diff options
| author | Jocelyn Turcotte <jocelyn.turcotte@digia.com> | 2014-08-08 14:30:41 +0200 |
|---|---|---|
| committer | Jocelyn Turcotte <jocelyn.turcotte@digia.com> | 2014-08-12 13:49:54 +0200 |
| commit | ab0a50979b9eb4dfa3320eff7e187e41efedf7a9 (patch) | |
| tree | 498dfb8a97ff3361a9f7486863a52bb4e26bb898 /chromium/third_party/libvpx/source/libvpx/vp8 | |
| parent | 4ce69f7403811819800e7c5ae1318b2647e778d1 (diff) | |
Update Chromium to beta version 37.0.2062.68
Change-Id: I188e3b5aff1bec75566014291b654eb19f5bc8ca
Reviewed-by: Andras Becsi <andras.becsi@digia.com>
Diffstat (limited to 'chromium/third_party/libvpx/source/libvpx/vp8')
155 files changed, 6171 insertions, 6053 deletions
diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/alloccommon.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/alloccommon.h index ea93c252280..93e99d76b1d 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/alloccommon.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/alloccommon.h @@ -9,15 +9,23 @@ */ -#ifndef __INC_ALLOCCOMMON_H -#define __INC_ALLOCCOMMON_H +#ifndef VP8_COMMON_ALLOCCOMMON_H_ +#define VP8_COMMON_ALLOCCOMMON_H_ #include "onyxc_int.h" +#ifdef __cplusplus +extern "C" { +#endif + void vp8_create_common(VP8_COMMON *oci); void vp8_remove_common(VP8_COMMON *oci); void vp8_de_alloc_frame_buffers(VP8_COMMON *oci); int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height); void vp8_setup_version(VP8_COMMON *oci); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_ALLOCCOMMON_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance16x16_armv6.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance16x16_armv6.asm index dc84c30daf5..39919579f80 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance16x16_armv6.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance16x16_armv6.asm @@ -53,7 +53,7 @@ loop orr r6, r6, r7 ; differences of all 4 pixels ; calculate total sum adds r8, r8, r4 ; add positive differences to sum - subs r8, r8, r5 ; substract negative differences from sum + subs r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -77,7 +77,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -101,7 +101,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -127,7 +127,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance8x8_armv6.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance8x8_armv6.asm index adc353d2006..915ee499309 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance8x8_armv6.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance8x8_armv6.asm @@ -51,7 +51,7 @@ loop orr r8, r8, r10 ; differences of all 4 pixels ; calculate total sum add r4, r4, r6 ; add positive differences to sum - sub r4, r4, r7 ; substract negative differences from sum + sub r4, r4, r7 ; subtract negative differences from sum ; calculate sse uxtb16 r7, r8 ; byte (two pixels) to halfwords @@ -77,7 +77,7 @@ loop ; calculate total sum add r4, r4, r6 ; add positive differences to sum - sub r4, r4, r7 ; substract negative differences from sum + sub r4, r4, r7 ; subtract negative differences from sum ; calculate sse uxtb16 r7, r8 ; byte (two pixels) to halfwords diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_h_armv6.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_h_armv6.asm index dd2ce685c8b..3668dc517a9 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_h_armv6.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_h_armv6.asm @@ -58,7 +58,7 @@ loop orr r6, r6, r7 ; differences of all 4 pixels ; calculate total sum adds r8, r8, r4 ; add positive differences to sum - subs r8, r8, r5 ; substract negative differences from sum + subs r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -89,7 +89,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -120,7 +120,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -153,7 +153,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_hv_armv6.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_hv_armv6.asm index f972d9b5bac..b4e0959d1b3 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_hv_armv6.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_hv_armv6.asm @@ -69,7 +69,7 @@ loop orr r6, r6, r7 ; differences of all 4 pixels ; calculate total sum adds r8, r8, r4 ; add positive differences to sum - subs r8, r8, r5 ; substract negative differences from sum + subs r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -111,7 +111,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -153,7 +153,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -195,7 +195,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_v_armv6.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_v_armv6.asm index f5da9c09eed..10863e2ec80 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_v_armv6.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/armv6/vp8_variance_halfpixvar16x16_v_armv6.asm @@ -59,7 +59,7 @@ loop orr r6, r6, r7 ; differences of all 4 pixels ; calculate total sum adds r8, r8, r4 ; add positive differences to sum - subs r8, r8, r5 ; substract negative differences from sum + subs r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -90,7 +90,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -121,7 +121,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords @@ -154,7 +154,7 @@ loop ; calculate total sum add r8, r8, r4 ; add positive differences to sum - sub r8, r8, r5 ; substract negative differences from sum + sub r8, r8, r5 ; subtract negative differences from sum ; calculate sse uxtb16 r5, r6 ; byte (two pixels) to halfwords diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/bilinearfilter_arm.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/bilinearfilter_arm.h index b7155d3f0a5..6b84e6f3b55 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/bilinearfilter_arm.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/bilinearfilter_arm.h @@ -9,8 +9,12 @@ */ -#ifndef BILINEARFILTER_ARM_H -#define BILINEARFILTER_ARM_H +#ifndef VP8_COMMON_ARM_BILINEARFILTER_ARM_H_ +#define VP8_COMMON_ARM_BILINEARFILTER_ARM_H_ + +#ifdef __cplusplus +extern "C" { +#endif extern void vp8_filter_block2d_bil_first_pass_armv6 ( @@ -32,4 +36,8 @@ extern void vp8_filter_block2d_bil_second_pass_armv6 const short *vp8_filter ); -#endif /* BILINEARFILTER_ARM_H */ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_ARM_BILINEARFILTER_ARM_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/dequantize_arm.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/dequantize_arm.c index 70e72aa4774..1f8157f0b11 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/dequantize_arm.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/dequantize_arm.c @@ -12,26 +12,9 @@ #include "vpx_config.h" #include "vp8/common/blockd.h" -#if HAVE_NEON -extern void vp8_dequantize_b_loop_neon(short *Q, short *DQC, short *DQ); -#endif - #if HAVE_MEDIA extern void vp8_dequantize_b_loop_v6(short *Q, short *DQC, short *DQ); -#endif - -#if HAVE_NEON - -void vp8_dequantize_b_neon(BLOCKD *d, short *DQC) -{ - short *DQ = d->dqcoeff; - short *Q = d->qcoeff; - - vp8_dequantize_b_loop_neon(Q, DQC, DQ); -} -#endif -#if HAVE_MEDIA void vp8_dequantize_b_v6(BLOCKD *d, short *DQC) { short *DQ = d->dqcoeff; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/loopfilter_arm.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/loopfilter_arm.c index 3bdc9675e71..f37ca636edc 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/loopfilter_arm.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/loopfilter_arm.c @@ -25,20 +25,24 @@ extern prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_armv6); extern prototype_loopfilter(vp8_mbloop_filter_vertical_edge_armv6); #endif -#if HAVE_NEON +#if HAVE_NEON_ASM || HAVE_NEON typedef void loopfilter_y_neon(unsigned char *src, int pitch, unsigned char blimit, unsigned char limit, unsigned char thresh); typedef void loopfilter_uv_neon(unsigned char *u, int pitch, unsigned char blimit, unsigned char limit, unsigned char thresh, unsigned char *v); +#endif +#if HAVE_NEON_ASM extern loopfilter_y_neon vp8_loop_filter_horizontal_edge_y_neon; extern loopfilter_y_neon vp8_loop_filter_vertical_edge_y_neon; -extern loopfilter_y_neon vp8_mbloop_filter_horizontal_edge_y_neon; -extern loopfilter_y_neon vp8_mbloop_filter_vertical_edge_y_neon; - extern loopfilter_uv_neon vp8_loop_filter_horizontal_edge_uv_neon; extern loopfilter_uv_neon vp8_loop_filter_vertical_edge_uv_neon; +#endif + +#if HAVE_NEON +extern loopfilter_y_neon vp8_mbloop_filter_horizontal_edge_y_neon; +extern loopfilter_y_neon vp8_mbloop_filter_vertical_edge_y_neon; extern loopfilter_uv_neon vp8_mbloop_filter_horizontal_edge_uv_neon; extern loopfilter_uv_neon vp8_mbloop_filter_vertical_edge_uv_neon; #endif @@ -146,7 +150,9 @@ void vp8_loop_filter_mbv_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsign if (u_ptr) vp8_mbloop_filter_vertical_edge_uv_neon(u_ptr, uv_stride, mblim, lim, hev_thr, v_ptr); } +#endif +#if HAVE_NEON_ASM /* Horizontal B Filtering */ void vp8_loop_filter_bh_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, int y_stride, int uv_stride, loop_filter_info *lfi) diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict16x16_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict16x16_neon.asm deleted file mode 100644 index e392786d43d..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict16x16_neon.asm +++ /dev/null @@ -1,357 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_bilinear_predict16x16_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; r4 unsigned char *dst_ptr, -; stack(r5) int dst_pitch - -|vp8_bilinear_predict16x16_neon| PROC - push {r4-r5, lr} - - adr r12, bifilter16_coeff - ldr r4, [sp, #12] ;load parameters from stack - ldr r5, [sp, #16] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq secondpass_bfilter16x16_only - - add r2, r12, r2, lsl #3 ;calculate filter location - - cmp r3, #0 ;skip second_pass filter if yoffset=0 - - vld1.s32 {d31}, [r2] ;load first_pass filter - - beq firstpass_bfilter16x16_only - - sub sp, sp, #272 ;reserve space on stack for temporary storage - vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data - mov lr, sp - vld1.u8 {d5, d6, d7}, [r0], r1 - - mov r2, #3 ;loop counter - vld1.u8 {d8, d9, d10}, [r0], r1 - - vdup.8 d0, d31[0] ;first_pass filter (d0 d1) - vld1.u8 {d11, d12, d13}, [r0], r1 - - vdup.8 d1, d31[4] - -;First Pass: output_height lines x output_width columns (17x16) -filt_blk2d_fp16x16_loop_neon - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q8, d3, d0 - vmull.u8 q9, d5, d0 - vmull.u8 q10, d6, d0 - vmull.u8 q11, d8, d0 - vmull.u8 q12, d9, d0 - vmull.u8 q13, d11, d0 - vmull.u8 q14, d12, d0 - - vext.8 d2, d2, d3, #1 ;construct src_ptr[1] - vext.8 d5, d5, d6, #1 - vext.8 d8, d8, d9, #1 - vext.8 d11, d11, d12, #1 - - vmlal.u8 q7, d2, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q9, d5, d1 - vmlal.u8 q11, d8, d1 - vmlal.u8 q13, d11, d1 - - vext.8 d3, d3, d4, #1 - vext.8 d6, d6, d7, #1 - vext.8 d9, d9, d10, #1 - vext.8 d12, d12, d13, #1 - - vmlal.u8 q8, d3, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q10, d6, d1 - vmlal.u8 q12, d9, d1 - vmlal.u8 q14, d12, d1 - - subs r2, r2, #1 - - vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d15, q8, #7 - vqrshrn.u16 d16, q9, #7 - vqrshrn.u16 d17, q10, #7 - vqrshrn.u16 d18, q11, #7 - vqrshrn.u16 d19, q12, #7 - vqrshrn.u16 d20, q13, #7 - - vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data - vqrshrn.u16 d21, q14, #7 - vld1.u8 {d5, d6, d7}, [r0], r1 - - vst1.u8 {d14, d15, d16, d17}, [lr]! ;store result - vld1.u8 {d8, d9, d10}, [r0], r1 - vst1.u8 {d18, d19, d20, d21}, [lr]! - vld1.u8 {d11, d12, d13}, [r0], r1 - - bne filt_blk2d_fp16x16_loop_neon - -;First-pass filtering for rest 5 lines - vld1.u8 {d14, d15, d16}, [r0], r1 - - vmull.u8 q9, d2, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q10, d3, d0 - vmull.u8 q11, d5, d0 - vmull.u8 q12, d6, d0 - vmull.u8 q13, d8, d0 - vmull.u8 q14, d9, d0 - - vext.8 d2, d2, d3, #1 ;construct src_ptr[1] - vext.8 d5, d5, d6, #1 - vext.8 d8, d8, d9, #1 - - vmlal.u8 q9, d2, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q11, d5, d1 - vmlal.u8 q13, d8, d1 - - vext.8 d3, d3, d4, #1 - vext.8 d6, d6, d7, #1 - vext.8 d9, d9, d10, #1 - - vmlal.u8 q10, d3, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q12, d6, d1 - vmlal.u8 q14, d9, d1 - - vmull.u8 q1, d11, d0 - vmull.u8 q2, d12, d0 - vmull.u8 q3, d14, d0 - vmull.u8 q4, d15, d0 - - vext.8 d11, d11, d12, #1 ;construct src_ptr[1] - vext.8 d14, d14, d15, #1 - - vmlal.u8 q1, d11, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q3, d14, d1 - - vext.8 d12, d12, d13, #1 - vext.8 d15, d15, d16, #1 - - vmlal.u8 q2, d12, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q4, d15, d1 - - vqrshrn.u16 d10, q9, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d11, q10, #7 - vqrshrn.u16 d12, q11, #7 - vqrshrn.u16 d13, q12, #7 - vqrshrn.u16 d14, q13, #7 - vqrshrn.u16 d15, q14, #7 - vqrshrn.u16 d16, q1, #7 - vqrshrn.u16 d17, q2, #7 - vqrshrn.u16 d18, q3, #7 - vqrshrn.u16 d19, q4, #7 - - vst1.u8 {d10, d11, d12, d13}, [lr]! ;store result - vst1.u8 {d14, d15, d16, d17}, [lr]! - vst1.u8 {d18, d19}, [lr]! - -;Second pass: 16x16 -;secondpass_filter - add r3, r12, r3, lsl #3 - sub lr, lr, #272 - - vld1.u32 {d31}, [r3] ;load second_pass filter - - vld1.u8 {d22, d23}, [lr]! ;load src data - - vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1) - vdup.8 d1, d31[4] - mov r12, #4 ;loop counter - -filt_blk2d_sp16x16_loop_neon - vld1.u8 {d24, d25}, [lr]! - vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0]) - vld1.u8 {d26, d27}, [lr]! - vmull.u8 q2, d23, d0 - vld1.u8 {d28, d29}, [lr]! - vmull.u8 q3, d24, d0 - vld1.u8 {d30, d31}, [lr]! - - vmull.u8 q4, d25, d0 - vmull.u8 q5, d26, d0 - vmull.u8 q6, d27, d0 - vmull.u8 q7, d28, d0 - vmull.u8 q8, d29, d0 - - vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * vp8_filter[1]) - vmlal.u8 q2, d25, d1 - vmlal.u8 q3, d26, d1 - vmlal.u8 q4, d27, d1 - vmlal.u8 q5, d28, d1 - vmlal.u8 q6, d29, d1 - vmlal.u8 q7, d30, d1 - vmlal.u8 q8, d31, d1 - - subs r12, r12, #1 - - vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d3, q2, #7 - vqrshrn.u16 d4, q3, #7 - vqrshrn.u16 d5, q4, #7 - vqrshrn.u16 d6, q5, #7 - vqrshrn.u16 d7, q6, #7 - vqrshrn.u16 d8, q7, #7 - vqrshrn.u16 d9, q8, #7 - - vst1.u8 {d2, d3}, [r4], r5 ;store result - vst1.u8 {d4, d5}, [r4], r5 - vst1.u8 {d6, d7}, [r4], r5 - vmov q11, q15 - vst1.u8 {d8, d9}, [r4], r5 - - bne filt_blk2d_sp16x16_loop_neon - - add sp, sp, #272 - - pop {r4-r5,pc} - -;-------------------- -firstpass_bfilter16x16_only - mov r2, #4 ;loop counter - vdup.8 d0, d31[0] ;first_pass filter (d0 d1) - vdup.8 d1, d31[4] - -;First Pass: output_height lines x output_width columns (16x16) -filt_blk2d_fpo16x16_loop_neon - vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data - vld1.u8 {d5, d6, d7}, [r0], r1 - vld1.u8 {d8, d9, d10}, [r0], r1 - vld1.u8 {d11, d12, d13}, [r0], r1 - - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q8, d3, d0 - vmull.u8 q9, d5, d0 - vmull.u8 q10, d6, d0 - vmull.u8 q11, d8, d0 - vmull.u8 q12, d9, d0 - vmull.u8 q13, d11, d0 - vmull.u8 q14, d12, d0 - - vext.8 d2, d2, d3, #1 ;construct src_ptr[1] - vext.8 d5, d5, d6, #1 - vext.8 d8, d8, d9, #1 - vext.8 d11, d11, d12, #1 - - vmlal.u8 q7, d2, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q9, d5, d1 - vmlal.u8 q11, d8, d1 - vmlal.u8 q13, d11, d1 - - vext.8 d3, d3, d4, #1 - vext.8 d6, d6, d7, #1 - vext.8 d9, d9, d10, #1 - vext.8 d12, d12, d13, #1 - - vmlal.u8 q8, d3, d1 ;(src_ptr[0] * vp8_filter[1]) - vmlal.u8 q10, d6, d1 - vmlal.u8 q12, d9, d1 - vmlal.u8 q14, d12, d1 - - subs r2, r2, #1 - - vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d15, q8, #7 - vqrshrn.u16 d16, q9, #7 - vqrshrn.u16 d17, q10, #7 - vqrshrn.u16 d18, q11, #7 - vqrshrn.u16 d19, q12, #7 - vqrshrn.u16 d20, q13, #7 - vst1.u8 {d14, d15}, [r4], r5 ;store result - vqrshrn.u16 d21, q14, #7 - - vst1.u8 {d16, d17}, [r4], r5 - vst1.u8 {d18, d19}, [r4], r5 - vst1.u8 {d20, d21}, [r4], r5 - - bne filt_blk2d_fpo16x16_loop_neon - pop {r4-r5,pc} - -;--------------------- -secondpass_bfilter16x16_only -;Second pass: 16x16 -;secondpass_filter - add r3, r12, r3, lsl #3 - mov r12, #4 ;loop counter - vld1.u32 {d31}, [r3] ;load second_pass filter - vld1.u8 {d22, d23}, [r0], r1 ;load src data - - vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1) - vdup.8 d1, d31[4] - -filt_blk2d_spo16x16_loop_neon - vld1.u8 {d24, d25}, [r0], r1 - vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0]) - vld1.u8 {d26, d27}, [r0], r1 - vmull.u8 q2, d23, d0 - vld1.u8 {d28, d29}, [r0], r1 - vmull.u8 q3, d24, d0 - vld1.u8 {d30, d31}, [r0], r1 - - vmull.u8 q4, d25, d0 - vmull.u8 q5, d26, d0 - vmull.u8 q6, d27, d0 - vmull.u8 q7, d28, d0 - vmull.u8 q8, d29, d0 - - vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * vp8_filter[1]) - vmlal.u8 q2, d25, d1 - vmlal.u8 q3, d26, d1 - vmlal.u8 q4, d27, d1 - vmlal.u8 q5, d28, d1 - vmlal.u8 q6, d29, d1 - vmlal.u8 q7, d30, d1 - vmlal.u8 q8, d31, d1 - - vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d3, q2, #7 - vqrshrn.u16 d4, q3, #7 - vqrshrn.u16 d5, q4, #7 - vqrshrn.u16 d6, q5, #7 - vqrshrn.u16 d7, q6, #7 - vqrshrn.u16 d8, q7, #7 - vqrshrn.u16 d9, q8, #7 - - vst1.u8 {d2, d3}, [r4], r5 ;store result - subs r12, r12, #1 - vst1.u8 {d4, d5}, [r4], r5 - vmov q11, q15 - vst1.u8 {d6, d7}, [r4], r5 - vst1.u8 {d8, d9}, [r4], r5 - - bne filt_blk2d_spo16x16_loop_neon - pop {r4-r5,pc} - - ENDP - -;----------------- - -bifilter16_coeff - DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112 - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict4x4_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict4x4_neon.asm deleted file mode 100644 index 0ac62436f97..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict4x4_neon.asm +++ /dev/null @@ -1,130 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_bilinear_predict4x4_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; r4 unsigned char *dst_ptr, -; stack(lr) int dst_pitch - -|vp8_bilinear_predict4x4_neon| PROC - push {r4, lr} - - adr r12, bifilter4_coeff - ldr r4, [sp, #8] ;load parameters from stack - ldr lr, [sp, #12] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq skip_firstpass_filter - -;First pass: output_height lines x output_width columns (5x4) - vld1.u8 {d2}, [r0], r1 ;load src data - add r2, r12, r2, lsl #3 ;calculate Hfilter location (2coeffsx4bytes=8bytes) - - vld1.u8 {d3}, [r0], r1 - vld1.u32 {d31}, [r2] ;first_pass filter - - vld1.u8 {d4}, [r0], r1 - vdup.8 d0, d31[0] ;first_pass filter (d0-d1) - vld1.u8 {d5}, [r0], r1 - vdup.8 d1, d31[4] - vld1.u8 {d6}, [r0], r1 - - vshr.u64 q4, q1, #8 ;construct src_ptr[1] - vshr.u64 q5, q2, #8 - vshr.u64 d12, d6, #8 - - vzip.32 d2, d3 ;put 2-line data in 1 register (src_ptr[0]) - vzip.32 d4, d5 - vzip.32 d8, d9 ;put 2-line data in 1 register (src_ptr[1]) - vzip.32 d10, d11 - - vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q8, d4, d0 - vmull.u8 q9, d6, d0 - - vmlal.u8 q7, d8, d1 ;(src_ptr[1] * vp8_filter[1]) - vmlal.u8 q8, d10, d1 - vmlal.u8 q9, d12, d1 - - vqrshrn.u16 d28, q7, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d29, q8, #7 - vqrshrn.u16 d30, q9, #7 - -;Second pass: 4x4 -secondpass_filter - cmp r3, #0 ;skip second_pass filter if yoffset=0 - beq skip_secondpass_filter - - add r3, r12, r3, lsl #3 ;calculate Vfilter location - vld1.u32 {d31}, [r3] ;load second_pass filter - - vdup.8 d0, d31[0] ;second_pass filter parameters (d0-d5) - vdup.8 d1, d31[4] - - vmull.u8 q1, d28, d0 - vmull.u8 q2, d29, d0 - - vext.8 d26, d28, d29, #4 ;construct src_ptr[pixel_step] - vext.8 d27, d29, d30, #4 - - vmlal.u8 q1, d26, d1 - vmlal.u8 q2, d27, d1 - - add r0, r4, lr - add r1, r0, lr - add r2, r1, lr - - vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d3, q2, #7 - - vst1.32 {d2[0]}, [r4] ;store result - vst1.32 {d2[1]}, [r0] - vst1.32 {d3[0]}, [r1] - vst1.32 {d3[1]}, [r2] - - pop {r4, pc} - -;-------------------- -skip_firstpass_filter - - vld1.32 {d28[0]}, [r0], r1 ;load src data - vld1.32 {d28[1]}, [r0], r1 - vld1.32 {d29[0]}, [r0], r1 - vld1.32 {d29[1]}, [r0], r1 - vld1.32 {d30[0]}, [r0], r1 - - b secondpass_filter - -;--------------------- -skip_secondpass_filter - vst1.32 {d28[0]}, [r4], lr ;store result - vst1.32 {d28[1]}, [r4], lr - vst1.32 {d29[0]}, [r4], lr - vst1.32 {d29[1]}, [r4], lr - - pop {r4, pc} - - ENDP - -;----------------- - -bifilter4_coeff - DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112 - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict8x4_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict8x4_neon.asm deleted file mode 100644 index 41f5c45ffe7..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict8x4_neon.asm +++ /dev/null @@ -1,135 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_bilinear_predict8x4_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; r4 unsigned char *dst_ptr, -; stack(lr) int dst_pitch - -|vp8_bilinear_predict8x4_neon| PROC - push {r4, lr} - - adr r12, bifilter8x4_coeff - ldr r4, [sp, #8] ;load parameters from stack - ldr lr, [sp, #12] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq skip_firstpass_filter - -;First pass: output_height lines x output_width columns (5x8) - add r2, r12, r2, lsl #3 ;calculate filter location - - vld1.u8 {q1}, [r0], r1 ;load src data - vld1.u32 {d31}, [r2] ;load first_pass filter - vld1.u8 {q2}, [r0], r1 - vdup.8 d0, d31[0] ;first_pass filter (d0 d1) - vld1.u8 {q3}, [r0], r1 - vdup.8 d1, d31[4] - vld1.u8 {q4}, [r0], r1 - - vmull.u8 q6, d2, d0 ;(src_ptr[0] * vp8_filter[0]) - vld1.u8 {q5}, [r0], r1 - vmull.u8 q7, d4, d0 - vmull.u8 q8, d6, d0 - vmull.u8 q9, d8, d0 - vmull.u8 q10, d10, d0 - - vext.8 d3, d2, d3, #1 ;construct src_ptr[-1] - vext.8 d5, d4, d5, #1 - vext.8 d7, d6, d7, #1 - vext.8 d9, d8, d9, #1 - vext.8 d11, d10, d11, #1 - - vmlal.u8 q6, d3, d1 ;(src_ptr[1] * vp8_filter[1]) - vmlal.u8 q7, d5, d1 - vmlal.u8 q8, d7, d1 - vmlal.u8 q9, d9, d1 - vmlal.u8 q10, d11, d1 - - vqrshrn.u16 d22, q6, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d23, q7, #7 - vqrshrn.u16 d24, q8, #7 - vqrshrn.u16 d25, q9, #7 - vqrshrn.u16 d26, q10, #7 - -;Second pass: 4x8 -secondpass_filter - cmp r3, #0 ;skip second_pass filter if yoffset=0 - beq skip_secondpass_filter - - add r3, r12, r3, lsl #3 - add r0, r4, lr - - vld1.u32 {d31}, [r3] ;load second_pass filter - add r1, r0, lr - - vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1) - vdup.8 d1, d31[4] - - vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q2, d23, d0 - vmull.u8 q3, d24, d0 - vmull.u8 q4, d25, d0 - - vmlal.u8 q1, d23, d1 ;(src_ptr[pixel_step] * vp8_filter[1]) - vmlal.u8 q2, d24, d1 - vmlal.u8 q3, d25, d1 - vmlal.u8 q4, d26, d1 - - add r2, r1, lr - - vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d3, q2, #7 - vqrshrn.u16 d4, q3, #7 - vqrshrn.u16 d5, q4, #7 - - vst1.u8 {d2}, [r4] ;store result - vst1.u8 {d3}, [r0] - vst1.u8 {d4}, [r1] - vst1.u8 {d5}, [r2] - - pop {r4, pc} - -;-------------------- -skip_firstpass_filter - vld1.u8 {d22}, [r0], r1 ;load src data - vld1.u8 {d23}, [r0], r1 - vld1.u8 {d24}, [r0], r1 - vld1.u8 {d25}, [r0], r1 - vld1.u8 {d26}, [r0], r1 - - b secondpass_filter - -;--------------------- -skip_secondpass_filter - vst1.u8 {d22}, [r4], lr ;store result - vst1.u8 {d23}, [r4], lr - vst1.u8 {d24}, [r4], lr - vst1.u8 {d25}, [r4], lr - - pop {r4, pc} - - ENDP - -;----------------- - -bifilter8x4_coeff - DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112 - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict8x8_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict8x8_neon.asm deleted file mode 100644 index c4711bc4d4a..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict8x8_neon.asm +++ /dev/null @@ -1,183 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_bilinear_predict8x8_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; r4 unsigned char *dst_ptr, -; stack(lr) int dst_pitch - -|vp8_bilinear_predict8x8_neon| PROC - push {r4, lr} - - adr r12, bifilter8_coeff - ldr r4, [sp, #8] ;load parameters from stack - ldr lr, [sp, #12] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq skip_firstpass_filter - -;First pass: output_height lines x output_width columns (9x8) - add r2, r12, r2, lsl #3 ;calculate filter location - - vld1.u8 {q1}, [r0], r1 ;load src data - vld1.u32 {d31}, [r2] ;load first_pass filter - vld1.u8 {q2}, [r0], r1 - vdup.8 d0, d31[0] ;first_pass filter (d0 d1) - vld1.u8 {q3}, [r0], r1 - vdup.8 d1, d31[4] - vld1.u8 {q4}, [r0], r1 - - vmull.u8 q6, d2, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q7, d4, d0 - vmull.u8 q8, d6, d0 - vmull.u8 q9, d8, d0 - - vext.8 d3, d2, d3, #1 ;construct src_ptr[-1] - vext.8 d5, d4, d5, #1 - vext.8 d7, d6, d7, #1 - vext.8 d9, d8, d9, #1 - - vmlal.u8 q6, d3, d1 ;(src_ptr[1] * vp8_filter[1]) - vmlal.u8 q7, d5, d1 - vmlal.u8 q8, d7, d1 - vmlal.u8 q9, d9, d1 - - vld1.u8 {q1}, [r0], r1 ;load src data - vqrshrn.u16 d22, q6, #7 ;shift/round/saturate to u8 - vld1.u8 {q2}, [r0], r1 - vqrshrn.u16 d23, q7, #7 - vld1.u8 {q3}, [r0], r1 - vqrshrn.u16 d24, q8, #7 - vld1.u8 {q4}, [r0], r1 - vqrshrn.u16 d25, q9, #7 - - ;first_pass filtering on the rest 5-line data - vld1.u8 {q5}, [r0], r1 - - vmull.u8 q6, d2, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q7, d4, d0 - vmull.u8 q8, d6, d0 - vmull.u8 q9, d8, d0 - vmull.u8 q10, d10, d0 - - vext.8 d3, d2, d3, #1 ;construct src_ptr[-1] - vext.8 d5, d4, d5, #1 - vext.8 d7, d6, d7, #1 - vext.8 d9, d8, d9, #1 - vext.8 d11, d10, d11, #1 - - vmlal.u8 q6, d3, d1 ;(src_ptr[1] * vp8_filter[1]) - vmlal.u8 q7, d5, d1 - vmlal.u8 q8, d7, d1 - vmlal.u8 q9, d9, d1 - vmlal.u8 q10, d11, d1 - - vqrshrn.u16 d26, q6, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d27, q7, #7 - vqrshrn.u16 d28, q8, #7 - vqrshrn.u16 d29, q9, #7 - vqrshrn.u16 d30, q10, #7 - -;Second pass: 8x8 -secondpass_filter - cmp r3, #0 ;skip second_pass filter if yoffset=0 - beq skip_secondpass_filter - - add r3, r12, r3, lsl #3 - add r0, r4, lr - - vld1.u32 {d31}, [r3] ;load second_pass filter - add r1, r0, lr - - vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1) - vdup.8 d1, d31[4] - - vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0]) - vmull.u8 q2, d23, d0 - vmull.u8 q3, d24, d0 - vmull.u8 q4, d25, d0 - vmull.u8 q5, d26, d0 - vmull.u8 q6, d27, d0 - vmull.u8 q7, d28, d0 - vmull.u8 q8, d29, d0 - - vmlal.u8 q1, d23, d1 ;(src_ptr[pixel_step] * vp8_filter[1]) - vmlal.u8 q2, d24, d1 - vmlal.u8 q3, d25, d1 - vmlal.u8 q4, d26, d1 - vmlal.u8 q5, d27, d1 - vmlal.u8 q6, d28, d1 - vmlal.u8 q7, d29, d1 - vmlal.u8 q8, d30, d1 - - vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8 - vqrshrn.u16 d3, q2, #7 - vqrshrn.u16 d4, q3, #7 - vqrshrn.u16 d5, q4, #7 - vqrshrn.u16 d6, q5, #7 - vqrshrn.u16 d7, q6, #7 - vqrshrn.u16 d8, q7, #7 - vqrshrn.u16 d9, q8, #7 - - vst1.u8 {d2}, [r4] ;store result - vst1.u8 {d3}, [r0] - vst1.u8 {d4}, [r1], lr - vst1.u8 {d5}, [r1], lr - vst1.u8 {d6}, [r1], lr - vst1.u8 {d7}, [r1], lr - vst1.u8 {d8}, [r1], lr - vst1.u8 {d9}, [r1], lr - - pop {r4, pc} - -;-------------------- -skip_firstpass_filter - vld1.u8 {d22}, [r0], r1 ;load src data - vld1.u8 {d23}, [r0], r1 - vld1.u8 {d24}, [r0], r1 - vld1.u8 {d25}, [r0], r1 - vld1.u8 {d26}, [r0], r1 - vld1.u8 {d27}, [r0], r1 - vld1.u8 {d28}, [r0], r1 - vld1.u8 {d29}, [r0], r1 - vld1.u8 {d30}, [r0], r1 - - b secondpass_filter - -;--------------------- -skip_secondpass_filter - vst1.u8 {d22}, [r4], lr ;store result - vst1.u8 {d23}, [r4], lr - vst1.u8 {d24}, [r4], lr - vst1.u8 {d25}, [r4], lr - vst1.u8 {d26}, [r4], lr - vst1.u8 {d27}, [r4], lr - vst1.u8 {d28}, [r4], lr - vst1.u8 {d29}, [r4], lr - - pop {r4, pc} - - ENDP - -;----------------- - -bifilter8_coeff - DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112 - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict_neon.c new file mode 100644 index 00000000000..e1c3c2be7df --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/bilinearpredict_neon.c @@ -0,0 +1,696 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +static const uint16_t bifilter4_coeff[8][2] = { + {128, 0}, + {112, 16}, + { 96, 32}, + { 80, 48}, + { 64, 64}, + { 48, 80}, + { 32, 96}, + { 16, 112} +}; + +void vp8_bilinear_predict4x4_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8; + uint8x8_t d26u8, d27u8, d28u8, d29u8, d30u8; + uint32x2_t d28u32, d29u32, d30u32; + uint8x16_t q1u8, q2u8; + uint16x8_t q1u16, q2u16; + uint16x8_t q7u16, q8u16, q9u16; + uint64x2_t q4u64, q5u64; + uint64x1_t d12u64; + uint32x2x2_t d0u32x2, d1u32x2, d2u32x2, d3u32x2; + + if (xoffset == 0) { // skip_1stpass_filter + d28u32 = vld1_lane_u32((const uint32_t *)src_ptr, d28u32, 0); + src_ptr += src_pixels_per_line; + d28u32 = vld1_lane_u32((const uint32_t *)src_ptr, d28u32, 1); + src_ptr += src_pixels_per_line; + d29u32 = vld1_lane_u32((const uint32_t *)src_ptr, d29u32, 0); + src_ptr += src_pixels_per_line; + d29u32 = vld1_lane_u32((const uint32_t *)src_ptr, d29u32, 1); + src_ptr += src_pixels_per_line; + d30u32 = vld1_lane_u32((const uint32_t *)src_ptr, d30u32, 0); + d28u8 = vreinterpret_u8_u32(d28u32); + d29u8 = vreinterpret_u8_u32(d29u32); + d30u8 = vreinterpret_u8_u32(d30u32); + } else { + d2u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d3u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d4u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d5u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d6u8 = vld1_u8(src_ptr); + + q1u8 = vcombine_u8(d2u8, d3u8); + q2u8 = vcombine_u8(d4u8, d5u8); + + d0u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][1]); + + q4u64 = vshrq_n_u64(vreinterpretq_u64_u8(q1u8), 8); + q5u64 = vshrq_n_u64(vreinterpretq_u64_u8(q2u8), 8); + d12u64 = vshr_n_u64(vreinterpret_u64_u8(d6u8), 8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q1u8)), + vreinterpret_u32_u8(vget_high_u8(q1u8))); + d1u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q2u8)), + vreinterpret_u32_u8(vget_high_u8(q2u8))); + d2u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q4u64)), + vreinterpret_u32_u64(vget_high_u64(q4u64))); + d3u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q5u64)), + vreinterpret_u32_u64(vget_high_u64(q5u64))); + + q7u16 = vmull_u8(vreinterpret_u8_u32(d0u32x2.val[0]), d0u8); + q8u16 = vmull_u8(vreinterpret_u8_u32(d1u32x2.val[0]), d0u8); + q9u16 = vmull_u8(d6u8, d0u8); + + q7u16 = vmlal_u8(q7u16, vreinterpret_u8_u32(d2u32x2.val[0]), d1u8); + q8u16 = vmlal_u8(q8u16, vreinterpret_u8_u32(d3u32x2.val[0]), d1u8); + q9u16 = vmlal_u8(q9u16, vreinterpret_u8_u64(d12u64), d1u8); + + d28u8 = vqrshrn_n_u16(q7u16, 7); + d29u8 = vqrshrn_n_u16(q8u16, 7); + d30u8 = vqrshrn_n_u16(q9u16, 7); + } + + // secondpass_filter + if (yoffset == 0) { // skip_2ndpass_filter + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d28u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d28u8), 1); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d29u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d29u8), 1); + } else { + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + q1u16 = vmull_u8(d28u8, d0u8); + q2u16 = vmull_u8(d29u8, d0u8); + + d26u8 = vext_u8(d28u8, d29u8, 4); + d27u8 = vext_u8(d29u8, d30u8, 4); + + q1u16 = vmlal_u8(q1u16, d26u8, d1u8); + q2u16 = vmlal_u8(q2u16, d27u8, d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 1); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 1); + } + return; +} + +void vp8_bilinear_predict8x4_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8; + uint8x8_t d7u8, d9u8, d11u8, d22u8, d23u8, d24u8, d25u8, d26u8; + uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8; + uint16x8_t q1u16, q2u16, q3u16, q4u16; + uint16x8_t q6u16, q7u16, q8u16, q9u16, q10u16; + + if (xoffset == 0) { // skip_1stpass_filter + d22u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d23u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d24u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d25u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d26u8 = vld1_u8(src_ptr); + } else { + q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q5u8 = vld1q_u8(src_ptr); + + d0u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][1]); + + q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + + d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1); + d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1); + d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d11u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + + q6u16 = vmlal_u8(q6u16, d3u8, d1u8); + q7u16 = vmlal_u8(q7u16, d5u8, d1u8); + q8u16 = vmlal_u8(q8u16, d7u8, d1u8); + q9u16 = vmlal_u8(q9u16, d9u8, d1u8); + q10u16 = vmlal_u8(q10u16, d11u8, d1u8); + + d22u8 = vqrshrn_n_u16(q6u16, 7); + d23u8 = vqrshrn_n_u16(q7u16, 7); + d24u8 = vqrshrn_n_u16(q8u16, 7); + d25u8 = vqrshrn_n_u16(q9u16, 7); + d26u8 = vqrshrn_n_u16(q10u16, 7); + } + + // secondpass_filter + if (yoffset == 0) { // skip_2ndpass_filter + vst1_u8((uint8_t *)dst_ptr, d22u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d23u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d24u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d25u8); + } else { + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + q1u16 = vmull_u8(d22u8, d0u8); + q2u16 = vmull_u8(d23u8, d0u8); + q3u16 = vmull_u8(d24u8, d0u8); + q4u16 = vmull_u8(d25u8, d0u8); + + q1u16 = vmlal_u8(q1u16, d23u8, d1u8); + q2u16 = vmlal_u8(q2u16, d24u8, d1u8); + q3u16 = vmlal_u8(q3u16, d25u8, d1u8); + q4u16 = vmlal_u8(q4u16, d26u8, d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + + vst1_u8((uint8_t *)dst_ptr, d2u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d3u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d4u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d5u8); + } + return; +} + +void vp8_bilinear_predict8x8_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8, d11u8; + uint8x8_t d22u8, d23u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8, d30u8; + uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8; + uint16x8_t q1u16, q2u16, q3u16, q4u16, q5u16; + uint16x8_t q6u16, q7u16, q8u16, q9u16, q10u16; + + if (xoffset == 0) { // skip_1stpass_filter + d22u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d23u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d24u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d25u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d26u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d27u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d28u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d29u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line; + d30u8 = vld1_u8(src_ptr); + } else { + q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + + d0u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][1]); + + q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + + d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1); + d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1); + d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + + q6u16 = vmlal_u8(q6u16, d3u8, d1u8); + q7u16 = vmlal_u8(q7u16, d5u8, d1u8); + q8u16 = vmlal_u8(q8u16, d7u8, d1u8); + q9u16 = vmlal_u8(q9u16, d9u8, d1u8); + + d22u8 = vqrshrn_n_u16(q6u16, 7); + d23u8 = vqrshrn_n_u16(q7u16, 7); + d24u8 = vqrshrn_n_u16(q8u16, 7); + d25u8 = vqrshrn_n_u16(q9u16, 7); + + // first_pass filtering on the rest 5-line data + q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q5u8 = vld1q_u8(src_ptr); + + q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + + d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1); + d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1); + d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d11u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + + q6u16 = vmlal_u8(q6u16, d3u8, d1u8); + q7u16 = vmlal_u8(q7u16, d5u8, d1u8); + q8u16 = vmlal_u8(q8u16, d7u8, d1u8); + q9u16 = vmlal_u8(q9u16, d9u8, d1u8); + q10u16 = vmlal_u8(q10u16, d11u8, d1u8); + + d26u8 = vqrshrn_n_u16(q6u16, 7); + d27u8 = vqrshrn_n_u16(q7u16, 7); + d28u8 = vqrshrn_n_u16(q8u16, 7); + d29u8 = vqrshrn_n_u16(q9u16, 7); + d30u8 = vqrshrn_n_u16(q10u16, 7); + } + + // secondpass_filter + if (yoffset == 0) { // skip_2ndpass_filter + vst1_u8((uint8_t *)dst_ptr, d22u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d23u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d24u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d25u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d26u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d27u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d28u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d29u8); + } else { + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + q1u16 = vmull_u8(d22u8, d0u8); + q2u16 = vmull_u8(d23u8, d0u8); + q3u16 = vmull_u8(d24u8, d0u8); + q4u16 = vmull_u8(d25u8, d0u8); + q5u16 = vmull_u8(d26u8, d0u8); + q6u16 = vmull_u8(d27u8, d0u8); + q7u16 = vmull_u8(d28u8, d0u8); + q8u16 = vmull_u8(d29u8, d0u8); + + q1u16 = vmlal_u8(q1u16, d23u8, d1u8); + q2u16 = vmlal_u8(q2u16, d24u8, d1u8); + q3u16 = vmlal_u8(q3u16, d25u8, d1u8); + q4u16 = vmlal_u8(q4u16, d26u8, d1u8); + q5u16 = vmlal_u8(q5u16, d27u8, d1u8); + q6u16 = vmlal_u8(q6u16, d28u8, d1u8); + q7u16 = vmlal_u8(q7u16, d29u8, d1u8); + q8u16 = vmlal_u8(q8u16, d30u8, d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + d6u8 = vqrshrn_n_u16(q5u16, 7); + d7u8 = vqrshrn_n_u16(q6u16, 7); + d8u8 = vqrshrn_n_u16(q7u16, 7); + d9u8 = vqrshrn_n_u16(q8u16, 7); + + vst1_u8((uint8_t *)dst_ptr, d2u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d3u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d4u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d5u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d6u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d7u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d8u8); dst_ptr += dst_pitch; + vst1_u8((uint8_t *)dst_ptr, d9u8); + } + return; +} + +void vp8_bilinear_predict16x16_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + int i; + unsigned char tmp[272]; + unsigned char *tmpp; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d10u8, d11u8, d12u8, d13u8, d14u8, d15u8, d16u8, d17u8, d18u8; + uint8x8_t d19u8, d20u8, d21u8; + uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8, q6u8, q7u8, q8u8, q9u8, q10u8; + uint8x16_t q11u8, q12u8, q13u8, q14u8, q15u8; + uint16x8_t q1u16, q2u16, q3u16, q4u16, q5u16, q6u16, q7u16, q8u16; + uint16x8_t q9u16, q10u16, q11u16, q12u16, q13u16, q14u16; + + if (xoffset == 0) { // secondpass_bfilter16x16_only + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + q11u8 = vld1q_u8(src_ptr); + src_ptr += src_pixels_per_line; + for (i = 4; i > 0; i--) { + q12u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q13u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q14u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + q15u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line; + + q1u16 = vmull_u8(vget_low_u8(q11u8), d0u8); + q2u16 = vmull_u8(vget_high_u8(q11u8), d0u8); + q3u16 = vmull_u8(vget_low_u8(q12u8), d0u8); + q4u16 = vmull_u8(vget_high_u8(q12u8), d0u8); + q5u16 = vmull_u8(vget_low_u8(q13u8), d0u8); + q6u16 = vmull_u8(vget_high_u8(q13u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q14u8), d0u8); + q8u16 = vmull_u8(vget_high_u8(q14u8), d0u8); + + q1u16 = vmlal_u8(q1u16, vget_low_u8(q12u8), d1u8); + q2u16 = vmlal_u8(q2u16, vget_high_u8(q12u8), d1u8); + q3u16 = vmlal_u8(q3u16, vget_low_u8(q13u8), d1u8); + q4u16 = vmlal_u8(q4u16, vget_high_u8(q13u8), d1u8); + q5u16 = vmlal_u8(q5u16, vget_low_u8(q14u8), d1u8); + q6u16 = vmlal_u8(q6u16, vget_high_u8(q14u8), d1u8); + q7u16 = vmlal_u8(q7u16, vget_low_u8(q15u8), d1u8); + q8u16 = vmlal_u8(q8u16, vget_high_u8(q15u8), d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + d6u8 = vqrshrn_n_u16(q5u16, 7); + d7u8 = vqrshrn_n_u16(q6u16, 7); + d8u8 = vqrshrn_n_u16(q7u16, 7); + d9u8 = vqrshrn_n_u16(q8u16, 7); + + q1u8 = vcombine_u8(d2u8, d3u8); + q2u8 = vcombine_u8(d4u8, d5u8); + q3u8 = vcombine_u8(d6u8, d7u8); + q4u8 = vcombine_u8(d8u8, d9u8); + + q11u8 = q15u8; + + vst1q_u8((uint8_t *)dst_ptr, q1u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q2u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q3u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q4u8); dst_ptr += dst_pitch; + } + return; + } + + if (yoffset == 0) { // firstpass_bfilter16x16_only + d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]); + + for (i = 4; i > 0 ; i--) { + d2u8 = vld1_u8(src_ptr); + d3u8 = vld1_u8(src_ptr + 8); + d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d5u8 = vld1_u8(src_ptr); + d6u8 = vld1_u8(src_ptr + 8); + d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d8u8 = vld1_u8(src_ptr); + d9u8 = vld1_u8(src_ptr + 8); + d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d11u8 = vld1_u8(src_ptr); + d12u8 = vld1_u8(src_ptr + 8); + d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + q7u16 = vmull_u8(d2u8, d0u8); + q8u16 = vmull_u8(d3u8, d0u8); + q9u16 = vmull_u8(d5u8, d0u8); + q10u16 = vmull_u8(d6u8, d0u8); + q11u16 = vmull_u8(d8u8, d0u8); + q12u16 = vmull_u8(d9u8, d0u8); + q13u16 = vmull_u8(d11u8, d0u8); + q14u16 = vmull_u8(d12u8, d0u8); + + d2u8 = vext_u8(d2u8, d3u8, 1); + d5u8 = vext_u8(d5u8, d6u8, 1); + d8u8 = vext_u8(d8u8, d9u8, 1); + d11u8 = vext_u8(d11u8, d12u8, 1); + + q7u16 = vmlal_u8(q7u16, d2u8, d1u8); + q9u16 = vmlal_u8(q9u16, d5u8, d1u8); + q11u16 = vmlal_u8(q11u16, d8u8, d1u8); + q13u16 = vmlal_u8(q13u16, d11u8, d1u8); + + d3u8 = vext_u8(d3u8, d4u8, 1); + d6u8 = vext_u8(d6u8, d7u8, 1); + d9u8 = vext_u8(d9u8, d10u8, 1); + d12u8 = vext_u8(d12u8, d13u8, 1); + + q8u16 = vmlal_u8(q8u16, d3u8, d1u8); + q10u16 = vmlal_u8(q10u16, d6u8, d1u8); + q12u16 = vmlal_u8(q12u16, d9u8, d1u8); + q14u16 = vmlal_u8(q14u16, d12u8, d1u8); + + d14u8 = vqrshrn_n_u16(q7u16, 7); + d15u8 = vqrshrn_n_u16(q8u16, 7); + d16u8 = vqrshrn_n_u16(q9u16, 7); + d17u8 = vqrshrn_n_u16(q10u16, 7); + d18u8 = vqrshrn_n_u16(q11u16, 7); + d19u8 = vqrshrn_n_u16(q12u16, 7); + d20u8 = vqrshrn_n_u16(q13u16, 7); + d21u8 = vqrshrn_n_u16(q14u16, 7); + + q7u8 = vcombine_u8(d14u8, d15u8); + q8u8 = vcombine_u8(d16u8, d17u8); + q9u8 = vcombine_u8(d18u8, d19u8); + q10u8 =vcombine_u8(d20u8, d21u8); + + vst1q_u8((uint8_t *)dst_ptr, q7u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q8u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q9u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q10u8); dst_ptr += dst_pitch; + } + return; + } + + d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]); + + d2u8 = vld1_u8(src_ptr); + d3u8 = vld1_u8(src_ptr + 8); + d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d5u8 = vld1_u8(src_ptr); + d6u8 = vld1_u8(src_ptr + 8); + d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d8u8 = vld1_u8(src_ptr); + d9u8 = vld1_u8(src_ptr + 8); + d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d11u8 = vld1_u8(src_ptr); + d12u8 = vld1_u8(src_ptr + 8); + d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + // First Pass: output_height lines x output_width columns (17x16) + tmpp = tmp; + for (i = 3; i > 0; i--) { + q7u16 = vmull_u8(d2u8, d0u8); + q8u16 = vmull_u8(d3u8, d0u8); + q9u16 = vmull_u8(d5u8, d0u8); + q10u16 = vmull_u8(d6u8, d0u8); + q11u16 = vmull_u8(d8u8, d0u8); + q12u16 = vmull_u8(d9u8, d0u8); + q13u16 = vmull_u8(d11u8, d0u8); + q14u16 = vmull_u8(d12u8, d0u8); + + d2u8 = vext_u8(d2u8, d3u8, 1); + d5u8 = vext_u8(d5u8, d6u8, 1); + d8u8 = vext_u8(d8u8, d9u8, 1); + d11u8 = vext_u8(d11u8, d12u8, 1); + + q7u16 = vmlal_u8(q7u16, d2u8, d1u8); + q9u16 = vmlal_u8(q9u16, d5u8, d1u8); + q11u16 = vmlal_u8(q11u16, d8u8, d1u8); + q13u16 = vmlal_u8(q13u16, d11u8, d1u8); + + d3u8 = vext_u8(d3u8, d4u8, 1); + d6u8 = vext_u8(d6u8, d7u8, 1); + d9u8 = vext_u8(d9u8, d10u8, 1); + d12u8 = vext_u8(d12u8, d13u8, 1); + + q8u16 = vmlal_u8(q8u16, d3u8, d1u8); + q10u16 = vmlal_u8(q10u16, d6u8, d1u8); + q12u16 = vmlal_u8(q12u16, d9u8, d1u8); + q14u16 = vmlal_u8(q14u16, d12u8, d1u8); + + d14u8 = vqrshrn_n_u16(q7u16, 7); + d15u8 = vqrshrn_n_u16(q8u16, 7); + d16u8 = vqrshrn_n_u16(q9u16, 7); + d17u8 = vqrshrn_n_u16(q10u16, 7); + d18u8 = vqrshrn_n_u16(q11u16, 7); + d19u8 = vqrshrn_n_u16(q12u16, 7); + d20u8 = vqrshrn_n_u16(q13u16, 7); + d21u8 = vqrshrn_n_u16(q14u16, 7); + + d2u8 = vld1_u8(src_ptr); + d3u8 = vld1_u8(src_ptr + 8); + d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d5u8 = vld1_u8(src_ptr); + d6u8 = vld1_u8(src_ptr + 8); + d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d8u8 = vld1_u8(src_ptr); + d9u8 = vld1_u8(src_ptr + 8); + d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + d11u8 = vld1_u8(src_ptr); + d12u8 = vld1_u8(src_ptr + 8); + d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + q7u8 = vcombine_u8(d14u8, d15u8); + q8u8 = vcombine_u8(d16u8, d17u8); + q9u8 = vcombine_u8(d18u8, d19u8); + q10u8 = vcombine_u8(d20u8, d21u8); + + vst1q_u8((uint8_t *)tmpp, q7u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q8u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q9u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q10u8); tmpp += 16; + } + + // First-pass filtering for rest 5 lines + d14u8 = vld1_u8(src_ptr); + d15u8 = vld1_u8(src_ptr + 8); + d16u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line; + + q9u16 = vmull_u8(d2u8, d0u8); + q10u16 = vmull_u8(d3u8, d0u8); + q11u16 = vmull_u8(d5u8, d0u8); + q12u16 = vmull_u8(d6u8, d0u8); + q13u16 = vmull_u8(d8u8, d0u8); + q14u16 = vmull_u8(d9u8, d0u8); + + d2u8 = vext_u8(d2u8, d3u8, 1); + d5u8 = vext_u8(d5u8, d6u8, 1); + d8u8 = vext_u8(d8u8, d9u8, 1); + + q9u16 = vmlal_u8(q9u16, d2u8, d1u8); + q11u16 = vmlal_u8(q11u16, d5u8, d1u8); + q13u16 = vmlal_u8(q13u16, d8u8, d1u8); + + d3u8 = vext_u8(d3u8, d4u8, 1); + d6u8 = vext_u8(d6u8, d7u8, 1); + d9u8 = vext_u8(d9u8, d10u8, 1); + + q10u16 = vmlal_u8(q10u16, d3u8, d1u8); + q12u16 = vmlal_u8(q12u16, d6u8, d1u8); + q14u16 = vmlal_u8(q14u16, d9u8, d1u8); + + q1u16 = vmull_u8(d11u8, d0u8); + q2u16 = vmull_u8(d12u8, d0u8); + q3u16 = vmull_u8(d14u8, d0u8); + q4u16 = vmull_u8(d15u8, d0u8); + + d11u8 = vext_u8(d11u8, d12u8, 1); + d14u8 = vext_u8(d14u8, d15u8, 1); + + q1u16 = vmlal_u8(q1u16, d11u8, d1u8); + q3u16 = vmlal_u8(q3u16, d14u8, d1u8); + + d12u8 = vext_u8(d12u8, d13u8, 1); + d15u8 = vext_u8(d15u8, d16u8, 1); + + q2u16 = vmlal_u8(q2u16, d12u8, d1u8); + q4u16 = vmlal_u8(q4u16, d15u8, d1u8); + + d10u8 = vqrshrn_n_u16(q9u16, 7); + d11u8 = vqrshrn_n_u16(q10u16, 7); + d12u8 = vqrshrn_n_u16(q11u16, 7); + d13u8 = vqrshrn_n_u16(q12u16, 7); + d14u8 = vqrshrn_n_u16(q13u16, 7); + d15u8 = vqrshrn_n_u16(q14u16, 7); + d16u8 = vqrshrn_n_u16(q1u16, 7); + d17u8 = vqrshrn_n_u16(q2u16, 7); + d18u8 = vqrshrn_n_u16(q3u16, 7); + d19u8 = vqrshrn_n_u16(q4u16, 7); + + q5u8 = vcombine_u8(d10u8, d11u8); + q6u8 = vcombine_u8(d12u8, d13u8); + q7u8 = vcombine_u8(d14u8, d15u8); + q8u8 = vcombine_u8(d16u8, d17u8); + q9u8 = vcombine_u8(d18u8, d19u8); + + vst1q_u8((uint8_t *)tmpp, q5u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q6u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q7u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q8u8); tmpp += 16; + vst1q_u8((uint8_t *)tmpp, q9u8); + + // secondpass_filter + d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]); + d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]); + + tmpp = tmp; + q11u8 = vld1q_u8(tmpp); + tmpp += 16; + for (i = 4; i > 0; i--) { + q12u8 = vld1q_u8(tmpp); tmpp += 16; + q13u8 = vld1q_u8(tmpp); tmpp += 16; + q14u8 = vld1q_u8(tmpp); tmpp += 16; + q15u8 = vld1q_u8(tmpp); tmpp += 16; + + q1u16 = vmull_u8(vget_low_u8(q11u8), d0u8); + q2u16 = vmull_u8(vget_high_u8(q11u8), d0u8); + q3u16 = vmull_u8(vget_low_u8(q12u8), d0u8); + q4u16 = vmull_u8(vget_high_u8(q12u8), d0u8); + q5u16 = vmull_u8(vget_low_u8(q13u8), d0u8); + q6u16 = vmull_u8(vget_high_u8(q13u8), d0u8); + q7u16 = vmull_u8(vget_low_u8(q14u8), d0u8); + q8u16 = vmull_u8(vget_high_u8(q14u8), d0u8); + + q1u16 = vmlal_u8(q1u16, vget_low_u8(q12u8), d1u8); + q2u16 = vmlal_u8(q2u16, vget_high_u8(q12u8), d1u8); + q3u16 = vmlal_u8(q3u16, vget_low_u8(q13u8), d1u8); + q4u16 = vmlal_u8(q4u16, vget_high_u8(q13u8), d1u8); + q5u16 = vmlal_u8(q5u16, vget_low_u8(q14u8), d1u8); + q6u16 = vmlal_u8(q6u16, vget_high_u8(q14u8), d1u8); + q7u16 = vmlal_u8(q7u16, vget_low_u8(q15u8), d1u8); + q8u16 = vmlal_u8(q8u16, vget_high_u8(q15u8), d1u8); + + d2u8 = vqrshrn_n_u16(q1u16, 7); + d3u8 = vqrshrn_n_u16(q2u16, 7); + d4u8 = vqrshrn_n_u16(q3u16, 7); + d5u8 = vqrshrn_n_u16(q4u16, 7); + d6u8 = vqrshrn_n_u16(q5u16, 7); + d7u8 = vqrshrn_n_u16(q6u16, 7); + d8u8 = vqrshrn_n_u16(q7u16, 7); + d9u8 = vqrshrn_n_u16(q8u16, 7); + + q1u8 = vcombine_u8(d2u8, d3u8); + q2u8 = vcombine_u8(d4u8, d5u8); + q3u8 = vcombine_u8(d6u8, d7u8); + q4u8 = vcombine_u8(d8u8, d9u8); + + q11u8 = q15u8; + + vst1q_u8((uint8_t *)dst_ptr, q1u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q2u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q3u8); dst_ptr += dst_pitch; + vst1q_u8((uint8_t *)dst_ptr, q4u8); dst_ptr += dst_pitch; + } + return; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/buildintrapredictorsmby_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/buildintrapredictorsmby_neon.asm index e3ea91fe6c0..a8730aa04ef 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/buildintrapredictorsmby_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/buildintrapredictorsmby_neon.asm @@ -26,6 +26,7 @@ |vp8_build_intra_predictors_mby_neon_func| PROC push {r4-r8, lr} + vpush {d8-d15} cmp r3, #0 beq case_dc_pred @@ -37,8 +38,8 @@ beq case_tm_pred case_dc_pred - ldr r4, [sp, #24] ; Up - ldr r5, [sp, #28] ; Left + ldr r4, [sp, #88] ; Up + ldr r5, [sp, #92] ; Left ; Default the DC average to 128 mov r12, #128 @@ -143,6 +144,7 @@ skip_dc_pred_up_left vst1.u8 {q0}, [r1]! vst1.u8 {q0}, [r1]! + vpop {d8-d15} pop {r4-r8,pc} case_v_pred ; Copy down above row @@ -165,6 +167,7 @@ case_v_pred vst1.u8 {q0}, [r1]! vst1.u8 {q0}, [r1]! vst1.u8 {q0}, [r1]! + vpop {d8-d15} pop {r4-r8,pc} case_h_pred @@ -224,6 +227,7 @@ case_h_pred vst1.u8 {q2}, [r1]! vst1.u8 {q3}, [r1]! + vpop {d8-d15} pop {r4-r8,pc} case_tm_pred @@ -293,6 +297,7 @@ case_tm_pred_loop subs r12, r12, #1 bne case_tm_pred_loop + vpop {d8-d15} pop {r4-r8,pc} ENDP @@ -307,6 +312,7 @@ case_tm_pred_loop |vp8_build_intra_predictors_mby_s_neon_func| PROC push {r4-r8, lr} + vpush {d8-d15} mov r1, r0 ; unsigned char *ypred_ptr = x->dst.y_buffer; //x->Predictor; @@ -320,8 +326,8 @@ case_tm_pred_loop beq case_tm_pred_s case_dc_pred_s - ldr r4, [sp, #24] ; Up - ldr r5, [sp, #28] ; Left + ldr r4, [sp, #88] ; Up + ldr r5, [sp, #92] ; Left ; Default the DC average to 128 mov r12, #128 @@ -426,6 +432,7 @@ skip_dc_pred_up_left_s vst1.u8 {q0}, [r1], r2 vst1.u8 {q0}, [r1], r2 + vpop {d8-d15} pop {r4-r8,pc} case_v_pred_s ; Copy down above row @@ -448,6 +455,8 @@ case_v_pred_s vst1.u8 {q0}, [r1], r2 vst1.u8 {q0}, [r1], r2 vst1.u8 {q0}, [r1], r2 + + vpop {d8-d15} pop {r4-r8,pc} case_h_pred_s @@ -507,6 +516,7 @@ case_h_pred_s vst1.u8 {q2}, [r1], r2 vst1.u8 {q3}, [r1], r2 + vpop {d8-d15} pop {r4-r8,pc} case_tm_pred_s @@ -576,6 +586,7 @@ case_tm_pred_loop_s subs r12, r12, #1 bne case_tm_pred_loop_s + vpop {d8-d15} pop {r4-r8,pc} ENDP diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem16x16_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem16x16_neon.asm deleted file mode 100644 index bda4b965442..