Update bundled libwebp to version 1.3.1

[ChangeLog][Third-Party Code] Update bundled libwebp to version 1.3.1

Pick-to: 6.6 6.5 6.2 5.15
Change-Id: I04b4a9fdb8da7cd3627d4ffaa15326c85858c425
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>

22 files changed, 242 insertions, 83 deletions

diff --git a/src/3rdparty/libwebp/src/dsp/alpha_processing.c b/src/3rdparty/libwebp/src/dsp/alpha_processing.c
index 1892929..1d152f2 100644
--- a/src/3rdparty/libwebp/src/dsp/alpha_processing.c
+++ b/src/3rdparty/libwebp/src/dsp/alpha_processing.c
@@ -425,6 +425,7 @@ void (*WebPAlphaReplace)(uint32_t* src, int length, uint32_t color);
 //------------------------------------------------------------------------------
 // Init function
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPInitAlphaProcessingMIPSdspR2(void);
 extern void WebPInitAlphaProcessingSSE2(void);
 extern void WebPInitAlphaProcessingSSE41(void);
diff --git a/src/3rdparty/libwebp/src/dsp/cost.c b/src/3rdparty/libwebp/src/dsp/cost.c
index 460ec4f..73d2140 100644
--- a/src/3rdparty/libwebp/src/dsp/cost.c
+++ b/src/3rdparty/libwebp/src/dsp/cost.c
@@ -374,6 +374,7 @@ static void SetResidualCoeffs_C(const int16_t* const coeffs,
 VP8GetResidualCostFunc VP8GetResidualCost;
 VP8SetResidualCoeffsFunc VP8SetResidualCoeffs;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8EncDspCostInitMIPS32(void);
 extern void VP8EncDspCostInitMIPSdspR2(void);
 extern void VP8EncDspCostInitSSE2(void);
diff --git a/src/3rdparty/libwebp/src/dsp/cost_neon.c b/src/3rdparty/libwebp/src/dsp/cost_neon.c
index 8cc8ce5..6582669 100644
--- a/src/3rdparty/libwebp/src/dsp/cost_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/cost_neon.c
@@ -29,7 +29,7 @@ static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
   const uint8x16_t eob = vcombine_u8(vqmovn_u16(eob_0), vqmovn_u16(eob_1));
   const uint8x16_t masked = vandq_u8(eob, vld1q_u8(position));
 
-#ifdef __aarch64__
+#if WEBP_AARCH64
   res->last = vmaxvq_u8(masked) - 1;
 #else
   const uint8x8_t eob_8x8 = vmax_u8(vget_low_u8(masked), vget_high_u8(masked));
@@ -43,7 +43,7 @@ static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
 
   vst1_lane_s32(&res->last, vreinterpret_s32_u32(eob_32x2), 0);
   --res->last;
-#endif  // __aarch64__
+#endif  // WEBP_AARCH64
 
   res->coeffs = coeffs;
 }
diff --git a/src/3rdparty/libwebp/src/dsp/cpu.c b/src/3rdparty/libwebp/src/dsp/cpu.c
index 62de73f..2234c77 100644
--- a/src/3rdparty/libwebp/src/dsp/cpu.c
+++ b/src/3rdparty/libwebp/src/dsp/cpu.c
@@ -173,6 +173,7 @@ static int x86CPUInfo(CPUFeature feature) {
   }
   return 0;
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = x86CPUInfo;
 #elif defined(WEBP_ANDROID_NEON)  // NB: needs to be before generic NEON test.
 static int AndroidCPUInfo(CPUFeature feature) {
@@ -184,6 +185,7 @@ static int AndroidCPUInfo(CPUFeature feature) {
   }
   return 0;
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = AndroidCPUInfo;
 #elif defined(EMSCRIPTEN) // also needs to be before generic NEON test
 // Use compile flags as an indicator of SIMD support instead of a runtime check.
@@ -208,6 +210,7 @@ static int wasmCPUInfo(CPUFeature feature) {
   }
   return 0;
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = wasmCPUInfo;
 #elif defined(WEBP_HAVE_NEON)
 // In most cases this function doesn't check for NEON support (it's assumed by
@@ -236,6 +239,7 @@ static int armCPUInfo(CPUFeature feature) {
   return 1;
 #endif
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = armCPUInfo;
 #elif defined(WEBP_USE_MIPS32) || defined(WEBP_USE_MIPS_DSP_R2) || \
       defined(WEBP_USE_MSA)
@@ -247,7 +251,9 @@ static int mipsCPUInfo(CPUFeature feature) {
   }
 
