 hash_chain, refs_array);
+    return GetBackwardReferences(width, height, argb, quality,
+                                 lz77_types_to_try, cache_bits, hash_chain,
+                                 refs_tmp1, refs_tmp2);
   }
 }
diff --git a/src/3rdparty/libwebp/src/enc/backward_references_enc.h b/src/3rdparty/libwebp/src/enc/backward_references_enc.h
index 3a19aa7..103ddfd 100644
--- a/src/3rdparty/libwebp/src/enc/backward_references_enc.h
+++ b/src/3rdparty/libwebp/src/enc/backward_references_enc.h
@@ -10,13 +10,13 @@
 // Author: Jyrki Alakuijala (jyrki@google.com)
 //
 
-#ifndef WEBP_ENC_BACKWARD_REFERENCES_H_
-#define WEBP_ENC_BACKWARD_REFERENCES_H_
+#ifndef WEBP_ENC_BACKWARD_REFERENCES_ENC_H_
+#define WEBP_ENC_BACKWARD_REFERENCES_ENC_H_
 
 #include <assert.h>
 #include <stdlib.h>
-#include "../webp/types.h"
-#include "../webp/format_constants.h"
+#include "src/webp/types.h"
+#include "src/webp/format_constants.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -91,11 +91,6 @@ static WEBP_INLINE uint32_t PixOrCopyLength(const PixOrCopy* const p) {
   return p->len;
 }
 
-static WEBP_INLINE uint32_t PixOrCopyArgb(const PixOrCopy* const p) {
-  assert(p->mode == kLiteral);
-  return p->argb_or_distance;
-}
-
 static WEBP_INLINE uint32_t PixOrCopyCacheIdx(const PixOrCopy* const p) {
   assert(p->mode == kCacheIdx);
   assert(p->argb_or_distance < (1U << MAX_COLOR_CACHE_BITS));
@@ -113,6 +108,16 @@ static WEBP_INLINE uint32_t PixOrCopyDistance(const PixOrCopy* const p) {
 #define HASH_BITS 18
 #define HASH_SIZE (1 << HASH_BITS)
 
+// If you change this, you need MAX_LENGTH_BITS + WINDOW_SIZE_BITS <= 32 as it
+// is used in VP8LHashChain.
+#define MAX_LENGTH_BITS 12
+#define WINDOW_SIZE_BITS 20
+// We want the max value to be attainable and stored in MAX_LENGTH_BITS bits.
+#define MAX_LENGTH ((1 << MAX_LENGTH_BITS) - 1)
+#if MAX_LENGTH_BITS + WINDOW_SIZE_BITS > 32
+#error "MAX_LENGTH_BITS + WINDOW_SIZE_BITS > 32"
+#endif
+
 typedef struct VP8LHashChain VP8LHashChain;
 struct VP8LHashChain {
   // The 20 most significant bits contain the offset at which the best match
@@ -134,6 +139,24 @@ int VP8LHashChainFill(VP8LHashChain* const p, int quality,
                       int low_effort);
 void VP8LHashChainClear(VP8LHashChain* const p);  // release memory
 
+static WEBP_INLINE int VP8LHashChainFindOffset(const VP8LHashChain* const p,
+                                               const int base_position) {
+  return p->offset_length_[base_position] >> MAX_LENGTH_BITS;
+}
+
+static WEBP_INLINE int VP8LHashChainFindLength(const VP8LHashChain* const p,
+                                               const int base_position) {
+  return p->offset_length_[base_position] & ((1U << MAX_LENGTH_BITS) - 1);
+}
+
+static WEBP_INLINE void VP8LHashChainFindCopy(const VP8LHashChain* const p,
+                                              int base_position,
+                                              int* const offset_ptr,
+                                              int* const length_ptr) {
+  *offset_ptr = VP8LHashChainFindOffset(p, base_position);
+  *length_ptr = VP8LHashChainFindLength(p, base_position);
+}
+
 // -----------------------------------------------------------------------------
 // VP8LBackwardRefs (block-based backward-references storage)
 
@@ -158,9 +181,6 @@ struct VP8LBackwardRefs {
 void VP8LBackwardRefsInit(VP8LBackwardRefs* const refs, int block_size);
 // Release memory for backward references.
 void VP8LBackwardRefsClear(VP8LBackwardRefs* const refs);
-// Copies the 'src' backward refs to the 'dst'. Returns 0 in case of error.
-int VP8LBackwardRefsCopy(const VP8LBackwardRefs* const src,
-                         VP8LBackwardRefs* const dst);
 
 // Cursor for iterating on references content
 typedef struct {
@@ -189,6 +209,12 @@ static WEBP_INLINE void VP8LRefsCursorNext(VP8LRefsCursor* const c) {
 // -----------------------------------------------------------------------------
 // Main entry points
 
+enum VP8LLZ77Type {
+  kLZ77Standard = 1,
+  kLZ77RLE = 2,
+  kLZ77Box = 4
+};
+
 // Evaluates best possible backward references for specified quality.
 // The input cache_bits to 'VP8LGetBackwardReferences' sets the maximum cache
 // bits to use (passing 0 implies disabling the local color cache).
@@ -197,11 +223,12 @@ static WEBP_INLINE void VP8LRefsCursorNext(VP8LRefsCursor* const c) {
 // refs[0] or refs[1].
 VP8LBackwardRefs* VP8LGetBackwardReferences(
     int width, int height, const uint32_t* const argb, int quality,
-    int low_effort, int* const cache_bits,
-    const VP8LHashChain* const hash_chain, VP8LBackwardRefs refs[2]);
+    int low_effort, int lz77_types_to_try, int* const cache_bits,
+    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs_tmp1,
+    VP8LBackwardRefs* const refs_tmp2);
 
 #ifdef __cplusplus
 }
 #endif
 
-#endif  // WEBP_ENC_BACKWARD_REFERENCES_H_
+#endif  // WEBP_ENC_BACKWARD_REFERENCES_ENC_H_
diff --git a/src/3rdparty/libwebp/src/enc/config_enc.c b/src/3rdparty/libwebp/src/enc/config_enc.c
index 4589dc0..9d48289 100644
--- a/src/3rdparty/libwebp/src/enc/config_enc.c
+++ b/src/3rdparty/libwebp/src/enc/config_enc.c
@@ -12,10 +12,10 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
-#include "../webp/encode.h"
+#include "src/webp/encode.h"
 
 //------------------------------------------------------------------------------
 // WebPConfig
diff --git a/src/3rdparty/libwebp/src/enc/cost_enc.c b/src/3rdparty/libwebp/src/enc/cost_enc.c
index c823f5a..48fd9bc 100644
--- a/src/3rdparty/libwebp/src/enc/cost_enc.c
+++ b/src/3rdparty/libwebp/src/enc/cost_enc.c
@@ -11,7 +11,7 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./cost_enc.h"
+#include "src/enc/cost_enc.h"
 
 //------------------------------------------------------------------------------
 // Level cost tables
diff --git a/src/3rdparty/libwebp/src/enc/cost_enc.h b/src/3rdparty/libwebp/src/enc/cost_enc.h
index 99e4b37..bdce1e6 100644
--- a/src/3rdparty/libwebp/src/enc/cost_enc.h
+++ b/src/3rdparty/libwebp/src/enc/cost_enc.h
@@ -11,12 +11,12 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_ENC_COST_H_
-#define WEBP_ENC_COST_H_
+#ifndef WEBP_ENC_COST_ENC_H_
+#define WEBP_ENC_COST_ENC_H_
 
 #include <assert.h>
 #include <stdlib.h>
-#include "./vp8i_enc.h"
+#include "src/enc/vp8i_enc.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -79,4 +79,4 @@ extern const uint16_t VP8FixedCostsI4[NUM_BMODES][NUM_BMODES][NUM_BMODES];
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_ENC_COST_H_ */
+#endif  /* WEBP_ENC_COST_ENC_H_ */
diff --git a/src/3rdparty/libwebp/src/enc/delta_palettization_enc.c b/src/3rdparty/libwebp/src/enc/delta_palettization_enc.c
index eaf0f05..a61c8e6 100644
--- a/src/3rdparty/libwebp/src/enc/delta_palettization_enc.c
+++ b/src/3rdparty/libwebp/src/enc/delta_palettization_enc.c
@@ -10,11 +10,11 @@
 // Author: Mislav Bradac (mislavm@google.com)
 //
 
-#include "./delta_palettization_enc.h"
+#include "src/enc/delta_palettization_enc.h"
 
 #ifdef WEBP_EXPERIMENTAL_FEATURES
-#include "../webp/types.h"
-#include "../dsp/lossless.h"
+#include "src/webp/types.h"
+#include "src/dsp/lossless.h"
 
 #define MK_COL(r, g, b) (((r) << 16) + ((g) << 8) + (b))
 
diff --git a/src/3rdparty/libwebp/src/enc/delta_palettization_enc.h b/src/3rdparty/libwebp/src/enc/delta_palettization_enc.h
index 63048ec..b15e2cd 100644
--- a/src/3rdparty/libwebp/src/enc/delta_palettization_enc.h
+++ b/src/3rdparty/libwebp/src/enc/delta_palettization_enc.h
@@ -10,11 +10,11 @@
 // Author: Mislav Bradac (mislavm@google.com)
 //
 
-#ifndef WEBP_ENC_DELTA_PALETTIZATION_H_
-#define WEBP_ENC_DELTA_PALETTIZATION_H_
+#ifndef WEBP_ENC_DELTA_PALETTIZATION_ENC_H_
+#define WEBP_ENC_DELTA_PALETTIZATION_ENC_H_
 
-#include "../webp/encode.h"
-#include "../enc/vp8li_enc.h"
+#include "src/webp/encode.h"
+#include "src/enc/vp8li_enc.h"
 
 // Replaces enc->argb_[] input by a palettizable approximation of it,
 // and generates optimal enc->palette_[].
@@ -22,4 +22,4 @@
 // if delta-palettization is not producing expected saving.
 WebPEncodingError WebPSearchOptimalDeltaPalette(VP8LEncoder* const enc);
 
-#endif  // WEBP_ENC_DELTA_PALETTIZATION_H_
+#endif  // WEBP_ENC_DELTA_PALETTIZATION_ENC_H_
diff --git a/src/3rdparty/libwebp/src/enc/filter_enc.c b/src/3rdparty/libwebp/src/enc/filter_enc.c
index 4bc3672..580800b 100644
--- a/src/3rdparty/libwebp/src/enc/filter_enc.c
+++ b/src/3rdparty/libwebp/src/enc/filter_enc.c
@@ -12,8 +12,8 @@
 // Author: somnath@google.com (Somnath Banerjee)
 
 #include <assert.h>
-#include "./vp8i_enc.h"
-#include "../dsp/dsp.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/dsp/dsp.h"
 
 // This table gives, for a given sharpness, the filtering strength to be
 // used (at least) in order to filter a given edge step delta.
@@ -65,6 +65,8 @@ int VP8FilterStrengthFromDelta(int sharpness, int delta) {
 //------------------------------------------------------------------------------
 // Paragraph 15.4: compute the inner-edge filtering strength
 
+#if !defined(WEBP_REDUCE_SIZE)
+
 static int GetILevel(int sharpness, int level) {
   if (sharpness > 0) {
     if (sharpness > 4) {
@@ -129,11 +131,14 @@ static double GetMBSSIM(const uint8_t* yuv1, const uint8_t* yuv2) {
   return sum;
 }
 
+#endif  // !defined(WEBP_REDUCE_SIZE)
+
 //------------------------------------------------------------------------------
 // Exposed APIs: Encoder should call the following 3 functions to adjust
 // loop filter strength
 
 void VP8InitFilter(VP8EncIterator* const it) {
+#if !defined(WEBP_REDUCE_SIZE)
   if (it->lf_stats_ != NULL) {
     int s, i;
     for (s = 0; s < NUM_MB_SEGMENTS; s++) {
@@ -143,9 +148,13 @@ void VP8InitFilter(VP8EncIterator* const it) {
     }
     VP8SSIMDspInit();
   }
+#else
+  (void)it;
+#endif
 }
 
 void VP8StoreFilterStats(VP8EncIterator* const it) {
+#if !defined(WEBP_REDUCE_SIZE)
   int d;
   VP8Encoder* const enc = it->enc_;
   const int s = it->mb_->segment_;
@@ -177,10 +186,14 @@ void VP8StoreFilterStats(VP8EncIterator* const it) {
     DoFilter(it, level);
     (*it->lf_stats_)[s][level] += GetMBSSIM(it->yuv_in_, it->yuv_out2_);
   }
+#else  // defined(WEBP_REDUCE_SIZE)
+  (void)it;
+#endif  // !defined(WEBP_REDUCE_SIZE)
 }
 
 void VP8AdjustFilterStrength(VP8EncIterator* const it) {
   VP8Encoder* const enc = it->enc_;
+#if !defined(WEBP_REDUCE_SIZE)
   if (it->lf_stats_ != NULL) {
     int s;
     for (s = 0; s < NUM_MB_SEGMENTS; s++) {
@@ -196,7 +209,10 @@ void VP8AdjustFilterStrength(VP8EncIterator* const it) {
       }
       enc->dqm_[s].fstrength_ = best_level;
     }
-  } else if (enc->config_->filter_strength > 0) {
+    return;
+  }
+#endif  // !defined(WEBP_REDUCE_SIZE)
+  if (enc->config_->filter_strength > 0) {
     int max_level = 0;
     int s;
     for (s = 0; s < NUM_MB_SEGMENTS; s++) {
diff --git a/src/3rdparty/libwebp/src/enc/frame_enc.c b/src/3rdparty/libwebp/src/enc/frame_enc.c
index abef523..2b0dc66 100644
--- a/src/3rdparty/libwebp/src/enc/frame_enc.c
+++ b/src/3rdparty/libwebp/src/enc/frame_enc.c
@@ -14,10 +14,10 @@
 #include <string.h>
 #include <math.h>
 
-#include "./cost_enc.h"
-#include "./vp8i_enc.h"
-#include "../dsp/dsp.h"
-#include "../webp/format_constants.h"  // RIFF constants
+#include "src/enc/cost_enc.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/dsp/dsp.h"
+#include "src/webp/format_constants.h"  // RIFF constants
 
 #define SEGMENT_VISU 0
 #define DEBUG_SEARCH 0    // useful to track search convergence
@@ -200,11 +200,13 @@ static void SetSegmentProbas(VP8Encoder* const enc) {
     const VP8MBInfo* const mb = &enc->mb_info_[n];
     p[mb->segment_]++;
   }
+#if !defined(WEBP_DISABLE_STATS)
   if (enc->pic_->stats != NULL) {
     for (n = 0; n < NUM_MB_SEGMENTS; ++n) {
       enc->pic_->stats->segment_size[n] = p[n];
     }
   }
+#endif
   if (enc->segment_hdr_.num_segments_ > 1) {
     uint8_t* const probas = enc->proba_.segments_;
     probas[0] = GetProba(p[0] + p[1], p[2] + p[3]);
@@ -452,6 +454,8 @@ static int RecordTokens(VP8EncIterator* const it, const VP8ModeScore* const rd,
 //------------------------------------------------------------------------------
 // ExtraInfo map / Debug function
 
+#if !defined(WEBP_DISABLE_STATS)
+
 #if SEGMENT_VISU
 static void SetBlock(uint8_t* p, int value, int size) {
   int y;
@@ -516,6 +520,20 @@ static void StoreSideInfo(const VP8EncIterator* const it) {
 #endif
 }
 
+#else  // defined(WEBP_DISABLE_STATS)
+static void ResetSSE(VP8Encoder* const enc) {
+  (void)enc;
+}
+static void StoreSideInfo(const VP8EncIterator* const it) {
+  VP8Encoder* const enc = it->enc_;
+  WebPPicture* const pic = enc->pic_;
+  if (pic->extra_info != NULL) {
+    memset(pic->extra_info, 0,
+           enc->mb_w_ * enc->mb_h_ * sizeof(*pic->extra_info));
+  }
+}
+#endif  // !defined(WEBP_DISABLE_STATS)
+
 static double GetPSNR(uint64_t mse, uint64_t size) {
   return (mse > 0 && size > 0) ? 10. * log10(255. * 255. * size / mse) : 99;
 }
@@ -640,7 +658,7 @@ static int StatLoop(VP8Encoder* const enc) {
 // Main loops
 //
 
-static const int kAverageBytesPerMB[8] = { 50, 24, 16, 9, 7, 5, 3, 2 };
+static const uint8_t kAverageBytesPerMB[8] = { 50, 24, 16, 9, 7, 5, 3, 2 };
 
 static int PreLoopInitialize(VP8Encoder* const enc) {
   int p;
@@ -670,6 +688,7 @@ static int PostLoopFinalize(VP8EncIterator* const it, int ok) {
   }
 
   if (ok) {      // All good. Finish up.
+#if !defined(WEBP_DISABLE_STATS)
     if (enc->pic_->stats != NULL) {  // finalize byte counters...
       int i, s;
       for (i = 0; i <= 2; ++i) {
@@ -678,6 +697,7 @@ static int PostLoopFinalize(VP8EncIterator* const it, int ok) {
         }
       }
     }
+#endif
     VP8AdjustFilterStrength(it);     // ...and store filter stats.
   } else {
     // Something bad happened -> need to do some memory cleanup.
diff --git a/src/3rdparty/libwebp/src/enc/histogram_enc.c b/src/3rdparty/libwebp/src/enc/histogram_enc.c
index 808b6f7..056a972 100644
--- a/src/3rdparty/libwebp/src/enc/histogram_enc.c
+++ b/src/3rdparty/libwebp/src/enc/histogram_enc.c
@@ -10,16 +10,16 @@
 // Author: Jyrki Alakuijala (jyrki@google.com)
 //
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
 #include <math.h>
 
-#include "./backward_references_enc.h"
-#include "./histogram_enc.h"
-#include "../dsp/lossless.h"
-#include "../dsp/lossless_common.h"
-#include "../utils/utils.h"
+#include "src/enc/backward_references_enc.h"
+#include "src/enc/histogram_enc.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+#include "src/utils/utils.h"
 
 #define MAX_COST 1.e38
 
@@ -76,7 +76,7 @@ void VP8LHistogramStoreRefs(const VP8LBackwardRefs* const refs,
                             VP8LHistogram* const histo) {
   VP8LRefsCursor c = VP8LRefsCursorInit(refs);
   while (VP8LRefsCursorOk(&c)) {
-    VP8LHistogramAddSinglePixOrCopy(histo, c.cur_pos);
+    VP8LHistogramAddSinglePixOrCopy(histo, c.cur_pos, NULL, 0);
     VP8LRefsCursorNext(&c);
   }
 }
@@ -138,7 +138,9 @@ VP8LHistogramSet* VP8LAllocateHistogramSet(int size, int cache_bits) {
 // -----------------------------------------------------------------------------
 
 void VP8LHistogramAddSinglePixOrCopy(VP8LHistogram* const histo,
-                                     const PixOrCopy* const v) {
+                                     const PixOrCopy* const v,
+                                     int (*const distance_modifier)(int, int),
+                                     int distance_modifier_arg0) {
   if (PixOrCopyIsLiteral(v)) {
     ++histo->alpha_[PixOrCopyLiteral(v, 3)];
     ++histo->red_[PixOrCopyLiteral(v, 2)];
@@ -152,7 +154,13 @@ void VP8LHistogramAddSinglePixOrCopy(VP8LHistogram* const histo,
     int code, extra_bits;
     VP8LPrefixEncodeBits(PixOrCopyLength(v), &code, &extra_bits);
     ++histo->literal_[NUM_LITERAL_CODES + code];
-    VP8LPrefixEncodeBits(PixOrCopyDistance(v), &code, &extra_bits);
+    if (distance_modifier == NULL) {
+      VP8LPrefixEncodeBits(PixOrCopyDistance(v), &code, &extra_bits);
+    } else {
+      VP8LPrefixEncodeBits(
+          distance_modifier(distance_modifier_arg0, PixOrCopyDistance(v)),
+          &code, &extra_bits);
+    }
     ++histo->distance_[code];
   }
 }
@@ -473,7 +481,7 @@ static void HistogramBuild(
   while (VP8LRefsCursorOk(&c)) {
     const PixOrCopy* const v = c.cur_pos;
     const int ix = (y >> histo_bits) * histo_xsize + (x >> histo_bits);
-    VP8LHistogramAddSinglePixOrCopy(histograms[ix], v);
+    VP8LHistogramAddSinglePixOrCopy(histograms[ix], v, NULL, 0);
     x += PixOrCopyLength(v);
     while (x >= xsize) {
       x -= xsize;
@@ -523,11 +531,12 @@ static void HistogramAnalyzeEntropyBin(VP8LHistogramSet* const image_histo,
 
 // Compact image_histo[] by merging some histograms with same bin_id together if
 // it's advantageous.
-static VP8LHistogram* HistogramCombineEntropyBin(
-    VP8LHistogramSet* const image_histo,
-    VP8LHistogram* cur_combo,
-    const uint16_t* const bin_map, int bin_map_size, int num_bins,
-    double combine_cost_factor, int low_effort) {
+static void HistogramCombineEntropyBin(VP8LHistogramSet* const image_histo,
+                                       VP8LHistogram* cur_combo,
+                                       const uint16_t* const bin_map,
+                                       int bin_map_size, int num_bins,
+                                       double combine_cost_factor,
+                                       int low_effort) {
   VP8LHistogram** const histograms = image_histo->histograms;
   int idx;
   // Work in-place: processed histograms are put at the beginning of
@@ -593,14 +602,13 @@ static VP8LHistogram* HistogramCombineEntropyBin(
       UpdateHistogramCost(histograms[idx]);
     }
   }
-  return cur_combo;
 }
 
+// Implement a Lehmer random number generator with a multiplicative constant of
+// 48271 and a modulo constant of 2^31 − 1.
 static uint32_t MyRand(uint32_t* const seed) {
-  *seed = (*seed * 16807ull) & 0xffffffffu;
-  if (*seed == 0) {
-    *seed = 1;
-  }
+  *seed = (uint32_t)(((uint64_t)(*seed) * 48271u) % 2147483647u);
+  assert(*seed > 0);
   return *seed;
 }
 
@@ -641,57 +649,75 @@ static int HistoQueueInit(HistoQueue* const histo_queue, const int max_index) {
 static void HistoQueueClear(HistoQueue* const histo_queue) {
   assert(histo_queue != NULL);
   WebPSafeFree(histo_queue->queue);
+  histo_queue->size = 0;
+  histo_queue->max_size = 0;
 }
 
-static void SwapHistogramPairs(HistogramPair *p1,
-                               HistogramPair *p2) {
-  const HistogramPair tmp = *p1;
-  *p1 = *p2;
-  *p2 = tmp;
+// Pop a specific pair in the queue by replacing it with the last one
+// and shrinking the queue.
+static void HistoQueuePopPair(HistoQueue* const histo_queue,
+                              HistogramPair* const pair) {
+  assert(pair >= histo_queue->queue &&
+         pair < (histo_queue->queue + histo_queue->size));
+  assert(histo_queue->size > 0);
+  *pair = histo_queue->queue[histo_queue->size - 1];
+  --histo_queue->size;
 }
 
-// Given a valid priority queue in range [0, queue_size) this function checks
-// whether histo_queue[queue_size] should be accepted and swaps it with the
-// front if it is smaller. Otherwise, it leaves it as is.
-static void UpdateQueueFront(HistoQueue* const histo_queue) {
-  if (histo_queue->queue[histo_queue->size].cost_diff >= 0) return;
-
-  if (histo_queue->queue[histo_queue->size].cost_diff <
-      histo_queue->queue[0].cost_diff) {
-    SwapHistogramPairs(histo_queue->queue,
-                       histo_queue->queue + histo_queue->size);
+// Check whether a pair in the queue should be updated as head or not.
+static void HistoQueueUpdateHead(HistoQueue* const histo_queue,
+                                 HistogramPair* const pair) {
+  assert(pair->cost_diff < 0.);
+  assert(pair >= histo_queue->queue &&
+         pair < (histo_queue->queue + histo_queue->size));
+  assert(histo_queue->size > 0);
+  if (pair->cost_diff < histo_queue->queue[0].cost_diff) {
+    // Replace the best pair.
+    const HistogramPair tmp = histo_queue->queue[0];
+    histo_queue->queue[0] = *pair;
+    *pair = tmp;
   }
-  ++histo_queue->size;
-
-  // We cannot add more elements than the capacity.
-  // The allocation adds an extra element to the official capacity so that
-  // histo_queue->queue[histo_queue->max_size] is read/written within bound.
-  assert(histo_queue->size <= histo_queue->max_size);
 }
 
-// -----------------------------------------------------------------------------
-
-static void PreparePair(VP8LHistogram** histograms, int idx1, int idx2,
-                        HistogramPair* const pair) {
-  VP8LHistogram* h1;
-  VP8LHistogram* h2;
+// Create a pair from indices "idx1" and "idx2" provided its cost
+// is inferior to "threshold", a negative entropy.
+// It returns the cost of the pair, or 0. if it superior to threshold.
+static double HistoQueuePush(HistoQueue* const histo_queue,
+                             VP8LHistogram** const histograms, int idx1,
+                             int idx2, double threshold) {
+  const VP8LHistogram* h1;
+  const VP8LHistogram* h2;
+  HistogramPair pair;
   double sum_cost;
 
+  assert(threshold <= 0.);
   if (idx1 > idx2) {
     const int tmp = idx2;
     idx2 = idx1;
     idx1 = tmp;
   }
-  pair->idx1 = idx1;
-  pair->idx2 = idx2;
+  pair.idx1 = idx1;
+  pair.idx2 = idx2;
   h1 = histograms[idx1];
   h2 = histograms[idx2];
   sum_cost = h1->bit_cost_ + h2->bit_cost_;
-  pair->cost_combo = 0.;
-  GetCombinedHistogramEntropy(h1, h2, sum_cost, &pair->cost_combo);
-  pair->cost_diff = pair->cost_combo - sum_cost;
+  pair.cost_combo = 0.;
+  GetCombinedHistogramEntropy(h1, h2, sum_cost + threshold, &pair.cost_combo);
+  pair.cost_diff = pair.cost_combo - sum_cost;
+
+  // Do not even consider the pair if it does not improve the entropy.
+  if (pair.cost_diff >= threshold) return 0.;
+
+  // We cannot add more elements than the capacity.
+  assert(histo_queue->size < histo_queue->max_size);
+  histo_queue->queue[histo_queue->size++] = pair;
+  HistoQueueUpdateHead(histo_queue, &histo_queue->queue[histo_queue->size - 1]);
+
+  return pair.cost_diff;
 }
 
+// -----------------------------------------------------------------------------
+
 // Combines histograms by continuously choosing the one with the highest cost
 // reduction.
 static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo) {
@@ -714,13 +740,11 @@ static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo) {
     clusters[i] = i;
     for (j = i + 1; j < image_histo_size; ++j) {
       // Initialize positions array.
-      PreparePair(histograms, i, j, &histo_queue.queue[histo_queue.size]);
-      UpdateQueueFront(&histo_queue);
+      HistoQueuePush(&histo_queue, histograms, i, j, 0.);
     }
   }
 
   while (image_histo_size > 1 && histo_queue.size > 0) {
-    HistogramPair* copy_to;
     const int idx1 = histo_queue.queue[0].idx1;
     const int idx2 = histo_queue.queue[0].idx2;
     HistogramAdd(histograms[idx2], histograms[idx1], histograms[idx1]);
@@ -733,31 +757,22 @@ static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo) {
     }
     --image_histo_size;
 
-    // Remove pairs intersecting the just combined best pair. This will
-    // therefore pop the head of the queue.
-    copy_to = histo_queue.queue;
-    for (i = 0; i < histo_queue.size; ++i) {
+    // Remove pairs intersecting the just combined best pair.
+    for (i = 0; i < histo_queue.size;) {
       HistogramPair* const p = histo_queue.queue + i;
       if (p->idx1 == idx1 || p->idx2 == idx1 ||
           p->idx1 == idx2 || p->idx2 == idx2) {
-        // Do not copy the invalid pair.
-        continue;
-      }
-      if (p->cost_diff < histo_queue.queue[0].cost_diff) {
-        // Replace the top of the queue if we found better.
-        SwapHistogramPairs(histo_queue.queue, p);
+        HistoQueuePopPair(&histo_queue, p);
+      } else {
+        HistoQueueUpdateHead(&histo_queue, p);
+        ++i;
       }
-      SwapHistogramPairs(copy_to, p);
-      ++copy_to;
     }
-    histo_queue.size = (int)(copy_to - histo_queue.queue);
 
     // Push new pairs formed with combined histogram to the queue.
     for (i = 0; i < image_histo_size; ++i) {
       if (clusters[i] != idx1) {
-        PreparePair(histograms, idx1, clusters[i],
-                    &histo_queue.queue[histo_queue.size]);
-        UpdateQueueFront(&histo_queue);
+        HistoQueuePush(&histo_queue, histograms, idx1, clusters[i], 0.);
       }
     }
   }
@@ -777,90 +792,130 @@ static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo) {
   return ok;
 }
 
-static void HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
-                                       VP8LHistogram* tmp_histo,
-                                       VP8LHistogram* best_combo,
-                                       int quality, int min_cluster_size) {
+// Perform histogram aggregation using a stochastic approach.
+// 'do_greedy' is set to 1 if a greedy approach needs to be performed
+// afterwards, 0 otherwise.
+static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
+                                      int min_cluster_size,
+                                      int* const do_greedy) {
   int iter;
-  uint32_t seed = 0;
+  uint32_t seed = 1;
   int tries_with_no_success = 0;
   int image_histo_size = image_histo->size;
-  const int iter_mult = (quality < 25) ? 2 : 2 + (quality - 25) / 8;
-  const int outer_iters = image_histo_size * iter_mult;
-  const int num_pairs = image_histo_size / 2;
+  const int outer_iters = image_histo_size;
   const int num_tries_no_success = outer_iters / 2;
-  int idx2_max = image_histo_size - 1;
-  int do_brute_dorce = 0;
   VP8LHistogram** const histograms = image_histo->histograms;
+  // Priority queue of histogram pairs. Its size of "kCostHeapSizeSqrt"^2
+  // impacts the quality of the compression and the speed: the smaller the
+  // faster but the worse for the compression.
+  HistoQueue histo_queue;
+  const int kHistoQueueSizeSqrt = 3;
+  int ok = 0;
 
+  if (!HistoQueueInit(&histo_queue, kHistoQueueSizeSqrt)) {
+    goto End;
+  }
   // Collapse similar histograms in 'image_histo'.
   ++min_cluster_size;
-  for (iter = 0;
-       iter < outer_iters && image_histo_size >= min_cluster_size;
+  for (iter = 0; iter < outer_iters && image_histo_size >= min_cluster_size &&
+                 ++tries_with_no_success < num_tries_no_success;
        ++iter) {
-    double best_cost_diff = 0.;
+    double best_cost =
+        (histo_queue.size == 0) ? 0. : histo_queue.queue[0].cost_diff;
     int best_idx1 = -1, best_idx2 = 1;
     int j;
-    int num_tries =
-        (num_pairs < image_histo_size) ? num_pairs : image_histo_size;
-    // Use a brute force approach if:
-    // - stochastic has not worked for a while and
-    // - if the number of iterations for brute force is less than the number of
-    // iterations if we never find a match ever again stochastically (hence
-    // num_tries times the number of remaining outer iterations).
-    do_brute_dorce =
-        (tries_with_no_success > 10) &&
-        (idx2_max * (idx2_max + 1) < 2 * num_tries * (outer_iters - iter));
-    if (do_brute_dorce) num_tries = idx2_max;
-
-    seed += iter;
-    for (j = 0; j < num_tries; ++j) {
-      double curr_cost_diff;
-      // Choose two histograms at random and try to combine them.
-      uint32_t idx1, idx2;
-      if (do_brute_dorce) {
-        // Use a brute force approach.
-        idx1 = (uint32_t)j;
-        idx2 = (uint32_t)idx2_max;
-      } else {
-        const uint32_t tmp = (j & 7) + 1;
-        const uint32_t diff =
-            (tmp < 3) ? tmp : MyRand(&seed) % (image_histo_size - 1);
-        idx1 = MyRand(&seed) % image_histo_size;
-        idx2 = (idx1 + diff + 1) % image_histo_size;
-        if (idx1 == idx2) {
-          continue;
-        }
-      }
+    const uint32_t rand_range = (image_histo_size - 1) * image_histo_size;
+    // image_histo_size / 2 was chosen empirically. Less means faster but worse
+    // compression.
+    const int num_tries = image_histo_size / 2;
 