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem16x16_neon.asm +++ /dev/null @@ -1,59 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_copy_mem16x16_neon| - ; ARM - ; REQUIRE8 - ; PRESERVE8 - - AREA Block, CODE, READONLY ; name this block of code -;void copy_mem16x16_neon( unsigned char *src, int src_stride, unsigned char *dst, int dst_stride) -;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= -|vp8_copy_mem16x16_neon| PROC - - vld1.u8 {q0}, [r0], r1 - vld1.u8 {q1}, [r0], r1 - vld1.u8 {q2}, [r0], r1 - vst1.u8 {q0}, [r2], r3 - vld1.u8 {q3}, [r0], r1 - vst1.u8 {q1}, [r2], r3 - vld1.u8 {q4}, [r0], r1 - vst1.u8 {q2}, [r2], r3 - vld1.u8 {q5}, [r0], r1 - vst1.u8 {q3}, [r2], r3 - vld1.u8 {q6}, [r0], r1 - vst1.u8 {q4}, [r2], r3 - vld1.u8 {q7}, [r0], r1 - vst1.u8 {q5}, [r2], r3 - vld1.u8 {q8}, [r0], r1 - vst1.u8 {q6}, [r2], r3 - vld1.u8 {q9}, [r0], r1 - vst1.u8 {q7}, [r2], r3 - vld1.u8 {q10}, [r0], r1 - vst1.u8 {q8}, [r2], r3 - vld1.u8 {q11}, [r0], r1 - vst1.u8 {q9}, [r2], r3 - vld1.u8 {q12}, [r0], r1 - vst1.u8 {q10}, [r2], r3 - vld1.u8 {q13}, [r0], r1 - vst1.u8 {q11}, [r2], r3 - vld1.u8 {q14}, [r0], r1 - vst1.u8 {q12}, [r2], r3 - vld1.u8 {q15}, [r0], r1 - vst1.u8 {q13}, [r2], r3 - vst1.u8 {q14}, [r2], r3 - vst1.u8 {q15}, [r2], r3 - - mov pc, lr - - ENDP ; |vp8_copy_mem16x16_neon| - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem8x4_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem8x4_neon.asm deleted file mode 100644 index 35c0f6708a5..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem8x4_neon.asm +++ /dev/null @@ -1,34 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_copy_mem8x4_neon| - ; ARM - ; REQUIRE8 - ; PRESERVE8 - - AREA Block, CODE, READONLY ; name this block of code -;void copy_mem8x4_neon( unsigned char *src, int src_stride, unsigned char *dst, int dst_stride) -;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= -|vp8_copy_mem8x4_neon| PROC - vld1.u8 {d0}, [r0], r1 - vld1.u8 {d1}, [r0], r1 - vst1.u8 {d0}, [r2], r3 - vld1.u8 {d2}, [r0], r1 - vst1.u8 {d1}, [r2], r3 - vld1.u8 {d3}, [r0], r1 - vst1.u8 {d2}, [r2], r3 - vst1.u8 {d3}, [r2], r3 - - mov pc, lr - - ENDP ; |vp8_copy_mem8x4_neon| - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem8x8_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem8x8_neon.asm deleted file mode 100644 index 1f5b9411bb5..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem8x8_neon.asm +++ /dev/null @@ -1,43 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_copy_mem8x8_neon| - ; ARM - ; REQUIRE8 - ; PRESERVE8 - - AREA Block, CODE, READONLY ; name this block of code -;void copy_mem8x8_neon( unsigned char *src, int src_stride, unsigned char *dst, int dst_stride) -;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= -|vp8_copy_mem8x8_neon| PROC - - vld1.u8 {d0}, [r0], r1 - vld1.u8 {d1}, [r0], r1 - vst1.u8 {d0}, [r2], r3 - vld1.u8 {d2}, [r0], r1 - vst1.u8 {d1}, [r2], r3 - vld1.u8 {d3}, [r0], r1 - vst1.u8 {d2}, [r2], r3 - vld1.u8 {d4}, [r0], r1 - vst1.u8 {d3}, [r2], r3 - vld1.u8 {d5}, [r0], r1 - vst1.u8 {d4}, [r2], r3 - vld1.u8 {d6}, [r0], r1 - vst1.u8 {d5}, [r2], r3 - vld1.u8 {d7}, [r0], r1 - vst1.u8 {d6}, [r2], r3 - vst1.u8 {d7}, [r2], r3 - - mov pc, lr - - ENDP ; |vp8_copy_mem8x8_neon| - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem_neon.c new file mode 100644 index 00000000000..deced115c14 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/copymem_neon.c @@ -0,0 +1,59 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +void vp8_copy_mem8x4_neon( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) { + uint8x8_t vtmp; + int r; + + for (r = 0; r < 4; r++) { + vtmp = vld1_u8(src); + vst1_u8(dst, vtmp); + src += src_stride; + dst += dst_stride; + } +} + +void vp8_copy_mem8x8_neon( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) { + uint8x8_t vtmp; + int r; + + for (r = 0; r < 8; r++) { + vtmp = vld1_u8(src); + vst1_u8(dst, vtmp); + src += src_stride; + dst += dst_stride; + } +} + +void vp8_copy_mem16x16_neon( + unsigned char *src, + int src_stride, + unsigned char *dst, + int dst_stride) { + int r; + uint8x16_t qtmp; + + for (r = 0; r < 16; r++) { + qtmp = vld1q_u8(src); + vst1q_u8(dst, qtmp); + src += src_stride; + dst += dst_stride; + } +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.asm deleted file mode 100644 index 79ff02c6940..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.asm +++ /dev/null @@ -1,54 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license and patent -; grant that can be found in the LICENSE file in the root of the source -; tree. All contributing project authors may be found in the AUTHORS -; file in the root of the source tree. -; - - - EXPORT |vp8_dc_only_idct_add_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -;void vp8_dc_only_idct_add_c(short input_dc, unsigned char *pred_ptr, -; int pred_stride, unsigned char *dst_ptr, -; int dst_stride) - -; r0 input_dc -; r1 pred_ptr -; r2 pred_stride -; r3 dst_ptr -; sp dst_stride - -|vp8_dc_only_idct_add_neon| PROC - add r0, r0, #4 - asr r0, r0, #3 - ldr r12, [sp] - vdup.16 q0, r0 - - vld1.32 {d2[0]}, [r1], r2 - vld1.32 {d2[1]}, [r1], r2 - vld1.32 {d4[0]}, [r1], r2 - vld1.32 {d4[1]}, [r1] - - vaddw.u8 q1, q0, d2 - vaddw.u8 q2, q0, d4 - - vqmovun.s16 d2, q1 - vqmovun.s16 d4, q2 - - vst1.32 {d2[0]}, [r3], r12 - vst1.32 {d2[1]}, [r3], r12 - vst1.32 {d4[0]}, [r3], r12 - vst1.32 {d4[1]}, [r3] - - bx lr - - ENDP - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.c new file mode 100644 index 00000000000..ad5f41d7dee --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dc_only_idct_add_neon.c @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +void vp8_dc_only_idct_add_neon( + int16_t input_dc, + unsigned char *pred_ptr, + int pred_stride, + unsigned char *dst_ptr, + int dst_stride) { + int i; + uint16_t a1 = ((input_dc + 4) >> 3); + uint32x2_t d2u32 = vdup_n_u32(0); + uint8x8_t d2u8; + uint16x8_t q1u16; + uint16x8_t qAdd; + + qAdd = vdupq_n_u16(a1); + + for (i = 0; i < 2; i++) { + d2u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d2u32, 0); + pred_ptr += pred_stride; + d2u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d2u32, 1); + pred_ptr += pred_stride; + + q1u16 = vaddw_u8(qAdd, vreinterpret_u8_u32(d2u32)); + d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q1u16)); + + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 0); + dst_ptr += dst_stride; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 1); + dst_ptr += dst_stride; + } +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequant_idct_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequant_idct_neon.asm deleted file mode 100644 index 602cce67697..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequant_idct_neon.asm +++ /dev/null @@ -1,131 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_dequant_idct_add_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 -;void vp8_dequant_idct_add_neon(short *input, short *dq, -; unsigned char *dest, int stride) -; r0 short *input, -; r1 short *dq, -; r2 unsigned char *dest -; r3 int stride - -|vp8_dequant_idct_add_neon| PROC - vld1.16 {q3, q4}, [r0] - vld1.16 {q5, q6}, [r1] - - add r1, r2, r3 ; r1 = dest + stride - lsl r3, #1 ; 2x stride - - vld1.32 {d14[0]}, [r2], r3 - vld1.32 {d14[1]}, [r1], r3 - vld1.32 {d15[0]}, [r2] - vld1.32 {d15[1]}, [r1] - - adr r12, cospi8sqrt2minus1 ; pointer to the first constant - - vmul.i16 q1, q3, q5 ;input for short_idct4x4llm_neon - vmul.i16 q2, q4, q6 - -;|short_idct4x4llm_neon| PROC - vld1.16 {d0}, [r12] - vswp d3, d4 ;q2(vp[4] vp[12]) - - vqdmulh.s16 q3, q2, d0[2] - vqdmulh.s16 q4, q2, d0[0] - - vqadd.s16 d12, d2, d3 ;a1 - vqsub.s16 d13, d2, d3 ;b1 - - vshr.s16 q3, q3, #1 - vshr.s16 q4, q4, #1 - - vqadd.s16 q3, q3, q2 - vqadd.s16 q4, q4, q2 - - vqsub.s16 d10, d6, d9 ;c1 - vqadd.s16 d11, d7, d8 ;d1 - - vqadd.s16 d2, d12, d11 - vqadd.s16 d3, d13, d10 - vqsub.s16 d4, d13, d10 - vqsub.s16 d5, d12, d11 - - vtrn.32 d2, d4 - vtrn.32 d3, d5 - vtrn.16 d2, d3 - vtrn.16 d4, d5 - -; memset(input, 0, 32) -- 32bytes - vmov.i16 q14, #0 - - vswp d3, d4 - vqdmulh.s16 q3, q2, d0[2] - vqdmulh.s16 q4, q2, d0[0] - - vqadd.s16 d12, d2, d3 ;a1 - vqsub.s16 d13, d2, d3 ;b1 - - vmov q15, q14 - - vshr.s16 q3, q3, #1 - vshr.s16 q4, q4, #1 - - vqadd.s16 q3, q3, q2 - vqadd.s16 q4, q4, q2 - - vqsub.s16 d10, d6, d9 ;c1 - vqadd.s16 d11, d7, d8 ;d1 - - vqadd.s16 d2, d12, d11 - vqadd.s16 d3, d13, d10 - vqsub.s16 d4, d13, d10 - vqsub.s16 d5, d12, d11 - - vst1.16 {q14, q15}, [r0] - - vrshr.s16 d2, d2, #3 - vrshr.s16 d3, d3, #3 - vrshr.s16 d4, d4, #3 - vrshr.s16 d5, d5, #3 - - vtrn.32 d2, d4 - vtrn.32 d3, d5 - vtrn.16 d2, d3 - vtrn.16 d4, d5 - - vaddw.u8 q1, q1, d14 - vaddw.u8 q2, q2, d15 - - sub r2, r2, r3 - sub r1, r1, r3 - - vqmovun.s16 d0, q1 - vqmovun.s16 d1, q2 - - vst1.32 {d0[0]}, [r2], r3 - vst1.32 {d0[1]}, [r1], r3 - vst1.32 {d1[0]}, [r2] - vst1.32 {d1[1]}, [r1] - - bx lr - - ENDP ; |vp8_dequant_idct_add_neon| - -; Constant Pool -cospi8sqrt2minus1 DCD 0x4e7b4e7b -sinpi8sqrt2 DCD 0x8a8c8a8c - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequant_idct_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequant_idct_neon.c new file mode 100644 index 00000000000..58e11922c76 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequant_idct_neon.c @@ -0,0 +1,142 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +static const int16_t cospi8sqrt2minus1 = 20091; +static const int16_t sinpi8sqrt2 = 35468; + +void vp8_dequant_idct_add_neon( + int16_t *input, + int16_t *dq, + unsigned char *dst, + int stride) { + unsigned char *dst0; + int32x2_t d14, d15; + int16x4_t d2, d3, d4, d5, d10, d11, d12, d13; + int16x8_t q1, q2, q3, q4, q5, q6; + int16x8_t qEmpty = vdupq_n_s16(0); + int32x2x2_t d2tmp0, d2tmp1; + int16x4x2_t d2tmp2, d2tmp3; + + d14 = d15 = vdup_n_s32(0); + + // load input + q3 = vld1q_s16(input); + vst1q_s16(input, qEmpty); + input += 8; + q4 = vld1q_s16(input); + vst1q_s16(input, qEmpty); + + // load dq + q5 = vld1q_s16(dq); + dq += 8; + q6 = vld1q_s16(dq); + + // load src from dst + dst0 = dst; + d14 = vld1_lane_s32((const int32_t *)dst0, d14, 0); + dst0 += stride; + d14 = vld1_lane_s32((const int32_t *)dst0, d14, 1); + dst0 += stride; + d15 = vld1_lane_s32((const int32_t *)dst0, d15, 0); + dst0 += stride; + d15 = vld1_lane_s32((const int32_t *)dst0, d15, 1); + + q1 = vreinterpretq_s16_u16(vmulq_u16(vreinterpretq_u16_s16(q3), + vreinterpretq_u16_s16(q5))); + q2 = vreinterpretq_s16_u16(vmulq_u16(vreinterpretq_u16_s16(q4), + vreinterpretq_u16_s16(q6))); + + d12 = vqadd_s16(vget_low_s16(q1), vget_low_s16(q2)); + d13 = vqsub_s16(vget_low_s16(q1), vget_low_s16(q2)); + + q2 = vcombine_s16(vget_high_s16(q1), vget_high_s16(q2)); + + q3 = vqdmulhq_n_s16(q2, sinpi8sqrt2); + q4 = vqdmulhq_n_s16(q2, cospi8sqrt2minus1); + + q3 = vshrq_n_s16(q3, 1); + q4 = vshrq_n_s16(q4, 1); + + q3 = vqaddq_s16(q3, q2); + q4 = vqaddq_s16(q4, q2); + + d10 = vqsub_s16(vget_low_s16(q3), vget_high_s16(q4)); + d11 = vqadd_s16(vget_high_s16(q3), vget_low_s16(q4)); + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + d2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + d2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + d2tmp2 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[0]), + vreinterpret_s16_s32(d2tmp1.val[0])); + d2tmp3 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[1]), + vreinterpret_s16_s32(d2tmp1.val[1])); + + // loop 2 + q2 = vcombine_s16(d2tmp2.val[1], d2tmp3.val[1]); + + q3 = vqdmulhq_n_s16(q2, sinpi8sqrt2); + q4 = vqdmulhq_n_s16(q2, cospi8sqrt2minus1); + + d12 = vqadd_s16(d2tmp2.val[0], d2tmp3.val[0]); + d13 = vqsub_s16(d2tmp2.val[0], d2tmp3.val[0]); + + q3 = vshrq_n_s16(q3, 1); + q4 = vshrq_n_s16(q4, 1); + + q3 = vqaddq_s16(q3, q2); + q4 = vqaddq_s16(q4, q2); + + d10 = vqsub_s16(vget_low_s16(q3), vget_high_s16(q4)); + d11 = vqadd_s16(vget_high_s16(q3), vget_low_s16(q4)); + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + d2 = vrshr_n_s16(d2, 3); + d3 = vrshr_n_s16(d3, 3); + d4 = vrshr_n_s16(d4, 3); + d5 = vrshr_n_s16(d5, 3); + + d2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + d2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + d2tmp2 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[0]), + vreinterpret_s16_s32(d2tmp1.val[0])); + d2tmp3 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[1]), + vreinterpret_s16_s32(d2tmp1.val[1])); + + q1 = vcombine_s16(d2tmp2.val[0], d2tmp2.val[1]); + q2 = vcombine_s16(d2tmp3.val[0], d2tmp3.val[1]); + + q1 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q1), + vreinterpret_u8_s32(d14))); + q2 = vreinterpretq_s16_u16(vaddw_u8(vreinterpretq_u16_s16(q2), + vreinterpret_u8_s32(d15))); + + d14 = vreinterpret_s32_u8(vqmovun_s16(q1)); + d15 = vreinterpret_s32_u8(vqmovun_s16(q2)); + + dst0 = dst; + vst1_lane_s32((int32_t *)dst0, d14, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d14, 1); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d15, 0); + dst0 += stride; + vst1_lane_s32((int32_t *)dst0, d15, 1); + return; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequantizeb_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequantizeb_neon.asm deleted file mode 100644 index c8e0c31f29c..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequantizeb_neon.asm +++ /dev/null @@ -1,34 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_dequantize_b_loop_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 -; r0 short *Q, -; r1 short *DQC -; r2 short *DQ -|vp8_dequantize_b_loop_neon| PROC - vld1.16 {q0, q1}, [r0] - vld1.16 {q2, q3}, [r1] - - vmul.i16 q4, q0, q2 - vmul.i16 q5, q1, q3 - - vst1.16 {q4, q5}, [r2] - - bx lr - - ENDP - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequantizeb_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequantizeb_neon.c new file mode 100644 index 00000000000..54e709dd3c3 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/dequantizeb_neon.c @@ -0,0 +1,25 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +#include "vp8/common/blockd.h" + +void vp8_dequantize_b_neon(BLOCKD *d, short *DQC) { + int16x8x2_t qQ, qDQC, qDQ; + + qQ = vld2q_s16(d->qcoeff); + qDQC = vld2q_s16(DQC); + + qDQ.val[0] = vmulq_s16(qQ.val[0], qDQC.val[0]); + qDQ.val[1] = vmulq_s16(qQ.val[1], qDQC.val[1]); + + vst2q_s16(d->dqcoeff, qDQ); +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_0_2x_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_0_2x_neon.asm index 6c29c55860d..3a3921081c4 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_0_2x_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_0_2x_neon.asm @@ -22,6 +22,7 @@ ; r3 stride |idct_dequant_0_2x_neon| PROC push {r4, r5} + vpush {d8-d15} add r12, r2, #4 vld1.32 {d2[0]}, [r2], r3 @@ -72,6 +73,7 @@ vst1.32 {d4[1]}, [r2] vst1.32 {d10[1]}, [r0] + vpop {d8-d15} pop {r4, r5} bx lr diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_full_2x_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_full_2x_neon.asm index d5dce63f6bd..8da0fa0b7ea 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_full_2x_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/idct_dequant_full_2x_neon.asm @@ -22,6 +22,8 @@ ; r2 *dst ; r3 stride |idct_dequant_full_2x_neon| PROC + vpush {d8-d15} + vld1.16 {q0, q1}, [r1] ; dq (same l/r) vld1.16 {q2, q3}, [r0] ; l q add r0, r0, #32 @@ -184,6 +186,7 @@ vst1.32 {d3[0]}, [r2] vst1.32 {d3[1]}, [r1] + vpop {d8-d15} bx lr ENDP ; |idct_dequant_full_2x_neon| diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/iwalsh_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/iwalsh_neon.asm deleted file mode 100644 index e8ea2a61976..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/iwalsh_neon.asm +++ /dev/null @@ -1,87 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - EXPORT |vp8_short_inv_walsh4x4_neon| - - ARM - REQUIRE8 - PRESERVE8 - - AREA |.text|, CODE, READONLY ; name this block of code - -;short vp8_short_inv_walsh4x4_neon(short *input, short *mb_dqcoeff) -|vp8_short_inv_walsh4x4_neon| PROC - - ; read in all four lines of values: d0->d3 - vld1.i16 {q0-q1}, [r0@128] - - ; first for loop - vadd.s16 d4, d0, d3 ;a = [0] + [12] - vadd.s16 d6, d1, d2 ;b = [4] + [8] - vsub.s16 d5, d0, d3 ;d = [0] - [12] - vsub.s16 d7, d1, d2 ;c = [4] - [8] - - vadd.s16 q0, q2, q3 ; a+b d+c - vsub.s16 q1, q2, q3 ; a-b d-c - - vtrn.32 d0, d2 ;d0: 0 1 8 9 - ;d2: 2 3 10 11 - vtrn.32 d1, d3 ;d1: 4 5 12 13 - ;d3: 6 7 14 15 - - vtrn.16 d0, d1 ;d0: 0 4 8 12 - ;d1: 1 5 9 13 - vtrn.16 d2, d3 ;d2: 2 6 10 14 - ;d3: 3 7 11 15 - - ; second for loop - - vadd.s16 d4, d0, d3 ;a = [0] + [3] - vadd.s16 d6, d1, d2 ;b = [1] + [2] - vsub.s16 d5, d0, d3 ;d = [0] - [3] - vsub.s16 d7, d1, d2 ;c = [1] - [2] - - vmov.i16 q8, #3 - - vadd.s16 q0, q2, q3 ; a+b d+c - vsub.s16 q1, q2, q3 ; a-b d-c - - vadd.i16 q0, q0, q8 ;e/f += 3 - vadd.i16 q1, q1, q8 ;g/h += 3 - - vshr.s16 q0, q0, #3 ;e/f >> 3 - vshr.s16 q1, q1, #3 ;g/h >> 3 - - mov r2, #64 - add r3, r1, #32 - - vst1.i16 d0[0], [r1],r2 - vst1.i16 d1[0], [r3],r2 - vst1.i16 d2[0], [r1],r2 - vst1.i16 d3[0], [r3],r2 - - vst1.i16 d0[1], [r1],r2 - vst1.i16 d1[1], [r3],r2 - vst1.i16 d2[1], [r1],r2 - vst1.i16 d3[1], [r3],r2 - - vst1.i16 d0[2], [r1],r2 - vst1.i16 d1[2], [r3],r2 - vst1.i16 d2[2], [r1],r2 - vst1.i16 d3[2], [r3],r2 - - vst1.i16 d0[3], [r1],r2 - vst1.i16 d1[3], [r3],r2 - vst1.i16 d2[3], [r1] - vst1.i16 d3[3], [r3] - - bx lr - ENDP ; |vp8_short_inv_walsh4x4_neon| - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/iwalsh_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/iwalsh_neon.c new file mode 100644 index 00000000000..6ea9dd712aa --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/iwalsh_neon.c @@ -0,0 +1,102 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +void vp8_short_inv_walsh4x4_neon( + int16_t *input, + int16_t *mb_dqcoeff) { + int16x8_t q0s16, q1s16, q2s16, q3s16; + int16x4_t d4s16, d5s16, d6s16, d7s16; + int16x4x2_t v2tmp0, v2tmp1; + int32x2x2_t v2tmp2, v2tmp3; + int16x8_t qAdd3; + + q0s16 = vld1q_s16(input); + q1s16 = vld1q_s16(input + 8); + + // 1st for loop + d4s16 = vadd_s16(vget_low_s16(q0s16), vget_high_s16(q1s16)); + d6s16 = vadd_s16(vget_high_s16(q0s16), vget_low_s16(q1s16)); + d5s16 = vsub_s16(vget_low_s16(q0s16), vget_high_s16(q1s16)); + d7s16 = vsub_s16(vget_high_s16(q0s16), vget_low_s16(q1s16)); + + q2s16 = vcombine_s16(d4s16, d5s16); + q3s16 = vcombine_s16(d6s16, d7s16); + + q0s16 = vaddq_s16(q2s16, q3s16); + q1s16 = vsubq_s16(q2s16, q3s16); + + v2tmp2 = vtrn_s32(vreinterpret_s32_s16(vget_low_s16(q0s16)), + vreinterpret_s32_s16(vget_low_s16(q1s16))); + v2tmp3 = vtrn_s32(vreinterpret_s32_s16(vget_high_s16(q0s16)), + vreinterpret_s32_s16(vget_high_s16(q1s16))); + v2tmp0 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[0]), + vreinterpret_s16_s32(v2tmp3.val[0])); + v2tmp1 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[1]), + vreinterpret_s16_s32(v2tmp3.val[1])); + + // 2nd for loop + d4s16 = vadd_s16(v2tmp0.val[0], v2tmp1.val[1]); + d6s16 = vadd_s16(v2tmp0.val[1], v2tmp1.val[0]); + d5s16 = vsub_s16(v2tmp0.val[0], v2tmp1.val[1]); + d7s16 = vsub_s16(v2tmp0.val[1], v2tmp1.val[0]); + q2s16 = vcombine_s16(d4s16, d5s16); + q3s16 = vcombine_s16(d6s16, d7s16); + + qAdd3 = vdupq_n_s16(3); + + q0s16 = vaddq_s16(q2s16, q3s16); + q1s16 = vsubq_s16(q2s16, q3s16); + + q0s16 = vaddq_s16(q0s16, qAdd3); + q1s16 = vaddq_s16(q1s16, qAdd3); + + q0s16 = vshrq_n_s16(q0s16, 3); + q1s16 = vshrq_n_s16(q1s16, 3); + + // store + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 0); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 0); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 0); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 0); + mb_dqcoeff += 16; + + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 1); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 1); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 1); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 1); + mb_dqcoeff += 16; + + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 2); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 2); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 2); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 2); + mb_dqcoeff += 16; + + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 3); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 3); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 3); + mb_dqcoeff += 16; + vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 3); + mb_dqcoeff += 16; + return; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfilter_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfilter_neon.asm index e44be0a1e34..c4f09c7753b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfilter_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfilter_neon.asm @@ -24,10 +24,12 @@ ; sp unsigned char thresh, |vp8_loop_filter_horizontal_edge_y_neon| PROC push {lr} + vpush {d8-d15} + vdup.u8 q0, r2 ; duplicate blimit vdup.u8 q1, r3 ; duplicate limit sub r2, r0, r1, lsl #2 ; move src pointer down by 4 lines - ldr r3, [sp, #4] ; load thresh + ldr r3, [sp, #68] ; load thresh add r12, r2, r1 add r1, r1, r1 @@ -52,6 +54,7 @@ vst1.u8 {q7}, [r2@128], r1 ; store oq0 vst1.u8 {q8}, [r12@128], r1 ; store oq1 + vpop {d8-d15} pop {pc} ENDP ; |vp8_loop_filter_horizontal_edge_y_neon| @@ -64,10 +67,12 @@ ; sp+4 unsigned char *v |vp8_loop_filter_horizontal_edge_uv_neon| PROC push {lr} + vpush {d8-d15} + vdup.u8 q0, r2 ; duplicate blimit vdup.u8 q1, r3 ; duplicate limit - ldr r12, [sp, #4] ; load thresh - ldr r2, [sp, #8] ; load v ptr + ldr r12, [sp, #68] ; load thresh + ldr r2, [sp, #72] ; load v ptr vdup.u8 q2, r12 ; duplicate thresh sub r3, r0, r1, lsl #2 ; move u pointer down by 4 lines @@ -104,6 +109,7 @@ vst1.u8 {d16}, [r0@64] ; store u oq1 vst1.u8 {d17}, [r2@64] ; store v oq1 + vpop {d8-d15} pop {pc} ENDP ; |vp8_loop_filter_horizontal_edge_uv_neon| @@ -120,11 +126,13 @@ |vp8_loop_filter_vertical_edge_y_neon| PROC push {lr} + vpush {d8-d15} + vdup.u8 q0, r2 ; duplicate blimit vdup.u8 q1, r3 ; duplicate limit sub r2, r0, #4 ; src ptr down by 4 columns add r1, r1, r1 - ldr r3, [sp, #4] ; load thresh + ldr r3, [sp, #68] ; load thresh add r12, r2, r1, asr #1 vld1.u8 {d6}, [r2], r1 @@ -194,6 +202,7 @@ vst4.8 {d14[6], d15[6], d16[6], d17[6]}, [r0] vst4.8 {d14[7], d15[7], d16[7], d17[7]}, [r12] + vpop {d8-d15} pop {pc} ENDP ; |vp8_loop_filter_vertical_edge_y_neon| @@ -210,9 +219,11 @@ ; sp+4 unsigned char *v |vp8_loop_filter_vertical_edge_uv_neon| PROC push {lr} + vpush {d8-d15} + vdup.u8 q0, r2 ; duplicate blimit sub r12, r0, #4 ; move u pointer down by 4 columns - ldr r2, [sp, #8] ; load v ptr + ldr r2, [sp, #72] ; load v ptr vdup.u8 q1, r3 ; duplicate limit sub r3, r2, #4 ; move v pointer down by 4 columns @@ -233,7 +244,7 @@ vld1.u8 {d20}, [r12] vld1.u8 {d21}, [r3] - ldr r12, [sp, #4] ; load thresh + ldr r12, [sp, #68] ; load thresh ;transpose to 8x16 matrix vtrn.32 q3, q7 @@ -281,6 +292,7 @@ vst4.8 {d10[7], d11[7], d12[7], d13[7]}, [r0] vst4.8 {d14[7], d15[7], d16[7], d17[7]}, [r2] + vpop {d8-d15} pop {pc} ENDP ; |vp8_loop_filter_vertical_edge_uv_neon| diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm deleted file mode 100644 index adf848b9c34..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm +++ /dev/null @@ -1,117 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - ;EXPORT |vp8_loop_filter_simple_horizontal_edge_neon| - EXPORT |vp8_loop_filter_bhs_neon| - EXPORT |vp8_loop_filter_mbhs_neon| - ARM - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -; r0 unsigned char *s, PRESERVE -; r1 int p, PRESERVE -; q1 limit, PRESERVE - -|vp8_loop_filter_simple_horizontal_edge_neon| PROC - - sub r3, r0, r1, lsl #1 ; move src pointer down by 2 lines - - vld1.u8 {q7}, [r0@128], r1 ; q0 - vld1.u8 {q5}, [r3@128], r1 ; p0 - vld1.u8 {q8}, [r0@128] ; q1 - vld1.u8 {q6}, [r3@128] ; p1 - - vabd.u8 q15, q6, q7 ; abs(p0 - q0) - vabd.u8 q14, q5, q8 ; abs(p1 - q1) - - vqadd.u8 q15, q15, q15 ; abs(p0 - q0) * 2 - vshr.u8 q14, q14, #1 ; abs(p1 - q1) / 2 - vmov.u8 q0, #0x80 ; 0x80 - vmov.s16 q13, #3 - vqadd.u8 q15, q15, q14 ; abs(p0 - q0) * 2 + abs(p1 - q1) / 2 - - veor q7, q7, q0 ; qs0: q0 offset to convert to a signed value - veor q6, q6, q0 ; ps0: p0 offset to convert to a signed value - veor q5, q5, q0 ; ps1: p1 offset to convert to a signed value - veor q8, q8, q0 ; qs1: q1 offset to convert to a signed value - - vcge.u8 q15, q1, q15 ; (abs(p0 - q0)*2 + abs(p1-q1)/2 > limit)*-1 - - vsubl.s8 q2, d14, d12 ; ( qs0 - ps0) - vsubl.s8 q3, d15, d13 - - vqsub.s8 q4, q5, q8 ; q4: vp8_filter = vp8_signed_char_clamp(ps1-qs1) - - vmul.s16 q2, q2, q13 ; 3 * ( qs0 - ps0) - vmul.s16 q3, q3, q13 - - vmov.u8 q10, #0x03 ; 0x03 - vmov.u8 q9, #0x04 ; 0x04 - - vaddw.s8 q2, q2, d8 ; vp8_filter + 3 * ( qs0 - ps0) - vaddw.s8 q3, q3, d9 - - vqmovn.s16 d8, q2 ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0)) - vqmovn.s16 d9, q3 - - vand q14, q4, q15 ; vp8_filter &= mask - - vqadd.s8 q2, q14, q10 ; Filter2 = vp8_signed_char_clamp(vp8_filter+3) - vqadd.s8 q3, q14, q9 ; Filter1 = vp8_signed_char_clamp(vp8_filter+4) - vshr.s8 q2, q2, #3 ; Filter2 >>= 3 - vshr.s8 q4, q3, #3 ; Filter1 >>= 3 - - sub r0, r0, r1 - - ;calculate output - vqadd.s8 q11, q6, q2 ; u = vp8_signed_char_clamp(ps0 + Filter2) - vqsub.s8 q10, q7, q4 ; u = vp8_signed_char_clamp(qs0 - Filter1) - - veor q6, q11, q0 ; *op0 = u^0x80 - veor q7, q10, q0 ; *oq0 = u^0x80 - - vst1.u8 {q6}, [r3@128] ; store op0 - vst1.u8 {q7}, [r0@128] ; store oq0 - - bx lr - ENDP ; |vp8_loop_filter_simple_horizontal_edge_neon| - -; r0 unsigned char *y -; r1 int ystride -; r2 const unsigned char *blimit - -|vp8_loop_filter_bhs_neon| PROC - push {r4, lr} - ldrb r3, [r2] ; load blim from mem - vdup.s8 q1, r3 ; duplicate blim - - add r0, r0, r1, lsl #2 ; src = y_ptr + 4 * y_stride - bl vp8_loop_filter_simple_horizontal_edge_neon - ; vp8_loop_filter_simple_horizontal_edge_neon preserves r0, r1 and q1 - add r0, r0, r1, lsl #2 ; src = y_ptr + 8* y_stride - bl vp8_loop_filter_simple_horizontal_edge_neon - add r0, r0, r1, lsl #2 ; src = y_ptr + 12 * y_stride - pop {r4, lr} - b vp8_loop_filter_simple_horizontal_edge_neon - ENDP ;|vp8_loop_filter_bhs_neon| - -; r0 unsigned char *y -; r1 int ystride -; r2 const unsigned char *blimit - -|vp8_loop_filter_mbhs_neon| PROC - ldrb r3, [r2] ; load blim from mem - vdup.s8 q1, r3 ; duplicate mblim - b vp8_loop_filter_simple_horizontal_edge_neon - ENDP ;|vp8_loop_filter_bhs_neon| - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c new file mode 100644 index 00000000000..b25686ffb88 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c @@ -0,0 +1,111 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> +#include "./vpx_config.h" + +static INLINE void vp8_loop_filter_simple_horizontal_edge_neon( + unsigned char *s, + int p, + const unsigned char *blimit) { + uint8_t *sp; + uint8x16_t qblimit, q0u8; + uint8x16_t q5u8, q6u8, q7u8, q8u8, q9u8, q10u8, q14u8, q15u8; + int16x8_t q2s16, q3s16, q13s16; + int8x8_t d8s8, d9s8; + int8x16_t q2s8, q3s8, q4s8, q10s8, q11s8, q14s8; + + qblimit = vdupq_n_u8(*blimit); + + sp = s - (p << 1); + q5u8 = vld1q_u8(sp); + sp += p; + q6u8 = vld1q_u8(sp); + sp += p; + q7u8 = vld1q_u8(sp); + sp += p; + q8u8 = vld1q_u8(sp); + + q15u8 = vabdq_u8(q6u8, q7u8); + q14u8 = vabdq_u8(q5u8, q8u8); + + q15u8 = vqaddq_u8(q15u8, q15u8); + q14u8 = vshrq_n_u8(q14u8, 1); + q0u8 = vdupq_n_u8(0x80); + q13s16 = vdupq_n_s16(3); + q15u8 = vqaddq_u8(q15u8, q14u8); + + q5u8 = veorq_u8(q5u8, q0u8); + q6u8 = veorq_u8(q6u8, q0u8); + q7u8 = veorq_u8(q7u8, q0u8); + q8u8 = veorq_u8(q8u8, q0u8); + + q15u8 = vcgeq_u8(qblimit, q15u8); + + q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7u8)), + vget_low_s8(vreinterpretq_s8_u8(q6u8))); + q3s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7u8)), + vget_high_s8(vreinterpretq_s8_u8(q6u8))); + + q4s8 = vqsubq_s8(vreinterpretq_s8_u8(q5u8), + vreinterpretq_s8_u8(q8u8)); + + q2s16 = vmulq_s16(q2s16, q13s16); + q3s16 = vmulq_s16(q3s16, q13s16); + + q10u8 = vdupq_n_u8(3); + q9u8 = vdupq_n_u8(4); + + q2s16 = vaddw_s8(q2s16, vget_low_s8(q4s8)); + q3s16 = vaddw_s8(q3s16, vget_high_s8(q4s8)); + + d8s8 = vqmovn_s16(q2s16); + d9s8 = vqmovn_s16(q3s16); + q4s8 = vcombine_s8(d8s8, d9s8); + + q14s8 = vandq_s8(q4s8, vreinterpretq_s8_u8(q15u8)); + + q2s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q10u8)); + q3s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q9u8)); + q2s8 = vshrq_n_s8(q2s8, 3); + q3s8 = vshrq_n_s8(q3s8, 3); + + q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6u8), q2s8); + q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q7u8), q3s8); + + q6u8 = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8); + q7u8 = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8); + + vst1q_u8(s, q7u8); + s -= p; + vst1q_u8(s, q6u8); + return; +} + +void vp8_loop_filter_bhs_neon( + unsigned char *y_ptr, + int y_stride, + const unsigned char *blimit) { + y_ptr += y_stride * 4; + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + y_ptr += y_stride * 4; + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + y_ptr += y_stride * 4; + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + return; +} + +void vp8_loop_filter_mbhs_neon( + unsigned char *y_ptr, + int y_stride, + const unsigned char *blimit) { + vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit); + return; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm index e690df2f7de..78d13c895aa 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm @@ -9,7 +9,6 @@ ; - ;EXPORT |vp8_loop_filter_simple_vertical_edge_neon| EXPORT |vp8_loop_filter_bvs_neon| EXPORT |vp8_loop_filter_mbvs_neon| ARM @@ -22,6 +21,8 @@ ; q1 limit, PRESERVE |vp8_loop_filter_simple_vertical_edge_neon| PROC + vpush {d8-d15} + sub r0, r0, #2 ; move src pointer down by 2 columns add r12, r1, r1 add r3, r0, r1 @@ -120,6 +121,7 @@ vst2.8 {d14[6], d15[6]}, [r0], r12 vst2.8 {d14[7], d15[7]}, [r3] + vpop {d8-d15} bx lr ENDP ; |vp8_loop_filter_simple_vertical_edge_neon| diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/mbloopfilter_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/mbloopfilter_neon.asm deleted file mode 100644 index f41c156df8b..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/mbloopfilter_neon.asm +++ /dev/null @@ -1,469 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_mbloop_filter_horizontal_edge_y_neon| - EXPORT |vp8_mbloop_filter_horizontal_edge_uv_neon| - EXPORT |vp8_mbloop_filter_vertical_edge_y_neon| - EXPORT |vp8_mbloop_filter_vertical_edge_uv_neon| - ARM - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -; void vp8_mbloop_filter_horizontal_edge_y_neon(unsigned char *src, int pitch, -; const unsigned char *blimit, -; const unsigned char *limit, -; const unsigned char *thresh) -; r0 unsigned char *src, -; r1 int pitch, -; r2 unsigned char blimit -; r3 unsigned char limit -; sp unsigned char thresh, -|vp8_mbloop_filter_horizontal_edge_y_neon| PROC - push {lr} - add r1, r1, r1 ; double stride - ldr r12, [sp, #4] ; load thresh - sub r0, r0, r1, lsl #1 ; move src pointer down by 4 lines - vdup.u8 q2, r12 ; thresh - add r12, r0, r1, lsr #1 ; move src pointer up by 1 line - - vld1.u8 {q3}, [r0@128], r1 ; p3 - vld1.u8 {q4}, [r12@128], r1 ; p2 - vld1.u8 {q5}, [r0@128], r1 ; p1 - vld1.u8 {q6}, [r12@128], r1 ; p0 - vld1.u8 {q7}, [r0@128], r1 ; q0 - vld1.u8 {q8}, [r12@128], r1 ; q1 - vld1.u8 {q9}, [r0@128], r1 ; q2 - vld1.u8 {q10}, [r12@128], r1 ; q3 - - bl vp8_mbloop_filter_neon - - sub r12, r12, r1, lsl #2 - add r0, r12, r1, lsr #1 - - vst1.u8 {q4}, [r12@128],r1 ; store op2 - vst1.u8 {q5}, [r0@128],r1 ; store op1 - vst1.u8 {q6}, [r12@128], r1 ; store op0 - vst1.u8 {q7}, [r0@128],r1 ; store oq0 - vst1.u8 {q8}, [r12@128] ; store oq1 - vst1.u8 {q9}, [r0@128] ; store oq2 - - pop {pc} - ENDP ; |vp8_mbloop_filter_horizontal_edge_y_neon| - -; void vp8_mbloop_filter_horizontal_edge_uv_neon(unsigned char *u, int pitch, -; const unsigned char *blimit, -; const unsigned char *limit, -; const unsigned char *thresh, -; unsigned char *v) -; r0 unsigned char *u, -; r1 int pitch, -; r2 unsigned char blimit -; r3 unsigned char limit -; sp unsigned char thresh, -; sp+4 unsigned char *v - -|vp8_mbloop_filter_horizontal_edge_uv_neon| PROC - push {lr} - ldr r12, [sp, #4] ; load thresh - sub r0, r0, r1, lsl #2 ; move u pointer down by 4 lines - vdup.u8 q2, r12 ; thresh - ldr r12, [sp, #8] ; load v ptr - sub r12, r12, r1, lsl #2 ; move v pointer down by 4 lines - - vld1.u8 {d6}, [r0@64], r1 ; p3 - vld1.u8 {d7}, [r12@64], r1 ; p3 - vld1.u8 {d8}, [r0@64], r1 ; p2 - vld1.u8 {d9}, [r12@64], r1 ; p2 - vld1.u8 {d10}, [r0@64], r1 ; p1 - vld1.u8 {d11}, [r12@64], r1 ; p1 - vld1.u8 {d12}, [r0@64], r1 ; p0 - vld1.u8 {d13}, [r12@64], r1 ; p0 - vld1.u8 {d14}, [r0@64], r1 ; q0 - vld1.u8 {d15}, [r12@64], r1 ; q0 - vld1.u8 {d16}, [r0@64], r1 ; q1 - vld1.u8 {d17}, [r12@64], r1 ; q1 - vld1.u8 {d18}, [r0@64], r1 ; q2 - vld1.u8 {d19}, [r12@64], r1 ; q2 - vld1.u8 {d20}, [r0@64], r1 ; q3 - vld1.u8 {d21}, [r12@64], r1 ; q3 - - bl vp8_mbloop_filter_neon - - sub r0, r0, r1, lsl #3 - sub r12, r12, r1, lsl #3 - - add r0, r0, r1 - add r12, r12, r1 - - vst1.u8 {d8}, [r0@64], r1 ; store u op2 - vst1.u8 {d9}, [r12@64], r1 ; store v op2 - vst1.u8 {d10}, [r0@64], r1 ; store u op1 - vst1.u8 {d11}, [r12@64], r1 ; store v op1 - vst1.u8 {d12}, [r0@64], r1 ; store u op0 - vst1.u8 {d13}, [r12@64], r1 ; store v op0 - vst1.u8 {d14}, [r0@64], r1 ; store u oq0 - vst1.u8 {d15}, [r12@64], r1 ; store v oq0 - vst1.u8 {d16}, [r0@64], r1 ; store u oq1 - vst1.u8 {d17}, [r12@64], r1 ; store v oq1 - vst1.u8 {d18}, [r0@64], r1 ; store u oq2 - vst1.u8 {d19}, [r12@64], r1 ; store v oq2 - - pop {pc} - ENDP ; |vp8_mbloop_filter_horizontal_edge_uv_neon| - -; void vp8_mbloop_filter_vertical_edge_y_neon(unsigned char *src, int pitch, -; const unsigned char *blimit, -; const unsigned char *limit, -; const unsigned char *thresh) -; r0 unsigned char *src, -; r1 int pitch, -; r2 unsigned char blimit -; r3 unsigned char limit -; sp unsigned char thresh, -|vp8_mbloop_filter_vertical_edge_y_neon| PROC - push {lr} - ldr r12, [sp, #4] ; load thresh - sub r0, r0, #4 ; move src pointer down by 4 columns - vdup.s8 q2, r12 ; thresh - add r12, r0, r1, lsl #3 ; move src pointer down by 8 lines - - vld1.u8 {d6}, [r0], r1 ; load first 8-line src data - vld1.u8 {d7}, [r12], r1 ; load second 8-line src data - vld1.u8 {d8}, [r0], r1 - vld1.u8 {d9}, [r12], r1 - vld1.u8 {d10}, [r0], r1 - vld1.u8 {d11}, [r12], r1 - vld1.u8 {d12}, [r0], r1 - vld1.u8 {d13}, [r12], r1 - vld1.u8 {d14}, [r0], r1 - vld1.u8 {d15}, [r12], r1 - vld1.u8 {d16}, [r0], r1 - vld1.u8 {d17}, [r12], r1 - vld1.u8 {d18}, [r0], r1 - vld1.u8 {d19}, [r12], r1 - vld1.u8 {d20}, [r0], r1 - vld1.u8 {d21}, [r12], r1 - - ;transpose to 8x16 matrix - vtrn.32 q3, q7 - vtrn.32 q4, q8 - vtrn.32 q5, q9 - vtrn.32 q6, q10 - - vtrn.16 q3, q5 - vtrn.16 q4, q6 - vtrn.16 q7, q9 - vtrn.16 q8, q10 - - vtrn.8 q3, q4 - vtrn.8 q5, q6 - vtrn.8 q7, q8 - vtrn.8 q9, q10 - - sub r0, r0, r1, lsl #3 - - bl vp8_mbloop_filter_neon - - sub r12, r12, r1, lsl #3 - - ;transpose to 16x8 matrix - vtrn.32 q3, q7 - vtrn.32 q4, q8 - vtrn.32 q5, q9 - vtrn.32 q6, q10 - - vtrn.16 q3, q5 - vtrn.16 q4, q6 - vtrn.16 q7, q9 - vtrn.16 q8, q10 - - vtrn.8 q3, q4 - vtrn.8 q5, q6 - vtrn.8 q7, q8 - vtrn.8 q9, q10 - - ;store op2, op1, op0, oq0, oq1, oq2 - vst1.8 {d6}, [r0], r1 - vst1.8 {d7}, [r12], r1 - vst1.8 {d8}, [r0], r1 - vst1.8 {d9}, [r12], r1 - vst1.8 {d10}, [r0], r1 - vst1.8 {d11}, [r12], r1 - vst1.8 {d12}, [r0], r1 - vst1.8 {d13}, [r12], r1 - vst1.8 {d14}, [r0], r1 - vst1.8 {d15}, [r12], r1 - vst1.8 {d16}, [r0], r1 - vst1.8 {d17}, [r12], r1 - vst1.8 {d18}, [r0], r1 - vst1.8 {d19}, [r12], r1 - vst1.8 {d20}, [r0] - vst1.8 {d21}, [r12] - - pop {pc} - ENDP ; |vp8_mbloop_filter_vertical_edge_y_neon| - -; void vp8_mbloop_filter_vertical_edge_uv_neon(unsigned char *u, int pitch, -; const unsigned char *blimit, -; const unsigned char *limit, -; const unsigned char *thresh, -; unsigned char *v) -; r0 unsigned char *u, -; r1 int pitch, -; r2 const signed char *flimit, -; r3 const signed char *limit, -; sp const signed char *thresh, -; sp+4 unsigned char *v -|vp8_mbloop_filter_vertical_edge_uv_neon| PROC - push {lr} - ldr r12, [sp, #4] ; load thresh - sub r0, r0, #4 ; move u pointer down by 4 columns - vdup.u8 q2, r12 ; thresh - ldr r12, [sp, #8] ; load v ptr - sub r12, r12, #4 ; move v pointer down by 4 columns - - vld1.u8 {d6}, [r0], r1 ;load u data - vld1.u8 {d7}, [r12], r1 ;load v data - vld1.u8 {d8}, [r0], r1 - vld1.u8 {d9}, [r12], r1 - vld1.u8 {d10}, [r0], r1 - vld1.u8 {d11}, [r12], r1 - vld1.u8 {d12}, [r0], r1 - vld1.u8 {d13}, [r12], r1 - vld1.u8 {d14}, [r0], r1 - vld1.u8 {d15}, [r12], r1 - vld1.u8 {d16}, [r0], r1 - vld1.u8 {d17}, [r12], r1 - vld1.u8 {d18}, [r0], r1 - vld1.u8 {d19}, [r12], r1 - vld1.u8 {d20}, [r0], r1 - vld1.u8 {d21}, [r12], r1 - - ;transpose to 8x16 matrix - vtrn.32 q3, q7 - vtrn.32 q4, q8 - vtrn.32 q5, q9 - vtrn.32 q6, q10 - - vtrn.16 q3, q5 - vtrn.16 q4, q6 - vtrn.16 q7, q9 - vtrn.16 q8, q10 - - vtrn.8 q3, q4 - vtrn.8 q5, q6 - vtrn.8 q7, q8 - vtrn.8 q9, q10 - - sub r0, r0, r1, lsl #3 - - bl vp8_mbloop_filter_neon - - sub r12, r12, r1, lsl #3 - - ;transpose to 16x8 matrix - vtrn.32 q3, q7 - vtrn.32 q4, q8 - vtrn.32 q5, q9 - vtrn.32 q6, q10 - - vtrn.16 q3, q5 - vtrn.16 q4, q6 - vtrn.16 q7, q9 - vtrn.16 q8, q10 - - vtrn.8 q3, q4 - vtrn.8 q5, q6 - vtrn.8 q7, q8 - vtrn.8 q9, q10 - - ;store op2, op1, op0, oq0, oq1, oq2 - vst1.8 {d6}, [r0], r1 - vst1.8 {d7}, [r12], r1 - vst1.8 {d8}, [r0], r1 - vst1.8 {d9}, [r12], r1 - vst1.8 {d10}, [r0], r1 - vst1.8 {d11}, [r12], r1 - vst1.8 {d12}, [r0], r1 - vst1.8 {d13}, [r12], r1 - vst1.8 {d14}, [r0], r1 - vst1.8 {d15}, [r12], r1 - vst1.8 {d16}, [r0], r1 - vst1.8 {d17}, [r12], r1 - vst1.8 {d18}, [r0], r1 - vst1.8 {d19}, [r12], r1 - vst1.8 {d20}, [r0] - vst1.8 {d21}, [r12] - - pop {pc} - ENDP ; |vp8_mbloop_filter_vertical_edge_uv_neon| - -; void vp8_mbloop_filter_neon() -; This is a helper function for the macroblock loopfilters. The individual -; functions do the necessary load, transpose (if necessary), preserve (if -; necessary) and store. - -; r0,r1 PRESERVE -; r2 mblimit -; r3 limit - -; q2 thresh -; q3 p3 PRESERVE -; q4 p2 -; q5 p1 -; q6 p0 -; q7 q0 -; q8 q1 -; q9 q2 -; q10 q3 PRESERVE - -|vp8_mbloop_filter_neon| PROC - - ; vp8_filter_mask - vabd.u8 q11, q3, q4 ; abs(p3 - p2) - vabd.u8 q12, q4, q5 ; abs(p2 - p1) - vabd.u8 q13, q5, q6 ; abs(p1 - p0) - vabd.u8 q14, q8, q7 ; abs(q1 - q0) - vabd.u8 q1, q9, q8 ; abs(q2 - q1) - vabd.u8 q0, q10, q9 ; abs(q3 - q2) - - vmax.u8 q11, q11, q12 - vmax.u8 q12, q13, q14 - vmax.u8 q1, q1, q0 - vmax.u8 q15, q11, q12 - - vabd.u8 q12, q6, q7 ; abs(p0 - q0) - - ; vp8_hevmask - vcgt.u8 q13, q13, q2 ; (abs(p1 - p0) > thresh) * -1 - vcgt.u8 q14, q14, q2 ; (abs(q1 - q0) > thresh) * -1 - vmax.u8 q15, q15, q1 - - vdup.u8 q1, r3 ; limit - vdup.u8 q2, r2 ; mblimit - - vmov.u8 q0, #0x80 ; 0x80 - - vcge.u8 q15, q1, q15 - - vabd.u8 q1, q5, q8 ; a = abs(p1 - q1) - vqadd.u8 q12, q12, q12 ; b = abs(p0 - q0) * 2 - vmov.u16 q11, #3 ; #3 - - ; vp8_filter - ; convert to signed - veor q7, q7, q0 ; qs0 - vshr.u8 q1, q1, #1 ; a = a / 2 - veor q6, q6, q0 ; ps0 - veor q5, q5, q0 ; ps1 - - vqadd.u8 q12, q12, q1 ; a = b + a - - veor q8, q8, q0 ; qs1 - veor q4, q4, q0 ; ps2 - veor q9, q9, q0 ; qs2 - - vorr q14, q13, q14 ; vp8_hevmask - - vcge.u8 q12, q2, q12 ; (a > flimit * 2 + limit) * -1 - - vsubl.s8 q2, d14, d12 ; qs0 - ps0 - vsubl.s8 q13, d15, d13 - - vqsub.s8 q1, q5, q8 ; vp8_filter = clamp(ps1-qs1) - - vmul.i16 q2, q2, q11 ; 3 * ( qs0 - ps0) - - vand q15, q15, q12 ; vp8_filter_mask - - vmul.i16 q13, q13, q11 - - vmov.u8 q12, #3 ; #3 - - vaddw.s8 q2, q2, d2 ; vp8_filter + 3 * ( qs0 - ps0) - vaddw.s8 q13, q13, d3 - - vmov.u8 q11, #4 ; #4 - - ; vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0)) - vqmovn.s16 d2, q2 - vqmovn.s16 d3, q13 - - vand q1, q1, q15 ; vp8_filter &= mask - - vmov.u16 q15, #63 ; #63 - - vand q13, q1, q14 ; Filter2 &= hev - - vqadd.s8 q2, q13, q11 ; Filter1 = clamp(Filter2+4) - vqadd.s8 q13, q13, q12 ; Filter2 = clamp(Filter2+3) - - vmov q0, q15 - - vshr.s8 q2, q2, #3 ; Filter1 >>= 3 - vshr.s8 q13, q13, #3 ; Filter2 >>= 3 - - vmov q11, q15 - vmov q12, q15 - - vqsub.s8 q7, q7, q2 ; qs0 = clamp(qs0 - Filter1) - - vqadd.s8 q6, q6, q13 ; ps0 = clamp(ps0 + Filter2) - - vbic q1, q1, q14 ; vp8_filter &= ~hev - - ; roughly 1/7th difference across boundary - ; roughly 2/7th difference across boundary - ; roughly 3/7th difference across boundary - - vmov.u8 d5, #9 ; #9 - vmov.u8 d4, #18 ; #18 - - vmov q13, q15 - vmov q14, q15 - - vmlal.s8 q0, d2, d5 ; 63 + Filter2 * 9 - vmlal.s8 q11, d3, d5 - vmov.u8 d5, #27 ; #27 - vmlal.s8 q12, d2, d4 ; 63 + Filter2 * 18 - vmlal.s8 q13, d3, d4 - vmlal.s8 q14, d2, d5 ; 63 + Filter2 * 27 - vmlal.s8 q15, d3, d5 - - vqshrn.s16 d0, q0, #7 ; u = clamp((63 + Filter2 * 9)>>7) - vqshrn.s16 d1, q11, #7 - vqshrn.s16 d24, q12, #7 ; u = clamp((63 + Filter2 * 18)>>7) - vqshrn.s16 d25, q13, #7 - vqshrn.s16 d28, q14, #7 ; u = clamp((63 + Filter2 * 27)>>7) - vqshrn.s16 d29, q15, #7 - - vmov.u8 q1, #0x80 ; 0x80 - - vqsub.s8 q11, q9, q0 ; s = clamp(qs2 - u) - vqadd.s8 q0, q4, q0 ; s = clamp(ps2 + u) - vqsub.s8 q13, q8, q12 ; s = clamp(qs1 - u) - vqadd.s8 q12, q5, q12 ; s = clamp(ps1 + u) - vqsub.s8 q15, q7, q14 ; s = clamp(qs0 - u) - vqadd.s8 q14, q6, q14 ; s = clamp(ps0 + u) - - veor q9, q11, q1 ; *oq2 = s^0x80 - veor q4, q0, q1 ; *op2 = s^0x80 - veor q8, q13, q1 ; *oq1 = s^0x80 - veor q5, q12, q1 ; *op2 = s^0x80 - veor q7, q15, q1 ; *oq0 = s^0x80 - veor q6, q14, q1 ; *op0 = s^0x80 - - bx lr - ENDP ; |vp8_mbloop_filter_neon| - -;----------------- - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/mbloopfilter_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/mbloopfilter_neon.c new file mode 100644 index 00000000000..5351f4be665 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/mbloopfilter_neon.c @@ -0,0 +1,625 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> +#include "./vpx_config.h" + +static INLINE void vp8_mbloop_filter_neon( + uint8x16_t qblimit, // mblimit + uint8x16_t qlimit, // limit + uint8x16_t qthresh, // thresh + uint8x16_t q3, // p2 + uint8x16_t q4, // p2 + uint8x16_t q5, // p1 + uint8x16_t q6, // p0 + uint8x16_t q7, // q0 + uint8x16_t q8, // q1 + uint8x16_t q9, // q2 + uint8x16_t q10, // q3 + uint8x16_t *q4r, // p1 + uint8x16_t *q5r, // p1 + uint8x16_t *q6r, // p0 + uint8x16_t *q7r, // q0 + uint8x16_t *q8r, // q1 + uint8x16_t *q9r) { // q1 + uint8x16_t q0u8, q1u8, q11u8, q12u8, q13u8, q14u8, q15u8; + int16x8_t q0s16, q2s16, q11s16, q12s16, q13s16, q14s16, q15s16; + int8x16_t q1s8, q6s8, q7s8, q2s8, q11s8, q13s8; + uint16x8_t q0u16, q11u16, q12u16, q13u16, q14u16, q15u16; + int8x16_t q0s8, q12s8, q14s8, q15s8; + int8x8_t d0, d1, d2, d3, d4, d5, d24, d25, d28, d29; + + q11u8 = vabdq_u8(q3, q4); + q12u8 = vabdq_u8(q4, q5); + q13u8 = vabdq_u8(q5, q6); + q14u8 = vabdq_u8(q8, q7); + q1u8 = vabdq_u8(q9, q8); + q0u8 = vabdq_u8(q10, q9); + + q11u8 = vmaxq_u8(q11u8, q12u8); + q12u8 = vmaxq_u8(q13u8, q14u8); + q1u8 = vmaxq_u8(q1u8, q0u8); + q15u8 = vmaxq_u8(q11u8, q12u8); + + q12u8 = vabdq_u8(q6, q7); + + // vp8_hevmask + q13u8 = vcgtq_u8(q13u8, qthresh); + q14u8 = vcgtq_u8(q14u8, qthresh); + q15u8 = vmaxq_u8(q15u8, q1u8); + + q15u8 = vcgeq_u8(qlimit, q15u8); + + q1u8 = vabdq_u8(q5, q8); + q12u8 = vqaddq_u8(q12u8, q12u8); + + // vp8_filter() function + // convert to signed + q0u8 = vdupq_n_u8(0x80); + q9 = veorq_u8(q9, q0u8); + q8 = veorq_u8(q8, q0u8); + q7 = veorq_u8(q7, q0u8); + q6 = veorq_u8(q6, q0u8); + q5 = veorq_u8(q5, q0u8); + q4 = veorq_u8(q4, q0u8); + + q1u8 = vshrq_n_u8(q1u8, 1); + q12u8 = vqaddq_u8(q12u8, q1u8); + + q14u8 = vorrq_u8(q13u8, q14u8); + q12u8 = vcgeq_u8(qblimit, q12u8); + + q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)), + vget_low_s8(vreinterpretq_s8_u8(q6))); + q13s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)), + vget_high_s8(vreinterpretq_s8_u8(q6))); + + q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5), + vreinterpretq_s8_u8(q8)); + + q11s16 = vdupq_n_s16(3); + q2s16 = vmulq_s16(q2s16, q11s16); + q13s16 = vmulq_s16(q13s16, q11s16); + + q15u8 = vandq_u8(q15u8, q12u8); + + q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8)); + q13s16 = vaddw_s8(q13s16, vget_high_s8(q1s8)); + + q12u8 = vdupq_n_u8(3); + q11u8 = vdupq_n_u8(4); + // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0)) + d2 = vqmovn_s16(q2s16); + d3 = vqmovn_s16(q13s16); + q1s8 = vcombine_s8(d2, d3); + q1s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q15u8)); + q13s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); + + q2s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q11u8)); + q13s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q12u8)); + q2s8 = vshrq_n_s8(q2s8, 3); + q13s8 = vshrq_n_s8(q13s8, 3); + + q7s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q2s8); + q6s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q13s8); + + q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); + + q0u16 = q11u16 = q12u16 = q13u16 = q14u16 = q15u16 = vdupq_n_u16(63); + d5 = vdup_n_s8(9); + d4 = vdup_n_s8(18); + + q0s16 = vmlal_s8(vreinterpretq_s16_u16(q0u16), vget_low_s8(q1s8), d5); + q11s16 = vmlal_s8(vreinterpretq_s16_u16(q11u16), vget_high_s8(q1s8), d5); + d5 = vdup_n_s8(27); + q12s16 = vmlal_s8(vreinterpretq_s16_u16(q12u16), vget_low_s8(q1s8), d4); + q13s16 = vmlal_s8(vreinterpretq_s16_u16(q13u16), vget_high_s8(q1s8), d4); + q14s16 = vmlal_s8(vreinterpretq_s16_u16(q14u16), vget_low_s8(q1s8), d5); + q15s16 = vmlal_s8(vreinterpretq_s16_u16(q15u16), vget_high_s8(q1s8), d5); + + d0 = vqshrn_n_s16(q0s16 , 7); + d1 = vqshrn_n_s16(q11s16, 7); + d24 = vqshrn_n_s16(q12s16, 7); + d25 = vqshrn_n_s16(q13s16, 7); + d28 = vqshrn_n_s16(q14s16, 7); + d29 = vqshrn_n_s16(q15s16, 7); + + q0s8 = vcombine_s8(d0, d1); + q12s8 = vcombine_s8(d24, d25); + q14s8 = vcombine_s8(d28, d29); + + q11s8 = vqsubq_s8(vreinterpretq_s8_u8(q9), q0s8); + q0s8 = vqaddq_s8(vreinterpretq_s8_u8(q4), q0s8); + q13s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q12s8); + q12s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q12s8); + q15s8 = vqsubq_s8((q7s8), q14s8); + q14s8 = vqaddq_s8((q6s8), q14s8); + + q1u8 = vdupq_n_u8(0x80); + *q9r = veorq_u8(vreinterpretq_u8_s8(q11s8), q1u8); + *q8r = veorq_u8(vreinterpretq_u8_s8(q13s8), q1u8); + *q7r = veorq_u8(vreinterpretq_u8_s8(q15s8), q1u8); + *q6r = veorq_u8(vreinterpretq_u8_s8(q14s8), q1u8); + *q5r = veorq_u8(vreinterpretq_u8_s8(q12s8), q1u8); + *q4r = veorq_u8(vreinterpretq_u8_s8(q0s8), q1u8); + return; +} + +void vp8_mbloop_filter_horizontal_edge_y_neon( + unsigned char *src, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh) { + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + src -= (pitch << 2); + + q3 = vld1q_u8(src); + src += pitch; + q4 = vld1q_u8(src); + src += pitch; + q5 = vld1q_u8(src); + src += pitch; + q6 = vld1q_u8(src); + src += pitch; + q7 = vld1q_u8(src); + src += pitch; + q8 = vld1q_u8(src); + src += pitch; + q9 = vld1q_u8(src); + src += pitch; + q10 = vld1q_u8(src); + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + src -= (pitch * 6); + vst1q_u8(src, q4); + src += pitch; + vst1q_u8(src, q5); + src += pitch; + vst1q_u8(src, q6); + src += pitch; + vst1q_u8(src, q7); + src += pitch; + vst1q_u8(src, q8); + src += pitch; + vst1q_u8(src, q9); + return; +} + +void vp8_mbloop_filter_horizontal_edge_uv_neon( + unsigned char *u, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh, + unsigned char *v) { + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + u -= (pitch << 2); + v -= (pitch << 2); + + d6 = vld1_u8(u); + u += pitch; + d7 = vld1_u8(v); + v += pitch; + d8 = vld1_u8(u); + u += pitch; + d9 = vld1_u8(v); + v += pitch; + d10 = vld1_u8(u); + u += pitch; + d11 = vld1_u8(v); + v += pitch; + d12 = vld1_u8(u); + u += pitch; + d13 = vld1_u8(v); + v += pitch; + d14 = vld1_u8(u); + u += pitch; + d15 = vld1_u8(v); + v += pitch; + d16 = vld1_u8(u); + u += pitch; + d17 = vld1_u8(v); + v += pitch; + d18 = vld1_u8(u); + u += pitch; + d19 = vld1_u8(v); + v += pitch; + d20 = vld1_u8(u); + d21 = vld1_u8(v); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + u -= (pitch * 6); + v -= (pitch * 6); + vst1_u8(u, vget_low_u8(q4)); + u += pitch; + vst1_u8(v, vget_high_u8(q4)); + v += pitch; + vst1_u8(u, vget_low_u8(q5)); + u += pitch; + vst1_u8(v, vget_high_u8(q5)); + v += pitch; + vst1_u8(u, vget_low_u8(q6)); + u += pitch; + vst1_u8(v, vget_high_u8(q6)); + v += pitch; + vst1_u8(u, vget_low_u8(q7)); + u += pitch; + vst1_u8(v, vget_high_u8(q7)); + v += pitch; + vst1_u8(u, vget_low_u8(q8)); + u += pitch; + vst1_u8(v, vget_high_u8(q8)); + v += pitch; + vst1_u8(u, vget_low_u8(q9)); + vst1_u8(v, vget_high_u8(q9)); + return; +} + +void vp8_mbloop_filter_vertical_edge_y_neon( + unsigned char *src, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh) { + unsigned char *s1, *s2; + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; + uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; + uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + s1 = src - 4; + s2 = s1 + 8 * pitch; + d6 = vld1_u8(s1); + s1 += pitch; + d7 = vld1_u8(s2); + s2 += pitch; + d8 = vld1_u8(s1); + s1 += pitch; + d9 = vld1_u8(s2); + s2 += pitch; + d10 = vld1_u8(s1); + s1 += pitch; + d11 = vld1_u8(s2); + s2 += pitch; + d12 = vld1_u8(s1); + s1 += pitch; + d13 = vld1_u8(s2); + s2 += pitch; + d14 = vld1_u8(s1); + s1 += pitch; + d15 = vld1_u8(s2); + s2 += pitch; + d16 = vld1_u8(s1); + s1 += pitch; + d17 = vld1_u8(s2); + s2 += pitch; + d18 = vld1_u8(s1); + s1 += pitch; + d19 = vld1_u8(s2); + s2 += pitch; + d20 = vld1_u8(s1); + d21 = vld1_u8(s2); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + s1 -= 7 * pitch; + s2 -= 7 * pitch; + + vst1_u8(s1, vget_low_u8(q3)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q3)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q4)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q4)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q5)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q5)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q6)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q6)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q7)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q7)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q8)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q8)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q9)); + s1 += pitch; + vst1_u8(s2, vget_high_u8(q9)); + s2 += pitch; + vst1_u8(s1, vget_low_u8(q10)); + vst1_u8(s2, vget_high_u8(q10)); + return; +} + +void vp8_mbloop_filter_vertical_edge_uv_neon( + unsigned char *u, + int pitch, + unsigned char blimit, + unsigned char limit, + unsigned char thresh, + unsigned char *v) { + unsigned char *us, *ud; + unsigned char *vs, *vd; + uint8x16_t qblimit, qlimit, qthresh, q3, q4; + uint8x16_t q5, q6, q7, q8, q9, q10; + uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; + uint8x8_t d15, d16, d17, d18, d19, d20, d21; + uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; + uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; + uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; + + qblimit = vdupq_n_u8(blimit); + qlimit = vdupq_n_u8(limit); + qthresh = vdupq_n_u8(thresh); + + us = u - 4; + vs = v - 4; + d6 = vld1_u8(us); + us += pitch; + d7 = vld1_u8(vs); + vs += pitch; + d8 = vld1_u8(us); + us += pitch; + d9 = vld1_u8(vs); + vs += pitch; + d10 = vld1_u8(us); + us += pitch; + d11 = vld1_u8(vs); + vs += pitch; + d12 = vld1_u8(us); + us += pitch; + d13 = vld1_u8(vs); + vs += pitch; + d14 = vld1_u8(us); + us += pitch; + d15 = vld1_u8(vs); + vs += pitch; + d16 = vld1_u8(us); + us += pitch; + d17 = vld1_u8(vs); + vs += pitch; + d18 = vld1_u8(us); + us += pitch; + d19 = vld1_u8(vs); + vs += pitch; + d20 = vld1_u8(us); + d21 = vld1_u8(vs); + + q3 = vcombine_u8(d6, d7); + q4 = vcombine_u8(d8, d9); + q5 = vcombine_u8(d10, d11); + q6 = vcombine_u8(d12, d13); + q7 = vcombine_u8(d14, d15); + q8 = vcombine_u8(d16, d17); + q9 = vcombine_u8(d18, d19); + q10 = vcombine_u8(d20, d21); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4, + q5, q6, q7, q8, q9, q10, + &q4, &q5, &q6, &q7, &q8, &q9); + + q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); + q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); + q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); + q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); + + q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), + vreinterpretq_u16_u32(q2tmp2.val[0])); + q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), + vreinterpretq_u16_u32(q2tmp3.val[0])); + q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), + vreinterpretq_u16_u32(q2tmp2.val[1])); + q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), + vreinterpretq_u16_u32(q2tmp3.val[1])); + + q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), + vreinterpretq_u8_u16(q2tmp5.val[0])); + q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), + vreinterpretq_u8_u16(q2tmp5.val[1])); + q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), + vreinterpretq_u8_u16(q2tmp7.val[0])); + q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), + vreinterpretq_u8_u16(q2tmp7.val[1])); + + q3 = q2tmp8.val[0]; + q4 = q2tmp8.val[1]; + q5 = q2tmp9.val[0]; + q6 = q2tmp9.val[1]; + q7 = q2tmp10.val[0]; + q8 = q2tmp10.val[1]; + q9 = q2tmp11.val[0]; + q10 = q2tmp11.val[1]; + + ud = u - 4; + vst1_u8(ud, vget_low_u8(q3)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q4)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q5)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q6)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q7)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q8)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q9)); + ud += pitch; + vst1_u8(ud, vget_low_u8(q10)); + + vd = v - 4; + vst1_u8(vd, vget_high_u8(q3)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q4)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q5)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q6)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q7)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q8)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q9)); + vd += pitch; + vst1_u8(vd, vget_high_u8(q10)); + return; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad16_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad16_neon.asm deleted file mode 100644 index d7c590e15a2..