diff options
| author | Allan Sandfeld Jensen <allan.jensen@theqtcompany.com> | 2016-05-09 14:22:11 +0200 |
|---|---|---|
| committer | Allan Sandfeld Jensen <allan.jensen@qt.io> | 2016-05-09 15:11:45 +0000 |
| commit | 2ddb2d3e14eef3de7dbd0cef553d669b9ac2361c (patch) | |
| tree | e75f511546c5fd1a173e87c1f9fb11d7ac8d1af3 /chromium/third_party/libyuv/source | |
| parent | a4f3d46271c57e8155ba912df46a05559d14726e (diff) | |
| download | qtwebengine-chromium-2ddb2d3e14eef3de7dbd0cef553d669b9ac2361c.tar.gz | |
BASELINE: Update Chromium to 51.0.2704.41
Also adds in all smaller components by reversing logic for exclusion.
Change-Id: Ibf90b506e7da088ea2f65dcf23f2b0992c504422
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
Diffstat (limited to 'chromium/third_party/libyuv/source')
21 files changed, 491 insertions, 574 deletions
diff --git a/chromium/third_party/libyuv/source/convert.cc b/chromium/third_party/libyuv/source/convert.cc index 5dc279f8a67..e332bc505c4 100644 --- a/chromium/third_party/libyuv/source/convert.cc +++ b/chromium/third_party/libyuv/source/convert.cc @@ -303,14 +303,14 @@ static int X420ToI420(const uint8* src_y, } } #endif -#if defined(HAS_SPLITUVROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_SPLITUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_uv, 4) && IS_ALIGNED(src_stride_uv, 4) && IS_ALIGNED(dst_u, 4) && IS_ALIGNED(dst_stride_u, 4) && IS_ALIGNED(dst_v, 4) && IS_ALIGNED(dst_stride_v, 4)) { - SplitUVRow = SplitUVRow_Any_MIPS_DSPR2; + SplitUVRow = SplitUVRow_Any_DSPR2; if (IS_ALIGNED(halfwidth, 16)) { - SplitUVRow = SplitUVRow_MIPS_DSPR2; + SplitUVRow = SplitUVRow_DSPR2; } } #endif diff --git a/chromium/third_party/libyuv/source/convert_argb.cc b/chromium/third_party/libyuv/source/convert_argb.cc index cf3d7228905..e586f7043ce 100644 --- a/chromium/third_party/libyuv/source/convert_argb.cc +++ b/chromium/third_party/libyuv/source/convert_argb.cc @@ -92,13 +92,13 @@ static int I420ToARGBMatrix(const uint8* src_y, int src_stride_y, } } #endif -#if defined(HAS_I422TOARGBROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 4) && +#if defined(HAS_I422TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 4) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) && IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) { - I422ToARGBRow = I422ToARGBRow_MIPS_DSPR2; + I422ToARGBRow = I422ToARGBRow_DSPR2; } #endif @@ -262,13 +262,13 @@ static int I422ToARGBMatrix(const uint8* src_y, int src_stride_y, } } #endif -#if defined(HAS_I422TOARGBROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 4) && +#if defined(HAS_I422TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 4) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) && IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) { - I422ToARGBRow = I422ToARGBRow_MIPS_DSPR2; + I422ToARGBRow = I422ToARGBRow_DSPR2; } #endif @@ -607,13 +607,13 @@ static int I420AlphaToARGBMatrix(const uint8* src_y, int src_stride_y, } } #endif -#if defined(HAS_I422ALPHATOARGBROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 4) && +#if defined(HAS_I422ALPHATOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 4) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) && IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) { - I422AlphaToARGBRow = I422AlphaToARGBRow_MIPS_DSPR2; + I422AlphaToARGBRow = I422AlphaToARGBRow_DSPR2; } #endif #if defined(HAS_ARGBATTENUATEROW_SSSE3) diff --git a/chromium/third_party/libyuv/source/convert_from.cc b/chromium/third_party/libyuv/source/convert_from.cc index 9c138d936da..3bc9eb1be45 100644 --- a/chromium/third_party/libyuv/source/convert_from.cc +++ b/chromium/third_party/libyuv/source/convert_from.cc @@ -445,7 +445,7 @@ int I420ToNV21(const uint8* src_y, int src_stride_y, return I420ToNV12(src_y, src_stride_y, src_v, src_stride_v, src_u, src_stride_u, - dst_y, src_stride_y, + dst_y, dst_stride_y, dst_vu, dst_stride_vu, width, height); } @@ -498,13 +498,13 @@ static int I420ToRGBAMatrix(const uint8* src_y, int src_stride_y, } } #endif -#if defined(HAS_I422TORGBAROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 4) && +#if defined(HAS_I422TORGBAROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 4) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) && IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) && IS_ALIGNED(dst_rgba, 4) && IS_ALIGNED(dst_stride_rgba, 4)) { - I422ToRGBARow = I422ToRGBARow_MIPS_DSPR2; + I422ToRGBARow = I422ToRGBARow_DSPR2; } #endif @@ -888,12 +888,12 @@ int I420ToRGB565Dither(const uint8* src_y, int src_stride_y, } } #endif -#if defined(HAS_I422TOARGBROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 4) && +#if defined(HAS_I422TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 4) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) && IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2)) { - I422ToARGBRow = I422ToARGBRow_MIPS_DSPR2; + I422ToARGBRow = I422ToARGBRow_DSPR2; } #endif #if defined(HAS_ARGBTORGB565DITHERROW_SSE2) diff --git a/chromium/third_party/libyuv/source/convert_from_argb.cc b/chromium/third_party/libyuv/source/convert_from_argb.cc index 6796343c0fb..2a8682b7eb4 100644 --- a/chromium/third_party/libyuv/source/convert_from_argb.cc +++ b/chromium/third_party/libyuv/source/convert_from_argb.cc @@ -109,13 +109,16 @@ int ARGBToI422(const uint8* src_argb, int src_stride_argb, uint8* dst_v, int dst_stride_v, int width, int height) { int y; - void (*ARGBToUV422Row)(const uint8* src_argb, uint8* dst_u, uint8* dst_v, - int width) = ARGBToUV422Row_C; + void (*ARGBToUVRow)(const uint8* src_argb0, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C; void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int width) = ARGBToYRow_C; - if (!src_argb || !dst_y || !dst_u || !dst_v || width <= 0 || height == 0) { + if (!src_argb || + !dst_y || !dst_u || !dst_v || + width <= 0 || height == 0) { return -1; } + // Negative height means invert the image. if (height < 0) { height = -height; src_argb = src_argb + (height - 1) * src_stride_argb; @@ -130,34 +133,22 @@ int ARGBToI422(const uint8* src_argb, int src_stride_argb, height = 1; src_stride_argb = dst_stride_y = dst_stride_u = dst_stride_v = 0; } -#if defined(HAS_ARGBTOUV422ROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3)) { - ARGBToUV422Row = ARGBToUV422Row_Any_SSSE3; - if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_SSSE3; - } - } -#endif -#if defined(HAS_ARGBTOUV422ROW_NEON) - if (TestCpuFlag(kCpuHasNEON)) { - ARGBToUV422Row = ARGBToUV422Row_Any_NEON; - if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_NEON; - } - } -#endif -#if defined(HAS_ARGBTOYROW_SSSE3) +#if defined(HAS_ARGBTOYROW_SSSE3) && defined(HAS_ARGBTOUVROW_SSSE3) if (TestCpuFlag(kCpuHasSSSE3)) { + ARGBToUVRow = ARGBToUVRow_Any_SSSE3; ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_SSSE3; ARGBToYRow = ARGBToYRow_SSSE3; } } #endif -#if defined(HAS_ARGBTOYROW_AVX2) +#if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { + ARGBToUVRow = ARGBToUVRow_Any_AVX2; ARGBToYRow = ARGBToYRow_Any_AVX2; if (IS_ALIGNED(width, 32)) { + ARGBToUVRow = ARGBToUVRow_AVX2; ARGBToYRow = ARGBToYRow_AVX2; } } @@ -170,9 +161,17 @@ int ARGBToI422(const uint8* src_argb, int src_stride_argb, } } #endif +#if defined(HAS_ARGBTOUVROW_NEON) + if (TestCpuFlag(kCpuHasNEON)) { + ARGBToUVRow = ARGBToUVRow_Any_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } + } +#endif for (y = 0; y < height; ++y) { - ARGBToUV422Row(src_argb, dst_u, dst_v, width); + ARGBToUVRow(src_argb, 0, dst_u, dst_v, width); ARGBToYRow(src_argb, dst_y, width); src_argb += src_stride_argb; dst_y += dst_stride_y; @@ -478,8 +477,8 @@ int ARGBToYUY2(const uint8* src_argb, int src_stride_argb, uint8* dst_yuy2, int dst_stride_yuy2, int