-      // Calculate cost reduction on combining.
-      curr_cost_diff = HistogramAddEval(histograms[idx1], histograms[idx2],
-                                        tmp_histo, best_cost_diff);
-      if (curr_cost_diff < best_cost_diff) {  // found a better pair?
-        HistogramSwap(&best_combo, &tmp_histo);
-        best_cost_diff = curr_cost_diff;
-        best_idx1 = idx1;
-        best_idx2 = idx2;
+    for (j = 0; j < num_tries; ++j) {
+      double curr_cost;
+      // Choose two different histograms at random and try to combine them.
+      const uint32_t tmp = MyRand(&seed) % rand_range;
+      const uint32_t idx1 = tmp / (image_histo_size - 1);
+      uint32_t idx2 = tmp % (image_histo_size - 1);
+      if (idx2 >= idx1) ++idx2;
+
+      // Calculate cost reduction on combination.
+      curr_cost =
+          HistoQueuePush(&histo_queue, histograms, idx1, idx2, best_cost);
+      if (curr_cost < 0) {  // found a better pair?
+        best_cost = curr_cost;
+        // Empty the queue if we reached full capacity.
+        if (histo_queue.size == histo_queue.max_size) break;
       }
     }
-    if (do_brute_dorce) --idx2_max;
-
-    if (best_idx1 >= 0) {
-      HistogramSwap(&best_combo, &histograms[best_idx1]);
-      // swap best_idx2 slot with last one (which is now unused)
-      --image_histo_size;
-      if (idx2_max >= image_histo_size) idx2_max = image_histo_size - 1;
-      if (best_idx2 != image_histo_size) {
-        HistogramSwap(&histograms[image_histo_size], &histograms[best_idx2]);
-        histograms[image_histo_size] = NULL;
-      }
-      tries_with_no_success = 0;
+    if (histo_queue.size == 0) continue;
+
+    // Merge the two best histograms.
+    best_idx1 = histo_queue.queue[0].idx1;
+    best_idx2 = histo_queue.queue[0].idx2;
+    assert(best_idx1 < best_idx2);
+    HistogramAddEval(histograms[best_idx1], histograms[best_idx2],
+                     histograms[best_idx1], 0);
+    // Swap the best_idx2 histogram with the last one (which is now unused).
+    --image_histo_size;
+    if (best_idx2 != image_histo_size) {
+      HistogramSwap(&histograms[image_histo_size], &histograms[best_idx2]);
     }
-    if (++tries_with_no_success >= num_tries_no_success || idx2_max == 0) {
-      break;
+    histograms[image_histo_size] = NULL;
+    // Parse the queue and update each pair that deals with best_idx1,
+    // best_idx2 or image_histo_size.
+    for (j = 0; j < histo_queue.size;) {
+      HistogramPair* const p = histo_queue.queue + j;
+      const int is_idx1_best = p->idx1 == best_idx1 || p->idx1 == best_idx2;
+      const int is_idx2_best = p->idx2 == best_idx1 || p->idx2 == best_idx2;
+      int do_eval = 0;
+      // The front pair could have been duplicated by a random pick so
+      // check for it all the time nevertheless.
+      if (is_idx1_best && is_idx2_best) {
+        HistoQueuePopPair(&histo_queue, p);
+        continue;
+      }
+      // Any pair containing one of the two best indices should only refer to
+      // best_idx1. Its cost should also be updated.
+      if (is_idx1_best) {
+        p->idx1 = best_idx1;
+        do_eval = 1;
+      } else if (is_idx2_best) {
+        p->idx2 = best_idx1;
+        do_eval = 1;
+      }
+      if (p->idx2 == image_histo_size) {
+        // No need to re-evaluate here as it does not involve a pair
+        // containing best_idx1 or best_idx2.
+        p->idx2 = best_idx2;
+      }
+      assert(p->idx2 < image_histo_size);
+      // Make sure the index order is respected.
+      if (p->idx1 > p->idx2) {
+        const int tmp = p->idx2;
+        p->idx2 = p->idx1;
+        p->idx1 = tmp;
+      }
+      if (do_eval) {
+        // Re-evaluate the cost of an updated pair.
+        GetCombinedHistogramEntropy(histograms[p->idx1], histograms[p->idx2], 0,
+                                    &p->cost_diff);
+        if (p->cost_diff >= 0.) {
+          HistoQueuePopPair(&histo_queue, p);
+          continue;
+        }
+      }
+      HistoQueueUpdateHead(&histo_queue, p);
+      ++j;
     }
+
+    tries_with_no_success = 0;
   }
   image_histo->size = image_histo_size;
+  *do_greedy = (image_histo->size <= min_cluster_size);
+  ok = 1;
+
+End:
+  HistoQueueClear(&histo_queue);
+  return ok;
 }
 
 // -----------------------------------------------------------------------------
@@ -925,7 +980,7 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
                              int quality, int low_effort,
                              int histo_bits, int cache_bits,
                              VP8LHistogramSet* const image_histo,
-                             VP8LHistogramSet* const tmp_histos,
+                             VP8LHistogram* const tmp_histo,
                              uint16_t* const histogram_symbols) {
   int ok = 0;
   const int histo_xsize = histo_bits ? VP8LSubSampleSize(xsize, histo_bits) : 1;
@@ -933,7 +988,6 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
   const int image_histo_raw_size = histo_xsize * histo_ysize;
   VP8LHistogramSet* const orig_histo =
       VP8LAllocateHistogramSet(image_histo_raw_size, cache_bits);
-  VP8LHistogram* cur_combo;
   // Don't attempt linear bin-partition heuristic for
   // histograms of small sizes (as bin_map will be very sparse) and
   // maximum quality q==100 (to preserve the compression gains at that level).
@@ -948,7 +1002,6 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
   // Copies the histograms and computes its bit_cost.
   HistogramCopyAndAnalyze(orig_histo, image_histo);
 
-  cur_combo = tmp_histos->histograms[1];  // pick up working slot
   if (entropy_combine) {
     const int bin_map_size = orig_histo->size;
     // Reuse histogram_symbols storage. By definition, it's guaranteed to be ok.
@@ -958,10 +1011,9 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
 
     HistogramAnalyzeEntropyBin(orig_histo, bin_map, low_effort);
     // Collapse histograms with similar entropy.
-    cur_combo = HistogramCombineEntropyBin(image_histo, cur_combo,
-                                           bin_map, bin_map_size,
-                                           entropy_combine_num_bins,
-                                           combine_cost_factor, low_effort);
+    HistogramCombineEntropyBin(image_histo, tmp_histo, bin_map, bin_map_size,
+                               entropy_combine_num_bins, combine_cost_factor,
+                               low_effort);
   }
 
   // Don't combine the histograms using stochastic and greedy heuristics for
@@ -970,10 +1022,11 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
     const float x = quality / 100.f;
     // cubic ramp between 1 and MAX_HISTO_GREEDY:
     const int threshold_size = (int)(1 + (x * x * x) * (MAX_HISTO_GREEDY - 1));
-    HistogramCombineStochastic(image_histo, tmp_histos->histograms[0],
-                               cur_combo, quality, threshold_size);
-    if ((image_histo->size <= threshold_size) &&
-        !HistogramCombineGreedy(image_histo)) {
+    int do_greedy;
+    if (!HistogramCombineStochastic(image_histo, threshold_size, &do_greedy)) {
+      goto Error;
+    }
+    if (do_greedy && !HistogramCombineGreedy(image_histo)) {
       goto Error;
     }
   }
diff --git a/src/3rdparty/libwebp/src/enc/histogram_enc.h b/src/3rdparty/libwebp/src/enc/histogram_enc.h
index a9d258a..15b1fbd 100644
--- a/src/3rdparty/libwebp/src/enc/histogram_enc.h
+++ b/src/3rdparty/libwebp/src/enc/histogram_enc.h
@@ -11,14 +11,14 @@
 //
 // Models the histograms of literal and distance codes.
 
-#ifndef WEBP_ENC_HISTOGRAM_H_
-#define WEBP_ENC_HISTOGRAM_H_
+#ifndef WEBP_ENC_HISTOGRAM_ENC_H_
+#define WEBP_ENC_HISTOGRAM_ENC_H_
 
 #include <string.h>
 
-#include "./backward_references_enc.h"
-#include "../webp/format_constants.h"
-#include "../webp/types.h"
+#include "src/enc/backward_references_enc.h"
+#include "src/webp/format_constants.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -90,7 +90,9 @@ VP8LHistogram* VP8LAllocateHistogram(int cache_bits);
 
 // Accumulate a token 'v' into a histogram.
 void VP8LHistogramAddSinglePixOrCopy(VP8LHistogram* const histo,
-                                     const PixOrCopy* const v);
+                                     const PixOrCopy* const v,
+                                     int (*const distance_modifier)(int, int),
+                                     int distance_modifier_arg0);
 
 static WEBP_INLINE int VP8LHistogramNumCodes(int palette_code_bits) {
   return NUM_LITERAL_CODES + NUM_LENGTH_CODES +
@@ -103,7 +105,7 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
                              int quality, int low_effort,
                              int histogram_bits, int cache_bits,
                              VP8LHistogramSet* const image_in,
-                             VP8LHistogramSet* const tmp_histos,
+                             VP8LHistogram* const tmp_histo,
                              uint16_t* const histogram_symbols);
 
 // Returns the entropy for the symbols in the input array.
@@ -120,4 +122,4 @@ double VP8LHistogramEstimateBits(const VP8LHistogram* const p);
 }
 #endif
 
-#endif  // WEBP_ENC_HISTOGRAM_H_
+#endif  // WEBP_ENC_HISTOGRAM_ENC_H_
diff --git a/src/3rdparty/libwebp/src/enc/iterator_enc.c b/src/3rdparty/libwebp/src/enc/iterator_enc.c
index e48d30b..cfacfd2 100644
--- a/src/3rdparty/libwebp/src/enc/iterator_enc.c
+++ b/src/3rdparty/libwebp/src/enc/iterator_enc.c
@@ -13,7 +13,7 @@
 
 #include <string.h>
 
-#include "./vp8i_enc.h"
+#include "src/enc/vp8i_enc.h"
 
 //------------------------------------------------------------------------------
 // VP8Iterator
diff --git a/src/3rdparty/libwebp/src/enc/near_lossless_enc.c b/src/3rdparty/libwebp/src/enc/near_lossless_enc.c
index 2bd03ab..cadd14c 100644
--- a/src/3rdparty/libwebp/src/enc/near_lossless_enc.c
+++ b/src/3rdparty/libwebp/src/enc/near_lossless_enc.c
@@ -17,18 +17,20 @@
 #include <assert.h>
 #include <stdlib.h>
 
-#include "../dsp/lossless_common.h"
-#include "../utils/utils.h"
-#include "./vp8i_enc.h"
+#include "src/dsp/lossless_common.h"
+#include "src/utils/utils.h"
+#include "src/enc/vp8li_enc.h"
+
+#if (WEBP_NEAR_LOSSLESS == 1)
 
 #define MIN_DIM_FOR_NEAR_LOSSLESS 64
 #define MAX_LIMIT_BITS             5
 
 // Quantizes the value up or down to a multiple of 1<<bits (or to 255),
 // choosing the closer one, resolving ties using bankers' rounding.
-static int FindClosestDiscretized(int a, int bits) {
-  const int mask = (1 << bits) - 1;
-  const int biased = a + (mask >> 1) + ((a >> bits) & 1);
+static uint32_t FindClosestDiscretized(uint32_t a, int bits) {
+  const uint32_t mask = (1u << bits) - 1;
+  const uint32_t biased = a + (mask >> 1) + ((a >> bits) & 1);
   assert(bits > 0);
   if (biased > 0xff) return 0xff;
   return biased & ~mask;
@@ -69,22 +71,30 @@ static int IsSmooth(const uint32_t* const prev_row,
 }
 
 // Adjusts pixel values of image with given maximum error.
-static void NearLossless(int xsize, int ysize, uint32_t* argb,
-                         int limit_bits, uint32_t* copy_buffer) {
+static void NearLossless(int xsize, int ysize, const uint32_t* argb_src,
+                         int stride, int limit_bits, uint32_t* copy_buffer,
+                         uint32_t* argb_dst) {
   int x, y;
   const int limit = 1 << limit_bits;
   uint32_t* prev_row = copy_buffer;
   uint32_t* curr_row = prev_row + xsize;
   uint32_t* next_row = curr_row + xsize;
-  memcpy(copy_buffer, argb, xsize * 2 * sizeof(argb[0]));
+  memcpy(curr_row, argb_src, xsize * sizeof(argb_src[0]));
+  memcpy(next_row, argb_src + stride, xsize * sizeof(argb_src[0]));
 
-  for (y = 1; y < ysize - 1; ++y) {
-    uint32_t* const curr_argb_row = argb + y * xsize;
-    uint32_t* const next_argb_row = curr_argb_row + xsize;
-    memcpy(next_row, next_argb_row, xsize * sizeof(argb[0]));
-    for (x = 1; x < xsize - 1; ++x) {
-      if (!IsSmooth(prev_row, curr_row, next_row, x, limit)) {
-        curr_argb_row[x] = ClosestDiscretizedArgb(curr_row[x], limit_bits);
+  for (y = 0; y < ysize; ++y, argb_src += stride, argb_dst += xsize) {
+    if (y == 0 || y == ysize - 1) {
+      memcpy(argb_dst, argb_src, xsize * sizeof(argb_src[0]));
+    } else {
+      memcpy(next_row, argb_src + stride, xsize * sizeof(argb_src[0]));
+      argb_dst[0] = argb_src[0];
+      argb_dst[xsize - 1] = argb_src[xsize - 1];
+      for (x = 1; x < xsize - 1; ++x) {
+        if (IsSmooth(prev_row, curr_row, next_row, x, limit)) {
+          argb_dst[x] = curr_row[x];
+        } else {
+          argb_dst[x] = ClosestDiscretizedArgb(curr_row[x], limit_bits);
+        }
       }
     }
     {
@@ -97,26 +107,45 @@ static void NearLossless(int xsize, int ysize, uint32_t* argb,
   }
 }
 
-int VP8ApplyNearLossless(int xsize, int ysize, uint32_t* argb, int quality) {
+int VP8ApplyNearLossless(const WebPPicture* const picture, int quality,
+                         uint32_t* const argb_dst) {
   int i;
+  const int xsize = picture->width;
+  const int ysize = picture->height;
+  const int stride = picture->argb_stride;
   uint32_t* const copy_buffer =
       (uint32_t*)WebPSafeMalloc(xsize * 3, sizeof(*copy_buffer));
   const int limit_bits = VP8LNearLosslessBits(quality);
-  assert(argb != NULL);
-  assert(limit_bits >= 0);
+  assert(argb_dst != NULL);
+  assert(limit_bits > 0);
   assert(limit_bits <= MAX_LIMIT_BITS);
   if (copy_buffer == NULL) {
     return 0;
   }
   // For small icon images, don't attempt to apply near-lossless compression.
-  if (xsize < MIN_DIM_FOR_NEAR_LOSSLESS && ysize < MIN_DIM_FOR_NEAR_LOSSLESS) {
+  if ((xsize < MIN_DIM_FOR_NEAR_LOSSLESS &&
+       ysize < MIN_DIM_FOR_NEAR_LOSSLESS) ||
+      ysize < 3) {
+    for (i = 0; i < ysize; ++i) {
+      memcpy(argb_dst + i * xsize, picture->argb + i * picture->argb_stride,
+             xsize * sizeof(*argb_dst));
+    }
     WebPSafeFree(copy_buffer);
     return 1;
   }
 
-  for (i = limit_bits; i != 0; --i) {
-    NearLossless(xsize, ysize, argb, i, copy_buffer);
+  NearLossless(xsize, ysize, picture->argb, stride, limit_bits, copy_buffer,
+               argb_dst);
+  for (i = limit_bits - 1; i != 0; --i) {
+    NearLossless(xsize, ysize, argb_dst, xsize, i, copy_buffer, argb_dst);
   }
   WebPSafeFree(copy_buffer);
   return 1;
 }
+#else  // (WEBP_NEAR_LOSSLESS == 1)
+
+// Define a stub to suppress compiler warnings.
+extern void VP8LNearLosslessStub(void);
+WEBP_TSAN_IGNORE_FUNCTION void VP8LNearLosslessStub(void) {}
+
+#endif  // (WEBP_NEAR_LOSSLESS == 1)
diff --git a/src/3rdparty/libwebp/src/enc/picture_csp_enc.c b/src/3rdparty/libwebp/src/enc/picture_csp_enc.c
index e5d1c75..d531dd0 100644
--- a/src/3rdparty/libwebp/src/enc/picture_csp_enc.c
+++ b/src/3rdparty/libwebp/src/enc/picture_csp_enc.c
@@ -15,10 +15,12 @@
 #include <stdlib.h>
 #include <math.h>
 
-#include "./vp8i_enc.h"
-#include "../utils/random_utils.h"
-#include "../utils/utils.h"
-#include "../dsp/yuv.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/utils/random_utils.h"
+#include "src/utils/utils.h"
+#include "src/dsp/dsp.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/yuv.h"
 
 // Uncomment to disable gamma-compression during RGB->U/V averaging
 #define USE_GAMMA_COMPRESSION
@@ -39,12 +41,15 @@ static const union {
 static int CheckNonOpaque(const uint8_t* alpha, int width, int height,
                           int x_step, int y_step) {
   if (alpha == NULL) return 0;
-  while (height-- > 0) {
-    int x;
-    for (x = 0; x < width * x_step; x += x_step) {
-      if (alpha[x] != 0xff) return 1;  // TODO(skal): check 4/8 bytes at a time.
+  WebPInitAlphaProcessing();
+  if (x_step == 1) {
+    for (; height-- > 0; alpha += y_step) {
+      if (WebPHasAlpha8b(alpha, width)) return 1;
+    }
+  } else {
+    for (; height-- > 0; alpha += y_step) {
+      if (WebPHasAlpha32b(alpha, width)) return 1;
     }
-    alpha += y_step;
   }
   return 0;
 }
@@ -56,15 +61,10 @@ int WebPPictureHasTransparency(const WebPPicture* picture) {
     return CheckNonOpaque(picture->a, picture->width, picture->height,
                           1, picture->a_stride);
   } else {
-    int x, y;
-    const uint32_t* argb = picture->argb;
-    if (argb == NULL) return 0;
-    for (y = 0; y < picture->height; ++y) {
-      for (x = 0; x < picture->width; ++x) {
-        if (argb[x] < 0xff000000u) return 1;   // test any alpha values != 0xff
-      }
-      argb += picture->argb_stride;
-    }
+    const int alpha_offset = ALPHA_IS_LAST ? 3 : 0;
+    return CheckNonOpaque((const uint8_t*)picture->argb + alpha_offset,
+                          picture->width, picture->height,
+                          4, picture->argb_stride * sizeof(*picture->argb));
   }
   return 0;
 }
@@ -171,7 +171,7 @@ typedef uint16_t fixed_y_t;   // unsigned type with extra SFIX precision for W
 #if defined(USE_GAMMA_COMPRESSION)
 
 // float variant of gamma-correction
-// We use tables of different size and precision for the Rec709
+// We use tables of different size and precision for the Rec709 / BT2020
 // transfer function.
 #define kGammaF (1./0.45)
 static float kGammaToLinearTabF[MAX_Y_T + 1];   // size scales with Y_FIX
@@ -183,8 +183,8 @@ static WEBP_TSAN_IGNORE_FUNCTION void InitGammaTablesF(void) {
     int v;
     const double norm = 1. / MAX_Y_T;
     const double scale = 1. / kGammaTabSize;
-    const double a = 0.099;
-    const double thresh = 0.018;
+    const double a = 0.09929682680944;
+    const double thresh = 0.018053968510807;
     for (v = 0; v <= MAX_Y_T; ++v) {
       const double g = norm * v;
       if (g <= thresh * 4.5) {
@@ -856,7 +856,6 @@ static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
     return 0;
   }
   if (has_alpha) {
-    WebPInitAlphaProcessing();
     assert(step == 4);
 #if defined(USE_GAMMA_COMPRESSION) && defined(USE_INVERSE_ALPHA_TABLE)
     assert(kAlphaFix + kGammaFix <= 31);
@@ -1085,40 +1084,45 @@ int WebPPictureYUVAToARGB(WebPPicture* picture) {
 // automatic import / conversion
 
 static int Import(WebPPicture* const picture,
-                  const uint8_t* const rgb, int rgb_stride,
+                  const uint8_t* rgb, int rgb_stride,
                   int step, int swap_rb, int import_alpha) {
   int y;
   const uint8_t* r_ptr = rgb + (swap_rb ? 2 : 0);
   const uint8_t* g_ptr = rgb + 1;
   const uint8_t* b_ptr = rgb + (swap_rb ? 0 : 2);
-  const uint8_t* a_ptr = import_alpha ? rgb + 3 : NULL;
   const int width = picture->width;
   const int height = picture->height;
 
   if (!picture->use_argb) {
+    const uint8_t* a_ptr = import_alpha ? rgb + 3 : NULL;
     return ImportYUVAFromRGBA(r_ptr, g_ptr, b_ptr, a_ptr, step, rgb_stride,
                               0.f /* no dithering */, 0, picture);
   }
   if (!WebPPictureAlloc(picture)) return 0;
 
-  VP8EncDspARGBInit();
+  VP8LDspInit();
+  WebPInitAlphaProcessing();
 
   if (import_alpha) {
     uint32_t* dst = picture->argb;
+    const int do_copy =
+        (!swap_rb && !ALPHA_IS_LAST) || (swap_rb && ALPHA_IS_LAST);
     assert(step == 4);
     for (y = 0; y < height; ++y) {
-      VP8PackARGB(a_ptr, r_ptr, g_ptr, b_ptr, width, dst);
-      a_ptr += rgb_stride;
-      r_ptr += rgb_stride;
-      g_ptr += rgb_stride;
-      b_ptr += rgb_stride;
+      if (do_copy) {
+        memcpy(dst, rgb, width * 4);
+      } else {
+        // RGBA input order. Need to swap R and B.
+        VP8LConvertBGRAToRGBA((const uint32_t*)rgb, width, (uint8_t*)dst);
+      }
+      rgb += rgb_stride;
       dst += picture->argb_stride;
     }
   } else {
     uint32_t* dst = picture->argb;
     assert(step >= 3);
     for (y = 0; y < height; ++y) {
-      VP8PackRGB(r_ptr, g_ptr, b_ptr, width, step, dst);
+      WebPPackRGB(r_ptr, g_ptr, b_ptr, width, step, dst);
       r_ptr += rgb_stride;
       g_ptr += rgb_stride;
       b_ptr += rgb_stride;
@@ -1130,12 +1134,7 @@ static int Import(WebPPicture* const picture,
 
 // Public API
 
-int WebPPictureImportRGB(WebPPicture* picture,
-                         const uint8_t* rgb, int rgb_stride) {
-  return (picture != NULL && rgb != NULL)
-             ? Import(picture, rgb, rgb_stride, 3, 0, 0)
-             : 0;
-}
+#if !defined(WEBP_REDUCE_CSP)
 
 int WebPPictureImportBGR(WebPPicture* picture,
                          const uint8_t* rgb, int rgb_stride) {
@@ -1144,31 +1143,41 @@ int WebPPictureImportBGR(WebPPicture* picture,
              : 0;
 }
 
-int WebPPictureImportRGBA(WebPPicture* picture,
+int WebPPictureImportBGRA(WebPPicture* picture,
                           const uint8_t* rgba, int rgba_stride) {
   return (picture != NULL && rgba != NULL)
-             ? Import(picture, rgba, rgba_stride, 4, 0, 1)
+             ? Import(picture, rgba, rgba_stride, 4, 1, 1)
              : 0;
 }
 
-int WebPPictureImportBGRA(WebPPicture* picture,
+
+int WebPPictureImportBGRX(WebPPicture* picture,
                           const uint8_t* rgba, int rgba_stride) {
   return (picture != NULL && rgba != NULL)
-             ? Import(picture, rgba, rgba_stride, 4, 1, 1)
+             ? Import(picture, rgba, rgba_stride, 4, 1, 0)
              : 0;
 }
 
-int WebPPictureImportRGBX(WebPPicture* picture,
+#endif   // WEBP_REDUCE_CSP
+
+int WebPPictureImportRGB(WebPPicture* picture,
+                         const uint8_t* rgb, int rgb_stride) {
+  return (picture != NULL && rgb != NULL)
+             ? Import(picture, rgb, rgb_stride, 3, 0, 0)
+             : 0;
+}
+
+int WebPPictureImportRGBA(WebPPicture* picture,
                           const uint8_t* rgba, int rgba_stride) {
   return (picture != NULL && rgba != NULL)
-             ? Import(picture, rgba, rgba_stride, 4, 0, 0)
+             ? Import(picture, rgba, rgba_stride, 4, 0, 1)
              : 0;
 }
 
-int WebPPictureImportBGRX(WebPPicture* picture,
+int WebPPictureImportRGBX(WebPPicture* picture,
                           const uint8_t* rgba, int rgba_stride) {
   return (picture != NULL && rgba != NULL)
-             ? Import(picture, rgba, rgba_stride, 4, 1, 0)
+             ? Import(picture, rgba, rgba_stride, 4, 0, 0)
              : 0;
 }
 
diff --git a/src/3rdparty/libwebp/src/enc/picture_enc.c b/src/3rdparty/libwebp/src/enc/picture_enc.c
index dfa6651..c691622 100644
--- a/src/3rdparty/libwebp/src/enc/picture_enc.c
+++ b/src/3rdparty/libwebp/src/enc/picture_enc.c
@@ -14,9 +14,9 @@
 #include <assert.h>
 #include <stdlib.h>
 
-#include "./vp8i_enc.h"
-#include "../dsp/dsp.h"
-#include "../utils/utils.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // WebPPicture
@@ -76,13 +76,12 @@ int WebPPictureAllocARGB(WebPPicture* const picture, int width, int height) {
     return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
   }
   // allocate a new buffer.
-  memory = WebPSafeMalloc(argb_size, sizeof(*picture->argb));
+  memory = WebPSafeMalloc(argb_size + WEBP_ALIGN_CST, sizeof(*picture->argb));
   if (memory == NULL) {
     return WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
   }
-  // TODO(skal): align plane to cache line?
   picture->memory_argb_ = memory;
-  picture->argb = (uint32_t*)memory;
+  picture->argb = (uint32_t*)WEBP_ALIGN(memory);
   picture->argb_stride = width;
   return 1;
 }
@@ -92,8 +91,8 @@ int WebPPictureAllocYUVA(WebPPicture* const picture, int width, int height) {
       (WebPEncCSP)((int)picture->colorspace & WEBP_CSP_UV_MASK);
   const int has_alpha = (int)picture->colorspace & WEBP_CSP_ALPHA_BIT;
   const int y_stride = width;
-  const int uv_width = (width + 1) >> 1;
-  const int uv_height = (height + 1) >> 1;
+  const int uv_width = (int)(((int64_t)width + 1) >> 1);
+  const int uv_height = (int)(((int64_t)height + 1) >> 1);
   const int uv_stride = uv_width;
   int a_width, a_stride;
   uint64_t y_size, uv_size, a_size, total_size;
@@ -118,8 +117,8 @@ int WebPPictureAllocYUVA(WebPPicture* const picture, int width, int height) {
   total_size = y_size + a_size + 2 * uv_size;
 
   // Security and validation checks
-  if (width <= 0 || height <= 0 ||         // luma/alpha param error
-      uv_width < 0 || uv_height < 0) {     // u/v param error
+  if (width <= 0 || height <= 0 ||           // luma/alpha param error
+      uv_width <= 0 || uv_height <= 0) {     // u/v param error
     return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
   }
   // allocate a new buffer.
@@ -271,9 +270,11 @@ size_t NAME(const uint8_t* in, int w, int h, int bps, float q,          \
 }
 
 ENCODE_FUNC(WebPEncodeRGB, WebPPictureImportRGB)
-ENCODE_FUNC(WebPEncodeBGR, WebPPictureImportBGR)
 ENCODE_FUNC(WebPEncodeRGBA, WebPPictureImportRGBA)
+#if !defined(WEBP_REDUCE_CSP)
+ENCODE_FUNC(WebPEncodeBGR, WebPPictureImportBGR)
 ENCODE_FUNC(WebPEncodeBGRA, WebPPictureImportBGRA)
+#endif  // WEBP_REDUCE_CSP
 
 #undef ENCODE_FUNC
 
@@ -284,9 +285,11 @@ size_t NAME(const uint8_t* in, int w, int h, int bps, uint8_t** out) {       \
 }
 
 LOSSLESS_ENCODE_FUNC(WebPEncodeLosslessRGB, WebPPictureImportRGB)
-LOSSLESS_ENCODE_FUNC(WebPEncodeLosslessBGR, WebPPictureImportBGR)
 LOSSLESS_ENCODE_FUNC(WebPEncodeLosslessRGBA, WebPPictureImportRGBA)
+#if !defined(WEBP_REDUCE_CSP)
+LOSSLESS_ENCODE_FUNC(WebPEncodeLosslessBGR, WebPPictureImportBGR)
 LOSSLESS_ENCODE_FUNC(WebPEncodeLosslessBGRA, WebPPictureImportBGRA)
+#endif  // WEBP_REDUCE_CSP
 
 #undef LOSSLESS_ENCODE_FUNC
 
diff --git a/src/3rdparty/libwebp/src/enc/picture_psnr_enc.c b/src/3rdparty/libwebp/src/enc/picture_psnr_enc.c
index 9c0b229..362a7c7 100644
--- a/src/3rdparty/libwebp/src/enc/picture_psnr_enc.c
+++ b/src/3rdparty/libwebp/src/enc/picture_psnr_enc.c
@@ -11,11 +11,15 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
+#include "src/webp/encode.h"
+
+#if !(defined(WEBP_DISABLE_STATS) || defined(WEBP_REDUCE_SIZE))
+
 #include <math.h>
 #include <stdlib.h>
 
-#include "./vp8i_enc.h"
-#include "../utils/utils.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/utils/utils.h"
 
 typedef double (*AccumulateFunc)(const uint8_t* src, int src_stride,
                                  const uint8_t* ref, int ref_stride,
@@ -210,4 +214,34 @@ int WebPPictureDistortion(const WebPPicture* src, const WebPPicture* ref,
   return ok;
 }
 
-//------------------------------------------------------------------------------
+#else  // defined(WEBP_DISABLE_STATS)
+int WebPPlaneDistortion(const uint8_t* src, size_t src_stride,
+                        const uint8_t* ref, size_t ref_stride,
+                        int width, int height, size_t x_step,
+                        int type, float* distortion, float* result) {
+  (void)src;
+  (void)src_stride;
+  (void)ref;
+  (void)ref_stride;
+  (void)width;
+  (void)height;
+  (void)x_step;
+  (void)type;
+  if (distortion == NULL || result == NULL) return 0;
+  *distortion = 0.f;
+  *result = 0.f;
+  return 1;
+}
+
+int WebPPictureDistortion(const WebPPicture* src, const WebPPicture* ref,
+                          int type, float results[5]) {
+  int i;
+  (void)src;
+  (void)ref;
+  (void)type;
+  if (results == NULL) return 0;
+  for (i = 0; i < 5; ++i) results[i] = 0.f;
+  return 1;
+}
+
+#endif  // !defined(WEBP_DISABLE_STATS)
diff --git a/src/3rdparty/libwebp/src/enc/picture_rescale_enc.c b/src/3rdparty/libwebp/src/enc/picture_rescale_enc.c
index 0b7181c..58a6ae7 100644
--- a/src/3rdparty/libwebp/src/enc/picture_rescale_enc.c
+++ b/src/3rdparty/libwebp/src/enc/picture_rescale_enc.c
@@ -11,12 +11,16 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
+#include "src/webp/encode.h"
+
+#if !defined(WEBP_REDUCE_SIZE)
+
 #include <assert.h>
 #include <stdlib.h>
 
-#include "./vp8i_enc.h"
-#include "../utils/rescaler_utils.h"
-#include "../utils/utils.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/utils/rescaler_utils.h"
+#include "src/utils/utils.h"
 
 #define HALVE(x) (((x) + 1) >> 1)
 
@@ -261,4 +265,45 @@ int WebPPictureRescale(WebPPicture* pic, int width, int height) {
   return 1;
 }
 
-//------------------------------------------------------------------------------
+#else  // defined(WEBP_REDUCE_SIZE)
+
+int WebPPictureCopy(const WebPPicture* src, WebPPicture* dst) {
+  (void)src;
+  (void)dst;
+  return 0;
+}
+
+int WebPPictureIsView(const WebPPicture* picture) {
+  (void)picture;
+  return 0;
+}
+
+int WebPPictureView(const WebPPicture* src,
+                    int left, int top, int width, int height,
+                    WebPPicture* dst) {
+  (void)src;
+  (void)left;
+  (void)top;
+  (void)width;
+  (void)height;
+  (void)dst;
+  return 0;
+}
+
+int WebPPictureCrop(WebPPicture* pic,
+                    int left, int top, int width, int height) {
+  (void)pic;
+  (void)left;
+  (void)top;
+  (void)width;
+  (void)height;
+  return 0;
+}
+
+int WebPPictureRescale(WebPPicture* pic, int width, int height) {
+  (void)pic;
+  (void)width;
+  (void)height;
+  return 0;
+}
+#endif  // !defined(WEBP_REDUCE_SIZE)
diff --git a/src/3rdparty/libwebp/src/enc/picture_tools_enc.c b/src/3rdparty/libwebp/src/enc/picture_tools_enc.c
index 895df51..be292d4 100644
--- a/src/3rdparty/libwebp/src/enc/picture_tools_enc.c
+++ b/src/3rdparty/libwebp/src/enc/picture_tools_enc.c
@@ -13,8 +13,8 @@
 
 #include <assert.h>
 
-#include "./vp8i_enc.h"
-#include "../dsp/yuv.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/dsp/yuv.h"
 
 static WEBP_INLINE uint32_t MakeARGB32(int r, int g, int b) {
   return (0xff000000u | (r << 16) | (g << 8) | b);
@@ -25,20 +25,7 @@ static WEBP_INLINE uint32_t MakeARGB32(int r, int g, int b) {
 
 #define SIZE 8
 #define SIZE2 (SIZE / 2)
-static int is_transparent_area(const uint8_t* ptr, int stride, int size) {
-  int y, x;
-  for (y = 0; y < size; ++y) {
-    for (x = 0; x < size; ++x) {
-      if (ptr[x]) {
-        return 0;
-      }
-    }
-    ptr += stride;
-  }
-  return 1;
-}
-
-static int is_transparent_argb_area(const uint32_t* ptr, int stride, int size) {
+static int IsTransparentARGBArea(const uint32_t* ptr, int stride, int size) {
   int y, x;
   for (y = 0; y < size; ++y) {
     for (x = 0; x < size; ++x) {
@@ -51,7 +38,7 @@ static int is_transparent_argb_area(const uint32_t* ptr, int stride, int size) {
   return 1;
 }
 
-static void flatten(uint8_t* ptr, int v, int stride, int size) {
+static void Flatten(uint8_t* ptr, int v, int stride, int size) {
   int y;
   for (y = 0; y < size; ++y) {
     memset(ptr, v, size);
@@ -59,7 +46,7 @@ static void flatten(uint8_t* ptr, int v, int stride, int size) {
   }
 }
 
-static void flatten_argb(uint32_t* ptr, uint32_t v, int stride, int size) {
+static void FlattenARGB(uint32_t* ptr, uint32_t v, int stride, int size) {
   int x, y;
   for (y = 0; y < size; ++y) {
     for (x = 0; x < size; ++x) ptr[x] = v;
@@ -67,54 +54,114 @@ static void flatten_argb(uint32_t* ptr, uint32_t v, int stride, int size) {
   }
 }
 