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad16_neon.asm +++ /dev/null @@ -1,207 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_sad16x16_neon| - EXPORT |vp8_sad16x8_neon| - - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -; r0 unsigned char *src_ptr -; r1 int src_stride -; r2 unsigned char *ref_ptr -; r3 int ref_stride -|vp8_sad16x16_neon| PROC -;; - vld1.8 {q0}, [r0], r1 - vld1.8 {q4}, [r2], r3 - - vld1.8 {q1}, [r0], r1 - vld1.8 {q5}, [r2], r3 - - vabdl.u8 q12, d0, d8 - vabdl.u8 q13, d1, d9 - - vld1.8 {q2}, [r0], r1 - vld1.8 {q6}, [r2], r3 - - vabal.u8 q12, d2, d10 - vabal.u8 q13, d3, d11 - - vld1.8 {q3}, [r0], r1 - vld1.8 {q7}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q13, d5, d13 - -;; - vld1.8 {q0}, [r0], r1 - vld1.8 {q4}, [r2], r3 - - vabal.u8 q12, d6, d14 - vabal.u8 q13, d7, d15 - - vld1.8 {q1}, [r0], r1 - vld1.8 {q5}, [r2], r3 - - vabal.u8 q12, d0, d8 - vabal.u8 q13, d1, d9 - - vld1.8 {q2}, [r0], r1 - vld1.8 {q6}, [r2], r3 - - vabal.u8 q12, d2, d10 - vabal.u8 q13, d3, d11 - - vld1.8 {q3}, [r0], r1 - vld1.8 {q7}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q13, d5, d13 - -;; - vld1.8 {q0}, [r0], r1 - vld1.8 {q4}, [r2], r3 - - vabal.u8 q12, d6, d14 - vabal.u8 q13, d7, d15 - - vld1.8 {q1}, [r0], r1 - vld1.8 {q5}, [r2], r3 - - vabal.u8 q12, d0, d8 - vabal.u8 q13, d1, d9 - - vld1.8 {q2}, [r0], r1 - vld1.8 {q6}, [r2], r3 - - vabal.u8 q12, d2, d10 - vabal.u8 q13, d3, d11 - - vld1.8 {q3}, [r0], r1 - vld1.8 {q7}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q13, d5, d13 - -;; - vld1.8 {q0}, [r0], r1 - vld1.8 {q4}, [r2], r3 - - vabal.u8 q12, d6, d14 - vabal.u8 q13, d7, d15 - - vld1.8 {q1}, [r0], r1 - vld1.8 {q5}, [r2], r3 - - vabal.u8 q12, d0, d8 - vabal.u8 q13, d1, d9 - - vld1.8 {q2}, [r0], r1 - vld1.8 {q6}, [r2], r3 - - vabal.u8 q12, d2, d10 - vabal.u8 q13, d3, d11 - - vld1.8 {q3}, [r0] - vld1.8 {q7}, [r2] - - vabal.u8 q12, d4, d12 - vabal.u8 q13, d5, d13 - - vabal.u8 q12, d6, d14 - vabal.u8 q13, d7, d15 - - vadd.u16 q0, q12, q13 - - vpaddl.u16 q1, q0 - vpaddl.u32 q0, q1 - - vadd.u32 d0, d0, d1 - - vmov.32 r0, d0[0] - - bx lr - - ENDP - -;============================== -;unsigned int vp8_sad16x8_c( -; unsigned char *src_ptr, -; int src_stride, -; unsigned char *ref_ptr, -; int ref_stride) -|vp8_sad16x8_neon| PROC - vld1.8 {q0}, [r0], r1 - vld1.8 {q4}, [r2], r3 - - vld1.8 {q1}, [r0], r1 - vld1.8 {q5}, [r2], r3 - - vabdl.u8 q12, d0, d8 - vabdl.u8 q13, d1, d9 - - vld1.8 {q2}, [r0], r1 - vld1.8 {q6}, [r2], r3 - - vabal.u8 q12, d2, d10 - vabal.u8 q13, d3, d11 - - vld1.8 {q3}, [r0], r1 - vld1.8 {q7}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q13, d5, d13 - - vld1.8 {q0}, [r0], r1 - vld1.8 {q4}, [r2], r3 - - vabal.u8 q12, d6, d14 - vabal.u8 q13, d7, d15 - - vld1.8 {q1}, [r0], r1 - vld1.8 {q5}, [r2], r3 - - vabal.u8 q12, d0, d8 - vabal.u8 q13, d1, d9 - - vld1.8 {q2}, [r0], r1 - vld1.8 {q6}, [r2], r3 - - vabal.u8 q12, d2, d10 - vabal.u8 q13, d3, d11 - - vld1.8 {q3}, [r0], r1 - vld1.8 {q7}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q13, d5, d13 - - vabal.u8 q12, d6, d14 - vabal.u8 q13, d7, d15 - - vadd.u16 q0, q12, q13 - - vpaddl.u16 q1, q0 - vpaddl.u32 q0, q1 - - vadd.u32 d0, d0, d1 - - vmov.32 r0, d0[0] - - bx lr - - ENDP - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad8_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad8_neon.asm deleted file mode 100644 index 23ba6df93a4..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad8_neon.asm +++ /dev/null @@ -1,209 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_sad8x8_neon| - EXPORT |vp8_sad8x16_neon| - EXPORT |vp8_sad4x4_neon| - - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 -; unsigned int vp8_sad8x8_c( -; unsigned char *src_ptr, -; int src_stride, -; unsigned char *ref_ptr, -; int ref_stride) - -|vp8_sad8x8_neon| PROC - vld1.8 {d0}, [r0], r1 - vld1.8 {d8}, [r2], r3 - - vld1.8 {d2}, [r0], r1 - vld1.8 {d10}, [r2], r3 - - vabdl.u8 q12, d0, d8 - - vld1.8 {d4}, [r0], r1 - vld1.8 {d12}, [r2], r3 - - vabal.u8 q12, d2, d10 - - vld1.8 {d6}, [r0], r1 - vld1.8 {d14}, [r2], r3 - - vabal.u8 q12, d4, d12 - - vld1.8 {d0}, [r0], r1 - vld1.8 {d8}, [r2], r3 - - vabal.u8 q12, d6, d14 - - vld1.8 {d2}, [r0], r1 - vld1.8 {d10}, [r2], r3 - - vabal.u8 q12, d0, d8 - - vld1.8 {d4}, [r0], r1 - vld1.8 {d12}, [r2], r3 - - vabal.u8 q12, d2, d10 - - vld1.8 {d6}, [r0], r1 - vld1.8 {d14}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q12, d6, d14 - - vpaddl.u16 q1, q12 - vpaddl.u32 q0, q1 - vadd.u32 d0, d0, d1 - - vmov.32 r0, d0[0] - - bx lr - - ENDP - -;============================ -;unsigned int vp8_sad8x16_c( -; unsigned char *src_ptr, -; int src_stride, -; unsigned char *ref_ptr, -; int ref_stride) - -|vp8_sad8x16_neon| PROC - vld1.8 {d0}, [r0], r1 - vld1.8 {d8}, [r2], r3 - - vld1.8 {d2}, [r0], r1 - vld1.8 {d10}, [r2], r3 - - vabdl.u8 q12, d0, d8 - - vld1.8 {d4}, [r0], r1 - vld1.8 {d12}, [r2], r3 - - vabal.u8 q12, d2, d10 - - vld1.8 {d6}, [r0], r1 - vld1.8 {d14}, [r2], r3 - - vabal.u8 q12, d4, d12 - - vld1.8 {d0}, [r0], r1 - vld1.8 {d8}, [r2], r3 - - vabal.u8 q12, d6, d14 - - vld1.8 {d2}, [r0], r1 - vld1.8 {d10}, [r2], r3 - - vabal.u8 q12, d0, d8 - - vld1.8 {d4}, [r0], r1 - vld1.8 {d12}, [r2], r3 - - vabal.u8 q12, d2, d10 - - vld1.8 {d6}, [r0], r1 - vld1.8 {d14}, [r2], r3 - - vabal.u8 q12, d4, d12 - - vld1.8 {d0}, [r0], r1 - vld1.8 {d8}, [r2], r3 - - vabal.u8 q12, d6, d14 - - vld1.8 {d2}, [r0], r1 - vld1.8 {d10}, [r2], r3 - - vabal.u8 q12, d0, d8 - - vld1.8 {d4}, [r0], r1 - vld1.8 {d12}, [r2], r3 - - vabal.u8 q12, d2, d10 - - vld1.8 {d6}, [r0], r1 - vld1.8 {d14}, [r2], r3 - - vabal.u8 q12, d4, d12 - - vld1.8 {d0}, [r0], r1 - vld1.8 {d8}, [r2], r3 - - vabal.u8 q12, d6, d14 - - vld1.8 {d2}, [r0], r1 - vld1.8 {d10}, [r2], r3 - - vabal.u8 q12, d0, d8 - - vld1.8 {d4}, [r0], r1 - vld1.8 {d12}, [r2], r3 - - vabal.u8 q12, d2, d10 - - vld1.8 {d6}, [r0], r1 - vld1.8 {d14}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q12, d6, d14 - - vpaddl.u16 q1, q12 - vpaddl.u32 q0, q1 - vadd.u32 d0, d0, d1 - - vmov.32 r0, d0[0] - - bx lr - - ENDP - -;=========================== -;unsigned int vp8_sad4x4_c( -; unsigned char *src_ptr, -; int src_stride, -; unsigned char *ref_ptr, -; int ref_stride) - -|vp8_sad4x4_neon| PROC - vld1.8 {d0}, [r0], r1 - vld1.8 {d8}, [r2], r3 - - vld1.8 {d2}, [r0], r1 - vld1.8 {d10}, [r2], r3 - - vabdl.u8 q12, d0, d8 - - vld1.8 {d4}, [r0], r1 - vld1.8 {d12}, [r2], r3 - - vabal.u8 q12, d2, d10 - - vld1.8 {d6}, [r0], r1 - vld1.8 {d14}, [r2], r3 - - vabal.u8 q12, d4, d12 - vabal.u8 q12, d6, d14 - - vpaddl.u16 d1, d24 - vpaddl.u32 d0, d1 - vmov.32 r0, d0[0] - - bx lr - - ENDP - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad_neon.c new file mode 100644 index 00000000000..6595ac0519b --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sad_neon.c @@ -0,0 +1,184 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +unsigned int vp8_sad8x8_neon( + unsigned char *src_ptr, + int src_stride, + unsigned char *ref_ptr, + int ref_stride) { + uint8x8_t d0, d8; + uint16x8_t q12; + uint32x4_t q1; + uint64x2_t q3; + uint32x2_t d5; + int i; + + d0 = vld1_u8(src_ptr); + src_ptr += src_stride; + d8 = vld1_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabdl_u8(d0, d8); + + for (i = 0; i < 7; i++) { + d0 = vld1_u8(src_ptr); + src_ptr += src_stride; + d8 = vld1_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabal_u8(q12, d0, d8); + } + + q1 = vpaddlq_u16(q12); + q3 = vpaddlq_u32(q1); + d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)), + vreinterpret_u32_u64(vget_high_u64(q3))); + + return vget_lane_u32(d5, 0); +} + +unsigned int vp8_sad8x16_neon( + unsigned char *src_ptr, + int src_stride, + unsigned char *ref_ptr, + int ref_stride) { + uint8x8_t d0, d8; + uint16x8_t q12; + uint32x4_t q1; + uint64x2_t q3; + uint32x2_t d5; + int i; + + d0 = vld1_u8(src_ptr); + src_ptr += src_stride; + d8 = vld1_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabdl_u8(d0, d8); + + for (i = 0; i < 15; i++) { + d0 = vld1_u8(src_ptr); + src_ptr += src_stride; + d8 = vld1_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabal_u8(q12, d0, d8); + } + + q1 = vpaddlq_u16(q12); + q3 = vpaddlq_u32(q1); + d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)), + vreinterpret_u32_u64(vget_high_u64(q3))); + + return vget_lane_u32(d5, 0); +} + +unsigned int vp8_sad4x4_neon( + unsigned char *src_ptr, + int src_stride, + unsigned char *ref_ptr, + int ref_stride) { + uint8x8_t d0, d8; + uint16x8_t q12; + uint32x2_t d1; + uint64x1_t d3; + int i; + + d0 = vld1_u8(src_ptr); + src_ptr += src_stride; + d8 = vld1_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabdl_u8(d0, d8); + + for (i = 0; i < 3; i++) { + d0 = vld1_u8(src_ptr); + src_ptr += src_stride; + d8 = vld1_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabal_u8(q12, d0, d8); + } + + d1 = vpaddl_u16(vget_low_u16(q12)); + d3 = vpaddl_u32(d1); + + return vget_lane_u32(vreinterpret_u32_u64(d3), 0); +} + +unsigned int vp8_sad16x16_neon( + unsigned char *src_ptr, + int src_stride, + unsigned char *ref_ptr, + int ref_stride) { + uint8x16_t q0, q4; + uint16x8_t q12, q13; + uint32x4_t q1; + uint64x2_t q3; + uint32x2_t d5; + int i; + + q0 = vld1q_u8(src_ptr); + src_ptr += src_stride; + q4 = vld1q_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabdl_u8(vget_low_u8(q0), vget_low_u8(q4)); + q13 = vabdl_u8(vget_high_u8(q0), vget_high_u8(q4)); + + for (i = 0; i < 15; i++) { + q0 = vld1q_u8(src_ptr); + src_ptr += src_stride; + q4 = vld1q_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabal_u8(q12, vget_low_u8(q0), vget_low_u8(q4)); + q13 = vabal_u8(q13, vget_high_u8(q0), vget_high_u8(q4)); + } + + q12 = vaddq_u16(q12, q13); + q1 = vpaddlq_u16(q12); + q3 = vpaddlq_u32(q1); + d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)), + vreinterpret_u32_u64(vget_high_u64(q3))); + + return vget_lane_u32(d5, 0); +} + +unsigned int vp8_sad16x8_neon( + unsigned char *src_ptr, + int src_stride, + unsigned char *ref_ptr, + int ref_stride) { + uint8x16_t q0, q4; + uint16x8_t q12, q13; + uint32x4_t q1; + uint64x2_t q3; + uint32x2_t d5; + int i; + + q0 = vld1q_u8(src_ptr); + src_ptr += src_stride; + q4 = vld1q_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabdl_u8(vget_low_u8(q0), vget_low_u8(q4)); + q13 = vabdl_u8(vget_high_u8(q0), vget_high_u8(q4)); + + for (i = 0; i < 7; i++) { + q0 = vld1q_u8(src_ptr); + src_ptr += src_stride; + q4 = vld1q_u8(ref_ptr); + ref_ptr += ref_stride; + q12 = vabal_u8(q12, vget_low_u8(q0), vget_low_u8(q4)); + q13 = vabal_u8(q13, vget_high_u8(q0), vget_high_u8(q4)); + } + + q12 = vaddq_u16(q12, q13); + q1 = vpaddlq_u16(q12); + q3 = vpaddlq_u32(q1); + d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)), + vreinterpret_u32_u64(vget_high_u64(q3))); + + return vget_lane_u32(d5, 0); +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/save_reg_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/save_reg_neon.asm deleted file mode 100644 index fd7002e7a9e..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/save_reg_neon.asm +++ /dev/null @@ -1,36 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_push_neon| - EXPORT |vp8_pop_neon| - - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -|vp8_push_neon| PROC - vst1.i64 {d8, d9, d10, d11}, [r0]! - vst1.i64 {d12, d13, d14, d15}, [r0]! - bx lr - - ENDP - -|vp8_pop_neon| PROC - vld1.i64 {d8, d9, d10, d11}, [r0]! - vld1.i64 {d12, d13, d14, d15}, [r0]! - bx lr - - ENDP - - END - diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.asm deleted file mode 100644 index 67d2ab0150d..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.asm +++ /dev/null @@ -1,139 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_short_idct4x4llm_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -;************************************************************* -;void vp8_short_idct4x4llm_c(short *input, unsigned char *pred, int pitch, -; unsigned char *dst, int stride) -;r0 short * input -;r1 short * pred -;r2 int pitch -;r3 unsigned char dst -;sp int stride -;************************************************************* - -; static const int cospi8sqrt2minus1=20091; -; static const int sinpi8sqrt2 =35468; -; static const int rounding = 0; - -; Optimization note: The resulted data from dequantization are signed -; 13-bit data that is in the range of [-4096, 4095]. This allows to -; use "vqdmulh"(neon) instruction since it won't go out of range -; (13+16+1=30bits<32bits). This instruction gives the high half -; result of the multiplication that is needed in IDCT. - -|vp8_short_idct4x4llm_neon| PROC - adr r12, idct_coeff - vld1.16 {q1, q2}, [r0] - vld1.16 {d0}, [r12] - - vswp d3, d4 ;q2(vp[4] vp[12]) - ldr r0, [sp] ; stride - - vqdmulh.s16 q3, q2, d0[2] - vqdmulh.s16 q4, q2, d0[0] - - vqadd.s16 d12, d2, d3 ;a1 - vqsub.s16 d13, d2, d3 ;b1 - - vshr.s16 q3, q3, #1 - vshr.s16 q4, q4, #1 - - vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number) - vqadd.s16 q4, q4, q2 - - ;d6 - c1:temp1 - ;d7 - d1:temp2 - ;d8 - d1:temp1 - ;d9 - c1:temp2 - - vqsub.s16 d10, d6, d9 ;c1 - vqadd.s16 d11, d7, d8 ;d1 - - vqadd.s16 d2, d12, d11 - vqadd.s16 d3, d13, d10 - vqsub.s16 d4, d13, d10 - vqsub.s16 d5, d12, d11 - - vtrn.32 d2, d4 - vtrn.32 d3, d5 - vtrn.16 d2, d3 - vtrn.16 d4, d5 - - vswp d3, d4 - - vqdmulh.s16 q3, q2, d0[2] - vqdmulh.s16 q4, q2, d0[0] - - vqadd.s16 d12, d2, d3 ;a1 - vqsub.s16 d13, d2, d3 ;b1 - - vshr.s16 q3, q3, #1 - vshr.s16 q4, q4, #1 - - vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number) - vqadd.s16 q4, q4, q2 - - vqsub.s16 d10, d6, d9 ;c1 - vqadd.s16 d11, d7, d8 ;d1 - - vqadd.s16 d2, d12, d11 - vqadd.s16 d3, d13, d10 - vqsub.s16 d4, d13, d10 - vqsub.s16 d5, d12, d11 - - vrshr.s16 d2, d2, #3 - vrshr.s16 d3, d3, #3 - vrshr.s16 d4, d4, #3 - vrshr.s16 d5, d5, #3 - - vtrn.32 d2, d4 - vtrn.32 d3, d5 - vtrn.16 d2, d3 - vtrn.16 d4, d5 - - ; load prediction data - vld1.32 d6[0], [r1], r2 - vld1.32 d6[1], [r1], r2 - vld1.32 d7[0], [r1], r2 - vld1.32 d7[1], [r1], r2 - - ; add prediction and residual - vaddw.u8 q1, q1, d6 - vaddw.u8 q2, q2, d7 - - vqmovun.s16 d1, q1 - vqmovun.s16 d2, q2 - - ; store to destination - vst1.32 d1[0], [r3], r0 - vst1.32 d1[1], [r3], r0 - vst1.32 d2[0], [r3], r0 - vst1.32 d2[1], [r3], r0 - - bx lr - - ENDP - -;----------------- - -idct_coeff - DCD 0x4e7b4e7b, 0x8a8c8a8c - -;20091, 20091, 35468, 35468 - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.c new file mode 100644 index 00000000000..373afa6ed35 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/shortidct4x4llm_neon.c @@ -0,0 +1,123 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +static const int16_t cospi8sqrt2minus1 = 20091; +static const int16_t sinpi8sqrt2 = 35468; + +void vp8_short_idct4x4llm_neon( + int16_t *input, + unsigned char *pred_ptr, + int pred_stride, + unsigned char *dst_ptr, + int dst_stride) { + int i; + uint32x2_t d6u32 = vdup_n_u32(0); + uint8x8_t d1u8; + int16x4_t d2, d3, d4, d5, d10, d11, d12, d13; + uint16x8_t q1u16; + int16x8_t q1s16, q2s16, q3s16, q4s16; + int32x2x2_t v2tmp0, v2tmp1; + int16x4x2_t v2tmp2, v2tmp3; + + d2 = vld1_s16(input); + d3 = vld1_s16(input + 4); + d4 = vld1_s16(input + 8); + d5 = vld1_s16(input + 12); + + // 1st for loop + q1s16 = vcombine_s16(d2, d4); // Swap d3 d4 here + q2s16 = vcombine_s16(d3, d5); + + q3s16 = vqdmulhq_n_s16(q2s16, sinpi8sqrt2); + q4s16 = vqdmulhq_n_s16(q2s16, cospi8sqrt2minus1); + + d12 = vqadd_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // a1 + d13 = vqsub_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // b1 + + q3s16 = vshrq_n_s16(q3s16, 1); + q4s16 = vshrq_n_s16(q4s16, 1); + + q3s16 = vqaddq_s16(q3s16, q2s16); + q4s16 = vqaddq_s16(q4s16, q2s16); + + d10 = vqsub_s16(vget_low_s16(q3s16), vget_high_s16(q4s16)); // c1 + d11 = vqadd_s16(vget_high_s16(q3s16), vget_low_s16(q4s16)); // d1 + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + v2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + v2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + v2tmp2 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[0]), + vreinterpret_s16_s32(v2tmp1.val[0])); + v2tmp3 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[1]), + vreinterpret_s16_s32(v2tmp1.val[1])); + + // 2nd for loop + q1s16 = vcombine_s16(v2tmp2.val[0], v2tmp3.val[0]); + q2s16 = vcombine_s16(v2tmp2.val[1], v2tmp3.val[1]); + + q3s16 = vqdmulhq_n_s16(q2s16, sinpi8sqrt2); + q4s16 = vqdmulhq_n_s16(q2s16, cospi8sqrt2minus1); + + d12 = vqadd_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // a1 + d13 = vqsub_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // b1 + + q3s16 = vshrq_n_s16(q3s16, 1); + q4s16 = vshrq_n_s16(q4s16, 1); + + q3s16 = vqaddq_s16(q3s16, q2s16); + q4s16 = vqaddq_s16(q4s16, q2s16); + + d10 = vqsub_s16(vget_low_s16(q3s16), vget_high_s16(q4s16)); // c1 + d11 = vqadd_s16(vget_high_s16(q3s16), vget_low_s16(q4s16)); // d1 + + d2 = vqadd_s16(d12, d11); + d3 = vqadd_s16(d13, d10); + d4 = vqsub_s16(d13, d10); + d5 = vqsub_s16(d12, d11); + + d2 = vrshr_n_s16(d2, 3); + d3 = vrshr_n_s16(d3, 3); + d4 = vrshr_n_s16(d4, 3); + d5 = vrshr_n_s16(d5, 3); + + v2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4)); + v2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5)); + v2tmp2 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[0]), + vreinterpret_s16_s32(v2tmp1.val[0])); + v2tmp3 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[1]), + vreinterpret_s16_s32(v2tmp1.val[1])); + + q1s16 = vcombine_s16(v2tmp2.val[0], v2tmp2.val[1]); + q2s16 = vcombine_s16(v2tmp3.val[0], v2tmp3.val[1]); + + // dc_only_idct_add + for (i = 0; i < 2; i++, q1s16 = q2s16) { + d6u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d6u32, 0); + pred_ptr += pred_stride; + d6u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d6u32, 1); + pred_ptr += pred_stride; + + q1u16 = vaddw_u8(vreinterpretq_u16_s16(q1s16), + vreinterpret_u8_u32(d6u32)); + d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q1u16)); + + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d1u8), 0); + dst_ptr += dst_stride; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d1u8), 1); + dst_ptr += dst_stride; + } + return; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict16x16_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict16x16_neon.asm deleted file mode 100644 index 9fdafd3609e..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict16x16_neon.asm +++ /dev/null @@ -1,490 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_sixtap_predict16x16_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -filter16_coeff - DCD 0, 0, 128, 0, 0, 0, 0, 0 - DCD 0, -6, 123, 12, -1, 0, 0, 0 - DCD 2, -11, 108, 36, -8, 1, 0, 0 - DCD 0, -9, 93, 50, -6, 0, 0, 0 - DCD 3, -16, 77, 77, -16, 3, 0, 0 - DCD 0, -6, 50, 93, -9, 0, 0, 0 - DCD 1, -8, 36, 108, -11, 2, 0, 0 - DCD 0, -1, 12, 123, -6, 0, 0, 0 - -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; r4 unsigned char *dst_ptr, -; stack(r5) int dst_pitch - -;Note: To take advantage of 8-bit mulplication instruction in NEON. First apply abs() to -; filter coeffs to make them u8. Then, use vmlsl for negtive coeffs. After multiplication, -; the result can be negtive. So, I treat the result as s16. But, since it is also possible -; that the result can be a large positive number (> 2^15-1), which could be confused as a -; negtive number. To avoid that error, apply filter coeffs in the order of 0, 1, 4 ,5 ,2, -; which ensures that the result stays in s16 range. Finally, saturated add the result by -; applying 3rd filter coeff. Same applys to other filter functions. - -|vp8_sixtap_predict16x16_neon| PROC - push {r4-r5, lr} - - adr r12, filter16_coeff - ldr r4, [sp, #12] ;load parameters from stack - ldr r5, [sp, #16] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq secondpass_filter16x16_only - - add r2, r12, r2, lsl #5 ;calculate filter location - - cmp r3, #0 ;skip second_pass filter if yoffset=0 - - vld1.s32 {q14, q15}, [r2] ;load first_pass filter - - beq firstpass_filter16x16_only - - sub sp, sp, #336 ;reserve space on stack for temporary storage - mov lr, sp - - vabs.s32 q12, q14 - vabs.s32 q13, q15 - - mov r2, #7 ;loop counter - sub r0, r0, #2 ;move srcptr back to (line-2) and (column-2) - sub r0, r0, r1, lsl #1 - - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vdup.8 d1, d24[4] - vdup.8 d2, d25[0] - vdup.8 d3, d25[4] - vdup.8 d4, d26[0] - vdup.8 d5, d26[4] - -;First Pass: output_height lines x output_width columns (21x16) -filt_blk2d_fp16x16_loop_neon - vld1.u8 {d6, d7, d8}, [r0], r1 ;load src data - vld1.u8 {d9, d10, d11}, [r0], r1 - vld1.u8 {d12, d13, d14}, [r0], r1 - - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vmull.u8 q8, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q9, d7, d0 - vmull.u8 q10, d9, d0 - vmull.u8 q11, d10, d0 - vmull.u8 q12, d12, d0 - vmull.u8 q13, d13, d0 - - vext.8 d28, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d29, d9, d10, #1 - vext.8 d30, d12, d13, #1 - - vmlsl.u8 q8, d28, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q10, d29, d1 - vmlsl.u8 q12, d30, d1 - - vext.8 d28, d7, d8, #1 - vext.8 d29, d10, d11, #1 - vext.8 d30, d13, d14, #1 - - vmlsl.u8 q9, d28, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q11, d29, d1 - vmlsl.u8 q13, d30, d1 - - vext.8 d28, d6, d7, #4 ;construct src_ptr[2] - vext.8 d29, d9, d10, #4 - vext.8 d30, d12, d13, #4 - - vmlsl.u8 q8, d28, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q10, d29, d4 - vmlsl.u8 q12, d30, d4 - - vext.8 d28, d7, d8, #4 - vext.8 d29, d10, d11, #4 - vext.8 d30, d13, d14, #4 - - vmlsl.u8 q9, d28, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q11, d29, d4 - vmlsl.u8 q13, d30, d4 - - vext.8 d28, d6, d7, #5 ;construct src_ptr[3] - vext.8 d29, d9, d10, #5 - vext.8 d30, d12, d13, #5 - - vmlal.u8 q8, d28, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q10, d29, d5 - vmlal.u8 q12, d30, d5 - - vext.8 d28, d7, d8, #5 - vext.8 d29, d10, d11, #5 - vext.8 d30, d13, d14, #5 - - vmlal.u8 q9, d28, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q11, d29, d5 - vmlal.u8 q13, d30, d5 - - vext.8 d28, d6, d7, #2 ;construct src_ptr[0] - vext.8 d29, d9, d10, #2 - vext.8 d30, d12, d13, #2 - - vmlal.u8 q8, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q10, d29, d2 - vmlal.u8 q12, d30, d2 - - vext.8 d28, d7, d8, #2 - vext.8 d29, d10, d11, #2 - vext.8 d30, d13, d14, #2 - - vmlal.u8 q9, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q11, d29, d2 - vmlal.u8 q13, d30, d2 - - vext.8 d28, d6, d7, #3 ;construct src_ptr[1] - vext.8 d29, d9, d10, #3 - vext.8 d30, d12, d13, #3 - - vext.8 d15, d7, d8, #3 - vext.8 d31, d10, d11, #3 - vext.8 d6, d13, d14, #3 - - vmull.u8 q4, d28, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q5, d29, d3 - vmull.u8 q6, d30, d3 - - vqadd.s16 q8, q4 ;sum of all (src_data*filter_parameters) - vqadd.s16 q10, q5 - vqadd.s16 q12, q6 - - vmull.u8 q6, d15, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q7, d31, d3 - vmull.u8 q3, d6, d3 - - subs r2, r2, #1 - - vqadd.s16 q9, q6 - vqadd.s16 q11, q7 - vqadd.s16 q13, q3 - - vqrshrun.s16 d6, q8, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q9, #7 - vqrshrun.s16 d8, q10, #7 - vqrshrun.s16 d9, q11, #7 - vqrshrun.s16 d10, q12, #7 - vqrshrun.s16 d11, q13, #7 - - vst1.u8 {d6, d7, d8}, [lr]! ;store result - vst1.u8 {d9, d10, d11}, [lr]! - - bne filt_blk2d_fp16x16_loop_neon - -;Second pass: 16x16 -;secondpass_filter - do first 8-columns and then second 8-columns - add r3, r12, r3, lsl #5 - sub lr, lr, #336 - - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - mov r3, #2 ;loop counter - - vabs.s32 q7, q5 - vabs.s32 q8, q6 - - mov r2, #16 - - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vdup.8 d1, d14[4] - vdup.8 d2, d15[0] - vdup.8 d3, d15[4] - vdup.8 d4, d16[0] - vdup.8 d5, d16[4] - -filt_blk2d_sp16x16_outloop_neon - vld1.u8 {d18}, [lr], r2 ;load src data - vld1.u8 {d19}, [lr], r2 - vld1.u8 {d20}, [lr], r2 - vld1.u8 {d21}, [lr], r2 - mov r12, #4 ;loop counter - vld1.u8 {d22}, [lr], r2 - -secondpass_inner_loop_neon - vld1.u8 {d23}, [lr], r2 ;load src data - vld1.u8 {d24}, [lr], r2 - vld1.u8 {d25}, [lr], r2 - vld1.u8 {d26}, [lr], r2 - - vmull.u8 q3, d18, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d19, d0 - vmull.u8 q5, d20, d0 - vmull.u8 q6, d21, d0 - - vmlsl.u8 q3, d19, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q4, d20, d1 - vmlsl.u8 q5, d21, d1 - vmlsl.u8 q6, d22, d1 - - vmlsl.u8 q3, d22, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d23, d4 - vmlsl.u8 q5, d24, d4 - vmlsl.u8 q6, d25, d4 - - vmlal.u8 q3, d20, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d21, d2 - vmlal.u8 q5, d22, d2 - vmlal.u8 q6, d23, d2 - - vmlal.u8 q3, d23, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q4, d24, d5 - vmlal.u8 q5, d25, d5 - vmlal.u8 q6, d26, d5 - - vmull.u8 q7, d21, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q8, d22, d3 - vmull.u8 q9, d23, d3 - vmull.u8 q10, d24, d3 - - subs r12, r12, #1 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q8, #7 - vqrshrun.s16 d8, q9, #7 - vqrshrun.s16 d9, q10, #7 - - vst1.u8 {d6}, [r4], r5 ;store result - vmov q9, q11 - vst1.u8 {d7}, [r4], r5 - vmov q10, q12 - vst1.u8 {d8}, [r4], r5 - vmov d22, d26 - vst1.u8 {d9}, [r4], r5 - - bne secondpass_inner_loop_neon - - subs r3, r3, #1 - sub lr, lr, #336 - add lr, lr, #8 - - sub r4, r4, r5, lsl #4 - add r4, r4, #8 - - bne filt_blk2d_sp16x16_outloop_neon - - add sp, sp, #336 - pop {r4-r5,pc} - -;-------------------- -firstpass_filter16x16_only - vabs.s32 q12, q14 - vabs.s32 q13, q15 - - mov r2, #8 ;loop counter - sub r0, r0, #2 ;move srcptr back to (column-2) - - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vdup.8 d1, d24[4] - vdup.8 d2, d25[0] - vdup.8 d3, d25[4] - vdup.8 d4, d26[0] - vdup.8 d5, d26[4] - -;First Pass: output_height lines x output_width columns (16x16) -filt_blk2d_fpo16x16_loop_neon - vld1.u8 {d6, d7, d8}, [r0], r1 ;load src data - vld1.u8 {d9, d10, d11}, [r0], r1 - - pld [r0] - pld [r0, r1] - - vmull.u8 q6, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q7, d7, d0 - vmull.u8 q8, d9, d0 - vmull.u8 q9, d10, d0 - - vext.8 d20, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d21, d9, d10, #1 - vext.8 d22, d7, d8, #1 - vext.8 d23, d10, d11, #1 - vext.8 d24, d6, d7, #4 ;construct src_ptr[2] - vext.8 d25, d9, d10, #4 - vext.8 d26, d7, d8, #4 - vext.8 d27, d10, d11, #4 - vext.8 d28, d6, d7, #5 ;construct src_ptr[3] - vext.8 d29, d9, d10, #5 - - vmlsl.u8 q6, d20, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d21, d1 - vmlsl.u8 q7, d22, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q9, d23, d1 - vmlsl.u8 q6, d24, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d25, d4 - vmlsl.u8 q7, d26, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q9, d27, d4 - vmlal.u8 q6, d28, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q8, d29, d5 - - vext.8 d20, d7, d8, #5 - vext.8 d21, d10, d11, #5 - vext.8 d22, d6, d7, #2 ;construct src_ptr[0] - vext.8 d23, d9, d10, #2 - vext.8 d24, d7, d8, #2 - vext.8 d25, d10, d11, #2 - - vext.8 d26, d6, d7, #3 ;construct src_ptr[1] - vext.8 d27, d9, d10, #3 - vext.8 d28, d7, d8, #3 - vext.8 d29, d10, d11, #3 - - vmlal.u8 q7, d20, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q9, d21, d5 - vmlal.u8 q6, d22, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d23, d2 - vmlal.u8 q7, d24, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q9, d25, d2 - - vmull.u8 q10, d26, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q11, d27, d3 - vmull.u8 q12, d28, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q15, d29, d3 - - vqadd.s16 q6, q10 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q11 - vqadd.s16 q7, q12 - vqadd.s16 q9, q15 - - subs r2, r2, #1 - - vqrshrun.s16 d6, q6, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q7, #7 - vqrshrun.s16 d8, q8, #7 - vqrshrun.s16 d9, q9, #7 - - vst1.u8 {q3}, [r4], r5 ;store result - vst1.u8 {q4}, [r4], r5 - - bne filt_blk2d_fpo16x16_loop_neon - - pop {r4-r5,pc} - -;-------------------- -secondpass_filter16x16_only -;Second pass: 16x16 - add r3, r12, r3, lsl #5 - sub r0, r0, r1, lsl #1 - - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - mov r3, #2 ;loop counter - - vabs.s32 q7, q5 - vabs.s32 q8, q6 - - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vdup.8 d1, d14[4] - vdup.8 d2, d15[0] - vdup.8 d3, d15[4] - vdup.8 d4, d16[0] - vdup.8 d5, d16[4] - -filt_blk2d_spo16x16_outloop_neon - vld1.u8 {d18}, [r0], r1 ;load src data - vld1.u8 {d19}, [r0], r1 - vld1.u8 {d20}, [r0], r1 - vld1.u8 {d21}, [r0], r1 - mov r12, #4 ;loop counter - vld1.u8 {d22}, [r0], r1 - -secondpass_only_inner_loop_neon - vld1.u8 {d23}, [r0], r1 ;load src data - vld1.u8 {d24}, [r0], r1 - vld1.u8 {d25}, [r0], r1 - vld1.u8 {d26}, [r0], r1 - - vmull.u8 q3, d18, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d19, d0 - vmull.u8 q5, d20, d0 - vmull.u8 q6, d21, d0 - - vmlsl.u8 q3, d19, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q4, d20, d1 - vmlsl.u8 q5, d21, d1 - vmlsl.u8 q6, d22, d1 - - vmlsl.u8 q3, d22, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d23, d4 - vmlsl.u8 q5, d24, d4 - vmlsl.u8 q6, d25, d4 - - vmlal.u8 q3, d20, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d21, d2 - vmlal.u8 q5, d22, d2 - vmlal.u8 q6, d23, d2 - - vmlal.u8 q3, d23, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q4, d24, d5 - vmlal.u8 q5, d25, d5 - vmlal.u8 q6, d26, d5 - - vmull.u8 q7, d21, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q8, d22, d3 - vmull.u8 q9, d23, d3 - vmull.u8 q10, d24, d3 - - subs r12, r12, #1 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q8, #7 - vqrshrun.s16 d8, q9, #7 - vqrshrun.s16 d9, q10, #7 - - vst1.u8 {d6}, [r4], r5 ;store result - vmov q9, q11 - vst1.u8 {d7}, [r4], r5 - vmov q10, q12 - vst1.u8 {d8}, [r4], r5 - vmov d22, d26 - vst1.u8 {d9}, [r4], r5 - - bne secondpass_only_inner_loop_neon - - subs r3, r3, #1 - sub r0, r0, r1, lsl #4 - sub r0, r0, r1, lsl #2 - sub r0, r0, r1 - add r0, r0, #8 - - sub r4, r4, r5, lsl #4 - add r4, r4, #8 - - bne filt_blk2d_spo16x16_outloop_neon - - pop {r4-r5,pc} - - ENDP - -;----------------- - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict4x4_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict4x4_neon.asm deleted file mode 100644 index a4222bc62c5..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict4x4_neon.asm +++ /dev/null @@ -1,422 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_sixtap_predict4x4_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -filter4_coeff - DCD 0, 0, 128, 0, 0, 0, 0, 0 - DCD 0, -6, 123, 12, -1, 0, 0, 0 - DCD 2, -11, 108, 36, -8, 1, 0, 0 - DCD 0, -9, 93, 50, -6, 0, 0, 0 - DCD 3, -16, 77, 77, -16, 3, 0, 0 - DCD 0, -6, 50, 93, -9, 0, 0, 0 - DCD 1, -8, 36, 108, -11, 2, 0, 0 - DCD 0, -1, 12, 123, -6, 0, 0, 0 - -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; stack(r4) unsigned char *dst_ptr, -; stack(lr) int dst_pitch - -|vp8_sixtap_predict4x4_neon| PROC - push {r4, lr} - - adr r12, filter4_coeff - ldr r4, [sp, #8] ;load parameters from stack - ldr lr, [sp, #12] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq secondpass_filter4x4_only - - add r2, r12, r2, lsl #5 ;calculate filter location - - cmp r3, #0 ;skip second_pass filter if yoffset=0 - vld1.s32 {q14, q15}, [r2] ;load first_pass filter - - beq firstpass_filter4x4_only - - vabs.s32 q12, q14 ;get abs(filer_parameters) - vabs.s32 q13, q15 - - sub r0, r0, #2 ;go back 2 columns of src data - sub r0, r0, r1, lsl #1 ;go back 2 lines of src data - -;First pass: output_height lines x output_width columns (9x4) - vld1.u8 {q3}, [r0], r1 ;load first 4-line src data - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vld1.u8 {q4}, [r0], r1 - vdup.8 d1, d24[4] - vld1.u8 {q5}, [r0], r1 - vdup.8 d2, d25[0] - vld1.u8 {q6}, [r0], r1 - vdup.8 d3, d25[4] - vdup.8 d4, d26[0] - vdup.8 d5, d26[4] - - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vext.8 d18, d6, d7, #5 ;construct src_ptr[3] - vext.8 d19, d8, d9, #5 - vext.8 d20, d10, d11, #5 - vext.8 d21, d12, d13, #5 - - vswp d7, d8 ;discard 2nd half data after src_ptr[3] is done - vswp d11, d12 - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[3]) - vzip.32 d20, d21 - vmull.u8 q7, d18, d5 ;(src_ptr[3] * vp8_filter[5]) - vmull.u8 q8, d20, d5 - - vmov q4, q3 ;keep original src data in q4 q6 - vmov q6, q5 - - vzip.32 d6, d7 ;construct src_ptr[-2], and put 2-line data together - vzip.32 d10, d11 - vshr.u64 q9, q4, #8 ;construct src_ptr[-1] - vshr.u64 q10, q6, #8 - vmlal.u8 q7, d6, d0 ;+(src_ptr[-2] * vp8_filter[0]) - vmlal.u8 q8, d10, d0 - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[-1]) - vzip.32 d20, d21 - vshr.u64 q3, q4, #32 ;construct src_ptr[2] - vshr.u64 q5, q6, #32 - vmlsl.u8 q7, d18, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d20, d1 - - vzip.32 d6, d7 ;put 2-line data in 1 register (src_ptr[2]) - vzip.32 d10, d11 - vshr.u64 q9, q4, #16 ;construct src_ptr[0] - vshr.u64 q10, q6, #16 - vmlsl.u8 q7, d6, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d10, d4 - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[0]) - vzip.32 d20, d21 - vshr.u64 q3, q4, #24 ;construct src_ptr[1] - vshr.u64 q5, q6, #24 - vmlal.u8 q7, d18, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d20, d2 - - vzip.32 d6, d7 ;put 2-line data in 1 register (src_ptr[1]) - vzip.32 d10, d11 - vmull.u8 q9, d6, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q10, d10, d3 - - vld1.u8 {q3}, [r0], r1 ;load rest 5-line src data - vld1.u8 {q4}, [r0], r1 - - vqadd.s16 q7, q9 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q10 - - vld1.u8 {q5}, [r0], r1 - vld1.u8 {q6}, [r0], r1 - - vqrshrun.s16 d27, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d28, q8, #7 - - ;First Pass on rest 5-line data - vld1.u8 {q11}, [r0], r1 - - vext.8 d18, d6, d7, #5 ;construct src_ptr[3] - vext.8 d19, d8, d9, #5 - vext.8 d20, d10, d11, #5 - vext.8 d21, d12, d13, #5 - - vswp d7, d8 ;discard 2nd half data after src_ptr[3] is done - vswp d11, d12 - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[3]) - vzip.32 d20, d21 - vext.8 d31, d22, d23, #5 ;construct src_ptr[3] - vmull.u8 q7, d18, d5 ;(src_ptr[3] * vp8_filter[5]) - vmull.u8 q8, d20, d5 - vmull.u8 q12, d31, d5 ;(src_ptr[3] * vp8_filter[5]) - - vmov q4, q3 ;keep original src data in q4 q6 - vmov q6, q5 - - vzip.32 d6, d7 ;construct src_ptr[-2], and put 2-line data together - vzip.32 d10, d11 - vshr.u64 q9, q4, #8 ;construct src_ptr[-1] - vshr.u64 q10, q6, #8 - - vmlal.u8 q7, d6, d0 ;+(src_ptr[-2] * vp8_filter[0]) - vmlal.u8 q8, d10, d0 - vmlal.u8 q12, d22, d0 ;(src_ptr[-2] * vp8_filter[0]) - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[-1]) - vzip.32 d20, d21 - vshr.u64 q3, q4, #32 ;construct src_ptr[2] - vshr.u64 q5, q6, #32 - vext.8 d31, d22, d23, #1 ;construct src_ptr[-1] - - vmlsl.u8 q7, d18, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d20, d1 - vmlsl.u8 q12, d31, d1 ;-(src_ptr[-1] * vp8_filter[1]) - - vzip.32 d6, d7 ;put 2-line data in 1 register (src_ptr[2]) - vzip.32 d10, d11 - vshr.u64 q9, q4, #16 ;construct src_ptr[0] - vshr.u64 q10, q6, #16 - vext.8 d31, d22, d23, #4 ;construct src_ptr[2] - - vmlsl.u8 q7, d6, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d10, d4 - vmlsl.u8 q12, d31, d4 ;-(src_ptr[2] * vp8_filter[4]) - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[0]) - vzip.32 d20, d21 - vshr.u64 q3, q4, #24 ;construct src_ptr[1] - vshr.u64 q5, q6, #24 - vext.8 d31, d22, d23, #2 ;construct src_ptr[0] - - vmlal.u8 q7, d18, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d20, d2 - vmlal.u8 q12, d31, d2 ;(src_ptr[0] * vp8_filter[2]) - - vzip.32 d6, d7 ;put 2-line data in 1 register (src_ptr[1]) - vzip.32 d10, d11 - vext.8 d31, d22, d23, #3 ;construct src_ptr[1] - vmull.u8 q9, d6, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q10, d10, d3 - vmull.u8 q11, d31, d3 ;(src_ptr[1] * vp8_filter[3]) - - add r3, r12, r3, lsl #5 - - vqadd.s16 q7, q9 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q10 - vqadd.s16 q12, q11 - - vext.8 d23, d27, d28, #4 - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - - vqrshrun.s16 d29, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d30, q8, #7 - vqrshrun.s16 d31, q12, #7 - -;Second pass: 4x4 - vabs.s32 q7, q5 - vabs.s32 q8, q6 - - vext.8 d24, d28, d29, #4 - vext.8 d25, d29, d30, #4 - vext.8 d26, d30, d31, #4 - - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vdup.8 d1, d14[4] - vdup.8 d2, d15[0] - vdup.8 d3, d15[4] - vdup.8 d4, d16[0] - vdup.8 d5, d16[4] - - vmull.u8 q3, d27, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d28, d0 - - vmull.u8 q5, d25, d5 ;(src_ptr[3] * vp8_filter[5]) - vmull.u8 q6, d26, d5 - - vmlsl.u8 q3, d29, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d30, d4 - - vmlsl.u8 q5, d23, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q6, d24, d1 - - vmlal.u8 q3, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d29, d2 - - vmlal.u8 q5, d24, d3 ;(src_ptr[1] * vp8_filter[3]) - vmlal.u8 q6, d25, d3 - - add r0, r4, lr - add r1, r0, lr - add r2, r1, lr - - vqadd.s16 q5, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q6, q4 - - vqrshrun.s16 d3, q5, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d4, q6, #7 - - vst1.32 {d3[0]}, [r4] ;store result - vst1.32 {d3[1]}, [r0] - vst1.32 {d4[0]}, [r1] - vst1.32 {d4[1]}, [r2] - - pop {r4, pc} - - -;--------------------- -firstpass_filter4x4_only - vabs.s32 q12, q14 ;get abs(filer_parameters) - vabs.s32 q13, q15 - - sub r0, r0, #2 ;go back 2 columns of src data - -;First pass: output_height lines x output_width columns (4x4) - vld1.u8 {q3}, [r0], r1 ;load first 4-line src data - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vld1.u8 {q4}, [r0], r1 - vdup.8 d1, d24[4] - vld1.u8 {q5}, [r0], r1 - vdup.8 d2, d25[0] - vld1.u8 {q6}, [r0], r1 - - vdup.8 d3, d25[4] - vdup.8 d4, d26[0] - vdup.8 d5, d26[4] - - vext.8 d18, d6, d7, #5 ;construct src_ptr[3] - vext.8 d19, d8, d9, #5 - vext.8 d20, d10, d11, #5 - vext.8 d21, d12, d13, #5 - - vswp d7, d8 ;discard 2nd half data after src_ptr[3] is done - vswp d11, d12 - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[3]) - vzip.32 d20, d21 - vmull.u8 q7, d18, d5 ;(src_ptr[3] * vp8_filter[5]) - vmull.u8 q8, d20, d5 - - vmov q4, q3 ;keep original src data in q4 q6 - vmov q6, q5 - - vzip.32 d6, d7 ;construct src_ptr[-2], and put 2-line data together - vzip.32 d10, d11 - vshr.u64 q9, q4, #8 ;construct src_ptr[-1] - vshr.u64 q10, q6, #8 - vmlal.u8 q7, d6, d0 ;+(src_ptr[-2] * vp8_filter[0]) - vmlal.u8 q8, d10, d0 - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[-1]) - vzip.32 d20, d21 - vshr.u64 q3, q4, #32 ;construct src_ptr[2] - vshr.u64 q5, q6, #32 - vmlsl.u8 q7, d18, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d20, d1 - - vzip.32 d6, d7 ;put 2-line data in 1 register (src_ptr[2]) - vzip.32 d10, d11 - vshr.u64 q9, q4, #16 ;construct src_ptr[0] - vshr.u64 q10, q6, #16 - vmlsl.u8 q7, d6, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d10, d4 - - vzip.32 d18, d19 ;put 2-line data in 1 register (src_ptr[0]) - vzip.32 d20, d21 - vshr.u64 q3, q4, #24 ;construct src_ptr[1] - vshr.u64 q5, q6, #24 - vmlal.u8 q7, d18, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d20, d2 - - vzip.32 d6, d7 ;put 2-line data in 1 register (src_ptr[1]) - vzip.32 d10, d11 - vmull.u8 q9, d6, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q10, d10, d3 - - add r0, r4, lr - add r1, r0, lr - add r2, r1, lr - - vqadd.s16 q7, q9 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q10 - - vqrshrun.s16 d27, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d28, q8, #7 - - vst1.32 {d27[0]}, [r4] ;store result - vst1.32 {d27[1]}, [r0] - vst1.32 {d28[0]}, [r1] - vst1.32 {d28[1]}, [r2] - - pop {r4, pc} - - -;--------------------- -secondpass_filter4x4_only - sub r0, r0, r1, lsl #1 - add r3, r12, r3, lsl #5 - - vld1.32 {d27[0]}, [r0], r1 ;load src data - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - vld1.32 {d27[1]}, [r0], r1 - vabs.s32 q7, q5 - vld1.32 {d28[0]}, [r0], r1 - vabs.s32 q8, q6 - vld1.32 {d28[1]}, [r0], r1 - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vld1.32 {d29[0]}, [r0], r1 - vdup.8 d1, d14[4] - vld1.32 {d29[1]}, [r0], r1 - vdup.8 d2, d15[0] - vld1.32 {d30[0]}, [r0], r1 - vdup.8 d3, d15[4] - vld1.32 {d30[1]}, [r0], r1 - vdup.8 d4, d16[0] - vld1.32 {d31[0]}, [r0], r1 - vdup.8 d5, d16[4] - - vext.8 d23, d27, d28, #4 - vext.8 d24, d28, d29, #4 - vext.8 d25, d29, d30, #4 - vext.8 d26, d30, d31, #4 - - vmull.u8 q3, d27, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d28, d0 - - vmull.u8 q5, d25, d5 ;(src_ptr[3] * vp8_filter[5]) - vmull.u8 q6, d26, d5 - - vmlsl.u8 q3, d29, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d30, d4 - - vmlsl.u8 q5, d23, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q6, d24, d1 - - vmlal.u8 q3, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d29, d2 - - vmlal.u8 q5, d24, d3 ;(src_ptr[1] * vp8_filter[3]) - vmlal.u8 q6, d25, d3 - - add r0, r4, lr - add r1, r0, lr - add r2, r1, lr - - vqadd.s16 q5, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q6, q4 - - vqrshrun.s16 d3, q5, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d4, q6, #7 - - vst1.32 {d3[0]}, [r4] ;store result - vst1.32 {d3[1]}, [r0] - vst1.32 {d4[0]}, [r1] - vst1.32 {d4[1]}, [r2] - - pop {r4, pc} - - ENDP - -;----------------- - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict8x4_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict8x4_neon.asm deleted file mode 100644 index a57ec015f2c..