diff options
author | Hangyu Kuang <hkuang@google.com> | 2016-07-06 14:21:45 -0700 |
---|---|---|
committer | Hangyu Kuang <hkuang@google.com> | 2016-07-08 09:51:10 -0700 |
commit | f047e7ca6983218eed7703c7afd51fed7bd3b5c9 (patch) | |
tree | 2667579566b6270c21ee4b495b4cd119af5ccf5b /files/source/row_common.cc | |
parent | bb74e3e19b98261031216de8cadcef34cccd9e4a (diff) | |
download | libyuv-f047e7ca6983218eed7703c7afd51fed7bd3b5c9.tar.gz |
Update libyuv to r1602 version to get best performance.android-cts_7.1_r1android-cts-7.1_r9android-cts-7.1_r8android-cts-7.1_r7android-cts-7.1_r6android-cts-7.1_r5android-cts-7.1_r4android-cts-7.1_r3android-cts-7.1_r29android-cts-7.1_r28android-cts-7.1_r27android-cts-7.1_r26android-cts-7.1_r25android-cts-7.1_r24android-cts-7.1_r23android-cts-7.1_r22android-cts-7.1_r21android-cts-7.1_r20android-cts-7.1_r2android-cts-7.1_r19android-cts-7.1_r18android-cts-7.1_r17android-cts-7.1_r16android-cts-7.1_r15android-cts-7.1_r14android-cts-7.1_r13android-cts-7.1_r12android-cts-7.1_r11android-cts-7.1_r10android-cts-7.1_r1android-7.1.2_r9android-7.1.2_r8android-7.1.2_r6android-7.1.2_r5android-7.1.2_r4android-7.1.2_r39android-7.1.2_r38android-7.1.2_r37android-7.1.2_r36android-7.1.2_r33android-7.1.2_r32android-7.1.2_r30android-7.1.2_r3android-7.1.2_r29android-7.1.2_r28android-7.1.2_r27android-7.1.2_r25android-7.1.2_r24android-7.1.2_r23android-7.1.2_r2android-7.1.2_r19android-7.1.2_r18android-7.1.2_r17android-7.1.2_r16android-7.1.2_r15android-7.1.2_r14android-7.1.2_r13android-7.1.2_r12android-7.1.2_r11android-7.1.2_r10android-7.1.2_r1android-7.1.1_r9android-7.1.1_r8android-7.1.1_r7android-7.1.1_r61android-7.1.1_r60android-7.1.1_r6android-7.1.1_r59android-7.1.1_r58android-7.1.1_r57android-7.1.1_r56android-7.1.1_r55android-7.1.1_r54android-7.1.1_r53android-7.1.1_r52android-7.1.1_r51android-7.1.1_r50android-7.1.1_r49android-7.1.1_r48android-7.1.1_r47android-7.1.1_r46android-7.1.1_r45android-7.1.1_r44android-7.1.1_r43android-7.1.1_r42android-7.1.1_r41android-7.1.1_r40android-7.1.1_r4android-7.1.1_r39android-7.1.1_r38android-7.1.1_r35android-7.1.1_r33android-7.1.1_r32android-7.1.1_r31android-7.1.1_r3android-7.1.1_r28android-7.1.1_r27android-7.1.1_r26android-7.1.1_r25android-7.1.1_r24android-7.1.1_r23android-7.1.1_r22android-7.1.1_r21android-7.1.1_r20android-7.1.1_r2android-7.1.1_r17android-7.1.1_r16android-7.1.1_r15android-7.1.1_r14android-7.1.1_r13android-7.1.1_r12android-7.1.1_r11android-7.1.1_r10android-7.1.1_r1android-7.1.0_r7android-7.1.0_r6android-7.1.0_r5android-7.1.0_r4android-7.1.0_r3android-7.1.0_r2android-7.1.0_r1nougat-mr2.3-releasenougat-mr2.2-releasenougat-mr2.1-releasenougat-mr2-security-releasenougat-mr2-releasenougat-mr2-pixel-releasenougat-mr2-devnougat-mr1.8-releasenougat-mr1.7-releasenougat-mr1.6-releasenougat-mr1.5-releasenougat-mr1.4-releasenougat-mr1.3-releasenougat-mr1.2-releasenougat-mr1.1-releasenougat-mr1-volantis-releasenougat-mr1-security-releasenougat-mr1-releasenougat-mr1-flounder-releasenougat-mr1-devnougat-mr1-cts-releasenougat-dr1-release
Bug: 29870647
Change-Id: I8ec9fab7f55765fa33ebe7ba1c7ad2147f418de2
Diffstat (limited to 'files/source/row_common.cc')
-rw-r--r-- | files/source/row_common.cc | 2553 |
1 files changed, 1967 insertions, 586 deletions
diff --git a/files/source/row_common.cc b/files/source/row_common.cc index c5f3ce05..32d2f686 100644 --- a/files/source/row_common.cc +++ b/files/source/row_common.cc @@ -4,13 +4,13 @@ * 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 + * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree. */ #include "libyuv/row.h" -#include <string.h> // For memcpy +#include <string.h> // For memcpy and memset. #include "libyuv/basic_types.h" @@ -19,56 +19,60 @@ namespace libyuv { extern "C" { #endif -void BGRAToARGBRow_C(const uint8* src_bgra, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { - // To support in-place conversion. - uint8 a = src_bgra[0]; - uint8 r = src_bgra[1]; - uint8 g = src_bgra[2]; - uint8 b = src_bgra[3]; - dst_argb[0] = b; - dst_argb[1] = g; - dst_argb[2] = r; - dst_argb[3] = a; - dst_argb += 4; - src_bgra += 4; - } +// llvm x86 is poor at ternary operator, so use branchless min/max. + +#define USE_BRANCHLESS 1 +#if USE_BRANCHLESS +static __inline int32 clamp0(int32 v) { + return ((-(v) >> 31) & (v)); } -void ABGRToARGBRow_C(const uint8* src_abgr, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { - // To support in-place conversion. - uint8 r = src_abgr[0]; - uint8 g = src_abgr[1]; - uint8 b = src_abgr[2]; - uint8 a = src_abgr[3]; - dst_argb[0] = b; - dst_argb[1] = g; - dst_argb[2] = r; - dst_argb[3] = a; - dst_argb += 4; - src_abgr += 4; - } +static __inline int32 clamp255(int32 v) { + return (((255 - (v)) >> 31) | (v)) & 255; } -void RGBAToARGBRow_C(const uint8* src_abgr, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { - // To support in-place conversion. - uint8 a = src_abgr[0]; - uint8 b = src_abgr[1]; - uint8 g = src_abgr[2]; - uint8 r = src_abgr[3]; - dst_argb[0] = b; - dst_argb[1] = g; - dst_argb[2] = r; - dst_argb[3] = a; - dst_argb += 4; - src_abgr += 4; - } +static __inline uint32 Clamp(int32 val) { + int v = clamp0(val); + return (uint32)(clamp255(v)); +} + +static __inline uint32 Abs(int32 v) { + int m = v >> 31; + return (v + m) ^ m; +} +#else // USE_BRANCHLESS +static __inline int32 clamp0(int32 v) { + return (v < 0) ? 0 : v; +} + +static __inline int32 clamp255(int32 v) { + return (v > 255) ? 255 : v; } +static __inline uint32 Clamp(int32 val) { + int v = clamp0(val); + return (uint32)(clamp255(v)); +} + +static __inline uint32 Abs(int32 v) { + return (v < 0) ? -v : v; +} +#endif // USE_BRANCHLESS + +#ifdef LIBYUV_LITTLE_ENDIAN +#define WRITEWORD(p, v) *(uint32*)(p) = v +#else +static inline void WRITEWORD(uint8* p, uint32 v) { + p[0] = (uint8)(v & 255); + p[1] = (uint8)((v >> 8) & 255); + p[2] = (uint8)((v >> 16) & 255); + p[3] = (uint8)((v >> 24) & 255); +} +#endif + void RGB24ToARGBRow_C(const uint8* src_rgb24, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { uint8 b = src_rgb24[0]; uint8 g = src_rgb24[1]; uint8 r = src_rgb24[2]; @@ -82,7 +86,8 @@ void RGB24ToARGBRow_C(const uint8* src_rgb24, uint8* dst_argb, int width) { } void RAWToARGBRow_C(const uint8* src_raw, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { uint8 r = src_raw[0]; uint8 g = src_raw[1]; uint8 b = src_raw[2]; @@ -95,67 +100,72 @@ void RAWToARGBRow_C(const uint8* src_raw, uint8* dst_argb, int width) { } } -void RGB565ToARGBRow_C(const uint8* src_rgb, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { - uint8 b = src_rgb[0] & 0x1f; - uint8 g = (src_rgb[0] >> 5) | ((src_rgb[1] & 0x07) << 3); - uint8 r = src_rgb[1] >> 3; +void RAWToRGB24Row_C(const uint8* src_raw, uint8* dst_rgb24, int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 r = src_raw[0]; + uint8 g = src_raw[1]; + uint8 b = src_raw[2]; + dst_rgb24[0] = b; + dst_rgb24[1] = g; + dst_rgb24[2] = r; + dst_rgb24 += 3; + src_raw += 3; + } +} + +void RGB565ToARGBRow_C(const uint8* src_rgb565, uint8* dst_argb, int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 b = src_rgb565[0] & 0x1f; + uint8 g = (src_rgb565[0] >> 5) | ((src_rgb565[1] & 0x07) << 3); + uint8 r = src_rgb565[1] >> 3; dst_argb[0] = (b << 3) | (b >> 2); dst_argb[1] = (g << 2) | (g >> 4); dst_argb[2] = (r << 3) | (r >> 2); dst_argb[3] = 255u; dst_argb += 4; - src_rgb += 2; + src_rgb565 += 2; } } -void ARGB1555ToARGBRow_C(const uint8* src_rgb, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { - uint8 b = src_rgb[0] & 0x1f; - uint8 g = (src_rgb[0] >> 5) | ((src_rgb[1] & 0x03) << 3); - uint8 r = (src_rgb[1] & 0x7c) >> 2; - uint8 a = src_rgb[1] >> 7; +void ARGB1555ToARGBRow_C(const uint8* src_argb1555, uint8* dst_argb, + int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 b = src_argb1555[0] & 0x1f; + uint8 g = (src_argb1555[0] >> 5) | ((src_argb1555[1] & 0x03) << 3); + uint8 r = (src_argb1555[1] & 0x7c) >> 2; + uint8 a = src_argb1555[1] >> 7; dst_argb[0] = (b << 3) | (b >> 2); dst_argb[1] = (g << 3) | (g >> 2); dst_argb[2] = (r << 3) | (r >> 2); dst_argb[3] = -a; dst_argb += 4; - src_rgb += 2; + src_argb1555 += 2; } } -void ARGB4444ToARGBRow_C(const uint8* src_rgb, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { - uint8 b = src_rgb[0] & 0x0f; - uint8 g = src_rgb[0] >> 4; - uint8 r = src_rgb[1] & 0x0f; - uint8 a = src_rgb[1] >> 4; +void ARGB4444ToARGBRow_C(const uint8* src_argb4444, uint8* dst_argb, + int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 b = src_argb4444[0] & 0x0f; + uint8 g = src_argb4444[0] >> 4; + uint8 r = src_argb4444[1] & 0x0f; + uint8 a = src_argb4444[1] >> 4; dst_argb[0] = (b << 4) | b; dst_argb[1] = (g << 4) | g; dst_argb[2] = (r << 4) | r; dst_argb[3] = (a << 4) | a; dst_argb += 4; - src_rgb += 2; - } -} - -void ARGBToRGBARow_C(const uint8* src_argb, uint8* dst_rgb, int width) { - for (int x = 0; x < width; ++x) { - uint8 b = src_argb[0]; - uint8 g = src_argb[1]; - uint8 r = src_argb[2]; - uint8 a = src_argb[3]; - dst_rgb[0] = a; - dst_rgb[1] = b; - dst_rgb[2] = g; - dst_rgb[3] = r; - dst_rgb += 4; - src_argb += 4; + src_argb4444 += 2; } } void ARGBToRGB24Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { uint8 b = src_argb[0]; uint8 g = src_argb[1]; uint8 r = src_argb[2]; @@ -168,7 +178,8 @@ void ARGBToRGB24Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { } void ARGBToRAWRow_C(const uint8* src_argb, uint8* dst_rgb, int width) { - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { uint8 b = src_argb[0]; uint8 g = src_argb[1]; uint8 r = src_argb[2]; @@ -180,17 +191,17 @@ void ARGBToRAWRow_C(const uint8* src_argb, uint8* dst_rgb, int width) { } } -// TODO(fbarchard): support big endian CPU void ARGBToRGB565Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { - for (int x = 0; x < width - 1; x += 2) { + int x; + for (x = 0; x < width - 1; x += 2) { uint8 b0 = src_argb[0] >> 3; uint8 g0 = src_argb[1] >> 2; uint8 r0 = src_argb[2] >> 3; uint8 b1 = src_argb[4] >> 3; uint8 g1 = src_argb[5] >> 2; uint8 r1 = src_argb[6] >> 3; - *reinterpret_cast<uint32*>(dst_rgb) = b0 | (g0 << 5) | (r0 << 11) | - (b1 << 16) | (g1 << 21) | (r1 << 27); + WRITEWORD(dst_rgb, b0 | (g0 << 5) | (r0 << 11) | + (b1 << 16) | (g1 << 21) | (r1 << 27)); dst_rgb += 4; src_argb += 8; } @@ -198,12 +209,47 @@ void ARGBToRGB565Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { uint8 b0 = src_argb[0] >> 3; uint8 g0 = src_argb[1] >> 2; uint8 r0 = src_argb[2] >> 3; - *reinterpret_cast<uint16*>(dst_rgb) = b0 | (g0 << 5) | (r0 << 11); + *(uint16*)(dst_rgb) = b0 | (g0 << 5) | (r0 << 11); + } +} + +// dither4 is a row of 4 values from 4x4 dither matrix. +// The 4x4 matrix contains values to increase RGB. When converting to +// fewer bits (565) this provides an ordered dither. +// The order in the 4x4 matrix in first byte is upper left. +// The 4 values are passed as an int, then referenced as an array, so +// endian will not affect order of the original matrix. But the dither4 +// will containing the first pixel in the lower byte for little endian +// or the upper byte for big endian. +void ARGBToRGB565DitherRow_C(const uint8* src_argb, uint8* dst_rgb, + const uint32 dither4, int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + int dither0 = ((const unsigned char*)(&dither4))[x & 3]; + int dither1 = ((const unsigned char*)(&dither4))[(x + 1) & 3]; + uint8 b0 = clamp255(src_argb[0] + dither0) >> 3; + uint8 g0 = clamp255(src_argb[1] + dither0) >> 2; + uint8 r0 = clamp255(src_argb[2] + dither0) >> 3; + uint8 b1 = clamp255(src_argb[4] + dither1) >> 3; + uint8 g1 = clamp255(src_argb[5] + dither1) >> 2; + uint8 r1 = clamp255(src_argb[6] + dither1) >> 3; + WRITEWORD(dst_rgb, b0 | (g0 << 5) | (r0 << 11) | + (b1 << 16) | (g1 << 21) | (r1 << 27)); + dst_rgb += 4; + src_argb += 8; + } + if (width & 1) { + int dither0 = ((const unsigned char*)(&dither4))[(width - 1) & 3]; + uint8 b0 = clamp255(src_argb[0] + dither0) >> 3; + uint8 g0 = clamp255(src_argb[1] + dither0) >> 2; + uint8 r0 = clamp255(src_argb[2] + dither0) >> 3; + *(uint16*)(dst_rgb) = b0 | (g0 << 5) | (r0 << 11); } } void ARGBToARGB1555Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { - for (int x = 0; x < width - 1; x += 2) { + int x; + for (x = 0; x < width - 1; x += 2) { uint8 b0 = src_argb[0] >> 3; uint8 g0 = src_argb[1] >> 3; uint8 r0 = src_argb[2] >> 3; @@ -212,7 +258,7 @@ void ARGBToARGB1555Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { uint8 g1 = src_argb[5] >> 3; uint8 r1 = src_argb[6] >> 3; uint8 a1 = src_argb[7] >> 7; - *reinterpret_cast<uint32*>(dst_rgb) = + *(uint32*)(dst_rgb) = b0 | (g0 << 5) | (r0 << 10) | (a0 << 15) | (b1 << 16) | (g1 << 21) | (r1 << 26) | (a1 << 31); dst_rgb += 4; @@ -223,13 +269,14 @@ void ARGBToARGB1555Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { uint8 g0 = src_argb[1] >> 3; uint8 r0 = src_argb[2] >> 3; uint8 a0 = src_argb[3] >> 7; - *reinterpret_cast<uint16*>(dst_rgb) = + *(uint16*)(dst_rgb) = b0 | (g0 << 5) | (r0 << 10) | (a0 << 15); } } void ARGBToARGB4444Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { - for (int x = 0; x < width - 1; x += 2) { + int x; + for (x = 0; x < width - 1; x += 2) { uint8 b0 = src_argb[0] >> 4; uint8 g0 = src_argb[1] >> 4; uint8 r0 = src_argb[2] >> 4; @@ -238,7 +285,7 @@ void ARGBToARGB4444Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { uint8 g1 = src_argb[5] >> 4; uint8 r1 = src_argb[6] >> 4; uint8 a1 = src_argb[7] >> 4; - *reinterpret_cast<uint32*>(dst_rgb) = + *(uint32*)(dst_rgb) = b0 | (g0 << 4) | (r0 << 8) | (a0 << 12) | (b1 << 16) | (g1 << 20) | (r1 << 24) | (a1 << 28); dst_rgb += 4; @@ -249,44 +296,46 @@ void ARGBToARGB4444Row_C(const uint8* src_argb, uint8* dst_rgb, int width) { uint8 g0 = src_argb[1] >> 4; uint8 r0 = src_argb[2] >> 4; uint8 a0 = src_argb[3] >> 4; - *reinterpret_cast<uint16*>(dst_rgb) = + *(uint16*)(dst_rgb) = b0 | (g0 << 4) | (r0 << 8) | (a0 << 12); } } static __inline int RGBToY(uint8 r, uint8 g, uint8 b) { - return (( 66 * r + 129 * g + 25 * b + 128) >> 8) + 16; + return (66 * r + 129 * g + 25 * b + 0x1080) >> 8; } static __inline int RGBToU(uint8 r, uint8 g, uint8 b) { - return ((-38 * r - 74 * g + 112 * b + 128) >> 8) + 128; + return (112 * b - 74 * g - 38 * r + 0x8080) >> 8; } static __inline int RGBToV(uint8 r, uint8 g, uint8 b) { - return ((112 * r - 94 * g - 18 * b + 128) >> 8) + 128; + return (112 * r - 94 * g - 18 * b + 0x8080) >> 8; } -#define MAKEROWY(NAME, R, G, B) \ +#define MAKEROWY(NAME, R, G, B, BPP) \ void NAME ## ToYRow_C(const uint8* src_argb0, uint8* dst_y, int width) { \ - for (int x = 0; x < width; ++x) { \ + int x; \ + for (x = 0; x < width; ++x) { \ dst_y[0] = RGBToY(src_argb0[R], src_argb0[G], src_argb0[B]); \ - src_argb0 += 4; \ + src_argb0 += BPP; \ dst_y += 1; \ } \ } \ void NAME ## ToUVRow_C(const uint8* src_rgb0, int src_stride_rgb, \ uint8* dst_u, uint8* dst_v, int width) { \ const uint8* src_rgb1 = src_rgb0 + src_stride_rgb; \ - for (int x = 0; x < width - 1; x += 2) { \ - uint8 ab = (src_rgb0[B] + src_rgb0[B + 4] + \ - src_rgb1[B] + src_rgb1[B + 4]) >> 2; \ - uint8 ag = (src_rgb0[G] + src_rgb0[G + 4] + \ - src_rgb1[G] + src_rgb1[G + 4]) >> 2; \ - uint8 ar = (src_rgb0[R] + src_rgb0[R + 4] + \ - src_rgb1[R] + src_rgb1[R + 4]) >> 2; \ + int x; \ + for (x = 0; x < width - 1; x += 2) { \ + uint8 ab = (src_rgb0[B] + src_rgb0[B + BPP] + \ + src_rgb1[B] + src_rgb1[B + BPP]) >> 2; \ + uint8 ag = (src_rgb0[G] + src_rgb0[G + BPP] + \ + src_rgb1[G] + src_rgb1[G + BPP]) >> 2; \ + uint8 ar = (src_rgb0[R] + src_rgb0[R + BPP] + \ + src_rgb1[R] + src_rgb1[R + BPP]) >> 2; \ dst_u[0] = RGBToU(ar, ag, ab); \ dst_v[0] = RGBToV(ar, ag, ab); \ - src_rgb0 += 8; \ - src_rgb1 += 8; \ + src_rgb0 += BPP * 2; \ + src_rgb1 += BPP * 2; \ dst_u += 1; \ dst_v += 1; \ } \ @@ -299,21 +348,333 @@ void NAME ## ToUVRow_C(const uint8* src_rgb0, int src_stride_rgb, \ } \ } -MAKEROWY(ARGB, 2, 1, 0) -MAKEROWY(BGRA, 1, 2, 3) -MAKEROWY(ABGR, 0, 1, 2) -MAKEROWY(RGBA, 3, 2, 1) +MAKEROWY(ARGB, 2, 1, 0, 4) +MAKEROWY(BGRA, 1, 2, 3, 4) +MAKEROWY(ABGR, 0, 1, 2, 4) +MAKEROWY(RGBA, 3, 2, 1, 4) +MAKEROWY(RGB24, 2, 1, 0, 3) +MAKEROWY(RAW, 0, 1, 2, 3) +#undef MAKEROWY + +// JPeg uses a variation on BT.601-1 full range +// y = 0.29900 * r + 0.58700 * g + 0.11400 * b +// u = -0.16874 * r - 0.33126 * g + 0.50000 * b + center +// v = 0.50000 * r - 0.41869 * g - 0.08131 * b + center +// BT.601 Mpeg range uses: +// b 0.1016 * 255 = 25.908 = 25 +// g 0.5078 * 255 = 129.489 = 129 +// r 0.2578 * 255 = 65.739 = 66 +// JPeg 8 bit Y (not used): +// b 0.11400 * 256 = 29.184 = 29 +// g 0.58700 * 256 = 150.272 = 150 +// r 0.29900 * 256 = 76.544 = 77 +// JPeg 7 bit Y: +// b 0.11400 * 128 = 14.592 = 15 +// g 0.58700 * 128 = 75.136 = 75 +// r 0.29900 * 128 = 38.272 = 38 +// JPeg 8 bit U: +// b 0.50000 * 255 = 127.5 = 127 +// g -0.33126 * 255 = -84.4713 = -84 +// r -0.16874 * 255 = -43.0287 = -43 +// JPeg 8 bit V: +// b -0.08131 * 255 = -20.73405 = -20 +// g -0.41869 * 255 = -106.76595 = -107 +// r 0.50000 * 255 = 127.5 = 127 + +static __inline int RGBToYJ(uint8 r, uint8 g, uint8 b) { + return (38 * r + 75 * g + 15 * b + 64) >> 7; +} + +static __inline int RGBToUJ(uint8 r, uint8 g, uint8 b) { + return (127 * b - 84 * g - 43 * r + 0x8080) >> 8; +} +static __inline int RGBToVJ(uint8 r, uint8 g, uint8 b) { + return (127 * r - 107 * g - 20 * b + 0x8080) >> 8; +} + +#define AVGB(a, b) (((a) + (b) + 1) >> 1) + +#define MAKEROWYJ(NAME, R, G, B, BPP) \ +void NAME ## ToYJRow_C(const uint8* src_argb0, uint8* dst_y, int width) { \ + int x; \ + for (x = 0; x < width; ++x) { \ + dst_y[0] = RGBToYJ(src_argb0[R], src_argb0[G], src_argb0[B]); \ + src_argb0 += BPP; \ + dst_y += 1; \ + } \ +} \ +void NAME ## ToUVJRow_C(const uint8* src_rgb0, int src_stride_rgb, \ + uint8* dst_u, uint8* dst_v, int width) { \ + const uint8* src_rgb1 = src_rgb0 + src_stride_rgb; \ + int x; \ + for (x = 0; x < width - 1; x += 2) { \ + uint8 ab = AVGB(AVGB(src_rgb0[B], src_rgb1[B]), \ + AVGB(src_rgb0[B + BPP], src_rgb1[B + BPP])); \ + uint8 ag = AVGB(AVGB(src_rgb0[G], src_rgb1[G]), \ + AVGB(src_rgb0[G + BPP], src_rgb1[G + BPP])); \ + uint8 ar = AVGB(AVGB(src_rgb0[R], src_rgb1[R]), \ + AVGB(src_rgb0[R + BPP], src_rgb1[R + BPP])); \ + dst_u[0] = RGBToUJ(ar, ag, ab); \ + dst_v[0] = RGBToVJ(ar, ag, ab); \ + src_rgb0 += BPP * 2; \ + src_rgb1 += BPP * 2; \ + dst_u += 1; \ + dst_v += 1; \ + } \ + if (width & 1) { \ + uint8 ab = AVGB(src_rgb0[B], src_rgb1[B]); \ + uint8 ag = AVGB(src_rgb0[G], src_rgb1[G]); \ + uint8 ar = AVGB(src_rgb0[R], src_rgb1[R]); \ + dst_u[0] = RGBToUJ(ar, ag, ab); \ + dst_v[0] = RGBToVJ(ar, ag, ab); \ + } \ +} + +MAKEROWYJ(ARGB, 2, 1, 0, 4) +#undef MAKEROWYJ + +void RGB565ToYRow_C(const uint8* src_rgb565, uint8* dst_y, int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 b = src_rgb565[0] & 0x1f; + uint8 g = (src_rgb565[0] >> 5) | ((src_rgb565[1] & 0x07) << 3); + uint8 r = src_rgb565[1] >> 3; + b = (b << 3) | (b >> 2); + g = (g << 2) | (g >> 4); + r = (r << 3) | (r >> 2); + dst_y[0] = RGBToY(r, g, b); + src_rgb565 += 2; + dst_y += 1; + } +} + +void ARGB1555ToYRow_C(const uint8* src_argb1555, uint8* dst_y, int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 b = src_argb1555[0] & 0x1f; + uint8 g = (src_argb1555[0] >> 5) | ((src_argb1555[1] & 0x03) << 3); + uint8 r = (src_argb1555[1] & 0x7c) >> 2; + b = (b << 3) | (b >> 2); + g = (g << 3) | (g >> 2); + r = (r << 3) | (r >> 2); + dst_y[0] = RGBToY(r, g, b); + src_argb1555 += 2; + dst_y += 1; + } +} + +void ARGB4444ToYRow_C(const uint8* src_argb4444, uint8* dst_y, int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 b = src_argb4444[0] & 0x0f; + uint8 g = src_argb4444[0] >> 4; + uint8 r = src_argb4444[1] & 0x0f; + b = (b << 4) | b; + g = (g << 4) | g; + r = (r << 4) | r; + dst_y[0] = RGBToY(r, g, b); + src_argb4444 += 2; + dst_y += 1; + } +} + +void RGB565ToUVRow_C(const uint8* src_rgb565, int src_stride_rgb565, + uint8* dst_u, uint8* dst_v, int width) { + const uint8* next_rgb565 = src_rgb565 + src_stride_rgb565; + int x; + for (x = 0; x < width - 1; x += 2) { + uint8 b0 = src_rgb565[0] & 0x1f; + uint8 g0 = (src_rgb565[0] >> 5) | ((src_rgb565[1] & 0x07) << 3); + uint8 r0 = src_rgb565[1] >> 3; + uint8 b1 = src_rgb565[2] & 0x1f; + uint8 g1 = (src_rgb565[2] >> 5) | ((src_rgb565[3] & 0x07) << 3); + uint8 r1 = src_rgb565[3] >> 3; + uint8 b2 = next_rgb565[0] & 0x1f; + uint8 g2 = (next_rgb565[0] >> 5) | ((next_rgb565[1] & 0x07) << 3); + uint8 r2 = next_rgb565[1] >> 3; + uint8 b3 = next_rgb565[2] & 0x1f; + uint8 g3 = (next_rgb565[2] >> 5) | ((next_rgb565[3] & 0x07) << 3); + uint8 r3 = next_rgb565[3] >> 3; + uint8 b = (b0 + b1 + b2 + b3); // 565 * 4 = 787. + uint8 g = (g0 + g1 + g2 + g3); + uint8 r = (r0 + r1 + r2 + r3); + b = (b << 1) | (b >> 6); // 787 -> 888. + r = (r << 1) | (r >> 6); + dst_u[0] = RGBToU(r, g, b); + dst_v[0] = RGBToV(r, g, b); + src_rgb565 += 4; + next_rgb565 += 4; + dst_u += 1; + dst_v += 1; + } + if (width & 1) { + uint8 b0 = src_rgb565[0] & 0x1f; + uint8 g0 = (src_rgb565[0] >> 5) | ((src_rgb565[1] & 0x07) << 3); + uint8 r0 = src_rgb565[1] >> 3; + uint8 b2 = next_rgb565[0] & 0x1f; + uint8 g2 = (next_rgb565[0] >> 5) | ((next_rgb565[1] & 0x07) << 3); + uint8 r2 = next_rgb565[1] >> 3; + uint8 b = (b0 + b2); // 565 * 2 = 676. + uint8 g = (g0 + g2); + uint8 r = (r0 + r2); + b = (b << 2) | (b >> 4); // 676 -> 888 + g = (g << 1) | (g >> 6); + r = (r << 2) | (r >> 4); + dst_u[0] = RGBToU(r, g, b); + dst_v[0] = RGBToV(r, g, b); + } +} + +void ARGB1555ToUVRow_C(const uint8* src_argb1555, int src_stride_argb1555, + uint8* dst_u, uint8* dst_v, int width) { + const uint8* next_argb1555 = src_argb1555 + src_stride_argb1555; + int x; + for (x = 0; x < width - 1; x += 2) { + uint8 b0 = src_argb1555[0] & 0x1f; + uint8 g0 = (src_argb1555[0] >> 5) | ((src_argb1555[1] & 0x03) << 3); + uint8 r0 = (src_argb1555[1] & 0x7c) >> 2; + uint8 b1 = src_argb1555[2] & 0x1f; + uint8 g1 = (src_argb1555[2] >> 5) | ((src_argb1555[3] & 0x03) << 3); + uint8 r1 = (src_argb1555[3] & 0x7c) >> 2; + uint8 b2 = next_argb1555[0] & 0x1f; + uint8 g2 = (next_argb1555[0] >> 5) | ((next_argb1555[1] & 0x03) << 3); + uint8 r2 = (next_argb1555[1] & 0x7c) >> 2; + uint8 b3 = next_argb1555[2] & 0x1f; + uint8 g3 = (next_argb1555[2] >> 5) | ((next_argb1555[3] & 0x03) << 3); + uint8 r3 = (next_argb1555[3] & 0x7c) >> 2; + uint8 b = (b0 + b1 + b2 + b3); // 555 * 4 = 777. + uint8 g = (g0 + g1 + g2 + g3); + uint8 r = (r0 + r1 + r2 + r3); + b = (b << 1) | (b >> 6); // 777 -> 888. + g = (g << 1) | (g >> 6); + r = (r << 1) | (r >> 6); + dst_u[0] = RGBToU(r, g, b); + dst_v[0] = RGBToV(r, g, b); + src_argb1555 += 4; + next_argb1555 += 4; + dst_u += 1; + dst_v += 1; + } + if (width & 1) { + uint8 b0 = src_argb1555[0] & 0x1f; + uint8 g0 = (src_argb1555[0] >> 