+// Smoothen the luma components of transparent pixels. Return true if the whole
+// block is transparent.
+static int SmoothenBlock(const uint8_t* a_ptr, int a_stride, uint8_t* y_ptr,
+                         int y_stride, int width, int height) {
+  int sum = 0, count = 0;
+  int x, y;
+  const uint8_t* alpha_ptr = a_ptr;
+  uint8_t* luma_ptr = y_ptr;
+  for (y = 0; y < height; ++y) {
+    for (x = 0; x < width; ++x) {
+      if (alpha_ptr[x] != 0) {
+        ++count;
+        sum += luma_ptr[x];
+      }
+    }
+    alpha_ptr += a_stride;
+    luma_ptr += y_stride;
+  }
+  if (count > 0 && count < width * height) {
+    const uint8_t avg_u8 = (uint8_t)(sum / count);
+    alpha_ptr = a_ptr;
+    luma_ptr = y_ptr;
+    for (y = 0; y < height; ++y) {
+      for (x = 0; x < width; ++x) {
+        if (alpha_ptr[x] == 0) luma_ptr[x] = avg_u8;
+      }
+      alpha_ptr += a_stride;
+      luma_ptr += y_stride;
+    }
+  }
+  return (count == 0);
+}
+
 void WebPCleanupTransparentArea(WebPPicture* pic) {
   int x, y, w, h;
   if (pic == NULL) return;
   w = pic->width / SIZE;
   h = pic->height / SIZE;
 
-  // note: we ignore the left-overs on right/bottom
+  // note: we ignore the left-overs on right/bottom, except for SmoothenBlock().
   if (pic->use_argb) {
     uint32_t argb_value = 0;
     for (y = 0; y < h; ++y) {
       int need_reset = 1;
       for (x = 0; x < w; ++x) {
         const int off = (y * pic->argb_stride + x) * SIZE;
-        if (is_transparent_argb_area(pic->argb + off, pic->argb_stride, SIZE)) {
+        if (IsTransparentARGBArea(pic->argb + off, pic->argb_stride, SIZE)) {
           if (need_reset) {
             argb_value = pic->argb[off];
             need_reset = 0;
           }
-          flatten_argb(pic->argb + off, argb_value, pic->argb_stride, SIZE);
+          FlattenARGB(pic->argb + off, argb_value, pic->argb_stride, SIZE);
         } else {
           need_reset = 1;
         }
       }
     }
   } else {
-    const uint8_t* const a_ptr = pic->a;
+    const int width = pic->width;
+    const int height = pic->height;
+    const int y_stride = pic->y_stride;
+    const int uv_stride = pic->uv_stride;
+    const int a_stride = pic->a_stride;
+    uint8_t* y_ptr = pic->y;
+    uint8_t* u_ptr = pic->u;
+    uint8_t* v_ptr = pic->v;
+    const uint8_t* a_ptr = pic->a;
     int values[3] = { 0 };
-    if (a_ptr == NULL) return;    // nothing to do
-    for (y = 0; y < h; ++y) {
+    if (a_ptr == NULL || y_ptr == NULL || u_ptr == NULL || v_ptr == NULL) {
+      return;
+    }
+    for (y = 0; y + SIZE <= height; y += SIZE) {
       int need_reset = 1;
-      for (x = 0; x < w; ++x) {
-        const int off_a = (y * pic->a_stride + x) * SIZE;
-        const int off_y = (y * pic->y_stride + x) * SIZE;
-        const int off_uv = (y * pic->uv_stride + x) * SIZE2;
-        if (is_transparent_area(a_ptr + off_a, pic->a_stride, SIZE)) {
+      for (x = 0; x + SIZE <= width; x += SIZE) {
+        if (SmoothenBlock(a_ptr + x, a_stride, y_ptr + x, y_stride,
+                          SIZE, SIZE)) {
           if (need_reset) {
-            values[0] = pic->y[off_y];
-            values[1] = pic->u[off_uv];
-            values[2] = pic->v[off_uv];
+            values[0] = y_ptr[x];
+            values[1] = u_ptr[x >> 1];
+            values[2] = v_ptr[x >> 1];
             need_reset = 0;
           }
-          flatten(pic->y + off_y, values[0], pic->y_stride, SIZE);
-          flatten(pic->u + off_uv, values[1], pic->uv_stride, SIZE2);
-          flatten(pic->v + off_uv, values[2], pic->uv_stride, SIZE2);
+          Flatten(y_ptr + x,        values[0], y_stride,  SIZE);
+          Flatten(u_ptr + (x >> 1), values[1], uv_stride, SIZE2);
+          Flatten(v_ptr + (x >> 1), values[2], uv_stride, SIZE2);
         } else {
           need_reset = 1;
         }
       }
+      if (x < width) {
+        SmoothenBlock(a_ptr + x, a_stride, y_ptr + x, y_stride,
+                      width - x, SIZE);
+      }
+      a_ptr += SIZE * a_stride;
+      y_ptr += SIZE * y_stride;
+      u_ptr += SIZE2 * uv_stride;
+      v_ptr += SIZE2 * uv_stride;
+    }
+    if (y < height) {
+      const int sub_height = height - y;
+      for (x = 0; x + SIZE <= width; x += SIZE) {
+        SmoothenBlock(a_ptr + x, a_stride, y_ptr + x, y_stride,
+                      SIZE, sub_height);
+      }
+      if (x < width) {
+        SmoothenBlock(a_ptr + x, a_stride, y_ptr + x, y_stride,
+                      width - x, sub_height);
+      }
     }
   }
 }
@@ -144,9 +191,9 @@ void WebPCleanupTransparentAreaLossless(WebPPicture* const pic) {
 // Blend color and remove transparency info
 
 #define BLEND(V0, V1, ALPHA) \
-    ((((V0) * (255 - (ALPHA)) + (V1) * (ALPHA)) * 0x101) >> 16)
+    ((((V0) * (255 - (ALPHA)) + (V1) * (ALPHA)) * 0x101 + 256) >> 16)
 #define BLEND_10BIT(V0, V1, ALPHA) \
-    ((((V0) * (1020 - (ALPHA)) + (V1) * (ALPHA)) * 0x101) >> 18)
+    ((((V0) * (1020 - (ALPHA)) + (V1) * (ALPHA)) * 0x101 + 1024) >> 18)
 
 void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
   const int red = (background_rgb >> 16) & 0xff;
diff --git a/src/3rdparty/libwebp/src/enc/predictor_enc.c b/src/3rdparty/libwebp/src/enc/predictor_enc.c
index 0639b74..f3715f5 100644
--- a/src/3rdparty/libwebp/src/enc/predictor_enc.c
+++ b/src/3rdparty/libwebp/src/enc/predictor_enc.c
@@ -14,9 +14,9 @@
 //          Urvang Joshi (urvang@google.com)
 //          Vincent Rabaud (vrabaud@google.com)
 
-#include "../dsp/lossless.h"
-#include "../dsp/lossless_common.h"
-#include "./vp8li_enc.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+#include "src/enc/vp8li_enc.h"
 
 #define MAX_DIFF_COST (1e30f)
 
@@ -26,7 +26,6 @@ static const uint32_t kMaskAlpha = 0xff000000;
 
 // Mostly used to reduce code size + readability
 static WEBP_INLINE int GetMin(int a, int b) { return (a > b) ? b : a; }
-static WEBP_INLINE int GetMax(int a, int b) { return (a < b) ? b : a; }
 
 //------------------------------------------------------------------------------
 // Methods to calculate Entropy (Shannon).
@@ -90,6 +89,9 @@ static WEBP_INLINE void PredictBatch(int mode, int x_start, int y,
   }
 }
 
+#if (WEBP_NEAR_LOSSLESS == 1)
+static WEBP_INLINE int GetMax(int a, int b) { return (a < b) ? b : a; }
+
 static int MaxDiffBetweenPixels(uint32_t p1, uint32_t p2) {
   const int diff_a = abs((int)(p1 >> 24) - (int)(p2 >> 24));
   const int diff_r = abs((int)((p1 >> 16) & 0xff) - (int)((p2 >> 16) & 0xff));
@@ -180,6 +182,7 @@ static uint8_t NearLosslessComponent(uint8_t value, uint8_t predict,
 // max_quantization which is a power of 2, smaller than max_diff). Take care if
 // value and predict have undergone subtract green, which means that red and
 // blue are represented as offsets from green.
+#define NEAR_LOSSLESS_DIFF(a, b) (uint8_t)((((int)(a) - (int)(b))) & 0xff)
 static uint32_t NearLossless(uint32_t value, uint32_t predict,
                              int max_quantization, int max_diff,
                              int used_subtract_green) {
@@ -196,7 +199,7 @@ static uint32_t NearLossless(uint32_t value, uint32_t predict,
   }
   if ((value >> 24) == 0 || (value >> 24) == 0xff) {
     // Preserve transparency of fully transparent or fully opaque pixels.
-    a = ((value >> 24) - (predict >> 24)) & 0xff;
+    a = NEAR_LOSSLESS_DIFF(value >> 24, predict >> 24);
   } else {
     a = NearLosslessComponent(value >> 24, predict >> 24, 0xff, quantization);
   }
@@ -209,15 +212,17 @@ static uint32_t NearLossless(uint32_t value, uint32_t predict,
     // The amount by which green has been adjusted during quantization. It is
     // subtracted from red and blue for compensation, to avoid accumulating two
     // quantization errors in them.
-    green_diff = (new_green - (value >> 8)) & 0xff;
+    green_diff = NEAR_LOSSLESS_DIFF(new_green, value >> 8);
   }
-  r = NearLosslessComponent(((value >> 16) - green_diff) & 0xff,
+  r = NearLosslessComponent(NEAR_LOSSLESS_DIFF(value >> 16, green_diff),
                             (predict >> 16) & 0xff, 0xff - new_green,
                             quantization);
-  b = NearLosslessComponent((value - green_diff) & 0xff, predict & 0xff,
-                            0xff - new_green, quantization);
+  b = NearLosslessComponent(NEAR_LOSSLESS_DIFF(value, green_diff),
+                            predict & 0xff, 0xff - new_green, quantization);
   return ((uint32_t)a << 24) | ((uint32_t)r << 16) | ((uint32_t)g << 8) | b;
 }
+#undef NEAR_LOSSLESS_DIFF
+#endif  // (WEBP_NEAR_LOSSLESS == 1)
 
 // Stores the difference between the pixel and its prediction in "out".
 // In case of a lossy encoding, updates the source image to avoid propagating
@@ -244,6 +249,7 @@ static WEBP_INLINE void GetResidual(
       } else {
         predict = pred_func(current_row[x - 1], upper_row + x);
       }
+#if (WEBP_NEAR_LOSSLESS == 1)
       if (max_quantization == 1 || mode == 0 || y == 0 || y == height - 1 ||
           x == 0 || x == width - 1) {
         residual = VP8LSubPixels(current_row[x], predict);
@@ -254,6 +260,13 @@ static WEBP_INLINE void GetResidual(
         current_row[x] = VP8LAddPixels(predict, residual);
         // x is never 0 here so we do not need to update upper_row like below.
       }
+#else
+      (void)max_diffs;
+      (void)height;
+      (void)max_quantization;
+      (void)used_subtract_green;
+      residual = VP8LSubPixels(current_row[x], predict);
+#endif
       if ((current_row[x] & kMaskAlpha) == 0) {
         // If alpha is 0, cleanup RGB. We can choose the RGB values of the
         // residual for best compression. The prediction of alpha itself can be
@@ -296,11 +309,12 @@ static int GetBestPredictorForTile(int width, int height,
   const int max_x = GetMin(tile_size, width - start_x);
   // Whether there exist columns just outside the tile.
   const int have_left = (start_x > 0);
-  const int have_right = (max_x < width - start_x);
   // Position and size of the strip covering the tile and adjacent columns if
   // they exist.
   const int context_start_x = start_x - have_left;
-  const int context_width = max_x + have_left + have_right;
+#if (WEBP_NEAR_LOSSLESS == 1)
+  const int context_width = max_x + have_left + (max_x < width - start_x);
+#endif
   const int tiles_per_row = VP8LSubSampleSize(width, bits);
   // Prediction modes of the left and above neighbor tiles.
   const int left_mode = (tile_x > 0) ?
@@ -352,10 +366,12 @@ static int GetBestPredictorForTile(int width, int height,
       memcpy(current_row + context_start_x,
              argb + y * width + context_start_x,
              sizeof(*argb) * (max_x + have_left + (y + 1 < height)));
+#if (WEBP_NEAR_LOSSLESS == 1)
       if (max_quantization > 1 && y >= 1 && y + 1 < height) {
         MaxDiffsForRow(context_width, width, argb + y * width + context_start_x,
                        max_diffs + context_start_x, used_subtract_green);
       }
+#endif
 
       GetResidual(width, height, upper_row, current_row, max_diffs, mode,
                   start_x, start_x + max_x, y, max_quantization, exact,
@@ -405,7 +421,9 @@ static void CopyImageWithPrediction(int width, int height,
   uint32_t* upper_row = argb_scratch;
   uint32_t* current_row = upper_row + width + 1;
   uint8_t* current_max_diffs = (uint8_t*)(current_row + width + 1);
+#if (WEBP_NEAR_LOSSLESS == 1)
   uint8_t* lower_max_diffs = current_max_diffs + width;
+#endif
   int y;
 
   for (y = 0; y < height; ++y) {
@@ -420,6 +438,7 @@ static void CopyImageWithPrediction(int width, int height,
       PredictBatch(kPredLowEffort, 0, y, width, current_row, upper_row,
                    argb + y * width);
     } else {
+#if (WEBP_NEAR_LOSSLESS == 1)
       if (max_quantization > 1) {
         // Compute max_diffs for the lower row now, because that needs the
         // contents of argb for the current row, which we will overwrite with
@@ -432,6 +451,7 @@ static void CopyImageWithPrediction(int width, int height,
                          used_subtract_green);
         }
       }
+#endif
       for (x = 0; x < width;) {
         const int mode =
             (modes[(y >> bits) * tiles_per_row + (x >> bits)] >> 8) & 0xff;
diff --git a/src/3rdparty/libwebp/src/enc/quant_enc.c b/src/3rdparty/libwebp/src/enc/quant_enc.c
index b118fb2..3b1a312 100644
--- a/src/3rdparty/libwebp/src/enc/quant_enc.c
+++ b/src/3rdparty/libwebp/src/enc/quant_enc.c
@@ -15,8 +15,8 @@
 #include <math.h>
 #include <stdlib.h>  // for abs()
 
-#include "./vp8i_enc.h"
-#include "./cost_enc.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/enc/cost_enc.h"
 
 #define DO_TRELLIS_I4  1
 #define DO_TRELLIS_I16 1   // not a huge gain, but ok at low bitrate.
@@ -457,11 +457,11 @@ void VP8SetSegmentParams(VP8Encoder* const enc, float quality) {
 // Form the predictions in cache
 
 // Must be ordered using {DC_PRED, TM_PRED, V_PRED, H_PRED} as index
-const int VP8I16ModeOffsets[4] = { I16DC16, I16TM16, I16VE16, I16HE16 };
-const int VP8UVModeOffsets[4] = { C8DC8, C8TM8, C8VE8, C8HE8 };
+const uint16_t VP8I16ModeOffsets[4] = { I16DC16, I16TM16, I16VE16, I16HE16 };
+const uint16_t VP8UVModeOffsets[4] = { C8DC8, C8TM8, C8VE8, C8HE8 };
 
 // Must be indexed using {B_DC_PRED -> B_HU_PRED} as index
-const int VP8I4ModeOffsets[NUM_BMODES] = {
+const uint16_t VP8I4ModeOffsets[NUM_BMODES] = {
   I4DC4, I4TM4, I4VE4, I4HE4, I4RD4, I4VR4, I4LD4, I4VL4, I4HD4, I4HU4
 };
 
@@ -492,14 +492,14 @@ void VP8MakeIntra4Preds(const VP8EncIterator* const it) {
 // |YYYY|....| 12
 // +----+----+
 
-const int VP8Scan[16] = {  // Luma
+const uint16_t VP8Scan[16] = {  // Luma
   0 +  0 * BPS,  4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS,
   0 +  4 * BPS,  4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS,
   0 +  8 * BPS,  4 +  8 * BPS, 8 +  8 * BPS, 12 +  8 * BPS,
   0 + 12 * BPS,  4 + 12 * BPS, 8 + 12 * BPS, 12 + 12 * BPS,
 };
 
-static const int VP8ScanUV[4 + 4] = {
+static const uint16_t VP8ScanUV[4 + 4] = {
   0 + 0 * BPS,   4 + 0 * BPS, 0 + 4 * BPS,  4 + 4 * BPS,    // U
   8 + 0 * BPS,  12 + 0 * BPS, 8 + 4 * BPS, 12 + 4 * BPS     // V
 };
@@ -1162,7 +1162,7 @@ static void RefineUsingDistortion(VP8EncIterator* const it,
     const uint8_t* const src = it->yuv_in_ + Y_OFF_ENC;
     for (mode = 0; mode < NUM_PRED_MODES; ++mode) {
       const uint8_t* const ref = it->yuv_p_ + VP8I16ModeOffsets[mode];
-      const score_t score = VP8SSE16x16(src, ref) * RD_DISTO_MULT
+      const score_t score = (score_t)VP8SSE16x16(src, ref) * RD_DISTO_MULT
                           + VP8FixedCostsI16[mode] * lambda_d_i16;
       if (mode > 0 && VP8FixedCostsI16[mode] > bit_limit) {
         continue;
diff --git a/src/3rdparty/libwebp/src/enc/syntax_enc.c b/src/3rdparty/libwebp/src/enc/syntax_enc.c
index 90665bd..a9e5a6c 100644
--- a/src/3rdparty/libwebp/src/enc/syntax_enc.c
+++ b/src/3rdparty/libwebp/src/enc/syntax_enc.c
@@ -13,10 +13,10 @@
 
 #include <assert.h>
 
-#include "../utils/utils.h"
-#include "../webp/format_constants.h"  // RIFF constants
-#include "../webp/mux_types.h"         // ALPHA_FLAG
-#include "./vp8i_enc.h"
+#include "src/utils/utils.h"
+#include "src/webp/format_constants.h"  // RIFF constants
+#include "src/webp/mux_types.h"         // ALPHA_FLAG
+#include "src/enc/vp8i_enc.h"
 
 //------------------------------------------------------------------------------
 // Helper functions
@@ -289,11 +289,17 @@ static int GeneratePartition0(VP8Encoder* const enc) {
 
   pos3 = VP8BitWriterPos(bw);
 
+#if !defined(WEBP_DISABLE_STATS)
   if (enc->pic_->stats) {
     enc->pic_->stats->header_bytes[0] = (int)((pos2 - pos1 + 7) >> 3);
     enc->pic_->stats->header_bytes[1] = (int)((pos3 - pos2 + 7) >> 3);
     enc->pic_->stats->alpha_data_size = (int)enc->alpha_data_size_;
   }
+#else
+  (void)pos1;
+  (void)pos2;
+  (void)pos3;
+#endif
   if (bw->error_) {
     return WebPEncodingSetError(enc->pic_, VP8_ENC_ERROR_OUT_OF_MEMORY);
   }
diff --git a/src/3rdparty/libwebp/src/enc/token_enc.c b/src/3rdparty/libwebp/src/enc/token_enc.c
index 02a0d72..3a2192a 100644
--- a/src/3rdparty/libwebp/src/enc/token_enc.c
+++ b/src/3rdparty/libwebp/src/enc/token_enc.c
@@ -20,9 +20,9 @@
 #include <stdlib.h>
 #include <string.h>
 
-#include "./cost_enc.h"
-#include "./vp8i_enc.h"
-#include "../utils/utils.h"
+#include "src/enc/cost_enc.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/utils/utils.h"
 
 #if !defined(DISABLE_TOKEN_BUFFER)
 
@@ -195,39 +195,6 @@ int VP8RecordCoeffTokens(int ctx, const struct VP8Residual* const res,
 #undef TOKEN_ID
 
 //------------------------------------------------------------------------------
-// This function works, but isn't currently used. Saved for later.
-
-#if 0
-
-static void Record(int bit, proba_t* const stats) {
-  proba_t p = *stats;
-  if (p >= 0xffff0000u) {               // an overflow is inbound.
-    p = ((p + 1u) >> 1) & 0x7fff7fffu;  // -> divide the stats by 2.
-  }
-  // record bit count (lower 16 bits) and increment total count (upper 16 bits).
-  p += 0x00010000u + bit;
-  *stats = p;
-}
-
-void VP8TokenToStats(const VP8TBuffer* const b, proba_t* const stats) {
-  const VP8Tokens* p = b->pages_;
-  while (p != NULL) {
-    const int N = (p->next_ == NULL) ? b->left_ : 0;
-    int n = MAX_NUM_TOKEN;
-    const token_t* const tokens = TOKEN_DATA(p);
-    while (n-- > N) {
-      const token_t token = tokens[n];
-      if (!(token & FIXED_PROBA_BIT)) {
-        Record((token >> 15) & 1, stats + (token & 0x3fffu));
-      }
-    }
-    p = p->next_;
-  }
-}
-
-#endif   // 0
-
-//------------------------------------------------------------------------------
 // Final coding pass, with known probabilities
 
 int VP8EmitTokens(VP8TBuffer* const b, VP8BitWriter* const bw,
@@ -283,8 +250,9 @@ size_t VP8EstimateTokenSize(VP8TBuffer* const b, const uint8_t* const probas) {
 
 #else     // DISABLE_TOKEN_BUFFER
 
-void VP8TBufferInit(VP8TBuffer* const b) {
+void VP8TBufferInit(VP8TBuffer* const b, int page_size) {
   (void)b;
+  (void)page_size;
 }
 void VP8TBufferClear(VP8TBuffer* const b) {
   (void)b;
diff --git a/src/3rdparty/libwebp/src/enc/tree_enc.c b/src/3rdparty/libwebp/src/enc/tree_enc.c
index 2c40fe7..64ed283 100644
--- a/src/3rdparty/libwebp/src/enc/tree_enc.c
+++ b/src/3rdparty/libwebp/src/enc/tree_enc.c
@@ -11,7 +11,7 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./vp8i_enc.h"
+#include "src/enc/vp8i_enc.h"
 
 //------------------------------------------------------------------------------
 // Default probabilities
diff --git a/src/3rdparty/libwebp/src/enc/vp8i_enc.h b/src/3rdparty/libwebp/src/enc/vp8i_enc.h
index 93c95ec..3463491 100644
--- a/src/3rdparty/libwebp/src/enc/vp8i_enc.h
+++ b/src/3rdparty/libwebp/src/enc/vp8i_enc.h
@@ -11,16 +11,16 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_ENC_VP8ENCI_H_
-#define WEBP_ENC_VP8ENCI_H_
+#ifndef WEBP_ENC_VP8I_ENC_H_
+#define WEBP_ENC_VP8I_ENC_H_
 
 #include <string.h>     // for memcpy()
-#include "../dec/common_dec.h"
-#include "../dsp/dsp.h"
-#include "../utils/bit_writer_utils.h"
-#include "../utils/thread_utils.h"
-#include "../utils/utils.h"
-#include "../webp/encode.h"
+#include "src/dec/common_dec.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/bit_writer_utils.h"
+#include "src/utils/thread_utils.h"
+#include "src/utils/utils.h"
+#include "src/webp/encode.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -32,7 +32,7 @@ extern "C" {
 // version numbers
 #define ENC_MAJ_VERSION 0
 #define ENC_MIN_VERSION 6
-#define ENC_REV_VERSION 0
+#define ENC_REV_VERSION 1
 
 enum { MAX_LF_LEVELS = 64,       // Maximum loop filter level
        MAX_VARIABLE_LEVEL = 67,  // last (inclusive) level with variable cost
@@ -75,10 +75,10 @@ typedef enum {   // Rate-distortion optimization levels
 #define U_OFF_ENC    (16)
 #define V_OFF_ENC    (16 + 8)
 
-extern const int VP8Scan[16];           // in quant.c
-extern const int VP8UVModeOffsets[4];   // in analyze.c
-extern const int VP8I16ModeOffsets[4];
-extern const int VP8I4ModeOffsets[NUM_BMODES];
+extern const uint16_t VP8Scan[16];
+extern const uint16_t VP8UVModeOffsets[4];
+extern const uint16_t VP8I16ModeOffsets[4];
+extern const uint16_t VP8I4ModeOffsets[NUM_BMODES];
 
 // Layout of prediction blocks
 // intra 16x16
@@ -330,9 +330,6 @@ int VP8RecordCoeffTokens(int ctx, const struct VP8Residual* const res,
 // Estimate the final coded size given a set of 'probas'.
 size_t VP8EstimateTokenSize(VP8TBuffer* const b, const uint8_t* const probas);
 
-// unused for now
-void VP8TokenToStats(const VP8TBuffer* const b, proba_t* const stats);
-
 #endif  // !DISABLE_TOKEN_BUFFER
 
 //------------------------------------------------------------------------------
@@ -502,19 +499,10 @@ int WebPPictureAllocYUVA(WebPPicture* const picture, int width, int height);
 // compressibility (no guarantee, though). Assumes that pic->use_argb is true.
 void WebPCleanupTransparentAreaLossless(WebPPicture* const pic);
 
-  // in near_lossless.c
-// Near lossless preprocessing in RGB color-space.
-int VP8ApplyNearLossless(int xsize, int ysize, uint32_t* argb, int quality);
-// Near lossless adjustment for predictors.
-void VP8ApplyNearLosslessPredict(int xsize, int ysize, int pred_bits,
-                                 const uint32_t* argb_orig,
-                                 uint32_t* argb, uint32_t* argb_scratch,
-                                 const uint32_t* const transform_data,
-                                 int quality, int subtract_green);
 //------------------------------------------------------------------------------
 
 #ifdef __cplusplus
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_ENC_VP8ENCI_H_ */
+#endif  /* WEBP_ENC_VP8I_ENC_H_ */
diff --git a/src/3rdparty/libwebp/src/enc/vp8l_enc.c b/src/3rdparty/libwebp/src/enc/vp8l_enc.c
index b1a793d..312e521 100644
--- a/src/3rdparty/libwebp/src/enc/vp8l_enc.c
+++ b/src/3rdparty/libwebp/src/enc/vp8l_enc.c
@@ -15,20 +15,19 @@
 #include <assert.h>
 #include <stdlib.h>
 
-#include "./backward_references_enc.h"
-#include "./histogram_enc.h"
-#include "./vp8i_enc.h"
-#include "./vp8li_enc.h"
-#include "../dsp/lossless.h"
-#include "../dsp/lossless_common.h"
-#include "../utils/bit_writer_utils.h"
-#include "../utils/huffman_encode_utils.h"
-#include "../utils/utils.h"
-#include "../webp/format_constants.h"
-
-#include "./delta_palettization_enc.h"
-
-#define PALETTE_KEY_RIGHT_SHIFT   22  // Key for 1K buffer.
+#include "src/enc/backward_references_enc.h"
+#include "src/enc/histogram_enc.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/enc/vp8li_enc.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+#include "src/utils/bit_writer_utils.h"
+#include "src/utils/huffman_encode_utils.h"
+#include "src/utils/utils.h"
+#include "src/webp/format_constants.h"
+
+#include "src/enc/delta_palettization_enc.h"
+
 // Maximum number of histogram images (sub-blocks).
 #define MAX_HUFF_IMAGE_SIZE       2600
 
@@ -128,7 +127,10 @@ static int AnalyzeAndCreatePalette(const WebPPicture* const pic,
                                    uint32_t palette[MAX_PALETTE_SIZE],
                                    int* const palette_size) {
   const int num_colors = WebPGetColorPalette(pic, palette);
-  if (num_colors > MAX_PALETTE_SIZE) return 0;
+  if (num_colors > MAX_PALETTE_SIZE) {
+    *palette_size = 0;
+    return 0;
+  }
   *palette_size = num_colors;
   qsort(palette, num_colors, sizeof(*palette), PaletteCompareColorsForQsort);
   if (!low_effort && PaletteHasNonMonotonousDeltas(palette, num_colors)) {
@@ -188,22 +190,33 @@ static WEBP_INLINE uint32_t HashPix(uint32_t pix) {
 static int AnalyzeEntropy(const uint32_t* argb,
                           int width, int height, int argb_stride,
                           int use_palette,
+                          int palette_size, int transform_bits,
                           EntropyIx* const min_entropy_ix,
                           int* const red_and_blue_always_zero) {
   // Allocate histogram set with cache_bits = 0.
-  uint32_t* const histo =
-      (uint32_t*)WebPSafeCalloc(kHistoTotal, sizeof(*histo) * 256);
+  uint32_t* histo;
+
+  if (use_palette && palette_size <= 16) {
+    // In the case of small palettes, we pack 2, 4 or 8 pixels together. In
+    // practice, small palettes are better than any other transform.
+    *min_entropy_ix = kPalette;
+    *red_and_blue_always_zero = 1;
+    return 1;
+  }
+  histo = (uint32_t*)WebPSafeCalloc(kHistoTotal, sizeof(*histo) * 256);
   if (histo != NULL) {
     int i, x, y;
-    const uint32_t* prev_row = argb;
-    const uint32_t* curr_row = argb + argb_stride;
-    for (y = 1; y < height; ++y) {
-      uint32_t prev_pix = curr_row[0];
-      for (x = 1; x < width; ++x) {
+    const uint32_t* prev_row = NULL;
+    const uint32_t* curr_row = argb;
+    uint32_t pix_prev = argb[0];  // Skip the first pixel.
+    for (y = 0; y < height; ++y) {
+      for (x = 0; x < width; ++x) {
         const uint32_t pix = curr_row[x];
-        const uint32_t pix_diff = VP8LSubPixels(pix, prev_pix);
-        if ((pix_diff == 0) || (pix == prev_row[x])) continue;
-        prev_pix = pix;
+        const uint32_t pix_diff = VP8LSubPixels(pix, pix_prev);
+        pix_prev = pix;
+        if ((pix_diff == 0) || (prev_row != NULL && pix == prev_row[x])) {
+          continue;
+        }
         AddSingle(pix,
                   &histo[kHistoAlpha * 256],
                   &histo[kHistoRed * 256],
@@ -264,8 +277,24 @@ static int AnalyzeEntropy(const uint32_t* argb,
           entropy_comp[kHistoRedPredSubGreen] +
           entropy_comp[kHistoGreenPred] +
           entropy_comp[kHistoBluePredSubGreen];
-      // Palette mode seems more efficient in a breakeven case. Bias with 1.0.
-      entropy[kPalette] = entropy_comp[kHistoPalette] - 1.0;
+      entropy[kPalette] = entropy_comp[kHistoPalette];
+
+      // When including transforms, there is an overhead in bits from
+      // storing them. This overhead is small but matters for small images.
+      // For spatial, there are 14 transformations.
+      entropy[kSpatial] += VP8LSubSampleSize(width, transform_bits) *
+                           VP8LSubSampleSize(height, transform_bits) *
+                           VP8LFastLog2(14);
+      // For color transforms: 24 as only 3 channels are considered in a
+      // ColorTransformElement.
+      entropy[kSpatialSubGreen] += VP8LSubSampleSize(width, transform_bits) *
+                                   VP8LSubSampleSize(height, transform_bits) *
+                                   VP8LFastLog2(24);
+      // For palettes, add the cost of storing the palette.
+      // We empirically estimate the cost of a compressed entry as 8 bits.
+      // The palette is differential-coded when compressed hence a much
+      // lower cost than sizeof(uint32_t)*8.
+      entropy[kPalette] += palette_size * 8;
 
       *min_entropy_ix = kDirect;
       for (k = kDirect + 1; k <= last_mode_to_analyze; ++k) {
@@ -273,6 +302,7 @@ static int AnalyzeEntropy(const uint32_t* argb,
           *min_entropy_ix = (EntropyIx)k;
         }
       }
+      assert((int)*min_entropy_ix <= last_mode_to_analyze);
       *red_and_blue_always_zero = 1;
       // Let's check if the histogram of the chosen entropy mode has
       // non-zero red and blue values. If all are zero, we can later skip
@@ -325,60 +355,95 @@ static int GetTransformBits(int method, int histo_bits) {
   return res;
 }
 
-static int AnalyzeAndInit(VP8LEncoder* const enc) {
+// Set of parameters to be used in each iteration of the cruncher.
+#define CRUNCH_CONFIGS_LZ77_MAX 2
+typedef struct {
+  int entropy_idx_;
+  int lz77s_types_to_try_[CRUNCH_CONFIGS_LZ77_MAX];
+  int lz77s_types_to_try_size_;
+} CrunchConfig;
+
+#define CRUNCH_CONFIGS_MAX kNumEntropyIx
+
+static int EncoderAnalyze(VP8LEncoder* const enc,
+                          CrunchConfig crunch_configs[CRUNCH_CONFIGS_MAX],
+                          int* const crunch_configs_size,
+                          int* const red_and_blue_always_zero) {
   const WebPPicture* const pic = enc->pic_;
   const int width = pic->width;
   const int height = pic->height;
-  const int pix_cnt = width * height;
   const WebPConfig* const config = enc->config_;
   const int method = config->method;
   const int low_effort = (config->method == 0);
-  // we round the block size up, so we're guaranteed to have
-  // at max MAX_REFS_BLOCK_PER_IMAGE blocks used:
-  int refs_block_size = (pix_cnt - 1) / MAX_REFS_BLOCK_PER_IMAGE + 1;
+  int i;
+  int use_palette;
+  int n_lz77s;
   assert(pic != NULL && pic->argb != NULL);
 
-  enc->use_cross_color_ = 0;
-  enc->use_predict_ = 0;
-  enc->use_subtract_green_ = 0;
-  enc->use_palette_ =
+  use_palette =
       AnalyzeAndCreatePalette(pic, low_effort,
                               enc->palette_, &enc->palette_size_);
 