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict8x4_neon.asm +++ /dev/null @@ -1,473 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_sixtap_predict8x4_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -filter8_coeff - DCD 0, 0, 128, 0, 0, 0, 0, 0 - DCD 0, -6, 123, 12, -1, 0, 0, 0 - DCD 2, -11, 108, 36, -8, 1, 0, 0 - DCD 0, -9, 93, 50, -6, 0, 0, 0 - DCD 3, -16, 77, 77, -16, 3, 0, 0 - DCD 0, -6, 50, 93, -9, 0, 0, 0 - DCD 1, -8, 36, 108, -11, 2, 0, 0 - DCD 0, -1, 12, 123, -6, 0, 0, 0 - -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; r4 unsigned char *dst_ptr, -; stack(r5) int dst_pitch - -|vp8_sixtap_predict8x4_neon| PROC - push {r4-r5, lr} - - adr r12, filter8_coeff - ldr r4, [sp, #12] ;load parameters from stack - ldr r5, [sp, #16] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq secondpass_filter8x4_only - - add r2, r12, r2, lsl #5 ;calculate filter location - - cmp r3, #0 ;skip second_pass filter if yoffset=0 - - vld1.s32 {q14, q15}, [r2] ;load first_pass filter - - beq firstpass_filter8x4_only - - sub sp, sp, #32 ;reserve space on stack for temporary storage - vabs.s32 q12, q14 - vabs.s32 q13, q15 - - sub r0, r0, #2 ;move srcptr back to (line-2) and (column-2) - mov lr, sp - sub r0, r0, r1, lsl #1 - - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vdup.8 d1, d24[4] - vdup.8 d2, d25[0] - -;First pass: output_height lines x output_width columns (9x8) - vld1.u8 {q3}, [r0], r1 ;load src data - vdup.8 d3, d25[4] - vld1.u8 {q4}, [r0], r1 - vdup.8 d4, d26[0] - vld1.u8 {q5}, [r0], r1 - vdup.8 d5, d26[4] - vld1.u8 {q6}, [r0], r1 - - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vmull.u8 q7, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q8, d8, d0 - vmull.u8 q9, d10, d0 - vmull.u8 q10, d12, d0 - - vext.8 d28, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d29, d8, d9, #1 - vext.8 d30, d10, d11, #1 - vext.8 d31, d12, d13, #1 - - vmlsl.u8 q7, d28, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d29, d1 - vmlsl.u8 q9, d30, d1 - vmlsl.u8 q10, d31, d1 - - vext.8 d28, d6, d7, #4 ;construct src_ptr[2] - vext.8 d29, d8, d9, #4 - vext.8 d30, d10, d11, #4 - vext.8 d31, d12, d13, #4 - - vmlsl.u8 q7, d28, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d29, d4 - vmlsl.u8 q9, d30, d4 - vmlsl.u8 q10, d31, d4 - - vext.8 d28, d6, d7, #2 ;construct src_ptr[0] - vext.8 d29, d8, d9, #2 - vext.8 d30, d10, d11, #2 - vext.8 d31, d12, d13, #2 - - vmlal.u8 q7, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d29, d2 - vmlal.u8 q9, d30, d2 - vmlal.u8 q10, d31, d2 - - vext.8 d28, d6, d7, #5 ;construct src_ptr[3] - vext.8 d29, d8, d9, #5 - vext.8 d30, d10, d11, #5 - vext.8 d31, d12, d13, #5 - - vmlal.u8 q7, d28, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q8, d29, d5 - vmlal.u8 q9, d30, d5 - vmlal.u8 q10, d31, d5 - - vext.8 d28, d6, d7, #3 ;construct src_ptr[1] - vext.8 d29, d8, d9, #3 - vext.8 d30, d10, d11, #3 - vext.8 d31, d12, d13, #3 - - vmull.u8 q3, d28, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q4, d29, d3 - vmull.u8 q5, d30, d3 - vmull.u8 q6, d31, d3 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vld1.u8 {q3}, [r0], r1 ;load src data - - vqrshrun.s16 d22, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d23, q8, #7 - vqrshrun.s16 d24, q9, #7 - vqrshrun.s16 d25, q10, #7 - - vld1.u8 {q4}, [r0], r1 - vst1.u8 {d22}, [lr]! ;store result - vld1.u8 {q5}, [r0], r1 - vst1.u8 {d23}, [lr]! - vld1.u8 {q6}, [r0], r1 - vst1.u8 {d24}, [lr]! - vld1.u8 {q7}, [r0], r1 - vst1.u8 {d25}, [lr]! - - ;first_pass filtering on the rest 5-line data - vmull.u8 q8, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q9, d8, d0 - vmull.u8 q10, d10, d0 - vmull.u8 q11, d12, d0 - vmull.u8 q12, d14, d0 - - vext.8 d27, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d28, d8, d9, #1 - vext.8 d29, d10, d11, #1 - vext.8 d30, d12, d13, #1 - vext.8 d31, d14, d15, #1 - - vmlsl.u8 q8, d27, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q9, d28, d1 - vmlsl.u8 q10, d29, d1 - vmlsl.u8 q11, d30, d1 - vmlsl.u8 q12, d31, d1 - - vext.8 d27, d6, d7, #4 ;construct src_ptr[2] - vext.8 d28, d8, d9, #4 - vext.8 d29, d10, d11, #4 - vext.8 d30, d12, d13, #4 - vext.8 d31, d14, d15, #4 - - vmlsl.u8 q8, d27, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q9, d28, d4 - vmlsl.u8 q10, d29, d4 - vmlsl.u8 q11, d30, d4 - vmlsl.u8 q12, d31, d4 - - vext.8 d27, d6, d7, #2 ;construct src_ptr[0] - vext.8 d28, d8, d9, #2 - vext.8 d29, d10, d11, #2 - vext.8 d30, d12, d13, #2 - vext.8 d31, d14, d15, #2 - - vmlal.u8 q8, d27, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q9, d28, d2 - vmlal.u8 q10, d29, d2 - vmlal.u8 q11, d30, d2 - vmlal.u8 q12, d31, d2 - - vext.8 d27, d6, d7, #5 ;construct src_ptr[3] - vext.8 d28, d8, d9, #5 - vext.8 d29, d10, d11, #5 - vext.8 d30, d12, d13, #5 - vext.8 d31, d14, d15, #5 - - vmlal.u8 q8, d27, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q9, d28, d5 - vmlal.u8 q10, d29, d5 - vmlal.u8 q11, d30, d5 - vmlal.u8 q12, d31, d5 - - vext.8 d27, d6, d7, #3 ;construct src_ptr[1] - vext.8 d28, d8, d9, #3 - vext.8 d29, d10, d11, #3 - vext.8 d30, d12, d13, #3 - vext.8 d31, d14, d15, #3 - - vmull.u8 q3, d27, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q4, d28, d3 - vmull.u8 q5, d29, d3 - vmull.u8 q6, d30, d3 - vmull.u8 q7, d31, d3 - - vqadd.s16 q8, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q9, q4 - vqadd.s16 q10, q5 - vqadd.s16 q11, q6 - vqadd.s16 q12, q7 - - vqrshrun.s16 d26, q8, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d27, q9, #7 - vqrshrun.s16 d28, q10, #7 - vqrshrun.s16 d29, q11, #7 ;load intermediate data from stack - vqrshrun.s16 d30, q12, #7 - -;Second pass: 8x4 -;secondpass_filter - add r3, r12, r3, lsl #5 - sub lr, lr, #32 - - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - vld1.u8 {q11}, [lr]! - - vabs.s32 q7, q5 - vabs.s32 q8, q6 - - vld1.u8 {q12}, [lr]! - - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vdup.8 d1, d14[4] - vdup.8 d2, d15[0] - vdup.8 d3, d15[4] - vdup.8 d4, d16[0] - vdup.8 d5, d16[4] - - vmull.u8 q3, d22, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d23, d0 - vmull.u8 q5, d24, d0 - vmull.u8 q6, d25, d0 - - vmlsl.u8 q3, d23, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q4, d24, d1 - vmlsl.u8 q5, d25, d1 - vmlsl.u8 q6, d26, d1 - - vmlsl.u8 q3, d26, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d27, d4 - vmlsl.u8 q5, d28, d4 - vmlsl.u8 q6, d29, d4 - - vmlal.u8 q3, d24, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d25, d2 - vmlal.u8 q5, d26, d2 - vmlal.u8 q6, d27, d2 - - vmlal.u8 q3, d27, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q4, d28, d5 - vmlal.u8 q5, d29, d5 - vmlal.u8 q6, d30, d5 - - vmull.u8 q7, d25, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q8, d26, d3 - vmull.u8 q9, d27, d3 - vmull.u8 q10, d28, d3 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q8, #7 - vqrshrun.s16 d8, q9, #7 - vqrshrun.s16 d9, q10, #7 - - vst1.u8 {d6}, [r4], r5 ;store result - vst1.u8 {d7}, [r4], r5 - vst1.u8 {d8}, [r4], r5 - vst1.u8 {d9}, [r4], r5 - - add sp, sp, #32 - pop {r4-r5,pc} - -;-------------------- -firstpass_filter8x4_only - vabs.s32 q12, q14 - vabs.s32 q13, q15 - - sub r0, r0, #2 ;move srcptr back to (line-2) and (column-2) - vld1.u8 {q3}, [r0], r1 ;load src data - - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vld1.u8 {q4}, [r0], r1 - vdup.8 d1, d24[4] - vld1.u8 {q5}, [r0], r1 - vdup.8 d2, d25[0] - vld1.u8 {q6}, [r0], r1 - vdup.8 d3, d25[4] - vdup.8 d4, d26[0] - vdup.8 d5, d26[4] - -;First pass: output_height lines x output_width columns (4x8) - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vmull.u8 q7, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q8, d8, d0 - vmull.u8 q9, d10, d0 - vmull.u8 q10, d12, d0 - - vext.8 d28, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d29, d8, d9, #1 - vext.8 d30, d10, d11, #1 - vext.8 d31, d12, d13, #1 - - vmlsl.u8 q7, d28, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d29, d1 - vmlsl.u8 q9, d30, d1 - vmlsl.u8 q10, d31, d1 - - vext.8 d28, d6, d7, #4 ;construct src_ptr[2] - vext.8 d29, d8, d9, #4 - vext.8 d30, d10, d11, #4 - vext.8 d31, d12, d13, #4 - - vmlsl.u8 q7, d28, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d29, d4 - vmlsl.u8 q9, d30, d4 - vmlsl.u8 q10, d31, d4 - - vext.8 d28, d6, d7, #2 ;construct src_ptr[0] - vext.8 d29, d8, d9, #2 - vext.8 d30, d10, d11, #2 - vext.8 d31, d12, d13, #2 - - vmlal.u8 q7, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d29, d2 - vmlal.u8 q9, d30, d2 - vmlal.u8 q10, d31, d2 - - vext.8 d28, d6, d7, #5 ;construct src_ptr[3] - vext.8 d29, d8, d9, #5 - vext.8 d30, d10, d11, #5 - vext.8 d31, d12, d13, #5 - - vmlal.u8 q7, d28, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q8, d29, d5 - vmlal.u8 q9, d30, d5 - vmlal.u8 q10, d31, d5 - - vext.8 d28, d6, d7, #3 ;construct src_ptr[1] - vext.8 d29, d8, d9, #3 - vext.8 d30, d10, d11, #3 - vext.8 d31, d12, d13, #3 - - vmull.u8 q3, d28, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q4, d29, d3 - vmull.u8 q5, d30, d3 - vmull.u8 q6, d31, d3 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vqrshrun.s16 d22, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d23, q8, #7 - vqrshrun.s16 d24, q9, #7 - vqrshrun.s16 d25, q10, #7 - - vst1.u8 {d22}, [r4], r5 ;store result - vst1.u8 {d23}, [r4], r5 - vst1.u8 {d24}, [r4], r5 - vst1.u8 {d25}, [r4], r5 - - pop {r4-r5,pc} - -;--------------------- -secondpass_filter8x4_only -;Second pass: 8x4 - add r3, r12, r3, lsl #5 - sub r0, r0, r1, lsl #1 - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - vabs.s32 q7, q5 - vabs.s32 q8, q6 - - vld1.u8 {d22}, [r0], r1 - vld1.u8 {d23}, [r0], r1 - vld1.u8 {d24}, [r0], r1 - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vld1.u8 {d25}, [r0], r1 - vdup.8 d1, d14[4] - vld1.u8 {d26}, [r0], r1 - vdup.8 d2, d15[0] - vld1.u8 {d27}, [r0], r1 - vdup.8 d3, d15[4] - vld1.u8 {d28}, [r0], r1 - vdup.8 d4, d16[0] - vld1.u8 {d29}, [r0], r1 - vdup.8 d5, d16[4] - vld1.u8 {d30}, [r0], r1 - - vmull.u8 q3, d22, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d23, d0 - vmull.u8 q5, d24, d0 - vmull.u8 q6, d25, d0 - - vmlsl.u8 q3, d23, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q4, d24, d1 - vmlsl.u8 q5, d25, d1 - vmlsl.u8 q6, d26, d1 - - vmlsl.u8 q3, d26, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d27, d4 - vmlsl.u8 q5, d28, d4 - vmlsl.u8 q6, d29, d4 - - vmlal.u8 q3, d24, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d25, d2 - vmlal.u8 q5, d26, d2 - vmlal.u8 q6, d27, d2 - - vmlal.u8 q3, d27, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q4, d28, d5 - vmlal.u8 q5, d29, d5 - vmlal.u8 q6, d30, d5 - - vmull.u8 q7, d25, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q8, d26, d3 - vmull.u8 q9, d27, d3 - vmull.u8 q10, d28, d3 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q8, #7 - vqrshrun.s16 d8, q9, #7 - vqrshrun.s16 d9, q10, #7 - - vst1.u8 {d6}, [r4], r5 ;store result - vst1.u8 {d7}, [r4], r5 - vst1.u8 {d8}, [r4], r5 - vst1.u8 {d9}, [r4], r5 - - pop {r4-r5,pc} - - ENDP - -;----------------- - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict8x8_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict8x8_neon.asm deleted file mode 100644 index 00ed5aeefe3..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict8x8_neon.asm +++ /dev/null @@ -1,524 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_sixtap_predict8x8_neon| - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -filter8_coeff - DCD 0, 0, 128, 0, 0, 0, 0, 0 - DCD 0, -6, 123, 12, -1, 0, 0, 0 - DCD 2, -11, 108, 36, -8, 1, 0, 0 - DCD 0, -9, 93, 50, -6, 0, 0, 0 - DCD 3, -16, 77, 77, -16, 3, 0, 0 - DCD 0, -6, 50, 93, -9, 0, 0, 0 - DCD 1, -8, 36, 108, -11, 2, 0, 0 - DCD 0, -1, 12, 123, -6, 0, 0, 0 - -; r0 unsigned char *src_ptr, -; r1 int src_pixels_per_line, -; r2 int xoffset, -; r3 int yoffset, -; stack(r4) unsigned char *dst_ptr, -; stack(r5) int dst_pitch - -|vp8_sixtap_predict8x8_neon| PROC - push {r4-r5, lr} - - adr r12, filter8_coeff - - ldr r4, [sp, #12] ;load parameters from stack - ldr r5, [sp, #16] ;load parameters from stack - - cmp r2, #0 ;skip first_pass filter if xoffset=0 - beq secondpass_filter8x8_only - - add r2, r12, r2, lsl #5 ;calculate filter location - - cmp r3, #0 ;skip second_pass filter if yoffset=0 - - vld1.s32 {q14, q15}, [r2] ;load first_pass filter - - beq firstpass_filter8x8_only - - sub sp, sp, #64 ;reserve space on stack for temporary storage - mov lr, sp - - vabs.s32 q12, q14 - vabs.s32 q13, q15 - - mov r2, #2 ;loop counter - sub r0, r0, #2 ;move srcptr back to (line-2) and (column-2) - sub r0, r0, r1, lsl #1 - - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vdup.8 d1, d24[4] - vdup.8 d2, d25[0] - -;First pass: output_height lines x output_width columns (13x8) - vld1.u8 {q3}, [r0], r1 ;load src data - vdup.8 d3, d25[4] - vld1.u8 {q4}, [r0], r1 - vdup.8 d4, d26[0] - vld1.u8 {q5}, [r0], r1 - vdup.8 d5, d26[4] - vld1.u8 {q6}, [r0], r1 - -filt_blk2d_fp8x8_loop_neon - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vmull.u8 q7, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q8, d8, d0 - vmull.u8 q9, d10, d0 - vmull.u8 q10, d12, d0 - - vext.8 d28, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d29, d8, d9, #1 - vext.8 d30, d10, d11, #1 - vext.8 d31, d12, d13, #1 - - vmlsl.u8 q7, d28, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d29, d1 - vmlsl.u8 q9, d30, d1 - vmlsl.u8 q10, d31, d1 - - vext.8 d28, d6, d7, #4 ;construct src_ptr[2] - vext.8 d29, d8, d9, #4 - vext.8 d30, d10, d11, #4 - vext.8 d31, d12, d13, #4 - - vmlsl.u8 q7, d28, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d29, d4 - vmlsl.u8 q9, d30, d4 - vmlsl.u8 q10, d31, d4 - - vext.8 d28, d6, d7, #2 ;construct src_ptr[0] - vext.8 d29, d8, d9, #2 - vext.8 d30, d10, d11, #2 - vext.8 d31, d12, d13, #2 - - vmlal.u8 q7, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d29, d2 - vmlal.u8 q9, d30, d2 - vmlal.u8 q10, d31, d2 - - vext.8 d28, d6, d7, #5 ;construct src_ptr[3] - vext.8 d29, d8, d9, #5 - vext.8 d30, d10, d11, #5 - vext.8 d31, d12, d13, #5 - - vmlal.u8 q7, d28, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q8, d29, d5 - vmlal.u8 q9, d30, d5 - vmlal.u8 q10, d31, d5 - - vext.8 d28, d6, d7, #3 ;construct src_ptr[1] - vext.8 d29, d8, d9, #3 - vext.8 d30, d10, d11, #3 - vext.8 d31, d12, d13, #3 - - vmull.u8 q3, d28, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q4, d29, d3 - vmull.u8 q5, d30, d3 - vmull.u8 q6, d31, d3 - - subs r2, r2, #1 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vld1.u8 {q3}, [r0], r1 ;load src data - - vqrshrun.s16 d22, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d23, q8, #7 - vqrshrun.s16 d24, q9, #7 - vqrshrun.s16 d25, q10, #7 - - vst1.u8 {d22}, [lr]! ;store result - vld1.u8 {q4}, [r0], r1 - vst1.u8 {d23}, [lr]! - vld1.u8 {q5}, [r0], r1 - vst1.u8 {d24}, [lr]! - vld1.u8 {q6}, [r0], r1 - vst1.u8 {d25}, [lr]! - - bne filt_blk2d_fp8x8_loop_neon - - ;first_pass filtering on the rest 5-line data - ;vld1.u8 {q3}, [r0], r1 ;load src data - ;vld1.u8 {q4}, [r0], r1 - ;vld1.u8 {q5}, [r0], r1 - ;vld1.u8 {q6}, [r0], r1 - vld1.u8 {q7}, [r0], r1 - - vmull.u8 q8, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q9, d8, d0 - vmull.u8 q10, d10, d0 - vmull.u8 q11, d12, d0 - vmull.u8 q12, d14, d0 - - vext.8 d27, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d28, d8, d9, #1 - vext.8 d29, d10, d11, #1 - vext.8 d30, d12, d13, #1 - vext.8 d31, d14, d15, #1 - - vmlsl.u8 q8, d27, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q9, d28, d1 - vmlsl.u8 q10, d29, d1 - vmlsl.u8 q11, d30, d1 - vmlsl.u8 q12, d31, d1 - - vext.8 d27, d6, d7, #4 ;construct src_ptr[2] - vext.8 d28, d8, d9, #4 - vext.8 d29, d10, d11, #4 - vext.8 d30, d12, d13, #4 - vext.8 d31, d14, d15, #4 - - vmlsl.u8 q8, d27, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q9, d28, d4 - vmlsl.u8 q10, d29, d4 - vmlsl.u8 q11, d30, d4 - vmlsl.u8 q12, d31, d4 - - vext.8 d27, d6, d7, #2 ;construct src_ptr[0] - vext.8 d28, d8, d9, #2 - vext.8 d29, d10, d11, #2 - vext.8 d30, d12, d13, #2 - vext.8 d31, d14, d15, #2 - - vmlal.u8 q8, d27, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q9, d28, d2 - vmlal.u8 q10, d29, d2 - vmlal.u8 q11, d30, d2 - vmlal.u8 q12, d31, d2 - - vext.8 d27, d6, d7, #5 ;construct src_ptr[3] - vext.8 d28, d8, d9, #5 - vext.8 d29, d10, d11, #5 - vext.8 d30, d12, d13, #5 - vext.8 d31, d14, d15, #5 - - vmlal.u8 q8, d27, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q9, d28, d5 - vmlal.u8 q10, d29, d5 - vmlal.u8 q11, d30, d5 - vmlal.u8 q12, d31, d5 - - vext.8 d27, d6, d7, #3 ;construct src_ptr[1] - vext.8 d28, d8, d9, #3 - vext.8 d29, d10, d11, #3 - vext.8 d30, d12, d13, #3 - vext.8 d31, d14, d15, #3 - - vmull.u8 q3, d27, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q4, d28, d3 - vmull.u8 q5, d29, d3 - vmull.u8 q6, d30, d3 - vmull.u8 q7, d31, d3 - - vqadd.s16 q8, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q9, q4 - vqadd.s16 q10, q5 - vqadd.s16 q11, q6 - vqadd.s16 q12, q7 - - add r3, r12, r3, lsl #5 - - vqrshrun.s16 d26, q8, #7 ;shift/round/saturate to u8 - sub lr, lr, #64 - vqrshrun.s16 d27, q9, #7 - vld1.u8 {q9}, [lr]! ;load intermediate data from stack - vqrshrun.s16 d28, q10, #7 - vld1.u8 {q10}, [lr]! - - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - - vqrshrun.s16 d29, q11, #7 - vld1.u8 {q11}, [lr]! - - vabs.s32 q7, q5 - vabs.s32 q8, q6 - - vqrshrun.s16 d30, q12, #7 - vld1.u8 {q12}, [lr]! - -;Second pass: 8x8 - mov r3, #2 ;loop counter - - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vdup.8 d1, d14[4] - vdup.8 d2, d15[0] - vdup.8 d3, d15[4] - vdup.8 d4, d16[0] - vdup.8 d5, d16[4] - -filt_blk2d_sp8x8_loop_neon - vmull.u8 q3, d18, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d19, d0 - vmull.u8 q5, d20, d0 - vmull.u8 q6, d21, d0 - - vmlsl.u8 q3, d19, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q4, d20, d1 - vmlsl.u8 q5, d21, d1 - vmlsl.u8 q6, d22, d1 - - vmlsl.u8 q3, d22, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d23, d4 - vmlsl.u8 q5, d24, d4 - vmlsl.u8 q6, d25, d4 - - vmlal.u8 q3, d20, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d21, d2 - vmlal.u8 q5, d22, d2 - vmlal.u8 q6, d23, d2 - - vmlal.u8 q3, d23, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q4, d24, d5 - vmlal.u8 q5, d25, d5 - vmlal.u8 q6, d26, d5 - - vmull.u8 q7, d21, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q8, d22, d3 - vmull.u8 q9, d23, d3 - vmull.u8 q10, d24, d3 - - subs r3, r3, #1 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q8, #7 - vqrshrun.s16 d8, q9, #7 - vqrshrun.s16 d9, q10, #7 - - vmov q9, q11 - vst1.u8 {d6}, [r4], r5 ;store result - vmov q10, q12 - vst1.u8 {d7}, [r4], r5 - vmov q11, q13 - vst1.u8 {d8}, [r4], r5 - vmov q12, q14 - vst1.u8 {d9}, [r4], r5 - vmov d26, d30 - - bne filt_blk2d_sp8x8_loop_neon - - add sp, sp, #64 - pop {r4-r5,pc} - -;--------------------- -firstpass_filter8x8_only - ;add r2, r12, r2, lsl #5 ;calculate filter location - ;vld1.s32 {q14, q15}, [r2] ;load first_pass filter - vabs.s32 q12, q14 - vabs.s32 q13, q15 - - mov r2, #2 ;loop counter - sub r0, r0, #2 ;move srcptr back to (line-2) and (column-2) - - vdup.8 d0, d24[0] ;first_pass filter (d0-d5) - vdup.8 d1, d24[4] - vdup.8 d2, d25[0] - vdup.8 d3, d25[4] - vdup.8 d4, d26[0] - vdup.8 d5, d26[4] - -;First pass: output_height lines x output_width columns (8x8) -filt_blk2d_fpo8x8_loop_neon - vld1.u8 {q3}, [r0], r1 ;load src data - vld1.u8 {q4}, [r0], r1 - vld1.u8 {q5}, [r0], r1 - vld1.u8 {q6}, [r0], r1 - - pld [r0] - pld [r0, r1] - pld [r0, r1, lsl #1] - - vmull.u8 q7, d6, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q8, d8, d0 - vmull.u8 q9, d10, d0 - vmull.u8 q10, d12, d0 - - vext.8 d28, d6, d7, #1 ;construct src_ptr[-1] - vext.8 d29, d8, d9, #1 - vext.8 d30, d10, d11, #1 - vext.8 d31, d12, d13, #1 - - vmlsl.u8 q7, d28, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q8, d29, d1 - vmlsl.u8 q9, d30, d1 - vmlsl.u8 q10, d31, d1 - - vext.8 d28, d6, d7, #4 ;construct src_ptr[2] - vext.8 d29, d8, d9, #4 - vext.8 d30, d10, d11, #4 - vext.8 d31, d12, d13, #4 - - vmlsl.u8 q7, d28, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q8, d29, d4 - vmlsl.u8 q9, d30, d4 - vmlsl.u8 q10, d31, d4 - - vext.8 d28, d6, d7, #2 ;construct src_ptr[0] - vext.8 d29, d8, d9, #2 - vext.8 d30, d10, d11, #2 - vext.8 d31, d12, d13, #2 - - vmlal.u8 q7, d28, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q8, d29, d2 - vmlal.u8 q9, d30, d2 - vmlal.u8 q10, d31, d2 - - vext.8 d28, d6, d7, #5 ;construct src_ptr[3] - vext.8 d29, d8, d9, #5 - vext.8 d30, d10, d11, #5 - vext.8 d31, d12, d13, #5 - - vmlal.u8 q7, d28, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q8, d29, d5 - vmlal.u8 q9, d30, d5 - vmlal.u8 q10, d31, d5 - - vext.8 d28, d6, d7, #3 ;construct src_ptr[1] - vext.8 d29, d8, d9, #3 - vext.8 d30, d10, d11, #3 - vext.8 d31, d12, d13, #3 - - vmull.u8 q3, d28, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q4, d29, d3 - vmull.u8 q5, d30, d3 - vmull.u8 q6, d31, d3 - ; - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - subs r2, r2, #1 - - vqrshrun.s16 d22, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d23, q8, #7 - vqrshrun.s16 d24, q9, #7 - vqrshrun.s16 d25, q10, #7 - - vst1.u8 {d22}, [r4], r5 ;store result - vst1.u8 {d23}, [r4], r5 - vst1.u8 {d24}, [r4], r5 - vst1.u8 {d25}, [r4], r5 - - bne filt_blk2d_fpo8x8_loop_neon - - pop {r4-r5,pc} - -;--------------------- -secondpass_filter8x8_only - sub r0, r0, r1, lsl #1 - add r3, r12, r3, lsl #5 - - vld1.u8 {d18}, [r0], r1 ;load src data - vld1.s32 {q5, q6}, [r3] ;load second_pass filter - vld1.u8 {d19}, [r0], r1 - vabs.s32 q7, q5 - vld1.u8 {d20}, [r0], r1 - vabs.s32 q8, q6 - vld1.u8 {d21}, [r0], r1 - mov r3, #2 ;loop counter - vld1.u8 {d22}, [r0], r1 - vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5) - vld1.u8 {d23}, [r0], r1 - vdup.8 d1, d14[4] - vld1.u8 {d24}, [r0], r1 - vdup.8 d2, d15[0] - vld1.u8 {d25}, [r0], r1 - vdup.8 d3, d15[4] - vld1.u8 {d26}, [r0], r1 - vdup.8 d4, d16[0] - vld1.u8 {d27}, [r0], r1 - vdup.8 d5, d16[4] - vld1.u8 {d28}, [r0], r1 - vld1.u8 {d29}, [r0], r1 - vld1.u8 {d30}, [r0], r1 - -;Second pass: 8x8 -filt_blk2d_spo8x8_loop_neon - vmull.u8 q3, d18, d0 ;(src_ptr[-2] * vp8_filter[0]) - vmull.u8 q4, d19, d0 - vmull.u8 q5, d20, d0 - vmull.u8 q6, d21, d0 - - vmlsl.u8 q3, d19, d1 ;-(src_ptr[-1] * vp8_filter[1]) - vmlsl.u8 q4, d20, d1 - vmlsl.u8 q5, d21, d1 - vmlsl.u8 q6, d22, d1 - - vmlsl.u8 q3, d22, d4 ;-(src_ptr[2] * vp8_filter[4]) - vmlsl.u8 q4, d23, d4 - vmlsl.u8 q5, d24, d4 - vmlsl.u8 q6, d25, d4 - - vmlal.u8 q3, d20, d2 ;(src_ptr[0] * vp8_filter[2]) - vmlal.u8 q4, d21, d2 - vmlal.u8 q5, d22, d2 - vmlal.u8 q6, d23, d2 - - vmlal.u8 q3, d23, d5 ;(src_ptr[3] * vp8_filter[5]) - vmlal.u8 q4, d24, d5 - vmlal.u8 q5, d25, d5 - vmlal.u8 q6, d26, d5 - - vmull.u8 q7, d21, d3 ;(src_ptr[1] * vp8_filter[3]) - vmull.u8 q8, d22, d3 - vmull.u8 q9, d23, d3 - vmull.u8 q10, d24, d3 - - subs r3, r3, #1 - - vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters) - vqadd.s16 q8, q4 - vqadd.s16 q9, q5 - vqadd.s16 q10, q6 - - vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8 - vqrshrun.s16 d7, q8, #7 - vqrshrun.s16 d8, q9, #7 - vqrshrun.s16 d9, q10, #7 - - vmov q9, q11 - vst1.u8 {d6}, [r4], r5 ;store result - vmov q10, q12 - vst1.u8 {d7}, [r4], r5 - vmov q11, q13 - vst1.u8 {d8}, [r4], r5 - vmov q12, q14 - vst1.u8 {d9}, [r4], r5 - vmov d26, d30 - - bne filt_blk2d_spo8x8_loop_neon - - pop {r4-r5,pc} - - ENDP - -;----------------- - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict_neon.c new file mode 100644 index 00000000000..7a4d9e05128 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/sixtappredict_neon.c @@ -0,0 +1,1752 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +#ifdef _MSC_VER +#define __builtin_prefetch(x) +#endif + +static const int8_t vp8_sub_pel_filters[8][8] = { + {0, 0, 128, 0, 0, 0, 0, 0}, /* note that 1/8 pel positionyys are */ + {0, -6, 123, 12, -1, 0, 0, 0}, /* just as per alpha -0.5 bicubic */ + {2, -11, 108, 36, -8, 1, 0, 0}, /* New 1/4 pel 6 tap filter */ + {0, -9, 93, 50, -6, 0, 0, 0}, + {3, -16, 77, 77, -16, 3, 0, 0}, /* New 1/2 pel 6 tap filter */ + {0, -6, 50, 93, -9, 0, 0, 0}, + {1, -8, 36, 108, -11, 2, 0, 0}, /* New 1/4 pel 6 tap filter */ + {0, -1, 12, 123, -6, 0, 0, 0}, +}; + +void vp8_sixtap_predict4x4_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + unsigned char *src; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d18u8, d19u8, d20u8, d21u8; + uint8x8_t d23u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8, d30u8, d31u8; + int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8; + uint32x2_t d27u32, d28u32, d29u32, d30u32, d31u32; + uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16; + uint16x8_t q8u16, q9u16, q10u16, q11u16, q12u16; + int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16; + uint8x16_t q3u8, q4u8, q5u8, q6u8, q11u8; + uint64x2_t q3u64, q4u64, q5u64, q6u64, q9u64, q10u64; + uint32x2x2_t d0u32x2, d1u32x2; + + if (xoffset == 0) { // secondpass_filter4x4_only + // load second_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // load src data + src = src_ptr - src_pixels_per_line * 2; + d27u32 = vld1_lane_u32((const uint32_t *)src, d27u32, 0); + src += src_pixels_per_line; + d27u32 = vld1_lane_u32((const uint32_t *)src, d27u32, 1); + src += src_pixels_per_line; + d28u32 = vld1_lane_u32((const uint32_t *)src, d28u32, 0); + src += src_pixels_per_line; + d28u32 = vld1_lane_u32((const uint32_t *)src, d28u32, 1); + src += src_pixels_per_line; + d29u32 = vld1_lane_u32((const uint32_t *)src, d29u32, 0); + src += src_pixels_per_line; + d29u32 = vld1_lane_u32((const uint32_t *)src, d29u32, 1); + src += src_pixels_per_line; + d30u32 = vld1_lane_u32((const uint32_t *)src, d30u32, 0); + src += src_pixels_per_line; + d30u32 = vld1_lane_u32((const uint32_t *)src, d30u32, 1); + src += src_pixels_per_line; + d31u32 = vld1_lane_u32((const uint32_t *)src, d31u32, 0); + + d27u8 = vreinterpret_u8_u32(d27u32); + d28u8 = vreinterpret_u8_u32(d28u32); + d29u8 = vreinterpret_u8_u32(d29u32); + d30u8 = vreinterpret_u8_u32(d30u32); + d31u8 = vreinterpret_u8_u32(d31u32); + + d23u8 = vext_u8(d27u8, d28u8, 4); + d24u8 = vext_u8(d28u8, d29u8, 4); + d25u8 = vext_u8(d29u8, d30u8, 4); + d26u8 = vext_u8(d30u8, d31u8, 4); + + q3u16 = vmull_u8(d27u8, d0u8); + q4u16 = vmull_u8(d28u8, d0u8); + q5u16 = vmull_u8(d25u8, d5u8); + q6u16 = vmull_u8(d26u8, d5u8); + + q3u16 = vmlsl_u8(q3u16, d29u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d30u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d23u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d24u8, d1u8); + + q3u16 = vmlal_u8(q3u16, d28u8, d2u8); + q4u16 = vmlal_u8(q4u16, d29u8, d2u8); + q5u16 = vmlal_u8(q5u16, d24u8, d3u8); + q6u16 = vmlal_u8(q6u16, d25u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + + q5s16 = vqaddq_s16(q5s16, q3s16); + q6s16 = vqaddq_s16(q6s16, q4s16); + + d3u8 = vqrshrun_n_s16(q5s16, 7); + d4u8 = vqrshrun_n_s16(q6s16, 7); + + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 1); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d4u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d4u8), 1); + return; + } + + // load first_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // First pass: output_height lines x output_width columns (9x4) + + if (yoffset == 0) // firstpass_filter4x4_only + src = src_ptr - 2; + else + src = src_ptr - 2 - (src_pixels_per_line * 2); + + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + + d18u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d19u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d20u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d21u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + + // vswp here + q3u8 = vcombine_u8(vget_low_u8(q3u8), vget_low_u8(q4u8)); + q5u8 = vcombine_u8(vget_low_u8(q5u8), vget_low_u8(q6u8)); + + d0u32x2 = vzip_u32(vreinterpret_u32_u8(d18u8), // d18 d19 + vreinterpret_u32_u8(d19u8)); + d1u32x2 = vzip_u32(vreinterpret_u32_u8(d20u8), // d20 d21 + vreinterpret_u32_u8(d21u8)); + q7u16 = vmull_u8(vreinterpret_u8_u32(d0u32x2.val[0]), d5u8); + q8u16 = vmull_u8(vreinterpret_u8_u32(d1u32x2.val[0]), d5u8); + + // keep original src data in q4 q6 + q4u64 = vreinterpretq_u64_u8(q3u8); + q6u64 = vreinterpretq_u64_u8(q5u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q3u8)), // d6 d7 + vreinterpret_u32_u8(vget_high_u8(q3u8))); + d1u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q5u8)), // d10 d11 + vreinterpret_u32_u8(vget_high_u8(q5u8))); + q9u64 = vshrq_n_u64(q4u64, 8); + q10u64 = vshrq_n_u64(q6u64, 8); + q7u16 = vmlal_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d0u8); + q8u16 = vmlal_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d0u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q9u64)), // d18 d19 + vreinterpret_u32_u64(vget_high_u64(q9u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q10u64)), // d20 d211 + vreinterpret_u32_u64(vget_high_u64(q10u64))); + q3u64 = vshrq_n_u64(q4u64, 32); + q5u64 = vshrq_n_u64(q6u64, 32); + q7u16 = vmlsl_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d1u8); + q8u16 = vmlsl_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d1u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q3u64)), // d6 d7 + vreinterpret_u32_u64(vget_high_u64(q3u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q5u64)), // d10 d11 + vreinterpret_u32_u64(vget_high_u64(q5u64))); + q9u64 = vshrq_n_u64(q4u64, 16); + q10u64 = vshrq_n_u64(q6u64, 16); + q7u16 = vmlsl_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d4u8); + q8u16 = vmlsl_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d4u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q9u64)), // d18 d19 + vreinterpret_u32_u64(vget_high_u64(q9u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q10u64)), // d20 d211 + vreinterpret_u32_u64(vget_high_u64(q10u64))); + q3u64 = vshrq_n_u64(q4u64, 24); + q5u64 = vshrq_n_u64(q6u64, 24); + q7u16 = vmlal_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d2u8); + q8u16 = vmlal_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d2u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q3u64)), // d6 d7 + vreinterpret_u32_u64(vget_high_u64(q3u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q5u64)), // d10 d11 + vreinterpret_u32_u64(vget_high_u64(q5u64))); + q9u16 = vmull_u8(vreinterpret_u8_u32(d0u32x2.val[0]), d3u8); + q10u16 = vmull_u8(vreinterpret_u8_u32(d1u32x2.val[0]), d3u8); + + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q7s16 = vqaddq_s16(q7s16, q9s16); + q8s16 = vqaddq_s16(q8s16, q10s16); + + d27u8 = vqrshrun_n_s16(q7s16, 7); + d28u8 = vqrshrun_n_s16(q8s16, 7); + + if (yoffset == 0) { // firstpass_filter4x4_only + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d27u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d27u8), 1); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d28u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d28u8), 1); + return; + } + + // First Pass on rest 5-line data + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + q11u8 = vld1q_u8(src); + + d18u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d19u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d20u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d21u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + + // vswp here + q3u8 = vcombine_u8(vget_low_u8(q3u8), vget_low_u8(q4u8)); + q5u8 = vcombine_u8(vget_low_u8(q5u8), vget_low_u8(q6u8)); + + d0u32x2 = vzip_u32(vreinterpret_u32_u8(d18u8), // d18 d19 + vreinterpret_u32_u8(d19u8)); + d1u32x2 = vzip_u32(vreinterpret_u32_u8(d20u8), // d20 d21 + vreinterpret_u32_u8(d21u8)); + d31u8 = vext_u8(vget_low_u8(q11u8), vget_high_u8(q11u8), 5); + q7u16 = vmull_u8(vreinterpret_u8_u32(d0u32x2.val[0]), d5u8); + q8u16 = vmull_u8(vreinterpret_u8_u32(d1u32x2.val[0]), d5u8); + q12u16 = vmull_u8(d31u8, d5u8); + + q4u64 = vreinterpretq_u64_u8(q3u8); + q6u64 = vreinterpretq_u64_u8(q5u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q3u8)), // d6 d7 + vreinterpret_u32_u8(vget_high_u8(q3u8))); + d1u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q5u8)), // d10 d11 + vreinterpret_u32_u8(vget_high_u8(q5u8))); + q9u64 = vshrq_n_u64(q4u64, 8); + q10u64 = vshrq_n_u64(q6u64, 8); + q7u16 = vmlal_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d0u8); + q8u16 = vmlal_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d0u8); + q12u16 = vmlal_u8(q12u16, vget_low_u8(q11u8), d0u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q9u64)), // d18 d19 + vreinterpret_u32_u64(vget_high_u64(q9u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q10u64)), // d20 d211 + vreinterpret_u32_u64(vget_high_u64(q10u64))); + q3u64 = vshrq_n_u64(q4u64, 32); + q5u64 = vshrq_n_u64(q6u64, 32); + d31u8 = vext_u8(vget_low_u8(q11u8), vget_high_u8(q11u8), 1); + q7u16 = vmlsl_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d1u8); + q8u16 = vmlsl_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d1u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d1u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q3u64)), // d6 d7 + vreinterpret_u32_u64(vget_high_u64(q3u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q5u64)), // d10 d11 + vreinterpret_u32_u64(vget_high_u64(q5u64))); + q9u64 = vshrq_n_u64(q4u64, 16); + q10u64 = vshrq_n_u64(q6u64, 16); + d31u8 = vext_u8(vget_low_u8(q11u8), vget_high_u8(q11u8), 4); + q7u16 = vmlsl_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d4u8); + q8u16 = vmlsl_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d4u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d4u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q9u64)), // d18 d19 + vreinterpret_u32_u64(vget_high_u64(q9u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q10u64)), // d20 d211 + vreinterpret_u32_u64(vget_high_u64(q10u64))); + q3u64 = vshrq_n_u64(q4u64, 24); + q5u64 = vshrq_n_u64(q6u64, 24); + d31u8 = vext_u8(vget_low_u8(q11u8), vget_high_u8(q11u8), 2); + q7u16 = vmlal_u8(q7u16, vreinterpret_u8_u32(d0u32x2.val[0]), d2u8); + q8u16 = vmlal_u8(q8u16, vreinterpret_u8_u32(d1u32x2.val[0]), d2u8); + q12u16 = vmlal_u8(q12u16, d31u8, d2u8); + + d0u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q3u64)), // d6 d7 + vreinterpret_u32_u64(vget_high_u64(q3u64))); + d1u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q5u64)), // d10 d11 + vreinterpret_u32_u64(vget_high_u64(q5u64))); + d31u8 = vext_u8(vget_low_u8(q11u8), vget_high_u8(q11u8), 3); + q9u16 = vmull_u8(vreinterpret_u8_u32(d0u32x2.val[0]), d3u8); + q10u16 = vmull_u8(vreinterpret_u8_u32(d1u32x2.val[0]), d3u8); + q11u16 = vmull_u8(d31u8, d3u8); + + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + q7s16 = vqaddq_s16(q7s16, q9s16); + q8s16 = vqaddq_s16(q8s16, q10s16); + q12s16 = vqaddq_s16(q12s16, q11s16); + + d29u8 = vqrshrun_n_s16(q7s16, 7); + d30u8 = vqrshrun_n_s16(q8s16, 7); + d31u8 = vqrshrun_n_s16(q12s16, 7); + + // Second pass: 4x4 + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + d23u8 = vext_u8(d27u8, d28u8, 4); + d24u8 = vext_u8(d28u8, d29u8, 4); + d25u8 = vext_u8(d29u8, d30u8, 4); + d26u8 = vext_u8(d30u8, d31u8, 4); + + q3u16 = vmull_u8(d27u8, d0u8); + q4u16 = vmull_u8(d28u8, d0u8); + q5u16 = vmull_u8(d25u8, d5u8); + q6u16 = vmull_u8(d26u8, d5u8); + + q3u16 = vmlsl_u8(q3u16, d29u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d30u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d23u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d24u8, d1u8); + + q3u16 = vmlal_u8(q3u16, d28u8, d2u8); + q4u16 = vmlal_u8(q4u16, d29u8, d2u8); + q5u16 = vmlal_u8(q5u16, d24u8, d3u8); + q6u16 = vmlal_u8(q6u16, d25u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + + q5s16 = vqaddq_s16(q5s16, q3s16); + q6s16 = vqaddq_s16(q6s16, q4s16); + + d3u8 = vqrshrun_n_s16(q5s16, 7); + d4u8 = vqrshrun_n_s16(q6s16, 7); + + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 1); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d4u8), 0); + dst_ptr += dst_pitch; + vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d4u8), 1); + return; +} + +void vp8_sixtap_predict8x4_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + unsigned char *src; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d22u8, d23u8, d24u8, d25u8, d26u8; + uint8x8_t d27u8, d28u8, d29u8, d30u8, d31u8; + int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8; + uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16; + uint16x8_t q8u16, q9u16, q10u16, q11u16, q12u16; + int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16; + uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8; + + if (xoffset == 0) { // secondpass_filter8x4_only + // load second_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // load src data + src = src_ptr - src_pixels_per_line * 2; + d22u8 = vld1_u8(src); + src += src_pixels_per_line; + d23u8 = vld1_u8(src); + src += src_pixels_per_line; + d24u8 = vld1_u8(src); + src += src_pixels_per_line; + d25u8 = vld1_u8(src); + src += src_pixels_per_line; + d26u8 = vld1_u8(src); + src += src_pixels_per_line; + d27u8 = vld1_u8(src); + src += src_pixels_per_line; + d28u8 = vld1_u8(src); + src += src_pixels_per_line; + d29u8 = vld1_u8(src); + src += src_pixels_per_line; + d30u8 = vld1_u8(src); + + q3u16 = vmull_u8(d22u8, d0u8); + q4u16 = vmull_u8(d23u8, d0u8); + q5u16 = vmull_u8(d24u8, d0u8); + q6u16 = vmull_u8(d25u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d23u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d24u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d25u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d26u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d26u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d27u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d28u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d29u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d24u8, d2u8); + q4u16 = vmlal_u8(q4u16, d25u8, d2u8); + q5u16 = vmlal_u8(q5u16, d26u8, d2u8); + q6u16 = vmlal_u8(q6u16, d27u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d27u8, d5u8); + q4u16 = vmlal_u8(q4u16, d28u8, d5u8); + q5u16 = vmlal_u8(q5u16, d29u8, d5u8); + q6u16 = vmlal_u8(q6u16, d30u8, d5u8); + + q7u16 = vmull_u8(d25u8, d3u8); + q8u16 = vmull_u8(d26u8, d3u8); + q9u16 = vmull_u8(d27u8, d3u8); + q10u16 = vmull_u8(d28u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + return; + } + + // load first_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // First pass: output_height lines x output_width columns (9x4) + if (yoffset == 0) // firstpass_filter4x4_only + src = src_ptr - 2; + else + src = src_ptr - 2 - (src_pixels_per_line * 2); + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + + q7u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + + q7u16 = vmlsl_u8(q7u16, d28u8, d1u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d1u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + + q7u16 = vmlsl_u8(q7u16, d28u8, d4u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d4u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + + q7u16 = vmlal_u8(q7u16, d28u8, d2u8); + q8u16 = vmlal_u8(q8u16, d29u8, d2u8); + q9u16 = vmlal_u8(q9u16, d30u8, d2u8); + q10u16 = vmlal_u8(q10u16, d31u8, d2u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + + q7u16 = vmlal_u8(q7u16, d28u8, d5u8); + q8u16 = vmlal_u8(q8u16, d29u8, d5u8); + q9u16 = vmlal_u8(q9u16, d30u8, d5u8); + q10u16 = vmlal_u8(q10u16, d31u8, d5u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + + q3u16 = vmull_u8(d28u8, d3u8); + q4u16 = vmull_u8(d29u8, d3u8); + q5u16 = vmull_u8(d30u8, d3u8); + q6u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d22u8 = vqrshrun_n_s16(q7s16, 7); + d23u8 = vqrshrun_n_s16(q8s16, 7); + d24u8 = vqrshrun_n_s16(q9s16, 7); + d25u8 = vqrshrun_n_s16(q10s16, 7); + + if (yoffset == 0) { // firstpass_filter8x4_only + vst1_u8(dst_ptr, d22u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d23u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d24u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d25u8); + return; + } + + // First Pass on rest 5-line data + src += src_pixels_per_line; + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + q7u8 = vld1q_u8(src); + + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q11u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + q12u16 = vmull_u8(vget_low_u8(q7u8), d0u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 1); + + q8u16 = vmlsl_u8(q8u16, d27u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d1u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d1u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d1u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 4); + + q8u16 = vmlsl_u8(q8u16, d27u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d4u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d4u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d4u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 2); + + q8u16 = vmlal_u8(q8u16, d27u8, d2u8); + q9u16 = vmlal_u8(q9u16, d28u8, d2u8); + q10u16 = vmlal_u8(q10u16, d29u8, d2u8); + q11u16 = vmlal_u8(q11u16, d30u8, d2u8); + q12u16 = vmlal_u8(q12u16, d31u8, d2u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 5); + + q8u16 = vmlal_u8(q8u16, d27u8, d5u8); + q9u16 = vmlal_u8(q9u16, d28u8, d5u8); + q10u16 = vmlal_u8(q10u16, d29u8, d5u8); + q11u16 = vmlal_u8(q11u16, d30u8, d5u8); + q12u16 = vmlal_u8(q12u16, d31u8, d5u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 3); + + q3u16 = vmull_u8(d27u8, d3u8); + q4u16 = vmull_u8(d28u8, d3u8); + q5u16 = vmull_u8(d29u8, d3u8); + q6u16 = vmull_u8(d30u8, d3u8); + q7u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + + q8s16 = vqaddq_s16(q8s16, q3s16); + q9s16 = vqaddq_s16(q9s16, q4s16); + q10s16 = vqaddq_s16(q10s16, q5s16); + q11s16 = vqaddq_s16(q11s16, q6s16); + q12s16 = vqaddq_s16(q12s16, q7s16); + + d26u8 = vqrshrun_n_s16(q8s16, 7); + d27u8 = vqrshrun_n_s16(q9s16, 7); + d28u8 = vqrshrun_n_s16(q10s16, 7); + d29u8 = vqrshrun_n_s16(q11s16, 7); + d30u8 = vqrshrun_n_s16(q12s16, 7); + + // Second pass: 8x4 + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + q3u16 = vmull_u8(d22u8, d0u8); + q4u16 = vmull_u8(d23u8, d0u8); + q5u16 = vmull_u8(d24u8, d0u8); + q6u16 = vmull_u8(d25u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d23u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d24u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d25u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d26u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d26u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d27u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d28u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d29u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d24u8, d2u8); + q4u16 = vmlal_u8(q4u16, d25u8, d2u8); + q5u16 = vmlal_u8(q5u16, d26u8, d2u8); + q6u16 = vmlal_u8(q6u16, d27u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d27u8, d5u8); + q4u16 = vmlal_u8(q4u16, d28u8, d5u8); + q5u16 = vmlal_u8(q5u16, d29u8, d5u8); + q6u16 = vmlal_u8(q6u16, d30u8, d5u8); + + q7u16 = vmull_u8(d25u8, d3u8); + q8u16 = vmull_u8(d26u8, d3u8); + q9u16 = vmull_u8(d27u8, d3u8); + q10u16 = vmull_u8(d28u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + return; +} + +void vp8_sixtap_predict8x8_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + unsigned char *src, *tmpp; + unsigned char tmp[64]; + int i; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d18u8, d19u8, d20u8, d21u8, d22u8, d23u8, d24u8, d25u8; + uint8x8_t d26u8, d27u8, d28u8, d29u8, d30u8, d31u8; + int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8; + uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16; + uint16x8_t q8u16, q9u16, q10u16, q11u16, q12u16; + int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16; + int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16; + uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8, q9u8, q10u8, q11u8, q12u8; + + if (xoffset == 0) { // secondpass_filter8x8_only + // load second_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // load src data + src = src_ptr - src_pixels_per_line * 2; + d18u8 = vld1_u8(src); + src += src_pixels_per_line; + d19u8 = vld1_u8(src); + src += src_pixels_per_line; + d20u8 = vld1_u8(src); + src += src_pixels_per_line; + d21u8 = vld1_u8(src); + src += src_pixels_per_line; + d22u8 = vld1_u8(src); + src += src_pixels_per_line; + d23u8 = vld1_u8(src); + src += src_pixels_per_line; + d24u8 = vld1_u8(src); + src += src_pixels_per_line; + d25u8 = vld1_u8(src); + src += src_pixels_per_line; + d26u8 = vld1_u8(src); + src += src_pixels_per_line; + d27u8 = vld1_u8(src); + src += src_pixels_per_line; + d28u8 = vld1_u8(src); + src += src_pixels_per_line; + d29u8 = vld1_u8(src); + src += src_pixels_per_line; + d30u8 = vld1_u8(src); + + for (i = 2; i > 0; i--) { + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + d23u8 = d27u8; + d24u8 = d28u8; + d25u8 = d29u8; + d26u8 = d30u8; + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + dst_ptr += dst_pitch; + } + return; + } + + // load first_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // First pass: output_height lines x output_width columns (9x4) + if (yoffset == 0) // firstpass_filter4x4_only + src = src_ptr - 2; + else + src = src_ptr - 2 - (src_pixels_per_line * 2); + + tmpp = tmp; + for (i = 2; i > 0; i--) { + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + + __builtin_prefetch(src); + __builtin_prefetch(src + src_pixels_per_line); + __builtin_prefetch(src + src_pixels_per_line * 2); + + q7u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q8u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + + q7u16 = vmlsl_u8(q7u16, d28u8, d1u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d1u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + + q7u16 = vmlsl_u8(q7u16, d28u8, d4u8); + q8u16 = vmlsl_u8(q8u16, d29u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d30u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d31u8, d4u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + + q7u16 = vmlal_u8(q7u16, d28u8, d2u8); + q8u16 = vmlal_u8(q8u16, d29u8, d2u8); + q9u16 = vmlal_u8(q9u16, d30u8, d2u8); + q10u16 = vmlal_u8(q10u16, d31u8, d2u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + + q7u16 = vmlal_u8(q7u16, d28u8, d5u8); + q8u16 = vmlal_u8(q8u16, d29u8, d5u8); + q9u16 = vmlal_u8(q9u16, d30u8, d5u8); + q10u16 = vmlal_u8(q10u16, d31u8, d5u8); + + d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + + q3u16 = vmull_u8(d28u8, d3u8); + q4u16 = vmull_u8(d29u8, d3u8); + q5u16 = vmull_u8(d30u8, d3u8); + q6u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d22u8 = vqrshrun_n_s16(q7s16, 7); + d23u8 = vqrshrun_n_s16(q8s16, 7); + d24u8 = vqrshrun_n_s16(q9s16, 7); + d25u8 = vqrshrun_n_s16(q10s16, 7); + + if (yoffset == 0) { // firstpass_filter8x4_only + vst1_u8(dst_ptr, d22u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d23u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d24u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d25u8); + dst_ptr += dst_pitch; + } else { + vst1_u8(tmpp, d22u8); + tmpp += 8; + vst1_u8(tmpp, d23u8); + tmpp += 8; + vst1_u8(tmpp, d24u8); + tmpp += 8; + vst1_u8(tmpp, d25u8); + tmpp += 8; + } + } + if (yoffset == 0) + return; + + // First Pass on rest 5-line data + q3u8 = vld1q_u8(src); + src += src_pixels_per_line; + q4u8 = vld1q_u8(src); + src += src_pixels_per_line; + q5u8 = vld1q_u8(src); + src += src_pixels_per_line; + q6u8 = vld1q_u8(src); + src += src_pixels_per_line; + q7u8 = vld1q_u8(src); + + q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8); + q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8); + q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8); + q11u16 = vmull_u8(vget_low_u8(q6u8), d0u8); + q12u16 = vmull_u8(vget_low_u8(q7u8), d0u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 1); + + q8u16 = vmlsl_u8(q8u16, d27u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d1u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d1u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d1u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 4); + + q8u16 = vmlsl_u8(q8u16, d27u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d28u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d4u8); + q11u16 = vmlsl_u8(q11u16, d30u8, d4u8); + q12u16 = vmlsl_u8(q12u16, d31u8, d4u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 2); + + q8u16 = vmlal_u8(q8u16, d27u8, d2u8); + q9u16 = vmlal_u8(q9u16, d28u8, d2u8); + q10u16 = vmlal_u8(q10u16, d29u8, d2u8); + q11u16 = vmlal_u8(q11u16, d30u8, d2u8); + q12u16 = vmlal_u8(q12u16, d31u8, d2u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 5); + + q8u16 = vmlal_u8(q8u16, d27u8, d5u8); + q9u16 = vmlal_u8(q9u16, d28u8, d5u8); + q10u16 = vmlal_u8(q10u16, d29u8, d5u8); + q11u16 = vmlal_u8(q11u16, d30u8, d5u8); + q12u16 = vmlal_u8(q12u16, d31u8, d5u8); + + d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3); + d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3); + d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3); + d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3); + d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 3); + + q3u16 = vmull_u8(d27u8, d3u8); + q4u16 = vmull_u8(d28u8, d3u8); + q5u16 = vmull_u8(d29u8, d3u8); + q6u16 = vmull_u8(d30u8, d3u8); + q7u16 = vmull_u8(d31u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + + q8s16 = vqaddq_s16(q8s16, q3s16); + q9s16 = vqaddq_s16(q9s16, q4s16); + q10s16 = vqaddq_s16(q10s16, q5s16); + q11s16 = vqaddq_s16(q11s16, q6s16); + q12s16 = vqaddq_s16(q12s16, q7s16); + + d26u8 = vqrshrun_n_s16(q8s16, 7); + d27u8 = vqrshrun_n_s16(q9s16, 7); + d28u8 = vqrshrun_n_s16(q10s16, 7); + d29u8 = vqrshrun_n_s16(q11s16, 7); + d30u8 = vqrshrun_n_s16(q12s16, 7); + + // Second pass: 8x8 + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + tmpp = tmp; + q9u8 = vld1q_u8(tmpp); + tmpp += 16; + q10u8 = vld1q_u8(tmpp); + tmpp += 16; + q11u8 = vld1q_u8(tmpp); + tmpp += 16; + q12u8 = vld1q_u8(tmpp); + + d18u8 = vget_low_u8(q9u8); + d19u8 = vget_high_u8(q9u8); + d20u8 = vget_low_u8(q10u8); + d21u8 = vget_high_u8(q10u8); + d22u8 = vget_low_u8(q11u8); + d23u8 = vget_high_u8(q11u8); + d24u8 = vget_low_u8(q12u8); + d25u8 = vget_high_u8(q12u8); + + for (i = 2; i > 0; i--) { + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + d23u8 = d27u8; + d24u8 = d28u8; + d25u8 = d29u8; + d26u8 = d30u8; + + vst1_u8(dst_ptr, d6u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d7u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d8u8); + dst_ptr += dst_pitch; + vst1_u8(dst_ptr, d9u8); + dst_ptr += dst_pitch; + } + return; +} + +void vp8_sixtap_predict16x16_neon( + unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + unsigned char *dst_ptr, + int dst_pitch) { + unsigned char *src, *src_tmp, *dst, *tmpp; + unsigned char tmp[336]; + int i, j; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8; + uint8x8_t d10u8, d11u8, d12u8, d13u8, d14u8, d15u8, d18u8, d19u8; + uint8x8_t d20u8, d21u8, d22u8, d23u8, d24u8, d25u8, d26u8, d27u8; + uint8x8_t d28u8, d29u8, d30u8, d31u8; + int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8; + uint8x16_t q3u8, q4u8; + uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16, q8u16, q9u16, q10u16; + uint16x8_t q11u16, q12u16, q13u16, q15u16; + int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16, q8s16, q9s16, q10s16; + int16x8_t q11s16, q12s16, q13s16, q15s16; + + if (xoffset == 0) { // secondpass_filter8x8_only + // load second_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // load src data + src_tmp = src_ptr - src_pixels_per_line * 2; + for (i = 0; i < 2; i++) { + src = src_tmp + i * 8; + dst = dst_ptr + i * 8; + d18u8 = vld1_u8(src); + src += src_pixels_per_line; + d19u8 = vld1_u8(src); + src += src_pixels_per_line; + d20u8 = vld1_u8(src); + src += src_pixels_per_line; + d21u8 = vld1_u8(src); + src += src_pixels_per_line; + d22u8 = vld1_u8(src); + src += src_pixels_per_line; + for (j = 0; j < 4; j++) { + d23u8 = vld1_u8(src); + src += src_pixels_per_line; + d24u8 = vld1_u8(src); + src += src_pixels_per_line; + d25u8 = vld1_u8(src); + src += src_pixels_per_line; + d26u8 = vld1_u8(src); + src += src_pixels_per_line; + + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + + vst1_u8(dst, d6u8); + dst += dst_pitch; + vst1_u8(dst, d7u8); + dst += dst_pitch; + vst1_u8(dst, d8u8); + dst += dst_pitch; + vst1_u8(dst, d9u8); + dst += dst_pitch; + } + } + return; + } + + // load first_pass filter + dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + // First pass: output_height lines x output_width columns (9x4) + if (yoffset == 0) { // firstpass_filter4x4_only + src = src_ptr - 2; + dst = dst_ptr; + for (i = 0; i < 8; i++) { + d6u8 = vld1_u8(src); + d7u8 = vld1_u8(src + 8); + d8u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + d9u8 = vld1_u8(src); + d10u8 = vld1_u8(src + 8); + d11u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + + __builtin_prefetch(src); + __builtin_prefetch(src + src_pixels_per_line); + + q6u16 = vmull_u8(d6u8, d0u8); + q7u16 = vmull_u8(d7u8, d0u8); + q8u16 = vmull_u8(d9u8, d0u8); + q9u16 = vmull_u8(d10u8, d0u8); + + d20u8 = vext_u8(d6u8, d7u8, 1); + d21u8 = vext_u8(d9u8, d10u8, 1); + d22u8 = vext_u8(d7u8, d8u8, 1); + d23u8 = vext_u8(d10u8, d11u8, 1); + d24u8 = vext_u8(d6u8, d7u8, 4); + d25u8 = vext_u8(d9u8, d10u8, 4); + d26u8 = vext_u8(d7u8, d8u8, 4); + d27u8 = vext_u8(d10u8, d11u8, 4); + d28u8 = vext_u8(d6u8, d7u8, 5); + d29u8 = vext_u8(d9u8, d10u8, 5); + + q6u16 = vmlsl_u8(q6u16, d20u8, d1u8); + q8u16 = vmlsl_u8(q8u16, d21u8, d1u8); + q7u16 = vmlsl_u8(q7u16, d22u8, d1u8); + q9u16 = vmlsl_u8(q9u16, d23u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d24u8, d4u8); + q8u16 = vmlsl_u8(q8u16, d25u8, d4u8); + q7u16 = vmlsl_u8(q7u16, d26u8, d4u8); + q9u16 = vmlsl_u8(q9u16, d27u8, d4u8); + q6u16 = vmlal_u8(q6u16, d28u8, d5u8); + q8u16 = vmlal_u8(q8u16, d29u8, d5u8); + + d20u8 = vext_u8(d7u8, d8u8, 5); + d21u8 = vext_u8(d10u8, d11u8, 5); + d22u8 = vext_u8(d6u8, d7u8, 2); + d23u8 = vext_u8(d9u8, d10u8, 2); + d24u8 = vext_u8(d7u8, d8u8, 2); + d25u8 = vext_u8(d10u8, d11u8, 2); + d26u8 = vext_u8(d6u8, d7u8, 3); + d27u8 = vext_u8(d9u8, d10u8, 3); + d28u8 = vext_u8(d7u8, d8u8, 3); + d29u8 = vext_u8(d10u8, d11u8, 3); + + q7u16 = vmlal_u8(q7u16, d20u8, d5u8); + q9u16 = vmlal_u8(q9u16, d21u8, d5u8); + q6u16 = vmlal_u8(q6u16, d22u8, d2u8); + q8u16 = vmlal_u8(q8u16, d23u8, d2u8); + q7u16 = vmlal_u8(q7u16, d24u8, d2u8); + q9u16 = vmlal_u8(q9u16, d25u8, d2u8); + + q10u16 = vmull_u8(d26u8, d3u8); + q11u16 = vmull_u8(d27u8, d3u8); + q12u16 = vmull_u8(d28u8, d3u8); + q15u16 = vmull_u8(d29u8, d3u8); + + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + q15s16 = vreinterpretq_s16_u16(q15u16); + + q6s16 = vqaddq_s16(q6s16, q10s16); + q8s16 = vqaddq_s16(q8s16, q11s16); + q7s16 = vqaddq_s16(q7s16, q12s16); + q9s16 = vqaddq_s16(q9s16, q15s16); + + d6u8 = vqrshrun_n_s16(q6s16, 7); + d7u8 = vqrshrun_n_s16(q7s16, 7); + d8u8 = vqrshrun_n_s16(q8s16, 7); + d9u8 = vqrshrun_n_s16(q9s16, 7); + + q3u8 = vcombine_u8(d6u8, d7u8); + q4u8 = vcombine_u8(d8u8, d9u8); + vst1q_u8(dst, q3u8); + dst += dst_pitch; + vst1q_u8(dst, q4u8); + dst += dst_pitch; + } + return; + } + + src = src_ptr - 2 - src_pixels_per_line * 2; + tmpp = tmp; + for (i = 0; i < 7; i++) { + d6u8 = vld1_u8(src); + d7u8 = vld1_u8(src + 8); + d8u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + d9u8 = vld1_u8(src); + d10u8 = vld1_u8(src + 8); + d11u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + d12u8 = vld1_u8(src); + d13u8 = vld1_u8(src + 8); + d14u8 = vld1_u8(src + 16); + src += src_pixels_per_line; + + __builtin_prefetch(src); + __builtin_prefetch(src + src_pixels_per_line); + __builtin_prefetch(src + src_pixels_per_line * 2); + + q8u16 = vmull_u8(d6u8, d0u8); + q9u16 = vmull_u8(d7u8, d0u8); + q10u16 = vmull_u8(d9u8, d0u8); + q11u16 = vmull_u8(d10u8, d0u8); + q12u16 = vmull_u8(d12u8, d0u8); + q13u16 = vmull_u8(d13u8, d0u8); + + d28u8 = vext_u8(d6u8, d7u8, 1); + d29u8 = vext_u8(d9u8, d10u8, 1); + d30u8 = vext_u8(d12u8, d13u8, 1); + q8u16 = vmlsl_u8(q8u16, d28u8, d1u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d1u8); + q12u16 = vmlsl_u8(q12u16, d30u8, d1u8); + d28u8 = vext_u8(d7u8, d8u8, 1); + d29u8 = vext_u8(d10u8, d11u8, 1); + d30u8 = vext_u8(d13u8, d14u8, 1); + q9u16 = vmlsl_u8(q9u16, d28u8, d1u8); + q11u16 = vmlsl_u8(q11u16, d29u8, d1u8); + q13u16 = vmlsl_u8(q13u16, d30u8, d1u8); + + d28u8 = vext_u8(d6u8, d7u8, 4); + d29u8 = vext_u8(d9u8, d10u8, 4); + d30u8 = vext_u8(d12u8, d13u8, 4); + q8u16 = vmlsl_u8(q8u16, d28u8, d4u8); + q10u16 = vmlsl_u8(q10u16, d29u8, d4u8); + q12u16 = vmlsl_u8(q12u16, d30u8, d4u8); + d28u8 = vext_u8(d7u8, d8u8, 4); + d29u8 = vext_u8(d10u8, d11u8, 4); + d30u8 = vext_u8(d13u8, d14u8, 4); + q9u16 = vmlsl_u8(q9u16, d28u8, d4u8); + q11u16 = vmlsl_u8(q11u16, d29u8, d4u8); + q13u16 = vmlsl_u8(q13u16, d30u8, d4u8); + + d28u8 = vext_u8(d6u8, d7u8, 5); + d29u8 = vext_u8(d9u8, d10u8, 5); + d30u8 = vext_u8(d12u8, d13u8, 5); + q8u16 = vmlal_u8(q8u16, d28u8, d5u8); + q10u16 = vmlal_u8(q10u16, d29u8, d5u8); + q12u16 = vmlal_u8(q12u16, d30u8, d5u8); + d28u8 = vext_u8(d7u8, d8u8, 5); + d29u8 = vext_u8(d10u8, d11u8, 5); + d30u8 = vext_u8(d13u8, d14u8, 5); + q9u16 = vmlal_u8(q9u16, d28u8, d5u8); + q11u16 = vmlal_u8(q11u16, d29u8, d5u8); + q13u16 = vmlal_u8(q13u16, d30u8, d5u8); + + d28u8 = vext_u8(d6u8, d7u8, 2); + d29u8 = vext_u8(d9u8, d10u8, 2); + d30u8 = vext_u8(d12u8, d13u8, 2); + q8u16 = vmlal_u8(q8u16, d28u8, d2u8); + q10u16 = vmlal_u8(q10u16, d29u8, d2u8); + q12u16 = vmlal_u8(q12u16, d30u8, d2u8); + d28u8 = vext_u8(d7u8, d8u8, 2); + d29u8 = vext_u8(d10u8, d11u8, 2); + d30u8 = vext_u8(d13u8, d14u8, 2); + q9u16 = vmlal_u8(q9u16, d28u8, d2u8); + q11u16 = vmlal_u8(q11u16, d29u8, d2u8); + q13u16 = vmlal_u8(q13u16, d30u8, d2u8); + + d28u8 = vext_u8(d6u8, d7u8, 3); + d29u8 = vext_u8(d9u8, d10u8, 3); + d30u8 = vext_u8(d12u8, d13u8, 3); + d15u8 = vext_u8(d7u8, d8u8, 3); + d31u8 = vext_u8(d10u8, d11u8, 3); + d6u8 = vext_u8(d13u8, d14u8, 3); + q4u16 = vmull_u8(d28u8, d3u8); + q5u16 = vmull_u8(d29u8, d3u8); + q6u16 = vmull_u8(d30u8, d3u8); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + q12s16 = vreinterpretq_s16_u16(q12u16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q10s16 = vqaddq_s16(q10s16, q5s16); + q12s16 = vqaddq_s16(q12s16, q6s16); + + q6u16 = vmull_u8(d15u8, d3u8); + q7u16 = vmull_u8(d31u8, d3u8); + q3u16 = vmull_u8(d6u8, d3u8); + q3s16 = vreinterpretq_s16_u16(q3u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q11s16 = vreinterpretq_s16_u16(q11u16); + q13s16 = vreinterpretq_s16_u16(q13u16); + q9s16 = vqaddq_s16(q9s16, q6s16); + q11s16 = vqaddq_s16(q11s16, q7s16); + q13s16 = vqaddq_s16(q13s16, q3s16); + + d6u8 = vqrshrun_n_s16(q8s16, 7); + d7u8 = vqrshrun_n_s16(q9s16, 7); + d8u8 = vqrshrun_n_s16(q10s16, 7); + d9u8 = vqrshrun_n_s16(q11s16, 7); + d10u8 = vqrshrun_n_s16(q12s16, 7); + d11u8 = vqrshrun_n_s16(q13s16, 7); + + vst1_u8(tmpp, d6u8); + tmpp += 8; + vst1_u8(tmpp, d7u8); + tmpp += 8; + vst1_u8(tmpp, d8u8); + tmpp += 8; + vst1_u8(tmpp, d9u8); + tmpp += 8; + vst1_u8(tmpp, d10u8); + tmpp += 8; + vst1_u8(tmpp, d11u8); + tmpp += 8; + } + + // Second pass: 16x16 + dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]); + d0s8 = vdup_lane_s8(dtmps8, 0); + d1s8 = vdup_lane_s8(dtmps8, 1); + d2s8 = vdup_lane_s8(dtmps8, 2); + d3s8 = vdup_lane_s8(dtmps8, 3); + d4s8 = vdup_lane_s8(dtmps8, 4); + d5s8 = vdup_lane_s8(dtmps8, 5); + d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8)); + d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8)); + d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8)); + d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8)); + d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8)); + d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8)); + + for (i = 0; i < 2; i++) { + dst = dst_ptr + 8 * i; + tmpp = tmp + 8 * i; + d18u8 = vld1_u8(tmpp); + tmpp += 16; + d19u8 = vld1_u8(tmpp); + tmpp += 16; + d20u8 = vld1_u8(tmpp); + tmpp += 16; + d21u8 = vld1_u8(tmpp); + tmpp += 16; + d22u8 = vld1_u8(tmpp); + tmpp += 16; + for (j = 0; j < 4; j++) { + d23u8 = vld1_u8(tmpp); + tmpp += 16; + d24u8 = vld1_u8(tmpp); + tmpp += 16; + d25u8 = vld1_u8(tmpp); + tmpp += 16; + d26u8 = vld1_u8(tmpp); + tmpp += 16; + + q3u16 = vmull_u8(d18u8, d0u8); + q4u16 = vmull_u8(d19u8, d0u8); + q5u16 = vmull_u8(d20u8, d0u8); + q6u16 = vmull_u8(d21u8, d0u8); + + q3u16 = vmlsl_u8(q3u16, d19u8, d1u8); + q4u16 = vmlsl_u8(q4u16, d20u8, d1u8); + q5u16 = vmlsl_u8(q5u16, d21u8, d1u8); + q6u16 = vmlsl_u8(q6u16, d22u8, d1u8); + + q3u16 = vmlsl_u8(q3u16, d22u8, d4u8); + q4u16 = vmlsl_u8(q4u16, d23u8, d4u8); + q5u16 = vmlsl_u8(q5u16, d24u8, d4u8); + q6u16 = vmlsl_u8(q6u16, d25u8, d4u8); + + q3u16 = vmlal_u8(q3u16, d20u8, d2u8); + q4u16 = vmlal_u8(q4u16, d21u8, d2u8); + q5u16 = vmlal_u8(q5u16, d22u8, d2u8); + q6u16 = vmlal_u8(q6u16, d23u8, d2u8); + + q3u16 = vmlal_u8(q3u16, d23u8, d5u8); + q4u16 = vmlal_u8(q4u16, d24u8, d5u8); + q5u16 = vmlal_u8(q5u16, d25u8, d5u8); + q6u16 = vmlal_u8(q6u16, d26u8, d5u8); + + q7u16 = vmull_u8(d21u8, d3u8); + q8u16 = vmull_u8(d22u8, d3u8); + q9u16 = vmull_u8(d23u8, d3u8); + q10u16 = vmull_u8(d24u8, d3u8); + + q3s16 = vreinterpretq_s16_u16(q3u16); + q4s16 = vreinterpretq_s16_u16(q4u16); + q5s16 = vreinterpretq_s16_u16(q5u16); + q6s16 = vreinterpretq_s16_u16(q6u16); + q7s16 = vreinterpretq_s16_u16(q7u16); + q8s16 = vreinterpretq_s16_u16(q8u16); + q9s16 = vreinterpretq_s16_u16(q9u16); + q10s16 = vreinterpretq_s16_u16(q10u16); + + q7s16 = vqaddq_s16(q7s16, q3s16); + q8s16 = vqaddq_s16(q8s16, q4s16); + q9s16 = vqaddq_s16(q9s16, q5s16); + q10s16 = vqaddq_s16(q10s16, q6s16); + + d6u8 = vqrshrun_n_s16(q7s16, 7); + d7u8 = vqrshrun_n_s16(q8s16, 7); + d8u8 = vqrshrun_n_s16(q9s16, 7); + d9u8 = vqrshrun_n_s16(q10s16, 7); + + d18u8 = d22u8; + d19u8 = d23u8; + d20u8 = d24u8; + d21u8 = d25u8; + d22u8 = d26u8; + + vst1_u8(dst, d6u8); + dst += dst_pitch; + vst1_u8(dst, d7u8); + dst += dst_pitch; + vst1_u8(dst, d8u8); + dst += dst_pitch; + vst1_u8(dst, d9u8); + dst += dst_pitch; + } + } + return; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/variance_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/variance_neon.asm deleted file mode 100644 index e3b48327d3f..