5) | ((src_argb1555[1] & 0x03) << 3); + uint8 r0 = (src_argb1555[1] & 0x7c) >> 2; + uint8 b2 = next_argb1555[0] & 0x1f; + uint8 g2 = (next_argb1555[0] >> 5) | ((next_argb1555[1] & 0x03) << 3); + uint8 r2 = next_argb1555[1] >> 3; + uint8 b = (b0 + b2); // 555 * 2 = 666. + uint8 g = (g0 + g2); + uint8 r = (r0 + r2); + b = (b << 2) | (b >> 4); // 666 -> 888. + g = (g << 2) | (g >> 4); + r = (r << 2) | (r >> 4); + dst_u[0] = RGBToU(r, g, b); + dst_v[0] = RGBToV(r, g, b); + } +} + +void ARGB4444ToUVRow_C(const uint8* src_argb4444, int src_stride_argb4444, + uint8* dst_u, uint8* dst_v, int width) { + const uint8* next_argb4444 = src_argb4444 + src_stride_argb4444; + int x; + for (x = 0; x < width - 1; x += 2) { + uint8 b0 = src_argb4444[0] & 0x0f; + uint8 g0 = src_argb4444[0] >> 4; + uint8 r0 = src_argb4444[1] & 0x0f; + uint8 b1 = src_argb4444[2] & 0x0f; + uint8 g1 = src_argb4444[2] >> 4; + uint8 r1 = src_argb4444[3] & 0x0f; + uint8 b2 = next_argb4444[0] & 0x0f; + uint8 g2 = next_argb4444[0] >> 4; + uint8 r2 = next_argb4444[1] & 0x0f; + uint8 b3 = next_argb4444[2] & 0x0f; + uint8 g3 = next_argb4444[2] >> 4; + uint8 r3 = next_argb4444[3] & 0x0f; + uint8 b = (b0 + b1 + b2 + b3); // 444 * 4 = 666. + uint8 g = (g0 + g1 + g2 + g3); + uint8 r = (r0 + r1 + r2 + r3); + b = (b << 2) | (b >> 4); // 666 -> 888. + g = (g << 2) | (g >> 4); + r = (r << 2) | (r >> 4); + dst_u[0] = RGBToU(r, g, b); + dst_v[0] = RGBToV(r, g, b); + src_argb4444 += 4; + next_argb4444 += 4; + dst_u += 1; + dst_v += 1; + } + if (width & 1) { + uint8 b0 = src_argb4444[0] & 0x0f; + uint8 g0 = src_argb4444[0] >> 4; + uint8 r0 = src_argb4444[1] & 0x0f; + uint8 b2 = next_argb4444[0] & 0x0f; + uint8 g2 = next_argb4444[0] >> 4; + uint8 r2 = next_argb4444[1] & 0x0f; + uint8 b = (b0 + b2); // 444 * 2 = 555. + uint8 g = (g0 + g2); + uint8 r = (r0 + r2); + b = (b << 3) | (b >> 2); // 555 -> 888. + g = (g << 3) | (g >> 2); + r = (r << 3) | (r >> 2); + dst_u[0] = RGBToU(r, g, b); + dst_v[0] = RGBToV(r, g, b); + } +} + +void ARGBToUV444Row_C(const uint8* src_argb, + uint8* dst_u, uint8* dst_v, int width) { + int x; + for (x = 0; x < width; ++x) { + uint8 ab = src_argb[0]; + uint8 ag = src_argb[1]; + uint8 ar = src_argb[2]; + dst_u[0] = RGBToU(ar, ag, ab); + dst_v[0] = RGBToV(ar, ag, ab); + src_argb += 4; + dst_u += 1; + dst_v += 1; + } +} -// http://en.wikipedia.org/wiki/Grayscale. -// 0.11 * B + 0.59 * G + 0.30 * R -// Coefficients rounded to multiple of 2 for consistency with SSSE3 version. -static __inline int RGBToGray(uint8 r, uint8 g, uint8 b) { - return (( 76 * r + 152 * g + 28 * b) >> 8); +void ARGBToUV411Row_C(const uint8* src_argb, + uint8* dst_u, uint8* dst_v, int width) { + int x; + for (x = 0; x < width - 3; x += 4) { + uint8 ab = (src_argb[0] + src_argb[4] + src_argb[8] + src_argb[12]) >> 2; + uint8 ag = (src_argb[1] + src_argb[5] + src_argb[9] + src_argb[13]) >> 2; + uint8 ar = (src_argb[2] + src_argb[6] + src_argb[10] + src_argb[14]) >> 2; + dst_u[0] = RGBToU(ar, ag, ab); + dst_v[0] = RGBToV(ar, ag, ab); + src_argb += 16; + dst_u += 1; + dst_v += 1; + } + // Odd width handling mimics 'any' function which replicates last pixel. + if ((width & 3) == 3) { + uint8 ab = (src_argb[0] + src_argb[4] + src_argb[8] + src_argb[8]) >> 2; + uint8 ag = (src_argb[1] + src_argb[5] + src_argb[9] + src_argb[9]) >> 2; + uint8 ar = (src_argb[2] + src_argb[6] + src_argb[10] + src_argb[10]) >> 2; + dst_u[0] = RGBToU(ar, ag, ab); + dst_v[0] = RGBToV(ar, ag, ab); + } else if ((width & 3) == 2) { + uint8 ab = (src_argb[0] + src_argb[4]) >> 1; + uint8 ag = (src_argb[1] + src_argb[5]) >> 1; + uint8 ar = (src_argb[2] + src_argb[6]) >> 1; + dst_u[0] = RGBToU(ar, ag, ab); + dst_v[0] = RGBToV(ar, ag, ab); + } else if ((width & 3) == 1) { + uint8 ab = src_argb[0]; + uint8 ag = src_argb[1]; + uint8 ar = src_argb[2]; + dst_u[0] = RGBToU(ar, ag, ab); + dst_v[0] = RGBToV(ar, ag, ab); + } } void ARGBGrayRow_C(const uint8* src_argb, uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { - uint8 y = RGBToGray(src_argb[2], src_argb[1], src_argb[0]); + int x; + for (x = 0; x < width; ++x) { + uint8 y = RGBToYJ(src_argb[2], src_argb[1], src_argb[0]); dst_argb[2] = dst_argb[1] = dst_argb[0] = y; dst_argb[3] = src_argb[3]; dst_argb += 4; @@ -323,7 +684,8 @@ void ARGBGrayRow_C(const uint8* src_argb, uint8* dst_argb, int width) { // Convert a row of image to Sepia tone. void ARGBSepiaRow_C(uint8* dst_argb, int width) { - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { int b = dst_argb[0]; int g = dst_argb[1]; int r = dst_argb[2]; @@ -331,60 +693,44 @@ void ARGBSepiaRow_C(uint8* dst_argb, int width) { int sg = (b * 22 + g * 88 + r * 45) >> 7; int sr = (b * 24 + g * 98 + r * 50) >> 7; // b does not over flow. a is preserved from original. - if (sg > 255) { - sg = 255; - } - if (sr > 255) { - sr = 255; - } dst_argb[0] = sb; - dst_argb[1] = sg; - dst_argb[2] = sr; + dst_argb[1] = clamp255(sg); + dst_argb[2] = clamp255(sr); dst_argb += 4; } } // Apply color matrix to a row of image. Matrix is signed. -void ARGBColorMatrixRow_C(uint8* dst_argb, const int8* matrix_argb, int width) { - for (int x = 0; x < width; ++x) { - int b = dst_argb[0]; - int g = dst_argb[1]; - int r = dst_argb[2]; - int a = dst_argb[3]; +// TODO(fbarchard): Consider adding rounding (+32). +void ARGBColorMatrixRow_C(const uint8* src_argb, uint8* dst_argb, + const int8* matrix_argb, int width) { + int x; + for (x = 0; x < width; ++x) { + int b = src_argb[0]; + int g = src_argb[1]; + int r = src_argb[2]; + int a = src_argb[3]; int sb = (b * matrix_argb[0] + g * matrix_argb[1] + - r * matrix_argb[2] + a * matrix_argb[3]) >> 7; + r * matrix_argb[2] + a * matrix_argb[3]) >> 6; int sg = (b * matrix_argb[4] + g * matrix_argb[5] + - r * matrix_argb[6] + a * matrix_argb[7]) >> 7; + r * matrix_argb[6] + a * matrix_argb[7]) >> 6; int sr = (b * matrix_argb[8] + g * matrix_argb[9] + - r * matrix_argb[10] + a * matrix_argb[11]) >> 7; - if (sb < 0) { - sb = 0; - } - if (sb > 255) { - sb = 255; - } - if (sg < 0) { - sg = 0; - } - if (sg > 255) { - sg = 255; - } - if (sr < 0) { - sr = 0; - } - if (sr > 255) { - sr = 255; - } - dst_argb[0] = sb; - dst_argb[1] = sg; - dst_argb[2] = sr; + r * matrix_argb[10] + a * matrix_argb[11]) >> 6; + int sa = (b * matrix_argb[12] + g * matrix_argb[13] + + r * matrix_argb[14] + a * matrix_argb[15]) >> 6; + dst_argb[0] = Clamp(sb); + dst_argb[1] = Clamp(sg); + dst_argb[2] = Clamp(sr); + dst_argb[3] = Clamp(sa); + src_argb += 4; dst_argb += 4; } } // Apply color table to a row of image. void ARGBColorTableRow_C(uint8* dst_argb, const uint8* table_argb, int width) { - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { int b = dst_argb[0]; int g = dst_argb[1]; int r = dst_argb[2]; @@ -397,9 +743,24 @@ void ARGBColorTableRow_C(uint8* dst_argb, const uint8* table_argb, int width) { } } +// Apply color table to a row of image. +void RGBColorTableRow_C(uint8* dst_argb, const uint8* table_argb, int width) { + int x; + for (x = 0; x < width; ++x) { + int b = dst_argb[0]; + int g = dst_argb[1]; + int r = dst_argb[2]; + dst_argb[0] = table_argb[b * 4 + 0]; + dst_argb[1] = table_argb[g * 4 + 1]; + dst_argb[2] = table_argb[r * 4 + 2]; + dst_argb += 4; + } +} + void ARGBQuantizeRow_C(uint8* dst_argb, int scale, int interval_size, int interval_offset, int width) { - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { int b = dst_argb[0]; int g = dst_argb[1]; int r = dst_argb[2]; @@ -410,9 +771,192 @@ void ARGBQuantizeRow_C(uint8* dst_argb, int scale, int interval_size, } } -void I400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int width) { +#define REPEAT8(v) (v) | ((v) << 8) +#define SHADE(f, v) v * f >> 24 + +void ARGBShadeRow_C(const uint8* src_argb, uint8* dst_argb, int width, + uint32 value) { + const uint32 b_scale = REPEAT8(value & 0xff); + const uint32 g_scale = REPEAT8((value >> 8) & 0xff); + const uint32 r_scale = REPEAT8((value >> 16) & 0xff); + const uint32 a_scale = REPEAT8(value >> 24); + + int i; + for (i = 0; i < width; ++i) { + const uint32 b = REPEAT8(src_argb[0]); + const uint32 g = REPEAT8(src_argb[1]); + const uint32 r = REPEAT8(src_argb[2]); + const uint32 a = REPEAT8(src_argb[3]); + dst_argb[0] = SHADE(b, b_scale); + dst_argb[1] = SHADE(g, g_scale); + dst_argb[2] = SHADE(r, r_scale); + dst_argb[3] = SHADE(a, a_scale); + src_argb += 4; + dst_argb += 4; + } +} +#undef REPEAT8 +#undef SHADE + +#define REPEAT8(v) (v) | ((v) << 8) +#define SHADE(f, v) v * f >> 16 + +void ARGBMultiplyRow_C(const uint8* src_argb0, const uint8* src_argb1, + uint8* dst_argb, int width) { + int i; + for (i = 0; i < width; ++i) { + const uint32 b = REPEAT8(src_argb0[0]); + const uint32 g = REPEAT8(src_argb0[1]); + const uint32 r = REPEAT8(src_argb0[2]); + const uint32 a = REPEAT8(src_argb0[3]); + const uint32 b_scale = src_argb1[0]; + const uint32 g_scale = src_argb1[1]; + const uint32 r_scale = src_argb1[2]; + const uint32 a_scale = src_argb1[3]; + dst_argb[0] = SHADE(b, b_scale); + dst_argb[1] = SHADE(g, g_scale); + dst_argb[2] = SHADE(r, r_scale); + dst_argb[3] = SHADE(a, a_scale); + src_argb0 += 4; + src_argb1 += 4; + dst_argb += 4; + } +} +#undef REPEAT8 +#undef SHADE + +#define SHADE(f, v) clamp255(v + f) + +void ARGBAddRow_C(const uint8* src_argb0, const uint8* src_argb1, + uint8* dst_argb, int width) { + int i; + for (i = 0; i < width; ++i) { + const int b = src_argb0[0]; + const int g = src_argb0[1]; + const int r = src_argb0[2]; + const int a = src_argb0[3]; + const int b_add = src_argb1[0]; + const int g_add = src_argb1[1]; + const int r_add = src_argb1[2]; + const int a_add = src_argb1[3]; + dst_argb[0] = SHADE(b, b_add); + dst_argb[1] = SHADE(g, g_add); + dst_argb[2] = SHADE(r, r_add); + dst_argb[3] = SHADE(a, a_add); + src_argb0 += 4; + src_argb1 += 4; + dst_argb += 4; + } +} +#undef SHADE + +#define SHADE(f, v) clamp0(f - v) + +void ARGBSubtractRow_C(const uint8* src_argb0, const uint8* src_argb1, + uint8* dst_argb, int width) { + int i; + for (i = 0; i < width; ++i) { + const int b = src_argb0[0]; + const int g = src_argb0[1]; + const int r = src_argb0[2]; + const int a = src_argb0[3]; + const int b_sub = src_argb1[0]; + const int g_sub = src_argb1[1]; + const int r_sub = src_argb1[2]; + const int a_sub = src_argb1[3]; + dst_argb[0] = SHADE(b, b_sub); + dst_argb[1] = SHADE(g, g_sub); + dst_argb[2] = SHADE(r, r_sub); + dst_argb[3] = SHADE(a, a_sub); + src_argb0 += 4; + src_argb1 += 4; + dst_argb += 4; + } +} +#undef SHADE + +// Sobel functions which mimics SSSE3. +void SobelXRow_C(const uint8* src_y0, const uint8* src_y1, const uint8* src_y2, + uint8* dst_sobelx, int width) { + int i; + for (i = 0; i < width; ++i) { + int a = src_y0[i]; + int b = src_y1[i]; + int c = src_y2[i]; + int a_sub = src_y0[i + 2]; + int b_sub = src_y1[i + 2]; + int c_sub = src_y2[i + 2]; + int a_diff = a - a_sub; + int b_diff = b - b_sub; + int c_diff = c - c_sub; + int sobel = Abs(a_diff + b_diff * 2 + c_diff); + dst_sobelx[i] = (uint8)(clamp255(sobel)); + } +} + +void SobelYRow_C(const uint8* src_y0, const uint8* src_y1, + uint8* dst_sobely, int width) { + int i; + for (i = 0; i < width; ++i) { + int a = src_y0[i + 0]; + int b = src_y0[i + 1]; + int c = src_y0[i + 2]; + int a_sub = src_y1[i + 0]; + int b_sub = src_y1[i + 1]; + int c_sub = src_y1[i + 2]; + int a_diff = a - a_sub; + int b_diff = b - b_sub; + int c_diff = c - c_sub; + int sobel = Abs(a_diff + b_diff * 2 + c_diff); + dst_sobely[i] = (uint8)(clamp255(sobel)); + } +} + +void SobelRow_C(const uint8* src_sobelx, const uint8* src_sobely, + uint8* dst_argb, int width) { + int i; + for (i = 0; i < width; ++i) { + int r = src_sobelx[i]; + int b = src_sobely[i]; + int s = clamp255(r + b); + dst_argb[0] = (uint8)(s); + dst_argb[1] = (uint8)(s); + dst_argb[2] = (uint8)(s); + dst_argb[3] = (uint8)(255u); + dst_argb += 4; + } +} + +void SobelToPlaneRow_C(const uint8* src_sobelx, const uint8* src_sobely, + uint8* dst_y, int width) { + int i; + for (i = 0; i < width; ++i) { + int r = src_sobelx[i]; + int b = src_sobely[i]; + int s = clamp255(r + b); + dst_y[i] = (uint8)(s); + } +} + +void SobelXYRow_C(const uint8* src_sobelx, const uint8* src_sobely, + uint8* dst_argb, int width) { + int i; + for (i = 0; i < width; ++i) { + int r = src_sobelx[i]; + int b = src_sobely[i]; + int g = clamp255(r + b); + dst_argb[0] = (uint8)(b); + dst_argb[1] = (uint8)(g); + dst_argb[2] = (uint8)(r); + dst_argb[3] = (uint8)(255u); + dst_argb += 4; + } +} + +void J400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int width) { // Copy a Y to RGB. - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { uint8 y = src_y[0]; dst_argb[2] = dst_argb[1] = dst_argb[0] = y; dst_argb[3] = 255u; @@ -421,250 +965,786 @@ void I400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int width) { } } -// C reference code that mimics the YUV assembly. +// TODO(fbarchard): Unify these structures to be platform independent. +// TODO(fbarchard): Generate SIMD structures from float matrix. + +// BT.601 YUV to RGB reference +// R = (Y - 16) * 1.164 - V * -1.596 +// G = (Y - 16) * 1.164 - U * 0.391 - V * 0.813 +// B = (Y - 16) * 1.164 - U * -2.018 + +// Y contribution to R,G,B. Scale and bias. +#define YG 18997 /* round(1.164 * 64 * 256 * 256 / 257) */ +#define YGB -1160 /* 1.164 * 64 * -16 + 64 / 2 */ + +// U and V contributions to R,G,B. +#define UB -128 /* max(-128, round(-2.018 * 64)) */ +#define UG 25 /* round(0.391 * 64) */ +#define VG 52 /* round(0.813 * 64) */ +#define VR -102 /* round(-1.596 * 64) */ + +// Bias values to subtract 16 from Y and 128 from U and V. +#define BB (UB * 128 + YGB) +#define BG (UG * 128 + VG * 128 + YGB) +#define BR (VR * 128 + YGB) + +#if defined(__aarch64__) +const YuvConstants SIMD_ALIGNED(kYuvI601Constants) = { + { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR }, + { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR }, + { UG, VG, UG, VG, UG, VG, UG, VG }, + { UG, VG, UG, VG, UG, VG, UG, VG }, + { BB, BG, BR, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +const YuvConstants SIMD_ALIGNED(kYvuI601Constants) = { + { -VR, -UB, -VR, -UB, -VR, -UB, -VR, -UB }, + { -VR, -UB, -VR, -UB, -VR, -UB, -VR, -UB }, + { VG, UG, VG, UG, VG, UG, VG, UG }, + { VG, UG, VG, UG, VG, UG, VG, UG }, + { BR, BG, BB, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +#elif defined(__arm__) +const YuvConstants SIMD_ALIGNED(kYuvI601Constants) = { + { -UB, -UB, -UB, -UB, -VR, -VR, -VR, -VR, 0, 0, 0, 0, 0, 0, 0, 0 }, + { UG, UG, UG, UG, VG, VG, VG, VG, 0, 0, 0, 0, 0, 0, 0, 0 }, + { BB, BG, BR, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +const YuvConstants SIMD_ALIGNED(kYvuI601Constants) = { + { -VR, -VR, -VR, -VR, -UB, -UB, -UB, -UB, 0, 0, 0, 0, 0, 0, 0, 0 }, + { VG, VG, VG, VG, UG, UG, UG, UG, 0, 0, 0, 0, 0, 0, 0, 0 }, + { BR, BG, BB, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +#else +const YuvConstants SIMD_ALIGNED(kYuvI601Constants) = { + { UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, + UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0 }, + { UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, + UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG }, + { 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, + 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; +const YuvConstants SIMD_ALIGNED(kYvuI601Constants) = { + { VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, + VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0 }, + { VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, + VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG }, + { 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, + 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; +#endif -#define YG 74 /* static_cast<int8>(1.164 * 64 + 0.5) */ +#undef BB +#undef BG +#undef BR +#undef YGB +#undef UB +#undef UG +#undef VG +#undef VR +#undef YG + +// JPEG YUV to RGB reference +// * R = Y - V * -1.40200 +// * G = Y - U * 0.34414 - V * 0.71414 +// * B = Y - U * -1.77200 + +// Y contribution to R,G,B. Scale and bias. +#define YG 16320 /* round(1.000 * 64 * 256 * 256 / 257) */ +#define YGB 32 /* 64 / 2 */ + +// U and V contributions to R,G,B. +#define UB -113 /* round(-1.77200 * 64) */ +#define UG 22 /* round(0.34414 * 64) */ +#define VG 46 /* round(0.71414 * 64) */ +#define VR -90 /* round(-1.40200 * 64) */ + +// Bias values to round, and subtract 128 from U and V. +#define BB (UB * 128 + YGB) +#define BG (UG * 128 + VG * 128 + YGB) +#define BR (VR * 128 + YGB) + +#if defined(__aarch64__) +const YuvConstants SIMD_ALIGNED(kYuvJPEGConstants) = { + { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR }, + { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR }, + { UG, VG, UG, VG, UG, VG, UG, VG }, + { UG, VG, UG, VG, UG, VG, UG, VG }, + { BB, BG, BR, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +const YuvConstants SIMD_ALIGNED(kYvuJPEGConstants) = { + { -VR, -UB, -VR, -UB, -VR, -UB, -VR, -UB }, + { -VR, -UB, -VR, -UB, -VR, -UB, -VR, -UB }, + { VG, UG, VG, UG, VG, UG, VG, UG }, + { VG, UG, VG, UG, VG, UG, VG, UG }, + { BR, BG, BB, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +#elif defined(__arm__) +const YuvConstants SIMD_ALIGNED(kYuvJPEGConstants) = { + { -UB, -UB, -UB, -UB, -VR, -VR, -VR, -VR, 0, 0, 0, 0, 0, 0, 0, 0 }, + { UG, UG, UG, UG, VG, VG, VG, VG, 0, 0, 0, 0, 0, 0, 0, 0 }, + { BB, BG, BR, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +const YuvConstants SIMD_ALIGNED(kYvuJPEGConstants) = { + { -VR, -VR, -VR, -VR, -UB, -UB, -UB, -UB, 0, 0, 0, 0, 0, 0, 0, 0 }, + { VG, VG, VG, VG, UG, UG, UG, UG, 0, 0, 0, 0, 0, 0, 0, 0 }, + { BR, BG, BB, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +#else +const YuvConstants SIMD_ALIGNED(kYuvJPEGConstants) = { + { UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, + UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0 }, + { UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, + UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG }, + { 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, + 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; +const YuvConstants SIMD_ALIGNED(kYvuJPEGConstants) = { + { VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, + VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0 }, + { VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, + VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG }, + { 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, + 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; +#endif -#define UB 127 /* min(63,static_cast<int8>(2.018 * 64)) */ -#define UG -25 /* static_cast<int8>(-0.391 * 64 - 0.5) */ -#define UR 0 +#undef BB +#undef BG +#undef BR +#undef YGB +#undef UB +#undef UG +#undef VG +#undef VR +#undef YG + +// BT.709 YUV to RGB reference +// * R = Y - V * -1.28033 +// * G = Y - U * 0.21482 - V * 0.38059 +// * B = Y - U * -2.12798 + +// Y contribution to R,G,B. Scale and bias. +#define YG 16320 /* round(1.000 * 64 * 256 * 256 / 257) */ +#define YGB 32 /* 64 / 2 */ + +// TODO(fbarchard): Find way to express 2.12 instead of 2.0. +// U and V contributions to R,G,B. +#define UB -128 /* max(-128, round(-2.12798 * 64)) */ +#define UG 14 /* round(0.21482 * 64) */ +#define VG 24 /* round(0.38059 * 64) */ +#define VR -82 /* round(-1.28033 * 64) */ + +// Bias values to round, and subtract 128 from U and V. +#define BB (UB * 128 + YGB) +#define BG (UG * 128 + VG * 128 + YGB) +#define BR (VR * 128 + YGB) + +#if defined(__aarch64__) +const YuvConstants SIMD_ALIGNED(kYuvH709Constants) = { + { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR }, + { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR }, + { UG, VG, UG, VG, UG, VG, UG, VG }, + { UG, VG, UG, VG, UG, VG, UG, VG }, + { BB, BG, BR, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +const YuvConstants SIMD_ALIGNED(kYvuH709Constants) = { + { -VR, -UB, -VR, -UB, -VR, -UB, -VR, -UB }, + { -VR, -UB, -VR, -UB, -VR, -UB, -VR, -UB }, + { VG, UG, VG, UG, VG, UG, VG, UG }, + { VG, UG, VG, UG, VG, UG, VG, UG }, + { BR, BG, BB, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +#elif defined(__arm__) +const YuvConstants SIMD_ALIGNED(kYuvH709Constants) = { + { -UB, -UB, -UB, -UB, -VR, -VR, -VR, -VR, 0, 0, 0, 0, 0, 0, 0, 0 }, + { UG, UG, UG, UG, VG, VG, VG, VG, 0, 0, 0, 0, 0, 0, 0, 0 }, + { BB, BG, BR, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +const YuvConstants SIMD_ALIGNED(kYvuH709Constants) = { + { -VR, -VR, -VR, -VR, -UB, -UB, -UB, -UB, 0, 0, 0, 0, 0, 0, 0, 0 }, + { VG, VG, VG, VG, UG, UG, UG, UG, 0, 0, 0, 0, 0, 0, 0, 0 }, + { BR, BG, BB, 0, 0, 0, 0, 0 }, + { 0x0101 * YG, 0, 0, 0 } +}; +#else +const YuvConstants SIMD_ALIGNED(kYuvH709Constants) = { + { UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, + UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0 }, + { UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, + UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG }, + { 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, + 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; +const YuvConstants SIMD_ALIGNED(kYvuH709Constants) = { + { VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, + VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0, VR, 0 }, + { VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, + VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG, VG, UG }, + { 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, + 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB, 0, UB }, + { BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR, BR }, + { BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG, BG }, + { BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB, BB }, + { YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG, YG } +}; +#endif -#define VB 0 -#define VG -52 /* static_cast<int8>(-0.813 * 64 - 0.5) */ -#define VR 102 /* static_cast<int8>(1.596 * 64 + 0.5) */ +#undef BB +#undef BG +#undef BR +#undef YGB +#undef UB +#undef UG +#undef VG +#undef VR +#undef YG -// Bias -#define BB UB * 128 + VB * 128 -#define BG UG * 128 + VG * 128 -#define BR UR * 128 + VR * 128 +// C reference code that mimics the YUV assembly. +static __inline void YuvPixel(uint8 y, uint8 u, uint8 v, + uint8* b, uint8* g, uint8* r, + const struct YuvConstants* yuvconstants) { +#if defined(__aarch64__) + int ub = -yuvconstants->kUVToRB[0]; + int ug = yuvconstants->kUVToG[0]; + int vg = yuvconstants->kUVToG[1]; + int vr = -yuvconstants->kUVToRB[1]; + int bb = yuvconstants->kUVBiasBGR[0]; + int bg = yuvconstants->kUVBiasBGR[1]; + int br = yuvconstants->kUVBiasBGR[2]; + int yg = yuvconstants->kYToRgb[0] / 0x0101; +#elif defined(__arm__) + int ub = -yuvconstants->kUVToRB[0]; + int ug = yuvconstants->kUVToG[0]; + int vg = yuvconstants->kUVToG[4]; + int