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = mipsCPUInfo;
 #else
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = NULL;
 #endif
diff --git a/src/3rdparty/libwebp/src/dsp/cpu.h b/src/3rdparty/libwebp/src/dsp/cpu.h
index de32a39..581ecbd 100644
--- a/src/3rdparty/libwebp/src/dsp/cpu.h
+++ b/src/3rdparty/libwebp/src/dsp/cpu.h
@@ -14,10 +14,10 @@
 #ifndef WEBP_DSP_CPU_H_
 #define WEBP_DSP_CPU_H_
 
-#include <stddef.h>
-
 #include <qglobal.h>
 
+#include <stddef.h>
+
 #ifdef HAVE_CONFIG_H
 #include "src/webp/config.h"
 #endif
@@ -45,6 +45,9 @@
 #define __has_builtin(x) 0
 #endif
 
+//------------------------------------------------------------------------------
+// x86 defines.
+
 #if !defined(HAVE_CONFIG_H)
 #if defined(_MSC_VER) && _MSC_VER > 1310 && \
     (defined(_M_X64) || defined(_M_IX86)) && !defined(__clang__)
@@ -82,6 +85,9 @@
 #undef WEBP_MSC_SSE41
 #undef WEBP_MSC_SSE2
 
+//------------------------------------------------------------------------------
+// Arm defines.
+
 // 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(__aarch64__)) &&       \
@@ -100,17 +106,27 @@
 // inclusion of arm64_neon.h; Visual Studio 2019 includes this file in
 // arm_neon.h. Compile errors were seen with Visual Studio 2019 16.4 with
 // vtbl4_u8(); a fix was made in 16.6.
-#if defined(_MSC_VER) && ((_MSC_VER >= 1700 && defined(_M_ARM)) || \
-                          (_MSC_VER >= 1926 && defined(_M_ARM64))) && \
-                         !defined(__clang__) && (QT_CONFIG_neon == 1)
+#if defined(_MSC_VER) && \
+    ((_MSC_VER >= 1700 && defined(_M_ARM)) || \
+     (_MSC_VER >= 1926 && (defined(_M_ARM64) || defined(_M_ARM64EC)))) && \
+    !defined(__clang__) && (QT_CONFIG_neon == 1)
 #define WEBP_USE_NEON
 #define WEBP_USE_INTRINSICS
 #endif
 
+#if defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
+#define WEBP_AARCH64 1
+#else
+#define WEBP_AARCH64 0
+#endif
+
 #if defined(WEBP_USE_NEON) && !defined(WEBP_HAVE_NEON)
 #define WEBP_HAVE_NEON
 #endif
 
+//------------------------------------------------------------------------------
+// MIPS defines.
+
 #if defined(__mips__) && !defined(__mips64) && defined(__mips_isa_rev) && \
     (__mips_isa_rev >= 1) && (__mips_isa_rev < 6)
 #define WEBP_USE_MIPS32
@@ -126,6 +142,8 @@
 #define WEBP_USE_MSA
 #endif
 
+//------------------------------------------------------------------------------
+
 #ifndef WEBP_DSP_OMIT_C_CODE
 #define WEBP_DSP_OMIT_C_CODE 1
 #endif
@@ -136,13 +154,14 @@
 #define WEBP_NEON_OMIT_C_CODE 0
 #endif
 
-#if !(LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 8) || \
-      defined(__aarch64__))
+#if !(LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 8) || WEBP_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)
@@ -244,16 +263,7 @@ typedef enum {
   kMSA
 } CPUFeature;
 
-#ifdef __cplusplus
-extern "C" {
-#endif
-
 // returns true if the CPU supports the feature.
 typedef int (*VP8CPUInfo)(CPUFeature feature);
-WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
-
-#ifdef __cplusplus
-}    // extern "C"
-#endif
 