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/variance_neon.asm +++ /dev/null @@ -1,276 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE 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. -; - - - EXPORT |vp8_variance16x16_neon| - EXPORT |vp8_variance16x8_neon| - EXPORT |vp8_variance8x16_neon| - EXPORT |vp8_variance8x8_neon| - - ARM - REQUIRE8 - PRESERVE8 - - AREA ||.text||, CODE, READONLY, ALIGN=2 - -; r0 unsigned char *src_ptr -; r1 int source_stride -; r2 unsigned char *ref_ptr -; r3 int recon_stride -; stack unsigned int *sse -|vp8_variance16x16_neon| PROC - vmov.i8 q8, #0 ;q8 - sum - vmov.i8 q9, #0 ;q9, q10 - sse - vmov.i8 q10, #0 - - mov r12, #8 - -variance16x16_neon_loop - vld1.8 {q0}, [r0], r1 ;Load up source and reference - vld1.8 {q2}, [r2], r3 - vld1.8 {q1}, [r0], r1 - vld1.8 {q3}, [r2], r3 - - vsubl.u8 q11, d0, d4 ;calculate diff - vsubl.u8 q12, d1, d5 - vsubl.u8 q13, d2, d6 - vsubl.u8 q14, d3, d7 - - ;VPADAL adds adjacent pairs of elements of a vector, and accumulates - ;the results into the elements of the destination vector. The explanation - ;in ARM guide is wrong. - vpadal.s16 q8, q11 ;calculate sum - vmlal.s16 q9, d22, d22 ;calculate sse - vmlal.s16 q10, d23, d23 - - subs r12, r12, #1 - - vpadal.s16 q8, q12 - vmlal.s16 q9, d24, d24 - vmlal.s16 q10, d25, d25 - vpadal.s16 q8, q13 - vmlal.s16 q9, d26, d26 - vmlal.s16 q10, d27, d27 - vpadal.s16 q8, q14 - vmlal.s16 q9, d28, d28 - vmlal.s16 q10, d29, d29 - - bne variance16x16_neon_loop - - vadd.u32 q10, q9, q10 ;accumulate sse - vpaddl.s32 q0, q8 ;accumulate sum - - ldr r12, [sp] ;load *sse from stack - - vpaddl.u32 q1, q10 - vadd.s64 d0, d0, d1 - vadd.u64 d1, d2, d3 - - ;vmov.32 r0, d0[0] ;this instruction costs a lot - ;vmov.32 r1, d1[0] - ;mul r0, r0, r0 - ;str r1, [r12] - ;sub r0, r1, r0, lsr #8 - - ; while sum is signed, sum * sum is always positive and must be treated as - ; unsigned to avoid propagating the sign bit. - vmull.s32 q5, d0, d0 - vst1.32 {d1[0]}, [r12] ;store sse - vshr.u32 d10, d10, #8 - vsub.u32 d0, d1, d10 - - vmov.32 r0, d0[0] ;return - bx lr - - ENDP - -;================================ -;unsigned int vp8_variance16x8_c( -; unsigned char *src_ptr, -; int source_stride, -; unsigned char *ref_ptr, -; int recon_stride, -; unsigned int *sse) -|vp8_variance16x8_neon| PROC - vmov.i8 q8, #0 ;q8 - sum - vmov.i8 q9, #0 ;q9, q10 - sse - vmov.i8 q10, #0 - - mov r12, #4 - -variance16x8_neon_loop - vld1.8 {q0}, [r0], r1 ;Load up source and reference - vld1.8 {q2}, [r2], r3 - vld1.8 {q1}, [r0], r1 - vld1.8 {q3}, [r2], r3 - - vsubl.u8 q11, d0, d4 ;calculate diff - vsubl.u8 q12, d1, d5 - vsubl.u8 q13, d2, d6 - vsubl.u8 q14, d3, d7 - - vpadal.s16 q8, q11 ;calculate sum - vmlal.s16 q9, d22, d22 ;calculate sse - vmlal.s16 q10, d23, d23 - - subs r12, r12, #1 - - vpadal.s16 q8, q12 - vmlal.s16 q9, d24, d24 - vmlal.s16 q10, d25, d25 - vpadal.s16 q8, q13 - vmlal.s16 q9, d26, d26 - vmlal.s16 q10, d27, d27 - vpadal.s16 q8, q14 - vmlal.s16 q9, d28, d28 - vmlal.s16 q10, d29, d29 - - bne variance16x8_neon_loop - - vadd.u32 q10, q9, q10 ;accumulate sse - vpaddl.s32 q0, q8 ;accumulate sum - - ldr r12, [sp] ;load *sse from stack - - vpaddl.u32 q1, q10 - vadd.s64 d0, d0, d1 - vadd.u64 d1, d2, d3 - - vmull.s32 q5, d0, d0 - vst1.32 {d1[0]}, [r12] ;store sse - vshr.u32 d10, d10, #7 - vsub.u32 d0, d1, d10 - - vmov.32 r0, d0[0] ;return - bx lr - - ENDP - -;================================= -;unsigned int vp8_variance8x16_c( -; unsigned char *src_ptr, -; int source_stride, -; unsigned char *ref_ptr, -; int recon_stride, -; unsigned int *sse) - -|vp8_variance8x16_neon| PROC - vmov.i8 q8, #0 ;q8 - sum - vmov.i8 q9, #0 ;q9, q10 - sse - vmov.i8 q10, #0 - - mov r12, #8 - -variance8x16_neon_loop - vld1.8 {d0}, [r0], r1 ;Load up source and reference - vld1.8 {d4}, [r2], r3 - vld1.8 {d2}, [r0], r1 - vld1.8 {d6}, [r2], r3 - - vsubl.u8 q11, d0, d4 ;calculate diff - vsubl.u8 q12, d2, d6 - - vpadal.s16 q8, q11 ;calculate sum - vmlal.s16 q9, d22, d22 ;calculate sse - vmlal.s16 q10, d23, d23 - - subs r12, r12, #1 - - vpadal.s16 q8, q12 - vmlal.s16 q9, d24, d24 - vmlal.s16 q10, d25, d25 - - bne variance8x16_neon_loop - - vadd.u32 q10, q9, q10 ;accumulate sse - vpaddl.s32 q0, q8 ;accumulate sum - - ldr r12, [sp] ;load *sse from stack - - vpaddl.u32 q1, q10 - vadd.s64 d0, d0, d1 - vadd.u64 d1, d2, d3 - - vmull.s32 q5, d0, d0 - vst1.32 {d1[0]}, [r12] ;store sse - vshr.u32 d10, d10, #7 - vsub.u32 d0, d1, d10 - - vmov.32 r0, d0[0] ;return - bx lr - - ENDP - -;================================== -; r0 unsigned char *src_ptr -; r1 int source_stride -; r2 unsigned char *ref_ptr -; r3 int recon_stride -; stack unsigned int *sse -|vp8_variance8x8_neon| PROC - vmov.i8 q8, #0 ;q8 - sum - vmov.i8 q9, #0 ;q9, q10 - sse - vmov.i8 q10, #0 - - mov r12, #2 - -variance8x8_neon_loop - vld1.8 {d0}, [r0], r1 ;Load up source and reference - vld1.8 {d4}, [r2], r3 - vld1.8 {d1}, [r0], r1 - vld1.8 {d5}, [r2], r3 - vld1.8 {d2}, [r0], r1 - vld1.8 {d6}, [r2], r3 - vld1.8 {d3}, [r0], r1 - vld1.8 {d7}, [r2], r3 - - vsubl.u8 q11, d0, d4 ;calculate diff - vsubl.u8 q12, d1, d5 - vsubl.u8 q13, d2, d6 - vsubl.u8 q14, d3, d7 - - vpadal.s16 q8, q11 ;calculate sum - vmlal.s16 q9, d22, d22 ;calculate sse - vmlal.s16 q10, d23, d23 - - subs r12, r12, #1 - - vpadal.s16 q8, q12 - vmlal.s16 q9, d24, d24 - vmlal.s16 q10, d25, d25 - vpadal.s16 q8, q13 - vmlal.s16 q9, d26, d26 - vmlal.s16 q10, d27, d27 - vpadal.s16 q8, q14 - vmlal.s16 q9, d28, d28 - vmlal.s16 q10, d29, d29 - - bne variance8x8_neon_loop - - vadd.u32 q10, q9, q10 ;accumulate sse - vpaddl.s32 q0, q8 ;accumulate sum - - ldr r12, [sp] ;load *sse from stack - - vpaddl.u32 q1, q10 - vadd.s64 d0, d0, d1 - vadd.u64 d1, d2, d3 - - vmull.s32 q5, d0, d0 - vst1.32 {d1[0]}, [r12] ;store sse - vshr.u32 d10, d10, #6 - vsub.u32 d0, d1, d10 - - vmov.32 r0, d0[0] ;return - bx lr - - ENDP - - END diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/variance_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/variance_neon.c new file mode 100644 index 00000000000..afd2dc3d1e2 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/variance_neon.c @@ -0,0 +1,323 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +#ifdef _MSC_VER +#define __builtin_prefetch(x) +#endif + +unsigned int vp8_variance16x16_neon( + const unsigned char *src_ptr, + int source_stride, + const unsigned char *ref_ptr, + int recon_stride, + unsigned int *sse) { + int i; + int16x4_t d22s16, d23s16, d24s16, d25s16, d26s16, d27s16, d28s16, d29s16; + uint32x2_t d0u32, d10u32; + int64x1_t d0s64, d1s64; + uint8x16_t q0u8, q1u8, q2u8, q3u8; + uint16x8_t q11u16, q12u16, q13u16, q14u16; + int32x4_t q8s32, q9s32, q10s32; + int64x2_t q0s64, q1s64, q5s64; + + q8s32 = vdupq_n_s32(0); + q9s32 = vdupq_n_s32(0); + q10s32 = vdupq_n_s32(0); + + for (i = 0; i < 8; i++) { + q0u8 = vld1q_u8(src_ptr); + src_ptr += source_stride; + q1u8 = vld1q_u8(src_ptr); + src_ptr += source_stride; + __builtin_prefetch(src_ptr); + + q2u8 = vld1q_u8(ref_ptr); + ref_ptr += recon_stride; + q3u8 = vld1q_u8(ref_ptr); + ref_ptr += recon_stride; + __builtin_prefetch(ref_ptr); + + q11u16 = vsubl_u8(vget_low_u8(q0u8), vget_low_u8(q2u8)); + q12u16 = vsubl_u8(vget_high_u8(q0u8), vget_high_u8(q2u8)); + q13u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q3u8)); + q14u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q3u8)); + + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q11u16)); + q9s32 = vmlal_s16(q9s32, d22s16, d22s16); + q10s32 = vmlal_s16(q10s32, d23s16, d23s16); + + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q12u16)); + q9s32 = vmlal_s16(q9s32, d24s16, d24s16); + q10s32 = vmlal_s16(q10s32, d25s16, d25s16); + + d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16)); + d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q13u16)); + q9s32 = vmlal_s16(q9s32, d26s16, d26s16); + q10s32 = vmlal_s16(q10s32, d27s16, d27s16); + + d28s16 = vreinterpret_s16_u16(vget_low_u16(q14u16)); + d29s16 = vreinterpret_s16_u16(vget_high_u16(q14u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q14u16)); + q9s32 = vmlal_s16(q9s32, d28s16, d28s16); + q10s32 = vmlal_s16(q10s32, d29s16, d29s16); + } + + q10s32 = vaddq_s32(q10s32, q9s32); + q0s64 = vpaddlq_s32(q8s32); + q1s64 = vpaddlq_s32(q10s32); + + d0s64 = vadd_s64(vget_low_s64(q0s64), vget_high_s64(q0s64)); + d1s64 = vadd_s64(vget_low_s64(q1s64), vget_high_s64(q1s64)); + + q5s64 = vmull_s32(vreinterpret_s32_s64(d0s64), + vreinterpret_s32_s64(d0s64)); + vst1_lane_u32((uint32_t *)sse, vreinterpret_u32_s64(d1s64), 0); + + d10u32 = vshr_n_u32(vreinterpret_u32_s64(vget_low_s64(q5s64)), 8); + d0u32 = vsub_u32(vreinterpret_u32_s64(d1s64), d10u32); + + return vget_lane_u32(d0u32, 0); +} + +unsigned int vp8_variance16x8_neon( + const unsigned char *src_ptr, + int source_stride, + const unsigned char *ref_ptr, + int recon_stride, + unsigned int *sse) { + int i; + int16x4_t d22s16, d23s16, d24s16, d25s16, d26s16, d27s16, d28s16, d29s16; + uint32x2_t d0u32, d10u32; + int64x1_t d0s64, d1s64; + uint8x16_t q0u8, q1u8, q2u8, q3u8; + uint16x8_t q11u16, q12u16, q13u16, q14u16; + int32x4_t q8s32, q9s32, q10s32; + int64x2_t q0s64, q1s64, q5s64; + + q8s32 = vdupq_n_s32(0); + q9s32 = vdupq_n_s32(0); + q10s32 = vdupq_n_s32(0); + + for (i = 0; i < 4; i++) { // variance16x8_neon_loop + q0u8 = vld1q_u8(src_ptr); + src_ptr += source_stride; + q1u8 = vld1q_u8(src_ptr); + src_ptr += source_stride; + __builtin_prefetch(src_ptr); + + q2u8 = vld1q_u8(ref_ptr); + ref_ptr += recon_stride; + q3u8 = vld1q_u8(ref_ptr); + ref_ptr += recon_stride; + __builtin_prefetch(ref_ptr); + + q11u16 = vsubl_u8(vget_low_u8(q0u8), vget_low_u8(q2u8)); + q12u16 = vsubl_u8(vget_high_u8(q0u8), vget_high_u8(q2u8)); + q13u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q3u8)); + q14u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q3u8)); + + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q11u16)); + q9s32 = vmlal_s16(q9s32, d22s16, d22s16); + q10s32 = vmlal_s16(q10s32, d23s16, d23s16); + + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q12u16)); + q9s32 = vmlal_s16(q9s32, d24s16, d24s16); + q10s32 = vmlal_s16(q10s32, d25s16, d25s16); + + d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16)); + d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q13u16)); + q9s32 = vmlal_s16(q9s32, d26s16, d26s16); + q10s32 = vmlal_s16(q10s32, d27s16, d27s16); + + d28s16 = vreinterpret_s16_u16(vget_low_u16(q14u16)); + d29s16 = vreinterpret_s16_u16(vget_high_u16(q14u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q14u16)); + q9s32 = vmlal_s16(q9s32, d28s16, d28s16); + q10s32 = vmlal_s16(q10s32, d29s16, d29s16); + } + + q10s32 = vaddq_s32(q10s32, q9s32); + q0s64 = vpaddlq_s32(q8s32); + q1s64 = vpaddlq_s32(q10s32); + + d0s64 = vadd_s64(vget_low_s64(q0s64), vget_high_s64(q0s64)); + d1s64 = vadd_s64(vget_low_s64(q1s64), vget_high_s64(q1s64)); + + q5s64 = vmull_s32(vreinterpret_s32_s64(d0s64), + vreinterpret_s32_s64(d0s64)); + vst1_lane_u32((uint32_t *)sse, vreinterpret_u32_s64(d1s64), 0); + + d10u32 = vshr_n_u32(vreinterpret_u32_s64(vget_low_s64(q5s64)), 7); + d0u32 = vsub_u32(vreinterpret_u32_s64(d1s64), d10u32); + + return vget_lane_u32(d0u32, 0); +} + +unsigned int vp8_variance8x16_neon( + const unsigned char *src_ptr, + int source_stride, + const unsigned char *ref_ptr, + int recon_stride, + unsigned int *sse) { + int i; + uint8x8_t d0u8, d2u8, d4u8, d6u8; + int16x4_t d22s16, d23s16, d24s16, d25s16; + uint32x2_t d0u32, d10u32; + int64x1_t d0s64, d1s64; + uint16x8_t q11u16, q12u16; + int32x4_t q8s32, q9s32, q10s32; + int64x2_t q0s64, q1s64, q5s64; + + q8s32 = vdupq_n_s32(0); + q9s32 = vdupq_n_s32(0); + q10s32 = vdupq_n_s32(0); + + for (i = 0; i < 8; i++) { // variance8x16_neon_loop + d0u8 = vld1_u8(src_ptr); + src_ptr += source_stride; + d2u8 = vld1_u8(src_ptr); + src_ptr += source_stride; + __builtin_prefetch(src_ptr); + + d4u8 = vld1_u8(ref_ptr); + ref_ptr += recon_stride; + d6u8 = vld1_u8(ref_ptr); + ref_ptr += recon_stride; + __builtin_prefetch(ref_ptr); + + q11u16 = vsubl_u8(d0u8, d4u8); + q12u16 = vsubl_u8(d2u8, d6u8); + + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q11u16)); + q9s32 = vmlal_s16(q9s32, d22s16, d22s16); + q10s32 = vmlal_s16(q10s32, d23s16, d23s16); + + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q12u16)); + q9s32 = vmlal_s16(q9s32, d24s16, d24s16); + q10s32 = vmlal_s16(q10s32, d25s16, d25s16); + } + + q10s32 = vaddq_s32(q10s32, q9s32); + q0s64 = vpaddlq_s32(q8s32); + q1s64 = vpaddlq_s32(q10s32); + + d0s64 = vadd_s64(vget_low_s64(q0s64), vget_high_s64(q0s64)); + d1s64 = vadd_s64(vget_low_s64(q1s64), vget_high_s64(q1s64)); + + q5s64 = vmull_s32(vreinterpret_s32_s64(d0s64), + vreinterpret_s32_s64(d0s64)); + vst1_lane_u32((uint32_t *)sse, vreinterpret_u32_s64(d1s64), 0); + + d10u32 = vshr_n_u32(vreinterpret_u32_s64(vget_low_s64(q5s64)), 7); + d0u32 = vsub_u32(vreinterpret_u32_s64(d1s64), d10u32); + + return vget_lane_u32(d0u32, 0); +} + +unsigned int vp8_variance8x8_neon( + const unsigned char *src_ptr, + int source_stride, + const unsigned char *ref_ptr, + int recon_stride, + unsigned int *sse) { + int i; + uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8; + int16x4_t d22s16, d23s16, d24s16, d25s16, d26s16, d27s16, d28s16, d29s16; + uint32x2_t d0u32, d10u32; + int64x1_t d0s64, d1s64; + uint16x8_t q11u16, q12u16, q13u16, q14u16; + int32x4_t q8s32, q9s32, q10s32; + int64x2_t q0s64, q1s64, q5s64; + + q8s32 = vdupq_n_s32(0); + q9s32 = vdupq_n_s32(0); + q10s32 = vdupq_n_s32(0); + + for (i = 0; i < 2; i++) { // variance8x8_neon_loop + d0u8 = vld1_u8(src_ptr); + src_ptr += source_stride; + d1u8 = vld1_u8(src_ptr); + src_ptr += source_stride; + d2u8 = vld1_u8(src_ptr); + src_ptr += source_stride; + d3u8 = vld1_u8(src_ptr); + src_ptr += source_stride; + + d4u8 = vld1_u8(ref_ptr); + ref_ptr += recon_stride; + d5u8 = vld1_u8(ref_ptr); + ref_ptr += recon_stride; + d6u8 = vld1_u8(ref_ptr); + ref_ptr += recon_stride; + d7u8 = vld1_u8(ref_ptr); + ref_ptr += recon_stride; + + q11u16 = vsubl_u8(d0u8, d4u8); + q12u16 = vsubl_u8(d1u8, d5u8); + q13u16 = vsubl_u8(d2u8, d6u8); + q14u16 = vsubl_u8(d3u8, d7u8); + + d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16)); + d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q11u16)); + q9s32 = vmlal_s16(q9s32, d22s16, d22s16); + q10s32 = vmlal_s16(q10s32, d23s16, d23s16); + + d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16)); + d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q12u16)); + q9s32 = vmlal_s16(q9s32, d24s16, d24s16); + q10s32 = vmlal_s16(q10s32, d25s16, d25s16); + + d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16)); + d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q13u16)); + q9s32 = vmlal_s16(q9s32, d26s16, d26s16); + q10s32 = vmlal_s16(q10s32, d27s16, d27s16); + + d28s16 = vreinterpret_s16_u16(vget_low_u16(q14u16)); + d29s16 = vreinterpret_s16_u16(vget_high_u16(q14u16)); + q8s32 = vpadalq_s16(q8s32, vreinterpretq_s16_u16(q14u16)); + q9s32 = vmlal_s16(q9s32, d28s16, d28s16); + q10s32 = vmlal_s16(q10s32, d29s16, d29s16); + } + + q10s32 = vaddq_s32(q10s32, q9s32); + q0s64 = vpaddlq_s32(q8s32); + q1s64 = vpaddlq_s32(q10s32); + + d0s64 = vadd_s64(vget_low_s64(q0s64), vget_high_s64(q0s64)); + d1s64 = vadd_s64(vget_low_s64(q1s64), vget_high_s64(q1s64)); + + q5s64 = vmull_s32(vreinterpret_s32_s64(d0s64), + vreinterpret_s32_s64(d0s64)); + vst1_lane_u32((uint32_t *)sse, vreinterpret_u32_s64(d1s64), 0); + + d10u32 = vshr_n_u32(vreinterpret_u32_s64(vget_low_s64(q5s64)), 6); + d0u32 = vsub_u32(vreinterpret_u32_s64(d1s64), d10u32); + + return vget_lane_u32(d0u32, 0); +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm index 9d22c52521c..adc5b7e3a78 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm @@ -31,11 +31,12 @@ bilinear_taps_coeff |vp8_sub_pixel_variance16x16_neon_func| PROC push {r4-r6, lr} + vpush {d8-d15} adr r12, bilinear_taps_coeff - ldr r4, [sp, #16] ;load *dst_ptr from stack - ldr r5, [sp, #20] ;load dst_pixels_per_line from stack - ldr r6, [sp, #24] ;load *sse from stack + ldr r4, [sp, #80] ;load *dst_ptr from stack + ldr r5, [sp, #84] ;load dst_pixels_per_line from stack + ldr r6, [sp, #88] ;load *sse from stack cmp r2, #0 ;skip first_pass filter if xoffset=0 beq secondpass_bfilter16x16_only @@ -416,6 +417,7 @@ sub_pixel_variance16x16_neon_loop add sp, sp, #528 vmov.32 r0, d0[0] ;return + vpop {d8-d15} pop {r4-r6,pc} ENDP diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16s_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16s_neon.asm index 155be4fc54b..b0829af7547 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16s_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16s_neon.asm @@ -31,9 +31,10 @@ ;================================================ |vp8_variance_halfpixvar16x16_h_neon| PROC push {lr} + vpush {d8-d15} mov r12, #4 ;loop counter - ldr lr, [sp, #4] ;load *sse from stack + ldr lr, [sp, #68] ;load *sse from stack vmov.i8 q8, #0 ;q8 - sum vmov.i8 q9, #0 ;q9, q10 - sse vmov.i8 q10, #0 @@ -116,6 +117,8 @@ vp8_filt_fpo16x16s_4_0_loop_neon vsub.u32 d0, d1, d10 vmov.32 r0, d0[0] ;return + + vpop {d8-d15} pop {pc} ENDP @@ -131,11 +134,12 @@ vp8_filt_fpo16x16s_4_0_loop_neon ;================================================ |vp8_variance_halfpixvar16x16_v_neon| PROC push {lr} + vpush {d8-d15} mov r12, #4 ;loop counter vld1.u8 {q0}, [r0], r1 ;load src data - ldr lr, [sp, #4] ;load *sse from stack + ldr lr, [sp, #68] ;load *sse from stack vmov.i8 q8, #0 ;q8 - sum vmov.i8 q9, #0 ;q9, q10 - sse @@ -212,6 +216,8 @@ vp8_filt_spo16x16s_0_4_loop_neon vsub.u32 d0, d1, d10 vmov.32 r0, d0[0] ;return + + vpop {d8-d15} pop {pc} ENDP @@ -227,10 +233,11 @@ vp8_filt_spo16x16s_0_4_loop_neon ;================================================ |vp8_variance_halfpixvar16x16_hv_neon| PROC push {lr} + vpush {d8-d15} vld1.u8 {d0, d1, d2, d3}, [r0], r1 ;load src data - ldr lr, [sp, #4] ;load *sse from stack + ldr lr, [sp, #68] ;load *sse from stack vmov.i8 q13, #0 ;q8 - sum vext.8 q1, q0, q1, #1 ;construct src_ptr[1] @@ -331,6 +338,8 @@ vp8_filt16x16s_4_4_loop_neon vsub.u32 d0, d1, d10 vmov.32 r0, d0[0] ;return + + vpop {d8-d15} pop {pc} ENDP @@ -349,10 +358,11 @@ vp8_filt16x16s_4_4_loop_neon |vp8_sub_pixel_variance16x16s_neon| PROC push {r4, lr} + vpush {d8-d15} - ldr r4, [sp, #8] ;load *dst_ptr from stack - ldr r12, [sp, #12] ;load dst_pixels_per_line from stack - ldr lr, [sp, #16] ;load *sse from stack + ldr r4, [sp, #72] ;load *dst_ptr from stack + ldr r12, [sp, #76] ;load dst_pixels_per_line from stack + ldr lr, [sp, #80] ;load *sse from stack cmp r2, #0 ;skip first_pass filter if xoffset=0 beq secondpass_bfilter16x16s_only @@ -566,6 +576,7 @@ sub_pixel_variance16x16s_neon_loop add sp, sp, #256 vmov.32 r0, d0[0] ;return + vpop {d8-d15} pop {r4, pc} ENDP diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance8x8_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance8x8_neon.asm index f6b6847537f..9d9f9e0772a 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance8x8_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/neon/vp8_subpixelvariance8x8_neon.asm @@ -26,11 +26,12 @@ |vp8_sub_pixel_variance8x8_neon| PROC push {r4-r5, lr} + vpush {d8-d15} adr r12, bilinear_taps_coeff - ldr r4, [sp, #12] ;load *dst_ptr from stack - ldr r5, [sp, #16] ;load dst_pixels_per_line from stack - ldr lr, [sp, #20] ;load *sse from stack + ldr r4, [sp, #76] ;load *dst_ptr from stack + ldr r5, [sp, #80] ;load dst_pixels_per_line from stack + ldr lr, [sp, #84] ;load *sse from stack cmp r2, #0 ;skip first_pass filter if xoffset=0 beq skip_firstpass_filter @@ -210,6 +211,8 @@ sub_pixel_variance8x8_neon_loop vsub.u32 d0, d1, d10 vmov.32 r0, d0[0] ;return + + vpop {d8-d15} pop {r4-r5, pc} ENDP diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/reconintra_arm.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/reconintra_arm.c index 2874896e8e2..e55a33cbb18 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/reconintra_arm.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/reconintra_arm.c @@ -14,7 +14,7 @@ #include "vp8/common/blockd.h" #include "vpx_mem/vpx_mem.h" -#if HAVE_NEON +#if HAVE_NEON_ASM extern void vp8_build_intra_predictors_mby_neon_func( unsigned char *y_buffer, unsigned char *ypred_ptr, diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/variance_arm.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/variance_arm.c index 467a509420e..e3f7083b611 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/variance_arm.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/arm/variance_arm.c @@ -95,7 +95,7 @@ unsigned int vp8_sub_pixel_variance16x16_armv6 #endif /* HAVE_MEDIA */ -#if HAVE_NEON +#if HAVE_NEON_ASM extern unsigned int vp8_sub_pixel_variance16x16_neon_func ( diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/blockd.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/blockd.h index f7ff5776352..ea1a6a4adfd 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/blockd.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/blockd.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_BLOCKD_H -#define __INC_BLOCKD_H +#ifndef VP8_COMMON_BLOCKD_H_ +#define VP8_COMMON_BLOCKD_H_ void vpx_log(const char *format, ...); @@ -20,6 +20,10 @@ void vpx_log(const char *format, ...); #include "treecoder.h" #include "vpx_ports/mem.h" +#ifdef __cplusplus +extern "C" { +#endif + /*#define DCPRED 1*/ #define DCPREDSIMTHRESH 0 #define DCPREDCNTTHRESH 3 @@ -297,4 +301,8 @@ typedef struct macroblockd extern void vp8_build_block_doffsets(MACROBLOCKD *x); extern void vp8_setup_block_dptrs(MACROBLOCKD *x); -#endif /* __INC_BLOCKD_H */ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_BLOCKD_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/coefupdateprobs.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/coefupdateprobs.h index 9e194dc9a4d..d96a19e7478 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/coefupdateprobs.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/coefupdateprobs.h @@ -8,6 +8,12 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_COMMON_COEFUPDATEPROBS_H_ +#define VP8_COMMON_COEFUPDATEPROBS_H_ + +#ifdef __cplusplus +extern "C" { +#endif /* Update probabilities for the nodes in the token entropy tree. Generated file included by entropy.c */ @@ -183,3 +189,9 @@ const vp8_prob vp8_coef_update_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTE }, }, }; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_COEFUPDATEPROBS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/common.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/common.h index 2cc1c544cdf..17262d6983c 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/common.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/common.h @@ -9,8 +9,8 @@ */ -#ifndef common_h -#define common_h 1 +#ifndef VP8_COMMON_COMMON_H_ +#define VP8_COMMON_COMMON_H_ #include <assert.h> @@ -18,6 +18,13 @@ #include "vpx_mem/vpx_mem.h" +#ifdef __cplusplus +extern "C" { +#endif + +#define MIN(x, y) (((x) < (y)) ? (x) : (y)) +#define MAX(x, y) (((x) > (y)) ? (x) : (y)) + /* Only need this for fixed-size arrays, for structs just assign. */ #define vp8_copy( Dest, Src) { \ @@ -37,4 +44,8 @@ #define vp8_zero_array( Dest, N) vpx_memset( Dest, 0, N * sizeof( *Dest)); -#endif /* common_h */ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_COMMON_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/default_coef_probs.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/default_coef_probs.h index 0d195636bcb..4d69e4be664 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/default_coef_probs.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/default_coef_probs.h @@ -8,6 +8,12 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_COMMON_DEFAULT_COEF_PROBS_H_ +#define VP8_COMMON_DEFAULT_COEF_PROBS_H_ + +#ifdef __cplusplus +extern "C" { +#endif /*Generated file, included by entropy.c*/ @@ -186,3 +192,9 @@ static const vp8_prob default_coef_probs [BLOCK_TYPES] } } }; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_DEFAULT_COEF_PROBS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/entropy.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/entropy.h index 5389bc1de4e..a90bab4bac2 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/entropy.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/entropy.h @@ -9,12 +9,16 @@ */ -#ifndef __INC_ENTROPY_H -#define __INC_ENTROPY_H +#ifndef VP8_COMMON_ENTROPY_H_ +#define VP8_COMMON_ENTROPY_H_ #include "treecoder.h" #include "blockd.h" +#ifdef __cplusplus +extern "C" { +#endif + /* Coefficient token alphabet */ #define ZERO_TOKEN 0 /* 0 Extra Bits 0+0 */ @@ -98,4 +102,8 @@ extern DECLARE_ALIGNED(16, const short, vp8_default_zig_zag_mask[16]); extern const int vp8_mb_feature_data_bits[MB_LVL_MAX]; void vp8_coef_tree_initialize(void); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_ENTROPY_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymode.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymode.h index 1df0f641e49..81bdfc4b8bd 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymode.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymode.h @@ -9,12 +9,16 @@ */ -#ifndef __INC_ENTROPYMODE_H -#define __INC_ENTROPYMODE_H +#ifndef VP8_COMMON_ENTROPYMODE_H_ +#define VP8_COMMON_ENTROPYMODE_H_ #include "onyxc_int.h" #include "treecoder.h" +#ifdef __cplusplus +extern "C" { +#endif + typedef enum { SUBMVREF_NORMAL, @@ -77,4 +81,8 @@ void vp8_init_mbmode_probs(VP8_COMMON *x); void vp8_default_bmode_probs(vp8_prob dest [VP8_BINTRAMODES-1]); void vp8_kf_default_bmode_probs(vp8_prob dest [VP8_BINTRAMODES] [VP8_BINTRAMODES] [VP8_BINTRAMODES-1]); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_ENTROPYMODE_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymv.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymv.h index 2db1e385bae..42840d58ad2 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymv.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/entropymv.h @@ -9,11 +9,15 @@ */ -#ifndef __INC_ENTROPYMV_H -#define __INC_ENTROPYMV_H +#ifndef VP8_COMMON_ENTROPYMV_H_ +#define VP8_COMMON_ENTROPYMV_H_ #include "treecoder.h" +#ifdef __cplusplus +extern "C" { +#endif + enum { mv_max = 1023, /* max absolute value of a MV component */ @@ -41,4 +45,8 @@ typedef struct mv_context extern const MV_CONTEXT vp8_mv_update_probs[2], vp8_default_mv_context[2]; +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_ENTROPYMV_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/extend.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/extend.h index 74a0b177d94..068f4ac5236 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/extend.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/extend.h @@ -9,11 +9,15 @@ */ -#ifndef __INC_EXTEND_H -#define __INC_EXTEND_H +#ifndef VP8_COMMON_EXTEND_H_ +#define VP8_COMMON_EXTEND_H_ #include "vpx_scale/yv12config.h" +#ifdef __cplusplus +extern "C" { +#endif + void vp8_extend_mb_row(YV12_BUFFER_CONFIG *ybf, unsigned char *YPtr, unsigned char *UPtr, unsigned char *VPtr); void vp8_copy_and_extend_frame(YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst); @@ -22,4 +26,8 @@ void vp8_copy_and_extend_frame_with_rect(YV12_BUFFER_CONFIG *src, int srcy, int srcx, int srch, int srcw); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_EXTEND_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/filter.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/filter.h index ccda7c8d020..cfba775fce4 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/filter.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/filter.h @@ -9,11 +9,15 @@ */ -#ifndef FILTER_H -#define FILTER_H +#ifndef VP8_COMMON_FILTER_H_ +#define VP8_COMMON_FILTER_H_ #include "vpx_ports/mem.h" +#ifdef __cplusplus +extern "C" { +#endif + #define BLOCK_HEIGHT_WIDTH 4 #define VP8_FILTER_WEIGHT 128 #define VP8_FILTER_SHIFT 7 @@ -21,4 +25,8 @@ extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters[8][2]); extern DECLARE_ALIGNED(16, const short, vp8_sub_pel_filters[8][6]); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_FILTER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/findnearmv.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/findnearmv.h index c60e463614b..3c8c0506f67 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/findnearmv.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/findnearmv.h @@ -9,14 +9,18 @@ */ -#ifndef __INC_FINDNEARMV_H -#define __INC_FINDNEARMV_H +#ifndef VP8_COMMON_FINDNEARMV_H_ +#define VP8_COMMON_FINDNEARMV_H_ #include "mv.h" #include "blockd.h" #include "modecont.h" #include "treecoder.h" +#ifdef __cplusplus +extern "C" { +#endif + static void mv_bias(int refmb_ref_frame_sign_bias, int refframe, int_mv *mvp, const int *ref_frame_sign_bias) @@ -179,4 +183,8 @@ static B_PREDICTION_MODE above_block_mode(const MODE_INFO *cur_mb, int b, int mi return (cur_mb->bmi + b - 4)->as_mode; } +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_FINDNEARMV_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/header.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/header.h index 3e98eeb3c34..e27bca16bd7 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/header.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/header.h @@ -9,8 +9,12 @@ */ -#ifndef __INC_HEADER_H -#define __INC_HEADER_H +#ifndef VP8_COMMON_HEADER_H_ +#define VP8_COMMON_HEADER_H_ + +#ifdef __cplusplus +extern "C" { +#endif /* 24 bits total */ typedef struct @@ -40,4 +44,8 @@ typedef struct #endif +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_HEADER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/invtrans.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/invtrans.h index 9262640d570..affe57e3d6e 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/invtrans.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/invtrans.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_INVTRANS_H -#define __INC_INVTRANS_H +#ifndef VP8_COMMON_INVTRANS_H_ +#define VP8_COMMON_INVTRANS_H_ #include "vpx_config.h" #include "vp8_rtcd.h" @@ -21,6 +21,10 @@ #include "vpx_mem/vpx_mem.h" #endif +#ifdef __cplusplus +extern "C" { +#endif + static void eob_adjust(char *eobs, short *diff) { /* eob adjust.... the idct can only skip if both the dc and eob are zero */ @@ -59,4 +63,8 @@ static void vp8_inverse_transform_mby(MACROBLOCKD *xd) xd->dst.y_buffer, xd->dst.y_stride, xd->eobs); } +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_INVTRANS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.c index 19857a7e9fb..7a07e76fc41 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.c @@ -15,7 +15,6 @@ #include "onyxc_int.h" #include "vpx_mem/vpx_mem.h" -typedef unsigned char uc; static void lf_init_lut(loop_filter_info_n *lfi) { diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.h index 1e47f349073..20a6bd375b6 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/loopfilter.h @@ -9,13 +9,17 @@ */ -#ifndef loopfilter_h -#define loopfilter_h +#ifndef VP8_COMMON_LOOPFILTER_H_ +#define VP8_COMMON_LOOPFILTER_H_ #include "vpx_ports/mem.h" #include "vpx_config.h" #include "vp8_rtcd.h" +#ifdef __cplusplus +extern "C" { +#endif + #define MAX_LOOP_FILTER 63 /* fraction of total macroblock rows to be used in fast filter level picking */ /* has to be > 2 */ @@ -102,4 +106,8 @@ void vp8_loop_filter_row_simple(struct VP8Common *cm, int mb_row, int post_ystride, int post_uvstride, unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_LOOPFILTER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/modecont.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/modecont.h index 24db88295f9..ff34c33c557 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/modecont.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/modecont.h @@ -9,9 +9,17 @@ */ -#ifndef __INC_MODECONT_H -#define __INC_MODECONT_H +#ifndef VP8_COMMON_MODECONT_H_ +#define VP8_COMMON_MODECONT_H_ + +#ifdef __cplusplus +extern "C" { +#endif extern const int vp8_mode_contexts[6][4]; +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_MODECONT_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/mv.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/mv.h index b3f919db2bd..111ccd63c72 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/mv.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/mv.h @@ -9,10 +9,14 @@ */ -#ifndef __INC_MV_H -#define __INC_MV_H +#ifndef VP8_COMMON_MV_H_ +#define VP8_COMMON_MV_H_ #include "vpx/vpx_integer.h" +#ifdef __cplusplus +extern "C" { +#endif + typedef struct { short row; @@ -25,4 +29,8 @@ typedef union int_mv MV as_mv; } int_mv; /* facilitates faster equality tests and copies */ +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_MV_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/onyx.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/onyx.h index 30c4cbbca63..119e40cdc9c 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/onyx.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/onyx.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_VP8_H -#define __INC_VP8_H +#ifndef VP8_COMMON_ONYX_H_ +#define VP8_COMMON_ONYX_H_ #ifdef __cplusplus extern "C" @@ -39,8 +39,8 @@ extern "C" typedef enum { - USAGE_STREAM_FROM_SERVER = 0x0, - USAGE_LOCAL_FILE_PLAYBACK = 0x1, + USAGE_LOCAL_FILE_PLAYBACK = 0x0, + USAGE_STREAM_FROM_SERVER = 0x1, USAGE_CONSTRAINED_QUALITY = 0x2, USAGE_CONSTANT_QUALITY = 0x3 } END_USAGE; @@ -267,4 +267,4 @@ extern "C" } #endif -#endif +#endif // VP8_COMMON_ONYX_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxc_int.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxc_int.h index e9bb7af26b1..6d89865c600 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxc_int.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxc_int.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_VP8C_INT_H -#define __INC_VP8C_INT_H +#ifndef VP8_COMMON_ONYXC_INT_H_ +#define VP8_COMMON_ONYXC_INT_H_ #include "vpx_config.h" #include "vp8_rtcd.h" @@ -26,6 +26,10 @@ #include "header.h" /*#endif*/ +#ifdef __cplusplus +extern "C" { +#endif + #define MINQ 0 #define MAXQ 127 #define QINDEX_RANGE (MAXQ + 1) @@ -174,4 +178,8 @@ typedef struct VP8Common int cpu_caps; } VP8_COMMON; +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_ONYXC_INT_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxd.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxd.h index 97c81c130a0..e37b29f32cf 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxd.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/onyxd.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_VP8D_H -#define __INC_VP8D_H +#ifndef VP8_COMMON_ONYXD_H_ +#define VP8_COMMON_ONYXD_H_ /* Create/destroy static data structures. */ @@ -60,4 +60,4 @@ extern "C" #endif -#endif +#endif // VP8_COMMON_ONYXD_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.c index dd998f16e8a..8e546d5bfbe 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.c @@ -71,11 +71,6 @@ static const unsigned char MV_REFERENCE_FRAME_colors[MAX_REF_FRAMES][3] = }; #endif -static const short kernel5[] = -{ - 1, 1, 4, 1, 1 -}; - const short vp8_rv[] = { 8, 5, 2, 2, 8, 12, 4, 9, 8, 3, @@ -308,13 +303,14 @@ void vp8_mbpost_proc_down_c(unsigned char *dst, int pitch, int rows, int cols, i { d[r&15] = (rv2[r&127] + sum + s[0]) >> 4; } - - s[-8*pitch] = d[(r-8)&15]; + if (r >= 8) + s[-8*pitch] = d[(r-8)&15]; s += pitch; } } } +#if CONFIG_POSTPROC static void vp8_de_mblock(YV12_BUFFER_CONFIG *post, int q) { @@ -387,6 +383,7 @@ void vp8_deblock(VP8_COMMON *cm, vp8_yv12_copy_frame(source, post); } } +#endif #if !(CONFIG_TEMPORAL_DENOISING) void vp8_de_noise(VP8_COMMON *cm, diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.h index 495a2c906f6..33d0a7f025b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/postproc.h @@ -9,8 +9,8 @@ */ -#ifndef POSTPROC_H -#define POSTPROC_H +#ifndef VP8_COMMON_POSTPROC_H_ +#define VP8_COMMON_POSTPROC_H_ #include "vpx_ports/mem.h" struct postproc_state @@ -26,6 +26,10 @@ struct postproc_state }; #include "onyxc_int.h" #include "ppflags.h" + +#ifdef __cplusplus +extern "C" { +#endif int vp8_post_proc_frame(struct VP8Common *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t *flags); @@ -47,4 +51,8 @@ void vp8_deblock(struct VP8Common *oci, #define MFQE_PRECISION 4 void vp8_multiframe_quality_enhance(struct VP8Common *cm); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_POSTPROC_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/ppflags.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/ppflags.h index 665e21fd965..768224aad5e 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/ppflags.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/ppflags.h @@ -9,8 +9,12 @@ */ -#ifndef __INC_PPFLAGS_H -#define __INC_PPFLAGS_H +#ifndef VP8_COMMON_PPFLAGS_H_ +#define VP8_COMMON_PPFLAGS_H_ + +#ifdef __cplusplus +extern "C" { +#endif enum { VP8D_NOFILTERING = 0, @@ -38,4 +42,8 @@ typedef struct int display_mv_flag; } vp8_ppflags_t; +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_PPFLAGS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/pragmas.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/pragmas.h index 99fee5ae232..329cc8275c2 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/pragmas.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/pragmas.h @@ -8,8 +8,12 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_COMMON_PRAGMAS_H_ +#define VP8_COMMON_PRAGMAS_H_ - +#ifdef __cplusplus +extern "C" { +#endif #ifdef __INTEL_COMPILER #pragma warning(disable:997 1011 170) @@ -17,3 +21,9 @@ #ifdef _MSC_VER #pragma warning(disable:4799) #endif + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_PRAGMAS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/quant_common.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/quant_common.h index cb64d8eb8d8..700b5e6d726 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/quant_common.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/quant_common.h @@ -8,14 +8,27 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_COMMON_QUANT_COMMON_H_ +#define VP8_COMMON_QUANT_COMMON_H_ + #include "string.h" #include "blockd.h" #include "onyxc_int.h" +#ifdef __cplusplus +extern "C" { +#endif + extern int vp8_ac_yquant(int QIndex); extern int vp8_dc_quant(int QIndex, int Delta); extern int vp8_dc2quant(int QIndex, int Delta); extern int vp8_ac2quant(int QIndex, int Delta); extern int vp8_dc_uv_quant(int QIndex, int Delta); extern int vp8_ac_uv_quant(int QIndex, int Delta); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_QUANT_COMMON_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/reconinter.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/reconinter.h index 233c02e5ba5..ba979b9664a 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/reconinter.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/reconinter.h @@ -9,8 +9,12 @@ */ -#ifndef __INC_RECONINTER_H -#define __INC_RECONINTER_H +#ifndef VP8_COMMON_RECONINTER_H_ +#define VP8_COMMON_RECONINTER_H_ + +#ifdef __cplusplus +extern "C" { +#endif extern void vp8_build_inter_predictors_mb(MACROBLOCKD *x); extern void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x, @@ -32,4 +36,8 @@ extern void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, extern void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x); extern void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_RECONINTER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/reconintra4x4.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/reconintra4x4.h index d2b0d43461e..ed59c9edd4c 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/reconintra4x4.