vr = -yuvconstants->kUVToRB[4]; + int bb = yuvconstants->kUVBiasBGR[0]; + int bg = yuvconstants->kUVBiasBGR[1]; + int br = yuvconstants->kUVBiasBGR[2]; + int yg = yuvconstants->kYToRgb[0] / 0x0101; +#else + int ub = yuvconstants->kUVToB[0]; + int ug = yuvconstants->kUVToG[0]; + int vg = yuvconstants->kUVToG[1]; + int vr = yuvconstants->kUVToR[1]; + int bb = yuvconstants->kUVBiasB[0]; + int bg = yuvconstants->kUVBiasG[0]; + int br = yuvconstants->kUVBiasR[0]; + int yg = yuvconstants->kYToRgb[0]; +#endif -static __inline uint32 Clip(int32 val) { - if (val < 0) { - return static_cast<uint32>(0); - } else if (val > 255) { - return static_cast<uint32>(255); - } - return static_cast<uint32>(val); + uint32 y1 = (uint32)(y * 0x0101 * yg) >> 16; + *b = Clamp((int32)(-(u * ub ) + y1 + bb) >> 6); + *g = Clamp((int32)(-(u * ug + v * vg) + y1 + bg) >> 6); + *r = Clamp((int32)(-( v * vr) + y1 + br) >> 6); } -static __inline void YuvPixel(uint8 y, uint8 u, uint8 v, uint8* rgb_buf, - int ashift, int rshift, int gshift, int bshift) { - int32 y1 = (static_cast<int32>(y) - 16) * YG; - uint32 b = Clip(static_cast<int32>((u * UB + v * VB) - (BB) + y1) >> 6); - uint32 g = Clip(static_cast<int32>((u * UG + v * VG) - (BG) + y1) >> 6); - uint32 r = Clip(static_cast<int32>((u * UR + v * VR) - (BR) + y1) >> 6); - *reinterpret_cast<uint32*>(rgb_buf) = (b << bshift) | - (g << gshift) | - (r << rshift) | - (255u << ashift); -} +// Y contribution to R,G,B. Scale and bias. +#define YG 18997 /* round(1.164 * 64 * 256 * 256 / 257) */ +#define YGB -1160 /* 1.164 * 64 * -16 + 64 / 2 */ -static __inline void YuvPixel2(uint8 y, uint8 u, uint8 v, - uint8* b, uint8* g, uint8* r) { - int32 y1 = (static_cast<int32>(y) - 16) * YG; - *b = Clip(static_cast<int32>((u * UB + v * VB) - (BB) + y1) >> 6); - *g = Clip(static_cast<int32>((u * UG + v * VG) - (BG) + y1) >> 6); - *r = Clip(static_cast<int32>((u * UR + v * VR) - (BR) + y1) >> 6); +// C reference code that mimics the YUV assembly. +static __inline void YPixel(uint8 y, uint8* b, uint8* g, uint8* r) { + uint32 y1 = (uint32)(y * 0x0101 * YG) >> 16; + *b = Clamp((int32)(y1 + YGB) >> 6); + *g = Clamp((int32)(y1 + YGB) >> 6); + *r = Clamp((int32)(y1 + YGB) >> 6); } -void I444ToARGBRow_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +#undef YG +#undef YGB + +#if !defined(LIBYUV_DISABLE_NEON) && \ + (defined(__ARM_NEON__) || defined(__aarch64__) || defined(LIBYUV_NEON)) +// C mimic assembly. +// TODO(fbarchard): Remove subsampling from Neon. +void I444ToARGBRow_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* rgb_buf, + const struct YuvConstants* yuvconstants, + int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + uint8 u = (src_u[0] + src_u[1] + 1) >> 1; + uint8 v = (src_v[0] + src_v[1] + 1) >> 1; + YuvPixel(src_y[0], u, v, rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, + yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_y[1], u, v, rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, + yuvconstants); + rgb_buf[7] = 255; + src_y += 2; + src_u += 2; + src_v += 2; + rgb_buf += 8; // Advance 2 pixels. + } + if (width & 1) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + } +} +#else +void I444ToARGBRow_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width; ++x) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf, 24, 16, 8, 0); - y_buf += 1; - u_buf += 1; - v_buf += 1; + int x; + for (x = 0; x < width; ++x) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + src_y += 1; + src_u += 1; + src_v += 1; rgb_buf += 4; // Advance 1 pixel. } } +#endif // Also used for 420 -void I422ToARGBRow_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +void I422ToARGBRow_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 24, 16, 8, 0); - YuvPixel(y_buf[1], u_buf[0], v_buf[0], rgb_buf + 4, 24, 16, 8, 0); - y_buf += 2; - u_buf += 1; - v_buf += 1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_y[1], src_u[0], src_v[0], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = 255; + src_y += 2; + src_u += 1; + src_v += 1; + rgb_buf += 8; // Advance 2 pixels. + } + if (width & 1) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + } +} + +void I422AlphaToARGBRow_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + const uint8* src_a, + uint8* rgb_buf, + const struct YuvConstants* yuvconstants, + int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = src_a[0]; + YuvPixel(src_y[1], src_u[0], src_v[0], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = src_a[1]; + src_y += 2; + src_u += 1; + src_v += 1; + src_a += 2; rgb_buf += 8; // Advance 2 pixels. } if (width & 1) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 24, 16, 8, 0); + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = src_a[0]; } } -void I422ToRGB24Row_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +void I422ToRGB24Row_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel2(y_buf[0], u_buf[0], v_buf[0], - rgb_buf + 0, rgb_buf + 1, rgb_buf + 2); - YuvPixel2(y_buf[1], u_buf[0], v_buf[0], - rgb_buf + 3, rgb_buf + 4, rgb_buf + 5); - y_buf += 2; - u_buf += 1; - v_buf += 1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + YuvPixel(src_y[1], src_u[0], src_v[0], + rgb_buf + 3, rgb_buf + 4, rgb_buf + 5, yuvconstants); + src_y += 2; + src_u += 1; + src_v += 1; rgb_buf += 6; // Advance 2 pixels. } if (width & 1) { - YuvPixel2(y_buf[0], u_buf[0], v_buf[0], - rgb_buf + 0, rgb_buf + 1, rgb_buf + 2); - } -} - -void I422ToRAWRow_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* rgb_buf, - int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel2(y_buf[0], u_buf[0], v_buf[0], - rgb_buf + 2, rgb_buf + 1, rgb_buf + 0); - YuvPixel2(y_buf[1], u_buf[0], v_buf[0], - rgb_buf + 5, rgb_buf + 4, rgb_buf + 3); - y_buf += 2; - u_buf += 1; - v_buf += 1; - rgb_buf += 6; // Advance 2 pixels. + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + } +} + +void I422ToARGB4444Row_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb4444, + const struct YuvConstants* yuvconstants, + int width) { + uint8 b0; + uint8 g0; + uint8 r0; + uint8 b1; + uint8 g1; + uint8 r1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_u[0], src_v[0], &b0, &g0, &r0, yuvconstants); + YuvPixel(src_y[1], src_u[0], src_v[0], &b1, &g1, &r1, yuvconstants); + b0 = b0 >> 4; + g0 = g0 >> 4; + r0 = r0 >> 4; + b1 = b1 >> 4; + g1 = g1 >> 4; + r1 = r1 >> 4; + *(uint32*)(dst_argb4444) = b0 | (g0 << 4) | (r0 << 8) | + (b1 << 16) | (g1 << 20) | (r1 << 24) | 0xf000f000; + src_y += 2; + src_u += 1; + src_v += 1; + dst_argb4444 += 4; // Advance 2 pixels. } if (width & 1) { - YuvPixel2(y_buf[0], u_buf[0], v_buf[0], - rgb_buf + 0, rgb_buf + 1, rgb_buf + 2); + YuvPixel(src_y[0], src_u[0], src_v[0], &b0, &g0, &r0, yuvconstants); + b0 = b0 >> 4; + g0 = g0 >> 4; + r0 = r0 >> 4; + *(uint16*)(dst_argb4444) = b0 | (g0 << 4) | (r0 << 8) | + 0xf000; } } -void I411ToARGBRow_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +void I422ToARGB1555Row_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb1555, + const struct YuvConstants* yuvconstants, + int width) { + uint8 b0; + uint8 g0; + uint8 r0; + uint8 b1; + uint8 g1; + uint8 r1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_u[0], src_v[0], &b0, &g0, &r0, yuvconstants); + YuvPixel(src_y[1], src_u[0], src_v[0], &b1, &g1, &r1, yuvconstants); + b0 = b0 >> 3; + g0 = g0 >> 3; + r0 = r0 >> 3; + b1 = b1 >> 3; + g1 = g1 >> 3; + r1 = r1 >> 3; + *(uint32*)(dst_argb1555) = b0 | (g0 << 5) | (r0 << 10) | + (b1 << 16) | (g1 << 21) | (r1 << 26) | 0x80008000; + src_y += 2; + src_u += 1; + src_v += 1; + dst_argb1555 += 4; // Advance 2 pixels. + } + if (width & 1) { + YuvPixel(src_y[0], src_u[0], src_v[0], &b0, &g0, &r0, yuvconstants); + b0 = b0 >> 3; + g0 = g0 >> 3; + r0 = r0 >> 3; + *(uint16*)(dst_argb1555) = b0 | (g0 << 5) | (r0 << 10) | + 0x8000; + } +} + +void I422ToRGB565Row_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_rgb565, + const struct YuvConstants* yuvconstants, + int width) { + uint8 b0; + uint8 g0; + uint8 r0; + uint8 b1; + uint8 g1; + uint8 r1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_u[0], src_v[0], &b0, &g0, &r0, yuvconstants); + YuvPixel(src_y[1], src_u[0], src_v[0], &b1, &g1, &r1, yuvconstants); + b0 = b0 >> 3; + g0 = g0 >> 2; + r0 = r0 >> 3; + b1 = b1 >> 3; + g1 = g1 >> 2; + r1 = r1 >> 3; + *(uint32*)(dst_rgb565) = b0 | (g0 << 5) | (r0 << 11) | + (b1 << 16) | (g1 << 21) | (r1 << 27); + src_y += 2; + src_u += 1; + src_v += 1; + dst_rgb565 += 4; // Advance 2 pixels. + } + if (width & 1) { + YuvPixel(src_y[0], src_u[0], src_v[0], &b0, &g0, &r0, yuvconstants); + b0 = b0 >> 3; + g0 = g0 >> 2; + r0 = r0 >> 3; + *(uint16*)(dst_rgb565) = b0 | (g0 << 5) | (r0 << 11); + } +} + +void I411ToARGBRow_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 3; x += 4) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 24, 16, 8, 0); - YuvPixel(y_buf[1], u_buf[0], v_buf[0], rgb_buf + 4, 24, 16, 8, 0); - YuvPixel(y_buf[2], u_buf[0], v_buf[0], rgb_buf + 8, 24, 16, 8, 0); - YuvPixel(y_buf[3], u_buf[0], v_buf[0], rgb_buf + 12, 24, 16, 8, 0); - y_buf += 4; - u_buf += 1; - v_buf += 1; + int x; + for (x = 0; x < width - 3; x += 4) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_y[1], src_u[0], src_v[0], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = 255; + YuvPixel(src_y[2], src_u[0], src_v[0], + rgb_buf + 8, rgb_buf + 9, rgb_buf + 10, yuvconstants); + rgb_buf[11] = 255; + YuvPixel(src_y[3], src_u[0], src_v[0], + rgb_buf + 12, rgb_buf + 13, rgb_buf + 14, yuvconstants); + rgb_buf[15] = 255; + src_y += 4; + src_u += 1; + src_v += 1; rgb_buf += 16; // Advance 4 pixels. } if (width & 2) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 24, 16, 8, 0); - YuvPixel(y_buf[1], u_buf[0], v_buf[0], rgb_buf + 4, 24, 16, 8, 0); - y_buf += 2; + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_y[1], src_u[0], src_v[0], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = 255; + src_y += 2; rgb_buf += 8; // Advance 2 pixels. } if (width & 1) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 24, 16, 8, 0); + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; } } -void NV12ToARGBRow_C(const uint8* y_buf, - const uint8* uv_buf, +void NV12ToARGBRow_C(const uint8* src_y, + const uint8* src_uv, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel(y_buf[0], uv_buf[0], uv_buf[1], rgb_buf + 0, 24, 16, 8, 0); - YuvPixel(y_buf[1], uv_buf[0], uv_buf[1], rgb_buf + 4, 24, 16, 8, 0); - y_buf += 2; - uv_buf += 2; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_uv[0], src_uv[1], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_y[1], src_uv[0], src_uv[1], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = 255; + src_y += 2; + src_uv += 2; rgb_buf += 8; // Advance 2 pixels. } if (width & 1) { - YuvPixel(y_buf[0], uv_buf[0], uv_buf[1], rgb_buf + 0, 24, 16, 8, 0); + YuvPixel(src_y[0], src_uv[0], src_uv[1], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; } } -void NV21ToARGBRow_C(const uint8* y_buf, - const uint8* vu_buf, +void NV21ToARGBRow_C(const uint8* src_y, + const uint8* src_vu, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel(y_buf[0], vu_buf[1], vu_buf[0], rgb_buf + 0, 24, 16, 8, 0); - YuvPixel(y_buf[1], vu_buf[1], vu_buf[0], rgb_buf + 4, 24, 16, 8, 0); - y_buf += 2; - vu_buf += 2; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_vu[1], src_vu[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_y[1], src_vu[1], src_vu[0], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = 255; + src_y += 2; + src_vu += 2; rgb_buf += 8; // Advance 2 pixels. } if (width & 1) { - YuvPixel(y_buf[0], vu_buf[1], vu_buf[0], rgb_buf + 0, 24, 16, 8, 0); + YuvPixel(src_y[0], src_vu[1], src_vu[0], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + } +} + +void NV12ToRGB565Row_C(const uint8* src_y, + const uint8* src_uv, + uint8* dst_rgb565, + const struct YuvConstants* yuvconstants, + int width) { + uint8 b0; + uint8 g0; + uint8 r0; + uint8 b1; + uint8 g1; + uint8 r1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_uv[0], src_uv[1], &b0, &g0, &r0, yuvconstants); + YuvPixel(src_y[1], src_uv[0], src_uv[1], &b1, &g1, &r1, yuvconstants); + b0 = b0 >> 3; + g0 = g0 >> 2; + r0 = r0 >> 3; + b1 = b1 >> 3; + g1 = g1 >> 2; + r1 = r1 >> 3; + *(uint32*)(dst_rgb565) = b0 | (g0 << 5) | (r0 << 11) | + (b1 << 16) | (g1 << 21) | (r1 << 27); + src_y += 2; + src_uv += 2; + dst_rgb565 += 4; // Advance 2 pixels. + } + if (width & 1) { + YuvPixel(src_y[0], src_uv[0], src_uv[1], &b0, &g0, &r0, yuvconstants); + b0 = b0 >> 3; + g0 = g0 >> 2; + r0 = r0 >> 3; + *(uint16*)(dst_rgb565) = b0 | (g0 << 5) | (r0 << 11); } } -void I422ToBGRARow_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +void YUY2ToARGBRow_C(const uint8* src_yuy2, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 0, 8, 16, 24); - YuvPixel(y_buf[1], u_buf[0], v_buf[0], rgb_buf + 4, 0, 8, 16, 24); - y_buf += 2; - u_buf += 1; - v_buf += 1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_yuy2[0], src_yuy2[1], src_yuy2[3], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_yuy2[2], src_yuy2[1], src_yuy2[3], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = 255; + src_yuy2 += 4; rgb_buf += 8; // Advance 2 pixels. } if (width & 1) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf, 0, 8, 16, 24); + YuvPixel(src_yuy2[0], src_yuy2[1], src_yuy2[3], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; } } -void I422ToABGRRow_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +void UYVYToARGBRow_C(const uint8* src_uyvy, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 24, 0, 8, 16); - YuvPixel(y_buf[1], u_buf[0], v_buf[0], rgb_buf + 4, 24, 0, 8, 16); - y_buf += 2; - u_buf += 1; - v_buf += 1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_uyvy[1], src_uyvy[0], src_uyvy[2], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; + YuvPixel(src_uyvy[3], src_uyvy[0], src_uyvy[2], + rgb_buf + 4, rgb_buf + 5, rgb_buf + 6, yuvconstants); + rgb_buf[7] = 255; + src_uyvy += 4; rgb_buf += 8; // Advance 2 pixels. } if (width & 1) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 24, 0, 8, 16); + YuvPixel(src_uyvy[1], src_uyvy[0], src_uyvy[2], + rgb_buf + 0, rgb_buf + 1, rgb_buf + 2, yuvconstants); + rgb_buf[3] = 255; } } -void I422ToRGBARow_C(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, +void I422ToRGBARow_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, uint8* rgb_buf, + const struct YuvConstants* yuvconstants, int width) { - for (int x = 0; x < width - 1; x += 2) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 0, 24, 16, 8); - YuvPixel(y_buf[1], u_buf[0], v_buf[0], rgb_buf + 4, 0, 24, 16, 8); - y_buf += 2; - u_buf += 1; - v_buf += 1; + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 1, rgb_buf + 2, rgb_buf + 3, yuvconstants); + rgb_buf[0] = 255; + YuvPixel(src_y[1], src_u[0], src_v[0], + rgb_buf + 5, rgb_buf + 6, rgb_buf + 7, yuvconstants); + rgb_buf[4] = 255; + src_y += 2; + src_u += 1; + src_v += 1; rgb_buf += 8; // Advance 2 pixels. } if (width & 1) { - YuvPixel(y_buf[0], u_buf[0], v_buf[0], rgb_buf + 0, 0, 24, 16, 8); + YuvPixel(src_y[0], src_u[0], src_v[0], + rgb_buf + 1, rgb_buf + 2, rgb_buf + 3, yuvconstants); + rgb_buf[0] = 255; } } -void YToARGBRow_C(const uint8* y_buf, uint8* rgb_buf, int width) { - for (int x = 0; x < width; ++x) { - YuvPixel(y_buf[0], 128, 128, rgb_buf, 24, 16, 8, 0); - y_buf += 1; - rgb_buf += 4; // Advance 1 pixel. +void I400ToARGBRow_C(const uint8* src_y, uint8* rgb_buf, int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + YPixel(src_y[0], rgb_buf + 0, rgb_buf + 1, rgb_buf + 2); + rgb_buf[3] = 255; + YPixel(src_y[1], rgb_buf + 4, rgb_buf + 5, rgb_buf + 6); + rgb_buf[7] = 255; + src_y += 2; + rgb_buf += 8; // Advance 2 pixels. + } + if (width & 1) { + YPixel(src_y[0], rgb_buf + 0, rgb_buf + 1, rgb_buf + 2); + rgb_buf[3] = 255; } } void MirrorRow_C(const uint8* src, uint8* dst, int width) { + int x; src += width - 1; - for (int x = 0; x < width - 1; x += 2) { + for (x = 0; x < width - 1; x += 2) { dst[x] = src[0]; dst[x + 1] = src[-1]; src -= 2; @@ -674,9 +1754,10 @@ void MirrorRow_C(const uint8* src, uint8* dst, int width) { } } -void MirrorRowUV_C(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { +void MirrorUVRow_C(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { + int x; src_uv += (width - 1) << 1; - for (int x = 0; x < width - 1; x += 2) { + for (x = 0; x < width - 1; x += 2) { dst_u[x] = src_uv[0]; dst_u[x + 1] = src_uv[-2]; dst_v[x] = src_uv[1]; @@ -690,10 +1771,11 @@ void MirrorRowUV_C(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { } void ARGBMirrorRow_C(const uint8* src, uint8* dst, int width) { - const uint32* src32 = reinterpret_cast<const uint32*>(src); - uint32* dst32 = reinterpret_cast<uint32*>(dst); + int x; + const uint32* src32 = (const uint32*)(src); + uint32* dst32 = (uint32*)(dst); src32 += width - 1; - for (int x = 0; x < width - 1; x += 2) { + for (x = 0; x < width - 1; x += 2) { dst32[x] = src32[0]; dst32[x + 1] = src32[-1]; src32 -= 2; @@ -703,8 +1785,9 @@ void ARGBMirrorRow_C(const uint8* src, uint8* dst, int width) { } } -void SplitUV_C(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { - for (int x = 0; x < width - 1; x += 2) { +void SplitUVRow_C(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { + int x; + for (x = 0; x < width - 1; x += 2) { dst_u[x] = src_uv[0]; dst_u[x + 1] = src_uv[2]; dst_v[x] = src_uv[1]; @@ -717,29 +1800,39 @@ void SplitUV_C(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { } } +void MergeUVRow_C(const uint8* src_u, const uint8* src_v, uint8* dst_uv, + int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + dst_uv[0] = src_u[x]; + dst_uv[1] = src_v[x]; + dst_uv[2] = src_u[x + 1]; + dst_uv[3] = src_v[x + 1]; + dst_uv += 4; + } + if (width & 1) { + dst_uv[0] = src_u[width - 1]; + dst_uv[1] = src_v[width - 1]; + } +} + void CopyRow_C(const uint8* src, uint8* dst, int count) { memcpy(dst, src, count); } -void SetRow8_C(uint8* dst, uint32 v8, int count) { -#ifdef _MSC_VER - // VC will generate rep stosb. - for (int x = 0; x < count; ++x) { - dst[x] = v8; - } -#else - memset(dst, v8, count); -#endif +void CopyRow_16_C(const uint16* src, uint16* dst, int count) { + memcpy(dst, src, count * 2); +} + +void SetRow_C(uint8* dst, uint8 v8, int width) { + memset(dst, v8, width); } -void SetRows32_C(uint8* dst, uint32 v32, int width, - int dst_stride, int height) { - for (int y = 0; y < height; ++y) { - uint32* d = reinterpret_cast<uint32*>(dst); - for (int x = 0; x < width; ++x) { - d[x] = v32; - } - dst += dst_stride; +void ARGBSetRow_C(uint8* dst_argb, uint32 v32, int width) { + uint32* d = (uint32*)(dst_argb); + int x; + for (x = 0; x < width; ++x) { + d[x] = v32; } } @@ -747,7 +1840,8 @@ void SetRows32_C(uint8* dst, uint32 v32, int width, void YUY2ToUVRow_C(const uint8* src_yuy2, int src_stride_yuy2, uint8* dst_u, uint8* dst_v, int width) { // Output a row of UV values, filtering 2 rows of YUY2. - for (int x = 0; x < width; x += 2) { + int x; + for (x = 0; x < width; x += 2) { dst_u[0] = (src_yuy2[1] + src_yuy2[src_stride_yuy2 + 1] + 1) >> 1; dst_v[0] = (src_yuy2[3] + src_yuy2[src_stride_yuy2 + 3] + 1) >> 1; src_yuy2 += 4; @@ -760,7 +1854,8 @@ void YUY2ToUVRow_C(const uint8* src_yuy2, int src_stride_yuy2, void YUY2ToUV422Row_C(const uint8* src_yuy2, uint8* dst_u, uint8* dst_v, int width) { // Output a row of UV values. - for (int x = 0; x < width; x += 2) { + int x; + for (x = 0; x < width; x += 2) { dst_u[0] = src_yuy2[1]; dst_v[0] = src_yuy2[3]; src_yuy2 += 4; @@ -772,7 +1867,8 @@ void YUY2ToUV422Row_C(const uint8* src_yuy2, // Copy row of YUY2 Y's (422) into Y (420/422). void YUY2ToYRow_C(const uint8* src_yuy2, uint8* dst_y, int width) { // Output a row of Y values. - for (int x = 0; x < width - 1; x += 2) { + int x; + for (x = 0; x < width - 1; x += 2) { dst_y[x] = src_yuy2[0]; dst_y[x + 1] = src_yuy2[2]; src_yuy2 += 4; @@ -786,7 +1882,8 @@ void YUY2ToYRow_C(const uint8* src_yuy2, uint8* dst_y, int width) { void UYVYToUVRow_C(const uint8* src_uyvy, int src_stride_uyvy, uint8* dst_u, uint8* dst_v, int width) { // Output a row of UV values. - for (int x = 0; x < width; x += 2) { + int x; + for (x = 0; x < width; x += 2) { dst_u[0] = (src_uyvy[0] + src_uyvy[src_stride_uyvy + 0] + 1) >> 1; dst_v[0] = (src_uyvy[2] + src_uyvy[src_stride_uyvy + 2] + 1) >> 1; src_uyvy += 4; @@ -799,7 +1896,8 @@ void UYVYToUVRow_C(const uint8* src_uyvy, int src_stride_uyvy, void UYVYToUV422Row_C(const uint8* src_uyvy, uint8* dst_u, uint8* dst_v, int width) { // Output a row of UV values. - for (int x = 0; x < width; x += 2) { + int x; + for (x = 0; x < width; x += 2) { dst_u[0] = src_uyvy[0]; dst_v[0] = src_uyvy[2]; src_uyvy += 4; @@ -811,7 +1909,8 @@ void UYVYToUV422Row_C(const uint8* src_uyvy, // Copy row of UYVY Y's (422) into Y (420/422). void UYVYToYRow_C(const uint8* src_uyvy, uint8* dst_y, int width) { // Output a row of Y values. - for (int x = 0; x < width - 1; x += 2) { + int x; + for (x = 0; x < width - 1; x += 2) { dst_y[x] = src_uyvy[1]; dst_y[x + 1] = src_uyvy[3]; src_uyvy += 4; @@ -828,7 +1927,8 @@ void UYVYToYRow_C(const uint8* src_uyvy, uint8* dst_y, int width) { // This code mimics the SSSE3 version for better testability. void ARGBBlendRow_C(const uint8* src_argb0, const uint8* src_argb1, uint8* dst_argb, int width) { - for (int x = 0; x < width - 1; x += 2) { + int x; + for (x = 0; x < width - 1; x += 2) { uint32 fb = src_argb0[0]; uint32 fg = src_argb0[1]; uint32 fr = src_argb0[2]; @@ -872,12 +1972,32 @@ void ARGBBlendRow_C(const uint8* src_argb0, const uint8* src_argb1, } } #undef BLEND + +#define UBLEND(f, b, a) (((a) * f) + ((255 - a) * b) + 255) >> 8 +void BlendPlaneRow_C(const uint8* src0, const uint8* src1, + const uint8* alpha, uint8* dst, int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + dst[0] = UBLEND(src0[0], src1[0], alpha[0]); + dst[1] = UBLEND(src0[1], src1[1], alpha[1]); + src0 += 2; + src1 += 2; + alpha += 2; + dst += 2; + } + if (width & 1) { + dst[0] = UBLEND(src0[0], src1[0], alpha[0]); + } +} +#undef UBLEND + #define ATTENUATE(f, a) (a | (a << 8)) * (f | (f << 8)) >> 24 // Multiply source RGB by alpha and store to destination. // This code mimics the SSSE3 version for better testability. void ARGBAttenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width) { - for (int i = 0; i < width - 1; i += 2) { + int i; + for (i = 0; i < width - 1; i += 2) { uint32 b = src_argb[0]; uint32 g = src_argb[1]; uint32 r = src_argb[2]; @@ -916,10 +2036,10 @@ void ARGBAttenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width) { // g = (g * 255 + (a / 2)) / a; // r = (r * 255 + (a / 2)) / a; // Reciprocal method is off by 1 on some values. ie 125 -// 8.16 fixed point inverse table -#define T(a) 0x10000 / a -uint32 