   // TODO(jyrki): replace the decision to be based on an actual estimate
   // of entropy, or even spatial variance of entropy.
-  enc->histo_bits_ = GetHistoBits(method, enc->use_palette_,
+  enc->histo_bits_ = GetHistoBits(method, use_palette,
                                   pic->width, pic->height);
   enc->transform_bits_ = GetTransformBits(method, enc->histo_bits_);
 
   if (low_effort) {
     // AnalyzeEntropy is somewhat slow.
-    enc->use_predict_ = !enc->use_palette_;
-    enc->use_subtract_green_ = !enc->use_palette_;
-    enc->use_cross_color_ = 0;
+    crunch_configs[0].entropy_idx_ = use_palette ? kPalette : kSpatialSubGreen;
+    n_lz77s = 1;
+    *crunch_configs_size = 1;
   } else {
-    int red_and_blue_always_zero;
     EntropyIx min_entropy_ix;
-    if (!AnalyzeEntropy(pic->argb, width, height, pic->argb_stride,
-                        enc->use_palette_, &min_entropy_ix,
-                        &red_and_blue_always_zero)) {
+    // Try out multiple LZ77 on images with few colors.
+    n_lz77s = (enc->palette_size_ > 0 && enc->palette_size_ <= 16) ? 2 : 1;
+    if (!AnalyzeEntropy(pic->argb, width, height, pic->argb_stride, use_palette,
+                        enc->palette_size_, enc->transform_bits_,
+                        &min_entropy_ix, red_and_blue_always_zero)) {
       return 0;
     }
-    enc->use_palette_ = (min_entropy_ix == kPalette);
-    enc->use_subtract_green_ =
-        (min_entropy_ix == kSubGreen) || (min_entropy_ix == kSpatialSubGreen);
-    enc->use_predict_ =
-        (min_entropy_ix == kSpatial) || (min_entropy_ix == kSpatialSubGreen);
-    enc->use_cross_color_ = red_and_blue_always_zero ? 0 : enc->use_predict_;
+    if (method == 6 && config->quality == 100) {
+      // Go brute force on all transforms.
+      *crunch_configs_size = 0;
+      for (i = 0; i < kNumEntropyIx; ++i) {
+        if (i != kPalette || use_palette) {
+          assert(*crunch_configs_size < CRUNCH_CONFIGS_MAX);
+          crunch_configs[(*crunch_configs_size)++].entropy_idx_ = i;
+        }
+      }
+    } else {
+      // Only choose the guessed best transform.
+      *crunch_configs_size = 1;
+      crunch_configs[0].entropy_idx_ = min_entropy_ix;
+    }
+  }
+  // Fill in the different LZ77s.
+  assert(n_lz77s <= CRUNCH_CONFIGS_LZ77_MAX);
+  for (i = 0; i < *crunch_configs_size; ++i) {
+    int j;
+    for (j = 0; j < n_lz77s; ++j) {
+      crunch_configs[i].lz77s_types_to_try_[j] =
+          (j == 0) ? kLZ77Standard | kLZ77RLE : kLZ77Box;
+    }
+    crunch_configs[i].lz77s_types_to_try_size_ = n_lz77s;
   }
+  return 1;
+}
 
+static int EncoderInit(VP8LEncoder* const enc) {
+  const WebPPicture* const pic = enc->pic_;
+  const int width = pic->width;
+  const int height = pic->height;
+  const int pix_cnt = width * height;
+  // we round the block size up, so we're guaranteed to have
+  // at most MAX_REFS_BLOCK_PER_IMAGE blocks used:
+  const int refs_block_size = (pix_cnt - 1) / MAX_REFS_BLOCK_PER_IMAGE + 1;
+  int i;
   if (!VP8LHashChainInit(&enc->hash_chain_, pix_cnt)) return 0;
 
-  // palette-friendly input typically uses less literals
-  //  -> reduce block size a bit
-  if (enc->use_palette_) refs_block_size /= 2;
-  VP8LBackwardRefsInit(&enc->refs_[0], refs_block_size);
-  VP8LBackwardRefsInit(&enc->refs_[1], refs_block_size);
+  for (i = 0; i < 3; ++i) VP8LBackwardRefsInit(&enc->refs_[i], refs_block_size);
 
   return 1;
 }
@@ -571,11 +636,16 @@ static void StoreFullHuffmanCode(VP8LBitWriter* const bw,
     length = write_trimmed_length ? trimmed_length : num_tokens;
     VP8LPutBits(bw, write_trimmed_length, 1);
     if (write_trimmed_length) {
-      const int nbits = VP8LBitsLog2Ceiling(trimmed_length - 1);
-      const int nbitpairs = (nbits == 0) ? 1 : (nbits + 1) / 2;
-      VP8LPutBits(bw, nbitpairs - 1, 3);
-      assert(trimmed_length >= 2);
-      VP8LPutBits(bw, trimmed_length - 2, nbitpairs * 2);
+      if (trimmed_length == 2) {
+        VP8LPutBits(bw, 0, 3 + 2);     // nbitpairs=1, trimmed_length=2
+      } else {
+        const int nbits = BitsLog2Floor(trimmed_length - 2);
+        const int nbitpairs = nbits / 2 + 1;
+        assert(trimmed_length > 2);
+        assert(nbitpairs - 1 < 8);
+        VP8LPutBits(bw, nbitpairs - 1, 3);
+        VP8LPutBits(bw, trimmed_length - 2, nbitpairs * 2);
+      }
     }
     StoreHuffmanTreeToBitMask(bw, tokens, length, &huffman_code);
   }
@@ -642,7 +712,7 @@ static WEBP_INLINE void WriteHuffmanCodeWithExtraBits(
 
 static WebPEncodingError StoreImageToBitMask(
     VP8LBitWriter* const bw, int width, int histo_bits,
-    VP8LBackwardRefs* const refs,
+    const VP8LBackwardRefs* const refs,
     const uint16_t* histogram_symbols,
     const HuffmanTreeCode* const huffman_codes) {
   const int histo_xsize = histo_bits ? VP8LSubSampleSize(width, histo_bits) : 1;
@@ -665,7 +735,7 @@ static WebPEncodingError StoreImageToBitMask(
       codes = huffman_codes + 5 * histogram_ix;
     }
     if (PixOrCopyIsLiteral(v)) {
-      static const int order[] = { 1, 2, 0, 3 };
+      static const uint8_t order[] = { 1, 2, 0, 3 };
       int k;
       for (k = 0; k < 4; ++k) {
         const int code = PixOrCopyLiteral(v, order[k]);
@@ -705,7 +775,8 @@ static WebPEncodingError StoreImageToBitMask(
 static WebPEncodingError EncodeImageNoHuffman(VP8LBitWriter* const bw,
                                               const uint32_t* const argb,
                                               VP8LHashChain* const hash_chain,
-                                              VP8LBackwardRefs refs_array[2],
+                                              VP8LBackwardRefs* const refs_tmp1,
+                                              VP8LBackwardRefs* const refs_tmp2,
                                               int width, int height,
                                               int quality, int low_effort) {
   int i;
@@ -730,8 +801,9 @@ static WebPEncodingError EncodeImageNoHuffman(VP8LBitWriter* const bw,
     err = VP8_ENC_ERROR_OUT_OF_MEMORY;
     goto Error;
   }
-  refs = VP8LGetBackwardReferences(width, height, argb, quality, 0, &cache_bits,
-                                   hash_chain, refs_array);
+  refs = VP8LGetBackwardReferences(width, height, argb, quality, 0,
+                                   kLZ77Standard | kLZ77RLE, &cache_bits,
+                                   hash_chain, refs_tmp1, refs_tmp2);
   if (refs == NULL) {
     err = VP8_ENC_ERROR_OUT_OF_MEMORY;
     goto Error;
@@ -788,39 +860,37 @@ static WebPEncodingError EncodeImageNoHuffman(VP8LBitWriter* const bw,
   return err;
 }
 
-static WebPEncodingError EncodeImageInternal(VP8LBitWriter* const bw,
-                                             const uint32_t* const argb,
-                                             VP8LHashChain* const hash_chain,
-                                             VP8LBackwardRefs refs_array[2],
-                                             int width, int height, int quality,
-                                             int low_effort,
-                                             int use_cache, int* cache_bits,
-                                             int histogram_bits,
-                                             size_t init_byte_position,
-                                             int* const hdr_size,
-                                             int* const data_size) {
+static WebPEncodingError EncodeImageInternal(
+    VP8LBitWriter* const bw, const uint32_t* const argb,
+    VP8LHashChain* const hash_chain, VP8LBackwardRefs refs_array[3], int width,
+    int height, int quality, int low_effort, int use_cache,
+    const CrunchConfig* const config, int* cache_bits, int histogram_bits,
+    size_t init_byte_position, int* const hdr_size, int* const data_size) {
   WebPEncodingError err = VP8_ENC_OK;
   const uint32_t histogram_image_xysize =
       VP8LSubSampleSize(width, histogram_bits) *
       VP8LSubSampleSize(height, histogram_bits);
   VP8LHistogramSet* histogram_image = NULL;
-  VP8LHistogramSet* tmp_histos = NULL;
+  VP8LHistogram* tmp_histo = NULL;
   int histogram_image_size = 0;
   size_t bit_array_size = 0;
-  HuffmanTree* huff_tree = NULL;
+  HuffmanTree* const huff_tree = (HuffmanTree*)WebPSafeMalloc(
+      3ULL * CODE_LENGTH_CODES, sizeof(*huff_tree));
   HuffmanTreeToken* tokens = NULL;
   HuffmanTreeCode* huffman_codes = NULL;
-  VP8LBackwardRefs refs;
-  VP8LBackwardRefs* best_refs;
+  VP8LBackwardRefs* refs_best;
+  VP8LBackwardRefs* refs_tmp;
   uint16_t* const histogram_symbols =
       (uint16_t*)WebPSafeMalloc(histogram_image_xysize,
                                 sizeof(*histogram_symbols));
+  int lz77s_idx;
+  VP8LBitWriter bw_init = *bw, bw_best;
+  int hdr_size_tmp;
   assert(histogram_bits >= MIN_HUFFMAN_BITS);
   assert(histogram_bits <= MAX_HUFFMAN_BITS);
   assert(hdr_size != NULL);
   assert(data_size != NULL);
 
-  VP8LBackwardRefsInit(&refs, refs_array[0].block_size_);
   if (histogram_symbols == NULL) {
     err = VP8_ENC_ERROR_OUT_OF_MEMORY;
     goto Error;
@@ -836,142 +906,162 @@ static WebPEncodingError EncodeImageInternal(VP8LBitWriter* const bw,
   // 'best_refs' is the reference to the best backward refs and points to one
   // of refs_array[0] or refs_array[1].
   // Calculate backward references from ARGB image.
-  if (!VP8LHashChainFill(hash_chain, quality, argb, width, height,
-                         low_effort)) {
-    err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-    goto Error;
-  }
-  best_refs = VP8LGetBackwardReferences(width, height, argb, quality,
-                                        low_effort, cache_bits, hash_chain,
-                                        refs_array);
-  if (best_refs == NULL || !VP8LBackwardRefsCopy(best_refs, &refs)) {
-    err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-    goto Error;
-  }
-  histogram_image =
-      VP8LAllocateHistogramSet(histogram_image_xysize, *cache_bits);
-  tmp_histos = VP8LAllocateHistogramSet(2, *cache_bits);
-  if (histogram_image == NULL || tmp_histos == NULL) {
+  if (huff_tree == NULL ||
+      !VP8LHashChainFill(hash_chain, quality, argb, width, height,
+                         low_effort) ||
+      !VP8LBitWriterInit(&bw_best, 0) ||
+      (config->lz77s_types_to_try_size_ > 1 &&
+       !VP8LBitWriterClone(bw, &bw_best))) {
     err = VP8_ENC_ERROR_OUT_OF_MEMORY;
     goto Error;
   }
+  for (lz77s_idx = 0; lz77s_idx < config->lz77s_types_to_try_size_;
+       ++lz77s_idx) {
+    refs_best = VP8LGetBackwardReferences(
+        width, height, argb, quality, low_effort,
+        config->lz77s_types_to_try_[lz77s_idx], cache_bits, hash_chain,
+        &refs_array[0], &refs_array[1]);
+    if (refs_best == NULL) {
+      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+      goto Error;
+    }
+    // Keep the best references aside and use the other element from the first
+    // two as a temporary for later usage.
+    refs_tmp = &refs_array[refs_best == &refs_array[0] ? 1 : 0];
+
+    histogram_image =
+        VP8LAllocateHistogramSet(histogram_image_xysize, *cache_bits);
+    tmp_histo = VP8LAllocateHistogram(*cache_bits);
+    if (histogram_image == NULL || tmp_histo == NULL) {
+      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+      goto Error;
+    }
 
-  // Build histogram image and symbols from backward references.
-  if (!VP8LGetHistoImageSymbols(width, height, &refs, quality, low_effort,
-                                histogram_bits, *cache_bits, histogram_image,
-                                tmp_histos, histogram_symbols)) {
-    err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-    goto Error;
-  }
-  // Create Huffman bit lengths and codes for each histogram image.
-  histogram_image_size = histogram_image->size;
-  bit_array_size = 5 * histogram_image_size;
-  huffman_codes = (HuffmanTreeCode*)WebPSafeCalloc(bit_array_size,
-                                                   sizeof(*huffman_codes));
-  // Note: some histogram_image entries may point to tmp_histos[], so the latter
-  // need to outlive the following call to GetHuffBitLengthsAndCodes().
-  if (huffman_codes == NULL ||
-      !GetHuffBitLengthsAndCodes(histogram_image, huffman_codes)) {
-    err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-    goto Error;
-  }
-  // Free combined histograms.
-  VP8LFreeHistogramSet(histogram_image);
-  histogram_image = NULL;
+    // Build histogram image and symbols from backward references.
+    if (!VP8LGetHistoImageSymbols(width, height, refs_best, quality, low_effort,
+                                  histogram_bits, *cache_bits, histogram_image,
+                                  tmp_histo, histogram_symbols)) {
+      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+      goto Error;
+    }
+    // Create Huffman bit lengths and codes for each histogram image.
+    histogram_image_size = histogram_image->size;
+    bit_array_size = 5 * histogram_image_size;
+    huffman_codes = (HuffmanTreeCode*)WebPSafeCalloc(bit_array_size,
+                                                     sizeof(*huffman_codes));
+    // Note: some histogram_image entries may point to tmp_histos[], so the
+    // latter need to outlive the following call to GetHuffBitLengthsAndCodes().
+    if (huffman_codes == NULL ||
+        !GetHuffBitLengthsAndCodes(histogram_image, huffman_codes)) {
+      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+      goto Error;
+    }
+    // Free combined histograms.
+    VP8LFreeHistogramSet(histogram_image);
+    histogram_image = NULL;
 
-  // Free scratch histograms.
-  VP8LFreeHistogramSet(tmp_histos);
-  tmp_histos = NULL;
+    // Free scratch histograms.
+    VP8LFreeHistogram(tmp_histo);
+    tmp_histo = NULL;
 
-  // Color Cache parameters.
-  if (*cache_bits > 0) {
-    VP8LPutBits(bw, 1, 1);
-    VP8LPutBits(bw, *cache_bits, 4);
-  } else {
-    VP8LPutBits(bw, 0, 1);
-  }
+    // Color Cache parameters.
+    if (*cache_bits > 0) {
+      VP8LPutBits(bw, 1, 1);
+      VP8LPutBits(bw, *cache_bits, 4);
+    } else {
+      VP8LPutBits(bw, 0, 1);
+    }
 
-  // Huffman image + meta huffman.
-  {
-    const int write_histogram_image = (histogram_image_size > 1);
-    VP8LPutBits(bw, write_histogram_image, 1);
-    if (write_histogram_image) {
-      uint32_t* const histogram_argb =
-          (uint32_t*)WebPSafeMalloc(histogram_image_xysize,
-                                    sizeof(*histogram_argb));
-      int max_index = 0;
-      uint32_t i;
-      if (histogram_argb == NULL) {
-        err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-        goto Error;
-      }
-      for (i = 0; i < histogram_image_xysize; ++i) {
-        const int symbol_index = histogram_symbols[i] & 0xffff;
-        histogram_argb[i] = (symbol_index << 8);
-        if (symbol_index >= max_index) {
-          max_index = symbol_index + 1;
+    // Huffman image + meta huffman.
+    {
+      const int write_histogram_image = (histogram_image_size > 1);
+      VP8LPutBits(bw, write_histogram_image, 1);
+      if (write_histogram_image) {
+        uint32_t* const histogram_argb =
+            (uint32_t*)WebPSafeMalloc(histogram_image_xysize,
+                                      sizeof(*histogram_argb));
+        int max_index = 0;
+        uint32_t i;
+        if (histogram_argb == NULL) {
+          err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+          goto Error;
+        }
+        for (i = 0; i < histogram_image_xysize; ++i) {
+          const int symbol_index = histogram_symbols[i] & 0xffff;
+          histogram_argb[i] = (symbol_index << 8);
+          if (symbol_index >= max_index) {
+            max_index = symbol_index + 1;
+          }
         }
+        histogram_image_size = max_index;
+
+        VP8LPutBits(bw, histogram_bits - 2, 3);
+        err = EncodeImageNoHuffman(
+            bw, histogram_argb, hash_chain, refs_tmp, &refs_array[2],
+            VP8LSubSampleSize(width, histogram_bits),
+            VP8LSubSampleSize(height, histogram_bits), quality, low_effort);
+        WebPSafeFree(histogram_argb);
+        if (err != VP8_ENC_OK) goto Error;
       }
-      histogram_image_size = max_index;
-
-      VP8LPutBits(bw, histogram_bits - 2, 3);
-      err = EncodeImageNoHuffman(bw, histogram_argb, hash_chain, refs_array,
-                                 VP8LSubSampleSize(width, histogram_bits),
-                                 VP8LSubSampleSize(height, histogram_bits),
-                                 quality, low_effort);
-      WebPSafeFree(histogram_argb);
-      if (err != VP8_ENC_OK) goto Error;
     }
-  }
 
-  // Store Huffman codes.
-  {
-    int i;
-    int max_tokens = 0;
-    huff_tree = (HuffmanTree*)WebPSafeMalloc(3ULL * CODE_LENGTH_CODES,
-                                             sizeof(*huff_tree));
-    if (huff_tree == NULL) {
-      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-      goto Error;
-    }
-    // Find maximum number of symbols for the huffman tree-set.
-    for (i = 0; i < 5 * histogram_image_size; ++i) {
-      HuffmanTreeCode* const codes = &huffman_codes[i];
-      if (max_tokens < codes->num_symbols) {
-        max_tokens = codes->num_symbols;
+    // Store Huffman codes.
+    {
+      int i;
+      int max_tokens = 0;
+      // Find maximum number of symbols for the huffman tree-set.
+      for (i = 0; i < 5 * histogram_image_size; ++i) {
+        HuffmanTreeCode* const codes = &huffman_codes[i];
+        if (max_tokens < codes->num_symbols) {
+          max_tokens = codes->num_symbols;
+        }
+      }
+      tokens = (HuffmanTreeToken*)WebPSafeMalloc(max_tokens, sizeof(*tokens));
+      if (tokens == NULL) {
+        err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+        goto Error;
+      }
+      for (i = 0; i < 5 * histogram_image_size; ++i) {
+        HuffmanTreeCode* const codes = &huffman_codes[i];
+        StoreHuffmanCode(bw, huff_tree, tokens, codes);
+        ClearHuffmanTreeIfOnlyOneSymbol(codes);
       }
     }
-    tokens = (HuffmanTreeToken*)WebPSafeMalloc(max_tokens,
-                                               sizeof(*tokens));
-    if (tokens == NULL) {
-      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-      goto Error;
+    // Store actual literals.
+    hdr_size_tmp = (int)(VP8LBitWriterNumBytes(bw) - init_byte_position);
+    err = StoreImageToBitMask(bw, width, histogram_bits, refs_best,
+                              histogram_symbols, huffman_codes);
+    // Keep track of the smallest image so far.
+    if (lz77s_idx == 0 ||
+        VP8LBitWriterNumBytes(bw) < VP8LBitWriterNumBytes(&bw_best)) {
+      *hdr_size = hdr_size_tmp;
+      *data_size =
+          (int)(VP8LBitWriterNumBytes(bw) - init_byte_position - *hdr_size);
+      VP8LBitWriterSwap(bw, &bw_best);
     }
-    for (i = 0; i < 5 * histogram_image_size; ++i) {
-      HuffmanTreeCode* const codes = &huffman_codes[i];
-      StoreHuffmanCode(bw, huff_tree, tokens, codes);
-      ClearHuffmanTreeIfOnlyOneSymbol(codes);
+    // Reset the bit writer for the following iteration if any.
+    if (config->lz77s_types_to_try_size_ > 1) VP8LBitWriterReset(&bw_init, bw);
+    WebPSafeFree(tokens);
+    tokens = NULL;
+    if (huffman_codes != NULL) {
+      WebPSafeFree(huffman_codes->codes);
+      WebPSafeFree(huffman_codes);
+      huffman_codes = NULL;
     }
   }
-
-  *hdr_size = (int)(VP8LBitWriterNumBytes(bw) - init_byte_position);
-  // Store actual literals.
-  err = StoreImageToBitMask(bw, width, histogram_bits, &refs,
-                            histogram_symbols, huffman_codes);
-  *data_size =
-        (int)(VP8LBitWriterNumBytes(bw) - init_byte_position - *hdr_size);
+  VP8LBitWriterSwap(bw, &bw_best);
 
  Error:
   WebPSafeFree(tokens);
   WebPSafeFree(huff_tree);
   VP8LFreeHistogramSet(histogram_image);
-  VP8LFreeHistogramSet(tmp_histos);
-  VP8LBackwardRefsClear(&refs);
+  VP8LFreeHistogram(tmp_histo);
   if (huffman_codes != NULL) {
     WebPSafeFree(huffman_codes->codes);
     WebPSafeFree(huffman_codes);
   }
   WebPSafeFree(histogram_symbols);
+  VP8LBitWriterWipeOut(&bw_best);
   return err;
 }
 
@@ -1005,11 +1095,11 @@ static WebPEncodingError ApplyPredictFilter(const VP8LEncoder* const enc,
   VP8LPutBits(bw, PREDICTOR_TRANSFORM, 2);
   assert(pred_bits >= 2);
   VP8LPutBits(bw, pred_bits - 2, 3);
-  return EncodeImageNoHuffman(bw, enc->transform_data_,
-                              (VP8LHashChain*)&enc->hash_chain_,
-                              (VP8LBackwardRefs*)enc->refs_,  // cast const away
-                              transform_width, transform_height,
-                              quality, low_effort);
+  return EncodeImageNoHuffman(
+      bw, enc->transform_data_, (VP8LHashChain*)&enc->hash_chain_,
+      (VP8LBackwardRefs*)&enc->refs_[0],  // cast const away
+      (VP8LBackwardRefs*)&enc->refs_[1], transform_width, transform_height,
+      quality, low_effort);
 }
 
 static WebPEncodingError ApplyCrossColorFilter(const VP8LEncoder* const enc,
@@ -1026,11 +1116,11 @@ static WebPEncodingError ApplyCrossColorFilter(const VP8LEncoder* const enc,
   VP8LPutBits(bw, CROSS_COLOR_TRANSFORM, 2);
   assert(ccolor_transform_bits >= 2);
   VP8LPutBits(bw, ccolor_transform_bits - 2, 3);
-  return EncodeImageNoHuffman(bw, enc->transform_data_,
-                              (VP8LHashChain*)&enc->hash_chain_,
-                              (VP8LBackwardRefs*)enc->refs_,  // cast const away
-                              transform_width, transform_height,
-                              quality, low_effort);
+  return EncodeImageNoHuffman(
+      bw, enc->transform_data_, (VP8LHashChain*)&enc->hash_chain_,
+      (VP8LBackwardRefs*)&enc->refs_[0],  // cast const away
+      (VP8LBackwardRefs*)&enc->refs_[1], transform_width, transform_height,
+      quality, low_effort);
 }
 
 // -----------------------------------------------------------------------------
@@ -1144,6 +1234,7 @@ static WebPEncodingError AllocateTransformBuffer(VP8LEncoder* const enc,
     }
     enc->transform_mem_ = mem;
     enc->transform_mem_size_ = (size_t)mem_size;
+    enc->argb_content_ = kEncoderNone;
   }
   enc->argb_ = mem;
   mem = (uint32_t*)WEBP_ALIGN(mem + image_size);
@@ -1164,11 +1255,13 @@ static WebPEncodingError MakeInputImageCopy(VP8LEncoder* const enc) {
   int y;
   err = AllocateTransformBuffer(enc, width, height);
   if (err != VP8_ENC_OK) return err;
+  if (enc->argb_content_ == kEncoderARGB) return VP8_ENC_OK;
   for (y = 0; y < height; ++y) {
     memcpy(enc->argb_ + y * width,
            picture->argb + y * picture->argb_stride,
            width * sizeof(*enc->argb_));
   }
+  enc->argb_content_ = kEncoderARGB;
   assert(enc->current_width_ == width);
   return VP8_ENC_OK;
 }
@@ -1215,12 +1308,13 @@ static WEBP_INLINE uint32_t ApplyPaletteHash0(uint32_t color) {
 
 static WEBP_INLINE uint32_t ApplyPaletteHash1(uint32_t color) {
   // Forget about alpha.
-  return ((color & 0x00ffffffu) * 4222244071u) >> (32 - PALETTE_INV_SIZE_BITS);
+  return ((uint32_t)((color & 0x00ffffffu) * 4222244071ull)) >>
+         (32 - PALETTE_INV_SIZE_BITS);
 }
 
 static WEBP_INLINE uint32_t ApplyPaletteHash2(uint32_t color) {
   // Forget about alpha.
-  return (color & 0x00ffffffu) * ((1u << 31) - 1) >>
+  return ((uint32_t)((color & 0x00ffffffu) * ((1ull << 31) - 1))) >>
          (32 - PALETTE_INV_SIZE_BITS);
 }
 
@@ -1346,6 +1440,7 @@ static WebPEncodingError MapImageFromPalette(VP8LEncoder* const enc,
   err = ApplyPalette(src, src_stride,
                      enc->argb_, enc->current_width_,
                      palette, palette_size, width, height, xbits);
+  enc->argb_content_ = kEncoderPalette;
   return err;
 }
 
@@ -1364,8 +1459,9 @@ static WebPEncodingError EncodePalette(VP8LBitWriter* const bw, int low_effort,
     tmp_palette[i] = VP8LSubPixels(palette[i], palette[i - 1]);
   }
   tmp_palette[0] = palette[0];
-  return EncodeImageNoHuffman(bw, tmp_palette, &enc->hash_chain_, enc->refs_,
-                              palette_size, 1, 20 /* quality */, low_effort);
+  return EncodeImageNoHuffman(bw, tmp_palette, &enc->hash_chain_,
+                              &enc->refs_[0], &enc->refs_[1], palette_size, 1,
+                              20 /* quality */, low_effort);
 }
 
 #ifdef WEBP_EXPERIMENTAL_FEATURES
@@ -1400,10 +1496,11 @@ static WebPEncodingError EncodeDeltaPalettePredictorImage(
   VP8LPutBits(bw, TRANSFORM_PRESENT, 1);
   VP8LPutBits(bw, PREDICTOR_TRANSFORM, 2);
   VP8LPutBits(bw, pred_bits - 2, 3);
-  err = EncodeImageNoHuffman(bw, predictors, &enc->hash_chain_,
-                             (VP8LBackwardRefs*)enc->refs_,  // cast const away
-                             transform_width, transform_height,
-                             quality, low_effort);
+  err = EncodeImageNoHuffman(
+      bw, predictors, &enc->hash_chain_,
+      (VP8LBackwardRefs*)&enc->refs_[0],  // cast const away
+      (VP8LBackwardRefs*)&enc->refs_[1],
+      transform_width, transform_height, quality, low_effort);
   WebPSafeFree(predictors);
   return err;
 }
@@ -1422,6 +1519,7 @@ static VP8LEncoder* VP8LEncoderNew(const WebPConfig* const config,
   }
   enc->config_ = config;
   enc->pic_ = picture;
+  enc->argb_content_ = kEncoderNone;
 
   VP8LEncDspInit();
 
@@ -1430,9 +1528,9 @@ static VP8LEncoder* VP8LEncoderNew(const WebPConfig* const config,
 
 static void VP8LEncoderDelete(VP8LEncoder* enc) {
   if (enc != NULL) {
+    int i;
     VP8LHashChainClear(&enc->hash_chain_);
-    VP8LBackwardRefsClear(&enc->refs_[0]);
-    VP8LBackwardRefsClear(&enc->refs_[1]);
+    for (i = 0; i < 3; ++i) VP8LBackwardRefsClear(&enc->refs_[i]);
     ClearTransformBuffer(enc);
     WebPSafeFree(enc);
   }
@@ -1441,134 +1539,347 @@ static void VP8LEncoderDelete(VP8LEncoder* enc) {
 // -----------------------------------------------------------------------------
 // Main call
 
-WebPEncodingError VP8LEncodeStream(const WebPConfig* const config,
-                                   const WebPPicture* const picture,
-                                   VP8LBitWriter* const bw, int use_cache) {
+typedef struct {
+  const WebPConfig* config_;
+  const WebPPicture* picture_;
+  VP8LBitWriter* bw_;
+  VP8LEncoder* enc_;
+  int use_cache_;
+  CrunchConfig crunch_configs_[CRUNCH_CONFIGS_MAX];
+  int num_crunch_configs_;
+  int red_and_blue_always_zero_;
+  WebPEncodingError err_;
+  WebPAuxStats* stats_;
+} StreamEncodeContext;
+
+static int EncodeStreamHook(void* input, void* data2) {
+  StreamEncodeContext* const params = (StreamEncodeContext*)input;
+  const WebPConfig* const config = params->config_;
+  const WebPPicture* const picture = params->picture_;
+  VP8LBitWriter* const bw = params->bw_;
+  VP8LEncoder* const enc = params->enc_;
+  const int use_cache = params->use_cache_;
+  const CrunchConfig* const crunch_configs = params->crunch_configs_;
+  const int num_crunch_configs = params->num_crunch_configs_;
+  const int red_and_blue_always_zero = params->red_and_blue_always_zero_;
+#if !defined(WEBP_DISABLE_STATS)
+  WebPAuxStats* const stats = params->stats_;
+#endif
   WebPEncodingError err = VP8_ENC_OK;
   const int quality = (int)config->quality;
   const int low_effort = (config->method == 0);
+#if (WEBP_NEAR_LOSSLESS == 1) || defined(WEBP_EXPERIMENTAL_FEATURES)
   const int width = picture->width;
+#endif
   const int height = picture->height;
-  VP8LEncoder* const enc = VP8LEncoderNew(config, picture);
   const size_t byte_position = VP8LBitWriterNumBytes(bw);
+#if (WEBP_NEAR_LOSSLESS == 1)
   int use_near_lossless = 0;
+#endif
   int hdr_size = 0;
   int data_size = 0;
   int use_delta_palette = 0;
+  int idx;
+  size_t best_size = 0;
+  VP8LBitWriter bw_init = *bw, bw_best;
+  (void)data2;
 
-  if (enc == NULL) {
-    err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-    goto Error;
-  }
-
-  // ---------------------------------------------------------------------------
-  // Analyze image (entropy, num_palettes etc)
-
-  if (!AnalyzeAndInit(enc)) {
+  if (!VP8LBitWriterInit(&bw_best, 0) ||
+      (num_crunch_configs > 1 && !VP8LBitWriterClone(bw, &bw_best))) {
     err = VP8_ENC_ERROR_OUT_OF_MEMORY;
     goto Error;
   }
 
-  // Apply near-lossless preprocessing.
-  use_near_lossless =
-      (config->near_lossless < 100) && !enc->use_palette_ && !enc->use_predict_;
-  if (use_near_lossless) {
-    if (!VP8ApplyNearLossless(width, height, picture->argb,
-                              config->near_lossless)) {
-      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-      goto Error;
+  for (idx = 0; idx < num_crunch_configs; ++idx) {
+    const int entropy_idx = crunch_configs[idx].entropy_idx_;
+    enc->use_palette_ = (entropy_idx == kPalette);
+    enc->use_subtract_green_ =
+        (entropy_idx == kSubGreen) || (entropy_idx == kSpatialSubGreen);
+    enc->use_predict_ =
+        (entropy_idx == kSpatial) || (entropy_idx == kSpatialSubGreen);
+    if (low_effort) {
+      enc->use_cross_color_ = 0;
+    } else {
+      enc->use_cross_color_ = red_and_blue_always_zero ? 0 : enc->use_predict_;
     }
-  }
+    // Reset any parameter in the encoder that is set in the previous iteration.
+    enc->cache_bits_ = 0;
+    VP8LBackwardRefsClear(&enc->refs_[0]);
+    VP8LBackwardRefsClear(&enc->refs_[1]);
 
-#ifdef WEBP_EXPERIMENTAL_FEATURES
-  if (config->use_delta_palette) {
-    enc->use_predict_ = 1;
-    enc->use_cross_color_ = 0;
-    enc->use_subtract_green_ = 0;
-    enc->use_palette_ = 1;
-    err = MakeInputImageCopy(enc);
-    if (err != VP8_ENC_OK) goto Error;
-    err = WebPSearchOptimalDeltaPalette(enc);
-    if (err != VP8_ENC_OK) goto Error;
-    if (enc->use_palette_) {
+#if (WEBP_NEAR_LOSSLESS == 1)
+    // Apply near-lossless preprocessing.
+    use_near_lossless = (config->near_lossless < 100) && !enc->use_palette_ &&
+                        !enc->use_predict_;
+    if (use_near_lossless) {
       err = AllocateTransformBuffer(enc, width, height);
       if (err != VP8_ENC_OK) goto Error;
-      err = EncodeDeltaPalettePredictorImage(bw, enc, quality, low_effort);
+      if ((enc->argb_content_ != kEncoderNearLossless) &&
+          !VP8ApplyNearLossless(picture, config->near_lossless, enc->argb_)) {
+        err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+        goto Error;
+      }
+      enc->argb_content_ = kEncoderNearLossless;
+    } else {
+      enc->argb_content_ = kEncoderNone;
+    }
+#else
+    enc->argb_content_ = kEncoderNone;
+#endif
+
+#ifdef WEBP_EXPERIMENTAL_FEATURES
+    if (config->use_delta_palette) {
+      enc->use_predict_ = 1;
+      enc->use_cross_color_ = 0;
+      enc->use_subtract_green_ = 0;
+      enc->use_palette_ = 1;
+      if (enc->argb_content_ != kEncoderNearLossless &&
+          enc->argb_content_ != kEncoderPalette) {
+        err = MakeInputImageCopy(enc);
+        if (err != VP8_ENC_OK) goto Error;
+      }
+      err = WebPSearchOptimalDeltaPalette(enc);
       if (err != VP8_ENC_OK) goto Error;
-      use_delta_palette = 1;
+      if (enc->use_palette_) {
+        err = AllocateTransformBuffer(enc, width, height);
+        if (err != VP8_ENC_OK) goto Error;
+        err = EncodeDeltaPalettePredictorImage(bw, enc, quality, low_effort);
+        if (err != VP8_ENC_OK) goto Error;
+        use_delta_palette = 1;
+      }
     }
-  }
 #endif  // WEBP_EXPERIMENTAL_FEATURES
 
-  // Encode palette
-  if (enc->use_palette_) {
-    err = EncodePalette(bw, low_effort, enc);
-    if (err != VP8_ENC_OK) goto Error;
-    err = MapImageFromPalette(enc, use_delta_palette);
-    if (err != VP8_ENC_OK) goto Error;
-    // If using a color cache, do not have it bigger than the number of colors.
-    if (use_cache && enc->palette_size_ < (1 << MAX_COLOR_CACHE_BITS)) {
-      enc->cache_bits_ = BitsLog2Floor(enc->palette_size_) + 1;
-    }
-  }
-  if (!use_delta_palette) {
-    // In case image is not packed.
-    if (enc->argb_ == NULL) {
-      err = MakeInputImageCopy(enc);
+    // Encode palette
+    if (enc->use_palette_) {
+      err = EncodePalette(bw, low_effort, enc);
+      if (err != VP8_ENC_OK) goto Error;
+      err = MapImageFromPalette(enc, use_delta_palette);
       if (err != VP8_ENC_OK) goto Error;
+      // If using a color cache, do not have it bigger than the number of
+      // colors.
+      if (use_cache && enc->palette_size_ < (1 << MAX_COLOR_CACHE_BITS)) {
+        enc->cache_bits_ = BitsLog2Floor(enc->palette_size_) + 1;
+      }
     }
+    if (!use_delta_palette) {
+      // In case image is not packed.
+      if (enc->argb_content_ != kEncoderNearLossless &&
+          enc->argb_content_ != kEncoderPalette) {
+        err = MakeInputImageCopy(enc);
+        if (err != VP8_ENC_OK) goto Error;
+      }
 
-    // -------------------------------------------------------------------------
-    // Apply transforms and write transform data.
+      // -----------------------------------------------------------------------
+      // Apply transforms and write transform data.
 