 #endif  // WEBP_DSP_CPU_H_
diff --git a/src/3rdparty/libwebp/src/dsp/dec.c b/src/3rdparty/libwebp/src/dsp/dec.c
index 537c701..33d8df8 100644
--- a/src/3rdparty/libwebp/src/dsp/dec.c
+++ b/src/3rdparty/libwebp/src/dsp/dec.c
@@ -734,6 +734,7 @@ VP8SimpleFilterFunc VP8SimpleHFilter16i;
 void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst,
                             int dst_stride);
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8DspInitSSE2(void);
 extern void VP8DspInitSSE41(void);
 extern void VP8DspInitNEON(void);
diff --git a/src/3rdparty/libwebp/src/dsp/dec_neon.c b/src/3rdparty/libwebp/src/dsp/dec_neon.c
index fa85170..22784cf 100644
--- a/src/3rdparty/libwebp/src/dsp/dec_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/dec_neon.c
@@ -1428,7 +1428,7 @@ static WEBP_INLINE void DC8_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x8_t A = vld1_u8(dst - BPS);  // top row
-#if defined(__aarch64__)
+#if WEBP_AARCH64
     const uint16_t p2 = vaddlv_u8(A);
     sum_top = vdupq_n_u16(p2);
 #else
@@ -1511,7 +1511,7 @@ static WEBP_INLINE void DC16_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x16_t A = vld1q_u8(dst - BPS);  // top row
-#if defined(__aarch64__)
+#if WEBP_AARCH64
     const uint16_t p3 = vaddlvq_u8(A);
     sum_top = vdupq_n_u16(p3);
 #else
diff --git a/src/3rdparty/libwebp/src/dsp/enc.c b/src/3rdparty/libwebp/src/dsp/enc.c
index ea47a3f..2ba97ba 100644
--- a/src/3rdparty/libwebp/src/dsp/enc.c
+++ b/src/3rdparty/libwebp/src/dsp/enc.c
@@ -732,6 +732,7 @@ VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT;
 VP8BlockCopy VP8Copy4x4;
 VP8BlockCopy VP8Copy16x8;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8EncDspInitSSE2(void);
 extern void VP8EncDspInitSSE41(void);
 extern void VP8EncDspInitNEON(void);
diff --git a/src/3rdparty/libwebp/src/dsp/enc_neon.c b/src/3rdparty/libwebp/src/dsp/enc_neon.c
index 3a04111..7148003 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_neon.c
@@ -764,7 +764,7 @@ static WEBP_INLINE void AccumulateSSE16_NEON(const uint8_t* const a,
 