fixed_invtbl8[256] = { - 0x0100, T(0x01), T(0x02), T(0x03), T(0x04), T(0x05), T(0x06), T(0x07), +// 8.8 fixed point inverse table with 1.0 in upper short and 1 / a in lower. +#define T(a) 0x01000000 + (0x10000 / a) +const uint32 fixed_invtbl8[256] = { + 0x01000000, 0x0100ffff, T(0x02), T(0x03), T(0x04), T(0x05), T(0x06), T(0x07), T(0x08), T(0x09), T(0x0a), T(0x0b), T(0x0c), T(0x0d), T(0x0e), T(0x0f), T(0x10), T(0x11), T(0x12), T(0x13), T(0x14), T(0x15), T(0x16), T(0x17), T(0x18), T(0x19), T(0x1a), T(0x1b), T(0x1c), T(0x1d), T(0x1e), T(0x1f), @@ -950,202 +2070,35 @@ uint32 fixed_invtbl8[256] = { T(0xe0), T(0xe1), T(0xe2), T(0xe3), T(0xe4), T(0xe5), T(0xe6), T(0xe7), T(0xe8), T(0xe9), T(0xea), T(0xeb), T(0xec), T(0xed), T(0xee), T(0xef), T(0xf0), T(0xf1), T(0xf2), T(0xf3), T(0xf4), T(0xf5), T(0xf6), T(0xf7), - T(0xf8), T(0xf9), T(0xfa), T(0xfb), T(0xfc), T(0xfd), T(0xfe), 0x0100 }; + T(0xf8), T(0xf9), T(0xfa), T(0xfb), T(0xfc), T(0xfd), T(0xfe), 0x01000100 }; #undef T void ARGBUnattenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width) { - for (int i = 0; i < width; ++i) { + int i; + for (i = 0; i < width; ++i) { uint32 b = src_argb[0]; uint32 g = src_argb[1]; uint32 r = src_argb[2]; const uint32 a = src_argb[3]; - if (a) { - const uint32 ia = fixed_invtbl8[a]; // 8.16 fixed point - b = (b * ia) >> 8; - g = (g * ia) >> 8; - r = (r * ia) >> 8; - // Clamping should not be necessary but is free in assembly. - if (b > 255) { - b = 255; - } - if (g > 255) { - g = 255; - } - if (r > 255) { - r = 255; - } - } - dst_argb[0] = b; - dst_argb[1] = g; - dst_argb[2] = r; + const uint32 ia = fixed_invtbl8[a] & 0xffff; // 8.8 fixed point + b = (b * ia) >> 8; + g = (g * ia) >> 8; + r = (r * ia) >> 8; + // Clamping should not be necessary but is free in assembly. + dst_argb[0] = clamp255(b); + dst_argb[1] = clamp255(g); + dst_argb[2] = clamp255(r); dst_argb[3] = a; src_argb += 4; dst_argb += 4; } } -// Wrappers to handle odd width -#define YANY(NAMEANY, I420TORGB_SSE, I420TORGB_C, UV_SHIFT) \ - void NAMEANY(const uint8* y_buf, \ - const uint8* u_buf, \ - const uint8* v_buf, \ - uint8* rgb_buf, \ - int width) { \ - int n = width & ~7; \ - I420TORGB_SSE(y_buf, u_buf, v_buf, rgb_buf, n); \ - I420TORGB_C(y_buf + n, \ - u_buf + (n >> UV_SHIFT), \ - v_buf + (n >> UV_SHIFT), \ - rgb_buf + n * 4, width & 7); \ - } - -// Wrappers to handle odd width -#define Y2NY(NAMEANY, NV12TORGB_SSE, NV12TORGB_C, UV_SHIFT) \ - void NAMEANY(const uint8* y_buf, \ - const uint8* uv_buf, \ - uint8* rgb_buf, \ - int width) { \ - int n = width & ~7; \ - NV12TORGB_SSE(y_buf, uv_buf, rgb_buf, n); \ - NV12TORGB_C(y_buf + n, \ - uv_buf + (n >> UV_SHIFT), \ - rgb_buf + n * 4, width & 7); \ - } - - -#ifdef HAS_I422TOARGBROW_SSSE3 -YANY(I444ToARGBRow_Any_SSSE3, I444ToARGBRow_Unaligned_SSSE3, I444ToARGBRow_C, 0) -YANY(I422ToARGBRow_Any_SSSE3, I422ToARGBRow_Unaligned_SSSE3, I422ToARGBRow_C, 1) -YANY(I411ToARGBRow_Any_SSSE3, I411ToARGBRow_Unaligned_SSSE3, I411ToARGBRow_C, 2) -Y2NY(NV12ToARGBRow_Any_SSSE3, NV12ToARGBRow_Unaligned_SSSE3, NV12ToARGBRow_C, 0) -Y2NY(NV21ToARGBRow_Any_SSSE3, NV21ToARGBRow_Unaligned_SSSE3, NV21ToARGBRow_C, 0) -YANY(I422ToBGRARow_Any_SSSE3, I422ToBGRARow_Unaligned_SSSE3, I422ToBGRARow_C, 1) -YANY(I422ToABGRRow_Any_SSSE3, I422ToABGRRow_Unaligned_SSSE3, I422ToABGRRow_C, 1) -#endif -#ifdef HAS_I422TORGB24ROW_SSSE3 -YANY(I422ToRGB24Row_Any_SSSE3, I422ToRGB24Row_Unaligned_SSSE3, \ - I422ToRGB24Row_C, 1) -YANY(I422ToRAWRow_Any_SSSE3, I422ToRAWRow_Unaligned_SSSE3, I422ToRAWRow_C, 1) -#endif -#ifdef HAS_I422TORGBAROW_SSSE3 -YANY(I422ToRGBARow_Any_SSSE3, I422ToRGBARow_Unaligned_SSSE3, I422ToRGBARow_C, 1) -#endif -#ifdef HAS_I422TOARGBROW_NEON -YANY(I422ToARGBRow_Any_NEON, I422ToARGBRow_NEON, I422ToARGBRow_C, 1) -YANY(I422ToBGRARow_Any_NEON, I422ToBGRARow_NEON, I422ToBGRARow_C, 1) -YANY(I422ToABGRRow_Any_NEON, I422ToABGRRow_NEON, I422ToABGRRow_C, 1) -YANY(I422ToRGBARow_Any_NEON, I422ToRGBARow_NEON, I422ToRGBARow_C, 1) -Y2NY(NV12ToARGBRow_Any_NEON, NV12ToARGBRow_NEON, NV12ToARGBRow_C, 0) -Y2NY(NV21ToARGBRow_Any_NEON, NV21ToARGBRow_NEON, NV21ToARGBRow_C, 0) -YANY(I422ToRGB24Row_Any_NEON, I422ToRGB24Row_NEON, I422ToRGB24Row_C, 1) -YANY(I422ToRAWRow_Any_NEON, I422ToRAWRow_NEON, I422ToRAWRow_C, 1) -#endif -#undef YANY - -#define RGBANY(NAMEANY, ARGBTORGB, BPP) \ - void NAMEANY(const uint8* argb_buf, \ - uint8* rgb_buf, \ - int width) { \ - SIMD_ALIGNED(uint8 row[kMaxStride]); \ - ARGBTORGB(argb_buf, row, width); \ - memcpy(rgb_buf, row, width * BPP); \ - } - -#if defined(HAS_ARGBTORGB24ROW_SSSE3) -RGBANY(ARGBToRGB24Row_Any_SSSE3, ARGBToRGB24Row_SSSE3, 3) -RGBANY(ARGBToRAWRow_Any_SSSE3, ARGBToRAWRow_SSSE3, 3) -RGBANY(ARGBToRGB565Row_Any_SSE2, ARGBToRGB565Row_SSE2, 2) -RGBANY(ARGBToARGB1555Row_Any_SSE2, ARGBToARGB1555Row_SSE2, 2) -RGBANY(ARGBToARGB4444Row_Any_SSE2, ARGBToARGB4444Row_SSE2, 2) -#endif -#if defined(HAS_ARGBTORGB24ROW_NEON) -RGBANY(ARGBToRGB24Row_Any_NEON, ARGBToRGB24Row_NEON, 3) -RGBANY(ARGBToRAWRow_Any_NEON, ARGBToRAWRow_NEON, 3) -#endif -#undef RGBANY - -#define YANY(NAMEANY, ARGBTOY_SSE, BPP) \ - void NAMEANY(const uint8* src_argb, uint8* dst_y, int width) { \ - ARGBTOY_SSE(src_argb, dst_y, width - 16); \ - ARGBTOY_SSE(src_argb + (width - 16) * BPP, dst_y + (width - 16), 16); \ - } - -#ifdef HAS_ARGBTOYROW_SSSE3 -YANY(ARGBToYRow_Any_SSSE3, ARGBToYRow_Unaligned_SSSE3, 4) -YANY(BGRAToYRow_Any_SSSE3, BGRAToYRow_Unaligned_SSSE3, 4) -YANY(ABGRToYRow_Any_SSSE3, ABGRToYRow_Unaligned_SSSE3, 4) -#endif -#ifdef HAS_RGBATOYROW_SSSE3 -YANY(RGBAToYRow_Any_SSSE3, RGBAToYRow_Unaligned_SSSE3, 4) -#endif -#ifdef HAS_YUY2TOYROW_SSE2 -YANY(YUY2ToYRow_Any_SSE2, YUY2ToYRow_Unaligned_SSE2, 2) -YANY(UYVYToYRow_Any_SSE2, UYVYToYRow_Unaligned_SSE2, 2) -#endif -#ifdef HAS_YUY2TOYROW_NEON -YANY(YUY2ToYRow_Any_NEON, YUY2ToYRow_NEON, 2) -YANY(UYVYToYRow_Any_NEON, UYVYToYRow_NEON, 2) -#endif -#undef YANY - -#define UVANY(NAMEANY, ANYTOUV_SSE, ANYTOUV_C, BPP) \ - void NAMEANY(const uint8* src_argb, int src_stride_argb, \ - uint8* dst_u, uint8* dst_v, int width) { \ - int n = width & ~15; \ - ANYTOUV_SSE(src_argb, src_stride_argb, dst_u, dst_v, n); \ - ANYTOUV_C(src_argb + n * BPP, src_stride_argb, \ - dst_u + (n >> 1), \ - dst_v + (n >> 1), \ - width & 15); \ - } - -#ifdef HAS_ARGBTOUVROW_SSSE3 -UVANY(ARGBToUVRow_Any_SSSE3, ARGBToUVRow_Unaligned_SSSE3, ARGBToUVRow_C, 4) -UVANY(BGRAToUVRow_Any_SSSE3, BGRAToUVRow_Unaligned_SSSE3, BGRAToUVRow_C, 4) -UVANY(ABGRToUVRow_Any_SSSE3, ABGRToUVRow_Unaligned_SSSE3, ABGRToUVRow_C, 4) -#endif -#ifdef HAS_RGBATOYROW_SSSE3 -UVANY(RGBAToUVRow_Any_SSSE3, RGBAToUVRow_Unaligned_SSSE3, RGBAToUVRow_C, 4) -#endif -#ifdef HAS_YUY2TOUVROW_SSE2 -UVANY(YUY2ToUVRow_Any_SSE2, YUY2ToUVRow_Unaligned_SSE2, YUY2ToUVRow_C, 2) -UVANY(UYVYToUVRow_Any_SSE2, UYVYToUVRow_Unaligned_SSE2, UYVYToUVRow_C, 2) -#endif -#ifdef HAS_YUY2TOUVROW_NEON -UVANY(YUY2ToUVRow_Any_NEON, YUY2ToUVRow_NEON, YUY2ToUVRow_C, 2) -UVANY(UYVYToUVRow_Any_NEON, UYVYToUVRow_NEON, UYVYToUVRow_C, 2) -#endif -#undef UVANY - -#define UV422ANY(NAMEANY, ANYTOUV_SSE, ANYTOUV_C, BPP) \ - void NAMEANY(const uint8* src_argb, \ - uint8* dst_u, uint8* dst_v, int width) { \ - int n = width & ~15; \ - ANYTOUV_SSE(src_argb, dst_u, dst_v, n); \ - ANYTOUV_C(src_argb + n * BPP, \ - dst_u + (n >> 1), \ - dst_v + (n >> 1), \ - width & 15); \ - } - -#ifdef HAS_YUY2TOUV422ROW_SSE2 -UV422ANY(YUY2ToUV422Row_Any_SSE2, YUY2ToUV422Row_Unaligned_SSE2, \ - YUY2ToUV422Row_C, 2) -UV422ANY(UYVYToUV422Row_Any_SSE2, UYVYToUV422Row_Unaligned_SSE2, \ - UYVYToUV422Row_C, 2) -#endif -#ifdef HAS_YUY2TOUV422ROW_NEON -UV422ANY(YUY2ToUV422Row_Any_NEON, YUY2ToUV422Row_NEON, \ - YUY2ToUV422Row_C, 2) -UV422ANY(UYVYToUV422Row_Any_NEON, UYVYToUV422Row_NEON, \ - UYVYToUV422Row_C, 2) -#endif -#undef UV422ANY - void ComputeCumulativeSumRow_C(const uint8* row, int32* cumsum, const int32* previous_cumsum, int width) { int32 row_sum[4] = {0, 0, 0, 0}; - for (int x = 0; x < width; ++x) { + int x; + for (x = 0; x < width; ++x) { row_sum[0] += row[x * 4 + 0]; row_sum[1] += row[x * 4 + 1]; row_sum[2] += row[x * 4 + 2]; @@ -1157,59 +2110,35 @@ void ComputeCumulativeSumRow_C(const uint8* row, int32* cumsum, } } -void CumulativeSumToAverage_C(const int32* tl, const int32* bl, - int w, int area, uint8* dst, int count) { +void CumulativeSumToAverageRow_C(const int32* tl, const int32* bl, + int w, int area, uint8* dst, int count) { float ooa = 1.0f / area; - for (int i = 0; i < count; ++i) { - dst[0] = static_cast<uint8>((bl[w + 0] + tl[0] - bl[0] - tl[w + 0]) * ooa); - dst[1] = static_cast<uint8>((bl[w + 1] + tl[1] - bl[1] - tl[w + 1]) * ooa); - dst[2] = static_cast<uint8>((bl[w + 2] + tl[2] - bl[2] - tl[w + 2]) * ooa); - dst[3] = static_cast<uint8>((bl[w + 3] + tl[3] - bl[3] - tl[w + 3]) * ooa); + int i; + for (i = 0; i < count; ++i) { + dst[0] = (uint8)((bl[w + 0] + tl[0] - bl[0] - tl[w + 0]) * ooa); + dst[1] = (uint8)((bl[w + 1] + tl[1] - bl[1] - tl[w + 1]) * ooa); + dst[2] = (uint8)((bl[w + 2] + tl[2] - bl[2] - tl[w + 2]) * ooa); + dst[3] = (uint8)((bl[w + 3] + tl[3] - bl[3] - tl[w + 3]) * ooa); dst += 4; tl += 4; bl += 4; } } -#define REPEAT8(v) (v) | ((v) << 8) -#define SHADE(f, v) v * f >> 24 - -void ARGBShadeRow_C(const uint8* src_argb, uint8* dst_argb, int width, - uint32 value) { - const uint32 b_scale = REPEAT8(value & 0xff); - const uint32 g_scale = REPEAT8((value >> 8) & 0xff); - const uint32 r_scale = REPEAT8((value >> 16) & 0xff); - const uint32 a_scale = REPEAT8(value >> 24); - - for (int i = 0; i < width; ++i) { - const uint32 b = REPEAT8(src_argb[0]); - const uint32 g = REPEAT8(src_argb[1]); - const uint32 r = REPEAT8(src_argb[2]); - const uint32 a = REPEAT8(src_argb[3]); - dst_argb[0] = SHADE(b, b_scale); - dst_argb[1] = SHADE(g, g_scale); - dst_argb[2] = SHADE(r, r_scale); - dst_argb[3] = SHADE(a, a_scale); - src_argb += 4; - dst_argb += 4; - } -} -#undef REPEAT8 -#undef SHADE - // Copy pixels from rotated source to destination row with a slope. LIBYUV_API void ARGBAffineRow_C(const uint8* src_argb, int src_argb_stride, uint8* dst_argb, const float* uv_dudv, int width) { + int i; // Render a row of pixels from source into a buffer. float uv[2]; uv[0] = uv_dudv[0]; uv[1] = uv_dudv[1]; - for (int i = 0; i < width; ++i) { - int x = static_cast<int>(uv[0]); - int y = static_cast<int>(uv[1]); - *reinterpret_cast<uint32*>(dst_argb) = - *reinterpret_cast<const uint32*>(src_argb + y * src_argb_stride + + for (i = 0; i < width; ++i) { + int x = (int)(uv[0]); + int y = (int)(uv[1]); + *(uint32*)(dst_argb) = + *(const uint32*)(src_argb + y * src_argb_stride + x * 4); dst_argb += 4; uv[0] += uv_dudv[2]; @@ -1217,29 +2146,481 @@ void ARGBAffineRow_C(const uint8* src_argb, int src_argb_stride, } } +// Blend 2 rows into 1. +static void HalfRow_C(const uint8* src_uv, ptrdiff_t src_uv_stride, + uint8* dst_uv, int width) { + int x; + for (x = 0; x < width; ++x) { + dst_uv[x] = (src_uv[x] + src_uv[src_uv_stride + x] + 1) >> 1; + } +} + +static void HalfRow_16_C(const uint16* src_uv, ptrdiff_t src_uv_stride, + uint16* dst_uv, int width) { + int x; + for (x = 0; x < width; ++x) { + dst_uv[x] = (src_uv[x] + src_uv[src_uv_stride + x] + 1) >> 1; + } +} + // C version 2x2 -> 2x1. -void ARGBInterpolateRow_C(uint8* dst_ptr, const uint8* src_ptr, - ptrdiff_t src_stride, - int dst_width, int source_y_fraction) { - int y1_fraction = source_y_fraction; +void