-    if (enc->use_subtract_green_) {
-      ApplySubtractGreen(enc, enc->current_width_, height, bw);
-    }
+      if (enc->use_subtract_green_) {
+        ApplySubtractGreen(enc, enc->current_width_, height, bw);
+      }
 
-    if (enc->use_predict_) {
-      err = ApplyPredictFilter(enc, enc->current_width_, height, quality,
-                               low_effort, enc->use_subtract_green_, bw);
-      if (err != VP8_ENC_OK) goto Error;
+      if (enc->use_predict_) {
+        err = ApplyPredictFilter(enc, enc->current_width_, height, quality,
+                                 low_effort, enc->use_subtract_green_, bw);
+        if (err != VP8_ENC_OK) goto Error;
+      }
+
+      if (enc->use_cross_color_) {
+        err = ApplyCrossColorFilter(enc, enc->current_width_, height, quality,
+                                    low_effort, bw);
+        if (err != VP8_ENC_OK) goto Error;
+      }
     }
 
-    if (enc->use_cross_color_) {
-      err = ApplyCrossColorFilter(enc, enc->current_width_,
-                                  height, quality, low_effort, bw);
-      if (err != VP8_ENC_OK) goto Error;
+    VP8LPutBits(bw, !TRANSFORM_PRESENT, 1);  // No more transforms.
+
+    // -------------------------------------------------------------------------
+    // Encode and write the transformed image.
+    err = EncodeImageInternal(bw, enc->argb_, &enc->hash_chain_, enc->refs_,
+                              enc->current_width_, height, quality, low_effort,
+                              use_cache, &crunch_configs[idx],
+                              &enc->cache_bits_, enc->histo_bits_,
+                              byte_position, &hdr_size, &data_size);
+    if (err != VP8_ENC_OK) goto Error;
+
+    // If we are better than what we already have.
+    if (idx == 0 || VP8LBitWriterNumBytes(bw) < best_size) {
+      best_size = VP8LBitWriterNumBytes(bw);
+      // Store the BitWriter.
+      VP8LBitWriterSwap(bw, &bw_best);
+#if !defined(WEBP_DISABLE_STATS)
+      // Update the stats.
+      if (stats != NULL) {
+        stats->lossless_features = 0;
+        if (enc->use_predict_) stats->lossless_features |= 1;
+        if (enc->use_cross_color_) stats->lossless_features |= 2;
+        if (enc->use_subtract_green_) stats->lossless_features |= 4;
+        if (enc->use_palette_) stats->lossless_features |= 8;
+        stats->histogram_bits = enc->histo_bits_;
+        stats->transform_bits = enc->transform_bits_;
+        stats->cache_bits = enc->cache_bits_;
+        stats->palette_size = enc->palette_size_;
+        stats->lossless_size = (int)(best_size - byte_position);
+        stats->lossless_hdr_size = hdr_size;
+        stats->lossless_data_size = data_size;
+      }
+#endif
     }
+    // Reset the bit writer for the following iteration if any.
+    if (num_crunch_configs > 1) VP8LBitWriterReset(&bw_init, bw);
   }
+  VP8LBitWriterSwap(&bw_best, bw);
 
-  VP8LPutBits(bw, !TRANSFORM_PRESENT, 1);  // No more transforms.
+Error:
+  VP8LBitWriterWipeOut(&bw_best);
+  params->err_ = err;
+  // The hook should return false in case of error.
+  return (err == VP8_ENC_OK);
+}
 
-  // ---------------------------------------------------------------------------
-  // Encode and write the transformed image.
-  err = EncodeImageInternal(bw, enc->argb_, &enc->hash_chain_, enc->refs_,
-                            enc->current_width_, height, quality, low_effort,
-                            use_cache, &enc->cache_bits_, enc->histo_bits_,
-                            byte_position, &hdr_size, &data_size);
-  if (err != VP8_ENC_OK) goto Error;
+WebPEncodingError VP8LEncodeStream(const WebPConfig* const config,
+                                   const WebPPicture* const picture,
+                                   VP8LBitWriter* const bw_main,
+                                   int use_cache) {
+  WebPEncodingError err = VP8_ENC_OK;
+  VP8LEncoder* const enc_main = VP8LEncoderNew(config, picture);
+  VP8LEncoder* enc_side = NULL;
+  CrunchConfig crunch_configs[CRUNCH_CONFIGS_MAX];
+  int num_crunch_configs_main, num_crunch_configs_side = 0;
+  int idx;
+  int red_and_blue_always_zero = 0;
+  WebPWorker worker_main, worker_side;
+  StreamEncodeContext params_main, params_side;
+  // The main thread uses picture->stats, the side thread uses stats_side.
+  WebPAuxStats stats_side;
+  VP8LBitWriter bw_side;
+  const WebPWorkerInterface* const worker_interface = WebPGetWorkerInterface();
+  int ok_main;
 
-  if (picture->stats != NULL) {
-    WebPAuxStats* const stats = picture->stats;
-    stats->lossless_features = 0;
-    if (enc->use_predict_) stats->lossless_features |= 1;
-    if (enc->use_cross_color_) stats->lossless_features |= 2;
-    if (enc->use_subtract_green_) stats->lossless_features |= 4;
-    if (enc->use_palette_) stats->lossless_features |= 8;
-    stats->histogram_bits = enc->histo_bits_;
-    stats->transform_bits = enc->transform_bits_;
-    stats->cache_bits = enc->cache_bits_;
-    stats->palette_size = enc->palette_size_;
-    stats->lossless_size = (int)(VP8LBitWriterNumBytes(bw) - byte_position);
-    stats->lossless_hdr_size = hdr_size;
-    stats->lossless_data_size = data_size;
+  // Analyze image (entropy, num_palettes etc)
+  if (enc_main == NULL ||
+      !EncoderAnalyze(enc_main, crunch_configs, &num_crunch_configs_main,
+                      &red_and_blue_always_zero) ||
+      !EncoderInit(enc_main) || !VP8LBitWriterInit(&bw_side, 0)) {
+    err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+    goto Error;
   }
 
- Error:
-  VP8LEncoderDelete(enc);
+  // Split the configs between the main and side threads (if any).
+  if (config->thread_level > 0) {
+    num_crunch_configs_side = num_crunch_configs_main / 2;
+    for (idx = 0; idx < num_crunch_configs_side; ++idx) {
+      params_side.crunch_configs_[idx] =
+          crunch_configs[num_crunch_configs_main - num_crunch_configs_side +
+                         idx];
+    }
+    params_side.num_crunch_configs_ = num_crunch_configs_side;
+  }
+  num_crunch_configs_main -= num_crunch_configs_side;
+  for (idx = 0; idx < num_crunch_configs_main; ++idx) {
+    params_main.crunch_configs_[idx] = crunch_configs[idx];
+  }
+  params_main.num_crunch_configs_ = num_crunch_configs_main;
+
+  // Fill in the parameters for the thread workers.
+  {
+    const int params_size = (num_crunch_configs_side > 0) ? 2 : 1;
+    for (idx = 0; idx < params_size; ++idx) {
+      // Create the parameters for each worker.
+      WebPWorker* const worker = (idx == 0) ? &worker_main : &worker_side;
+      StreamEncodeContext* const param =
+          (idx == 0) ? &params_main : &params_side;
+      param->config_ = config;
+      param->picture_ = picture;
+      param->use_cache_ = use_cache;
+      param->red_and_blue_always_zero_ = red_and_blue_always_zero;
+      if (idx == 0) {
+        param->stats_ = picture->stats;
+        param->bw_ = bw_main;
+        param->enc_ = enc_main;
+      } else {
+        param->stats_ = (picture->stats == NULL) ? NULL : &stats_side;
+        // Create a side bit writer.
+        if (!VP8LBitWriterClone(bw_main, &bw_side)) {
+          err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+          goto Error;
+        }
+        param->bw_ = &bw_side;
+        // Create a side encoder.
+        enc_side = VP8LEncoderNew(config, picture);
+        if (enc_side == NULL || !EncoderInit(enc_side)) {
+          err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+          goto Error;
+        }
+        // Copy the values that were computed for the main encoder.
+        enc_side->histo_bits_ = enc_main->histo_bits_;
+        enc_side->transform_bits_ = enc_main->transform_bits_;
+        enc_side->palette_size_ = enc_main->palette_size_;
+        memcpy(enc_side->palette_, enc_main->palette_,
+               sizeof(enc_main->palette_));
+        param->enc_ = enc_side;
+      }
+      // Create the workers.
+      worker_interface->Init(worker);
+      worker->data1 = param;
+      worker->data2 = NULL;
+      worker->hook = (WebPWorkerHook)EncodeStreamHook;
+    }
+  }
+
+  // Start the second thread if needed.
+  if (num_crunch_configs_side != 0) {
+    if (!worker_interface->Reset(&worker_side)) {
+      err = VP8_ENC_ERROR_OUT_OF_MEMORY;
+      goto Error;
+    }
+#if !defined(WEBP_DISABLE_STATS)
+    // This line is here and not in the param initialization above to remove a
+    // Clang static analyzer warning.
+    if (picture->stats != NULL) {
+      memcpy(&stats_side, picture->stats, sizeof(stats_side));
+    }
+#endif
+    // This line is only useful to remove a Clang static analyzer warning.
+    params_side.err_ = VP8_ENC_OK;
+    worker_interface->Launch(&worker_side);
+  }
+  // Execute the main thread.
+  worker_interface->Execute(&worker_main);
+  ok_main = worker_interface->Sync(&worker_main);
+  worker_interface->End(&worker_main);
+  if (num_crunch_configs_side != 0) {
+    // Wait for the second thread.
+    const int ok_side = worker_interface->Sync(&worker_side);
+    worker_interface->End(&worker_side);
+    if (!ok_main || !ok_side) {
+      err = ok_main ? params_side.err_ : params_main.err_;
+      goto Error;
+    }
+    if (VP8LBitWriterNumBytes(&bw_side) < VP8LBitWriterNumBytes(bw_main)) {
+      VP8LBitWriterSwap(bw_main, &bw_side);
+#if !defined(WEBP_DISABLE_STATS)
+      if (picture->stats != NULL) {
+        memcpy(picture->stats, &stats_side, sizeof(*picture->stats));
+      }
+#endif
+    }
+  } else {
+    if (!ok_main) {
+      err = params_main.err_;
+      goto Error;
+    }
+  }
+
+Error:
+  VP8LBitWriterWipeOut(&bw_side);
+  VP8LEncoderDelete(enc_main);
+  VP8LEncoderDelete(enc_side);
   return err;
 }
 
+#undef CRUNCH_CONFIGS_MAX
+#undef CRUNCH_CONFIGS_LZ77_MAX
+
 int VP8LEncodeImage(const WebPConfig* const config,
                     const WebPPicture* const picture) {
   int width, height;
@@ -1642,11 +1953,13 @@ int VP8LEncodeImage(const WebPConfig* const config,
 
   if (!WebPReportProgress(picture, 100, &percent)) goto UserAbort;
 
+#if !defined(WEBP_DISABLE_STATS)
   // Save size.
   if (picture->stats != NULL) {
     picture->stats->coded_size += (int)coded_size;
     picture->stats->lossless_size = (int)coded_size;
   }
+#endif
 
   if (picture->extra_info != NULL) {
     const int mb_w = (width + 15) >> 4;
diff --git a/src/3rdparty/libwebp/src/enc/vp8li_enc.h b/src/3rdparty/libwebp/src/enc/vp8li_enc.h
index 8c5fbcb..298a4a0 100644
--- a/src/3rdparty/libwebp/src/enc/vp8li_enc.h
+++ b/src/3rdparty/libwebp/src/enc/vp8li_enc.h
@@ -11,14 +11,23 @@
 //
 // Author: Vikas Arora (vikaas.arora@gmail.com)
 
-#ifndef WEBP_ENC_VP8LI_H_
-#define WEBP_ENC_VP8LI_H_
+#ifndef WEBP_ENC_VP8LI_ENC_H_
+#define WEBP_ENC_VP8LI_ENC_H_
 
-#include "./backward_references_enc.h"
-#include "./histogram_enc.h"
-#include "../utils/bit_writer_utils.h"
-#include "../webp/encode.h"
-#include "../webp/format_constants.h"
+#ifdef HAVE_CONFIG_H
+#include "src/webp/config.h"
+#endif
+// Either WEBP_NEAR_LOSSLESS is defined as 0 in config.h when compiling to
+// disable near-lossless, or it is enabled by default.
+#ifndef WEBP_NEAR_LOSSLESS
+#define WEBP_NEAR_LOSSLESS 1
+#endif
+
+#include "src/enc/backward_references_enc.h"
+#include "src/enc/histogram_enc.h"
+#include "src/utils/bit_writer_utils.h"
+#include "src/webp/encode.h"
+#include "src/webp/format_constants.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -27,16 +36,24 @@ extern "C" {
 // maximum value of transform_bits_ in VP8LEncoder.
 #define MAX_TRANSFORM_BITS 6
 
+typedef enum {
+  kEncoderNone = 0,
+  kEncoderARGB,
+  kEncoderNearLossless,
+  kEncoderPalette
+} VP8LEncoderARGBContent;
+
 typedef struct {
   const WebPConfig* config_;      // user configuration and parameters
   const WebPPicture* pic_;        // input picture.
 
-  uint32_t* argb_;                // Transformed argb image data.
-  uint32_t* argb_scratch_;        // Scratch memory for argb rows
-                                  // (used for prediction).
-  uint32_t* transform_data_;      // Scratch memory for transform data.
-  uint32_t* transform_mem_;       // Currently allocated memory.
-  size_t    transform_mem_size_;  // Currently allocated memory size.
+  uint32_t* argb_;                       // Transformed argb image data.
+  VP8LEncoderARGBContent argb_content_;  // Content type of the argb buffer.
+  uint32_t* argb_scratch_;               // Scratch memory for argb rows
+                                         // (used for prediction).
+  uint32_t* transform_data_;             // Scratch memory for transform data.
+  uint32_t* transform_mem_;              // Currently allocated memory.
+  size_t    transform_mem_size_;         // Currently allocated memory size.
 
   int       current_width_;       // Corresponds to packed image width.
 
@@ -54,8 +71,7 @@ typedef struct {
   uint32_t palette_[MAX_PALETTE_SIZE];
 
   // Some 'scratch' (potentially large) objects.
-  struct VP8LBackwardRefs refs_[2];  // Backward Refs array corresponding to
-                                     // LZ77 & RLE coding.
+  struct VP8LBackwardRefs refs_[3];  // Backward Refs array for temporaries.
   VP8LHashChain hash_chain_;         // HashChain data for constructing
                                      // backward references.
 } VP8LEncoder;
@@ -75,6 +91,13 @@ WebPEncodingError VP8LEncodeStream(const WebPConfig* const config,
                                    const WebPPicture* const picture,
                                    VP8LBitWriter* const bw, int use_cache);
 
+#if (WEBP_NEAR_LOSSLESS == 1)
+// in near_lossless.c
+// Near lossless preprocessing in RGB color-space.
+int VP8ApplyNearLossless(const WebPPicture* const picture, int quality,
+                         uint32_t* const argb_dst);
+#endif
+
 //------------------------------------------------------------------------------
 // Image transforms in predictor.c.
 
@@ -92,4 +115,4 @@ void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_ENC_VP8LI_H_ */
+#endif  /* WEBP_ENC_VP8LI_ENC_H_ */
diff --git a/src/3rdparty/libwebp/src/enc/webp_enc.c b/src/3rdparty/libwebp/src/enc/webp_enc.c
index f18461e..283cda8 100644
--- a/src/3rdparty/libwebp/src/enc/webp_enc.c
+++ b/src/3rdparty/libwebp/src/enc/webp_enc.c
@@ -16,10 +16,10 @@
 #include <string.h>
 #include <math.h>
 
-#include "./cost_enc.h"
-#include "./vp8i_enc.h"
-#include "./vp8li_enc.h"
-#include "../utils/utils.h"
+#include "src/enc/cost_enc.h"
+#include "src/enc/vp8i_enc.h"
+#include "src/enc/vp8li_enc.h"
+#include "src/utils/utils.h"
 
 // #define PRINT_MEMORY_INFO
 
@@ -207,7 +207,7 @@ static VP8Encoder* InitVP8Encoder(const WebPConfig* const config,
   enc->preds_w_ = preds_w;
   enc->mb_info_ = (VP8MBInfo*)mem;
   mem += info_size;
-  enc->preds_ = ((uint8_t*)mem) + 1 + enc->preds_w_;
+  enc->preds_ = mem + 1 + enc->preds_w_;
   mem += preds_size;
   enc->nz_ = 1 + (uint32_t*)WEBP_ALIGN(mem);
   mem += nz_size;
@@ -216,7 +216,7 @@ static VP8Encoder* InitVP8Encoder(const WebPConfig* const config,
 
   // top samples (all 16-aligned)
   mem = (uint8_t*)WEBP_ALIGN(mem);
-  enc->y_top_ = (uint8_t*)mem;
+  enc->y_top_ = mem;
   enc->uv_top_ = enc->y_top_ + top_stride;
   mem += 2 * top_stride;
   assert(mem <= (uint8_t*)enc + size);
@@ -256,6 +256,7 @@ static int DeleteVP8Encoder(VP8Encoder* enc) {
 
 //------------------------------------------------------------------------------
 
+#if !defined(WEBP_DISABLE_STATS)
 static double GetPSNR(uint64_t err, uint64_t size) {
   return (err > 0 && size > 0) ? 10. * log10(255. * 255. * size / err) : 99.;
 }
@@ -270,8 +271,10 @@ static void FinalizePSNR(const VP8Encoder* const enc) {
   stats->PSNR[3] = (float)GetPSNR(sse[0] + sse[1] + sse[2], size * 3 / 2);
   stats->PSNR[4] = (float)GetPSNR(sse[3], size);
 }
+#endif  // !defined(WEBP_DISABLE_STATS)
 
 static void StoreStats(VP8Encoder* const enc) {
+#if !defined(WEBP_DISABLE_STATS)
   WebPAuxStats* const stats = enc->pic_->stats;
   if (stats != NULL) {
     int i, s;
@@ -288,7 +291,9 @@ static void StoreStats(VP8Encoder* const enc) {
       stats->block_count[i] = enc->block_count_[i];
     }
   }
+#else  // defined(WEBP_DISABLE_STATS)
   WebPReportProgress(enc->pic_, 100, &enc->percent_);  // done!
+#endif  // !defined(WEBP_DISABLE_STATS)
 }
 
 int WebPEncodingSetError(const WebPPicture* const pic,
@@ -336,10 +341,6 @@ int WebPEncode(const WebPConfig* config, WebPPicture* pic) {
   if (!config->lossless) {
     VP8Encoder* enc = NULL;
 
-    if (!config->exact) {
-      WebPCleanupTransparentArea(pic);
-    }
-
     if (pic->use_argb || pic->y == NULL || pic->u == NULL || pic->v == NULL) {
       // Make sure we have YUVA samples.
       if (config->use_sharp_yuv || (config->preprocessing & 4)) {
@@ -361,6 +362,10 @@ int WebPEncode(const WebPConfig* config, WebPPicture* pic) {
       }
     }
 
+    if (!config->exact) {
+      WebPCleanupTransparentArea(pic);
+    }
+
     enc = InitVP8Encoder(config, pic);
     if (enc == NULL) return 0;  // pic->error is already set.
     // Note: each of the tasks below account for 20% in the progress report.
diff --git a/src/3rdparty/libwebp/src/mux/anim_encode.c b/src/3rdparty/libwebp/src/mux/anim_encode.c
index 6066388..7be9906 100644
--- a/src/3rdparty/libwebp/src/mux/anim_encode.c
+++ b/src/3rdparty/libwebp/src/mux/anim_encode.c
@@ -16,12 +16,12 @@
 #include <stdio.h>
 #include <stdlib.h>  // for abs()
 
-#include "../mux/animi.h"
-#include "../utils/utils.h"
-#include "../webp/decode.h"
-#include "../webp/encode.h"
-#include "../webp/format_constants.h"
-#include "../webp/mux.h"
+#include "src/mux/animi.h"
+#include "src/utils/utils.h"
+#include "src/webp/decode.h"
+#include "src/webp/encode.h"
+#include "src/webp/format_constants.h"
+#include "src/webp/mux.h"
 
 #if defined(_MSC_VER) && _MSC_VER < 1900
 #define snprintf _snprintf
@@ -35,7 +35,7 @@
 // Stores frame rectangle dimensions.
 typedef struct {
   int x_offset_, y_offset_, width_, height_;
-} FrameRect;
+} FrameRectangle;
 
 // Used to store two candidates of encoded data for an animation frame. One of
 // the two will be chosen later.
@@ -50,7 +50,7 @@ struct WebPAnimEncoder {
   const int canvas_height_;                 // Canvas height.
   const WebPAnimEncoderOptions options_;    // Global encoding options.
 
-  FrameRect prev_rect_;               // Previous WebP frame rectangle.
+  FrameRectangle prev_rect_;          // Previous WebP frame rectangle.
   WebPConfig last_config_;            // Cached in case a re-encode is needed.
   WebPConfig last_config_reversed_;   // If 'last_config_' uses lossless, then
                                       // this config uses lossy and vice versa;
@@ -206,7 +206,7 @@ static void ClearRectangle(WebPPicture* const picture,
 }
 
 static void WebPUtilClearPic(WebPPicture* const picture,
-                             const FrameRect* const rect) {
+                             const FrameRectangle* const rect) {
   if (rect != NULL) {
     ClearRectangle(picture, rect->x_offset_, rect->y_offset_,
                    rect->width_, rect->height_);
@@ -400,7 +400,7 @@ static WEBP_INLINE int ComparePixelsLossy(const uint32_t* src, int src_step,
   return 1;
 }
 
-static int IsEmptyRect(const FrameRect* const rect) {
+static int IsEmptyRect(const FrameRectangle* const rect) {
   return (rect->width_ == 0) || (rect->height_ == 0);
 }
 
@@ -413,7 +413,7 @@ static int QualityToMaxDiff(float quality) {
 // Assumes that an initial valid guess of change rectangle 'rect' is passed.
 static void MinimizeChangeRectangle(const WebPPicture* const src,
                                     const WebPPicture* const dst,
-                                    FrameRect* const rect,
+                                    FrameRectangle* const rect,
                                     int is_lossless, float quality) {
   int i, j;
   const ComparePixelsFunc compare_pixels =
@@ -498,7 +498,7 @@ static void MinimizeChangeRectangle(const WebPPicture* const src,
 }
 
 // Snap rectangle to even offsets (and adjust dimensions if needed).
-static WEBP_INLINE void SnapToEvenOffsets(FrameRect* const rect) {
+static WEBP_INLINE void SnapToEvenOffsets(FrameRectangle* const rect) {
   rect->width_ += (rect->x_offset_ & 1);
   rect->height_ += (rect->y_offset_ & 1);
   rect->x_offset_ &= ~1;
@@ -508,9 +508,9 @@ static WEBP_INLINE void SnapToEvenOffsets(FrameRect* const rect) {
 typedef struct {
   int should_try_;               // Should try this set of parameters.
   int empty_rect_allowed_;       // Frame with empty rectangle can be skipped.
-  FrameRect rect_ll_;            // Frame rectangle for lossless compression.
+  FrameRectangle rect_ll_;       // Frame rectangle for lossless compression.
   WebPPicture sub_frame_ll_;     // Sub-frame pic for lossless compression.
-  FrameRect rect_lossy_;         // Frame rectangle for lossy compression.
+  FrameRectangle rect_lossy_;    // Frame rectangle for lossy compression.
                                  // Could be smaller than rect_ll_ as pixels
                                  // with small diffs can be ignored.
   WebPPicture sub_frame_lossy_;  // Sub-frame pic for lossless compression.
@@ -538,7 +538,8 @@ static void SubFrameParamsFree(SubFrameParams* const params) {
 static int GetSubRect(const WebPPicture* const prev_canvas,
                       const WebPPicture* const curr_canvas, int is_key_frame,
                       int is_first_frame, int empty_rect_allowed,
-                      int is_lossless, float quality, FrameRect* const rect,
+                      int is_lossless, float quality,
+                      FrameRectangle* const rect,
                       WebPPicture* const sub_frame) {
   if (!is_key_frame || is_first_frame) {  // Optimize frame rectangle.
     // Note: This behaves as expected for first frame, as 'prev_canvas' is
@@ -594,7 +595,7 @@ int WebPAnimEncoderRefineRect(
     const WebPPicture* const prev_canvas, const WebPPicture* const curr_canvas,
     int is_lossless, float quality, int* const x_offset, int* const y_offset,
     int* const width, int* const height) {
-  FrameRect rect;
+  FrameRectangle rect;
   const int right = clip(*x_offset + *width, 0, curr_canvas->width);
   const int left = clip(*x_offset, 0, curr_canvas->width - 1);
   const int bottom = clip(*y_offset + *height, 0, curr_canvas->height);
@@ -620,7 +621,7 @@ int WebPAnimEncoderRefineRect(
 }
 
 static void DisposeFrameRectangle(int dispose_method,
-                                  const FrameRect* const rect,
+                                  const FrameRectangle* const rect,
                                   WebPPicture* const curr_canvas) {
   assert(rect != NULL);
   if (dispose_method == WEBP_MUX_DISPOSE_BACKGROUND) {
@@ -628,13 +629,13 @@ static void DisposeFrameRectangle(int dispose_method,
   }
 }
 
-static uint32_t RectArea(const FrameRect* const rect) {
+static uint32_t RectArea(const FrameRectangle* const rect) {
   return (uint32_t)rect->width_ * rect->height_;
 }
 
 static int IsLosslessBlendingPossible(const WebPPicture* const src,
                                       const WebPPicture* const dst,
-                                      const FrameRect* const rect) {
+                                      const FrameRectangle* const rect) {
   int i, j;
   assert(src->width == dst->width && src->height == dst->height);
   assert(rect->x_offset_ + rect->width_ <= dst->width);
@@ -656,7 +657,7 @@ static int IsLosslessBlendingPossible(const WebPPicture* const src,
 
 static int IsLossyBlendingPossible(const WebPPicture* const src,
                                    const WebPPicture* const dst,
-                                   const FrameRect* const rect,
+                                   const FrameRectangle* const rect,
                                    float quality) {
   const int max_allowed_diff_lossy = QualityToMaxDiff(quality);
   int i, j;
@@ -683,7 +684,7 @@ static int IsLossyBlendingPossible(const WebPPicture* const src,
 // transparent pixels.
 // Returns true if at least one pixel gets modified.
 static int IncreaseTransparency(const WebPPicture* const src,
-                                const FrameRect* const rect,
+                                const FrameRectangle* const rect,
                                 WebPPicture* const dst) {
   int i, j;
   int modified = 0;
@@ -709,7 +710,7 @@ static int IncreaseTransparency(const WebPPicture* const src,
 // Assumes lossy compression is being used.
 // Returns true if at least one pixel gets modified.
 static int FlattenSimilarBlocks(const WebPPicture* const src,
-                                const FrameRect* const rect,
+                                const FrameRectangle* const rect,
                                 WebPPicture* const dst, float quality) {
   const int max_allowed_diff_lossy = QualityToMaxDiff(quality);
   int i, j;
@@ -778,13 +779,13 @@ static int EncodeFrame(const WebPConfig* const config, WebPPicture* const pic,
 typedef struct {
   WebPMemoryWriter  mem_;
   WebPMuxFrameInfo  info_;
-  FrameRect         rect_;
+  FrameRectangle    rect_;
   int               evaluate_;  // True if this candidate should be evaluated.
 } Candidate;
 
 // Generates a candidate encoded frame given a picture and metadata.
 static WebPEncodingError EncodeCandidate(WebPPicture* const sub_frame,
-                                         const FrameRect* const rect,
+                                         const FrameRectangle* const rect,
                                          const WebPConfig* const encoder_config,
                                          int use_blending,
                                          Candidate* const candidate) {
@@ -958,7 +959,7 @@ static int IncreasePreviousDuration(WebPAnimEncoder* const enc, int duration) {
   if (new_duration >= MAX_DURATION) {  // Special case.
     // Separate out previous frame from earlier merged frames to avoid overflow.
     // We add a 1x1 transparent frame for the previous frame, with blending on.
-    const FrameRect rect = { 0, 0, 1, 1 };
+    const FrameRectangle rect = { 0, 0, 1, 1 };
     const uint8_t lossless_1x1_bytes[] = {
       0x52, 0x49, 0x46, 0x46, 0x14, 0x00, 0x00, 0x00, 0x57, 0x45, 0x42, 0x50,
       0x56, 0x50, 0x38, 0x4c, 0x08, 0x00, 0x00, 0x00, 0x2f, 0x00, 0x00, 0x00,
@@ -1223,7 +1224,7 @@ static int CacheFrame(WebPAnimEncoder* const enc,
       enc->prev_candidate_undecided_ = 0;
     } else {
       int64_t curr_delta;
-      FrameRect prev_rect_key, prev_rect_sub;
+      FrameRectangle prev_rect_key, prev_rect_sub;
 
       // Add this as a frame rectangle to enc.
       error_code = SetFrame(enc, config, 0, encoded_frame, &frame_skipped);
@@ -1535,7 +1536,8 @@ int WebPAnimEncoderAssemble(WebPAnimEncoder* enc, WebPData* webp_data) {
 
   if (!enc->got_null_frame_ && enc->in_frame_count_ > 1 && enc->count_ > 0) {
     // set duration of the last frame to be avg of durations of previous frames.
-    const double delta_time = enc->prev_timestamp_ - enc->first_timestamp_;
+    const double delta_time =
+        (uint32_t)enc->prev_timestamp_ - enc->first_timestamp_;
     const int average_duration = (int)(delta_time / (enc->in_frame_count_ - 1));
     if (!IncreasePreviousDuration(enc, average_duration)) {
       return 0;
diff --git a/src/3rdparty/libwebp/src/mux/animi.h b/src/3rdparty/libwebp/src/mux/animi.h
index cecaf1f..8889953 100644
--- a/src/3rdparty/libwebp/src/mux/animi.h
+++ b/src/3rdparty/libwebp/src/mux/animi.h
@@ -14,7 +14,7 @@
 #ifndef WEBP_MUX_ANIMI_H_
 #define WEBP_MUX_ANIMI_H_
 
-#include "../webp/mux.h"
+#include "src/webp/mux.h"
 
 #ifdef __cplusplus
 extern "C" {
diff --git a/src/3rdparty/libwebp/src/mux/muxedit.c b/src/3rdparty/libwebp/src/mux/muxedit.c
index d2c5305..7a027b3 100644
--- a/src/3rdparty/libwebp/src/mux/muxedit.c
+++ b/src/3rdparty/libwebp/src/mux/muxedit.c
@@ -13,8 +13,8 @@
 //          Vikas (vikasa@google.com)
 
 #include <assert.h>
-#include "./muxi.h"
-#include "../utils/utils.h"
+#include "src/mux/muxi.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Life of a mux object.
diff --git a/src/3rdparty/libwebp/src/mux/muxi.h b/src/3rdparty/libwebp/src/mux/muxi.h
index e6606aa..b73e3fb 100644
--- a/src/3rdparty/libwebp/src/mux/muxi.h
+++ b/src/3rdparty/libwebp/src/mux/muxi.h
@@ -15,9 +15,9 @@
 #define WEBP_MUX_MUXI_H_
 
 #include <stdlib.h>
-#include "../dec/vp8i_dec.h"
-#include "../dec/vp8li_dec.h"
-#include "../webp/mux.h"
+#include "src/dec/vp8i_dec.h"
+#include "src/dec/vp8li_dec.h"
+#include "src/webp/mux.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -28,7 +28,7 @@ extern "C" {
 
 #define MUX_MAJ_VERSION 0
 #define MUX_MIN_VERSION 4
-#define MUX_REV_VERSION 0
+#define MUX_REV_VERSION 1
 
 // Chunk object.
 typedef struct WebPChunk WebPChunk;
diff --git a/src/3rdparty/libwebp/src/mux/muxinternal.c b/src/3rdparty/libwebp/src/mux/muxinternal.c
index 387b57e..1473f10 100644
--- a/src/3rdparty/libwebp/src/mux/muxinternal.c
+++ b/src/3rdparty/libwebp/src/mux/muxinternal.c
@@ -13,8 +13,8 @@
 //          Vikas (vikasa@google.com)
 
 #include <assert.h>
-#include "./muxi.h"
-#include "../utils/utils.h"
+#include "src/mux/muxi.h"
+#include "src/utils/utils.h"
 