 // Horizontal sum of all four uint32_t values in 'sum'.
 static int SumToInt_NEON(uint32x4_t sum) {
-#if defined(__aarch64__)
+#if WEBP_AARCH64
   return (int)vaddvq_u32(sum);
 #else
   const uint64x2_t sum2 = vpaddlq_u32(sum);
@@ -865,7 +865,7 @@ static int QuantizeBlock_NEON(int16_t in[16], int16_t out[16],
   uint8x8x4_t shuffles;
   // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
   // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
   uint8x16x2_t all_out;
   INIT_VECTOR2(all_out, vreinterpretq_u8_s16(out0), vreinterpretq_u8_s16(out1));
diff --git a/src/3rdparty/libwebp/src/dsp/enc_sse2.c b/src/3rdparty/libwebp/src/dsp/enc_sse2.c
index 1d10556..010624a 100644
--- a/src/3rdparty/libwebp/src/dsp/enc_sse2.c
+++ b/src/3rdparty/libwebp/src/dsp/enc_sse2.c
@@ -25,9 +25,160 @@
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
-// Does one or two inverse transforms.
-static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
-                            int do_two) {
+// Does one inverse transform.
+static void ITransform_One_SSE2(const uint8_t* ref, const int16_t* in,
+                                uint8_t* dst) {
+  // This implementation makes use of 16-bit fixed point versions of two
+  // multiply constants:
+  //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
+  //    K2 = sqrt(2) * sin (pi/8) ~= 35468 / 2^16
+  //
+  // To be able to use signed 16-bit integers, we use the following trick to
+  // have constants within range:
+  // - Associated constants are obtained by subtracting the 16-bit fixed point
+  //   version of one:
+  //      k = K - (1 << 16)  =>  K = k + (1 << 16)
+  //      K1 = 85267  =>  k1 =  20091
+  //      K2 = 35468  =>  k2 = -30068
+  // - The multiplication of a variable by a constant become the sum of the
+  //   variable and the multiplication of that variable by the associated
+  //   constant:
+  //      (x * K) >> 16 = (x * (k + (1 << 16))) >> 16 = ((x * k ) >> 16) + x
+  const __m128i k1k2 = _mm_set_epi16(-30068, -30068, -30068, -30068,
+                                     20091, 20091, 20091, 20091);
+  const __m128i k2k1 = _mm_set_epi16(20091, 20091, 20091, 20091,
+                                     -30068, -30068, -30068, -30068);
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i zero_four = _mm_set_epi16(0, 0, 0, 0, 4, 4, 4, 4);
+  __m128i T01, T23;
+
+  // Load and concatenate the transform coefficients.
+  const __m128i in01 = _mm_loadu_si128((const __m128i*)&in[0]);
+  const __m128i in23 = _mm_loadu_si128((const __m128i*)&in[8]);
+  // a00 a10 a20 a30   a01 a11 a21 a31
+  // a02 a12 a22 a32   a03 a13 a23 a33
+
+  // Vertical pass and subsequent transpose.
+  {
+    const __m128i in1 = _mm_unpackhi_epi64(in01, in01);
+    const __m128i in3 = _mm_unpackhi_epi64(in23, in23);
+
+    // First pass, c and d calculations are longer because of the "trick"
+    // multiplications.
+    // c = MUL(in1, K2) - MUL(in3, K1) = MUL(in1, k2) - MUL(in3, k1) + in1 - in3
+    // d = MUL(in1, K1) + MUL(in3, K2) = MUL(in1, k1) + MUL(in3, k2) + in1 + in3
+    const __m128i a_d3 = _mm_add_epi16(in01, in23);
+    const __m128i b_c3 = _mm_sub_epi16(in01, in23);
+    const __m128i c1d1 = _mm_mulhi_epi16(in1, k2k1);
+    const __m128i c2d2 = _mm_mulhi_epi16(in3, k1k2);
+    const __m128i c3 = _mm_unpackhi_epi64(b_c3, b_c3);
+    const __m128i c4 = _mm_sub_epi16(c1d1, c2d2);
+    const __m128i c = _mm_add_epi16(c3, c4);
+    const __m128i d4u = _mm_add_epi16(c1d1, c2d2);
+    const __m128i du = _mm_add_epi16(a_d3, d4u);
+    const __m128i d = _mm_unpackhi_epi64(du, du);
+
+    // Second pass.
+    const __m128i comb_ab = _mm_unpacklo_epi64(a_d3, b_c3);
+    const __m128i comb_dc = _mm_unpacklo_epi64(d, c);
+
+    const __m128i tmp01 = _mm_add_epi16(comb_ab, comb_dc);
+    const __m128i tmp32 = _mm_sub_epi16(comb_ab, comb_dc);
+    const __m128i tmp23 = _mm_shuffle_epi32(tmp32, _MM_SHUFFLE(1, 0, 3, 2));
+
+    const __m128i transpose_0 = _mm_unpacklo_epi16(tmp01, tmp23);
+    const __m128i transpose_1 = _mm_unpackhi_epi16(tmp01, tmp23);
+    // a00 a20 a01 a21   a02 a22 a03 a23
+    // a10 a30 a11 a31   a12 a32 a13 a33
+
+    T01 = _mm_unpacklo_epi16(transpose_0, transpose_1);
+    T23 = _mm_unpackhi_epi16(transpose_0, transpose_1);
+    // a00 a10 a20 a30   a01 a11 a21 a31
+    // a02 a12 a22 a32   a03 a13 a23 a33
+  }
+
+  // Horizontal pass and subsequent transpose.
+  {
+    const __m128i T1 = _mm_unpackhi_epi64(T01, T01);
+    const __m128i T3 = _mm_unpackhi_epi64(T23, T23);
+
+    // First pass, c and d calculations are longer because of the "trick"
+    // multiplications.
+    const __m128i dc = _mm_add_epi16(T01, zero_four);
+
+    // c = MUL(T1, K2) - MUL(T3, K1) = MUL(T1, k2) - MUL(T3, k1) + T1 - T3
+    // d = MUL(T1, K1) + MUL(T3, K2) = MUL(T1, k1) + MUL(T3, k2) + T1 + T3
+    const __m128i a_d3 = _mm_add_epi16(dc, T23);
+    const __m128i b_c3 = _mm_sub_epi16(dc, T23);
+    const __m128i c1d1 = _mm_mulhi_epi16(T1, k2k1);
+    const __m128i c2d2 = _mm_mulhi_epi16(T3, k1k2);
+    const __m128i c3 = _mm_unpackhi_epi64(b_c3, b_c3);
+    const __m128i c4 = _mm_sub_epi16(c1d1, c2d2);
+    const __m128i c = _mm_add_epi16(c3, c4);
+    const __m128i d4u = _mm_add_epi16(c1d1, c2d2);
+    const __m128i du = _mm_add_epi16(a_d3, d4u);
+    const __m128i d = _mm_unpackhi_epi64(du, du);
+
+    // Second pass.
+    const __m128i comb_ab = _mm_unpacklo_epi64(a_d3, b_c3);
+    const __m128i comb_dc = _mm_unpacklo_epi64(d, c);
+
+    const __m128i tmp01 = _mm_add_epi16(comb_ab, comb_dc);
+    const __m128i tmp32 = _mm_sub_epi16(comb_ab, comb_dc);
+    const __m128i tmp23 = _mm_shuffle_epi32(tmp32, _MM_SHUFFLE(1, 0, 3, 2));
+
+    const __m128i shifted01 = _mm_srai_epi16(tmp01, 3);
+    const __m128i shifted23 = _mm_srai_epi16(tmp23, 3);
+    // a00 a01 a02 a03   a10 a11 a12 a13
+    // a20 a21 a22 a23   a30 a31 a32 a33
+
+    const __m128i transpose_0 = _mm_unpacklo_epi16(shifted01, shifted23);
+    const __m128i transpose_1 = _mm_unpackhi_epi16(shifted01, shifted23);
+    // a00 a20 a01 a21   a02 a22 a03 a23
+    // a10 a30 a11 a31   a12 a32 a13 a33
+
+    T01 = _mm_unpacklo_epi16(transpose_0, transpose_1);
+    T23 = _mm_unpackhi_epi16(transpose_0, transpose_1);
+    // a00 a10 a20 a30   a01 a11 a21 a31
+    // a02 a12 a22 a32   a03 a13 a23 a33
+  }
+
+  // Add inverse transform to 'ref' and store.
+  {
+    // Load the reference(s).
+    __m128i ref01, ref23, ref0123;
+    int32_t buf[4];
+
+    // Load four bytes/pixels per line.
+    const __m128i ref0 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[0 * BPS]));
+    const __m128i ref1 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[1 * BPS]));
+    const __m128i ref2 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[2 * BPS]));
+    const __m128i ref3 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[3 * BPS]));
+    ref01 = _mm_unpacklo_epi32(ref0, ref1);
+    ref23 = _mm_unpacklo_epi32(ref2, ref3);
+
+    // Convert to 16b.
+    ref01 = _mm_unpacklo_epi8(ref01, zero);
+    ref23 = _mm_unpacklo_epi8(ref23, zero);
+    // Add the inverse transform(s).
+    ref01 = _mm_add_epi16(ref01, T01);
+    ref23 = _mm_add_epi16(ref23, T23);
+    // Unsigned saturate to 8b.
+    ref0123 = _mm_packus_epi16(ref01, ref23);
+
+    _mm_storeu_si128((__m128i *)buf, ref0123);
+
+    // Store four bytes/pixels per line.
+    WebPInt32ToMem(&dst[0 * BPS], buf[0]);
+    WebPInt32ToMem(&dst[1 * BPS], buf[1]);
+    WebPInt32ToMem(&dst[2 * BPS], buf[2]);
+    WebPInt32ToMem(&dst[3 * BPS], buf[3]);
+  }
+}
+
+// Does two inverse transforms.
+static void ITransform_Two_SSE2(const uint8_t* ref, const int16_t* in,
+                                uint8_t* dst) {
   // This implementation makes use of 16-bit fixed point versions of two
   // multiply constants:
   //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
@@ -49,33 +200,21 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
   __m128i T0, T1, T2, T3;
 