InterpolateRow_C(uint8* dst_ptr, const uint8* src_ptr, + ptrdiff_t src_stride, + int width, int source_y_fraction) { + int y1_fraction = source_y_fraction ; int y0_fraction = 256 - y1_fraction; const uint8* src_ptr1 = src_ptr + src_stride; - uint8* end = dst_ptr + (dst_width << 2); - do { + int x; + if (y1_fraction == 0) { + memcpy(dst_ptr, src_ptr, width); + return; + } + if (y1_fraction == 128) { + HalfRow_C(src_ptr, src_stride, dst_ptr, width); + return; + } + for (x = 0; x < width - 1; x += 2) { + dst_ptr[0] = + (src_ptr[0] * y0_fraction + src_ptr1[0] * y1_fraction + 128) >> 8; + dst_ptr[1] = + (src_ptr[1] * y0_fraction + src_ptr1[1] * y1_fraction + 128) >> 8; + src_ptr += 2; + src_ptr1 += 2; + dst_ptr += 2; + } + if (width & 1) { + dst_ptr[0] = + (src_ptr[0] * y0_fraction + src_ptr1[0] * y1_fraction + 128) >> 8; + } +} + +void InterpolateRow_16_C(uint16* dst_ptr, const uint16* src_ptr, + ptrdiff_t src_stride, + int width, int source_y_fraction) { + int y1_fraction = source_y_fraction; + int y0_fraction = 256 - y1_fraction; + const uint16* src_ptr1 = src_ptr + src_stride; + int x; + if (source_y_fraction == 0) { + memcpy(dst_ptr, src_ptr, width * 2); + return; + } + if (source_y_fraction == 128) { + HalfRow_16_C(src_ptr, src_stride, dst_ptr, width); + return; + } + for (x = 0; x < width - 1; x += 2) { dst_ptr[0] = (src_ptr[0] * y0_fraction + src_ptr1[0] * y1_fraction) >> 8; dst_ptr[1] = (src_ptr[1] * y0_fraction + src_ptr1[1] * y1_fraction) >> 8; - dst_ptr[2] = (src_ptr[2] * y0_fraction + src_ptr1[2] * y1_fraction) >> 8; - dst_ptr[3] = (src_ptr[3] * y0_fraction + src_ptr1[3] * y1_fraction) >> 8; - dst_ptr[4] = (src_ptr[4] * y0_fraction + src_ptr1[4] * y1_fraction) >> 8; - dst_ptr[5] = (src_ptr[5] * y0_fraction + src_ptr1[5] * y1_fraction) >> 8; - dst_ptr[6] = (src_ptr[6] * y0_fraction + src_ptr1[6] * y1_fraction) >> 8; - dst_ptr[7] = (src_ptr[7] * y0_fraction + src_ptr1[7] * y1_fraction) >> 8; - src_ptr += 8; - src_ptr1 += 8; - dst_ptr += 8; - } while (dst_ptr < end); + src_ptr += 2; + src_ptr1 += 2; + dst_ptr += 2; + } + if (width & 1) { + dst_ptr[0] = (src_ptr[0] * y0_fraction + src_ptr1[0] * y1_fraction) >> 8; + } +} + +// Use first 4 shuffler values to reorder ARGB channels. +void ARGBShuffleRow_C(const uint8* src_argb, uint8* dst_argb, + const uint8* shuffler, int width) { + int index0 = shuffler[0]; + int index1 = shuffler[1]; + int index2 = shuffler[2]; + int index3 = shuffler[3]; + // Shuffle a row of ARGB. + int x; + for (x = 0; x < width; ++x) { + // To support in-place conversion. + uint8 b = src_argb[index0]; + uint8 g = src_argb[index1]; + uint8 r = src_argb[index2]; + uint8 a = src_argb[index3]; + dst_argb[0] = b; + dst_argb[1] = g; + dst_argb[2] = r; + dst_argb[3] = a; + src_argb += 4; + dst_argb += 4; + } +} + +void I422ToYUY2Row_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_frame, int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + dst_frame[0] = src_y[0]; + dst_frame[1] = src_u[0]; + dst_frame[2] = src_y[1]; + dst_frame[3] = src_v[0]; + dst_frame += 4; + src_y += 2; + src_u += 1; + src_v += 1; + } + if (width & 1) { + dst_frame[0] = src_y[0]; + dst_frame[1] = src_u[0]; + dst_frame[2] = 0; + dst_frame[3] = src_v[0]; + } } +void I422ToUYVYRow_C(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_frame, int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + dst_frame[0] = src_u[0]; + dst_frame[1] = src_y[0]; + dst_frame[2] = src_v[0]; + dst_frame[3] = src_y[1]; + dst_frame += 4; + src_y += 2; + src_u += 1; + src_v += 1; + } + if (width & 1) { + dst_frame[0] = src_u[0]; + dst_frame[1] = src_y[0]; + dst_frame[2] = src_v[0]; + dst_frame[3] = 0; + } +} + + +void ARGBPolynomialRow_C(const uint8* src_argb, + uint8* dst_argb, + const float* poly, + int width) { + int i; + for (i = 0; i < width; ++i) { + float b = (float)(src_argb[0]); + float g = (float)(src_argb[1]); + float r = (float)(src_argb[2]); + float a = (float)(src_argb[3]); + float b2 = b * b; + float g2 = g * g; + float r2 = r * r; + float a2 = a * a; + float db = poly[0] + poly[4] * b; + float dg = poly[1] + poly[5] * g; + float dr = poly[2] + poly[6] * r; + float da = poly[3] + poly[7] * a; + float b3 = b2 * b; + float g3 = g2 * g; + float r3 = r2 * r; + float a3 = a2 * a; + db += poly[8] * b2; + dg += poly[9] * g2; + dr += poly[10] * r2; + da += poly[11] * a2; + db += poly[12] * b3; + dg += poly[13] * g3; + dr += poly[14] * r3; + da += poly[15] * a3; + + dst_argb[0] = Clamp((int32)(db)); + dst_argb[1] = Clamp((int32)(dg)); + dst_argb[2] = Clamp((int32)(dr)); + dst_argb[3] = Clamp((int32)(da)); + src_argb += 4; + dst_argb += 4; + } +} + +void ARGBLumaColorTableRow_C(const uint8* src_argb, uint8* dst_argb, int width, + const uint8* luma, uint32 lumacoeff) { + uint32 bc = lumacoeff & 0xff; + uint32 gc = (lumacoeff >> 8) & 0xff; + uint32 rc = (lumacoeff >> 16) & 0xff; + + int i; + for (i = 0; i < width - 1; i += 2) { + // Luminance in rows, color values in columns. + const uint8* luma0 = ((src_argb[0] * bc + src_argb[1] * gc + + src_argb[2] * rc) & 0x7F00u) + luma; + const uint8* luma1; + dst_argb[0] = luma0[src_argb[0]]; + dst_argb[1] = luma0[src_argb[1]]; + dst_argb[2] = luma0[src_argb[2]]; + dst_argb[3] = src_argb[3]; + luma1 = ((src_argb[4] * bc + src_argb[5] * gc + + src_argb[6] * rc) & 0x7F00u) + luma; + dst_argb[4] = luma1[src_argb[4]]; + dst_argb[5] = luma1[src_argb[5]]; + dst_argb[6] = luma1[src_argb[6]]; + dst_argb[7] = src_argb[7]; + src_argb += 8; + dst_argb += 8; + } + if (width & 1) { + // Luminance in rows, color values in columns. + const uint8* luma0 = ((src_argb[0] * bc + src_argb[1] * gc + + src_argb[2] * rc) & 0x7F00u) + luma; + dst_argb[0] = luma0[src_argb[0]]; + dst_argb[1] = luma0[src_argb[1]]; + dst_argb[2] = luma0[src_argb[2]]; + dst_argb[3] = src_argb[3]; + } +} + +void ARGBCopyAlphaRow_C(const uint8* src, uint8* dst, int width) { + int i; + for (i = 0; i < width - 1; i += 2) { + dst[3] = src[3]; + dst[7] = src[7]; + dst += 8; + src += 8; + } + if (width & 1) { + dst[3] = src[3]; + } +} + +void ARGBExtractAlphaRow_C(const uint8* src_argb, uint8* dst_a, int width) { + int i; + for (i = 0; i < width - 1; i += 2) { + dst_a[0] = src_argb[3]; + dst_a[1] = src_argb[7]; + dst_a += 2; + src_argb += 8; + } + if (width & 1) { + dst_a[0] = src_argb[3]; + } +} + +void ARGBCopyYToAlphaRow_C(const uint8* src, uint8* dst, int width) { + int i; + for (i = 0; i < width - 1; i += 2) { + dst[3] = src[0]; + dst[7] = src[1]; + dst += 8; + src += 2; + } + if (width & 1) { + dst[3] = src[0]; + } +} + +// Maximum temporary width for wrappers to process at a time, in pixels. +#define MAXTWIDTH 2048 + +#if !(defined(_MSC_VER) && defined(_M_IX86)) && \ + defined(HAS_I422TORGB565ROW_SSSE3) +// row_win.cc has asm version, but GCC uses 2 step wrapper. +void I422ToRGB565Row_SSSE3(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_rgb565, + const struct YuvConstants* yuvconstants, + int width) { + SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, yuvconstants, twidth); + ARGBToRGB565Row_SSE2(row, dst_rgb565, twidth); + src_y += twidth; + src_u += twidth / 2; + src_v += twidth / 2; + dst_rgb565 += twidth * 2; + width -= twidth; + } +} +#endif + +#if defined(HAS_I422TOARGB1555ROW_SSSE3) +void I422ToARGB1555Row_SSSE3(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb1555, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, yuvconstants, twidth); + ARGBToARGB1555Row_SSE2(row, dst_argb1555, twidth); + src_y += twidth; + src_u += twidth / 2; + src_v += twidth / 2; + dst_argb1555 += twidth * 2; + width -= twidth; + } +} +#endif + +#if defined(HAS_I422TOARGB4444ROW_SSSE3) +void I422ToARGB4444Row_SSSE3(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb4444, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, yuvconstants, twidth); + ARGBToARGB4444Row_SSE2(row, dst_argb4444, twidth); + src_y += twidth; + src_u += twidth / 2; + src_v += twidth / 2; + dst_argb4444 += twidth * 2; + width -= twidth; + } +} +#endif + +#if defined(HAS_NV12TORGB565ROW_SSSE3) +void NV12ToRGB565Row_SSSE3(const uint8* src_y, + const uint8* src_uv, + uint8* dst_rgb565, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + NV12ToARGBRow_SSSE3(src_y, src_uv, row, yuvconstants, twidth); + ARGBToRGB565Row_SSE2(row, dst_rgb565, twidth); + src_y += twidth; + src_uv += twidth; + dst_rgb565 += twidth * 2; + width -= twidth; + } +} +#endif + +#if defined(HAS_I422TORGB565ROW_AVX2) +void I422ToRGB565Row_AVX2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_rgb565, + const struct YuvConstants* yuvconstants, + int width) { + SIMD_ALIGNED32(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + I422ToARGBRow_AVX2(src_y, src_u, src_v, row, yuvconstants, twidth); +#if defined(HAS_ARGBTORGB565ROW_AVX2) + ARGBToRGB565Row_AVX2(row, dst_rgb565, twidth); +#else + ARGBToRGB565Row_SSE2(row, dst_rgb565, twidth); +#endif + src_y += twidth; + src_u += twidth / 2; + src_v += twidth / 2; + dst_rgb565 += twidth * 2; + width -= twidth; + } +} +#endif + +#if defined(HAS_I422TOARGB1555ROW_AVX2) +void I422ToARGB1555Row_AVX2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb1555, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED32(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + I422ToARGBRow_AVX2(src_y, src_u, src_v, row, yuvconstants, twidth); +#if defined(HAS_ARGBTOARGB1555ROW_AVX2) + ARGBToARGB1555Row_AVX2(row, dst_argb1555, twidth); +#else + ARGBToARGB1555Row_SSE2(row, dst_argb1555, twidth); +#endif + src_y += twidth; + src_u += twidth / 2; + src_v += twidth / 2; + dst_argb1555 += twidth * 2; + width -= twidth; + } +} +#endif + +#if defined(HAS_I422TOARGB4444ROW_AVX2) +void I422ToARGB4444Row_AVX2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_argb4444, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED32(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + I422ToARGBRow_AVX2(src_y, src_u, src_v, row, yuvconstants, twidth); +#if defined(HAS_ARGBTOARGB4444ROW_AVX2) + ARGBToARGB4444Row_AVX2(row, dst_argb4444, twidth); +#else + ARGBToARGB4444Row_SSE2(row, dst_argb4444, twidth); +#endif + src_y += twidth; + src_u += twidth / 2; + src_v += twidth / 2; + dst_argb4444 += twidth * 2; + width -= twidth; + } +} +#endif + +#if defined(HAS_I422TORGB24ROW_AVX2) +void I422ToRGB24Row_AVX2(const uint8* src_y, + const uint8* src_u, + const uint8* src_v, + uint8* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED32(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + I422ToARGBRow_AVX2(src_y, src_u, src_v, row, yuvconstants, twidth); + // TODO(fbarchard): ARGBToRGB24Row_AVX2 + ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth); + src_y += twidth; + src_u += twidth / 2; + src_v += twidth / 2; + dst_rgb24 += twidth * 3; + width -= twidth; + } +} +#endif + +#if defined(HAS_NV12TORGB565ROW_AVX2) +void NV12ToRGB565Row_AVX2(const uint8* src_y, + const uint8* src_uv, + uint8* dst_rgb565, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED32(uint8 row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + NV12ToARGBRow_AVX2(src_y, src_uv, row, yuvconstants, twidth); +#if defined(HAS_ARGBTORGB565ROW_AVX2) + ARGBToRGB565Row_AVX2(row, dst_rgb565, twidth); +#else + ARGBToRGB565Row_SSE2(row, dst_rgb565, twidth); +#endif + src_y += twidth; + src_uv += twidth; + dst_rgb565 += twidth * 2; + width -= twidth; + } +} +#endif + #ifdef __cplusplus } // extern "C" } // namespace libyuv |