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/reconintra4x4.h @@ -9,10 +9,14 @@ */ -#ifndef __INC_RECONINTRA4x4_H -#define __INC_RECONINTRA4x4_H +#ifndef VP8_COMMON_RECONINTRA4X4_H_ +#define VP8_COMMON_RECONINTRA4X4_H_ #include "vp8/common/blockd.h" +#ifdef __cplusplus +extern "C" { +#endif + static void intra_prediction_down_copy(MACROBLOCKD *xd, unsigned char *above_right_src) { @@ -29,4 +33,8 @@ static void intra_prediction_down_copy(MACROBLOCKD *xd, *dst_ptr2 = *src_ptr; } +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_RECONINTRA4X4_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.pl b/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.pl new file mode 100644 index 00000000000..cbfd76a8d16 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.pl @@ -0,0 +1,561 @@ +sub vp8_common_forward_decls() { +print <<EOF +/* + * VP8 + */ + +struct blockd; +struct macroblockd; +struct loop_filter_info; + +/* Encoder forward decls */ +struct block; +struct macroblock; +struct variance_vtable; +union int_mv; +struct yv12_buffer_config; +EOF +} +forward_decls qw/vp8_common_forward_decls/; + +# +# system state +# +add_proto qw/void vp8_clear_system_state/, ""; +specialize qw/vp8_clear_system_state mmx/; +$vp8_clear_system_state_mmx=vpx_reset_mmx_state; + +# +# Dequant +# +add_proto qw/void vp8_dequantize_b/, "struct blockd*, short *dqc"; +specialize qw/vp8_dequantize_b mmx media neon/; +$vp8_dequantize_b_media=vp8_dequantize_b_v6; + +add_proto qw/void vp8_dequant_idct_add/, "short *input, short *dq, unsigned char *output, int stride"; +specialize qw/vp8_dequant_idct_add mmx media neon dspr2/; +$vp8_dequant_idct_add_media=vp8_dequant_idct_add_v6; +$vp8_dequant_idct_add_dspr2=vp8_dequant_idct_add_dspr2; + +add_proto qw/void vp8_dequant_idct_add_y_block/, "short *q, short *dq, unsigned char *dst, int stride, char *eobs"; +specialize qw/vp8_dequant_idct_add_y_block mmx sse2 media neon_asm dspr2/; +$vp8_dequant_idct_add_y_block_media=vp8_dequant_idct_add_y_block_v6; +$vp8_dequant_idct_add_y_block_neon_asm=vp8_dequant_idct_add_y_block_neon; +$vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2; + +add_proto qw/void vp8_dequant_idct_add_uv_block/, "short *q, short *dq, unsigned char *dst_u, unsigned char *dst_v, int stride, char *eobs"; +specialize qw/vp8_dequant_idct_add_uv_block mmx sse2 media neon_asm dspr2/; +$vp8_dequant_idct_add_uv_block_media=vp8_dequant_idct_add_uv_block_v6; +$vp8_dequant_idct_add_uv_block_neon_asm=vp8_dequant_idct_add_uv_block_neon; +$vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2; + +# +# Loopfilter +# +add_proto qw/void vp8_loop_filter_mbv/, "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"; +specialize qw/vp8_loop_filter_mbv mmx sse2 media neon dspr2/; +$vp8_loop_filter_mbv_media=vp8_loop_filter_mbv_armv6; +$vp8_loop_filter_mbv_dspr2=vp8_loop_filter_mbv_dspr2; + +add_proto qw/void vp8_loop_filter_bv/, "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"; +specialize qw/vp8_loop_filter_bv mmx sse2 media neon_asm dspr2/; +$vp8_loop_filter_bv_media=vp8_loop_filter_bv_armv6; +$vp8_loop_filter_bv_neon_asm=vp8_loop_filter_bv_neon; +$vp8_loop_filter_bv_dspr2=vp8_loop_filter_bv_dspr2; + +add_proto qw/void vp8_loop_filter_mbh/, "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"; +specialize qw/vp8_loop_filter_mbh mmx sse2 media neon dspr2/; +$vp8_loop_filter_mbh_media=vp8_loop_filter_mbh_armv6; +$vp8_loop_filter_mbh_dspr2=vp8_loop_filter_mbh_dspr2; + +add_proto qw/void vp8_loop_filter_bh/, "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"; +specialize qw/vp8_loop_filter_bh mmx sse2 media neon_asm dspr2/; +$vp8_loop_filter_bh_media=vp8_loop_filter_bh_armv6; +$vp8_loop_filter_bh_neon_asm=vp8_loop_filter_bh_neon; +$vp8_loop_filter_bh_dspr2=vp8_loop_filter_bh_dspr2; + + +add_proto qw/void vp8_loop_filter_simple_mbv/, "unsigned char *y, int ystride, const unsigned char *blimit"; +specialize qw/vp8_loop_filter_simple_mbv mmx sse2 media neon_asm/; +$vp8_loop_filter_simple_mbv_c=vp8_loop_filter_simple_vertical_edge_c; +$vp8_loop_filter_simple_mbv_mmx=vp8_loop_filter_simple_vertical_edge_mmx; +$vp8_loop_filter_simple_mbv_sse2=vp8_loop_filter_simple_vertical_edge_sse2; +$vp8_loop_filter_simple_mbv_media=vp8_loop_filter_simple_vertical_edge_armv6; +$vp8_loop_filter_simple_mbv_neon_asm=vp8_loop_filter_mbvs_neon; + +add_proto qw/void vp8_loop_filter_simple_mbh/, "unsigned char *y, int ystride, const unsigned char *blimit"; +specialize qw/vp8_loop_filter_simple_mbh mmx sse2 media neon/; +$vp8_loop_filter_simple_mbh_c=vp8_loop_filter_simple_horizontal_edge_c; +$vp8_loop_filter_simple_mbh_mmx=vp8_loop_filter_simple_horizontal_edge_mmx; +$vp8_loop_filter_simple_mbh_sse2=vp8_loop_filter_simple_horizontal_edge_sse2; +$vp8_loop_filter_simple_mbh_media=vp8_loop_filter_simple_horizontal_edge_armv6; +$vp8_loop_filter_simple_mbh_neon=vp8_loop_filter_mbhs_neon; + +add_proto qw/void vp8_loop_filter_simple_bv/, "unsigned char *y, int ystride, const unsigned char *blimit"; +specialize qw/vp8_loop_filter_simple_bv mmx sse2 media neon_asm/; +$vp8_loop_filter_simple_bv_c=vp8_loop_filter_bvs_c; +$vp8_loop_filter_simple_bv_mmx=vp8_loop_filter_bvs_mmx; +$vp8_loop_filter_simple_bv_sse2=vp8_loop_filter_bvs_sse2; +$vp8_loop_filter_simple_bv_media=vp8_loop_filter_bvs_armv6; +$vp8_loop_filter_simple_bv_neon_asm=vp8_loop_filter_bvs_neon; + +add_proto qw/void vp8_loop_filter_simple_bh/, "unsigned char *y, int ystride, const unsigned char *blimit"; +specialize qw/vp8_loop_filter_simple_bh mmx sse2 media neon/; +$vp8_loop_filter_simple_bh_c=vp8_loop_filter_bhs_c; +$vp8_loop_filter_simple_bh_mmx=vp8_loop_filter_bhs_mmx; +$vp8_loop_filter_simple_bh_sse2=vp8_loop_filter_bhs_sse2; +$vp8_loop_filter_simple_bh_media=vp8_loop_filter_bhs_armv6; +$vp8_loop_filter_simple_bh_neon=vp8_loop_filter_bhs_neon; + +# +# IDCT +# +#idct16 +add_proto qw/void vp8_short_idct4x4llm/, "short *input, unsigned char *pred, int pitch, unsigned char *dst, int dst_stride"; +specialize qw/vp8_short_idct4x4llm mmx media neon dspr2/; +$vp8_short_idct4x4llm_media=vp8_short_idct4x4llm_v6_dual; +$vp8_short_idct4x4llm_dspr2=vp8_short_idct4x4llm_dspr2; + +#iwalsh1 +add_proto qw/void vp8_short_inv_walsh4x4_1/, "short *input, short *output"; +specialize qw/vp8_short_inv_walsh4x4_1 dspr2/; +$vp8_short_inv_walsh4x4_1_dspr2=vp8_short_inv_walsh4x4_1_dspr2; +# no asm yet + +#iwalsh16 +add_proto qw/void vp8_short_inv_walsh4x4/, "short *input, short *output"; +specialize qw/vp8_short_inv_walsh4x4 mmx sse2 media neon dspr2/; +$vp8_short_inv_walsh4x4_media=vp8_short_inv_walsh4x4_v6; +$vp8_short_inv_walsh4x4_dspr2=vp8_short_inv_walsh4x4_dspr2; + +#idct1_scalar_add +add_proto qw/void vp8_dc_only_idct_add/, "short input, unsigned char *pred, int pred_stride, unsigned char *dst, int dst_stride"; +specialize qw/vp8_dc_only_idct_add mmx media neon dspr2/; +$vp8_dc_only_idct_add_media=vp8_dc_only_idct_add_v6; +$vp8_dc_only_idct_add_dspr2=vp8_dc_only_idct_add_dspr2; + +# +# RECON +# +add_proto qw/void vp8_copy_mem16x16/, "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_copy_mem16x16 mmx sse2 media neon dspr2/; +$vp8_copy_mem16x16_media=vp8_copy_mem16x16_v6; +$vp8_copy_mem16x16_dspr2=vp8_copy_mem16x16_dspr2; + +add_proto qw/void vp8_copy_mem8x8/, "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_copy_mem8x8 mmx media neon dspr2/; +$vp8_copy_mem8x8_media=vp8_copy_mem8x8_v6; +$vp8_copy_mem8x8_dspr2=vp8_copy_mem8x8_dspr2; + +add_proto qw/void vp8_copy_mem8x4/, "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_copy_mem8x4 mmx media neon dspr2/; +$vp8_copy_mem8x4_media=vp8_copy_mem8x4_v6; +$vp8_copy_mem8x4_dspr2=vp8_copy_mem8x4_dspr2; + +add_proto qw/void vp8_build_intra_predictors_mby_s/, "struct macroblockd *x, unsigned char * yabove_row, unsigned char * yleft, int left_stride, unsigned char * ypred_ptr, int y_stride"; +specialize qw/vp8_build_intra_predictors_mby_s sse2 ssse3/; +#TODO: fix assembly for neon + +add_proto qw/void vp8_build_intra_predictors_mbuv_s/, "struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row, unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride"; +specialize qw/vp8_build_intra_predictors_mbuv_s sse2 ssse3/; + +add_proto qw/void vp8_intra4x4_predict/, "unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left"; +specialize qw/vp8_intra4x4_predict media/; +$vp8_intra4x4_predict_media=vp8_intra4x4_predict_armv6; + +# +# Postproc +# +if (vpx_config("CONFIG_POSTPROC") eq "yes") { + add_proto qw/void vp8_mbpost_proc_down/, "unsigned char *dst, int pitch, int rows, int cols,int flimit"; + specialize qw/vp8_mbpost_proc_down mmx sse2/; + $vp8_mbpost_proc_down_sse2=vp8_mbpost_proc_down_xmm; + + add_proto qw/void vp8_mbpost_proc_across_ip/, "unsigned char *dst, int pitch, int rows, int cols,int flimit"; + specialize qw/vp8_mbpost_proc_across_ip sse2/; + $vp8_mbpost_proc_across_ip_sse2=vp8_mbpost_proc_across_ip_xmm; + + add_proto qw/void vp8_post_proc_down_and_across_mb_row/, "unsigned char *src, unsigned char *dst, int src_pitch, int dst_pitch, int cols, unsigned char *flimits, int size"; + specialize qw/vp8_post_proc_down_and_across_mb_row sse2/; + + add_proto qw/void vp8_plane_add_noise/, "unsigned char *s, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int w, unsigned int h, int pitch"; + specialize qw/vp8_plane_add_noise mmx sse2/; + $vp8_plane_add_noise_sse2=vp8_plane_add_noise_wmt; + + add_proto qw/void vp8_blend_mb_inner/, "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride"; + # no asm yet + + add_proto qw/void vp8_blend_mb_outer/, "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride"; + # no asm yet + + add_proto qw/void vp8_blend_b/, "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride"; + # no asm yet + + add_proto qw/void vp8_filter_by_weight16x16/, "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight"; + specialize qw/vp8_filter_by_weight16x16 sse2/; + + add_proto qw/void vp8_filter_by_weight8x8/, "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight"; + specialize qw/vp8_filter_by_weight8x8 sse2/; + + add_proto qw/void vp8_filter_by_weight4x4/, "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight"; + # no asm yet +} + +# +# Subpixel +# +add_proto qw/void vp8_sixtap_predict16x16/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_sixtap_predict16x16 mmx sse2 ssse3 media neon dspr2/; +$vp8_sixtap_predict16x16_media=vp8_sixtap_predict16x16_armv6; +$vp8_sixtap_predict16x16_dspr2=vp8_sixtap_predict16x16_dspr2; + +add_proto qw/void vp8_sixtap_predict8x8/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_sixtap_predict8x8 mmx sse2 ssse3 media neon dspr2/; +$vp8_sixtap_predict8x8_media=vp8_sixtap_predict8x8_armv6; +$vp8_sixtap_predict8x8_dspr2=vp8_sixtap_predict8x8_dspr2; + +add_proto qw/void vp8_sixtap_predict8x4/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_sixtap_predict8x4 mmx sse2 ssse3 media neon dspr2/; +$vp8_sixtap_predict8x4_media=vp8_sixtap_predict8x4_armv6; +$vp8_sixtap_predict8x4_dspr2=vp8_sixtap_predict8x4_dspr2; + +add_proto qw/void vp8_sixtap_predict4x4/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_sixtap_predict4x4 mmx ssse3 media neon dspr2/; +$vp8_sixtap_predict4x4_media=vp8_sixtap_predict4x4_armv6; +$vp8_sixtap_predict4x4_dspr2=vp8_sixtap_predict4x4_dspr2; + +add_proto qw/void vp8_bilinear_predict16x16/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_bilinear_predict16x16 mmx sse2 ssse3 media neon/; +$vp8_bilinear_predict16x16_media=vp8_bilinear_predict16x16_armv6; + +add_proto qw/void vp8_bilinear_predict8x8/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_bilinear_predict8x8 mmx sse2 ssse3 media neon/; +$vp8_bilinear_predict8x8_media=vp8_bilinear_predict8x8_armv6; + +add_proto qw/void vp8_bilinear_predict8x4/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_bilinear_predict8x4 mmx media neon/; +$vp8_bilinear_predict8x4_media=vp8_bilinear_predict8x4_armv6; + +add_proto qw/void vp8_bilinear_predict4x4/, "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"; +specialize qw/vp8_bilinear_predict4x4 mmx media neon/; +$vp8_bilinear_predict4x4_media=vp8_bilinear_predict4x4_armv6; + +# +# Whole-pixel Variance +# +add_proto qw/unsigned int vp8_variance4x4/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance4x4 mmx sse2/; +$vp8_variance4x4_sse2=vp8_variance4x4_wmt; + +add_proto qw/unsigned int vp8_variance8x8/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance8x8 mmx sse2 media neon/; +$vp8_variance8x8_sse2=vp8_variance8x8_wmt; +$vp8_variance8x8_media=vp8_variance8x8_armv6; + +add_proto qw/unsigned int vp8_variance8x16/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance8x16 mmx sse2 neon/; +$vp8_variance8x16_sse2=vp8_variance8x16_wmt; + +add_proto qw/unsigned int vp8_variance16x8/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance16x8 mmx sse2 neon/; +$vp8_variance16x8_sse2=vp8_variance16x8_wmt; + +add_proto qw/unsigned int vp8_variance16x16/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance16x16 mmx sse2 media neon/; +$vp8_variance16x16_sse2=vp8_variance16x16_wmt; +$vp8_variance16x16_media=vp8_variance16x16_armv6; + +# +# Sub-pixel Variance +# +add_proto qw/unsigned int vp8_sub_pixel_variance4x4/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"; +specialize qw/vp8_sub_pixel_variance4x4 mmx sse2/; +$vp8_sub_pixel_variance4x4_sse2=vp8_sub_pixel_variance4x4_wmt; + +add_proto qw/unsigned int vp8_sub_pixel_variance8x8/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"; +specialize qw/vp8_sub_pixel_variance8x8 mmx sse2 media neon_asm/; +$vp8_sub_pixel_variance8x8_sse2=vp8_sub_pixel_variance8x8_wmt; +$vp8_sub_pixel_variance8x8_media=vp8_sub_pixel_variance8x8_armv6; +$vp8_sub_pixel_variance8x8_neon_asm=vp8_sub_pixel_variance8x8_neon; + +add_proto qw/unsigned int vp8_sub_pixel_variance8x16/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"; +specialize qw/vp8_sub_pixel_variance8x16 mmx sse2/; +$vp8_sub_pixel_variance8x16_sse2=vp8_sub_pixel_variance8x16_wmt; + +add_proto qw/unsigned int vp8_sub_pixel_variance16x8/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"; +specialize qw/vp8_sub_pixel_variance16x8 mmx sse2 ssse3/; +$vp8_sub_pixel_variance16x8_sse2=vp8_sub_pixel_variance16x8_wmt; + +add_proto qw/unsigned int vp8_sub_pixel_variance16x16/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"; +specialize qw/vp8_sub_pixel_variance16x16 mmx sse2 ssse3 media neon_asm/; +$vp8_sub_pixel_variance16x16_sse2=vp8_sub_pixel_variance16x16_wmt; +$vp8_sub_pixel_variance16x16_media=vp8_sub_pixel_variance16x16_armv6; +$vp8_sub_pixel_variance16x16_neon_asm=vp8_sub_pixel_variance16x16_neon; + +add_proto qw/unsigned int vp8_variance_halfpixvar16x16_h/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance_halfpixvar16x16_h mmx sse2 media neon_asm/; +$vp8_variance_halfpixvar16x16_h_sse2=vp8_variance_halfpixvar16x16_h_wmt; +$vp8_variance_halfpixvar16x16_h_media=vp8_variance_halfpixvar16x16_h_armv6; +$vp8_variance_halfpixvar16x16_h_neon_asm=vp8_variance_halfpixvar16x16_h_neon; + +add_proto qw/unsigned int vp8_variance_halfpixvar16x16_v/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance_halfpixvar16x16_v mmx sse2 media neon_asm/; +$vp8_variance_halfpixvar16x16_v_sse2=vp8_variance_halfpixvar16x16_v_wmt; +$vp8_variance_halfpixvar16x16_v_media=vp8_variance_halfpixvar16x16_v_armv6; +$vp8_variance_halfpixvar16x16_v_neon_asm=vp8_variance_halfpixvar16x16_v_neon; + +add_proto qw/unsigned int vp8_variance_halfpixvar16x16_hv/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_variance_halfpixvar16x16_hv mmx sse2 media neon_asm/; +$vp8_variance_halfpixvar16x16_hv_sse2=vp8_variance_halfpixvar16x16_hv_wmt; +$vp8_variance_halfpixvar16x16_hv_media=vp8_variance_halfpixvar16x16_hv_armv6; +$vp8_variance_halfpixvar16x16_hv_neon_asm=vp8_variance_halfpixvar16x16_hv_neon; + +# +# Single block SAD +# +add_proto qw/unsigned int vp8_sad4x4/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"; +specialize qw/vp8_sad4x4 mmx sse2 neon/; +$vp8_sad4x4_sse2=vp8_sad4x4_wmt; + +add_proto qw/unsigned int vp8_sad8x8/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"; +specialize qw/vp8_sad8x8 mmx sse2 neon/; +$vp8_sad8x8_sse2=vp8_sad8x8_wmt; + +add_proto qw/unsigned int vp8_sad8x16/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"; +specialize qw/vp8_sad8x16 mmx sse2 neon/; +$vp8_sad8x16_sse2=vp8_sad8x16_wmt; + +add_proto qw/unsigned int vp8_sad16x8/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"; +specialize qw/vp8_sad16x8 mmx sse2 neon/; +$vp8_sad16x8_sse2=vp8_sad16x8_wmt; + +add_proto qw/unsigned int vp8_sad16x16/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"; +specialize qw/vp8_sad16x16 mmx sse2 sse3 media neon/; +$vp8_sad16x16_sse2=vp8_sad16x16_wmt; +$vp8_sad16x16_media=vp8_sad16x16_armv6; + +# +# Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally +# +add_proto qw/void vp8_sad4x4x3/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad4x4x3 sse3/; + +add_proto qw/void vp8_sad8x8x3/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad8x8x3 sse3/; + +add_proto qw/void vp8_sad8x16x3/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad8x16x3 sse3/; + +add_proto qw/void vp8_sad16x8x3/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad16x8x3 sse3 ssse3/; + +add_proto qw/void vp8_sad16x16x3/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad16x16x3 sse3 ssse3/; + +# Note the only difference in the following prototypes is that they return into +# an array of short +add_proto qw/void vp8_sad4x4x8/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array"; +specialize qw/vp8_sad4x4x8 sse4_1/; +$vp8_sad4x4x8_sse4_1=vp8_sad4x4x8_sse4; + +add_proto qw/void vp8_sad8x8x8/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array"; +specialize qw/vp8_sad8x8x8 sse4_1/; +$vp8_sad8x8x8_sse4_1=vp8_sad8x8x8_sse4; + +add_proto qw/void vp8_sad8x16x8/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array"; +specialize qw/vp8_sad8x16x8 sse4_1/; +$vp8_sad8x16x8_sse4_1=vp8_sad8x16x8_sse4; + +add_proto qw/void vp8_sad16x8x8/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array"; +specialize qw/vp8_sad16x8x8 sse4_1/; +$vp8_sad16x8x8_sse4_1=vp8_sad16x8x8_sse4; + +add_proto qw/void vp8_sad16x16x8/, "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array"; +specialize qw/vp8_sad16x16x8 sse4_1/; +$vp8_sad16x16x8_sse4_1=vp8_sad16x16x8_sse4; + +# +# Multi-block SAD, comparing a reference to N independent blocks +# +add_proto qw/void vp8_sad4x4x4d/, "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad4x4x4d sse3/; + +add_proto qw/void vp8_sad8x8x4d/, "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad8x8x4d sse3/; + +add_proto qw/void vp8_sad8x16x4d/, "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad8x16x4d sse3/; + +add_proto qw/void vp8_sad16x8x4d/, "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad16x8x4d sse3/; + +add_proto qw/void vp8_sad16x16x4d/, "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array"; +specialize qw/vp8_sad16x16x4d sse3/; + +# +# Encoder functions below this point. +# +if (vpx_config("CONFIG_VP8_ENCODER") eq "yes") { + +# +# Sum of squares (vector) +# +add_proto qw/unsigned int vp8_get_mb_ss/, "const short *"; +specialize qw/vp8_get_mb_ss mmx sse2/; + +# +# SSE (Sum Squared Error) +# +add_proto qw/unsigned int vp8_sub_pixel_mse16x16/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"; +specialize qw/vp8_sub_pixel_mse16x16 mmx sse2/; +$vp8_sub_pixel_mse16x16_sse2=vp8_sub_pixel_mse16x16_wmt; + +add_proto qw/unsigned int vp8_mse16x16/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse"; +specialize qw/vp8_mse16x16 mmx sse2 media neon_asm/; +$vp8_mse16x16_sse2=vp8_mse16x16_wmt; +$vp8_mse16x16_media=vp8_mse16x16_armv6; +$vp8_mse16x16_neon_asm=vp8_mse16x16_neon; + +add_proto qw/unsigned int vp8_get4x4sse_cs/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride"; +specialize qw/vp8_get4x4sse_cs mmx neon_asm/; +$vp8_get4x4sse_cs_neon_asm=vp8_get4x4sse_cs_neon; + +# +# Block copy +# +if ($opts{arch} =~ /x86/) { + add_proto qw/void vp8_copy32xn/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, int n"; + specialize qw/vp8_copy32xn sse2 sse3/; +} + +# +# Structured Similarity (SSIM) +# +if (vpx_config("CONFIG_INTERNAL_STATS") eq "yes") { + $opts{arch} eq "x86_64" and $sse2_on_x86_64 = "sse2"; + + add_proto qw/void vp8_ssim_parms_8x8/, "unsigned char *s, int sp, unsigned char *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr"; + specialize qw/vp8_ssim_parms_8x8/, "$sse2_on_x86_64"; + + add_proto qw/void vp8_ssim_parms_16x16/, "unsigned char *s, int sp, unsigned char *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr"; + specialize qw/vp8_ssim_parms_16x16/, "$sse2_on_x86_64"; +} + +# +# Forward DCT +# +add_proto qw/void vp8_short_fdct4x4/, "short *input, short *output, int pitch"; +specialize qw/vp8_short_fdct4x4 mmx sse2 media neon_asm/; +$vp8_short_fdct4x4_media=vp8_short_fdct4x4_armv6; +$vp8_short_fdct4x4_neon_asm=vp8_short_fdct4x4_neon; + +add_proto qw/void vp8_short_fdct8x4/, "short *input, short *output, int pitch"; +specialize qw/vp8_short_fdct8x4 mmx sse2 media neon_asm/; +$vp8_short_fdct8x4_media=vp8_short_fdct8x4_armv6; +$vp8_short_fdct8x4_neon_asm=vp8_short_fdct8x4_neon; + +add_proto qw/void vp8_short_walsh4x4/, "short *input, short *output, int pitch"; +specialize qw/vp8_short_walsh4x4 sse2 media neon_asm/; +$vp8_short_walsh4x4_media=vp8_short_walsh4x4_armv6; +$vp8_short_walsh4x4_neon_asm=vp8_short_walsh4x4_neon; + +# +# Quantizer +# +add_proto qw/void vp8_regular_quantize_b/, "struct block *, struct blockd *"; +specialize qw/vp8_regular_quantize_b sse2/; +# TODO(johann) Update sse4 implementation and re-enable +#$vp8_regular_quantize_b_sse4_1=vp8_regular_quantize_b_sse4; + +add_proto qw/void vp8_fast_quantize_b/, "struct block *, struct blockd *"; +specialize qw/vp8_fast_quantize_b sse2 ssse3 media neon_asm/; +$vp8_fast_quantize_b_media=vp8_fast_quantize_b_armv6; +$vp8_fast_quantize_b_neon_asm=vp8_fast_quantize_b_neon; + +add_proto qw/void vp8_regular_quantize_b_pair/, "struct block *b1, struct block *b2, struct blockd *d1, struct blockd *d2"; +# no asm yet + +add_proto qw/void vp8_fast_quantize_b_pair/, "struct block *b1, struct block *b2, struct blockd *d1, struct blockd *d2"; +specialize qw/vp8_fast_quantize_b_pair neon_asm/; +$vp8_fast_quantize_b_pair_neon_asm=vp8_fast_quantize_b_pair_neon; + +add_proto qw/void vp8_quantize_mb/, "struct macroblock *"; +specialize qw/vp8_quantize_mb neon/; + +add_proto qw/void vp8_quantize_mby/, "struct macroblock *"; +specialize qw/vp8_quantize_mby neon/; + +add_proto qw/void vp8_quantize_mbuv/, "struct macroblock *"; +specialize qw/vp8_quantize_mbuv neon/; + +# +# Block subtraction +# +add_proto qw/int vp8_block_error/, "short *coeff, short *dqcoeff"; +specialize qw/vp8_block_error mmx sse2/; +$vp8_block_error_sse2=vp8_block_error_xmm; + +add_proto qw/int vp8_mbblock_error/, "struct macroblock *mb, int dc"; +specialize qw/vp8_mbblock_error mmx sse2/; +$vp8_mbblock_error_sse2=vp8_mbblock_error_xmm; + +add_proto qw/int vp8_mbuverror/, "struct macroblock *mb"; +specialize qw/vp8_mbuverror mmx sse2/; +$vp8_mbuverror_sse2=vp8_mbuverror_xmm; + +add_proto qw/void vp8_subtract_b/, "struct block *be, struct blockd *bd, int pitch"; +specialize qw/vp8_subtract_b mmx sse2 media neon_asm/; +$vp8_subtract_b_media=vp8_subtract_b_armv6; +$vp8_subtract_b_neon_asm=vp8_subtract_b_neon; + +add_proto qw/void vp8_subtract_mby/, "short *diff, unsigned char *src, int src_stride, unsigned char *pred, int pred_stride"; +specialize qw/vp8_subtract_mby mmx sse2 media neon_asm/; +$vp8_subtract_mby_media=vp8_subtract_mby_armv6; +$vp8_subtract_mby_neon_asm=vp8_subtract_mby_neon; + +add_proto qw/void vp8_subtract_mbuv/, "short *diff, unsigned char *usrc, unsigned char *vsrc, int src_stride, unsigned char *upred, unsigned char *vpred, int pred_stride"; +specialize qw/vp8_subtract_mbuv mmx sse2 media neon_asm/; +$vp8_subtract_mbuv_media=vp8_subtract_mbuv_armv6; +$vp8_subtract_mbuv_neon_asm=vp8_subtract_mbuv_neon; + +# +# Motion search +# +add_proto qw/int vp8_full_search_sad/, "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, int sad_per_bit, int distance, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv"; +specialize qw/vp8_full_search_sad sse3 sse4_1/; +$vp8_full_search_sad_sse3=vp8_full_search_sadx3; +$vp8_full_search_sad_sse4_1=vp8_full_search_sadx8; + +add_proto qw/int vp8_refining_search_sad/, "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, int sad_per_bit, int distance, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv"; +specialize qw/vp8_refining_search_sad sse3/; +$vp8_refining_search_sad_sse3=vp8_refining_search_sadx4; + +add_proto qw/int vp8_diamond_search_sad/, "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, union int_mv *best_mv, int search_param, int sad_per_bit, int *num00, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv"; +$vp8_diamond_search_sad_sse3=vp8_diamond_search_sadx4; + +# +# Alt-ref Noise Reduction (ARNR) +# +if (vpx_config("CONFIG_REALTIME_ONLY") ne "yes") { + add_proto qw/void vp8_temporal_filter_apply/, "unsigned char *frame1, unsigned int stride, unsigned char *frame2, unsigned int block_size, int strength, int filter_weight, unsigned int *accumulator, unsigned short *count"; + specialize qw/vp8_temporal_filter_apply sse2/; +} + +# +# Pick Loopfilter +# +add_proto qw/void vp8_yv12_copy_partial_frame/, "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"; +specialize qw/vp8_yv12_copy_partial_frame neon_asm/; +$vp8_yv12_copy_partial_frame_neon_asm=vp8_yv12_copy_partial_frame_neon; + +# +# Denoiser filter +# +if (vpx_config("CONFIG_TEMPORAL_DENOISING") eq "yes") { + add_proto qw/int vp8_denoiser_filter/, "unsigned char *mc_running_avg_y, int mc_avg_y_stride, unsigned char *running_avg_y, int avg_y_stride, unsigned char *sig, int sig_stride, unsigned int motion_magnitude, int increase_denoising"; + specialize qw/vp8_denoiser_filter sse2 neon/; +} + +# End of encoder only functions +} +1; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.sh b/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.sh deleted file mode 100755 index 9ebf389d8c6..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.sh +++ /dev/null @@ -1,542 +0,0 @@ -vp8_common_forward_decls() { -cat <<EOF -/* - * VP8 - */ - -struct blockd; -struct macroblockd; -struct loop_filter_info; - -/* Encoder forward decls */ -struct block; -struct macroblock; -struct variance_vtable; -union int_mv; -struct yv12_buffer_config; -EOF -} -forward_decls vp8_common_forward_decls - -# -# system state -# -prototype void vp8_clear_system_state "" -specialize vp8_clear_system_state mmx -vp8_clear_system_state_mmx=vpx_reset_mmx_state - -# -# Dequant -# -prototype void vp8_dequantize_b "struct blockd*, short *dqc" -specialize vp8_dequantize_b mmx media neon -vp8_dequantize_b_media=vp8_dequantize_b_v6 - -prototype void vp8_dequant_idct_add "short *input, short *dq, unsigned char *output, int stride" -specialize vp8_dequant_idct_add mmx media neon dspr2 -vp8_dequant_idct_add_media=vp8_dequant_idct_add_v6 -vp8_dequant_idct_add_dspr2=vp8_dequant_idct_add_dspr2 - -prototype void vp8_dequant_idct_add_y_block "short *q, short *dq, unsigned char *dst, int stride, char *eobs" -specialize vp8_dequant_idct_add_y_block mmx sse2 media neon dspr2 -vp8_dequant_idct_add_y_block_media=vp8_dequant_idct_add_y_block_v6 -vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2 - -prototype void vp8_dequant_idct_add_uv_block "short *q, short *dq, unsigned char *dst_u, unsigned char *dst_v, int stride, char *eobs" -specialize vp8_dequant_idct_add_uv_block mmx sse2 media neon dspr2 -vp8_dequant_idct_add_uv_block_media=vp8_dequant_idct_add_uv_block_v6 -vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2 - -# -# Loopfilter -# -prototype void vp8_loop_filter_mbv "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi" -specialize vp8_loop_filter_mbv mmx sse2 media neon dspr2 -vp8_loop_filter_mbv_media=vp8_loop_filter_mbv_armv6 -vp8_loop_filter_mbv_dspr2=vp8_loop_filter_mbv_dspr2 - -prototype void vp8_loop_filter_bv "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi" -specialize vp8_loop_filter_bv mmx sse2 media neon dspr2 -vp8_loop_filter_bv_media=vp8_loop_filter_bv_armv6 -vp8_loop_filter_bv_dspr2=vp8_loop_filter_bv_dspr2 - -prototype void vp8_loop_filter_mbh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi" -specialize vp8_loop_filter_mbh mmx sse2 media neon dspr2 -vp8_loop_filter_mbh_media=vp8_loop_filter_mbh_armv6 -vp8_loop_filter_mbh_dspr2=vp8_loop_filter_mbh_dspr2 - -prototype void vp8_loop_filter_bh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi" -specialize vp8_loop_filter_bh mmx sse2 media neon dspr2 -vp8_loop_filter_bh_media=vp8_loop_filter_bh_armv6 -vp8_loop_filter_bh_dspr2=vp8_loop_filter_bh_dspr2 - - -prototype void vp8_loop_filter_simple_mbv "unsigned char *y, int ystride, const unsigned char *blimit" -specialize vp8_loop_filter_simple_mbv mmx sse2 media neon -vp8_loop_filter_simple_mbv_c=vp8_loop_filter_simple_vertical_edge_c -vp8_loop_filter_simple_mbv_mmx=vp8_loop_filter_simple_vertical_edge_mmx -vp8_loop_filter_simple_mbv_sse2=vp8_loop_filter_simple_vertical_edge_sse2 -vp8_loop_filter_simple_mbv_media=vp8_loop_filter_simple_vertical_edge_armv6 -vp8_loop_filter_simple_mbv_neon=vp8_loop_filter_mbvs_neon - -prototype void vp8_loop_filter_simple_mbh "unsigned char *y, int ystride, const unsigned char *blimit" -specialize vp8_loop_filter_simple_mbh mmx sse2 media neon -vp8_loop_filter_simple_mbh_c=vp8_loop_filter_simple_horizontal_edge_c -vp8_loop_filter_simple_mbh_mmx=vp8_loop_filter_simple_horizontal_edge_mmx -vp8_loop_filter_simple_mbh_sse2=vp8_loop_filter_simple_horizontal_edge_sse2 -vp8_loop_filter_simple_mbh_media=vp8_loop_filter_simple_horizontal_edge_armv6 -vp8_loop_filter_simple_mbh_neon=vp8_loop_filter_mbhs_neon - -prototype void vp8_loop_filter_simple_bv "unsigned char *y, int ystride, const unsigned char *blimit" -specialize vp8_loop_filter_simple_bv mmx sse2 media neon -vp8_loop_filter_simple_bv_c=vp8_loop_filter_bvs_c -vp8_loop_filter_simple_bv_mmx=vp8_loop_filter_bvs_mmx -vp8_loop_filter_simple_bv_sse2=vp8_loop_filter_bvs_sse2 -vp8_loop_filter_simple_bv_media=vp8_loop_filter_bvs_armv6 -vp8_loop_filter_simple_bv_neon=vp8_loop_filter_bvs_neon - -prototype void vp8_loop_filter_simple_bh "unsigned char *y, int ystride, const unsigned char *blimit" -specialize vp8_loop_filter_simple_bh mmx sse2 media neon -vp8_loop_filter_simple_bh_c=vp8_loop_filter_bhs_c -vp8_loop_filter_simple_bh_mmx=vp8_loop_filter_bhs_mmx -vp8_loop_filter_simple_bh_sse2=vp8_loop_filter_bhs_sse2 -vp8_loop_filter_simple_bh_media=vp8_loop_filter_bhs_armv6 -vp8_loop_filter_simple_bh_neon=vp8_loop_filter_bhs_neon - -# -# IDCT -# -#idct16 -prototype void vp8_short_idct4x4llm "short *input, unsigned char *pred, int pitch, unsigned char *dst, int dst_stride" -specialize vp8_short_idct4x4llm mmx media neon dspr2 -vp8_short_idct4x4llm_media=vp8_short_idct4x4llm_v6_dual -vp8_short_idct4x4llm_dspr2=vp8_short_idct4x4llm_dspr2 - -#iwalsh1 -prototype void vp8_short_inv_walsh4x4_1 "short *input, short *output" -specialize vp8_short_inv_walsh4x4_1 dspr2 -vp8_short_inv_walsh4x4_1_dspr2=vp8_short_inv_walsh4x4_1_dspr2 -# no asm yet - -#iwalsh16 -prototype void vp8_short_inv_walsh4x4 "short *input, short *output" -specialize vp8_short_inv_walsh4x4 mmx sse2 media neon dspr2 -vp8_short_inv_walsh4x4_media=vp8_short_inv_walsh4x4_v6 -vp8_short_inv_walsh4x4_dspr2=vp8_short_inv_walsh4x4_dspr2 - -#idct1_scalar_add -prototype void vp8_dc_only_idct_add "short input, unsigned char *pred, int pred_stride, unsigned char *dst, int dst_stride" -specialize vp8_dc_only_idct_add mmx media neon dspr2 -vp8_dc_only_idct_add_media=vp8_dc_only_idct_add_v6 -vp8_dc_only_idct_add_dspr2=vp8_dc_only_idct_add_dspr2 - -# -# RECON -# -prototype void vp8_copy_mem16x16 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch" -specialize vp8_copy_mem16x16 mmx sse2 media neon dspr2 -vp8_copy_mem16x16_media=vp8_copy_mem16x16_v6 -vp8_copy_mem16x16_dspr2=vp8_copy_mem16x16_dspr2 - -prototype void vp8_copy_mem8x8 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch" -specialize vp8_copy_mem8x8 mmx media neon dspr2 -vp8_copy_mem8x8_media=vp8_copy_mem8x8_v6 -vp8_copy_mem8x8_dspr2=vp8_copy_mem8x8_dspr2 - -prototype void vp8_copy_mem8x4 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch" -specialize vp8_copy_mem8x4 mmx media neon dspr2 -vp8_copy_mem8x4_media=vp8_copy_mem8x4_v6 -vp8_copy_mem8x4_dspr2=vp8_copy_mem8x4_dspr2 - -prototype void vp8_build_intra_predictors_mby_s "struct macroblockd *x, unsigned char * yabove_row, unsigned char * yleft, int left_stride, unsigned char * ypred_ptr, int y_stride" -specialize vp8_build_intra_predictors_mby_s sse2 ssse3 -#TODO: fix assembly for neon - -prototype void vp8_build_intra_predictors_mbuv_s "struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row, unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride" -specialize vp8_build_intra_predictors_mbuv_s sse2 ssse3 - -prototype void vp8_intra4x4_predict "unsigned char *Above, unsigned char *yleft, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left" -specialize vp8_intra4x4_predict media -vp8_intra4x4_predict_media=vp8_intra4x4_predict_armv6 - -# -# Postproc -# -if [ "$CONFIG_POSTPROC" = "yes" ]; then - prototype void vp8_mbpost_proc_down "unsigned char *dst, int pitch, int rows, int cols,int flimit" - specialize vp8_mbpost_proc_down mmx sse2 - vp8_mbpost_proc_down_sse2=vp8_mbpost_proc_down_xmm - - prototype void vp8_mbpost_proc_across_ip "unsigned char *dst, int pitch, int rows, int cols,int flimit" - specialize vp8_mbpost_proc_across_ip sse2 - vp8_mbpost_proc_across_ip_sse2=vp8_mbpost_proc_across_ip_xmm - - prototype void vp8_post_proc_down_and_across_mb_row "unsigned char *src, unsigned char *dst, int src_pitch, int dst_pitch, int cols, unsigned char *flimits, int size" - specialize vp8_post_proc_down_and_across_mb_row sse2 - - prototype void vp8_plane_add_noise "unsigned char *s, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int w, unsigned int h, int pitch" - specialize vp8_plane_add_noise mmx sse2 - vp8_plane_add_noise_sse2=vp8_plane_add_noise_wmt - - prototype void vp8_blend_mb_inner "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride" - # no asm yet - - prototype void vp8_blend_mb_outer "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride" - # no asm yet - - prototype void vp8_blend_b "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride" - # no asm yet - - prototype void vp8_filter_by_weight16x16 "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight" - specialize vp8_filter_by_weight16x16 sse2 - - prototype void vp8_filter_by_weight8x8 "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight" - specialize vp8_filter_by_weight8x8 sse2 - - prototype void vp8_filter_by_weight4x4 "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight" - # no asm yet -fi - -# -# Subpixel -# -prototype void vp8_sixtap_predict16x16 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_sixtap_predict16x16 mmx sse2 ssse3 media neon dspr2 -vp8_sixtap_predict16x16_media=vp8_sixtap_predict16x16_armv6 -vp8_sixtap_predict16x16_dspr2=vp8_sixtap_predict16x16_dspr2 - -prototype void vp8_sixtap_predict8x8 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_sixtap_predict8x8 mmx sse2 ssse3 media neon dspr2 -vp8_sixtap_predict8x8_media=vp8_sixtap_predict8x8_armv6 -vp8_sixtap_predict8x8_dspr2=vp8_sixtap_predict8x8_dspr2 - -prototype void vp8_sixtap_predict8x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_sixtap_predict8x4 mmx sse2 ssse3 media neon dspr2 -vp8_sixtap_predict8x4_media=vp8_sixtap_predict8x4_armv6 -vp8_sixtap_predict8x4_dspr2=vp8_sixtap_predict8x4_dspr2 - -prototype void vp8_sixtap_predict4x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_sixtap_predict4x4 mmx ssse3 media neon dspr2 -vp8_sixtap_predict4x4_media=vp8_sixtap_predict4x4_armv6 -vp8_sixtap_predict4x4_dspr2=vp8_sixtap_predict4x4_dspr2 - -prototype void vp8_bilinear_predict16x16 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_bilinear_predict16x16 mmx sse2 ssse3 media neon -vp8_bilinear_predict16x16_media=vp8_bilinear_predict16x16_armv6 - -prototype void vp8_bilinear_predict8x8 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_bilinear_predict8x8 mmx sse2 ssse3 media neon -vp8_bilinear_predict8x8_media=vp8_bilinear_predict8x8_armv6 - -prototype void vp8_bilinear_predict8x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_bilinear_predict8x4 mmx media neon -vp8_bilinear_predict8x4_media=vp8_bilinear_predict8x4_armv6 - -prototype void vp8_bilinear_predict4x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch" -specialize vp8_bilinear_predict4x4 mmx media neon -vp8_bilinear_predict4x4_media=vp8_bilinear_predict4x4_armv6 - -# -# Whole-pixel Variance -# -prototype unsigned int vp8_variance4x4 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance4x4 mmx sse2 -vp8_variance4x4_sse2=vp8_variance4x4_wmt - -prototype unsigned int vp8_variance8x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance8x8 mmx sse2 media neon -vp8_variance8x8_sse2=vp8_variance8x8_wmt -vp8_variance8x8_media=vp8_variance8x8_armv6 - -prototype unsigned int vp8_variance8x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance8x16 mmx sse2 neon -vp8_variance8x16_sse2=vp8_variance8x16_wmt - -prototype unsigned int vp8_variance16x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance16x8 mmx sse2 neon -vp8_variance16x8_sse2=vp8_variance16x8_wmt - -prototype unsigned int vp8_variance16x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance16x16 mmx sse2 media neon -vp8_variance16x16_sse2=vp8_variance16x16_wmt -vp8_variance16x16_media=vp8_variance16x16_armv6 - -# -# Sub-pixel Variance -# -prototype unsigned int vp8_sub_pixel_variance4x4 "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse" -specialize vp8_sub_pixel_variance4x4 mmx sse2 -vp8_sub_pixel_variance4x4_sse2=vp8_sub_pixel_variance4x4_wmt - -prototype unsigned int vp8_sub_pixel_variance8x8 "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse" -specialize vp8_sub_pixel_variance8x8 mmx sse2 media neon -vp8_sub_pixel_variance8x8_sse2=vp8_sub_pixel_variance8x8_wmt -vp8_sub_pixel_variance8x8_media=vp8_sub_pixel_variance8x8_armv6 - -prototype unsigned int vp8_sub_pixel_variance8x16 "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse" -specialize vp8_sub_pixel_variance8x16 mmx sse2 -vp8_sub_pixel_variance8x16_sse2=vp8_sub_pixel_variance8x16_wmt - -prototype unsigned int vp8_sub_pixel_variance16x8 "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse" -specialize vp8_sub_pixel_variance16x8 mmx sse2 ssse3 -vp8_sub_pixel_variance16x8_sse2=vp8_sub_pixel_variance16x8_wmt - -prototype unsigned int vp8_sub_pixel_variance16x16 "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse" -specialize vp8_sub_pixel_variance16x16 mmx sse2 ssse3 media neon -vp8_sub_pixel_variance16x16_sse2=vp8_sub_pixel_variance16x16_wmt -vp8_sub_pixel_variance16x16_media=vp8_sub_pixel_variance16x16_armv6 - -prototype unsigned int vp8_variance_halfpixvar16x16_h "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance_halfpixvar16x16_h mmx sse2 media neon -vp8_variance_halfpixvar16x16_h_sse2=vp8_variance_halfpixvar16x16_h_wmt -vp8_variance_halfpixvar16x16_h_media=vp8_variance_halfpixvar16x16_h_armv6 - -prototype unsigned int vp8_variance_halfpixvar16x16_v "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance_halfpixvar16x16_v mmx sse2 media neon -vp8_variance_halfpixvar16x16_v_sse2=vp8_variance_halfpixvar16x16_v_wmt -vp8_variance_halfpixvar16x16_v_media=vp8_variance_halfpixvar16x16_v_armv6 - -prototype unsigned int vp8_variance_halfpixvar16x16_hv "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_variance_halfpixvar16x16_hv mmx sse2 media neon -vp8_variance_halfpixvar16x16_hv_sse2=vp8_variance_halfpixvar16x16_hv_wmt -vp8_variance_halfpixvar16x16_hv_media=vp8_variance_halfpixvar16x16_hv_armv6 - -# -# Single block SAD -# -prototype unsigned int vp8_sad4x4 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad" -specialize vp8_sad4x4 mmx sse2 neon -vp8_sad4x4_sse2=vp8_sad4x4_wmt - -prototype unsigned int vp8_sad8x8 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad" -specialize vp8_sad8x8 mmx sse2 neon -vp8_sad8x8_sse2=vp8_sad8x8_wmt - -prototype unsigned int vp8_sad8x16 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad" -specialize vp8_sad8x16 mmx sse2 neon -vp8_sad8x16_sse2=vp8_sad8x16_wmt - -prototype unsigned int vp8_sad16x8 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad" -specialize vp8_sad16x8 mmx sse2 neon -vp8_sad16x8_sse2=vp8_sad16x8_wmt - -prototype unsigned int vp8_sad16x16 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad" -specialize vp8_sad16x16 mmx sse2 sse3 media neon -vp8_sad16x16_sse2=vp8_sad16x16_wmt -vp8_sad16x16_media=vp8_sad16x16_armv6 - -# -# Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally -# -prototype void vp8_sad4x4x3 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array" -specialize vp8_sad4x4x3 sse3 - -prototype void vp8_sad8x8x3 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array" -specialize vp8_sad8x8x3 sse3 - -prototype void vp8_sad8x16x3 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array" -specialize vp8_sad8x16x3 sse3 - -prototype void vp8_sad16x8x3 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array" -specialize vp8_sad16x8x3 sse3 ssse3 - -prototype void vp8_sad16x16x3 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sad_array" -specialize vp8_sad16x16x3 sse3 ssse3 - -# Note the only difference in the following prototypes is that they return into -# an array of short -prototype void vp8_sad4x4x8 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array" -specialize vp8_sad4x4x8 sse4_1 -vp8_sad4x4x8_sse4_1=vp8_sad4x4x8_sse4 - -prototype void vp8_sad8x8x8 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array" -specialize vp8_sad8x8x8 sse4_1 -vp8_sad8x8x8_sse4_1=vp8_sad8x8x8_sse4 - -prototype void vp8_sad8x16x8 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array" -specialize vp8_sad8x16x8 sse4_1 -vp8_sad8x16x8_sse4_1=vp8_sad8x16x8_sse4 - -prototype void vp8_sad16x8x8 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array" -specialize vp8_sad16x8x8 sse4_1 -vp8_sad16x8x8_sse4_1=vp8_sad16x8x8_sse4 - -prototype void vp8_sad16x16x8 "const unsigned char *src_ptr, int src_stride, const unsigned char *ref_ptr, int ref_stride, unsigned short *sad_array" -specialize vp8_sad16x16x8 sse4_1 -vp8_sad16x16x8_sse4_1=vp8_sad16x16x8_sse4 - -# -# Multi-block SAD, comparing a reference to N independent blocks -# -prototype void vp8_sad4x4x4d "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array" -specialize vp8_sad4x4x4d sse3 - -prototype void vp8_sad8x8x4d "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array" -specialize vp8_sad8x8x4d sse3 - -prototype void vp8_sad8x16x4d "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array" -specialize vp8_sad8x16x4d sse3 - -prototype void vp8_sad16x8x4d "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array" -specialize vp8_sad16x8x4d sse3 - -prototype void vp8_sad16x16x4d "const unsigned char *src_ptr, int src_stride, const unsigned char * const ref_ptr[], int ref_stride, unsigned int *sad_array" -specialize vp8_sad16x16x4d sse3 - -# -# Encoder functions below this point. -# -if [ "$CONFIG_VP8_ENCODER" = "yes" ]; then - -# -# Sum of squares (vector) -# -prototype unsigned int vp8_get_mb_ss "const short *" -specialize vp8_get_mb_ss mmx sse2 - -# -# SSE (Sum Squared Error) -# -prototype unsigned int vp8_sub_pixel_mse16x16 "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse" -specialize vp8_sub_pixel_mse16x16 mmx sse2 -vp8_sub_pixel_mse16x16_sse2=vp8_sub_pixel_mse16x16_wmt - -prototype unsigned int vp8_mse16x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse" -specialize vp8_mse16x16 mmx sse2 media neon -vp8_mse16x16_sse2=vp8_mse16x16_wmt -vp8_mse16x16_media=vp8_mse16x16_armv6 - -prototype unsigned int vp8_get4x4sse_cs "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride" -specialize vp8_get4x4sse_cs mmx neon - -# -# Block copy -# -case $arch in - x86*) - prototype void vp8_copy32xn "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, int n" - specialize vp8_copy32xn sse2 sse3 - ;; -esac - -# -# Structured Similarity (SSIM) -# -if [ "$CONFIG_INTERNAL_STATS" = "yes" ]; then - [ $arch = "x86_64" ] && sse2_on_x86_64=sse2 - - prototype void vp8_ssim_parms_8x8 "unsigned char *s, int sp, unsigned char *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr" - specialize vp8_ssim_parms_8x8 $sse2_on_x86_64 - - prototype void vp8_ssim_parms_16x16 "unsigned char *s, int sp, unsigned char *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr" - specialize vp8_ssim_parms_16x16 $sse2_on_x86_64 -fi - -# -# Forward DCT -# -prototype void vp8_short_fdct4x4 "short *input, short *output, int pitch" -specialize vp8_short_fdct4x4 mmx sse2 media neon -vp8_short_fdct4x4_media=vp8_short_fdct4x4_armv6 - -prototype void vp8_short_fdct8x4 "short *input, short *output, int pitch" -specialize vp8_short_fdct8x4 mmx sse2 media neon -vp8_short_fdct8x4_media=vp8_short_fdct8x4_armv6 - -prototype void vp8_short_walsh4x4 "short *input, short *output, int pitch" -specialize vp8_short_walsh4x4 sse2 media neon -vp8_short_walsh4x4_media=vp8_short_walsh4x4_armv6 - -# -# Quantizer -# -prototype void vp8_regular_quantize_b "struct block *, struct blockd *" -specialize vp8_regular_quantize_b sse2 #sse4_1 -# TODO(johann) Update sse4 implementation and re-enable -#vp8_regular_quantize_b_sse4_1=vp8_regular_quantize_b_sse4 - -prototype void vp8_fast_quantize_b "struct block *, struct blockd *" -specialize vp8_fast_quantize_b sse2 ssse3 media neon -vp8_fast_quantize_b_media=vp8_fast_quantize_b_armv6 - -prototype void vp8_regular_quantize_b_pair "struct block *b1, struct block *b2, struct blockd *d1, struct blockd *d2" -# no asm yet - -prototype void vp8_fast_quantize_b_pair "struct block *b1, struct block *b2, struct blockd *d1, struct blockd *d2" -specialize vp8_fast_quantize_b_pair neon - -prototype void vp8_quantize_mb "struct macroblock *" -specialize vp8_quantize_mb neon - -prototype void vp8_quantize_mby "struct macroblock *" -specialize vp8_quantize_mby neon - -prototype void vp8_quantize_mbuv "struct macroblock *" -specialize vp8_quantize_mbuv neon - -# -# Block subtraction -# -prototype int vp8_block_error "short *coeff, short *dqcoeff" -specialize vp8_block_error mmx sse2 -vp8_block_error_sse2=vp8_block_error_xmm - -prototype int vp8_mbblock_error "struct macroblock *mb, int dc" -specialize vp8_mbblock_error mmx sse2 -vp8_mbblock_error_sse2=vp8_mbblock_error_xmm - -prototype int vp8_mbuverror "struct macroblock *mb" -specialize vp8_mbuverror mmx sse2 -vp8_mbuverror_sse2=vp8_mbuverror_xmm - -prototype void vp8_subtract_b "struct block *be, struct blockd *bd, int pitch" -specialize vp8_subtract_b mmx sse2 media neon -vp8_subtract_b_media=vp8_subtract_b_armv6 - -prototype void vp8_subtract_mby "short *diff, unsigned char *src, int src_stride, unsigned char *pred, int pred_stride" -specialize vp8_subtract_mby mmx sse2 media neon -vp8_subtract_mby_media=vp8_subtract_mby_armv6 - -prototype void vp8_subtract_mbuv "short *diff, unsigned char *usrc, unsigned char *vsrc, int src_stride, unsigned char *upred, unsigned char *vpred, int pred_stride" -specialize vp8_subtract_mbuv mmx sse2 media neon -vp8_subtract_mbuv_media=vp8_subtract_mbuv_armv6 - -# -# Motion search -# -prototype int vp8_full_search_sad "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, int sad_per_bit, int distance, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv" -specialize vp8_full_search_sad sse3 sse4_1 -vp8_full_search_sad_sse3=vp8_full_search_sadx3 -vp8_full_search_sad_sse4_1=vp8_full_search_sadx8 - -prototype int vp8_refining_search_sad "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, int sad_per_bit, int distance, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv" -specialize vp8_refining_search_sad sse3 -vp8_refining_search_sad_sse3=vp8_refining_search_sadx4 - -prototype int vp8_diamond_search_sad "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, union int_mv *best_mv, int search_param, int sad_per_bit, int *num00, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv" -vp8_diamond_search_sad_sse3=vp8_diamond_search_sadx4 - -# -# Alt-ref Noise Reduction (ARNR) -# -if [ "$CONFIG_REALTIME_ONLY" != "yes" ]; then - prototype void vp8_temporal_filter_apply "unsigned char *frame1, unsigned int stride, unsigned char *frame2, unsigned int block_size, int strength, int filter_weight, unsigned int *accumulator, unsigned short *count" - specialize vp8_temporal_filter_apply sse2 -fi - -# -# Pick Loopfilter -# -prototype void vp8_yv12_copy_partial_frame "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc" -specialize vp8_yv12_copy_partial_frame neon - -# -# Denoiser filter -# -if [ "$CONFIG_TEMPORAL_DENOISING" = "yes" ]; then - prototype int vp8_denoiser_filter "struct yv12_buffer_config* mc_running_avg, struct yv12_buffer_config* running_avg, struct macroblock* signal, unsigned int motion_magnitude2, int y_offset, int uv_offset" - specialize vp8_denoiser_filter sse2 -fi - -# End of encoder only functions -fi diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/setupintrarecon.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/setupintrarecon.h index e515c3a005b..608f4a9ac25 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/setupintrarecon.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/setupintrarecon.h @@ -8,8 +8,14 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_COMMON_SETUPINTRARECON_H_ +#define VP8_COMMON_SETUPINTRARECON_H_ #include "vpx_scale/yv12config.h" + +#ifdef __cplusplus +extern "C" { +#endif extern void vp8_setup_intra_recon(YV12_BUFFER_CONFIG *ybf); extern void vp8_setup_intra_recon_top_line(YV12_BUFFER_CONFIG *ybf); @@ -31,3 +37,9 @@ void setup_intra_recon_left(unsigned char *y_buffer, for (i = 0; i < 8; i++) v_buffer[uv_stride *i] = (unsigned char) 129; } + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_SETUPINTRARECON_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/swapyv12buffer.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/swapyv12buffer.h index a6473ed92be..1d66cd3d62e 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/swapyv12buffer.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/swapyv12buffer.h @@ -9,11 +9,19 @@ */ -#ifndef SWAPYV12_BUFFER_H -#define SWAPYV12_BUFFER_H +#ifndef VP8_COMMON_SWAPYV12BUFFER_H_ +#define VP8_COMMON_SWAPYV12BUFFER_H_ #include "vpx_scale/yv12config.h" +#ifdef __cplusplus +extern "C" { +#endif + void vp8_swap_yv12_buffer(YV12_BUFFER_CONFIG *new_frame, YV12_BUFFER_CONFIG *last_frame); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_SWAPYV12BUFFER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/systemdependent.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/systemdependent.h index e6b0456f75b..3d44e37cf24 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/systemdependent.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/systemdependent.h @@ -8,8 +8,20 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_COMMON_SYSTEMDEPENDENT_H_ +#define VP8_COMMON_SYSTEMDEPENDENT_H_ #include "vpx_config.h" +#ifdef __cplusplus +extern "C" { +#endif + struct VP8Common; void vp8_machine_specific_config(struct VP8Common *); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_SYSTEMDEPENDENT_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/threading.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/threading.h index ed9e3e60dcb..01c82dbb805 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/threading.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/threading.h @@ -9,8 +9,12 @@ */ -#ifndef _PTHREAD_EMULATION -#define _PTHREAD_EMULATION +#ifndef VP8_COMMON_THREADING_H_ +#define VP8_COMMON_THREADING_H_ + +#ifdef __cplusplus +extern "C" { +#endif #if CONFIG_OS_SUPPORT && CONFIG_MULTITHREAD @@ -183,4 +187,8 @@ static inline int sem_destroy(sem_t * sem) #endif /* CONFIG_OS_SUPPORT && CONFIG_MULTITHREAD */ +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_THREADING_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/treecoder.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/treecoder.h index ebf51c5ed65..d22b7c570cb 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/treecoder.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/treecoder.h @@ -9,8 +9,12 @@ */ -#ifndef __INC_TREECODER_H -#define __INC_TREECODER_H +#ifndef VP8_COMMON_TREECODER_H_ +#define VP8_COMMON_TREECODER_H_ + +#ifdef __cplusplus +extern "C" { +#endif typedef unsigned char vp8bc_index_t; /* probability index */ @@ -87,4 +91,8 @@ void vp8bc_tree_probs_from_distribution( ); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_TREECODER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/variance.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/variance.h index 01193b8d724..89a32a72268 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/variance.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/variance.h @@ -9,11 +9,15 @@ */ -#ifndef VARIANCE_H -#define VARIANCE_H +#ifndef VP8_COMMON_VARIANCE_H_ +#define VP8_COMMON_VARIANCE_H_ #include "vpx_config.h" +#ifdef __cplusplus +extern "C" { +#endif + typedef unsigned int(*vp8_sad_fn_t)( const unsigned char *src_ptr, int source_stride, @@ -112,4 +116,8 @@ typedef struct variance_vtable #endif } vp8_variance_fn_ptr_t; +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_COMMON_VARIANCE_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/vp8_entropymodedata.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/vp8_entropymodedata.h index 13e9a92fc10..c4aed498970 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/vp8_entropymodedata.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/vp8_entropymodedata.h @@ -8,6 +8,12 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_COMMON_VP8_ENTROPYMODEDATA_H_ +#define VP8_COMMON_VP8_ENTROPYMODEDATA_H_ + +#ifdef __cplusplus +extern "C" { +#endif /*Generated file, included by entropymode.c*/ @@ -240,3 +246,9 @@ const vp8_prob vp8_kf_bmode_prob { 112, 19, 12, 61, 195, 128, 48, 4, 24 } } }; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_VP8_ENTROPYMODEDATA_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/filter_x86.h b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/filter_x86.h index cfadaeecbc1..d282841bee4 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/filter_x86.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/filter_x86.h @@ -8,11 +8,15 @@ * be found in the AUTHORS file in the root of the source tree. */ -#ifndef FILTER_X86_H -#define FILTER_X86_H +#ifndef VP8_COMMON_X86_FILTER_X86_H_ +#define VP8_COMMON_X86_FILTER_X86_H_ #include "vpx_ports/mem.h" +#ifdef __cplusplus +extern "C" { +#endif + /* x86 assembly specific copy of vp8/common/filter.c:vp8_bilinear_filters with * duplicated values */ @@ -22,4 +26,8 @@ extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_4[8][8]); /* duplicated 8x */ extern DECLARE_ALIGNED(16, const short, vp8_bilinear_filters_x86_8[8][16]); -#endif /* FILTER_X86_H */ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_COMMON_X86_FILTER_X86_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_mmx.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_mmx.asm index f388d247681..88a07b9f3fa 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_mmx.