 #define UNDEFINED_CHUNK_SIZE ((uint32_t)(-1))
 
@@ -504,6 +504,20 @@ WebPMuxError MuxValidate(const WebPMux* const mux) {
     if (!has_animation && (num_anim == 1 || num_frames > 0)) {
       return WEBP_MUX_INVALID_ARGUMENT;
     }
+    if (!has_animation) {
+      const WebPMuxImage* images = mux->images_;
+      // There can be only one image.
+      if (images == NULL || images->next_ != NULL) {
+        return WEBP_MUX_INVALID_ARGUMENT;
+      }
+      // Size must match.
+      if (mux->canvas_width_ > 0) {
+        if (images->width_ != mux->canvas_width_ ||
+            images->height_ != mux->canvas_height_) {
+          return WEBP_MUX_INVALID_ARGUMENT;
+        }
+      }
+    }
   }
 
   // Verify either VP8X chunk is present OR there is only one elem in
@@ -515,6 +529,7 @@ WebPMuxError MuxValidate(const WebPMux* const mux) {
   if (num_vp8x == 0 && num_images != 1) return WEBP_MUX_INVALID_ARGUMENT;
 
   // ALPHA_FLAG & alpha chunk(s) are consistent.
+  // Note: ALPHA_FLAG can be set when there is actually no Alpha data present.
   if (MuxHasAlpha(mux->images_)) {
     if (num_vp8x > 0) {
       // VP8X chunk is present, so it should contain ALPHA_FLAG.
@@ -525,8 +540,6 @@ WebPMuxError MuxValidate(const WebPMux* const mux) {
       if (err != WEBP_MUX_OK) return err;
       if (num_alpha > 0) return WEBP_MUX_INVALID_ARGUMENT;
     }
-  } else {  // Mux doesn't need alpha. So, ALPHA_FLAG should NOT be present.
-    if (flags & ALPHA_FLAG) return WEBP_MUX_INVALID_ARGUMENT;
   }
 
   return WEBP_MUX_OK;
diff --git a/src/3rdparty/libwebp/src/mux/muxread.c b/src/3rdparty/libwebp/src/mux/muxread.c
index 410acd9..0b55286 100644
--- a/src/3rdparty/libwebp/src/mux/muxread.c
+++ b/src/3rdparty/libwebp/src/mux/muxread.c
@@ -13,8 +13,8 @@
 //          Vikas (vikasa@google.com)
 
 #include <assert.h>
-#include "./muxi.h"
-#include "../utils/utils.h"
+#include "src/mux/muxi.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Helper method(s).
@@ -43,7 +43,7 @@ static WebPMuxError MuxGet(const WebPMux* const mux, CHUNK_INDEX idx,
   SWITCH_ID_LIST(IDX_ANIM, mux->anim_);
   SWITCH_ID_LIST(IDX_EXIF, mux->exif_);
   SWITCH_ID_LIST(IDX_XMP, mux->xmp_);
-  SWITCH_ID_LIST(IDX_UNKNOWN, mux->unknown_);
+  assert(idx != IDX_UNKNOWN);
   return WEBP_MUX_NOT_FOUND;
 }
 #undef SWITCH_ID_LIST
@@ -270,6 +270,9 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
     ChunkInit(&chunk);
   }
 
+  // Incomplete image.
+  if (wpi->is_partial_) goto Err;
+
   // Validate mux if complete.
   if (MuxValidate(mux) != WEBP_MUX_OK) goto Err;
 
diff --git a/src/3rdparty/libwebp/src/utils/bit_reader_inl_utils.h b/src/3rdparty/libwebp/src/utils/bit_reader_inl_utils.h
index fd7fb04..2ccc6ed 100644
--- a/src/3rdparty/libwebp/src/utils/bit_reader_inl_utils.h
+++ b/src/3rdparty/libwebp/src/utils/bit_reader_inl_utils.h
@@ -13,19 +13,19 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_UTILS_BIT_READER_INL_H_
-#define WEBP_UTILS_BIT_READER_INL_H_
+#ifndef WEBP_UTILS_BIT_READER_INL_UTILS_H_
+#define WEBP_UTILS_BIT_READER_INL_UTILS_H_
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
 #include <string.h>  // for memcpy
 
-#include "../dsp/dsp.h"
-#include "./bit_reader_utils.h"
-#include "./endian_inl_utils.h"
-#include "./utils.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/bit_reader_utils.h"
+#include "src/utils/endian_inl_utils.h"
+#include "src/utils/utils.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -187,4 +187,4 @@ static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
 }    // extern "C"
 #endif
 
-#endif   // WEBP_UTILS_BIT_READER_INL_H_
+#endif   // WEBP_UTILS_BIT_READER_INL_UTILS_H_
diff --git a/src/3rdparty/libwebp/src/utils/bit_reader_utils.c b/src/3rdparty/libwebp/src/utils/bit_reader_utils.c
index c3157e8..1500354 100644
--- a/src/3rdparty/libwebp/src/utils/bit_reader_utils.c
+++ b/src/3rdparty/libwebp/src/utils/bit_reader_utils.c
@@ -12,11 +12,11 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
-#include "./bit_reader_inl_utils.h"
-#include "../utils/utils.h"
+#include "src/utils/bit_reader_inl_utils.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // VP8BitReader
diff --git a/src/3rdparty/libwebp/src/utils/bit_reader_utils.h b/src/3rdparty/libwebp/src/utils/bit_reader_utils.h
index ec3426c..317d311 100644
--- a/src/3rdparty/libwebp/src/utils/bit_reader_utils.h
+++ b/src/3rdparty/libwebp/src/utils/bit_reader_utils.h
@@ -12,14 +12,14 @@
 // Author: Skal (pascal.massimino@gmail.com)
 //         Vikas Arora (vikaas.arora@gmail.com)
 
-#ifndef WEBP_UTILS_BIT_READER_H_
-#define WEBP_UTILS_BIT_READER_H_
+#ifndef WEBP_UTILS_BIT_READER_UTILS_H_
+#define WEBP_UTILS_BIT_READER_UTILS_H_
 
 #include <assert.h>
 #ifdef _MSC_VER
 #include <stdlib.h>  // _byteswap_ulong
 #endif
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -155,9 +155,10 @@ static WEBP_INLINE int VP8LIsEndOfStream(const VP8LBitReader* const br) {
 
 // For jumping over a number of bits in the bit stream when accessed with
 // VP8LPrefetchBits and VP8LFillBitWindow.
+// This function does *not* set br->eos_, since it's speed-critical.
+// Use with extreme care!
 static WEBP_INLINE void VP8LSetBitPos(VP8LBitReader* const br, int val) {
   br->bit_pos_ = val;
-  br->eos_ = VP8LIsEndOfStream(br);
 }
 
 // Advances the read buffer by 4 bytes to make room for reading next 32 bits.
@@ -171,4 +172,4 @@ static WEBP_INLINE void VP8LFillBitWindow(VP8LBitReader* const br) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_BIT_READER_H_ */
+#endif  /* WEBP_UTILS_BIT_READER_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/bit_writer_utils.c b/src/3rdparty/libwebp/src/utils/bit_writer_utils.c
index ab0c49d..f4f476c 100644
--- a/src/3rdparty/libwebp/src/utils/bit_writer_utils.c
+++ b/src/3rdparty/libwebp/src/utils/bit_writer_utils.c
@@ -16,9 +16,9 @@
 #include <string.h>   // for memcpy()
 #include <stdlib.h>
 
-#include "./bit_writer_utils.h"
-#include "./endian_inl_utils.h"
-#include "./utils.h"
+#include "src/utils/bit_writer_utils.h"
+#include "src/utils/endian_inl_utils.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // VP8BitWriter
@@ -239,6 +239,18 @@ int VP8LBitWriterInit(VP8LBitWriter* const bw, size_t expected_size) {
   return VP8LBitWriterResize(bw, expected_size);
 }
 
+int VP8LBitWriterClone(const VP8LBitWriter* const src,
+                       VP8LBitWriter* const dst) {
+  const size_t current_size = src->cur_ - src->buf_;
+  assert(src->cur_ >= src->buf_ && src->cur_ <= src->end_);
+  if (!VP8LBitWriterResize(dst, current_size)) return 0;
+  memcpy(dst->buf_, src->buf_, current_size);
+  dst->bits_ = src->bits_;
+  dst->used_ = src->used_;
+  dst->error_ = src->error_;
+  return 1;
+}
+
 void VP8LBitWriterWipeOut(VP8LBitWriter* const bw) {
   if (bw != NULL) {
     WebPSafeFree(bw->buf_);
@@ -246,6 +258,21 @@ void VP8LBitWriterWipeOut(VP8LBitWriter* const bw) {
   }
 }
 
+void VP8LBitWriterReset(const VP8LBitWriter* const bw_init,
+                        VP8LBitWriter* const bw) {
+  bw->bits_ = bw_init->bits_;
+  bw->used_ = bw_init->used_;
+  bw->cur_ = bw->buf_ + (bw_init->cur_ - bw_init->buf_);
+  assert(bw->cur_ <= bw->end_);
+  bw->error_ = bw_init->error_;
+}
+
+void VP8LBitWriterSwap(VP8LBitWriter* const src, VP8LBitWriter* const dst) {
+  const VP8LBitWriter tmp = *src;
+  *src = *dst;
+  *dst = tmp;
+}
+
 void VP8LPutBitsFlushBits(VP8LBitWriter* const bw) {
   // If needed, make some room by flushing some bits out.
   if (bw->cur_ + VP8L_WRITER_BYTES > bw->end_) {
diff --git a/src/3rdparty/libwebp/src/utils/bit_writer_utils.h b/src/3rdparty/libwebp/src/utils/bit_writer_utils.h
index 9c02bbc..2cf5976 100644
--- a/src/3rdparty/libwebp/src/utils/bit_writer_utils.h
+++ b/src/3rdparty/libwebp/src/utils/bit_writer_utils.h
@@ -11,10 +11,10 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_UTILS_BIT_WRITER_H_
-#define WEBP_UTILS_BIT_WRITER_H_
+#ifndef WEBP_UTILS_BIT_WRITER_UTILS_H_
+#define WEBP_UTILS_BIT_WRITER_UTILS_H_
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -100,16 +100,24 @@ typedef struct {
   int error_;
 } VP8LBitWriter;
 
-static WEBP_INLINE size_t VP8LBitWriterNumBytes(VP8LBitWriter* const bw) {
+static WEBP_INLINE size_t VP8LBitWriterNumBytes(const VP8LBitWriter* const bw) {
   return (bw->cur_ - bw->buf_) + ((bw->used_ + 7) >> 3);
 }
 
 // Returns false in case of memory allocation error.
 int VP8LBitWriterInit(VP8LBitWriter* const bw, size_t expected_size);
+// Returns false in case of memory allocation error.
+int VP8LBitWriterClone(const VP8LBitWriter* const src,
+                       VP8LBitWriter* const dst);
 // Finalize the bitstream coding. Returns a pointer to the internal buffer.
 uint8_t* VP8LBitWriterFinish(VP8LBitWriter* const bw);
 // Release any pending memory and zeroes the object.
 void VP8LBitWriterWipeOut(VP8LBitWriter* const bw);
+// Resets the cursor of the BitWriter bw to when it was like in bw_init.
+void VP8LBitWriterReset(const VP8LBitWriter* const bw_init,
+                        VP8LBitWriter* const bw);
+// Swaps the memory held by two BitWriters.
+void VP8LBitWriterSwap(VP8LBitWriter* const src, VP8LBitWriter* const dst);
 
 // Internal function for VP8LPutBits flushing 32 bits from the written state.
 void VP8LPutBitsFlushBits(VP8LBitWriter* const bw);
@@ -143,4 +151,4 @@ static WEBP_INLINE void VP8LPutBits(VP8LBitWriter* const bw,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_BIT_WRITER_H_ */
+#endif  /* WEBP_UTILS_BIT_WRITER_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/color_cache_utils.c b/src/3rdparty/libwebp/src/utils/color_cache_utils.c
index 0172590..b09f538 100644
--- a/src/3rdparty/libwebp/src/utils/color_cache_utils.c
+++ b/src/3rdparty/libwebp/src/utils/color_cache_utils.c
@@ -14,8 +14,8 @@
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
-#include "./color_cache_utils.h"
-#include "./utils.h"
+#include "src/utils/color_cache_utils.h"
+#include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
 // VP8LColorCache.
diff --git a/src/3rdparty/libwebp/src/utils/color_cache_utils.h b/src/3rdparty/libwebp/src/utils/color_cache_utils.h
index c373e6b..20b7be1 100644
--- a/src/3rdparty/libwebp/src/utils/color_cache_utils.h
+++ b/src/3rdparty/libwebp/src/utils/color_cache_utils.h
@@ -12,10 +12,12 @@
 // Authors: Jyrki Alakuijala (jyrki@google.com)
 //          Urvang Joshi (urvang@google.com)
 
-#ifndef WEBP_UTILS_COLOR_CACHE_H_
-#define WEBP_UTILS_COLOR_CACHE_H_
+#ifndef WEBP_UTILS_COLOR_CACHE_UTILS_H_
+#define WEBP_UTILS_COLOR_CACHE_UTILS_H_
 
-#include "../webp/types.h"
+#include <assert.h>
+
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -30,7 +32,7 @@ typedef struct {
 
 static const uint64_t kHashMul = 0x1e35a7bdull;
 
-static WEBP_INLINE int HashPix(uint32_t argb, int shift) {
+static WEBP_INLINE int VP8LHashPix(uint32_t argb, int shift) {
   return (int)(((argb * kHashMul) & 0xffffffffu) >> shift);
 }
 
@@ -48,19 +50,19 @@ static WEBP_INLINE void VP8LColorCacheSet(const VP8LColorCache* const cc,
 
 static WEBP_INLINE void VP8LColorCacheInsert(const VP8LColorCache* const cc,
                                              uint32_t argb) {
-  const int key = HashPix(argb, cc->hash_shift_);
+  const int key = VP8LHashPix(argb, cc->hash_shift_);
   cc->colors_[key] = argb;
 }
 
 static WEBP_INLINE int VP8LColorCacheGetIndex(const VP8LColorCache* const cc,
                                               uint32_t argb) {
-  return HashPix(argb, cc->hash_shift_);
+  return VP8LHashPix(argb, cc->hash_shift_);
 }
 
 // Return the key if cc contains argb, and -1 otherwise.
 static WEBP_INLINE int VP8LColorCacheContains(const VP8LColorCache* const cc,
                                               uint32_t argb) {
-  const int key = HashPix(argb, cc->hash_shift_);
+  const int key = VP8LHashPix(argb, cc->hash_shift_);
   return (cc->colors_[key] == argb) ? key : -1;
 }
 
@@ -82,4 +84,4 @@ void VP8LColorCacheClear(VP8LColorCache* const color_cache);
 }
 #endif
 
-#endif  // WEBP_UTILS_COLOR_CACHE_H_
+#endif  // WEBP_UTILS_COLOR_CACHE_UTILS_H_
diff --git a/src/3rdparty/libwebp/src/utils/endian_inl_utils.h b/src/3rdparty/libwebp/src/utils/endian_inl_utils.h
index e11260f..4b2f91d 100644
--- a/src/3rdparty/libwebp/src/utils/endian_inl_utils.h
+++ b/src/3rdparty/libwebp/src/utils/endian_inl_utils.h
@@ -9,15 +9,15 @@
 //
 // Endian related functions.
 
-#ifndef WEBP_UTILS_ENDIAN_INL_H_
-#define WEBP_UTILS_ENDIAN_INL_H_
+#ifndef WEBP_UTILS_ENDIAN_INL_UTILS_H_
+#define WEBP_UTILS_ENDIAN_INL_UTILS_H_
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
-#include "../dsp/dsp.h"
-#include "../webp/types.h"
+#include "src/dsp/dsp.h"
+#include "src/webp/types.h"
 
 // some endian fix (e.g.: mips-gcc doesn't define __BIG_ENDIAN__)
 #if !defined(WORDS_BIGENDIAN) && \
@@ -97,4 +97,4 @@ static WEBP_INLINE uint64_t BSwap64(uint64_t x) {
 #endif  // HAVE_BUILTIN_BSWAP64
 }
 
-#endif  // WEBP_UTILS_ENDIAN_INL_H_
+#endif  // WEBP_UTILS_ENDIAN_INL_UTILS_H_
diff --git a/src/3rdparty/libwebp/src/utils/filters_utils.c b/src/3rdparty/libwebp/src/utils/filters_utils.c
index 49c1d18..bbc2c34 100644
--- a/src/3rdparty/libwebp/src/utils/filters_utils.c
+++ b/src/3rdparty/libwebp/src/utils/filters_utils.c
@@ -11,7 +11,7 @@
 //
 // Author: Urvang (urvang@google.com)
 
-#include "./filters_utils.h"
+#include "src/utils/filters_utils.h"
 #include <stdlib.h>
 #include <string.h>
 
diff --git a/src/3rdparty/libwebp/src/utils/filters_utils.h b/src/3rdparty/libwebp/src/utils/filters_utils.h
index 088b132..410f2fc 100644
--- a/src/3rdparty/libwebp/src/utils/filters_utils.h
+++ b/src/3rdparty/libwebp/src/utils/filters_utils.h
@@ -11,11 +11,11 @@
 //
 // Author: Urvang (urvang@google.com)
 
-#ifndef WEBP_UTILS_FILTERS_H_
-#define WEBP_UTILS_FILTERS_H_
+#ifndef WEBP_UTILS_FILTERS_UTILS_H_
+#define WEBP_UTILS_FILTERS_UTILS_H_
 
-#include "../webp/types.h"
-#include "../dsp/dsp.h"
+#include "src/webp/types.h"
+#include "src/dsp/dsp.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -29,4 +29,4 @@ WEBP_FILTER_TYPE WebPEstimateBestFilter(const uint8_t* data,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_FILTERS_H_ */
+#endif  /* WEBP_UTILS_FILTERS_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/huffman_encode_utils.c b/src/3rdparty/libwebp/src/utils/huffman_encode_utils.c
index f950465..6f3b1bb 100644
--- a/src/3rdparty/libwebp/src/utils/huffman_encode_utils.c
+++ b/src/3rdparty/libwebp/src/utils/huffman_encode_utils.c
@@ -14,9 +14,9 @@
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
-#include "./huffman_encode_utils.h"
-#include "./utils.h"
-#include "../webp/format_constants.h"
+#include "src/utils/huffman_encode_utils.h"
+#include "src/utils/utils.h"
+#include "src/webp/format_constants.h"
 
 // -----------------------------------------------------------------------------
 // Util function to optimize the symbol map for RLE coding
diff --git a/src/3rdparty/libwebp/src/utils/huffman_encode_utils.h b/src/3rdparty/libwebp/src/utils/huffman_encode_utils.h
index a157165..3e6763c 100644
--- a/src/3rdparty/libwebp/src/utils/huffman_encode_utils.h
+++ b/src/3rdparty/libwebp/src/utils/huffman_encode_utils.h
@@ -11,10 +11,10 @@
 //
 // Entropy encoding (Huffman) for webp lossless
 
-#ifndef WEBP_UTILS_HUFFMAN_ENCODE_H_
-#define WEBP_UTILS_HUFFMAN_ENCODE_H_
+#ifndef WEBP_UTILS_HUFFMAN_ENCODE_UTILS_H_
+#define WEBP_UTILS_HUFFMAN_ENCODE_UTILS_H_
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -57,4 +57,4 @@ void VP8LCreateHuffmanTree(uint32_t* const histogram, int tree_depth_limit,
 }
 #endif
 
-#endif  // WEBP_UTILS_HUFFMAN_ENCODE_H_
+#endif  // WEBP_UTILS_HUFFMAN_ENCODE_UTILS_H_
diff --git a/src/3rdparty/libwebp/src/utils/huffman_utils.c b/src/3rdparty/libwebp/src/utils/huffman_utils.c
index 008b5d7..7a69963 100644
--- a/src/3rdparty/libwebp/src/utils/huffman_utils.c
+++ b/src/3rdparty/libwebp/src/utils/huffman_utils.c
@@ -14,9 +14,9 @@
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
-#include "./huffman_utils.h"
-#include "./utils.h"
-#include "../webp/format_constants.h"
+#include "src/utils/huffman_utils.h"
+#include "src/utils/utils.h"
+#include "src/webp/format_constants.h"
 
 // Huffman data read via DecodeImageStream is represented in two (red and green)
 // bytes.
diff --git a/src/3rdparty/libwebp/src/utils/huffman_utils.h b/src/3rdparty/libwebp/src/utils/huffman_utils.h
index c6dd6aa..ff7ef17 100644
--- a/src/3rdparty/libwebp/src/utils/huffman_utils.h
+++ b/src/3rdparty/libwebp/src/utils/huffman_utils.h
@@ -11,12 +11,12 @@
 //
 // Author: Urvang Joshi (urvang@google.com)
 
-#ifndef WEBP_UTILS_HUFFMAN_H_
-#define WEBP_UTILS_HUFFMAN_H_
+#ifndef WEBP_UTILS_HUFFMAN_UTILS_H_
+#define WEBP_UTILS_HUFFMAN_UTILS_H_
 
 #include <assert.h>
-#include "../webp/format_constants.h"
-#include "../webp/types.h"
+#include "src/webp/format_constants.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -85,4 +85,4 @@ int VP8LBuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
 }    // extern "C"
 #endif
 
-#endif  // WEBP_UTILS_HUFFMAN_H_
+#endif  // WEBP_UTILS_HUFFMAN_UTILS_H_
diff --git a/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.c b/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.c
index d4d23d3..3818a78 100644
--- a/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.c
+++ b/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.c
@@ -14,11 +14,11 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./quant_levels_dec_utils.h"
+#include "src/utils/quant_levels_dec_utils.h"
 
 #include <string.h>   // for memset
 
-#include "./utils.h"
+#include "src/utils/utils.h"
 
 // #define USE_DITHERING   // uncomment to enable ordered dithering (not vital)
 
@@ -71,10 +71,11 @@ typedef struct {
 
 //------------------------------------------------------------------------------
 
-#define CLIP_MASK (int)(~0U << (8 + DFIX))
+#define CLIP_8b_MASK (int)(~0U << (8 + DFIX))
 static WEBP_INLINE uint8_t clip_8b(int v) {
-  return (!(v & CLIP_MASK)) ? (uint8_t)(v >> DFIX) : (v < 0) ? 0u : 255u;
+  return (!(v & CLIP_8b_MASK)) ? (uint8_t)(v >> DFIX) : (v < 0) ? 0u : 255u;
 }
+#undef CLIP_8b_MASK
 
 // vertical accumulation
 static void VFilter(SmoothParams* const p) {
diff --git a/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.h b/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.h
index 59a1349..f822107 100644
--- a/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.h
+++ b/src/3rdparty/libwebp/src/utils/quant_levels_dec_utils.h
@@ -11,10 +11,10 @@
 //
 // Author:  Vikas Arora (vikasa@google.com)
 
-#ifndef WEBP_UTILS_QUANT_LEVELS_DEC_H_
-#define WEBP_UTILS_QUANT_LEVELS_DEC_H_
+#ifndef WEBP_UTILS_QUANT_LEVELS_DEC_UTILS_H_
+#define WEBP_UTILS_QUANT_LEVELS_DEC_UTILS_H_
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -32,4 +32,4 @@ int WebPDequantizeLevels(uint8_t* const data, int width, int height, int stride,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_QUANT_LEVELS_DEC_H_ */
+#endif  /* WEBP_UTILS_QUANT_LEVELS_DEC_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/quant_levels_utils.c b/src/3rdparty/libwebp/src/utils/quant_levels_utils.c
index 73174e8..d65ad3c 100644
--- a/src/3rdparty/libwebp/src/utils/quant_levels_utils.c
+++ b/src/3rdparty/libwebp/src/utils/quant_levels_utils.c
@@ -14,7 +14,7 @@
 
 #include <assert.h>
 
-#include "./quant_levels_utils.h"
+#include "src/utils/quant_levels_utils.h"
 
 #define NUM_SYMBOLS     256
 
diff --git a/src/3rdparty/libwebp/src/utils/quant_levels_utils.h b/src/3rdparty/libwebp/src/utils/quant_levels_utils.h
index 1cb5a32..75df2ba 100644
--- a/src/3rdparty/libwebp/src/utils/quant_levels_utils.h
+++ b/src/3rdparty/libwebp/src/utils/quant_levels_utils.h
@@ -11,12 +11,12 @@
 //
 // Author:  Vikas Arora (vikasa@google.com)
 
-#ifndef WEBP_UTILS_QUANT_LEVELS_H_
-#define WEBP_UTILS_QUANT_LEVELS_H_
+#ifndef WEBP_UTILS_QUANT_LEVELS_UTILS_H_
+#define WEBP_UTILS_QUANT_LEVELS_UTILS_H_
 
 #include <stdlib.h>
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -33,4 +33,4 @@ int QuantizeLevels(uint8_t* const data, int width, int height, int num_levels,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_QUANT_LEVELS_H_ */
+#endif  /* WEBP_UTILS_QUANT_LEVELS_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/random_utils.c b/src/3rdparty/libwebp/src/utils/random_utils.c
index 9f1e415..7edb3fe 100644
--- a/src/3rdparty/libwebp/src/utils/random_utils.c
+++ b/src/3rdparty/libwebp/src/utils/random_utils.c
@@ -12,7 +12,7 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <string.h>
-#include "./random_utils.h"
+#include "src/utils/random_utils.h"
 
 //------------------------------------------------------------------------------
 
diff --git a/src/3rdparty/libwebp/src/utils/random_utils.h b/src/3rdparty/libwebp/src/utils/random_utils.h
index c392a61..6d36c66 100644
--- a/src/3rdparty/libwebp/src/utils/random_utils.h
+++ b/src/3rdparty/libwebp/src/utils/random_utils.h
@@ -11,11 +11,11 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_UTILS_RANDOM_H_
-#define WEBP_UTILS_RANDOM_H_
+#ifndef WEBP_UTILS_RANDOM_UTILS_H_
+#define WEBP_UTILS_RANDOM_UTILS_H_
 
 #include <assert.h>
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -60,4 +60,4 @@ static WEBP_INLINE int VP8RandomBits(VP8Random* const rg, int num_bits) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_RANDOM_H_ */
+#endif  /* WEBP_UTILS_RANDOM_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/rescaler_utils.c b/src/3rdparty/libwebp/src/utils/rescaler_utils.c
index 0d1f80d..90e2ea7 100644
--- a/src/3rdparty/libwebp/src/utils/rescaler_utils.c
+++ b/src/3rdparty/libwebp/src/utils/rescaler_utils.c
@@ -14,8 +14,8 @@
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
-#include "../dsp/dsp.h"
-#include "./rescaler_utils.h"
+#include "src/dsp/dsp.h"
+#include "src/utils/rescaler_utils.h"
 
 //------------------------------------------------------------------------------
 
@@ -85,11 +85,13 @@ int WebPRescalerGetScaledDimensions(int src_width, int src_height,
 
     // if width is unspecified, scale original proportionally to height ratio.
     if (width == 0) {
-      width = (src_width * height + src_height / 2) / src_height;
+      width =
+          (int)(((uint64_t)src_width * height + src_height / 2) / src_height);
     }
     // if height is unspecified, scale original proportionally to width ratio.
     if (height == 0) {
-      height = (src_height * width + src_width / 2) / src_width;
+      height =
+          (int)(((uint64_t)src_height * width + src_width / 2) / src_width);
     }
     // Check if the overall dimensions still make sense.
     if (width <= 0 || height <= 0) {
diff --git a/src/3rdparty/libwebp/src/utils/rescaler_utils.h b/src/3rdparty/libwebp/src/utils/rescaler_utils.h
index 98b01a7..8890e6f 100644
--- a/src/3rdparty/libwebp/src/utils/rescaler_utils.h
+++ b/src/3rdparty/libwebp/src/utils/rescaler_utils.h
@@ -11,14 +11,14 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_UTILS_RESCALER_H_
-#define WEBP_UTILS_RESCALER_H_
+#ifndef WEBP_UTILS_RESCALER_UTILS_H_
+#define WEBP_UTILS_RESCALER_UTILS_H_
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #define WEBP_RESCALER_RFIX 32   // fixed-point precision for multiplies
 #define WEBP_RESCALER_ONE (1ull << WEBP_RESCALER_RFIX)
@@ -98,4 +98,4 @@ int WebPRescalerHasPendingOutput(const WebPRescaler* const rescaler) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_RESCALER_H_ */
+#endif  /* WEBP_UTILS_RESCALER_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/thread_utils.c b/src/3rdparty/libwebp/src/utils/thread_utils.c
index 1729060..2052b6b 100644
--- a/src/3rdparty/libwebp/src/utils/thread_utils.c
+++ b/src/3rdparty/libwebp/src/utils/thread_utils.c
@@ -13,8 +13,8 @@
 
 #include <assert.h>
 #include <string.h>   // for memset()
-#include "./thread_utils.h"
-#include "./utils.h"
+#include "src/utils/thread_utils.h"
+#include "src/utils/utils.h"
 
 #ifdef WEBP_USE_THREAD
 
@@ -50,11 +50,11 @@ typedef struct {
 
 #endif  // _WIN32
 
-struct WebPWorkerImpl {
+typedef struct {
   pthread_mutex_t mutex_;
   pthread_cond_t  condition_;
   pthread_t       thread_;
-};
+} WebPWorkerImpl;
 
 #if defined(_WIN32)
 
@@ -201,25 +201,24 @@ static int pthread_cond_wait(pthread_cond_t* const condition,
 
 //------------------------------------------------------------------------------
 
-static void Execute(WebPWorker* const worker);  // Forward declaration.
-
 static THREADFN ThreadLoop(void* ptr) {
   WebPWorker* const worker = (WebPWorker*)ptr;
+  WebPWorkerImpl* const impl = (WebPWorkerImpl*)worker->impl_;
   int done = 0;
   while (!done) {
-    pthread_mutex_lock(&worker->impl_->mutex_);
+    pthread_mutex_lock(&impl->mutex_);
     while (worker->status_ == OK) {   // wait in idling mode
-      pthread_cond_wait(&worker->impl_->condition_, &worker->impl_->mutex_);
+      pthread_cond_wait(&impl->condition_, &impl->mutex_);
     }
     if (worker->status_ == WORK) {
-      Execute(worker);
+      WebPGetWorkerInterface()->Execute(worker);
       worker->status_ = OK;
     } else if (worker->status_ == NOT_OK) {   // finish the worker
       done = 1;
     }
     // signal to the main thread that we're done (for Sync())
-    pthread_cond_signal(&worker->impl_->condition_);
-    pthread_mutex_unlock(&worker->impl_->mutex_);
+    pthread_cond_signal(&impl->condition_);
+    pthread_mutex_unlock(&impl->mutex_);
   }
   return THREAD_RETURN(NULL);    // Thread is finished
 }
@@ -229,21 +228,22 @@ static void ChangeState(WebPWorker* const worker, WebPWorkerStatus new_status) {
   // No-op when attempting to change state on a thread that didn't come up.
   // Checking status_ without acquiring the lock first would result in a data
   // race.
-  if (worker->impl_ == NULL) return;
+  WebPWorkerImpl* const impl = (WebPWorkerImpl*)worker->impl_;
+  if (impl == NULL) return;
 
-  pthread_mutex_lock(&worker->impl_->mutex_);
+  pthread_mutex_lock(&impl->mutex_);
   if (worker->status_ >= OK) {
     // wait for the worker to finish
     while (worker->status_ != OK) {
-      pthread_cond_wait(&worker->impl_->condition_, &worker->impl_->mutex_);
+      pthread_cond_wait(&impl->condition_, &impl->mutex_);
     }
     // assign new status and release the working thread if needed
     if (new_status != OK) {
       worker->status_ = new_status;
-      pthread_cond_signal(&worker->impl_->condition_);
+      pthread_cond_signal(&impl->condition_);
     }
   }
-  pthread_mutex_unlock(&worker->impl_->mutex_);
+  pthread_mutex_unlock(&impl->mutex_);
 }
 
 #endif  // WEBP_USE_THREAD
@@ -268,26 +268,28 @@ static int Reset(WebPWorker* const worker) {
   worker->had_error = 0;
   if (worker->status_ < OK) {
 #ifdef WEBP_USE_THREAD
-    worker->impl_ = (WebPWorkerImpl*)WebPSafeCalloc(1, sizeof(*worker->impl_));
+    WebPWorkerImpl* const impl =
+        (WebPWorkerImpl*)WebPSafeCalloc(1, sizeof(WebPWorkerImpl));
+    worker->impl_ = (void*)impl;
     if (worker->impl_ == NULL) {
       return 0;
     }
-    if (pthread_mutex_init(&worker->impl_->mutex_, NULL)) {
+    if (pthread_mutex_init(&impl->mutex_, NULL)) {
       goto Error;
     }
-    if (pthread_cond_init(&worker->impl_->condition_, NULL)) {
-      pthread_mutex_destroy(&worker->impl_->mutex_);
+    if (pthread_cond_init(&impl->condition_, NULL)) {
+      pthread_mutex_destroy(&impl->mutex_);
       goto Error;
     }
-    pthread_mutex_lock(&worker->impl_->mutex_);
-    ok = !pthread_create(&worker->impl_->thread_, NULL, ThreadLoop, worker);
+    pthread_mutex_lock(&impl->mutex_);
+    ok = !pthread_create(&impl->thread_, NULL, ThreadLoop, worker);
     if (ok) worker->status_ = OK;
-    pthread_mutex_unlock(&worker->impl_->mutex_);
+    pthread_mutex_unlock(&impl->mutex_);
     if (!ok) {
-      pthread_mutex_destroy(&worker->impl_->mutex_);
-      pthread_cond_destroy(&worker->impl_->condition_);
+      pthread_mutex_destroy(&impl->mutex_);
+      pthread_cond_destroy(&impl->condition_);
  Error:
-      WebPSafeFree(worker->impl_);
+      WebPSafeFree(impl);
       worker->impl_ = NULL;
       return 0;
     }
@@ -318,11 +320,12 @@ static void Launch(WebPWorker* const worker) {
 static void End(WebPWorker* const worker) {
 #ifdef WEBP_USE_THREAD
   if (worker->impl_ != NULL) {
+    WebPWorkerImpl* const impl = (WebPWorkerImpl*)worker->impl_;
     ChangeState(worker, NOT_OK);
-    pthread_join(worker->impl_->thread_, NULL);
-    pthread_mutex_destroy(&worker->impl_->mutex_);
-    pthread_cond_destroy(&worker->impl_->condition_);
-    WebPSafeFree(worker->impl_);
+    pthread_join(impl->thread_, NULL);
+    pthread_mutex_destroy(&impl->mutex_);
+    pthread_cond_destroy(&impl->condition_);
+    WebPSafeFree(impl);
     worker->impl_ = NULL;
   }
 #else
diff --git a/src/3rdparty/libwebp/src/utils/thread_utils.h b/src/3rdparty/libwebp/src/utils/thread_utils.h
index 8408311..c8ae6c9 100644
--- a/src/3rdparty/libwebp/src/utils/thread_utils.h
+++ b/src/3rdparty/libwebp/src/utils/thread_utils.h
@@ -11,14 +11,14 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#ifndef WEBP_UTILS_THREAD_H_
-#define WEBP_UTILS_THREAD_H_
+#ifndef WEBP_UTILS_THREAD_UTILS_H_
+#define WEBP_UTILS_THREAD_UTILS_H_
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
-#include "../webp/types.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -35,12 +35,9 @@ typedef enum {
 // arguments (data1 and data2), and should return false in case of error.
 typedef int (*WebPWorkerHook)(void*, void*);
 