   // Load and concatenate the transform coefficients (we'll do two inverse
-  // transforms in parallel). In the case of only one inverse transform, the
-  // second half of the vectors will just contain random value we'll never
-  // use nor store.
+  // transforms in parallel).
   __m128i in0, in1, in2, in3;
   {
-    in0 = _mm_loadl_epi64((const __m128i*)&in[0]);
-    in1 = _mm_loadl_epi64((const __m128i*)&in[4]);
-    in2 = _mm_loadl_epi64((const __m128i*)&in[8]);
-    in3 = _mm_loadl_epi64((const __m128i*)&in[12]);
-    // a00 a10 a20 a30   x x x x
-    // a01 a11 a21 a31   x x x x
-    // a02 a12 a22 a32   x x x x
-    // a03 a13 a23 a33   x x x x
-    if (do_two) {
-      const __m128i inB0 = _mm_loadl_epi64((const __m128i*)&in[16]);
-      const __m128i inB1 = _mm_loadl_epi64((const __m128i*)&in[20]);
-      const __m128i inB2 = _mm_loadl_epi64((const __m128i*)&in[24]);
-      const __m128i inB3 = _mm_loadl_epi64((const __m128i*)&in[28]);
-      in0 = _mm_unpacklo_epi64(in0, inB0);
-      in1 = _mm_unpacklo_epi64(in1, inB1);
-      in2 = _mm_unpacklo_epi64(in2, inB2);
-      in3 = _mm_unpacklo_epi64(in3, inB3);
-      // a00 a10 a20 a30   b00 b10 b20 b30
-      // a01 a11 a21 a31   b01 b11 b21 b31
-      // a02 a12 a22 a32   b02 b12 b22 b32
-      // a03 a13 a23 a33   b03 b13 b23 b33
-    }
+    const __m128i tmp0 = _mm_loadu_si128((const __m128i*)&in[0]);
+    const __m128i tmp1 = _mm_loadu_si128((const __m128i*)&in[8]);
+    const __m128i tmp2 = _mm_loadu_si128((const __m128i*)&in[16]);
+    const __m128i tmp3 = _mm_loadu_si128((const __m128i*)&in[24]);
+    in0 = _mm_unpacklo_epi64(tmp0, tmp2);
+    in1 = _mm_unpackhi_epi64(tmp0, tmp2);
+    in2 = _mm_unpacklo_epi64(tmp1, tmp3);
+    in3 = _mm_unpackhi_epi64(tmp1, tmp3);
+    // a00 a10 a20 a30   b00 b10 b20 b30
+    // a01 a11 a21 a31   b01 b11 b21 b31
+    // a02 a12 a22 a32   b02 b12 b22 b32
+    // a03 a13 a23 a33   b03 b13 b23 b33
   }
 