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_mmx.asm @@ -527,7 +527,7 @@ sym(vp8_loop_filter_vertical_edge_mmx): pxor mm7, [GLOBAL(t80)] ; unoffset ; mm7 = q1 - ; tranpose and write back + ; transpose and write back ; mm1 = 72 62 52 42 32 22 12 02 ; mm6 = 73 63 53 43 33 23 13 03 ; mm3 = 74 64 54 44 34 24 14 04 @@ -1289,7 +1289,7 @@ sym(vp8_mbloop_filter_vertical_edge_mmx): pxor mm6, [GLOBAL(t80)] ; mm6 = 71 61 51 41 31 21 11 01 pxor mm3, [GLOBAL(t80)] ; mm3 = 76 66 56 46 36 26 15 06 - ; tranpose and write back + ; transpose and write back movq mm0, [rdx] ; mm0 = 70 60 50 40 30 20 10 00 movq mm1, mm0 ; mm0 = 70 60 50 40 30 20 10 00 diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_sse2.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_sse2.asm index a66753ba806..1913abc69b0 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_sse2.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/loopfilter_sse2.asm @@ -958,7 +958,7 @@ sym(vp8_loop_filter_vertical_edge_sse2): ; start work on filters B_FILTER 2 - ; tranpose and write back - only work on q1, q0, p0, p1 + ; transpose and write back - only work on q1, q0, p0, p1 BV_TRANSPOSE ; store 16-line result @@ -1023,7 +1023,7 @@ sym(vp8_loop_filter_vertical_edge_uv_sse2): ; start work on filters B_FILTER 2 - ; tranpose and write back - only work on q1, q0, p0, p1 + ; transpose and write back - only work on q1, q0, p0, p1 BV_TRANSPOSE lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_mmx.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_mmx.asm index 5cf110b5326..8be3431f9b0 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_mmx.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_mmx.asm @@ -204,13 +204,16 @@ sym(vp8_mbpost_proc_down_mmx): and rcx, 15 movd DWORD PTR [rsp+rcx*4], mm1 ;d[rcx*4] + cmp edx, 8 + jl .skip_assignment + mov rcx, rdx sub rcx, 8 - and rcx, 15 movd mm1, DWORD PTR [rsp+rcx*4] ;d[rcx*4] - movd [rsi], mm1 + +.skip_assignment lea rsi, [rsi+rax] lea rdi, [rdi+rax] diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_sse2.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_sse2.asm index 00f84a31b21..f53daa7e508 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_sse2.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/postproc_sse2.asm @@ -425,13 +425,16 @@ sym(vp8_mbpost_proc_down_xmm): and rcx, 15 movq QWORD PTR [rsp + rcx*8], xmm1 ;d[rcx*8] + cmp edx, 8 + jl .skip_assignment + mov rcx, rdx sub rcx, 8 - and rcx, 15 movq mm0, [rsp + rcx*8] ;d[rcx*8] - movq [rsi], mm0 + +.skip_assignment lea rsi, [rsi+rax] lea rdi, [rdi+rax] diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/recon_sse2.asm b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/recon_sse2.asm index 1434bcd9379..7141f832463 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/recon_sse2.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/x86/recon_sse2.asm @@ -365,6 +365,7 @@ sym(vp8_intra_pred_uv_tm_%1): GET_GOT rbx push rsi push rdi + push rbx ; end prolog ; read top row @@ -395,8 +396,11 @@ sym(vp8_intra_pred_uv_tm_%1): movsxd rcx, dword ptr arg(1) ;dst_stride .vp8_intra_pred_uv_tm_%1_loop: - movd xmm3, [rsi] - movd xmm5, [rsi+rax] + mov bl, [rsi] + movd xmm3, ebx + + mov bl, [rsi+rax] + movd xmm5, ebx %ifidn %1, sse2 punpcklbw xmm3, xmm0 punpcklbw xmm5, xmm0 @@ -419,6 +423,7 @@ sym(vp8_intra_pred_uv_tm_%1): jnz .vp8_intra_pred_uv_tm_%1_loop ; begin epilog + pop rbx pop rdi pop rsi RESTORE_GOT @@ -486,10 +491,8 @@ sym(vp8_intra_pred_uv_ho_%1): SHADOW_ARGS_TO_STACK 5 push rsi push rdi -%ifidn %1, ssse3 -%ifndef GET_GOT_SAVE_ARG push rbx -%endif +%ifidn %1, ssse3 GET_GOT rbx %endif ; end prolog @@ -507,13 +510,16 @@ sym(vp8_intra_pred_uv_ho_%1): %ifidn %1, ssse3 lea rdx, [rcx*3] movdqa xmm2, [GLOBAL(dc_00001111)] - lea rbx, [rax*3] %endif %ifidn %1, mmx2 .vp8_intra_pred_uv_ho_%1_loop: - movd mm0, [rsi] - movd mm1, [rsi+rax] + mov bl, [rsi] + movd mm0, ebx + + mov bl, [rsi+rax] + movd mm1, ebx + punpcklbw mm0, mm0 punpcklbw mm1, mm1 pshufw mm0, mm0, 0x0 @@ -525,10 +531,19 @@ sym(vp8_intra_pred_uv_ho_%1): dec edx jnz .vp8_intra_pred_uv_ho_%1_loop %else - movd xmm0, [rsi] - movd xmm3, [rsi+rax] - movd xmm1, [rsi+rax*2] - movd xmm4, [rsi+rbx] + mov bl, [rsi] + movd xmm0, ebx + + mov bl, [rsi+rax] + movd xmm3, ebx + + mov bl, [rsi+rax*2] + movd xmm1, ebx + + lea rbx, [rax*3] + mov bl, [rsi+rbx] + movd xmm4, ebx + punpcklbw xmm0, xmm3 punpcklbw xmm1, xmm4 pshufb xmm0, xmm2 @@ -539,10 +554,20 @@ sym(vp8_intra_pred_uv_ho_%1): movhps [rdi+rdx], xmm1 lea rsi, [rsi+rax*4] lea rdi, [rdi+rcx*4] - movd xmm0, [rsi] - movd xmm3, [rsi+rax] - movd xmm1, [rsi+rax*2] - movd xmm4, [rsi+rbx] + + mov bl, [rsi] + movd xmm0, ebx + + mov bl, [rsi+rax] + movd xmm3, ebx + + mov bl, [rsi+rax*2] + movd xmm1, ebx + + lea rbx, [rax*3] + mov bl, [rsi+rbx] + movd xmm4, ebx + punpcklbw xmm0, xmm3 punpcklbw xmm1, xmm4 pshufb xmm0, xmm2 @@ -556,10 +581,8 @@ sym(vp8_intra_pred_uv_ho_%1): ; begin epilog %ifidn %1, ssse3 RESTORE_GOT -%ifndef GET_GOT_SAVE_ARG - pop rbx -%endif %endif + pop rbx pop rdi pop rsi UNSHADOW_ARGS @@ -893,6 +916,7 @@ sym(vp8_intra_pred_y_tm_%1): SAVE_XMM 7 push rsi push rdi + push rbx GET_GOT rbx ; end prolog @@ -926,8 +950,11 @@ sym(vp8_intra_pred_y_tm_%1): mov rdi, arg(0) ;dst; movsxd rcx, dword ptr arg(1) ;dst_stride vp8_intra_pred_y_tm_%1_loop: - movd xmm4, [rsi] - movd xmm5, [rsi+rax] + mov bl, [rsi] + movd xmm4, ebx + + mov bl, [rsi+rax] + movd xmm5, ebx %ifidn %1, sse2 punpcklbw xmm4, xmm0 punpcklbw xmm5, xmm0 @@ -956,6 +983,7 @@ vp8_intra_pred_y_tm_%1_loop: ; begin epilog RESTORE_GOT + pop rbx pop rdi pop rsi RESTORE_XMM @@ -1029,6 +1057,7 @@ sym(vp8_intra_pred_y_ho_sse2): SHADOW_ARGS_TO_STACK 5 push rsi push rdi + push rbx ; end prolog ;arg(2) not used @@ -1041,8 +1070,11 @@ sym(vp8_intra_pred_y_ho_sse2): movsxd rcx, dword ptr arg(1) ;dst_stride vp8_intra_pred_y_ho_sse2_loop: - movd xmm0, [rsi] - movd xmm1, [rsi+rax] + mov bl, [rsi] + movd xmm0, ebx + mov bl, [rsi+rax] + movd xmm1, ebx + ; FIXME use pshufb for ssse3 version punpcklbw xmm0, xmm0 punpcklbw xmm1, xmm1 @@ -1058,6 +1090,7 @@ vp8_intra_pred_y_ho_sse2_loop: jnz vp8_intra_pred_y_ho_sse2_loop ; begin epilog + pop rbx pop rdi pop rsi UNSHADOW_ARGS diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.c b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.c index 0007d7a7a3e..b874d4c46c0 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.c @@ -10,11 +10,12 @@ #include "dboolhuff.h" +#include "vp8/common/common.h" int vp8dx_start_decode(BOOL_DECODER *br, const unsigned char *source, unsigned int source_sz, - vp8_decrypt_cb *decrypt_cb, + vpx_decrypt_cb decrypt_cb, void *decrypt_state) { br->user_buffer_end = source+source_sz; @@ -39,7 +40,7 @@ void vp8dx_bool_decoder_fill(BOOL_DECODER *br) const unsigned char *bufptr = br->user_buffer; VP8_BD_VALUE value = br->value; int count = br->count; - int shift = VP8_BD_VALUE_SIZE - 8 - (count + 8); + int shift = VP8_BD_VALUE_SIZE - CHAR_BIT - (count + CHAR_BIT); size_t bytes_left = br->user_buffer_end - bufptr; size_t bits_left = bytes_left * CHAR_BIT; int x = (int)(shift + CHAR_BIT - bits_left); @@ -47,7 +48,7 @@ void vp8dx_bool_decoder_fill(BOOL_DECODER *br) unsigned char decrypted[sizeof(VP8_BD_VALUE) + 1]; if (br->decrypt_cb) { - size_t n = bytes_left > sizeof(decrypted) ? sizeof(decrypted) : bytes_left; + size_t n = MIN(sizeof(decrypted), bytes_left); br->decrypt_cb(br->decrypt_state, bufptr, decrypted, (int)n); bufptr = decrypted; } diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.h index 4c0ca1ce737..51c5adc28ab 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/dboolhuff.h @@ -9,16 +9,21 @@ */ -#ifndef DBOOLHUFF_H_ -#define DBOOLHUFF_H_ +#ifndef VP8_DECODER_DBOOLHUFF_H_ +#define VP8_DECODER_DBOOLHUFF_H_ #include <stddef.h> #include <limits.h> #include "vpx_config.h" #include "vpx_ports/mem.h" +#include "vpx/vp8dx.h" #include "vpx/vpx_integer.h" +#ifdef __cplusplus +extern "C" { +#endif + typedef size_t VP8_BD_VALUE; #define VP8_BD_VALUE_SIZE ((int)sizeof(VP8_BD_VALUE)*CHAR_BIT) @@ -28,12 +33,6 @@ typedef size_t VP8_BD_VALUE; Even relatively modest values like 100 would work fine.*/ #define VP8_LOTS_OF_BITS (0x40000000) -/*Decrypt n bytes of data from input -> output, using the decrypt_state - passed in VP8D_SET_DECRYPTOR. -*/ -typedef void (vp8_decrypt_cb)(void *decrypt_state, const unsigned char *input, - unsigned char *output, int count); - typedef struct { const unsigned char *user_buffer_end; @@ -41,7 +40,7 @@ typedef struct VP8_BD_VALUE value; int count; unsigned int range; - vp8_decrypt_cb *decrypt_cb; + vpx_decrypt_cb decrypt_cb; void *decrypt_state; } BOOL_DECODER; @@ -50,7 +49,7 @@ DECLARE_ALIGNED(16, extern const unsigned char, vp8_norm[256]); int vp8dx_start_decode(BOOL_DECODER *br, const unsigned char *source, unsigned int source_sz, - vp8_decrypt_cb *decrypt_cb, + vpx_decrypt_cb decrypt_cb, void *decrypt_state); void vp8dx_bool_decoder_fill(BOOL_DECODER *br); @@ -135,4 +134,8 @@ static int vp8dx_bool_error(BOOL_DECODER *br) return 0; } -#endif // DBOOLHUFF_H_ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DBOOLHUFF_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodframe.c b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodeframe.c index 16da78a2c1e..e7cf0d9b9c6 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodframe.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodeframe.c @@ -17,6 +17,7 @@ #include "vp8/common/reconintra4x4.h" #include "vp8/common/reconinter.h" #include "detokenize.h" +#include "vp8/common/common.h" #include "vp8/common/invtrans.h" #include "vp8/common/alloccommon.h" #include "vp8/common/entropymode.h" @@ -631,9 +632,17 @@ static void decode_mb_rows(VP8D_COMP *pbi) xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset; xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset; - xd->pre.y_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][0] + recon_yoffset; - xd->pre.u_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][1] + recon_uvoffset; - xd->pre.v_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][2] + recon_uvoffset; + if (xd->mode_info_context->mbmi.ref_frame >= LAST_FRAME) { + MV_REFERENCE_FRAME ref = xd->mode_info_context->mbmi.ref_frame; + xd->pre.y_buffer = ref_buffer[ref][0] + recon_yoffset; + xd->pre.u_buffer = ref_buffer[ref][1] + recon_uvoffset; + xd->pre.v_buffer = ref_buffer[ref][2] + recon_uvoffset; + } else { + // ref_frame is INTRA_FRAME, pre buffer should not be used. + xd->pre.y_buffer = 0; + xd->pre.u_buffer = 0; + xd->pre.v_buffer = 0; + } /* propagate errors from reference frames */ xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame]; @@ -680,7 +689,6 @@ static void decode_mb_rows(VP8D_COMP *pbi) vp8_loop_filter_row_simple(pc, lf_mic, mb_row-1, recon_y_stride, recon_uv_stride, lf_dst[0], lf_dst[1], lf_dst[2]); - if(mb_row > 1) { yv12_extend_frame_left_right_c(yv12_fb_new, @@ -691,10 +699,6 @@ static void decode_mb_rows(VP8D_COMP *pbi) eb_dst[0] += recon_y_stride * 16; eb_dst[1] += recon_uv_stride * 8; eb_dst[2] += recon_uv_stride * 8; - - if(mb_row == 2) - yv12_extend_frame_top_c(yv12_fb_new); - } lf_dst[0] += recon_y_stride * 16; @@ -713,13 +717,9 @@ static void decode_mb_rows(VP8D_COMP *pbi) eb_dst[0], eb_dst[1], eb_dst[2]); - eb_dst[0] += recon_y_stride * 16; eb_dst[1] += recon_uv_stride * 8; eb_dst[2] += recon_uv_stride * 8; - - if(mb_row == 1) - yv12_extend_frame_top_c(yv12_fb_new); } } } @@ -747,7 +747,7 @@ static void decode_mb_rows(VP8D_COMP *pbi) eb_dst[0], eb_dst[1], eb_dst[2]); - + yv12_extend_frame_top_c(yv12_fb_new); yv12_extend_frame_bottom_c(yv12_fb_new); } @@ -1019,8 +1019,7 @@ int vp8_decode_frame(VP8D_COMP *pbi) const unsigned char *clear = data; if (pbi->decrypt_cb) { - int n = (int)(data_end - data); - if (n > 10) n = 10; + int n = (int)MIN(sizeof(clear_buffer), data_end - data); pbi->decrypt_cb(pbi->decrypt_state, data, clear_buffer, n); clear = clear_buffer; } diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodemv.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodemv.h index 05a33d27f6a..f33b07351d3 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodemv.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decodemv.h @@ -8,11 +8,19 @@ * be found in the AUTHORS file in the root of the source tree. */ -#ifndef DECODEMV_H_ -#define DECODEMV_H_ +#ifndef VP8_DECODER_DECODEMV_H_ +#define VP8_DECODER_DECODEMV_H_ #include "onyxd_int.h" +#ifdef __cplusplus +extern "C" { +#endif + void vp8_decode_mode_mvs(VP8D_COMP *); -#endif // DECODEMV_H_ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DECODEMV_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decoderthreading.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decoderthreading.h index bc716e489e0..c563cf6e93a 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decoderthreading.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/decoderthreading.h @@ -8,8 +8,12 @@ * be found in the AUTHORS file in the root of the source tree. */ -#ifndef DECODERTHREADING_H_ -#define DECODERTHREADING_H_ +#ifndef VP8_DECODER_DECODERTHREADING_H_ +#define VP8_DECODER_DECODERTHREADING_H_ + +#ifdef __cplusplus +extern "C" { +#endif #if CONFIG_MULTITHREAD void vp8mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd); @@ -19,4 +23,8 @@ void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows); void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows); #endif -#endif // DECODERTHREADING_H_ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DECODERTHREADING_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/detokenize.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/detokenize.h index f2130b36178..f0b125444f0 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/detokenize.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/detokenize.h @@ -8,12 +8,20 @@ * be found in the AUTHORS file in the root of the source tree. */ -#ifndef DETOKENIZE_H_ -#define DETOKENIZE_H_ +#ifndef VP8_DECODER_DETOKENIZE_H_ +#define VP8_DECODER_DETOKENIZE_H_ #include "onyxd_int.h" +#ifdef __cplusplus +extern "C" { +#endif + void vp8_reset_mb_tokens_context(MACROBLOCKD *x); int vp8_decode_mb_tokens(VP8D_COMP *, MACROBLOCKD *); -#endif // DETOKENIZE_H +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_DETOKENIZE_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/ec_types.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/ec_types.h index b24bfd9439f..3af5ca86b4b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/ec_types.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/ec_types.h @@ -8,8 +8,12 @@ * be found in the AUTHORS file in the root of the source tree. */ -#ifndef VP8_DEC_EC_TYPES_H -#define VP8_DEC_EC_TYPES_H +#ifndef VP8_DECODER_EC_TYPES_H_ +#define VP8_DECODER_EC_TYPES_H_ + +#ifdef __cplusplus +extern "C" { +#endif #define MAX_OVERLAPS 16 @@ -47,4 +51,8 @@ typedef struct MV_REFERENCE_FRAME ref_frame; } EC_BLOCK; -#endif // VP8_DEC_EC_TYPES_H +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_EC_TYPES_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.c b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.c index 0b58c98fd88..4b304c83c78 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.c @@ -15,9 +15,7 @@ #include "decodemv.h" #include "vpx_mem/vpx_mem.h" #include "vp8/common/findnearmv.h" - -#define MIN(x,y) (((x)<(y))?(x):(y)) -#define MAX(x,y) (((x)>(y))?(x):(y)) +#include "vp8/common/common.h" #define FLOOR(x,q) ((x) & -(1 << (q))) diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.h index fb96b3605ec..9a1e024865d 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/error_concealment.h @@ -9,12 +9,16 @@ */ -#ifndef ERROR_CONCEALMENT_H_ -#define ERROR_CONCEALMENT_H_ +#ifndef VP8_DECODER_ERROR_CONCEALMENT_H_ +#define VP8_DECODER_ERROR_CONCEALMENT_H_ #include "onyxd_int.h" #include "ec_types.h" +#ifdef __cplusplus +extern "C" { +#endif + /* Allocate memory for the overlap lists */ int vp8_alloc_overlap_lists(VP8D_COMP *pbi); @@ -38,4 +42,8 @@ void vp8_interpolate_motion(MACROBLOCKD *mb, */ void vp8_conceal_corrupt_mb(MACROBLOCKD *xd); -#endif // ERROR_CONCEALMENT_H_ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_ERROR_CONCEALMENT_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_if.c b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_if.c index 2d9e343bc9f..29fea616bc8 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_if.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_if.c @@ -178,12 +178,6 @@ vpx_codec_err_t vp8dx_set_reference(VP8D_COMP *pbi, enum vpx_ref_frame_type ref_ return pbi->common.error.error_code; } -/*For ARM NEON, d8-d15 are callee-saved registers, and need to be saved by us.*/ -#if HAVE_NEON -extern void vp8_push_neon(int64_t *store); -extern void vp8_pop_neon(int64_t *store); -#endif - static int get_free_fb (VP8_COMMON *cm) { int i; @@ -307,9 +301,6 @@ int vp8dx_receive_compressed_data(VP8D_COMP *pbi, size_t size, const uint8_t *source, int64_t time_stamp) { -#if HAVE_NEON - int64_t dx_store_reg[8]; -#endif VP8_COMMON *cm = &pbi->common; int retcode = -1; @@ -319,15 +310,6 @@ int vp8dx_receive_compressed_data(VP8D_COMP *pbi, size_t size, if(retcode <= 0) return retcode; -#if HAVE_NEON -#if CONFIG_RUNTIME_CPU_DETECT - if (cm->cpu_caps & HAS_NEON) -#endif - { - vp8_push_neon(dx_store_reg); - } -#endif - cm->new_fb_idx = get_free_fb (cm); /* setup reference frames for vp8_decode_frame */ @@ -403,15 +385,6 @@ int vp8dx_receive_compressed_data(VP8D_COMP *pbi, size_t size, pbi->last_time_stamp = time_stamp; decode_exit: -#if HAVE_NEON -#if CONFIG_RUNTIME_CPU_DETECT - if (cm->cpu_caps & HAS_NEON) -#endif - { - vp8_pop_neon(dx_store_reg); - } -#endif - pbi->common.error.setjmp = 0; return retcode; } diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_int.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_int.h index 54a98f7cc32..aa2cc57f7b1 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_int.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/onyxd_int.h @@ -9,8 +9,8 @@ */ -#ifndef ONYXD_INT_H_ -#define ONYXD_INT_H_ +#ifndef VP8_DECODER_ONYXD_INT_H_ +#define VP8_DECODER_ONYXD_INT_H_ #include "vpx_config.h" #include "vp8/common/onyxd.h" @@ -22,6 +22,10 @@ #include "ec_types.h" #endif +#ifdef __cplusplus +extern "C" { +#endif + typedef struct { int ithread; @@ -122,7 +126,7 @@ typedef struct VP8D_COMP int independent_partitions; int frame_corrupt_residual; - vp8_decrypt_cb *decrypt_cb; + vpx_decrypt_cb decrypt_cb; void *decrypt_state; } VP8D_COMP; @@ -148,4 +152,8 @@ int vp8_remove_decoder_instances(struct frame_buffers *fb); } while(0) #endif -#endif // ONYXD_INT_H_ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_ONYXD_INT_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/treereader.h b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/treereader.h index 9393bb47857..35ee696000d 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/decoder/treereader.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/decoder/treereader.h @@ -9,12 +9,16 @@ */ -#ifndef TREEREADER_H_ -#define TREEREADER_H_ +#ifndef VP8_DECODER_TREEREADER_H_ +#define VP8_DECODER_TREEREADER_H_ #include "vp8/common/treecoder.h" #include "dboolhuff.h" +#ifdef __cplusplus +extern "C" { +#endif + typedef BOOL_DECODER vp8_reader; #define vp8_read vp8dx_decode_bool @@ -37,4 +41,8 @@ static int vp8_treed_read( return -i; } -#endif // TREEREADER_H_ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_DECODER_TREEREADER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/denoising_neon.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/denoising_neon.c new file mode 100644 index 00000000000..32ce65abf4c --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/denoising_neon.c @@ -0,0 +1,161 @@ +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <arm_neon.h> + +#include "vp8/encoder/denoising.h" +#include "vpx_mem/vpx_mem.h" +#include "./vp8_rtcd.h" + +/* + * The filter function was modified to reduce the computational complexity. + * + * Step 1: + * Instead of applying tap coefficients for each pixel, we calculated the + * pixel adjustments vs. pixel diff value ahead of time. + * adjustment = filtered_value - current_raw + * = (filter_coefficient * diff + 128) >> 8 + * where + * filter_coefficient = (255 << 8) / (256 + ((abs_diff * 330) >> 3)); + * filter_coefficient += filter_coefficient / + * (3 + motion_magnitude_adjustment); + * filter_coefficient is clamped to 0 ~ 255. + * + * Step 2: + * The adjustment vs. diff curve becomes flat very quick when diff increases. + * This allowed us to use only several levels to approximate the curve without + * changing the filtering algorithm too much. + * The adjustments were further corrected by checking the motion magnitude. + * The levels used are: + * diff level adjustment w/o adjustment w/ + * motion correction motion correction + * [-255, -16] 3 -6 -7 + * [-15, -8] 2 -4 -5 + * [-7, -4] 1 -3 -4 + * [-3, 3] 0 diff diff + * [4, 7] 1 3 4 + * [8, 15] 2 4 5 + * [16, 255] 3 6 7 + */ + +int vp8_denoiser_filter_neon(unsigned char *mc_running_avg_y, + int mc_running_avg_y_stride, + unsigned char *running_avg_y, + int running_avg_y_stride, + unsigned char *sig, int sig_stride, + unsigned int motion_magnitude, + int increase_denoising) { + /* If motion_magnitude is small, making the denoiser more aggressive by + * increasing the adjustment for each level, level1 adjustment is + * increased, the deltas stay the same. + */ + const uint8x16_t v_level1_adjustment = vdupq_n_u8( + (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 4 : 3); + const uint8x16_t v_delta_level_1_and_2 = vdupq_n_u8(1); + const uint8x16_t v_delta_level_2_and_3 = vdupq_n_u8(2); + const uint8x16_t v_level1_threshold = vdupq_n_u8(4); + const uint8x16_t v_level2_threshold = vdupq_n_u8(8); + const uint8x16_t v_level3_threshold = vdupq_n_u8(16); + int64x2_t v_sum_diff_total = vdupq_n_s64(0); + + /* Go over lines. */ + int i; + for (i = 0; i < 16; ++i) { + /* Load inputs. */ + const uint8x16_t v_sig = vld1q_u8(sig); + const uint8x16_t v_mc_running_avg_y = vld1q_u8(mc_running_avg_y); + + /* Calculate absolute difference and sign masks. */ + const uint8x16_t v_abs_diff = vabdq_u8(v_sig, v_mc_running_avg_y); + const uint8x16_t v_diff_pos_mask = vcltq_u8(v_sig, v_mc_running_avg_y); + const uint8x16_t v_diff_neg_mask = vcgtq_u8(v_sig, v_mc_running_avg_y); + + /* Figure out which level that put us in. */ + const uint8x16_t v_level1_mask = vcleq_u8(v_level1_threshold, + v_abs_diff); + const uint8x16_t v_level2_mask = vcleq_u8(v_level2_threshold, + v_abs_diff); + const uint8x16_t v_level3_mask = vcleq_u8(v_level3_threshold, + v_abs_diff); + + /* Calculate absolute adjustments for level 1, 2 and 3. */ + const uint8x16_t v_level2_adjustment = vandq_u8(v_level2_mask, + v_delta_level_1_and_2); + const uint8x16_t v_level3_adjustment = vandq_u8(v_level3_mask, + v_delta_level_2_and_3); + const uint8x16_t v_level1and2_adjustment = vaddq_u8(v_level1_adjustment, + v_level2_adjustment); + const uint8x16_t v_level1and2and3_adjustment = vaddq_u8( + v_level1and2_adjustment, v_level3_adjustment); + + /* Figure adjustment absolute value by selecting between the absolute + * difference if in level0 or the value for level 1, 2 and 3. + */ + const uint8x16_t v_abs_adjustment = vbslq_u8(v_level1_mask, + v_level1and2and3_adjustment, v_abs_diff); + + /* Calculate positive and negative adjustments. Apply them to the signal + * and accumulate them. Adjustments are less than eight and the maximum + * sum of them (7 * 16) can fit in a signed char. + */ + const uint8x16_t v_pos_adjustment = vandq_u8(v_diff_pos_mask, + v_abs_adjustment); + const uint8x16_t v_neg_adjustment = vandq_u8(v_diff_neg_mask, + v_abs_adjustment); + + uint8x16_t v_running_avg_y = vqaddq_u8(v_sig, v_pos_adjustment); + v_running_avg_y = vqsubq_u8(v_running_avg_y, v_neg_adjustment); + + /* Store results. */ + vst1q_u8(running_avg_y, v_running_avg_y); + + /* Sum all the accumulators to have the sum of all pixel differences + * for this macroblock. + */ + { + const int8x16_t v_sum_diff = + vqsubq_s8(vreinterpretq_s8_u8(v_pos_adjustment), + vreinterpretq_s8_u8(v_neg_adjustment)); + + const int16x8_t fe_dc_ba_98_76_54_32_10 = vpaddlq_s8(v_sum_diff); + + const int32x4_t fedc_ba98_7654_3210 = + vpaddlq_s16(fe_dc_ba_98_76_54_32_10); + + const int64x2_t fedcba98_76543210 = + vpaddlq_s32(fedc_ba98_7654_3210); + + v_sum_diff_total = vqaddq_s64(v_sum_diff_total, fedcba98_76543210); + } + + /* Update pointers for next iteration. */ + sig += sig_stride; + mc_running_avg_y += mc_running_avg_y_stride; + running_avg_y += running_avg_y_stride; + } + + /* Too much adjustments => copy block. */ + { + const int64x1_t x = vqadd_s64(vget_high_s64(v_sum_diff_total), + vget_low_s64(v_sum_diff_total)); + const int s0 = vget_lane_s32(vabs_s32(vreinterpret_s32_s64(x)), 0); + + if (s0 > SUM_DIFF_THRESHOLD) + return COPY_BLOCK; + } + + /* Tell above level that block was filtered. */ + running_avg_y -= running_avg_y_stride * 16; + sig -= sig_stride * 16; + + vp8_copy_mem16x16(running_avg_y, running_avg_y_stride, sig, sig_stride); + + return FILTER_BLOCK; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/subtract_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/subtract_neon.asm index 5bda78678db..840cb33d957 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/subtract_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/subtract_neon.asm @@ -65,8 +65,10 @@ ; unsigned char *pred, int pred_stride) |vp8_subtract_mby_neon| PROC push {r4-r7} + vpush {d8-d15} + mov r12, #4 - ldr r4, [sp, #16] ; pred_stride + ldr r4, [sp, #80] ; pred_stride mov r6, #32 ; "diff" stride x2 add r5, r0, #16 ; second diff pointer @@ -101,6 +103,7 @@ subtract_mby_loop subs r12, r12, #1 bne subtract_mby_loop + vpop {d8-d15} pop {r4-r7} bx lr ENDP @@ -112,9 +115,11 @@ subtract_mby_loop |vp8_subtract_mbuv_neon| PROC push {r4-r7} - ldr r4, [sp, #16] ; upred - ldr r5, [sp, #20] ; vpred - ldr r6, [sp, #24] ; pred_stride + vpush {d8-d15} + + ldr r4, [sp, #80] ; upred + ldr r5, [sp, #84] ; vpred + ldr r6, [sp, #88] ; pred_stride add r0, r0, #512 ; short *udiff = diff + 256; mov r12, #32 ; "diff" stride x2 add r7, r0, #16 ; second diff pointer @@ -191,6 +196,7 @@ subtract_mby_loop vst1.16 {q14}, [r0], r12 vst1.16 {q15}, [r7], r12 + vpop {d8-d15} pop {r4-r7} bx lr diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_memcpy_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_memcpy_neon.asm index 5b9f11e5935..d219e2d1424 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_memcpy_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_memcpy_neon.asm @@ -21,6 +21,7 @@ ;void vp8_memcpy_partial_neon(unsigned char *dst_ptr, unsigned char *src_ptr, ; int sz); |vp8_memcpy_partial_neon| PROC + vpush {d8-d15} ;pld [r1] ;preload pred data ;pld [r1, #128] ;pld [r1, #256] @@ -64,6 +65,7 @@ extra_copy_neon_loop bne extra_copy_neon_loop done_copy_neon_loop + vpop {d8-d15} bx lr ENDP diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm index 55edbf5129e..f82af3ee333 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm @@ -27,6 +27,8 @@ ;from vp8_variance(). |vp8_mse16x16_neon| PROC + vpush {q7} + vmov.i8 q7, #0 ;q7, q8, q9, q10 - sse vmov.i8 q8, #0 vmov.i8 q9, #0 @@ -62,7 +64,7 @@ mse16x16_neon_loop vadd.u32 q7, q7, q8 vadd.u32 q9, q9, q10 - ldr r12, [sp] ;load *sse from stack + ldr r12, [sp, #16] ;load *sse from stack vadd.u32 q10, q7, q9 vpaddl.u32 q1, q10 @@ -71,6 +73,7 @@ mse16x16_neon_loop vst1.32 {d0[0]}, [r12] vmov.32 r0, d0[0] + vpop {q7} bx lr ENDP @@ -82,6 +85,8 @@ mse16x16_neon_loop ; r2 unsigned char *ref_ptr, ; r3 int recon_stride |vp8_get4x4sse_cs_neon| PROC + vpush {q7} + vld1.8 {d0}, [r0], r1 ;Load up source and reference vld1.8 {d4}, [r2], r3 vld1.8 {d1}, [r0], r1 @@ -109,6 +114,8 @@ mse16x16_neon_loop vadd.u64 d0, d2, d3 vmov.32 r0, d0[0] + + vpop {q7} bx lr ENDP diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/bitstream.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/bitstream.h index 455a94fbed3..eef2d79e087 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/bitstream.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/bitstream.h @@ -9,8 +9,12 @@ */ -#ifndef __INC_BITSTREAM_H -#define __INC_BITSTREAM_H +#ifndef VP8_ENCODER_BITSTREAM_H_ +#define VP8_ENCODER_BITSTREAM_H_ + +#ifdef __cplusplus +extern "C" { +#endif #if HAVE_EDSP void vp8cx_pack_tokens_armv5(vp8_writer *w, const TOKENEXTRA *p, int xcount, @@ -43,4 +47,8 @@ void vp8_pack_tokens_c(vp8_writer *w, const TOKENEXTRA *p, int xcount); # define pack_mb_row_tokens(a,b) pack_mb_row_tokens_c(a,b) #endif +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_BITSTREAM_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/block.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/block.h index cf74c7aafc4..34879cf2ab4 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/block.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/block.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_BLOCK_H -#define __INC_BLOCK_H +#ifndef VP8_ENCODER_BLOCK_H_ +#define VP8_ENCODER_BLOCK_H_ #include "vp8/common/onyx.h" #include "vp8/common/blockd.h" @@ -18,6 +18,10 @@ #include "vp8/common/entropy.h" #include "vpx_ports/mem.h" +#ifdef __cplusplus +extern "C" { +#endif + #define MAX_MODES 20 #define MAX_ERROR_BINS 1024 @@ -121,6 +125,7 @@ typedef struct macroblock int optimize; int q_index; + int increase_denoising; #if CONFIG_TEMPORAL_DENOISING MB_PREDICTION_MODE best_sse_inter_mode; @@ -160,4 +165,8 @@ typedef struct macroblock } MACROBLOCK; +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_BLOCK_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/boolhuff.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/boolhuff.h index 39ab586b52b..61142157593 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/boolhuff.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/boolhuff.h @@ -16,12 +16,16 @@ * Description : Bool Coder header file. * ****************************************************************************/ -#ifndef __INC_BOOLHUFF_H -#define __INC_BOOLHUFF_H +#ifndef VP8_ENCODER_BOOLHUFF_H_ +#define VP8_ENCODER_BOOLHUFF_H_ #include "vpx_ports/mem.h" #include "vpx/internal/vpx_codec_internal.h" +#ifdef __cplusplus +extern "C" { +#endif + typedef struct { unsigned int lowvalue; @@ -125,4 +129,8 @@ static void vp8_encode_bool(BOOL_CODER *br, int bit, int probability) br->range = range; } +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_BOOLHUFF_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_cost.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_cost.h index e892765c69f..1cd3eec84ac 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_cost.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_cost.h @@ -8,6 +8,13 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_ENCODER_DCT_VALUE_COST_H_ +#define VP8_ENCODER_DCT_VALUE_COST_H_ + +#ifdef __cplusplus +extern "C" { +#endif + /* Generated file, included by tokenize.c */ /* Values generated by fill_value_tokens() */ @@ -356,3 +363,9 @@ static const short dct_value_cost[2048*2] = 8134, 8140, 8148, 8170, 8178, 8184, 8192, 8202, 8210, 8216, 8224, 8243, 8251, 8257, 8265, 8275 }; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_ENCODER_DCT_VALUE_COST_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_tokens.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_tokens.h index ef08eeddc77..c2aadefca79 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_tokens.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/dct_value_tokens.h @@ -8,6 +8,13 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_ENCODER_DCT_VALUE_TOKENS_H_ +#define VP8_ENCODER_DCT_VALUE_TOKENS_H_ + +#ifdef __cplusplus +extern "C" { +#endif + /* Generated file, included by tokenize.c */ /* Values generated by fill_value_tokens() */ @@ -697,3 +704,9 @@ static const TOKENVALUE dct_value_tokens[2048*2] = {10, 3942}, {10, 3944}, {10, 3946}, {10, 3948}, {10, 3950}, {10, 3952}, {10, 3954}, {10, 3956}, {10, 3958}, {10, 3960} }; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_ENCODER_DCT_VALUE_TOKENS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/defaultcoefcounts.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/defaultcoefcounts.h index 2c0f3ddf3b3..1e8e80484ae 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/defaultcoefcounts.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/defaultcoefcounts.h @@ -8,6 +8,13 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_ENCODER_DEFAULTCOEFCOUNTS_H_ +#define VP8_ENCODER_DEFAULTCOEFCOUNTS_H_ + +#ifdef __cplusplus +extern "C" { +#endif + /* Generated file, included by entropy.c */ static const unsigned int default_coef_counts[BLOCK_TYPES] @@ -221,3 +228,9 @@ static const unsigned int default_coef_counts[BLOCK_TYPES] }, }, }; + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_ENCODER_DEFAULTCOEFCOUNTS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.c index 7819265471e..1e645fbdfff 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.c @@ -21,6 +21,7 @@ static const unsigned int NOISE_MOTION_THRESHOLD = 25 * 25; */ static const unsigned int SSE_DIFF_THRESHOLD = 16 * 16 * 20; static const unsigned int SSE_THRESHOLD = 16 * 16 * 40; +static const unsigned int SSE_THRESHOLD_HIGH = 16 * 16 * 60; /* * The filter function was modified to reduce the computational complexity. @@ -51,27 +52,32 @@ static const unsigned int SSE_THRESHOLD = 16 * 16 * 40; * [16, 255] 6 7 */ -int vp8_denoiser_filter_c(YV12_BUFFER_CONFIG *mc_running_avg, - YV12_BUFFER_CONFIG *running_avg, MACROBLOCK *signal, - unsigned int motion_magnitude, int y_offset, - int uv_offset) +int vp8_denoiser_filter_c(unsigned char *mc_running_avg_y, int mc_avg_y_stride, + unsigned char *running_avg_y, int avg_y_stride, + unsigned char *sig, int sig_stride, + unsigned int motion_magnitude, + int increase_denoising) { - unsigned char *sig = signal->thismb; - int sig_stride = 16; - unsigned char *mc_running_avg_y = mc_running_avg->y_buffer + y_offset; - int mc_avg_y_stride = mc_running_avg->y_stride; - unsigned char *running_avg_y = running_avg->y_buffer + y_offset; - int avg_y_stride = running_avg->y_stride; - int r, c, i; + unsigned char *running_avg_y_start = running_avg_y; + unsigned char *sig_start = sig; + int sum_diff_thresh; + int r, c; int sum_diff = 0; int adj_val[3] = {3, 4, 6}; - + int shift_inc1 = 0; + int shift_inc2 = 1; /* If motion_magnitude is small, making the denoiser more aggressive by - * increasing the adjustment for each level. */ + * increasing the adjustment for each level. Add another increment for + * blocks that are labeled for increase denoising. */ if (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) { - for (i = 0; i < 3; i++) - adj_val[i] += 1; + if (increase_denoising) { + shift_inc1 = 1; + shift_inc2 = 2; + } + adj_val[0] += shift_inc2; + adj_val[1] += shift_inc2; + adj_val[2] += shift_inc2; } for (r = 0; r < 16; ++r) @@ -85,8 +91,9 @@ int vp8_denoiser_filter_c(YV12_BUFFER_CONFIG *mc_running_avg, diff = mc_running_avg_y[c] - sig[c]; absdiff = abs(diff); - /* When |diff| < 4, use pixel value from last denoised raw. */ - if (absdiff <= 3) + // When |diff| <= |3 + shift_inc1|, use pixel value from + // last denoised raw. + if (absdiff <= 3 + shift_inc1) { running_avg_y[c] = mc_running_avg_y[c]; sum_diff += diff; @@ -127,11 +134,12 @@ int vp8_denoiser_filter_c(YV12_BUFFER_CONFIG *mc_running_avg, running_avg_y += avg_y_stride; } - if (abs(sum_diff) > SUM_DIFF_THRESHOLD) + sum_diff_thresh= SUM_DIFF_THRESHOLD; + if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH; + if (abs(sum_diff) > sum_diff_thresh) return COPY_BLOCK; - vp8_copy_mem16x16(running_avg->y_buffer + y_offset, avg_y_stride, - signal->thismb, sig_stride); + vp8_copy_mem16x16(running_avg_y_start, avg_y_stride, sig_start, sig_stride); return FILTER_BLOCK; } @@ -192,7 +200,7 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser, int mv_row; int mv_col; unsigned int motion_magnitude2; - + unsigned int sse_thresh; MV_REFERENCE_FRAME frame = x->best_reference_frame; MV_REFERENCE_FRAME zero_frame = x->best_zeromv_reference_frame; @@ -277,7 +285,10 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser, mv_row = x->best_sse_mv.as_mv.row; mv_col = x->best_sse_mv.as_mv.col; motion_magnitude2 = mv_row * mv_row + mv_col * mv_col; - if (best_sse > SSE_THRESHOLD || motion_magnitude2 + sse_thresh = SSE_THRESHOLD; + if (x->increase_denoising) sse_thresh = SSE_THRESHOLD_HIGH; + + if (best_sse > sse_thresh || motion_magnitude2 > 8 * NOISE_MOTION_THRESHOLD) { decision = COPY_BLOCK; @@ -285,12 +296,18 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser, if (decision == FILTER_BLOCK) { + unsigned char *mc_running_avg_y = + denoiser->yv12_mc_running_avg.y_buffer + recon_yoffset; + int mc_avg_y_stride = denoiser->yv12_mc_running_avg.y_stride; + unsigned char *running_avg_y = + denoiser->yv12_running_avg[INTRA_FRAME].y_buffer + recon_yoffset; + int avg_y_stride = denoiser->yv12_running_avg[INTRA_FRAME].y_stride; + /* Filter. */ - decision = vp8_denoiser_filter(&denoiser->yv12_mc_running_avg, - &denoiser->yv12_running_avg[INTRA_FRAME], - x, - motion_magnitude2, - recon_yoffset, recon_uvoffset); + decision = vp8_denoiser_filter(mc_running_avg_y, mc_avg_y_stride, + running_avg_y, avg_y_stride, + x->thismb, 16, motion_magnitude2, + x->increase_denoising); } if (decision == COPY_BLOCK) { diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.h index b025f5cdf01..ae744d2efc0 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/denoising.h @@ -13,7 +13,12 @@ #include "block.h" +#ifdef __cplusplus +extern "C" { +#endif + #define SUM_DIFF_THRESHOLD (16 * 16 * 2) +#define SUM_DIFF_THRESHOLD_HIGH (16 * 16 * 3) #define MOTION_MAGNITUDE_THRESHOLD (8*3) enum vp8_denoiser_decision @@ -39,4 +44,8 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser, int recon_yoffset, int recon_uvoffset); -#endif /* VP8_ENCODER_DENOISING_H_ */ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_ENCODER_DENOISING_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeframe.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeframe.h index 4dd6ba0de18..e185c1035c0 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeframe.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeframe.h @@ -7,8 +7,12 @@ * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree. */ -#ifndef ENCODEFRAME_H -#define ENCODEFRAME_H +#ifndef VP8_ENCODER_ENCODEFRAME_H_ +#define VP8_ENCODER_ENCODEFRAME_H_ + +#ifdef __cplusplus +extern "C" { +#endif extern void vp8_activity_masking(VP8_COMP *cpi, MACROBLOCK *x); extern void vp8_build_block_offsets(MACROBLOCK *x); @@ -24,4 +28,8 @@ extern int vp8cx_encode_inter_macroblock(VP8_COMP *cpi, MACROBLOCK *x, extern int vp8cx_encode_intra_macroblock(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_ENCODEFRAME_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeintra.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeintra.h index be2141f2c63..a8d0284d29b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeintra.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodeintra.h @@ -9,13 +9,21 @@ */ -#ifndef _ENCODEINTRA_H_ -#define _ENCODEINTRA_H_ +#ifndef VP8_ENCODER_ENCODEINTRA_H_ +#define VP8_ENCODER_ENCODEINTRA_H_ #include "onyx_int.h" +#ifdef __cplusplus +extern "C" { +#endif + int vp8_encode_intra(VP8_COMP *cpi, MACROBLOCK *x, int use_dc_pred); void vp8_encode_intra16x16mby(MACROBLOCK *x); void vp8_encode_intra16x16mbuv(MACROBLOCK *x); void vp8_encode_intra4x4mby(MACROBLOCK *mb); void vp8_encode_intra4x4block(MACROBLOCK *x, int ib); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_ENCODEINTRA_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemb.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemb.h index 6badf7d901a..0b3ec875e7a 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemb.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemb.h @@ -9,10 +9,14 @@ */ -#ifndef __INC_ENCODEMB_H -#define __INC_ENCODEMB_H +#ifndef VP8_ENCODER_ENCODEMB_H_ +#define VP8_ENCODER_ENCODEMB_H_ #include "onyx_int.h" + +#ifdef __cplusplus +extern "C" { +#endif void vp8_encode_inter16x16(MACROBLOCK *x); void vp8_build_dcblock(MACROBLOCK *b); @@ -23,4 +27,8 @@ void vp8_transform_intra_mby(MACROBLOCK *x); void vp8_optimize_mby(MACROBLOCK *x); void vp8_optimize_mbuv(MACROBLOCK *x); void vp8_encode_inter16x16y(MACROBLOCK *x); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_ENCODEMB_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemv.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemv.h index a6116c133d0..722162ba212 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemv.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/encodemv.h @@ -9,13 +9,21 @@ */ -#ifndef __INC_ENCODEMV_H -#define __INC_ENCODEMV_H +#ifndef VP8_ENCODER_ENCODEMV_H_ +#define VP8_ENCODER_ENCODEMV_H_ #include "onyx_int.h" +#ifdef __cplusplus +extern "C" { +#endif + void vp8_write_mvprobs(VP8_COMP *); void vp8_encode_motion_vector(vp8_writer *, const MV *, const MV_CONTEXT *); void vp8_build_component_cost_table(int *mvcost[2], const MV_CONTEXT *mvc, int mvc_flag[2]); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_ENCODEMV_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.c index 968c7f3652e..98e5a7115db 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.c @@ -940,9 +940,9 @@ static int64_t estimate_modemvcost(VP8_COMP *cpi, /* Crude estimate of overhead cost from modes * << 9 is the normalization to (bits * 512) used in vp8_bits_per_mb */ - mode_cost =((((av_pct_inter - av_pct_motion) * zz_cost) + - (av_pct_motion * motion_cost) + - (av_intra * intra_cost)) * cpi->common.MBs) * 512; + mode_cost = (int64_t)((((av_pct_inter - av_pct_motion) * zz_cost) + + (av_pct_motion * motion_cost) + + (av_intra * intra_cost)) * cpi->common.MBs) * 512; return mv_cost + mode_cost; } @@ -2310,7 +2310,7 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) pct_extra = (pct_extra > 20) ? 20 : pct_extra; cpi->twopass.alt_extra_bits = - (cpi->twopass.gf_group_bits * pct_extra) / 100; + (int)(cpi->twopass.gf_group_bits * pct_extra) / 100; cpi->twopass.gf_group_bits -= cpi->twopass.alt_extra_bits; cpi->twopass.alt_extra_bits /= ((cpi->baseline_gf_interval-1)>>1); @@ -2386,7 +2386,7 @@ static void assign_std_frame_bits(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) target_frame_size = max_bits; if (target_frame_size > cpi->twopass.gf_group_bits) - target_frame_size = cpi->twopass.gf_group_bits; + target_frame_size = (int)cpi->twopass.gf_group_bits; } /* Adjust error and bits remaining */ @@ -2444,10 +2444,10 @@ void vp8_second_pass(VP8_COMP *cpi) find_next_key_frame(cpi, &this_frame_copy); /* Special case: Error error_resilient_mode mode does not make much - * sense for two pass but with its current meaning but this code is + * sense for two pass but with its current meaning this code is * designed to stop outlandish behaviour if someone does set it when * using two pass. It effectively disables GF groups. This is - * temporary code till we decide what should really happen in this + * temporary code until we decide what should really happen in this * case. */ if (cpi->oxcf.error_resilient_mode) @@ -2773,7 +2773,7 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) kf_group_intra_err += this_frame->intra_error; kf_group_coded_err += this_frame->coded_error; - /* load a the next frame's stats */ + /* Load the next frame's stats. */ vpx_memcpy(&last_frame, this_frame, sizeof(*this_frame)); input_stats(cpi, this_frame); diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.h index 95e1e5463c6..c409ebca8f1 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/firstpass.h @@ -9,8 +9,12 @@ */ -#if !defined __INC_FIRSTPASS_H -#define __INC_FIRSTPASS_H +#ifndef VP8_ENCODER_FIRSTPASS_H_ +#define VP8_ENCODER_FIRSTPASS_H_ + +#ifdef __cplusplus +extern "C" { +#endif extern void vp8_init_first_pass(VP8_COMP *cpi); extern void vp8_first_pass(VP8_COMP *cpi); @@ -21,4 +25,8 @@ extern void vp8_second_pass(VP8_COMP *cpi); extern void vp8_end_second_pass(VP8_COMP *cpi); extern size_t vp8_firstpass_stats_sz(unsigned int mb_count); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_FIRSTPASS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/lookahead.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/lookahead.h index cf56b75b7d5..cad68e639fb 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/lookahead.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/lookahead.h @@ -7,11 +7,15 @@ * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree. */ -#ifndef LOOKAHEAD_H -#define LOOKAHEAD_H +#ifndef VP8_ENCODER_LOOKAHEAD_H_ +#define VP8_ENCODER_LOOKAHEAD_H_ #include "vpx_scale/yv12config.h" #include "vpx/vpx_integer.h" +#ifdef __cplusplus +extern "C" { +#endif + struct lookahead_entry { YV12_BUFFER_CONFIG img; @@ -106,4 +110,8 @@ unsigned int vp8_lookahead_depth(struct lookahead_ctx *ctx); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_LOOKAHEAD_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.c index 0b11ea64a65..54abe76acd2 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.c @@ -17,6 +17,7 @@ #include <limits.h> #include <math.h> #include "vp8/common/findnearmv.h" +#include "vp8/common/common.h" #ifdef VP8_ENTROPY_STATS static int mv_ref_ct [31] [4] [2]; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.h index e36c51543cb..f284f7c3807 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mcomp.h @@ -9,12 +9,16 @@ */ -#ifndef __INC_MCOMP_H -#define __INC_MCOMP_H +#ifndef VP8_ENCODER_MCOMP_H_ +#define VP8_ENCODER_MCOMP_H_ #include "block.h" #include "vp8/common/variance.h" +#ifdef __cplusplus +extern "C" { +#endif + #ifdef VP8_ENTROPY_STATS extern void init_mv_ref_counts(); extern void accum_mv_refs(MB_PREDICTION_MODE, const int near_mv_ref_cts[4]); @@ -104,4 +108,8 @@ typedef int (*vp8_diamond_search_fn_t) int_mv *center_mv ); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_MCOMP_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/modecosts.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/modecosts.h index 99ef119d5dd..9281551c8d5 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/modecosts.