-// Platform-dependent implementation details for the worker.
-typedef struct WebPWorkerImpl WebPWorkerImpl;
-
 // Synchronization object used to launch job in the worker thread
 typedef struct {
-  WebPWorkerImpl* impl_;
+  void* impl_;            // platform-dependent implementation worker details
   WebPWorkerStatus status_;
   WebPWorkerHook hook;    // hook to call
   void* data1;            // first argument passed to 'hook'
@@ -78,11 +75,11 @@ typedef struct {
 // decoding takes place. The contents of the interface struct are copied, it
 // is safe to free the corresponding memory after this call. This function is
 // not thread-safe. Return false in case of invalid pointer or methods.
-WEBP_EXTERN(int) WebPSetWorkerInterface(
+WEBP_EXTERN int WebPSetWorkerInterface(
     const WebPWorkerInterface* const winterface);
 
 // Retrieve the currently set thread worker interface.
-WEBP_EXTERN(const WebPWorkerInterface*) WebPGetWorkerInterface(void);
+WEBP_EXTERN const WebPWorkerInterface* WebPGetWorkerInterface(void);
 
 //------------------------------------------------------------------------------
 
@@ -90,4 +87,4 @@ WEBP_EXTERN(const WebPWorkerInterface*) WebPGetWorkerInterface(void);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_THREAD_H_ */
+#endif  /* WEBP_UTILS_THREAD_UTILS_H_ */
diff --git a/src/3rdparty/libwebp/src/utils/utils.c b/src/3rdparty/libwebp/src/utils/utils.c
index 504d924..44d5c14 100644
--- a/src/3rdparty/libwebp/src/utils/utils.c
+++ b/src/3rdparty/libwebp/src/utils/utils.c
@@ -13,10 +13,11 @@
 
 #include <stdlib.h>
 #include <string.h>  // for memcpy()
-#include "../webp/decode.h"
-#include "../webp/encode.h"
-#include "../webp/format_constants.h"  // for MAX_PALETTE_SIZE
-#include "./utils.h"
+#include "src/webp/decode.h"
+#include "src/webp/encode.h"
+#include "src/webp/format_constants.h"  // for MAX_PALETTE_SIZE
+#include "src/utils/color_cache_utils.h"
+#include "src/utils/utils.h"
 
 // If PRINT_MEM_INFO is defined, extra info (like total memory used, number of
 // alloc/free etc) is printed. For debugging/tuning purpose only (it's slow,
@@ -252,7 +253,6 @@ int WebPGetColorPalette(const WebPPicture* const pic, uint32_t* const palette) {
   int num_colors = 0;
   uint8_t in_use[COLOR_HASH_SIZE] = { 0 };
   uint32_t colors[COLOR_HASH_SIZE];
-  static const uint64_t kHashMul = 0x1e35a7bdull;
   const uint32_t* argb = pic->argb;
   const int width = pic->width;
   const int height = pic->height;
@@ -267,7 +267,7 @@ int WebPGetColorPalette(const WebPPicture* const pic, uint32_t* const palette) {
         continue;
       }
       last_pix = argb[x];
-      key = ((last_pix * kHashMul) & 0xffffffffu) >> COLOR_HASH_RIGHT_SHIFT;
+      key = VP8LHashPix(last_pix, COLOR_HASH_RIGHT_SHIFT);
       while (1) {
         if (!in_use[key]) {
           colors[key] = last_pix;
diff --git a/src/3rdparty/libwebp/src/utils/utils.h b/src/3rdparty/libwebp/src/utils/utils.h
index 3ab4590..52921bf 100644
--- a/src/3rdparty/libwebp/src/utils/utils.h
+++ b/src/3rdparty/libwebp/src/utils/utils.h
@@ -16,14 +16,14 @@
 #define WEBP_UTILS_UTILS_H_
 
 #ifdef HAVE_CONFIG_H
-#include "../webp/config.h"
+#include "src/webp/config.h"
 #endif
 
 #include <assert.h>
 #include <limits.h>
 
-#include "../dsp/dsp.h"
-#include "../webp/types.h"
+#include "src/dsp/dsp.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -48,13 +48,13 @@ extern "C" {
 // somewhere (like: malloc(num_pixels * sizeof(*something))). That's why this
 // safe malloc() borrows the signature from calloc(), pointing at the dangerous
 // underlying multiply involved.
-WEBP_EXTERN(void*) WebPSafeMalloc(uint64_t nmemb, size_t size);
+WEBP_EXTERN void* WebPSafeMalloc(uint64_t nmemb, size_t size);
 // Note that WebPSafeCalloc() expects the second argument type to be 'size_t'
 // in order to favor the "calloc(num_foo, sizeof(foo))" pattern.
-WEBP_EXTERN(void*) WebPSafeCalloc(uint64_t nmemb, size_t size);
+WEBP_EXTERN void* WebPSafeCalloc(uint64_t nmemb, size_t size);
 
 // Companion deallocation function to the above allocations.
-WEBP_EXTERN(void) WebPSafeFree(void* const ptr);
+WEBP_EXTERN void WebPSafeFree(void* const ptr);
 
 //------------------------------------------------------------------------------
 // Alignment
@@ -66,7 +66,7 @@ WEBP_EXTERN(void) WebPSafeFree(void* const ptr);
 // memcpy() is the safe way of moving potentially unaligned 32b memory.
 static WEBP_INLINE uint32_t WebPMemToUint32(const uint8_t* const ptr) {
   uint32_t A;
-  memcpy(&A, (const int*)ptr, sizeof(A));
+  memcpy(&A, ptr, sizeof(A));
   return A;
 }
 static WEBP_INLINE void WebPUint32ToMem(uint8_t* const ptr, uint32_t val) {
@@ -112,12 +112,12 @@ static WEBP_INLINE void PutLE32(uint8_t* const data, uint32_t val) {
 #define WEBP_NEED_LOG_TABLE_8BIT
 extern const uint8_t WebPLogTable8bit[256];
 static WEBP_INLINE int WebPLog2FloorC(uint32_t n) {
-  int log = 0;
+  int log_value = 0;
   while (n >= 256) {
-    log += 8;
+    log_value += 8;
     n >>= 8;
   }
-  return log + WebPLogTable8bit[n];
+  return log_value + WebPLogTable8bit[n];
 }
 
 // Returns (int)floor(log2(n)). n must be > 0.
@@ -147,14 +147,14 @@ static WEBP_INLINE int BitsLog2Floor(uint32_t n) { return WebPLog2FloorC(n); }
 struct WebPPicture;
 
 // Copy width x height pixels from 'src' to 'dst' honoring the strides.
-WEBP_EXTERN(void) WebPCopyPlane(const uint8_t* src, int src_stride,
-                                uint8_t* dst, int dst_stride,
-                                int width, int height);
+WEBP_EXTERN void WebPCopyPlane(const uint8_t* src, int src_stride,
+                               uint8_t* dst, int dst_stride,
+                               int width, int height);
 
 // Copy ARGB pixels from 'src' to 'dst' honoring strides. 'src' and 'dst' are
 // assumed to be already allocated and using ARGB data.
-WEBP_EXTERN(void) WebPCopyPixels(const struct WebPPicture* const src,
-                                 struct WebPPicture* const dst);
+WEBP_EXTERN void WebPCopyPixels(const struct WebPPicture* const src,
+                                struct WebPPicture* const dst);
 
 //------------------------------------------------------------------------------
 // Unique colors.
@@ -166,8 +166,8 @@ WEBP_EXTERN(void) WebPCopyPixels(const struct WebPPicture* const src,
 // MAX_PALETTE_SIZE, also outputs the actual unique colors into 'palette'.
 // Note: 'palette' is assumed to be an array already allocated with at least
 // MAX_PALETTE_SIZE elements.
-WEBP_EXTERN(int) WebPGetColorPalette(const struct WebPPicture* const pic,
-                                     uint32_t* const palette);
+WEBP_EXTERN int WebPGetColorPalette(const struct WebPPicture* const pic,
+                                    uint32_t* const palette);
 
 //------------------------------------------------------------------------------
 
diff --git a/src/3rdparty/libwebp/src/webp/config.h b/src/3rdparty/libwebp/src/webp/config.h
index 731115b..5ca2e3d 100644
--- a/src/3rdparty/libwebp/src/webp/config.h
+++ b/src/3rdparty/libwebp/src/webp/config.h
@@ -79,7 +79,7 @@
 #define PACKAGE_NAME "libwebp"
 
 /* Define to the full name and version of this package. */
-#define PACKAGE_STRING "libwebp 0.6.0"
+#define PACKAGE_STRING "libwebp 0.6.1"
 
 /* Define to the one symbol short name of this package. */
 #define PACKAGE_TARNAME "libwebp"
@@ -88,7 +88,7 @@
 #define PACKAGE_URL "http://developers.google.com/speed/webp"
 
 /* Define to the version of this package. */
-#define PACKAGE_VERSION "0.6.0"
+#define PACKAGE_VERSION "0.6.1"
 
 /* Define to necessary symbol if this constant uses a non-standard name on
    your system. */
@@ -98,7 +98,7 @@
 /* #undef STDC_HEADERS */
 
 /* Version number of package */
-#define VERSION "0.6.0"
+#define VERSION "0.6.1"
 
 /* Enable experimental code */
 /* #undef WEBP_EXPERIMENTAL_FEATURES */
diff --git a/src/3rdparty/libwebp/src/webp/decode.h b/src/3rdparty/libwebp/src/webp/decode.h
index 4c5e74a..2165e96 100644
--- a/src/3rdparty/libwebp/src/webp/decode.h
+++ b/src/3rdparty/libwebp/src/webp/decode.h
@@ -36,39 +36,39 @@ typedef struct WebPDecoderConfig WebPDecoderConfig;
 
 // Return the decoder's version number, packed in hexadecimal using 8bits for
 // each of major/minor/revision. E.g: v2.5.7 is 0x020507.
-WEBP_EXTERN(int) WebPGetDecoderVersion(void);
+WEBP_EXTERN int WebPGetDecoderVersion(void);
 
 // Retrieve basic header information: width, height.
 // This function will also validate the header, returning true on success,
 // false otherwise. '*width' and '*height' are only valid on successful return.
 // Pointers 'width' and 'height' can be passed NULL if deemed irrelevant.
-WEBP_EXTERN(int) WebPGetInfo(const uint8_t* data, size_t data_size,
-                             int* width, int* height);
+WEBP_EXTERN int WebPGetInfo(const uint8_t* data, size_t data_size,
+                            int* width, int* height);
 
 // Decodes WebP images pointed to by 'data' and returns RGBA samples, along
 // with the dimensions in *width and *height. The ordering of samples in
 // memory is R, G, B, A, R, G, B, A... in scan order (endian-independent).
 // The returned pointer should be deleted calling WebPFree().
 // Returns NULL in case of error.
-WEBP_EXTERN(uint8_t*) WebPDecodeRGBA(const uint8_t* data, size_t data_size,
-                                     int* width, int* height);
+WEBP_EXTERN uint8_t* WebPDecodeRGBA(const uint8_t* data, size_t data_size,
+                                    int* width, int* height);
 
 // Same as WebPDecodeRGBA, but returning A, R, G, B, A, R, G, B... ordered data.
-WEBP_EXTERN(uint8_t*) WebPDecodeARGB(const uint8_t* data, size_t data_size,
-                                     int* width, int* height);
+WEBP_EXTERN uint8_t* WebPDecodeARGB(const uint8_t* data, size_t data_size,
+                                    int* width, int* height);
 
 // Same as WebPDecodeRGBA, but returning B, G, R, A, B, G, R, A... ordered data.
-WEBP_EXTERN(uint8_t*) WebPDecodeBGRA(const uint8_t* data, size_t data_size,
-                                     int* width, int* height);
+WEBP_EXTERN uint8_t* WebPDecodeBGRA(const uint8_t* data, size_t data_size,
+                                    int* width, int* height);
 
 // Same as WebPDecodeRGBA, but returning R, G, B, R, G, B... ordered data.
 // If the bitstream contains transparency, it is ignored.
-WEBP_EXTERN(uint8_t*) WebPDecodeRGB(const uint8_t* data, size_t data_size,
-                                    int* width, int* height);
+WEBP_EXTERN uint8_t* WebPDecodeRGB(const uint8_t* data, size_t data_size,
+                                   int* width, int* height);
 
 // Same as WebPDecodeRGB, but returning B, G, R, B, G, R... ordered data.
-WEBP_EXTERN(uint8_t*) WebPDecodeBGR(const uint8_t* data, size_t data_size,
-                                    int* width, int* height);
+WEBP_EXTERN uint8_t* WebPDecodeBGR(const uint8_t* data, size_t data_size,
+                                   int* width, int* height);
 
 
 // Decode WebP images pointed to by 'data' to Y'UV format(*). The pointer
@@ -80,13 +80,13 @@ WEBP_EXTERN(uint8_t*) WebPDecodeBGR(const uint8_t* data, size_t data_size,
 // have a common stride returned as '*uv_stride'.
 // Return NULL in case of error.
 // (*) Also named Y'CbCr. See: http://en.wikipedia.org/wiki/YCbCr
-WEBP_EXTERN(uint8_t*) WebPDecodeYUV(const uint8_t* data, size_t data_size,
-                                    int* width, int* height,
-                                    uint8_t** u, uint8_t** v,
-                                    int* stride, int* uv_stride);
+WEBP_EXTERN uint8_t* WebPDecodeYUV(const uint8_t* data, size_t data_size,
+                                   int* width, int* height,
+                                   uint8_t** u, uint8_t** v,
+                                   int* stride, int* uv_stride);
 
 // Releases memory returned by the WebPDecode*() functions above.
-WEBP_EXTERN(void) WebPFree(void* ptr);
+WEBP_EXTERN void WebPFree(void* ptr);
 
 // These five functions are variants of the above ones, that decode the image
 // directly into a pre-allocated buffer 'output_buffer'. The maximum storage
@@ -96,22 +96,22 @@ WEBP_EXTERN(void) WebPFree(void* ptr);
 // The parameter 'output_stride' specifies the distance (in bytes)
 // between scanlines. Hence, output_buffer_size is expected to be at least
 // output_stride x picture-height.
-WEBP_EXTERN(uint8_t*) WebPDecodeRGBAInto(
+WEBP_EXTERN uint8_t* WebPDecodeRGBAInto(
     const uint8_t* data, size_t data_size,
     uint8_t* output_buffer, size_t output_buffer_size, int output_stride);
-WEBP_EXTERN(uint8_t*) WebPDecodeARGBInto(
+WEBP_EXTERN uint8_t* WebPDecodeARGBInto(
     const uint8_t* data, size_t data_size,
     uint8_t* output_buffer, size_t output_buffer_size, int output_stride);
-WEBP_EXTERN(uint8_t*) WebPDecodeBGRAInto(
+WEBP_EXTERN uint8_t* WebPDecodeBGRAInto(
     const uint8_t* data, size_t data_size,
     uint8_t* output_buffer, size_t output_buffer_size, int output_stride);
 
 // RGB and BGR variants. Here too the transparency information, if present,
 // will be dropped and ignored.
-WEBP_EXTERN(uint8_t*) WebPDecodeRGBInto(
+WEBP_EXTERN uint8_t* WebPDecodeRGBInto(
     const uint8_t* data, size_t data_size,
     uint8_t* output_buffer, size_t output_buffer_size, int output_stride);
-WEBP_EXTERN(uint8_t*) WebPDecodeBGRInto(
+WEBP_EXTERN uint8_t* WebPDecodeBGRInto(
     const uint8_t* data, size_t data_size,
     uint8_t* output_buffer, size_t output_buffer_size, int output_stride);
 
@@ -122,7 +122,7 @@ WEBP_EXTERN(uint8_t*) WebPDecodeBGRInto(
 // 'u_size' and 'v_size' respectively.
 // Pointer to the luma plane ('*luma') is returned or NULL if an error occurred
 // during decoding (or because some buffers were found to be too small).
-WEBP_EXTERN(uint8_t*) WebPDecodeYUVInto(
+WEBP_EXTERN uint8_t* WebPDecodeYUVInto(
     const uint8_t* data, size_t data_size,
     uint8_t* luma, size_t luma_size, int luma_stride,
     uint8_t* u, size_t u_size, int u_stride,
@@ -213,7 +213,7 @@ struct WebPDecBuffer {
 };
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(int) WebPInitDecBufferInternal(WebPDecBuffer*, int);
+WEBP_EXTERN int WebPInitDecBufferInternal(WebPDecBuffer*, int);
 
 // Initialize the structure as empty. Must be called before any other use.
 // Returns false in case of version mismatch
@@ -223,7 +223,7 @@ static WEBP_INLINE int WebPInitDecBuffer(WebPDecBuffer* buffer) {
 
 // Free any memory associated with the buffer. Must always be called last.
 // Note: doesn't free the 'buffer' structure itself.
-WEBP_EXTERN(void) WebPFreeDecBuffer(WebPDecBuffer* buffer);
+WEBP_EXTERN void WebPFreeDecBuffer(WebPDecBuffer* buffer);
 
 //------------------------------------------------------------------------------
 // Enumeration of the status codes
@@ -277,7 +277,7 @@ typedef enum VP8StatusCode {
 // within valid bounds.
 // All other fields of WebPDecBuffer MUST remain constant between calls.
 // Returns NULL if the allocation failed.
-WEBP_EXTERN(WebPIDecoder*) WebPINewDecoder(WebPDecBuffer* output_buffer);
+WEBP_EXTERN WebPIDecoder* WebPINewDecoder(WebPDecBuffer* output_buffer);
 
 // This function allocates and initializes an incremental-decoder object, which
 // will output the RGB/A samples specified by 'csp' into a preallocated
@@ -289,7 +289,7 @@ WEBP_EXTERN(WebPIDecoder*) WebPINewDecoder(WebPDecBuffer* output_buffer);
 // colorspace 'csp' is taken into account for allocating this buffer. All other
 // parameters are ignored.
 // Returns NULL if the allocation failed, or if some parameters are invalid.
-WEBP_EXTERN(WebPIDecoder*) WebPINewRGB(
+WEBP_EXTERN WebPIDecoder* WebPINewRGB(
     WEBP_CSP_MODE csp,
     uint8_t* output_buffer, size_t output_buffer_size, int output_stride);
 
@@ -304,7 +304,7 @@ WEBP_EXTERN(WebPIDecoder*) WebPINewRGB(
 // In this case, the output buffer will be automatically allocated (using
 // MODE_YUVA) when decoding starts. All parameters are then ignored.
 // Returns NULL if the allocation failed or if a parameter is invalid.
-WEBP_EXTERN(WebPIDecoder*) WebPINewYUVA(
+WEBP_EXTERN WebPIDecoder* WebPINewYUVA(
     uint8_t* luma, size_t luma_size, int luma_stride,
     uint8_t* u, size_t u_size, int u_stride,
     uint8_t* v, size_t v_size, int v_stride,
@@ -312,19 +312,19 @@ WEBP_EXTERN(WebPIDecoder*) WebPINewYUVA(
 
 // Deprecated version of the above, without the alpha plane.
 // Kept for backward compatibility.
-WEBP_EXTERN(WebPIDecoder*) WebPINewYUV(
+WEBP_EXTERN WebPIDecoder* WebPINewYUV(
     uint8_t* luma, size_t luma_size, int luma_stride,
     uint8_t* u, size_t u_size, int u_stride,
     uint8_t* v, size_t v_size, int v_stride);
 
 // Deletes the WebPIDecoder object and associated memory. Must always be called
 // if WebPINewDecoder, WebPINewRGB or WebPINewYUV succeeded.
-WEBP_EXTERN(void) WebPIDelete(WebPIDecoder* idec);
+WEBP_EXTERN void WebPIDelete(WebPIDecoder* idec);
 
 // Copies and decodes the next available data. Returns VP8_STATUS_OK when
 // the image is successfully decoded. Returns VP8_STATUS_SUSPENDED when more
 // data is expected. Returns error in other cases.
-WEBP_EXTERN(VP8StatusCode) WebPIAppend(
+WEBP_EXTERN VP8StatusCode WebPIAppend(
     WebPIDecoder* idec, const uint8_t* data, size_t data_size);
 
 // A variant of the above function to be used when data buffer contains
@@ -332,7 +332,7 @@ WEBP_EXTERN(VP8StatusCode) WebPIAppend(
 // to the internal memory.
 // Note that the value of the 'data' pointer can change between calls to
 // WebPIUpdate, for instance when the data buffer is resized to fit larger data.
-WEBP_EXTERN(VP8StatusCode) WebPIUpdate(
+WEBP_EXTERN VP8StatusCode WebPIUpdate(
     WebPIDecoder* idec, const uint8_t* data, size_t data_size);
 
 // Returns the RGB/A image decoded so far. Returns NULL if output params
@@ -340,15 +340,16 @@ WEBP_EXTERN(VP8StatusCode) WebPIUpdate(
 // specified during call to WebPINewDecoder() or WebPINewRGB().
 // *last_y is the index of last decoded row in raster scan order. Some pointers
 // (*last_y, *width etc.) can be NULL if corresponding information is not
-// needed.
-WEBP_EXTERN(uint8_t*) WebPIDecGetRGB(
+// needed. The values in these pointers are only valid on successful (non-NULL)
+// return.
+WEBP_EXTERN uint8_t* WebPIDecGetRGB(
     const WebPIDecoder* idec, int* last_y,
     int* width, int* height, int* stride);
 
 // Same as above function to get a YUVA image. Returns pointer to the luma
 // plane or NULL in case of error. If there is no alpha information
 // the alpha pointer '*a' will be returned NULL.
-WEBP_EXTERN(uint8_t*) WebPIDecGetYUVA(
+WEBP_EXTERN uint8_t* WebPIDecGetYUVA(
     const WebPIDecoder* idec, int* last_y,
     uint8_t** u, uint8_t** v, uint8_t** a,
     int* width, int* height, int* stride, int* uv_stride, int* a_stride);
@@ -368,7 +369,7 @@ static WEBP_INLINE uint8_t* WebPIDecGetYUV(
 // Returns NULL in case the incremental decoder object is in an invalid state.
 // Otherwise returns the pointer to the internal representation. This structure
 // is read-only, tied to WebPIDecoder's lifespan and should not be modified.
-WEBP_EXTERN(const WebPDecBuffer*) WebPIDecodedArea(
+WEBP_EXTERN const WebPDecBuffer* WebPIDecodedArea(
     const WebPIDecoder* idec, int* left, int* top, int* width, int* height);
 
 //------------------------------------------------------------------------------
@@ -416,7 +417,7 @@ struct WebPBitstreamFeatures {
 };
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(VP8StatusCode) WebPGetFeaturesInternal(
+WEBP_EXTERN VP8StatusCode WebPGetFeaturesInternal(
     const uint8_t*, size_t, WebPBitstreamFeatures*, int);
 
 // Retrieve features from the bitstream. The *features structure is filled
@@ -457,7 +458,7 @@ struct WebPDecoderConfig {
 };
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(int) WebPInitDecoderConfigInternal(WebPDecoderConfig*, int);
+WEBP_EXTERN int WebPInitDecoderConfigInternal(WebPDecoderConfig*, int);
 
 // Initialize the configuration as empty. This function must always be
 // called first, unless WebPGetFeatures() is to be called.
@@ -477,14 +478,14 @@ static WEBP_INLINE int WebPInitDecoderConfig(WebPDecoderConfig* config) {
 // The return WebPIDecoder object must always be deleted calling WebPIDelete().
 // Returns NULL in case of error (and config->status will then reflect
 // the error condition, if available).
-WEBP_EXTERN(WebPIDecoder*) WebPIDecode(const uint8_t* data, size_t data_size,
-                                       WebPDecoderConfig* config);
+WEBP_EXTERN WebPIDecoder* WebPIDecode(const uint8_t* data, size_t data_size,
+                                      WebPDecoderConfig* config);
 
 // Non-incremental version. This version decodes the full data at once, taking
 // 'config' into account. Returns decoding status (which should be VP8_STATUS_OK
 // if the decoding was successful). Note that 'config' cannot be NULL.
-WEBP_EXTERN(VP8StatusCode) WebPDecode(const uint8_t* data, size_t data_size,
-                                      WebPDecoderConfig* config);
+WEBP_EXTERN VP8StatusCode WebPDecode(const uint8_t* data, size_t data_size,
+                                     WebPDecoderConfig* config);
 
 #ifdef __cplusplus
 }    // extern "C"
diff --git a/src/3rdparty/libwebp/src/webp/demux.h b/src/3rdparty/libwebp/src/webp/demux.h
index 454f691..555d641 100644
--- a/src/3rdparty/libwebp/src/webp/demux.h
+++ b/src/3rdparty/libwebp/src/webp/demux.h
@@ -71,7 +71,7 @@ typedef struct WebPAnimDecoderOptions WebPAnimDecoderOptions;
 
 // Returns the version number of the demux library, packed in hexadecimal using
 // 8bits for each of major/minor/revision. E.g: v2.5.7 is 0x020507.
-WEBP_EXTERN(int) WebPGetDemuxVersion(void);
+WEBP_EXTERN int WebPGetDemuxVersion(void);
 
 //------------------------------------------------------------------------------
 // Life of a Demux object
@@ -85,7 +85,7 @@ typedef enum WebPDemuxState {
 } WebPDemuxState;
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(WebPDemuxer*) WebPDemuxInternal(
+WEBP_EXTERN WebPDemuxer* WebPDemuxInternal(
     const WebPData*, int, WebPDemuxState*, int);
 
 // Parses the full WebP file given by 'data'. For single images the WebP file
@@ -109,27 +109,32 @@ static WEBP_INLINE WebPDemuxer* WebPDemuxPartial(
 }
 
 // Frees memory associated with 'dmux'.
-WEBP_EXTERN(void) WebPDemuxDelete(WebPDemuxer* dmux);
+WEBP_EXTERN void WebPDemuxDelete(WebPDemuxer* dmux);
 
 //------------------------------------------------------------------------------
 // Data/information extraction.
 
 typedef enum WebPFormatFeature {
-  WEBP_FF_FORMAT_FLAGS,  // Extended format flags present in the 'VP8X' chunk.
+  WEBP_FF_FORMAT_FLAGS,      // bit-wise combination of WebPFeatureFlags
+                             // corresponding to the 'VP8X' chunk (if present).
   WEBP_FF_CANVAS_WIDTH,
   WEBP_FF_CANVAS_HEIGHT,
-  WEBP_FF_LOOP_COUNT,
-  WEBP_FF_BACKGROUND_COLOR,
-  WEBP_FF_FRAME_COUNT    // Number of frames present in the demux object.
-                         // In case of a partial demux, this is the number of
-                         // frames seen so far, with the last frame possibly
-                         // being partial.
+  WEBP_FF_LOOP_COUNT,        // only relevant for animated file
+  WEBP_FF_BACKGROUND_COLOR,  // idem.
+  WEBP_FF_FRAME_COUNT        // Number of frames present in the demux object.
+                             // In case of a partial demux, this is the number
+                             // of frames seen so far, with the last frame
+                             // possibly being partial.
 } WebPFormatFeature;
 
 // Get the 'feature' value from the 'dmux'.
 // NOTE: values are only valid if WebPDemux() was used or WebPDemuxPartial()
 // returned a state > WEBP_DEMUX_PARSING_HEADER.
-WEBP_EXTERN(uint32_t) WebPDemuxGetI(
+// If 'feature' is WEBP_FF_FORMAT_FLAGS, the returned value is a bit-wise
+// combination of WebPFeatureFlags values.
+// If 'feature' is WEBP_FF_LOOP_COUNT, WEBP_FF_BACKGROUND_COLOR, the returned
+// value is only meaningful if the bitstream is animated.
+WEBP_EXTERN uint32_t WebPDemuxGetI(
     const WebPDemuxer* dmux, WebPFormatFeature feature);
 
 //------------------------------------------------------------------------------
@@ -159,20 +164,20 @@ struct WebPIterator {
 // Returns false if 'dmux' is NULL or frame 'frame_number' is not present.
 // Call WebPDemuxReleaseIterator() when use of the iterator is complete.
 // NOTE: 'dmux' must persist for the lifetime of 'iter'.
-WEBP_EXTERN(int) WebPDemuxGetFrame(
+WEBP_EXTERN int WebPDemuxGetFrame(
     const WebPDemuxer* dmux, int frame_number, WebPIterator* iter);
 
 // Sets 'iter->fragment' to point to the next ('iter->frame_num' + 1) or
 // previous ('iter->frame_num' - 1) frame. These functions do not loop.
 // Returns true on success, false otherwise.
-WEBP_EXTERN(int) WebPDemuxNextFrame(WebPIterator* iter);
-WEBP_EXTERN(int) WebPDemuxPrevFrame(WebPIterator* iter);
+WEBP_EXTERN int WebPDemuxNextFrame(WebPIterator* iter);
+WEBP_EXTERN int WebPDemuxPrevFrame(WebPIterator* iter);
 
 // Releases any memory associated with 'iter'.
 // Must be called before any subsequent calls to WebPDemuxGetChunk() on the same
 // iter. Also, must be called before destroying the associated WebPDemuxer with
 // WebPDemuxDelete().
-WEBP_EXTERN(void) WebPDemuxReleaseIterator(WebPIterator* iter);
+WEBP_EXTERN void WebPDemuxReleaseIterator(WebPIterator* iter);
 
 //------------------------------------------------------------------------------
 // Chunk iteration.
@@ -197,20 +202,20 @@ struct WebPChunkIterator {
 // payloads are accessed through WebPDemuxGetFrame() and related functions.
 // Call WebPDemuxReleaseChunkIterator() when use of the iterator is complete.
 // NOTE: 'dmux' must persist for the lifetime of the iterator.
-WEBP_EXTERN(int) WebPDemuxGetChunk(const WebPDemuxer* dmux,
-                                   const char fourcc[4], int chunk_number,
-                                   WebPChunkIterator* iter);
+WEBP_EXTERN int WebPDemuxGetChunk(const WebPDemuxer* dmux,
+                                  const char fourcc[4], int chunk_number,
+                                  WebPChunkIterator* iter);
 
 // Sets 'iter->chunk' to point to the next ('iter->chunk_num' + 1) or previous
 // ('iter->chunk_num' - 1) chunk. These functions do not loop.
 // Returns true on success, false otherwise.
-WEBP_EXTERN(int) WebPDemuxNextChunk(WebPChunkIterator* iter);
-WEBP_EXTERN(int) WebPDemuxPrevChunk(WebPChunkIterator* iter);
+WEBP_EXTERN int WebPDemuxNextChunk(WebPChunkIterator* iter);
+WEBP_EXTERN int WebPDemuxPrevChunk(WebPChunkIterator* iter);
 
 // Releases any memory associated with 'iter'.
 // Must be called before destroying the associated WebPDemuxer with
 // WebPDemuxDelete().
-WEBP_EXTERN(void) WebPDemuxReleaseChunkIterator(WebPChunkIterator* iter);
+WEBP_EXTERN void WebPDemuxReleaseChunkIterator(WebPChunkIterator* iter);
 
 //------------------------------------------------------------------------------
 // WebPAnimDecoder API
@@ -252,7 +257,7 @@ struct WebPAnimDecoderOptions {
 };
 
 // Internal, version-checked, entry point.
-WEBP_EXTERN(int) WebPAnimDecoderOptionsInitInternal(
+WEBP_EXTERN int WebPAnimDecoderOptionsInitInternal(
     WebPAnimDecoderOptions*, int);
 
 // Should always be called, to initialize a fresh WebPAnimDecoderOptions
@@ -266,7 +271,7 @@ static WEBP_INLINE int WebPAnimDecoderOptionsInit(
 }
 
 // Internal, version-checked, entry point.
-WEBP_EXTERN(WebPAnimDecoder*) WebPAnimDecoderNewInternal(
+WEBP_EXTERN WebPAnimDecoder* WebPAnimDecoderNewInternal(
     const WebPData*, const WebPAnimDecoderOptions*, int);
 
 // Creates and initializes a WebPAnimDecoder object.
@@ -301,8 +306,8 @@ struct WebPAnimInfo {
 //   info - (out) global information fetched from the animation.
 // Returns:
 //   True on success.
-WEBP_EXTERN(int) WebPAnimDecoderGetInfo(const WebPAnimDecoder* dec,
-                                        WebPAnimInfo* info);
+WEBP_EXTERN int WebPAnimDecoderGetInfo(const WebPAnimDecoder* dec,
+                                       WebPAnimInfo* info);
 
 // Fetch the next frame from 'dec' based on options supplied to
 // WebPAnimDecoderNew(). This will be a fully reconstructed canvas of size
@@ -316,8 +321,8 @@ WEBP_EXTERN(int) WebPAnimDecoderGetInfo(const WebPAnimDecoder* dec,
 // Returns:
 //   False if any of the arguments are NULL, or if there is a parsing or
 //   decoding error, or if there are no more frames. Otherwise, returns true.
-WEBP_EXTERN(int) WebPAnimDecoderGetNext(WebPAnimDecoder* dec,
-                                        uint8_t** buf, int* timestamp);
+WEBP_EXTERN int WebPAnimDecoderGetNext(WebPAnimDecoder* dec,
+                                       uint8_t** buf, int* timestamp);
 