   // Vertical pass and subsequent transpose.
@@ -148,19 +287,11 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
     const __m128i zero = _mm_setzero_si128();
     // Load the reference(s).
     __m128i ref0, ref1, ref2, ref3;
-    if (do_two) {
-      // Load eight bytes/pixels per line.
-      ref0 = _mm_loadl_epi64((const __m128i*)&ref[0 * BPS]);
-      ref1 = _mm_loadl_epi64((const __m128i*)&ref[1 * BPS]);
-      ref2 = _mm_loadl_epi64((const __m128i*)&ref[2 * BPS]);
-      ref3 = _mm_loadl_epi64((const __m128i*)&ref[3 * BPS]);
-    } else {
-      // Load four bytes/pixels per line.
-      ref0 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[0 * BPS]));
-      ref1 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[1 * BPS]));
-      ref2 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[2 * BPS]));
-      ref3 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[3 * BPS]));
-    }
+    // Load eight bytes/pixels per line.
+    ref0 = _mm_loadl_epi64((const __m128i*)&ref[0 * BPS]);
+    ref1 = _mm_loadl_epi64((const __m128i*)&ref[1 * BPS]);
+    ref2 = _mm_loadl_epi64((const __m128i*)&ref[2 * BPS]);
+    ref3 = _mm_loadl_epi64((const __m128i*)&ref[3 * BPS]);
     // Convert to 16b.
     ref0 = _mm_unpacklo_epi8(ref0, zero);
     ref1 = _mm_unpacklo_epi8(ref1, zero);
@@ -176,20 +307,21 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
     ref1 = _mm_packus_epi16(ref1, ref1);
     ref2 = _mm_packus_epi16(ref2, ref2);
     ref3 = _mm_packus_epi16(ref3, ref3);
-    // Store the results.
-    if (do_two) {
-      // Store eight bytes/pixels per line.
-      _mm_storel_epi64((__m128i*)&dst[0 * BPS], ref0);
-      _mm_storel_epi64((__m128i*)&dst[1 * BPS], ref1);
-      _mm_storel_epi64((__m128i*)&dst[2 * BPS], ref2);
-      _mm_storel_epi64((__m128i*)&dst[3 * BPS], ref3);
-    } else {
-      // Store four bytes/pixels per line.
-      WebPInt32ToMem(&dst[0 * BPS], _mm_cvtsi128_si32(ref0));
-      WebPInt32ToMem(&dst[1 * BPS], _mm_cvtsi128_si32(ref1));
-      WebPInt32ToMem(&dst[2 * BPS], _mm_cvtsi128_si32(ref2));
-      WebPInt32ToMem(&dst[3 * BPS], _mm_cvtsi128_si32(ref3));
-    }
+    // Store eight bytes/pixels per line.
+    _mm_storel_epi64((__m128i*)&dst[0 * BPS], ref0);
+    _mm_storel_epi64((__m128i*)&dst[1 * BPS], ref1);
+    _mm_storel_epi64((__m128i*)&dst[2 * BPS], ref2);
+    _mm_storel_epi64((__m128i*)&dst[3 * BPS], ref3);
+  }
+}
+
+// Does one or two inverse transforms.
+static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
+                            int do_two) {
+  if (do_two) {
+    ITransform_Two_SSE2(ref, in, dst);
+  } else {
+    ITransform_One_SSE2(ref, in, dst);
   }
 }
 