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/modecosts.h @@ -9,9 +9,17 @@ */ -#ifndef __INC_MODECOSTS_H -#define __INC_MODECOSTS_H +#ifndef VP8_ENCODER_MODECOSTS_H_ +#define VP8_ENCODER_MODECOSTS_H_ + +#ifdef __cplusplus +extern "C" { +#endif void vp8_init_mode_costs(VP8_COMP *x); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_MODECOSTS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.c index 71218cca146..8d96445f53c 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.c @@ -15,6 +15,7 @@ #include "mr_dissim.h" #include "vpx_mem/vpx_mem.h" #include "rdopt.h" +#include "vp8/common/common.h" void vp8_cal_low_res_mb_cols(VP8_COMP *cpi) { diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.h index f8cb135d52c..5a59ce62a61 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/mr_dissim.h @@ -9,12 +9,20 @@ */ -#ifndef __INC_MR_DISSIM_H -#define __INC_MR_DISSIM_H +#ifndef VP8_ENCODER_MR_DISSIM_H_ +#define VP8_ENCODER_MR_DISSIM_H_ #include "vpx_config.h" +#ifdef __cplusplus +extern "C" { +#endif + extern void vp8_cal_low_res_mb_cols(VP8_COMP *cpi); extern void vp8_cal_dissimilarity(VP8_COMP *cpi); extern void vp8_store_drop_frame_info(VP8_COMP *cpi); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_MR_DISSIM_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c index 7c079759971..e95e44fd521 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c @@ -19,7 +19,7 @@ #include "vp8/common/alloccommon.h" #include "mcomp.h" #include "firstpass.h" -#include "psnr.h" +#include "vpx/internal/vpx_psnr.h" #include "vpx_scale/vpx_scale.h" #include "vp8/common/extend.h" #include "ratectrl.h" @@ -1401,6 +1401,7 @@ static void update_layer_contexts (VP8_COMP *cpi) unsigned int i; double prev_layer_framerate=0; + assert(oxcf->number_of_layers <= VPX_TS_MAX_LAYERS); for (i=0; i<oxcf->number_of_layers; i++) { LAYER_CONTEXT *lc = &cpi->layer_context[i]; @@ -1623,6 +1624,12 @@ void vp8_change_config(VP8_COMP *cpi, VP8_CONFIG *oxcf) cpi->oxcf.maximum_buffer_size = rescale((int)cpi->oxcf.maximum_buffer_size, cpi->oxcf.target_bandwidth, 1000); + // Under a configuration change, where maximum_buffer_size may change, + // keep buffer level clipped to the maximum allowed buffer size. + if (cpi->bits_off_target > cpi->oxcf.maximum_buffer_size) { + cpi->bits_off_target = cpi->oxcf.maximum_buffer_size; + cpi->buffer_level = cpi->bits_off_target; + } /* Set up frame rate and related parameters rate control values. */ vp8_new_framerate(cpi, cpi->framerate); @@ -1754,8 +1761,11 @@ void vp8_change_config(VP8_COMP *cpi, VP8_CONFIG *oxcf) } +#ifndef M_LOG2_E #define M_LOG2_E 0.693147180559945309417 +#endif #define log2f(x) (log (x) / (float) M_LOG2_E) + static void cal_mvsadcosts(int *mvsadcost[2]) { int i = 1; @@ -2164,10 +2174,12 @@ void vp8_remove_compressor(VP8_COMP **ptr) 8.0 / 1000.0 / time_encoded; double samples = 3.0 / 2 * cpi->frames_in_layer[i] * lst_yv12->y_width * lst_yv12->y_height; - double total_psnr = vp8_mse2psnr(samples, 255.0, - cpi->total_error2[i]); - double total_psnr2 = vp8_mse2psnr(samples, 255.0, - cpi->total_error2_p[i]); + double total_psnr = + vpx_sse_to_psnr(samples, 255.0, + cpi->total_error2[i]); + double total_psnr2 = + vpx_sse_to_psnr(samples, 255.0, + cpi->total_error2_p[i]); double total_ssim = 100 * pow(cpi->sum_ssim[i] / cpi->sum_weights[i], 8.0); @@ -2184,9 +2196,9 @@ void vp8_remove_compressor(VP8_COMP **ptr) { double samples = 3.0 / 2 * cpi->count * lst_yv12->y_width * lst_yv12->y_height; - double total_psnr = vp8_mse2psnr(samples, 255.0, - cpi->total_sq_error); - double total_psnr2 = vp8_mse2psnr(samples, 255.0, + double total_psnr = vpx_sse_to_psnr(samples, 255.0, + cpi->total_sq_error); + double total_psnr2 = vpx_sse_to_psnr(samples, 255.0, cpi->total_sq_error2); double total_ssim = 100 * pow(cpi->summed_quality / cpi->summed_weights, 8.0); @@ -2516,8 +2528,8 @@ static void generate_psnr_packet(VP8_COMP *cpi) pkt.data.psnr.samples[3] = width * height; for (i = 0; i < 4; i++) - pkt.data.psnr.psnr[i] = vp8_mse2psnr(pkt.data.psnr.samples[i], 255.0, - (double)(pkt.data.psnr.sse[i])); + pkt.data.psnr.psnr[i] = vpx_sse_to_psnr(pkt.data.psnr.samples[i], 255.0, + (double)(pkt.data.psnr.sse[i])); vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt); } @@ -2675,8 +2687,8 @@ static int resize_key_frame(VP8_COMP *cpi) VP8_COMMON *cm = &cpi->common; /* Do we need to apply resampling for one pass cbr. - * In one pass this is more limited than in two pass cbr - * The test and any change is only made one per key frame sequence + * In one pass this is more limited than in two pass cbr. + * The test and any change is only made once per key frame sequence. */ if (cpi->oxcf.allow_spatial_resampling && (cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER)) { @@ -2699,7 +2711,7 @@ static int resize_key_frame(VP8_COMP *cpi) cm->vert_scale = (cm->vert_scale > NORMAL) ? cm->vert_scale - 1 : NORMAL; } - /* Get the new hieght and width */ + /* Get the new height and width */ Scale2Ratio(cm->horiz_scale, &hr, &hs); Scale2Ratio(cm->vert_scale, &vr, &vs); new_width = ((hs - 1) + (cpi->oxcf.Width * hr)) / hs; @@ -3574,7 +3586,8 @@ static void encode_frame_to_data_rate for (i=cpi->current_layer+1; i<cpi->oxcf.number_of_layers; i++) { LAYER_CONTEXT *lc = &cpi->layer_context[i]; - lc->bits_off_target += cpi->av_per_frame_bandwidth; + lc->bits_off_target += (int)(lc->target_bandwidth / + lc->framerate); if (lc->bits_off_target > lc->maximum_buffer_size) lc->bits_off_target = lc->maximum_buffer_size; lc->buffer_level = lc->bits_off_target; @@ -3807,7 +3820,7 @@ static void encode_frame_to_data_rate /* Setup background Q adjustment for error resilient mode. * For multi-layer encodes only enable this for the base layer. - */ + */ if (cpi->cyclic_refresh_mode_enabled) { if (cpi->current_layer==0) @@ -4620,45 +4633,43 @@ static void encode_frame_to_data_rate vp8_clear_system_state(); if (cpi->twopass.total_left_stats.coded_error != 0.0) - fprintf(f, "%10d %10d %10d %10d %10d %10d %10d %10d %10d %6d %6d" - "%6d %6d %6d %5d %5d %5d %8d %8.2f %10d %10.3f" - "%10.3f %8d\n", + fprintf(f, "%10d %10d %10d %10d %10d %10"PRId64" %10"PRId64 + "%10"PRId64" %10d %6d %6d %6d %6d %5d %5d %5d %8d " + "%8.2lf %"PRId64" %10.3lf %10"PRId64" %8d\n", cpi->common.current_video_frame, cpi->this_frame_target, cpi->projected_frame_size, (cpi->projected_frame_size - cpi->this_frame_target), - (int)cpi->total_target_vs_actual, + cpi->total_target_vs_actual, cpi->buffer_level, (cpi->oxcf.starting_buffer_level-cpi->bits_off_target), - (int)cpi->total_actual_bits, cm->base_qindex, + cpi->total_actual_bits, cm->base_qindex, cpi->active_best_quality, cpi->active_worst_quality, cpi->ni_av_qi, cpi->cq_target_quality, - cpi->zbin_over_quant, cm->refresh_golden_frame, cm->refresh_alt_ref_frame, cm->frame_type, cpi->gfu_boost, cpi->twopass.est_max_qcorrection_factor, - (int)cpi->twopass.bits_left, + cpi->twopass.bits_left, cpi->twopass.total_left_stats.coded_error, (double)cpi->twopass.bits_left / cpi->twopass.total_left_stats.coded_error, cpi->tot_recode_hits); else - fprintf(f, "%10d %10d %10d %10d %10d %10d %10d %10d %10d %6d %6d" - "%6d %6d %6d %5d %5d %5d %8d %8.2f %10d %10.3f" - "%8d\n", - cpi->common.current_video_frame, - cpi->this_frame_target, cpi->projected_frame_size, + fprintf(f, "%10d %10d %10d %10d %10d %10"PRId64" %10"PRId64 + "%10"PRId64" %10d %6d %6d %6d %6d %5d %5d %5d %8d " + "%8.2lf %"PRId64" %10.3lf %8d\n", + cpi->common.current_video_frame, cpi->this_frame_target, + cpi->projected_frame_size, (cpi->projected_frame_size - cpi->this_frame_target), - (int)cpi->total_target_vs_actual, + cpi->total_target_vs_actual, cpi->buffer_level, (cpi->oxcf.starting_buffer_level-cpi->bits_off_target), - (int)cpi->total_actual_bits, cm->base_qindex, + cpi->total_actual_bits, cm->base_qindex, cpi->active_best_quality, cpi->active_worst_quality, cpi->ni_av_qi, cpi->cq_target_quality, - cpi->zbin_over_quant, cm->refresh_golden_frame, cm->refresh_alt_ref_frame, cm->frame_type, cpi->gfu_boost, cpi->twopass.est_max_qcorrection_factor, - (int)cpi->twopass.bits_left, + cpi->twopass.bits_left, cpi->twopass.total_left_stats.coded_error, cpi->tot_recode_hits); @@ -4666,7 +4677,6 @@ static void encode_frame_to_data_rate { FILE *fmodes = fopen("Modes.stt", "a"); - int i; fprintf(fmodes, "%6d:%1d:%1d:%1d ", cpi->common.current_video_frame, @@ -4810,33 +4820,11 @@ static void Pass2Encode(VP8_COMP *cpi, unsigned long *size, unsigned char *dest, } #endif -/* For ARM NEON, d8-d15 are callee-saved registers, and need to be saved. */ -#if HAVE_NEON -extern void vp8_push_neon(int64_t *store); -extern void vp8_pop_neon(int64_t *store); -#endif - - int vp8_receive_raw_frame(VP8_COMP *cpi, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, int64_t time_stamp, int64_t end_time) { -#if HAVE_NEON - int64_t store_reg[8]; -#if CONFIG_RUNTIME_CPU_DETECT - VP8_COMMON *cm = &cpi->common; -#endif -#endif struct vpx_usec_timer timer; int res = 0; -#if HAVE_NEON -#if CONFIG_RUNTIME_CPU_DETECT - if (cm->cpu_caps & HAS_NEON) -#endif - { - vp8_push_neon(store_reg); - } -#endif - vpx_usec_timer_start(&timer); /* Reinit the lookahead buffer if the frame size changes */ @@ -4853,15 +4841,6 @@ int vp8_receive_raw_frame(VP8_COMP *cpi, unsigned int frame_flags, YV12_BUFFER_C vpx_usec_timer_mark(&timer); cpi->time_receive_data += vpx_usec_timer_elapsed(&timer); -#if HAVE_NEON -#if CONFIG_RUNTIME_CPU_DETECT - if (cm->cpu_caps & HAS_NEON) -#endif - { - vp8_pop_neon(store_reg); - } -#endif - return res; } @@ -4882,9 +4861,6 @@ static int frame_is_reference(const VP8_COMP *cpi) int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, unsigned char *dest_end, int64_t *time_stamp, int64_t *time_end, int flush) { -#if HAVE_NEON - int64_t store_reg[8]; -#endif VP8_COMMON *cm; struct vpx_usec_timer tsctimer; struct vpx_usec_timer ticktimer; @@ -4904,15 +4880,6 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l cpi->common.error.setjmp = 1; -#if HAVE_NEON -#if CONFIG_RUNTIME_CPU_DETECT - if (cm->cpu_caps & HAS_NEON) -#endif - { - vp8_push_neon(store_reg); - } -#endif - vpx_usec_timer_start(&cmptimer); cpi->source = NULL; @@ -4995,14 +4962,6 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l #endif -#if HAVE_NEON -#if CONFIG_RUNTIME_CPU_DETECT - if (cm->cpu_caps & HAS_NEON) -#endif - { - vp8_pop_neon(store_reg); - } -#endif return -1; } @@ -5065,6 +5024,7 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l unsigned int i; /* Update frame rates for each layer */ + assert(cpi->oxcf.number_of_layers <= VPX_TS_MAX_LAYERS); for (i=0; i<cpi->oxcf.number_of_layers; i++) { LAYER_CONTEXT *lc = &cpi->layer_context[i]; @@ -5267,7 +5227,7 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l int y_samples = orig->y_height * orig->y_width ; int uv_samples = orig->uv_height * orig->uv_width ; int t_samples = y_samples + 2 * uv_samples; - double sq_error, sq_error2; + double sq_error; ye = calc_plane_error(orig->y_buffer, orig->y_stride, recon->y_buffer, recon->y_stride, orig->y_width, orig->y_height); @@ -5280,16 +5240,17 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l sq_error = (double)(ye + ue + ve); - frame_psnr = vp8_mse2psnr(t_samples, 255.0, sq_error); + frame_psnr = vpx_sse_to_psnr(t_samples, 255.0, sq_error); - cpi->total_y += vp8_mse2psnr(y_samples, 255.0, (double)ye); - cpi->total_u += vp8_mse2psnr(uv_samples, 255.0, (double)ue); - cpi->total_v += vp8_mse2psnr(uv_samples, 255.0, (double)ve); + cpi->total_y += vpx_sse_to_psnr(y_samples, 255.0, (double)ye); + cpi->total_u += vpx_sse_to_psnr(uv_samples, 255.0, (double)ue); + cpi->total_v += vpx_sse_to_psnr(uv_samples, 255.0, (double)ve); cpi->total_sq_error += sq_error; cpi->total += frame_psnr; #if CONFIG_POSTPROC { YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer; + double sq_error2; double frame_psnr2, frame_ssim2 = 0; double weight = 0; @@ -5307,14 +5268,14 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l sq_error2 = (double)(ye + ue + ve); - frame_psnr2 = vp8_mse2psnr(t_samples, 255.0, sq_error2); + frame_psnr2 = vpx_sse_to_psnr(t_samples, 255.0, sq_error2); - cpi->totalp_y += vp8_mse2psnr(y_samples, - 255.0, (double)ye); - cpi->totalp_u += vp8_mse2psnr(uv_samples, - 255.0, (double)ue); - cpi->totalp_v += vp8_mse2psnr(uv_samples, - 255.0, (double)ve); + cpi->totalp_y += vpx_sse_to_psnr(y_samples, + 255.0, (double)ye); + cpi->totalp_u += vpx_sse_to_psnr(uv_samples, + 255.0, (double)ue); + cpi->totalp_v += vpx_sse_to_psnr(uv_samples, + 255.0, (double)ve); cpi->total_sq_error2 += sq_error2; cpi->totalp += frame_psnr2; @@ -5405,15 +5366,6 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l #endif #endif -#if HAVE_NEON -#if CONFIG_RUNTIME_CPU_DETECT - if (cm->cpu_caps & HAS_NEON) -#endif - { - vp8_pop_neon(store_reg); - } -#endif - cpi->common.error.setjmp = 0; return 0; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_int.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_int.h index 3ab0fe8bfa1..df17dff3478 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_int.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_int.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_VP8_INT_H -#define __INC_VP8_INT_H +#ifndef VP8_ENCODER_ONYX_INT_H_ +#define VP8_ENCODER_ONYX_INT_H_ #include <stdio.h> #include "vpx_config.h" @@ -33,6 +33,10 @@ #include "vp8/encoder/denoising.h" #endif +#ifdef __cplusplus +extern "C" { +#endif + #define MIN_GF_INTERVAL 4 #define DEFAULT_GF_INTERVAL 7 @@ -57,9 +61,6 @@ #define VP8_TEMPORAL_ALT_REF 1 #endif -#define MAX(x,y) (((x)>(y))?(x):(y)) -#define MIN(x,y) (((x)<(y))?(x):(y)) - typedef struct { int kf_indicated; @@ -721,4 +722,8 @@ void vp8_set_speed_features(VP8_COMP *cpi); "Failed to allocate "#lval);\ } while(0) #endif +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_ONYX_INT_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.c index c5279fed2a8..cf6a82f5a8f 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.c @@ -14,6 +14,7 @@ #include "onyx_int.h" #include "modecosts.h" #include "encodeintra.h" +#include "vp8/common/common.h" #include "vp8/common/entropymode.h" #include "pickinter.h" #include "vp8/common/findnearmv.h" @@ -1176,6 +1177,7 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, x->best_reference_frame = best_mbmode.ref_frame; best_sse = best_rd_sse; } + x->increase_denoising = 0; vp8_denoiser_denoise_mb(&cpi->denoiser, x, best_sse, zero_mv_sse, recon_yoffset, recon_uvoffset); diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.h index 35011cab358..cf3b1f8d49c 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/pickinter.h @@ -9,11 +9,15 @@ */ -#ifndef __INC_PICKINTER_H -#define __INC_PICKINTER_H +#ifndef VP8_ENCODER_PICKINTER_H_ +#define VP8_ENCODER_PICKINTER_H_ #include "vpx_config.h" #include "vp8/common/onyxc_int.h" +#ifdef __cplusplus +extern "C" { +#endif + extern void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int recon_uvoffset, int *returnrate, int *returndistortion, int *returnintra, @@ -24,4 +28,8 @@ extern int vp8_get_inter_mbpred_error(MACROBLOCK *mb, const vp8_variance_fn_ptr_t *vfp, unsigned int *sse, int_mv this_mv); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_PICKINTER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/psnr.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/psnr.c deleted file mode 100644 index b3a3d955290..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/psnr.c +++ /dev/null @@ -1,31 +0,0 @@ -/* - * Copyright (c) 2010 The WebM project authors. All Rights Reserved. - * - * Use of this source code is governed by a BSD-style license - * that can be found in the LICENSE 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. - */ - - -#include "vpx_scale/yv12config.h" -#include "math.h" -#include "vp8/common/systemdependent.h" /* for vp8_clear_system_state() */ - -#define MAX_PSNR 100 - -double vp8_mse2psnr(double Samples, double Peak, double Mse) -{ - double psnr; - - if ((double)Mse > 0.0) - psnr = 10.0 * log10(Peak * Peak * Samples / Mse); - else - psnr = MAX_PSNR; /* Limit to prevent / 0 */ - - if (psnr > MAX_PSNR) - psnr = MAX_PSNR; - - return psnr; -} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/psnr.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/psnr.h deleted file mode 100644 index 7f6269abe4c..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/psnr.h +++ /dev/null @@ -1,17 +0,0 @@ -/* - * Copyright (c) 2010 The WebM project authors. All Rights Reserved. - * - * Use of this source code is governed by a BSD-style license - * that can be found in the LICENSE 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. - */ - - -#ifndef __INC_PSNR_H -#define __INC_PSNR_H - -extern double vp8_mse2psnr(double Samples, double Peak, double Mse); - -#endif diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/quantize.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/quantize.h index d55496c5ffe..c739b2627b6 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/quantize.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/quantize.h @@ -9,8 +9,12 @@ */ -#ifndef __INC_QUANTIZE_H -#define __INC_QUANTIZE_H +#ifndef VP8_ENCODER_QUANTIZE_H_ +#define VP8_ENCODER_QUANTIZE_H_ + +#ifdef __cplusplus +extern "C" { +#endif struct VP8_COMP; struct macroblock; @@ -20,4 +24,8 @@ extern void vp8_update_zbin_extra(struct VP8_COMP *cpi, struct macroblock *x); extern void vp8cx_mb_init_quantizer(struct VP8_COMP *cpi, struct macroblock *x, int ok_to_skip); extern void vp8cx_init_quantizer(struct VP8_COMP *cpi); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_QUANTIZE_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.c index 1e8259cf488..c51650c3c26 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.c @@ -174,14 +174,6 @@ static const int kf_gf_boost_qlimits[QINDEX_RANGE] = 600, 600, 600, 600, 600, 600, 600, 600, }; -/* % adjustment to target kf size based on seperation from previous frame */ -static const int kf_boost_seperation_adjustment[16] = -{ - 30, 40, 50, 55, 60, 65, 70, 75, - 80, 85, 90, 95, 100, 100, 100, 100, -}; - - static const int gf_adjust_table[101] = { 100, @@ -956,6 +948,21 @@ static void calc_pframe_target_size(VP8_COMP *cpi) if (cpi->bits_off_target > cpi->oxcf.maximum_buffer_size) cpi->bits_off_target = (int)cpi->oxcf.maximum_buffer_size; cpi->buffer_level = cpi->bits_off_target; + + if (cpi->oxcf.number_of_layers > 1) { + unsigned int i; + + // Propagate bits saved by dropping the frame to higher layers. + for (i = cpi->current_layer + 1; i < cpi->oxcf.number_of_layers; + i++) { + LAYER_CONTEXT *lc = &cpi->layer_context[i]; + lc->bits_off_target += (int)(lc->target_bandwidth / + lc->framerate); + if (lc->bits_off_target > lc->maximum_buffer_size) + lc->bits_off_target = lc->maximum_buffer_size; + lc->buffer_level = lc->bits_off_target; + } + } } } @@ -1223,7 +1230,6 @@ int vp8_regulate_q(VP8_COMP *cpi, int target_bits_per_frame) { Q = cpi->oxcf.gold_q; } - } else { diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.h index c43f08d6dd2..829697f391f 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/ratectrl.h @@ -9,10 +9,15 @@ */ -#if !defined __INC_RATECTRL_H +#ifndef VP8_ENCODER_RATECTRL_H_ +#define VP8_ENCODER_RATECTRL_H_ #include "onyx_int.h" +#ifdef __cplusplus +extern "C" { +#endif + extern void vp8_save_coding_context(VP8_COMP *cpi); extern void vp8_restore_coding_context(VP8_COMP *cpi); @@ -25,4 +30,8 @@ extern void vp8_compute_frame_size_bounds(VP8_COMP *cpi, int *frame_under_shoot_ /* return of 0 means drop frame */ extern int vp8_pick_frame_size(VP8_COMP *cpi); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_RATECTRL_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.c index 5016cc4220f..387701c5733 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.c @@ -528,19 +528,16 @@ static int cost_coeffs(MACROBLOCK *mb, BLOCKD *b, int type, ENTROPY_CONTEXT *a, VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l); -# define QC( I) ( qcoeff_ptr [vp8_default_zig_zag1d[I]] ) - + assert(eob <= 16); for (; c < eob; c++) { - int v = QC(c); - int t = vp8_dct_value_tokens_ptr[v].Token; + const int v = qcoeff_ptr[vp8_default_zig_zag1d[c]]; + const int t = vp8_dct_value_tokens_ptr[v].Token; cost += mb->token_costs [type] [vp8_coef_bands[c]] [pt] [t]; cost += vp8_dct_value_cost_ptr[v]; pt = vp8_prev_token_class[t]; } -# undef QC - if (c < 16) cost += mb->token_costs [type] [vp8_coef_bands[c]] [pt] [DCT_EOB_TOKEN]; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.h index 1e11fa77dfc..fe21b8e283e 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/rdopt.h @@ -9,8 +9,12 @@ */ -#ifndef __INC_RDOPT_H -#define __INC_RDOPT_H +#ifndef VP8_ENCODER_RDOPT_H_ +#define VP8_ENCODER_RDOPT_H_ + +#ifdef __cplusplus +extern "C" { +#endif #define RDCOST(RM,DM,R,D) ( ((128+(R)*(RM)) >> 8) + (DM)*(D) ) @@ -130,4 +134,8 @@ extern void vp8_mv_pred ); void vp8_cal_sad(VP8_COMP *cpi, MACROBLOCKD *xd, MACROBLOCK *x, int recon_yoffset, int near_sadidx[]); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_RDOPT_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/segmentation.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/segmentation.h index 12815b087f4..6b5500594e6 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/segmentation.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/segmentation.h @@ -8,9 +8,21 @@ * be found in the AUTHORS file in the root of the source tree. */ +#ifndef VP8_ENCODER_SEGMENTATION_H_ +#define VP8_ENCODER_SEGMENTATION_H_ #include "string.h" #include "vp8/common/blockd.h" #include "onyx_int.h" +#ifdef __cplusplus +extern "C" { +#endif + extern void vp8_update_gf_useage_maps(VP8_COMP *cpi, VP8_COMMON *cm, MACROBLOCK *x); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_ENCODER_SEGMENTATION_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/temporal_filter.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/temporal_filter.c index 7e3af71ecba..4dc0d959221 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/temporal_filter.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/temporal_filter.c @@ -16,7 +16,6 @@ #include "vp8/common/alloccommon.h" #include "mcomp.h" #include "firstpass.h" -#include "psnr.h" #include "vpx_scale/vpx_scale.h" #include "vp8/common/extend.h" #include "ratectrl.h" @@ -99,6 +98,7 @@ void vp8_temporal_filter_apply_c unsigned int i, j, k; int modifier; int byte = 0; + const int rounding = strength > 0 ? 1 << (strength - 1) : 0; for (i = 0,k = 0; i < block_size; i++) { @@ -115,7 +115,7 @@ void vp8_temporal_filter_apply_c */ modifier *= modifier; modifier *= 3; - modifier += 1 << (strength - 1); + modifier += rounding; modifier >>= strength; if (modifier > 16) diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.c index 11559a72083..2dc8205278b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.c @@ -213,6 +213,7 @@ static void tokenize1st_order_b /* Luma */ for (block = 0; block < 16; block++, b++) { + const int eob = *b->eob; tmp1 = vp8_block2above[block]; tmp2 = vp8_block2left[block]; qcoeff_ptr = b->qcoeff; @@ -223,7 +224,7 @@ static void tokenize1st_order_b c = type ? 0 : 1; - if(c >= *b->eob) + if(c >= eob) { /* c = band for this case */ t->Token = DCT_EOB_TOKEN; @@ -250,7 +251,8 @@ static void tokenize1st_order_b t++; c++; - for (; c < *b->eob; c++) + assert(eob <= 16); + for (; c < eob; c++) { rc = vp8_default_zig_zag1d[c]; band = vp8_coef_bands[c]; @@ -286,6 +288,7 @@ static void tokenize1st_order_b /* Chroma */ for (block = 16; block < 24; block++, b++) { + const int eob = *b->eob; tmp1 = vp8_block2above[block]; tmp2 = vp8_block2left[block]; qcoeff_ptr = b->qcoeff; @@ -294,7 +297,7 @@ static void tokenize1st_order_b VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l); - if(!(*b->eob)) + if(!eob) { /* c = band for this case */ t->Token = DCT_EOB_TOKEN; @@ -321,7 +324,8 @@ static void tokenize1st_order_b t++; c = 1; - for (; c < *b->eob; c++) + assert(eob <= 16); + for (; c < eob; c++) { rc = vp8_default_zig_zag1d[c]; band = vp8_coef_bands[c]; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.h index 1e6cea11465..b73a9ee1c89 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/tokenize.h @@ -9,12 +9,16 @@ */ -#ifndef tokenize_h -#define tokenize_h +#ifndef VP8_ENCODER_TOKENIZE_H_ +#define VP8_ENCODER_TOKENIZE_H_ #include "vp8/common/entropy.h" #include "block.h" +#ifdef __cplusplus +extern "C" { +#endif + void vp8_tokenize_initialize(); typedef struct @@ -47,4 +51,8 @@ extern const short *const vp8_dct_value_cost_ptr; */ extern const TOKENVALUE *const vp8_dct_value_tokens_ptr; -#endif /* tokenize_h */ +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP8_ENCODER_TOKENIZE_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/treewriter.h b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/treewriter.h index 48574f33cba..cfb2730ab3b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/treewriter.h +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/treewriter.h @@ -9,8 +9,8 @@ */ -#ifndef __INC_TREEWRITER_H -#define __INC_TREEWRITER_H +#ifndef VP8_ENCODER_TREEWRITER_H_ +#define VP8_ENCODER_TREEWRITER_H_ /* Trees map alphabets into huffman-like codes suitable for an arithmetic bit coder. Timothy S Murphy 11 October 2004 */ @@ -19,6 +19,10 @@ #include "boolhuff.h" /* for now */ +#ifdef __cplusplus +extern "C" { +#endif + typedef BOOL_CODER vp8_writer; #define vp8_write vp8_encode_bool @@ -123,4 +127,8 @@ void vp8_cost_tokens2( int *Costs, const vp8_prob *, vp8_tree, int ); +#ifdef __cplusplus +} // extern "C" #endif + +#endif // VP8_ENCODER_TREEWRITER_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/denoising_sse2.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/denoising_sse2.c index cceb8263f24..5112f891e91 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/denoising_sse2.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/denoising_sse2.c @@ -22,26 +22,28 @@ union sum_union { signed char e[16]; }; -int vp8_denoiser_filter_sse2(YV12_BUFFER_CONFIG *mc_running_avg, - YV12_BUFFER_CONFIG *running_avg, - MACROBLOCK *signal, unsigned int motion_magnitude, - int y_offset, int uv_offset) +int vp8_denoiser_filter_sse2(unsigned char *mc_running_avg_y, + int mc_avg_y_stride, + unsigned char *running_avg_y, int avg_y_stride, + unsigned char *sig, int sig_stride, + unsigned int motion_magnitude, + int increase_denoising) { - unsigned char *sig = signal->thismb; - int sig_stride = 16; - unsigned char *mc_running_avg_y = mc_running_avg->y_buffer + y_offset; - int mc_avg_y_stride = mc_running_avg->y_stride; - unsigned char *running_avg_y = running_avg->y_buffer + y_offset; - int avg_y_stride = running_avg->y_stride; + unsigned char *running_avg_y_start = running_avg_y; + unsigned char *sig_start = sig; + int sum_diff_thresh; int r; + int shift_inc = (increase_denoising && + motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 1 : 0; __m128i acc_diff = _mm_setzero_si128(); const __m128i k_0 = _mm_setzero_si128(); - const __m128i k_4 = _mm_set1_epi8(4); + const __m128i k_4 = _mm_set1_epi8(4 + shift_inc); const __m128i k_8 = _mm_set1_epi8(8); const __m128i k_16 = _mm_set1_epi8(16); /* Modify each level's adjustment according to motion_magnitude. */ const __m128i l3 = _mm_set1_epi8( - (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 : 6); + (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? + 7 + shift_inc : 6); /* Difference between level 3 and level 2 is 2. */ const __m128i l32 = _mm_set1_epi8(2); /* Difference between level 2 and level 1 is 1. */ @@ -108,13 +110,14 @@ int vp8_denoiser_filter_sse2(YV12_BUFFER_CONFIG *mc_running_avg, + s.e[6] + s.e[7] + s.e[8] + s.e[9] + s.e[10] + s.e[11] + s.e[12] + s.e[13] + s.e[14] + s.e[15]; - if (abs(sum_diff) > SUM_DIFF_THRESHOLD) + sum_diff_thresh = SUM_DIFF_THRESHOLD; + if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH; + if (abs(sum_diff) > sum_diff_thresh) { return COPY_BLOCK; } } - vp8_copy_mem16x16(running_avg->y_buffer + y_offset, avg_y_stride, - signal->thismb, sig_stride); + vp8_copy_mem16x16(running_avg_y_start, avg_y_stride, sig_start, sig_stride); return FILTER_BLOCK; } diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/quantize_ssse3.asm b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/quantize_ssse3.asm deleted file mode 100644 index 7b1dc119f08..00000000000 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/quantize_ssse3.asm +++ /dev/null @@ -1,138 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license and patent -; grant that can be found in the LICENSE file in the root of the source -; tree. All contributing project authors may be found in the AUTHORS -; file in the root of the source tree. -; - - -%include "vpx_ports/x86_abi_support.asm" -%include "vp8_asm_enc_offsets.asm" - - -; void vp8_fast_quantize_b_ssse3 | arg -; (BLOCK *b, | 0 -; BLOCKD *d) | 1 -; - -global sym(vp8_fast_quantize_b_ssse3) PRIVATE -sym(vp8_fast_quantize_b_ssse3): - push rbp - mov rbp, rsp - GET_GOT rbx - -%if ABI_IS_32BIT - push rdi - push rsi -%else - %if LIBVPX_YASM_WIN64 - push rdi - push rsi - %endif -%endif - ; end prolog - -%if ABI_IS_32BIT - mov rdi, arg(0) ; BLOCK *b - mov rsi, arg(1) ; BLOCKD *d -%else - %if LIBVPX_YASM_WIN64 - mov rdi, rcx ; BLOCK *b - mov rsi, rdx ; BLOCKD *d - %else - ;mov rdi, rdi ; BLOCK *b - ;mov rsi, rsi ; BLOCKD *d - %endif -%endif - - mov rax, [rdi + vp8_block_coeff] - mov rcx, [rdi + vp8_block_round] - mov rdx, [rdi + vp8_block_quant_fast] - - ; coeff - movdqa xmm0, [rax] - movdqa xmm4, [rax + 16] - - ; round - movdqa xmm2, [rcx] - movdqa xmm3, [rcx + 16] - - movdqa xmm1, xmm0 - movdqa xmm5, xmm4 - - ; sz = z >> 15 - psraw xmm0, 15 - psraw xmm4, 15 - - pabsw xmm1, xmm1 - pabsw xmm5, xmm5 - - paddw xmm1, xmm2 - paddw xmm5, xmm3 - - ; quant_fast - pmulhw xmm1, [rdx] - pmulhw xmm5, [rdx + 16] - - mov rax, [rsi + vp8_blockd_qcoeff] - mov rdi, [rsi + vp8_blockd_dequant] - mov rcx, [rsi + vp8_blockd_dqcoeff] - - movdqa xmm2, xmm1 ;store y for getting eob - movdqa xmm3, xmm5 - - pxor xmm1, xmm0 - pxor xmm5, xmm4 - psubw xmm1, xmm0 - psubw xmm5, xmm4 - - movdqa [rax], xmm1 - movdqa [rax + 16], xmm5 - - movdqa xmm0, [rdi] - movdqa xmm4, [rdi + 16] - - pmullw xmm0, xmm1 - pmullw xmm4, xmm5 - pxor xmm1, xmm1 - - pcmpgtw xmm2, xmm1 ;calculate eob - pcmpgtw xmm3, xmm1 - packsswb xmm2, xmm3 - pshufb xmm2, [GLOBAL(zz_shuf)] - - pmovmskb edx, xmm2 - - movdqa [rcx], xmm0 ;store dqcoeff - movdqa [rcx + 16], xmm4 ;store dqcoeff - mov rcx, [rsi + vp8_blockd_eob] - - bsr eax, edx ;count 0 - add eax, 1 - - cmp edx, 0 ;if all 0, eob=0 - cmove eax, edx - - mov BYTE PTR [rcx], al ;store eob - - ; begin epilog -%if ABI_IS_32BIT - pop rsi - pop rdi -%else - %if LIBVPX_YASM_WIN64 - pop rsi - pop rdi - %endif -%endif - - RESTORE_GOT - pop rbp - ret - -SECTION_RODATA -align 16 -zz_shuf: - db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15 diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/quantize_ssse3.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/quantize_ssse3.c new file mode 100644 index 00000000000..448217ff412 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/x86/quantize_ssse3.c @@ -0,0 +1,114 @@ +/* + * Copyright (c) 2012 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE 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. + */ + +#include <tmmintrin.h> /* SSSE3 */ + +#include "vp8/encoder/block.h" + +/* bitscan reverse (bsr) */ +#if defined(_MSC_VER) +#include <intrin.h> +#pragma intrinsic(_BitScanReverse) +static int bsr(int mask) { + int eob; + _BitScanReverse(&eob, mask); + eob++; + if (mask == 0) + eob = 0; + return eob; +} +#else +static int bsr(int mask) { + int eob; +#if defined(__GNUC__) && __GNUC__ + __asm__ __volatile__("bsr %1, %0" : "=r" (eob) : "r" (mask) : "flags"); +#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) + asm volatile("bsr %1, %0" : "=r" (eob) : "r" (mask) : "flags"); +#endif + eob++; + if (mask == 0) + eob = 0; + return eob; +} +#endif + +void vp8_fast_quantize_b_ssse3(BLOCK *b, BLOCKD *d) { + int eob, mask; + + __m128i z0 = _mm_load_si128((__m128i *)(b->coeff)); + __m128i z1 = _mm_load_si128((__m128i *)(b->coeff + 8)); + __m128i round0 = _mm_load_si128((__m128i *)(b->round)); + __m128i round1 = _mm_load_si128((__m128i *)(b->round + 8)); + __m128i quant_fast0 = _mm_load_si128((__m128i *)(b->quant_fast)); + __m128i quant_fast1 = _mm_load_si128((__m128i *)(b->quant_fast + 8)); + __m128i dequant0 = _mm_load_si128((__m128i *)(d->dequant)); + __m128i dequant1 = _mm_load_si128((__m128i *)(d->dequant + 8)); + + __m128i sz0, sz1, x, x0, x1, y0, y1, zeros, abs0, abs1; + + DECLARE_ALIGNED(16, const uint8_t, pshufb_zig_zag_mask[16]) = + { 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15 }; + __m128i zig_zag = _mm_load_si128((const __m128i *)pshufb_zig_zag_mask); + + /* sign of z: z >> 15 */ + sz0 = _mm_srai_epi16(z0, 15); + sz1 = _mm_srai_epi16(z1, 15); + + /* x = abs(z) */ + x0 = _mm_abs_epi16(z0); + x1 = _mm_abs_epi16(z1); + + /* x += round */ + x0 = _mm_add_epi16(x0, round0); + x1 = _mm_add_epi16(x1, round1); + + /* y = (x * quant) >> 16 */ + y0 = _mm_mulhi_epi16(x0, quant_fast0); + y1 = _mm_mulhi_epi16(x1, quant_fast1); + + /* ASM saves Y for EOB */ + /* I think we can ignore that because adding the sign doesn't change anything + * and multiplying 0 by dequant is OK as well */ + abs0 = y0; + abs1 = y1; + + /* Restore the sign bit. */ + y0 = _mm_xor_si128(y0, sz0); + y1 = _mm_xor_si128(y1, sz1); + x0 = _mm_sub_epi16(y0, sz0); + x1 = _mm_sub_epi16(y1, sz1); + + /* qcoeff = x */ + _mm_store_si128((__m128i *)(d->qcoeff), x0); + _mm_store_si128((__m128i *)(d->qcoeff + 8), x1); + + /* x * dequant */ + x0 = _mm_mullo_epi16(x0, dequant0); + x1 = _mm_mullo_epi16(x1, dequant1); + + /* dqcoeff = x * dequant */ + _mm_store_si128((__m128i *)(d->dqcoeff), x0); + _mm_store_si128((__m128i *)(d->dqcoeff + 8), x1); + + zeros = _mm_setzero_si128(); + + x0 = _mm_cmpgt_epi16(abs0, zeros); + x1 = _mm_cmpgt_epi16(abs1, zeros); + + x = _mm_packs_epi16(x0, x1); + + x = _mm_shuffle_epi8(x, zig_zag); + + mask = _mm_movemask_epi8(x); + + eob = bsr(mask); + + *d->eob = 0xFF & eob; +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk index f98eb318d43..8282547ea2d 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk @@ -47,7 +47,7 @@ VP8_COMMON_SRCS-yes += common/quant_common.h VP8_COMMON_SRCS-yes += common/reconinter.h VP8_COMMON_SRCS-yes += common/reconintra4x4.h VP8_COMMON_SRCS-yes += common/rtcd.c -VP8_COMMON_SRCS-yes += common/rtcd_defs.sh +VP8_COMMON_SRCS-yes += common/rtcd_defs.pl VP8_COMMON_SRCS-yes += common/setupintrarecon.h VP8_COMMON_SRCS-yes += common/swapyv12buffer.h VP8_COMMON_SRCS-yes += common/systemdependent.h @@ -129,7 +129,6 @@ VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/dequantize_dspr2.c # common (c) VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/filter_arm.c VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/loopfilter_arm.c -VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/reconintra_arm.c VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/dequantize_arm.c VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/variance_arm.c @@ -159,36 +158,29 @@ VP8_COMMON_SRCS-$(HAVE_MEDIA) += common/arm/armv6/vp8_variance_halfpixvar16x16_ VP8_COMMON_SRCS-$(HAVE_MEDIA) += common/arm/armv6/vp8_variance_halfpixvar16x16_hv_armv6$(ASM) # common (neon) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/bilinearpredict4x4_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/bilinearpredict8x4_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/bilinearpredict8x8_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/bilinearpredict16x16_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/copymem8x4_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/copymem8x8_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/copymem16x16_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dc_only_idct_add_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/iwalsh_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfilter_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfiltersimplehorizontaledge_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfiltersimpleverticaledge_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/mbloopfilter_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/shortidct4x4llm_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sad8_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sad16_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict4x4_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict8x4_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict8x8_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict16x16_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/buildintrapredictorsmby_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/save_reg_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dequant_idct_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/idct_dequant_full_2x_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/idct_dequant_0_2x_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dequantizeb_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/idct_blk_neon.c -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/variance_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp8_subpixelvariance8x8_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp8_subpixelvariance16x16_neon$(ASM) -VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp8_subpixelvariance16x16s_neon$(ASM) +#VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/reconintra_arm.c +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/loopfilter_neon$(ASM) +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/loopfiltersimpleverticaledge_neon$(ASM) +#VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/buildintrapredictorsmby_neon$(ASM) +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/idct_blk_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/idct_dequant_0_2x_neon$(ASM) +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/idct_dequant_full_2x_neon$(ASM) +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp8_subpixelvariance8x8_neon$(ASM) +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp8_subpixelvariance16x16_neon$(ASM) +VP8_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp8_subpixelvariance16x16s_neon$(ASM) -$(eval $(call rtcd_h_template,vp8_rtcd,vp8/common/rtcd_defs.sh)) +# common (neon intrinsics) +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/bilinearpredict_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/copymem_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dc_only_idct_add_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dequant_idct_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dequantizeb_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/iwalsh_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfiltersimplehorizontaledge_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/mbloopfilter_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sad_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/shortidct4x4llm_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict_neon.c +VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/variance_neon.c + +$(eval $(call rtcd_h_template,vp8_rtcd,vp8/common/rtcd_defs.pl)) diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c index 19e9d270183..501dd3eb4ac 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c @@ -414,7 +414,6 @@ static vpx_codec_err_t set_vp8e_config(VP8_CONFIG *oxcf, printf("Sharpness: %d\n", oxcf->Sharpness); printf("cpu_used: %d\n", oxcf->cpu_used); printf("Mode: %d\n", oxcf->Mode); - printf("delete_first_pass_file: %d\n", oxcf->delete_first_pass_file); printf("auto_key: %d\n", oxcf->auto_key); printf("key_freq: %d\n", oxcf->key_freq); printf("end_usage: %d\n", oxcf->end_usage); @@ -751,9 +750,6 @@ static vpx_codec_err_t vp8e_encode(vpx_codec_alg_priv_t *ctx, if (!ctx->cfg.rc_target_bitrate) return res; - if (!ctx->cfg.rc_target_bitrate) - return res; - if (img) res = validate_img(ctx, img); @@ -890,7 +886,7 @@ static vpx_codec_err_t vp8e_encode(vpx_codec_alg_priv_t *ctx, VP8_COMP *cpi = (VP8_COMP *)ctx->cpi; /* Add the frame packet to the list of returned packets. */ - round = (vpx_codec_pts_t)1000000 + round = (vpx_codec_pts_t)10000000 * ctx->cfg.g_timebase.num / 2 - 1; delta = (dst_end_time_stamp - dst_time_stamp); pkt.kind = VPX_CODEC_CX_FRAME_PKT; @@ -1239,6 +1235,8 @@ static vpx_codec_enc_cfg_map_t vp8e_usage_cfg_map[] = 0, /* rc_dropframe_thresh */ 0, /* rc_resize_allowed */ + 1, /* rc_scaled_width */ + 1, /* rc_scaled_height */ 60, /* rc_resize_down_thresold */ 30, /* rc_resize_up_thresold */ @@ -1266,10 +1264,10 @@ static vpx_codec_enc_cfg_map_t vp8e_usage_cfg_map[] = 128, /* kf_max_dist */ #if VPX_ENCODER_ABI_VERSION == (1 + VPX_CODEC_ABI_VERSION) - 1, /* g_delete_first_pass_file */ "vp8.fpf" /* first pass filename */ #endif VPX_SS_DEFAULT_LAYERS, /* ss_number_layers */ + {0}, /* ss_target_bitrate */ 1, /* ts_number_layers */ {0}, /* ts_target_bitrate */ {0}, /* ts_rate_decimator */ diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_dx_iface.c b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_dx_iface.c index 871b8d38579..10cbc6a583a 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_dx_iface.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_dx_iface.c @@ -16,9 +16,10 @@ #include "vpx/vp8dx.h" #include "vpx/internal/vpx_codec_internal.h" #include "vpx_version.h" +#include "common/alloccommon.h" +#include "common/common.h" #include "common/onyxd.h" #include "decoder/onyxd_int.h" -#include "common/alloccommon.h" #include "vpx_mem/vpx_mem.h" #if CONFIG_ERROR_CONCEALMENT #include "decoder/error_concealment.h" @@ -41,19 +42,11 @@ typedef enum static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_t); -static const mem_req_t vp8_mem_req_segs[] = -{ - {VP8_SEG_ALG_PRIV, 0, 8, VPX_CODEC_MEM_ZERO, vp8_priv_sz}, - {VP8_SEG_MAX, 0, 0, 0, NULL} -}; - struct vpx_codec_alg_priv { vpx_codec_priv_t base; - vpx_codec_mmap_t mmaps[NELEMENTS(vp8_mem_req_segs)-1]; vpx_codec_dec_cfg_t cfg; vp8_stream_info_t si; - int defer_alloc; int decoder_init; int postproc_cfg_set; vp8_postproc_cfg_t postproc_cfg; @@ -64,7 +57,7 @@ struct vpx_codec_alg_priv int dbg_color_b_modes_flag; int dbg_display_mv_flag; #endif - vp8_decrypt_cb *decrypt_cb; + vpx_decrypt_cb decrypt_cb; void *decrypt_state; vpx_image_t img; int img_setup; @@ -84,19 +77,14 @@ static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_ return sizeof(vpx_codec_alg_priv_t); } -static void vp8_init_ctx(vpx_codec_ctx_t *ctx, const vpx_codec_mmap_t *mmap) +static void vp8_init_ctx(vpx_codec_ctx_t *ctx) { - int i; - - ctx->priv = mmap->base; + ctx->priv = + (vpx_codec_priv_t *)vpx_memalign(8, sizeof(vpx_codec_alg_priv_t)); + vpx_memset(ctx->priv, 0, sizeof(vpx_codec_alg_priv_t)); ctx->priv->sz = sizeof(*ctx->priv); ctx->priv->iface = ctx->iface; - ctx->priv->alg_priv = mmap->base; - - for (i = 0; i < NELEMENTS(ctx->priv->alg_priv->mmaps); i++) - ctx->priv->alg_priv->mmaps[i].id = vp8_mem_req_segs[i].id; - - ctx->priv->alg_priv->mmaps[0] = *mmap; + ctx->priv->alg_priv = (vpx_codec_alg_priv_t *)ctx->priv; ctx->priv->alg_priv->si.sz = sizeof(ctx->priv->alg_priv->si); ctx->priv->alg_priv->decrypt_cb = NULL; ctx->priv->alg_priv->decrypt_state = NULL; @@ -110,11 +98,6 @@ static void vp8_init_ctx(vpx_codec_ctx_t *ctx, const vpx_codec_mmap_t *mmap) } } -static void vp8_finalize_mmaps(vpx_codec_alg_priv_t *ctx) -{ - /* nothing to clean up */ -} - static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx, vpx_codec_priv_enc_mr_cfg_t *data) { @@ -129,17 +112,7 @@ static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx, */ if (!ctx->priv) { - vpx_codec_mmap_t mmap; - - mmap.id = vp8_mem_req_segs[0].id; - mmap.sz = sizeof(vpx_codec_alg_priv_t); - mmap.align = vp8_mem_req_segs[0].align; - mmap.flags = vp8_mem_req_segs[0].flags; - - res = vpx_mmap_alloc(&mmap); - if (res != VPX_CODEC_OK) return res; - - vp8_init_ctx(ctx, &mmap); + vp8_init_ctx(ctx); /* initialize number of fragments to zero */ ctx->priv->alg_priv->fragments.count = 0; @@ -148,7 +121,6 @@ static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx, (ctx->priv->alg_priv->base.init_flags & VPX_CODEC_USE_INPUT_FRAGMENTS); - ctx->priv->alg_priv->defer_alloc = 1; /*post processing level initialized to do nothing */ } @@ -175,15 +147,9 @@ static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx, static vpx_codec_err_t vp8_destroy(vpx_codec_alg_priv_t *ctx) { - int i; - vp8_remove_decoder_instances(&ctx->yv12_frame_buffers); - for (i = NELEMENTS(ctx->mmaps) - 1; i >= 0; i--) - { - if (ctx->mmaps[i].dtor) - ctx->mmaps[i].dtor(&ctx->mmaps[i]); - } + vpx_free(ctx); return VPX_CODEC_OK; } @@ -191,7 +157,7 @@ static vpx_codec_err_t vp8_destroy(vpx_codec_alg_priv_t *ctx) static vpx_codec_err_t vp8_peek_si_internal(const uint8_t *data, unsigned int data_sz, vpx_codec_stream_info_t *si, - vp8_decrypt_cb *decrypt_cb, + vpx_decrypt_cb decrypt_cb, void *decrypt_state) { vpx_codec_err_t res = VPX_CODEC_OK; @@ -212,7 +178,7 @@ static vpx_codec_err_t vp8_peek_si_internal(const uint8_t *data, const uint8_t *clear = data; if (decrypt_cb) { - int n = data_sz > 10 ? 10 : data_sz; + int n = MIN(sizeof(clear_buffer), data_sz); decrypt_cb(decrypt_state, data, clear_buffer, n); clear = clear_buffer; } @@ -389,74 +355,40 @@ static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t *ctx, if ((ctx->si.h != h) || (ctx->si.w != w)) resolution_change = 1; - /* Perform deferred allocations, if required */ - if (!res && ctx->defer_alloc) - { - int i; - - for (i = 1; !res && i < NELEMENTS(ctx->mmaps); i++) - { - vpx_codec_dec_cfg_t cfg; - - cfg.w = ctx->si.w; - cfg.h = ctx->si.h; - ctx->mmaps[i].id = vp8_mem_req_segs[i].id; - ctx->mmaps[i].sz = vp8_mem_req_segs[i].sz; - ctx->mmaps[i].align = vp8_mem_req_segs[i].align; - ctx->mmaps[i].flags = vp8_mem_req_segs[i].flags; - - if (!ctx->mmaps[i].sz) - ctx->mmaps[i].sz = vp8_mem_req_segs[i].calc_sz(&cfg, - ctx->base.init_flags); - - res = vpx_mmap_alloc(&ctx->mmaps[i]); - } - - if (!res) - vp8_finalize_mmaps(ctx); - - ctx->defer_alloc = 0; - } - /* Initialize the decoder instance on the first frame*/ if (!res && !ctx->decoder_init) { - res = vpx_validate_mmaps(&ctx->si, ctx->mmaps, - vp8_mem_req_segs, NELEMENTS(vp8_mem_req_segs), - ctx->base.init_flags); - - if (!res) - { - VP8D_CONFIG oxcf; - - oxcf.Width = ctx->si.w; - oxcf.Height = ctx->si.h; - oxcf.Version = 9; - oxcf.postprocess = 0; - oxcf.max_threads = ctx->cfg.threads; - oxcf.error_concealment = - (ctx->base.init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT); - - /* If postprocessing was enabled by the application and a - * configuration has not been provided, default it. - */ - if (!ctx->postproc_cfg_set - && (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)) - { - ctx->postproc_cfg.post_proc_flag = - VP8_DEBLOCK | VP8_DEMACROBLOCK | VP8_MFQE; - ctx->postproc_cfg.deblocking_level = 4; - ctx->postproc_cfg.noise_level = 0; - } - - res = vp8_create_decoder_instances(&ctx->yv12_frame_buffers, &oxcf); - ctx->yv12_frame_buffers.pbi[0]->decrypt_cb = ctx->decrypt_cb; - ctx->yv12_frame_buffers.pbi[0]->decrypt_state = ctx->decrypt_state; - } - - ctx->decoder_init = 1; + VP8D_CONFIG oxcf; + + oxcf.Width = ctx->si.w; + oxcf.Height = ctx->si.h; + oxcf.Version = 9; + oxcf.postprocess = 0; + oxcf.max_threads = ctx->cfg.threads; + oxcf.error_concealment = + (ctx->base.init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT); + + /* If postprocessing was enabled by the application and a + * configuration has not been provided, default it. + */ + if (!ctx->postproc_cfg_set + && (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)) { + ctx->postproc_cfg.post_proc_flag = + VP8_DEBLOCK | VP8_DEMACROBLOCK | VP8_MFQE; + ctx->postproc_cfg.deblocking_level = 4; + ctx->postproc_cfg.noise_level = 0; + } + + res = vp8_create_decoder_instances(&ctx->yv12_frame_buffers, &oxcf); + ctx->decoder_init = 1; } + /* Set these even if already initialized. The caller may have changed the + * decrypt config between frames. + */ + ctx->yv12_frame_buffers.pbi[0]->decrypt_cb = ctx->decrypt_cb; + ctx->yv12_frame_buffers.pbi[0]->decrypt_state = ctx->decrypt_state; + if (!res) { VP8D_COMP *pbi = ctx->yv12_frame_buffers.pbi[0]; @@ -618,89 +550,6 @@ static vpx_image_t *vp8_get_frame(vpx_codec_alg_priv_t *ctx, return img; } - -static -vpx_codec_err_t vp8_xma_get_mmap(const vpx_codec_ctx_t *ctx, - vpx_codec_mmap_t *mmap, - vpx_codec_iter_t *iter) -{ - vpx_codec_err_t res; - const mem_req_t *seg_iter = *iter; - - /* Get address of next segment request */ - do - { - if (!seg_iter) - seg_iter = vp8_mem_req_segs; - else if (seg_iter->id != VP8_SEG_MAX) - seg_iter++; - - *iter = (vpx_codec_iter_t)seg_iter; - - if (seg_iter->id != VP8_SEG_MAX) - { - mmap->id = seg_iter->id; - mmap->sz = seg_iter->sz; - mmap->align = seg_iter->align; - mmap->flags = seg_iter->flags; - - if (!seg_iter->sz) - mmap->sz = seg_iter->calc_sz(ctx->config.dec, ctx->init_flags); - - res = VPX_CODEC_OK; - } - else - res = VPX_CODEC_LIST_END; - } - while (!mmap->sz && res != VPX_CODEC_LIST_END); - - return res; -} - -static vpx_codec_err_t vp8_xma_set_mmap(vpx_codec_ctx_t *ctx, - const vpx_codec_mmap_t *mmap) -{ - vpx_codec_err_t res = VPX_CODEC_MEM_ERROR; - int i, done; - - if (!ctx->priv) - { - if (mmap->id == VP8_SEG_ALG_PRIV) - { - if (!ctx->priv) - { - vp8_init_ctx(ctx, mmap); - res = VPX_CODEC_OK; - } - } - } - - done = 1; - - if (!res && ctx->priv->alg_priv) - { - for (i = 0; i < NELEMENTS(ctx->priv->alg_priv->mmaps); i++) - { - if (ctx->priv->alg_priv->mmaps[i].id == mmap->id) - if (!ctx->priv->alg_priv->mmaps[i].base) - { - ctx->priv->alg_priv->mmaps[i] = *mmap; - res = VPX_CODEC_OK; - } - - done &= (ctx->priv->alg_priv->mmaps[i].base != NULL); - } - } - - if (done && !res) - { - vp8_finalize_mmaps(ctx->priv->alg_priv); - res = ctx->iface->init(ctx, NULL); - } - - return res; -} - static vpx_codec_err_t image2yuvconfig(const vpx_image_t *img, YV12_BUFFER_CONFIG *yv12) { @@ -877,7 +726,7 @@ static vpx_codec_err_t vp8_set_decryptor(vpx_codec_alg_priv_t *ctx, int ctrl_id, va_list args) { - vp8_decrypt_init *init = va_arg(args, vp8_decrypt_init *); + vpx_decrypt_init *init = va_arg(args, vpx_decrypt_init *); if (init) { @@ -904,7 +753,7 @@ vpx_codec_ctrl_fn_map_t vp8_ctf_maps[] = {VP8D_GET_LAST_REF_UPDATES, vp8_get_last_ref_updates}, {VP8D_GET_FRAME_CORRUPTED, vp8_get_frame_corrupted}, {VP8D_GET_LAST_REF_USED, vp8_get_last_ref_frame}, - {VP8D_SET_DECRYPTOR, vp8_set_decryptor}, + {VPXD_SET_DECRYPTOR, vp8_set_decryptor}, { -1, NULL}, }; @@ -922,13 +771,14 @@ CODEC_INTERFACE(vpx_codec_vp8_dx) = vp8_init, /* vpx_codec_init_fn_t init; */ vp8_destroy, /* vpx_codec_destroy_fn_t destroy; */ vp8_ctf_maps, /* vpx_codec_ctrl_fn_map_t *ctrl_maps; */ - vp8_xma_get_mmap, /* vpx_codec_get_mmap_fn_t get_mmap; */ - vp8_xma_set_mmap, /* vpx_codec_set_mmap_fn_t set_mmap; */ + NOT_IMPLEMENTED, /* vpx_codec_get_mmap_fn_t get_mmap; */ + NOT_IMPLEMENTED, /* vpx_codec_set_mmap_fn_t set_mmap; */ { vp8_peek_si, /* vpx_codec_peek_si_fn_t peek_si; */ vp8_get_si, /* vpx_codec_get_si_fn_t get_si; */ vp8_decode, /* vpx_codec_decode_fn_t decode; */ vp8_get_frame, /* vpx_codec_frame_get_fn_t frame_get; */ + NOT_IMPLEMENTED, }, { /* encoder functions */ NOT_IMPLEMENTED, diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx.mk b/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx.mk index cd091f39ae3..607382b4c40 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx.mk +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx.mk @@ -50,7 +50,6 @@ VP8_CX_SRCS-yes += encoder/mcomp.h VP8_CX_SRCS-yes += encoder/modecosts.h VP8_CX_SRCS-yes += encoder/onyx_int.h VP8_CX_SRCS-yes += encoder/pickinter.h -VP8_CX_SRCS-yes += encoder/psnr.h VP8_CX_SRCS-yes += encoder/quantize.h VP8_CX_SRCS-yes += encoder/ratectrl.h VP8_CX_SRCS-yes += encoder/rdopt.h @@ -61,7 +60,6 @@ VP8_CX_SRCS-yes += encoder/modecosts.c VP8_CX_SRCS-yes += encoder/onyx_if.c VP8_CX_SRCS-yes += encoder/pickinter.c VP8_CX_SRCS-yes += encoder/picklpf.c -VP8_CX_SRCS-yes += encoder/psnr.c VP8_CX_SRCS-yes += encoder/quantize.c VP8_CX_SRCS-yes += encoder/ratectrl.c VP8_CX_SRCS-yes += encoder/rdopt.c @@ -90,6 +88,7 @@ VP8_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp8_enc_stubs_mmx.c VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_sse2.asm VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/fwalsh_sse2.asm VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.c +VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3.c ifeq ($(CONFIG_TEMPORAL_DENOISING),yes) VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/denoising_sse2.c @@ -98,7 +97,6 @@ endif VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/subtract_sse2.asm VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp8_enc_stubs_sse2.c -VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3.asm VP8_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/quantize_sse4.asm VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/quantize_mmx.asm VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/encodeopt.asm diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx_arm.mk b/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx_arm.mk index b030ee57e2a..57330486380 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx_arm.mk +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8cx_arm.mk @@ -35,10 +35,12 @@ VP8_CX_SRCS-$(HAVE_MEDIA) += encoder/arm/armv6/walsh_v6$(ASM) #File list for neon # encoder -VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/fastquantizeb_neon$(ASM) -VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/picklpf_arm.c -VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/shortfdct_neon$(ASM) -VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/subtract_neon$(ASM) -VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp8_mse16x16_neon$(ASM) -VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp8_memcpy_neon$(ASM) -VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp8_shortwalsh4x4_neon$(ASM) +VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/fastquantizeb_neon$(ASM) +VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/picklpf_arm.c +VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/shortfdct_neon$(ASM) +VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/subtract_neon$(ASM) +VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/vp8_mse16x16_neon$(ASM) +VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/vp8_memcpy_neon$(ASM) +VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/vp8_shortwalsh4x4_neon$(ASM) + +VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/denoising_neon.c diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8dx.mk b/chromium/third_party/libvpx/source/libvpx/vp8/vp8dx.mk index 4a8f46708ed..892ed70f52b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8dx.mk +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8dx.mk @@ -22,7 +22,7 @@ VP8_DX_SRCS-yes += vp8_dx_iface.c VP8_DX_SRCS-yes += decoder/dboolhuff.c VP8_DX_SRCS-yes += decoder/decodemv.c -VP8_DX_SRCS-yes += decoder/decodframe.c +VP8_DX_SRCS-yes += decoder/decodeframe.c VP8_DX_SRCS-yes += decoder/detokenize.c VP8_DX_SRCS-$(CONFIG_ERROR_CONCEALMENT) += decoder/ec_types.h VP8_DX_SRCS-$(CONFIG_ERROR_CONCEALMENT) += decoder/error_concealment.h |