 // Check if there are more frames left to decode.
 // Parameters:
@@ -325,7 +330,7 @@ WEBP_EXTERN(int) WebPAnimDecoderGetNext(WebPAnimDecoder* dec,
 // Returns:
 //   True if 'dec' is not NULL and some frames are yet to be decoded.
 //   Otherwise, returns false.
-WEBP_EXTERN(int) WebPAnimDecoderHasMoreFrames(const WebPAnimDecoder* dec);
+WEBP_EXTERN int WebPAnimDecoderHasMoreFrames(const WebPAnimDecoder* dec);
 
 // Resets the WebPAnimDecoder object, so that next call to
 // WebPAnimDecoderGetNext() will restart decoding from 1st frame. This would be
@@ -333,7 +338,7 @@ WEBP_EXTERN(int) WebPAnimDecoderHasMoreFrames(const WebPAnimDecoder* dec);
 // info.loop_count times) without destroying and recreating the 'dec' object.
 // Parameters:
 //   dec - (in/out) decoder instance to be reset
-WEBP_EXTERN(void) WebPAnimDecoderReset(WebPAnimDecoder* dec);
+WEBP_EXTERN void WebPAnimDecoderReset(WebPAnimDecoder* dec);
 
 // Grab the internal demuxer object.
 // Getting the demuxer object can be useful if one wants to use operations only
@@ -343,13 +348,13 @@ WEBP_EXTERN(void) WebPAnimDecoderReset(WebPAnimDecoder* dec);
 //
 // Parameters:
 //   dec - (in) decoder instance from which the demuxer object is to be fetched.
-WEBP_EXTERN(const WebPDemuxer*) WebPAnimDecoderGetDemuxer(
+WEBP_EXTERN const WebPDemuxer* WebPAnimDecoderGetDemuxer(
     const WebPAnimDecoder* dec);
 
 // Deletes the WebPAnimDecoder object.
 // Parameters:
 //   dec - (in/out) decoder instance to be deleted
-WEBP_EXTERN(void) WebPAnimDecoderDelete(WebPAnimDecoder* dec);
+WEBP_EXTERN void WebPAnimDecoderDelete(WebPAnimDecoder* dec);
 
 #ifdef __cplusplus
 }    // extern "C"
diff --git a/src/3rdparty/libwebp/src/webp/encode.h b/src/3rdparty/libwebp/src/webp/encode.h
index 35fde1d..7ec3543 100644
--- a/src/3rdparty/libwebp/src/webp/encode.h
+++ b/src/3rdparty/libwebp/src/webp/encode.h
@@ -35,7 +35,7 @@ typedef struct WebPMemoryWriter WebPMemoryWriter;
 
 // Return the encoder's version number, packed in hexadecimal using 8bits for
 // each of major/minor/revision. E.g: v2.5.7 is 0x020507.
-WEBP_EXTERN(int) WebPGetEncoderVersion(void);
+WEBP_EXTERN int WebPGetEncoderVersion(void);
 
 //------------------------------------------------------------------------------
 // One-stop-shop call! No questions asked:
@@ -46,37 +46,37 @@ WEBP_EXTERN(int) WebPGetEncoderVersion(void);
 // These functions compress using the lossy format, and the quality_factor
 // can go from 0 (smaller output, lower quality) to 100 (best quality,
 // larger output).
-WEBP_EXTERN(size_t) WebPEncodeRGB(const uint8_t* rgb,
+WEBP_EXTERN size_t WebPEncodeRGB(const uint8_t* rgb,
+                                 int width, int height, int stride,
+                                 float quality_factor, uint8_t** output);
+WEBP_EXTERN size_t WebPEncodeBGR(const uint8_t* bgr,
+                                 int width, int height, int stride,
+                                 float quality_factor, uint8_t** output);
+WEBP_EXTERN size_t WebPEncodeRGBA(const uint8_t* rgba,
                                   int width, int height, int stride,
                                   float quality_factor, uint8_t** output);
-WEBP_EXTERN(size_t) WebPEncodeBGR(const uint8_t* bgr,
+WEBP_EXTERN size_t WebPEncodeBGRA(const uint8_t* bgra,
                                   int width, int height, int stride,
                                   float quality_factor, uint8_t** output);
-WEBP_EXTERN(size_t) WebPEncodeRGBA(const uint8_t* rgba,
-                                   int width, int height, int stride,
-                                   float quality_factor, uint8_t** output);
-WEBP_EXTERN(size_t) WebPEncodeBGRA(const uint8_t* bgra,
-                                   int width, int height, int stride,
-                                   float quality_factor, uint8_t** output);
 
 // These functions are the equivalent of the above, but compressing in a
 // lossless manner. Files are usually larger than lossy format, but will
 // not suffer any compression loss.
-WEBP_EXTERN(size_t) WebPEncodeLosslessRGB(const uint8_t* rgb,
+WEBP_EXTERN size_t WebPEncodeLosslessRGB(const uint8_t* rgb,
+                                         int width, int height, int stride,
+                                         uint8_t** output);
+WEBP_EXTERN size_t WebPEncodeLosslessBGR(const uint8_t* bgr,
+                                         int width, int height, int stride,
+                                         uint8_t** output);
+WEBP_EXTERN size_t WebPEncodeLosslessRGBA(const uint8_t* rgba,
                                           int width, int height, int stride,
                                           uint8_t** output);
-WEBP_EXTERN(size_t) WebPEncodeLosslessBGR(const uint8_t* bgr,
+WEBP_EXTERN size_t WebPEncodeLosslessBGRA(const uint8_t* bgra,
                                           int width, int height, int stride,
                                           uint8_t** output);
-WEBP_EXTERN(size_t) WebPEncodeLosslessRGBA(const uint8_t* rgba,
-                                           int width, int height, int stride,
-                                           uint8_t** output);
-WEBP_EXTERN(size_t) WebPEncodeLosslessBGRA(const uint8_t* bgra,
-                                           int width, int height, int stride,
-                                           uint8_t** output);
 
 // Releases memory returned by the WebPEncode*() functions above.
-WEBP_EXTERN(void) WebPFree(void* ptr);
+WEBP_EXTERN void WebPFree(void* ptr);
 
 //------------------------------------------------------------------------------
 // Coding parameters
@@ -93,12 +93,15 @@ typedef enum WebPImageHint {
 // Compression parameters.
 struct WebPConfig {
   int lossless;           // Lossless encoding (0=lossy(default), 1=lossless).
-  float quality;          // between 0 (smallest file) and 100 (biggest)
+  float quality;          // between 0 and 100. For lossy, 0 gives the smallest
+                          // size and 100 the largest. For lossless, this
+                          // parameter is the amount of effort put into the
+                          // compression: 0 is the fastest but gives larger
+                          // files compared to the slowest, but best, 100.
   int method;             // quality/speed trade-off (0=fast, 6=slower-better)
 
   WebPImageHint image_hint;  // Hint for image type (lossless only for now).
 
-  // Parameters related to lossy compression only:
   int target_size;        // if non-zero, set the desired target size in bytes.
                           // Takes precedence over the 'compression' parameter.
   float target_PSNR;      // if non-zero, specifies the minimal distortion to
@@ -159,7 +162,7 @@ typedef enum WebPPreset {
 } WebPPreset;
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(int) WebPConfigInitInternal(WebPConfig*, WebPPreset, float, int);
+WEBP_EXTERN int WebPConfigInitInternal(WebPConfig*, WebPPreset, float, int);
 
 // Should always be called, to initialize a fresh WebPConfig structure before
 // modification. Returns false in case of version mismatch. WebPConfigInit()
@@ -186,15 +189,15 @@ static WEBP_INLINE int WebPConfigPreset(WebPConfig* config,
 // speed and final compressed size.
 // This function will overwrite several fields from config: 'method', 'quality'
 // and 'lossless'. Returns false in case of parameter error.
-WEBP_EXTERN(int) WebPConfigLosslessPreset(WebPConfig* config, int level);
+WEBP_EXTERN int WebPConfigLosslessPreset(WebPConfig* config, int level);
 
 // Returns true if 'config' is non-NULL and all configuration parameters are
 // within their valid ranges.
-WEBP_EXTERN(int) WebPValidateConfig(const WebPConfig* config);
+WEBP_EXTERN int WebPValidateConfig(const WebPConfig* config);
 
 //------------------------------------------------------------------------------
 // Input / Output
-// Structure for storing auxiliary statistics (mostly for lossy encoding).
+// Structure for storing auxiliary statistics.
 
 struct WebPAuxStats {
   int coded_size;         // final size
@@ -242,16 +245,16 @@ struct WebPMemoryWriter {
 };
 
 // The following must be called first before any use.
-WEBP_EXTERN(void) WebPMemoryWriterInit(WebPMemoryWriter* writer);
+WEBP_EXTERN void WebPMemoryWriterInit(WebPMemoryWriter* writer);
 
 // The following must be called to deallocate writer->mem memory. The 'writer'
 // object itself is not deallocated.
-WEBP_EXTERN(void) WebPMemoryWriterClear(WebPMemoryWriter* writer);
+WEBP_EXTERN void WebPMemoryWriterClear(WebPMemoryWriter* writer);
 // The custom writer to be used with WebPMemoryWriter as custom_ptr. Upon
 // completion, writer.mem and writer.size will hold the coded data.
 // writer.mem must be freed by calling WebPMemoryWriterClear.
-WEBP_EXTERN(int) WebPMemoryWrite(const uint8_t* data, size_t data_size,
-                                 const WebPPicture* picture);
+WEBP_EXTERN int WebPMemoryWrite(const uint8_t* data, size_t data_size,
+                                const WebPPicture* picture);
 
 // Progress hook, called from time to time to report progress. It can return
 // false to request an abort of the encoding process, or true otherwise if
@@ -354,7 +357,7 @@ struct WebPPicture {
 };
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(int) WebPPictureInitInternal(WebPPicture*, int);
+WEBP_EXTERN int WebPPictureInitInternal(WebPPicture*, int);
 
 // Should always be called, to initialize the structure. Returns false in case
 // of version mismatch. WebPPictureInit() must have succeeded before using the
@@ -371,20 +374,20 @@ static WEBP_INLINE int WebPPictureInit(WebPPicture* picture) {
 // Allocate y/u/v buffers as per colorspace/width/height specification.
 // Note! This function will free the previous buffer if needed.
 // Returns false in case of memory error.
-WEBP_EXTERN(int) WebPPictureAlloc(WebPPicture* picture);
+WEBP_EXTERN int WebPPictureAlloc(WebPPicture* picture);
 
 // Release the memory allocated by WebPPictureAlloc() or WebPPictureImport*().
 // Note that this function does _not_ free the memory used by the 'picture'
 // object itself.
 // Besides memory (which is reclaimed) all other fields of 'picture' are
 // preserved.
-WEBP_EXTERN(void) WebPPictureFree(WebPPicture* picture);
+WEBP_EXTERN void WebPPictureFree(WebPPicture* picture);
 
 // Copy the pixels of *src into *dst, using WebPPictureAlloc. Upon return, *dst
 // will fully own the copied pixels (this is not a view). The 'dst' picture need
 // not be initialized as its content is overwritten.
 // Returns false in case of memory allocation error.
-WEBP_EXTERN(int) WebPPictureCopy(const WebPPicture* src, WebPPicture* dst);
+WEBP_EXTERN int WebPPictureCopy(const WebPPicture* src, WebPPicture* dst);
 
 // Compute the single distortion for packed planes of samples.
 // 'src' will be compared to 'ref', and the raw distortion stored into
@@ -393,19 +396,19 @@ WEBP_EXTERN(int) WebPPictureCopy(const WebPPicture* src, WebPPicture* dst);
 // 'x_step' is the horizontal stride (in bytes) between samples.
 // 'src/ref_stride' is the byte distance between rows.
 // Returns false in case of error (bad parameter, memory allocation error, ...).
-WEBP_EXTERN(int) WebPPlaneDistortion(const uint8_t* src, size_t src_stride,
-                                     const uint8_t* ref, size_t ref_stride,
-                                     int width, int height,
-                                     size_t x_step,
-                                     int type,   // 0 = PSNR, 1 = SSIM, 2 = LSIM
-                                     float* distortion, float* result);
+WEBP_EXTERN int WebPPlaneDistortion(const uint8_t* src, size_t src_stride,
+                                    const uint8_t* ref, size_t ref_stride,
+                                    int width, int height,
+                                    size_t x_step,
+                                    int type,   // 0 = PSNR, 1 = SSIM, 2 = LSIM
+                                    float* distortion, float* result);
 
 // Compute PSNR, SSIM or LSIM distortion metric between two pictures. Results
 // are in dB, stored in result[] in the B/G/R/A/All order. The distortion is
 // always performed using ARGB samples. Hence if the input is YUV(A), the
 // picture will be internally converted to ARGB (just for the measurement).
 // Warning: this function is rather CPU-intensive.
-WEBP_EXTERN(int) WebPPictureDistortion(
+WEBP_EXTERN int WebPPictureDistortion(
     const WebPPicture* src, const WebPPicture* ref,
     int metric_type,           // 0 = PSNR, 1 = SSIM, 2 = LSIM
     float result[5]);
@@ -418,8 +421,8 @@ WEBP_EXTERN(int) WebPPictureDistortion(
 // must be fully be comprised inside the 'src' source picture. If the source
 // picture uses the YUV420 colorspace, the top and left coordinates will be
 // snapped to even values.
-WEBP_EXTERN(int) WebPPictureCrop(WebPPicture* picture,
-                                 int left, int top, int width, int height);
+WEBP_EXTERN int WebPPictureCrop(WebPPicture* picture,
+                                int left, int top, int width, int height);
 
 // Extracts a view from 'src' picture into 'dst'. The rectangle for the view
 // is defined by the top-left corner pixel coordinates (left, top) as well
@@ -432,42 +435,42 @@ WEBP_EXTERN(int) WebPPictureCrop(WebPPicture* picture,
 // with WebPPictureInit() if it is different from 'src', since its content will
 // be overwritten.
 // Returns false in case of memory allocation error or invalid parameters.
-WEBP_EXTERN(int) WebPPictureView(const WebPPicture* src,
-                                 int left, int top, int width, int height,
-                                 WebPPicture* dst);
+WEBP_EXTERN int WebPPictureView(const WebPPicture* src,
+                                int left, int top, int width, int height,
+                                WebPPicture* dst);
 
 // Returns true if the 'picture' is actually a view and therefore does
 // not own the memory for pixels.
-WEBP_EXTERN(int) WebPPictureIsView(const WebPPicture* picture);
+WEBP_EXTERN int WebPPictureIsView(const WebPPicture* picture);
 
 // Rescale a picture to new dimension width x height.
 // If either 'width' or 'height' (but not both) is 0 the corresponding
 // dimension will be calculated preserving the aspect ratio.
 // No gamma correction is applied.
 // Returns false in case of error (invalid parameter or insufficient memory).
-WEBP_EXTERN(int) WebPPictureRescale(WebPPicture* pic, int width, int height);
+WEBP_EXTERN int WebPPictureRescale(WebPPicture* pic, int width, int height);
 
 // Colorspace conversion function to import RGB samples.
 // Previous buffer will be free'd, if any.
 // *rgb buffer should have a size of at least height * rgb_stride.
 // Returns false in case of memory error.
-WEBP_EXTERN(int) WebPPictureImportRGB(
+WEBP_EXTERN int WebPPictureImportRGB(
     WebPPicture* picture, const uint8_t* rgb, int rgb_stride);
 // Same, but for RGBA buffer.
-WEBP_EXTERN(int) WebPPictureImportRGBA(
+WEBP_EXTERN int WebPPictureImportRGBA(
     WebPPicture* picture, const uint8_t* rgba, int rgba_stride);
 // Same, but for RGBA buffer. Imports the RGB direct from the 32-bit format
 // input buffer ignoring the alpha channel. Avoids needing to copy the data
 // to a temporary 24-bit RGB buffer to import the RGB only.
-WEBP_EXTERN(int) WebPPictureImportRGBX(
+WEBP_EXTERN int WebPPictureImportRGBX(
     WebPPicture* picture, const uint8_t* rgbx, int rgbx_stride);
 
 // Variants of the above, but taking BGR(A|X) input.
-WEBP_EXTERN(int) WebPPictureImportBGR(
+WEBP_EXTERN int WebPPictureImportBGR(
     WebPPicture* picture, const uint8_t* bgr, int bgr_stride);
-WEBP_EXTERN(int) WebPPictureImportBGRA(
+WEBP_EXTERN int WebPPictureImportBGRA(
     WebPPicture* picture, const uint8_t* bgra, int bgra_stride);
-WEBP_EXTERN(int) WebPPictureImportBGRX(
+WEBP_EXTERN int WebPPictureImportBGRX(
     WebPPicture* picture, const uint8_t* bgrx, int bgrx_stride);
 
 // Converts picture->argb data to the YUV420A format. The 'colorspace'
@@ -476,14 +479,14 @@ WEBP_EXTERN(int) WebPPictureImportBGRX(
 // non-opaque transparent values is detected, and 'colorspace' will be
 // adjusted accordingly. Note that this method is lossy.
 // Returns false in case of error.
-WEBP_EXTERN(int) WebPPictureARGBToYUVA(WebPPicture* picture,
-                                       WebPEncCSP /*colorspace = WEBP_YUV420*/);
+WEBP_EXTERN int WebPPictureARGBToYUVA(WebPPicture* picture,
+                                      WebPEncCSP /*colorspace = WEBP_YUV420*/);
 
 // Same as WebPPictureARGBToYUVA(), but the conversion is done using
 // pseudo-random dithering with a strength 'dithering' between
 // 0.0 (no dithering) and 1.0 (maximum dithering). This is useful
 // for photographic picture.
-WEBP_EXTERN(int) WebPPictureARGBToYUVADithered(
+WEBP_EXTERN int WebPPictureARGBToYUVADithered(
     WebPPicture* picture, WebPEncCSP colorspace, float dithering);
 
 // Performs 'sharp' RGBA->YUVA420 downsampling and colorspace conversion.
@@ -491,9 +494,9 @@ WEBP_EXTERN(int) WebPPictureARGBToYUVADithered(
 // method is roughly 2x slower than WebPPictureARGBToYUVA() but produces better
 // and sharper YUV representation.
 // Returns false in case of error.
-WEBP_EXTERN(int) WebPPictureSharpARGBToYUVA(WebPPicture* picture);
+WEBP_EXTERN int WebPPictureSharpARGBToYUVA(WebPPicture* picture);
 // kept for backward compatibility:
-WEBP_EXTERN(int) WebPPictureSmartARGBToYUVA(WebPPicture* picture);
+WEBP_EXTERN int WebPPictureSmartARGBToYUVA(WebPPicture* picture);
 
 // Converts picture->yuv to picture->argb and sets picture->use_argb to true.
 // The input format must be YUV_420 or YUV_420A. The conversion from YUV420 to
@@ -501,22 +504,22 @@ WEBP_EXTERN(int) WebPPictureSmartARGBToYUVA(WebPPicture* picture);
 // Note that the use of this colorspace is discouraged if one has access to the
 // raw ARGB samples, since using YUV420 is comparatively lossy.
 // Returns false in case of error.
-WEBP_EXTERN(int) WebPPictureYUVAToARGB(WebPPicture* picture);
+WEBP_EXTERN int WebPPictureYUVAToARGB(WebPPicture* picture);
 
 // Helper function: given a width x height plane of RGBA or YUV(A) samples
-// clean-up the YUV or RGB samples under fully transparent area, to help
-// compressibility (no guarantee, though).
-WEBP_EXTERN(void) WebPCleanupTransparentArea(WebPPicture* picture);
+// clean-up or smoothen the YUV or RGB samples under fully transparent area,
+// to help compressibility (no guarantee, though).
+WEBP_EXTERN void WebPCleanupTransparentArea(WebPPicture* picture);
 
 // Scan the picture 'picture' for the presence of non fully opaque alpha values.
 // Returns true in such case. Otherwise returns false (indicating that the
 // alpha plane can be ignored altogether e.g.).
-WEBP_EXTERN(int) WebPPictureHasTransparency(const WebPPicture* picture);
+WEBP_EXTERN int WebPPictureHasTransparency(const WebPPicture* picture);
 
 // Remove the transparency information (if present) by blending the color with
 // the background color 'background_rgb' (specified as 24bit RGB triplet).
 // After this call, all alpha values are reset to 0xff.
-WEBP_EXTERN(void) WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb);
+WEBP_EXTERN void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb);
 
 //------------------------------------------------------------------------------
 // Main call
@@ -531,7 +534,7 @@ WEBP_EXTERN(void) WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb);
 // the former for lossy encoding, and the latter for lossless encoding
 // (when config.lossless is true). Automatic conversion from one format to
 // another is provided but they both incur some loss.
-WEBP_EXTERN(int) WebPEncode(const WebPConfig* config, WebPPicture* picture);
+WEBP_EXTERN int WebPEncode(const WebPConfig* config, WebPPicture* picture);
 
 //------------------------------------------------------------------------------
 
diff --git a/src/3rdparty/libwebp/src/webp/mux.h b/src/3rdparty/libwebp/src/webp/mux.h
index daccc65..28bb4a4 100644
--- a/src/3rdparty/libwebp/src/webp/mux.h
+++ b/src/3rdparty/libwebp/src/webp/mux.h
@@ -98,13 +98,13 @@ typedef enum WebPChunkId {
 
 // Returns the version number of the mux library, packed in hexadecimal using
 // 8bits for each of major/minor/revision. E.g: v2.5.7 is 0x020507.
-WEBP_EXTERN(int) WebPGetMuxVersion(void);
+WEBP_EXTERN int WebPGetMuxVersion(void);
 
 //------------------------------------------------------------------------------
 // Life of a Mux object
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(WebPMux*) WebPNewInternal(int);
+WEBP_EXTERN WebPMux* WebPNewInternal(int);
 
 // Creates an empty mux object.
 // Returns:
@@ -117,13 +117,13 @@ static WEBP_INLINE WebPMux* WebPMuxNew(void) {
 // Deletes the mux object.
 // Parameters:
 //   mux - (in/out) object to be deleted
-WEBP_EXTERN(void) WebPMuxDelete(WebPMux* mux);
+WEBP_EXTERN void WebPMuxDelete(WebPMux* mux);
 
 //------------------------------------------------------------------------------
 // Mux creation.
 
 // Internal, version-checked, entry point
-WEBP_EXTERN(WebPMux*) WebPMuxCreateInternal(const WebPData*, int, int);
+WEBP_EXTERN WebPMux* WebPMuxCreateInternal(const WebPData*, int, int);
 
 // Creates a mux object from raw data given in WebP RIFF format.
 // Parameters:
@@ -160,7 +160,7 @@ static WEBP_INLINE WebPMux* WebPMuxCreate(const WebPData* bitstream,
 //                               or if fourcc corresponds to an image chunk.
 //   WEBP_MUX_MEMORY_ERROR - on memory allocation error.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxSetChunk(
+WEBP_EXTERN WebPMuxError WebPMuxSetChunk(
     WebPMux* mux, const char fourcc[4], const WebPData* chunk_data,
     int copy_data);
 
@@ -176,7 +176,7 @@ WEBP_EXTERN(WebPMuxError) WebPMuxSetChunk(
 //                               or if fourcc corresponds to an image chunk.
 //   WEBP_MUX_NOT_FOUND - If mux does not contain a chunk with the given id.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxGetChunk(
+WEBP_EXTERN WebPMuxError WebPMuxGetChunk(
     const WebPMux* mux, const char fourcc[4], WebPData* chunk_data);
 
 // Deletes the chunk with the given 'fourcc' from the mux object.
@@ -189,7 +189,7 @@ WEBP_EXTERN(WebPMuxError) WebPMuxGetChunk(
 //                               or if fourcc corresponds to an image chunk.
 //   WEBP_MUX_NOT_FOUND - If mux does not contain a chunk with the given fourcc.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxDeleteChunk(
+WEBP_EXTERN WebPMuxError WebPMuxDeleteChunk(
     WebPMux* mux, const char fourcc[4]);
 
 //------------------------------------------------------------------------------
@@ -222,7 +222,7 @@ struct WebPMuxFrameInfo {
 //   WEBP_MUX_INVALID_ARGUMENT - if mux is NULL or bitstream is NULL.
 //   WEBP_MUX_MEMORY_ERROR - on memory allocation error.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxSetImage(
+WEBP_EXTERN WebPMuxError WebPMuxSetImage(
     WebPMux* mux, const WebPData* bitstream, int copy_data);
 
 // Adds a frame at the end of the mux object.
@@ -241,7 +241,7 @@ WEBP_EXTERN(WebPMuxError) WebPMuxSetImage(
 //                               or if content of 'frame' is invalid.
 //   WEBP_MUX_MEMORY_ERROR - on memory allocation error.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxPushFrame(
+WEBP_EXTERN WebPMuxError WebPMuxPushFrame(
     WebPMux* mux, const WebPMuxFrameInfo* frame, int copy_data);
 
 // Gets the nth frame from the mux object.
@@ -259,7 +259,7 @@ WEBP_EXTERN(WebPMuxError) WebPMuxPushFrame(
 //   WEBP_MUX_BAD_DATA - if nth frame chunk in mux is invalid.
 //   WEBP_MUX_MEMORY_ERROR - on memory allocation error.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxGetFrame(
+WEBP_EXTERN WebPMuxError WebPMuxGetFrame(
     const WebPMux* mux, uint32_t nth, WebPMuxFrameInfo* frame);
 
 // Deletes a frame from the mux object.
@@ -272,7 +272,7 @@ WEBP_EXTERN(WebPMuxError) WebPMuxGetFrame(
 //   WEBP_MUX_NOT_FOUND - If there are less than nth frames in the mux object
 //                        before deletion.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxDeleteFrame(WebPMux* mux, uint32_t nth);
+WEBP_EXTERN WebPMuxError WebPMuxDeleteFrame(WebPMux* mux, uint32_t nth);
 
 //------------------------------------------------------------------------------
 // Animation.
@@ -296,7 +296,7 @@ struct WebPMuxAnimParams {
 //   WEBP_MUX_INVALID_ARGUMENT - if mux or params is NULL.
 //   WEBP_MUX_MEMORY_ERROR - on memory allocation error.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxSetAnimationParams(
+WEBP_EXTERN WebPMuxError WebPMuxSetAnimationParams(
     WebPMux* mux, const WebPMuxAnimParams* params);
 
 // Gets the animation parameters from the mux object.
@@ -307,7 +307,7 @@ WEBP_EXTERN(WebPMuxError) WebPMuxSetAnimationParams(
 //   WEBP_MUX_INVALID_ARGUMENT - if mux or params is NULL.
 //   WEBP_MUX_NOT_FOUND - if ANIM chunk is not present in mux object.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxGetAnimationParams(
+WEBP_EXTERN WebPMuxError WebPMuxGetAnimationParams(
     const WebPMux* mux, WebPMuxAnimParams* params);
 
 //------------------------------------------------------------------------------
@@ -328,8 +328,8 @@ WEBP_EXTERN(WebPMuxError) WebPMuxGetAnimationParams(
 //   WEBP_MUX_INVALID_ARGUMENT - if mux is NULL; or
 //                               width or height are invalid or out of bounds
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxSetCanvasSize(WebPMux* mux,
-                                               int width, int height);
+WEBP_EXTERN WebPMuxError WebPMuxSetCanvasSize(WebPMux* mux,
+                                              int width, int height);
 
 // Gets the canvas size from the mux object.
 // Note: This method assumes that the VP8X chunk, if present, is up-to-date.
@@ -343,8 +343,8 @@ WEBP_EXTERN(WebPMuxError) WebPMuxSetCanvasSize(WebPMux* mux,
 //   WEBP_MUX_INVALID_ARGUMENT - if mux, width or height is NULL.
 //   WEBP_MUX_BAD_DATA - if VP8X/VP8/VP8L chunk or canvas size is invalid.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxGetCanvasSize(const WebPMux* mux,
-                                               int* width, int* height);
+WEBP_EXTERN WebPMuxError WebPMuxGetCanvasSize(const WebPMux* mux,
+                                              int* width, int* height);
 
 // Gets the feature flags from the mux object.
 // Note: This method assumes that the VP8X chunk, if present, is up-to-date.
@@ -359,8 +359,8 @@ WEBP_EXTERN(WebPMuxError) WebPMuxGetCanvasSize(const WebPMux* mux,
 //   WEBP_MUX_INVALID_ARGUMENT - if mux or flags is NULL.
 //   WEBP_MUX_BAD_DATA - if VP8X/VP8/VP8L chunk or canvas size is invalid.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxGetFeatures(const WebPMux* mux,
-                                             uint32_t* flags);
+WEBP_EXTERN WebPMuxError WebPMuxGetFeatures(const WebPMux* mux,
+                                            uint32_t* flags);
 
 // Gets number of chunks with the given 'id' in the mux object.
 // Parameters:
@@ -370,8 +370,8 @@ WEBP_EXTERN(WebPMuxError) WebPMuxGetFeatures(const WebPMux* mux,
 // Returns:
 //   WEBP_MUX_INVALID_ARGUMENT - if mux, or num_elements is NULL.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxNumChunks(const WebPMux* mux,
-                                           WebPChunkId id, int* num_elements);
+WEBP_EXTERN WebPMuxError WebPMuxNumChunks(const WebPMux* mux,
+                                          WebPChunkId id, int* num_elements);
 
 // Assembles all chunks in WebP RIFF format and returns in 'assembled_data'.
 // This function also validates the mux object.
@@ -388,8 +388,8 @@ WEBP_EXTERN(WebPMuxError) WebPMuxNumChunks(const WebPMux* mux,
 //   WEBP_MUX_INVALID_ARGUMENT - if mux or assembled_data is NULL.
 //   WEBP_MUX_MEMORY_ERROR - on memory allocation error.
 //   WEBP_MUX_OK - on success.
-WEBP_EXTERN(WebPMuxError) WebPMuxAssemble(WebPMux* mux,
-                                          WebPData* assembled_data);
+WEBP_EXTERN WebPMuxError WebPMuxAssemble(WebPMux* mux,
+                                         WebPData* assembled_data);
 
 //------------------------------------------------------------------------------
 // WebPAnimEncoder API
@@ -442,7 +442,7 @@ struct WebPAnimEncoderOptions {
 };
 
 // Internal, version-checked, entry point.
-WEBP_EXTERN(int) WebPAnimEncoderOptionsInitInternal(
+WEBP_EXTERN int WebPAnimEncoderOptionsInitInternal(
     WebPAnimEncoderOptions*, int);
 
 // Should always be called, to initialize a fresh WebPAnimEncoderOptions
@@ -455,7 +455,7 @@ static WEBP_INLINE int WebPAnimEncoderOptionsInit(
 }
 
 // Internal, version-checked, entry point.
-WEBP_EXTERN(WebPAnimEncoder*) WebPAnimEncoderNewInternal(
+WEBP_EXTERN WebPAnimEncoder* WebPAnimEncoderNewInternal(
     int, int, const WebPAnimEncoderOptions*, int);
 
 // Creates and initializes a WebPAnimEncoder object.
@@ -490,7 +490,7 @@ static WEBP_INLINE WebPAnimEncoder* WebPAnimEncoderNew(
 // Returns:
 //   On error, returns false and frame->error_code is set appropriately.
 //   Otherwise, returns true.
-WEBP_EXTERN(int) WebPAnimEncoderAdd(
+WEBP_EXTERN int WebPAnimEncoderAdd(
     WebPAnimEncoder* enc, struct WebPPicture* frame, int timestamp_ms,
     const struct WebPConfig* config);
 
@@ -503,8 +503,8 @@ WEBP_EXTERN(int) WebPAnimEncoderAdd(
 //   webp_data - (out) generated WebP bitstream.
 // Returns:
 //   True on success.
-WEBP_EXTERN(int) WebPAnimEncoderAssemble(WebPAnimEncoder* enc,
-                                         WebPData* webp_data);
+WEBP_EXTERN int WebPAnimEncoderAssemble(WebPAnimEncoder* enc,
+                                        WebPData* webp_data);
 
 // Get error string corresponding to the most recent call using 'enc'. The
 // returned string is owned by 'enc' and is valid only until the next call to
@@ -514,12 +514,12 @@ WEBP_EXTERN(int) WebPAnimEncoderAssemble(WebPAnimEncoder* enc,
 // Returns:
 //   NULL if 'enc' is NULL. Otherwise, returns the error string if the last call
 //   to 'enc' had an error, or an empty string if the last call was a success.
-WEBP_EXTERN(const char*) WebPAnimEncoderGetError(WebPAnimEncoder* enc);
+WEBP_EXTERN const char* WebPAnimEncoderGetError(WebPAnimEncoder* enc);
 
 // Deletes the WebPAnimEncoder object.
 // Parameters:
 //   enc - (in/out) object to be deleted
-WEBP_EXTERN(void) WebPAnimEncoderDelete(WebPAnimEncoder* enc);
+WEBP_EXTERN void WebPAnimEncoderDelete(WebPAnimEncoder* enc);
 
 //------------------------------------------------------------------------------
 
diff --git a/src/3rdparty/libwebp/src/webp/types.h b/src/3rdparty/libwebp/src/webp/types.h
index 98fff35..989a763 100644
--- a/src/3rdparty/libwebp/src/webp/types.h
+++ b/src/3rdparty/libwebp/src/webp/types.h
@@ -40,9 +40,9 @@ typedef long long int int64_t;
 // This explicitly marks library functions and allows for changing the
 // signature for e.g., Windows DLL builds.
 # if defined(__GNUC__) && __GNUC__ >= 4
-#  define WEBP_EXTERN(type) extern __attribute__ ((visibility ("default"))) type
+#  define WEBP_EXTERN extern __attribute__ ((visibility ("default")))
 # else
-#  define WEBP_EXTERN(type) extern type
+#  define WEBP_EXTERN extern
 # endif  /* __GNUC__ >= 4 */
 #endif  /* WEBP_EXTERN */
author	Eirik Aavitsland <eirik.aavitsland@qt.io>	2018-04-12 14:25:27 +0200
committer	Eirik Aavitsland <eirik.aavitsland@qt.io>	2018-04-12 21:22:06 +0000
commit	b83a0a9460432abb82218da247710a1aaf321336 (patch)
tree	2bd4017481cc3644034d99f1649064434478c904
parent	4e10f35483db6d3396fe54dcf3d895d7850f8335 (diff)