diff --git a/src/3rdparty/libwebp/src/dsp/filters.c b/src/3rdparty/libwebp/src/dsp/filters.c
index 4506567..85eee50 100644
--- a/src/3rdparty/libwebp/src/dsp/filters.c
+++ b/src/3rdparty/libwebp/src/dsp/filters.c
@@ -233,6 +233,7 @@ static void GradientUnfilter_C(const uint8_t* prev, const uint8_t* in,
 WebPFilterFunc WebPFilters[WEBP_FILTER_LAST];
 WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8FiltersInitMIPSdspR2(void);
 extern void VP8FiltersInitMSA(void);
 extern void VP8FiltersInitNEON(void);
diff --git a/src/3rdparty/libwebp/src/dsp/lossless.c b/src/3rdparty/libwebp/src/dsp/lossless.c
index fb86e58..9f81209 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless.c
@@ -588,6 +588,7 @@ VP8LConvertFunc VP8LConvertBGRAToBGR;
 VP8LMapARGBFunc VP8LMapColor32b;
 VP8LMapAlphaFunc VP8LMapColor8b;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8LDspInitSSE2(void);
 extern void VP8LDspInitSSE41(void);
 extern void VP8LDspInitNEON(void);
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc.c b/src/3rdparty/libwebp/src/dsp/lossless_enc.c
index b1f9f26..cde1280 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc.c
@@ -791,6 +791,7 @@ VP8LBundleColorMapFunc VP8LBundleColorMap;
 VP8LPredictorAddSubFunc VP8LPredictorsSub[16];
 VP8LPredictorAddSubFunc VP8LPredictorsSub_C[16];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8LEncDspInitSSE2(void);
 extern void VP8LEncDspInitSSE41(void);
 extern void VP8LEncDspInitNEON(void);
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c b/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c
index 7c7b73f..e32c796 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_enc_neon.c
@@ -25,7 +25,7 @@
 
 // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
 // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
 #define USE_VTBLQ
 #endif
diff --git a/src/3rdparty/libwebp/src/dsp/lossless_neon.c b/src/3rdparty/libwebp/src/dsp/lossless_neon.c
index 89e3e01..ddc9b61 100644
--- a/src/3rdparty/libwebp/src/dsp/lossless_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/lossless_neon.c
@@ -498,7 +498,7 @@ static void PredictorAdd13_NEON(const uint32_t* in, const uint32_t* upper,
 
 // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
 // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
 #define USE_VTBLQ
 #endif
diff --git a/src/3rdparty/libwebp/src/dsp/neon.h b/src/3rdparty/libwebp/src/dsp/neon.h
index c591f9b..14acb40 100644
--- a/src/3rdparty/libwebp/src/dsp/neon.h
+++ b/src/3rdparty/libwebp/src/dsp/neon.h
@@ -21,7 +21,7 @@
 // Right now, some intrinsics functions seem slower, so we disable them
 // 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__)
+#if LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 9) || WEBP_AARCH64
 #define WEBP_USE_INTRINSICS   // use intrinsics when possible
 #endif
 
@@ -46,7 +46,7 @@
 // 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_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
+#if !(LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 8) || WEBP_AARCH64)
 #define WORK_AROUND_GCC
 #endif
 
diff --git a/src/3rdparty/libwebp/src/dsp/quant.h b/src/3rdparty/libwebp/src/dsp/quant.h
index fc099bf..bf7734c 100644
--- a/src/3rdparty/libwebp/src/dsp/quant.h
+++ b/src/3rdparty/libwebp/src/dsp/quant.h
@@ -22,7 +22,7 @@
 #define IsFlat IsFlat_NEON
 
 static uint32_t horizontal_add_uint32x4(const uint32x4_t a) {
-#if defined(__aarch64__)
+#if WEBP_AARCH64
   return vaddvq_u32(a);
 #else
   const uint64x2_t b = vpaddlq_u32(a);
diff --git a/src/3rdparty/libwebp/src/dsp/rescaler.c b/src/3rdparty/libwebp/src/dsp/rescaler.c
index 14620ce..325d8be 100644
--- a/src/3rdparty/libwebp/src/dsp/rescaler.c
+++ b/src/3rdparty/libwebp/src/dsp/rescaler.c
@@ -197,6 +197,7 @@ WebPRescalerImportRowFunc WebPRescalerImportRowShrink;
 WebPRescalerExportRowFunc WebPRescalerExportRowExpand;
 WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPRescalerDspInitSSE2(void);
 extern void WebPRescalerDspInitMIPS32(void);
 extern void WebPRescalerDspInitMIPSdspR2(void);
diff --git a/src/3rdparty/libwebp/src/dsp/ssim.c b/src/3rdparty/libwebp/src/dsp/ssim.c
index f85c2e6..9a1341e 100644
--- a/src/3rdparty/libwebp/src/dsp/ssim.c
+++ b/src/3rdparty/libwebp/src/dsp/ssim.c
@@ -137,6 +137,7 @@ VP8SSIMGetClippedFunc VP8SSIMGetClipped;
 VP8AccumulateSSEFunc VP8AccumulateSSE;
 #endif
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8SSIMDspInitSSE2(void);
 
 WEBP_DSP_INIT_FUNC(VP8SSIMDspInit) {
diff --git a/src/3rdparty/libwebp/src/dsp/upsampling.c b/src/3rdparty/libwebp/src/dsp/upsampling.c
index 87f771f..983b9c4 100644
--- a/src/3rdparty/libwebp/src/dsp/upsampling.c
+++ b/src/3rdparty/libwebp/src/dsp/upsampling.c
@@ -215,6 +215,7 @@ static void EmptyYuv444Func(const uint8_t* y,
 
 WebPYUV444Converter WebPYUV444Converters[MODE_LAST];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPInitYUV444ConvertersMIPSdspR2(void);
 extern void WebPInitYUV444ConvertersSSE2(void);
 extern void WebPInitYUV444ConvertersSSE41(void);
diff --git a/src/3rdparty/libwebp/src/dsp/upsampling_neon.c b/src/3rdparty/libwebp/src/dsp/upsampling_neon.c
index 6ba71a7..bbc000c 100644
--- a/src/3rdparty/libwebp/src/dsp/upsampling_neon.c
+++ b/src/3rdparty/libwebp/src/dsp/upsampling_neon.c
@@ -111,7 +111,7 @@ static const int16_t kCoeffs1[4] = { 19077, 26149, 6419, 13320 };
   vst4_u8(out, v255_r_g_b);                                             \
 } while (0)
 
-#if !defined(WEBP_SWAP_16BIT_CSP)
+#if (WEBP_SWAP_16BIT_CSP == 0)
 #define ZIP_U8(lo, hi) vzip_u8((lo), (hi))
 #else
 #define ZIP_U8(lo, hi) vzip_u8((hi), (lo))
diff --git a/src/3rdparty/libwebp/src/dsp/yuv.c b/src/3rdparty/libwebp/src/dsp/yuv.c
index d16c13d..8a04b85 100644
--- a/src/3rdparty/libwebp/src/dsp/yuv.c
+++ b/src/3rdparty/libwebp/src/dsp/yuv.c
@@ -70,6 +70,7 @@ void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
 
 WebPSamplerRowFunc WebPSamplers[MODE_LAST];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPInitSamplersSSE2(void);
 extern void WebPInitSamplersSSE41(void);
 extern void WebPInitSamplersMIPS32(void);