width, int height) { int y; - void (*ARGBToUV422Row)(const uint8* src_argb, uint8* dst_u, uint8* dst_v, - int width) = ARGBToUV422Row_C; + void (*ARGBToUVRow)(const uint8* src_argb, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C; void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int width) = ARGBToYRow_C; void (*I422ToYUY2Row)(const uint8* src_y, const uint8* src_u, @@ -502,34 +501,22 @@ int ARGBToYUY2(const uint8* src_argb, int src_stride_argb, height = 1; src_stride_argb = dst_stride_yuy2 = 0; } -#if defined(HAS_ARGBTOUV422ROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3)) { - ARGBToUV422Row = ARGBToUV422Row_Any_SSSE3; - if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_SSSE3; - } - } -#endif -#if defined(HAS_ARGBTOUV422ROW_NEON) - if (TestCpuFlag(kCpuHasNEON)) { - ARGBToUV422Row = ARGBToUV422Row_Any_NEON; - if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_NEON; - } - } -#endif -#if defined(HAS_ARGBTOYROW_SSSE3) +#if defined(HAS_ARGBTOYROW_SSSE3) && defined(HAS_ARGBTOUVROW_SSSE3) if (TestCpuFlag(kCpuHasSSSE3)) { + ARGBToUVRow = ARGBToUVRow_Any_SSSE3; ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_SSSE3; ARGBToYRow = ARGBToYRow_SSSE3; } } #endif -#if defined(HAS_ARGBTOYROW_AVX2) +#if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { + ARGBToUVRow = ARGBToUVRow_Any_AVX2; ARGBToYRow = ARGBToYRow_Any_AVX2; if (IS_ALIGNED(width, 32)) { + ARGBToUVRow = ARGBToUVRow_AVX2; ARGBToYRow = ARGBToYRow_AVX2; } } @@ -542,7 +529,14 @@ int ARGBToYUY2(const uint8* src_argb, int src_stride_argb, } } #endif - +#if defined(HAS_ARGBTOUVROW_NEON) + if (TestCpuFlag(kCpuHasNEON)) { + ARGBToUVRow = ARGBToUVRow_Any_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } + } +#endif #if defined(HAS_I422TOYUY2ROW_SSE2) if (TestCpuFlag(kCpuHasSSE2)) { I422ToYUY2Row = I422ToYUY2Row_Any_SSE2; @@ -567,7 +561,7 @@ int ARGBToYUY2(const uint8* src_argb, int src_stride_argb, uint8* row_v = row_u + ((width + 63) & ~63) / 2; for (y = 0; y < height; ++y) { - ARGBToUV422Row(src_argb, row_u, row_v, width); + ARGBToUVRow(src_argb, 0, row_u, row_v, width); ARGBToYRow(src_argb, row_y, width); I422ToYUY2Row(row_y, row_u, row_v, dst_yuy2, width); src_argb += src_stride_argb; @@ -585,8 +579,8 @@ int ARGBToUYVY(const uint8* src_argb, int src_stride_argb, uint8* dst_uyvy, int dst_stride_uyvy, int width, int height) { int y; - void (*ARGBToUV422Row)(const uint8* src_argb, uint8* dst_u, uint8* dst_v, - int width) = ARGBToUV422Row_C; + void (*ARGBToUVRow)(const uint8* src_argb, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C; void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int width) = ARGBToYRow_C; void (*I422ToUYVYRow)(const uint8* src_y, const uint8* src_u, @@ -609,34 +603,22 @@ int ARGBToUYVY(const uint8* src_argb, int src_stride_argb, height = 1; src_stride_argb = dst_stride_uyvy = 0; } -#if defined(HAS_ARGBTOUV422ROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3)) { - ARGBToUV422Row = ARGBToUV422Row_Any_SSSE3; - if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_SSSE3; - } - } -#endif -#if defined(HAS_ARGBTOUV422ROW_NEON) - if (TestCpuFlag(kCpuHasNEON)) { - ARGBToUV422Row = ARGBToUV422Row_Any_NEON; - if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_NEON; - } - } -#endif -#if defined(HAS_ARGBTOYROW_SSSE3) +#if defined(HAS_ARGBTOYROW_SSSE3) && defined(HAS_ARGBTOUVROW_SSSE3) if (TestCpuFlag(kCpuHasSSSE3)) { + ARGBToUVRow = ARGBToUVRow_Any_SSSE3; ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_SSSE3; ARGBToYRow = ARGBToYRow_SSSE3; } } #endif -#if defined(HAS_ARGBTOYROW_AVX2) +#if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { + ARGBToUVRow = ARGBToUVRow_Any_AVX2; ARGBToYRow = ARGBToYRow_Any_AVX2; if (IS_ALIGNED(width, 32)) { + ARGBToUVRow = ARGBToUVRow_AVX2; ARGBToYRow = ARGBToYRow_AVX2; } } @@ -649,7 +631,14 @@ int ARGBToUYVY(const uint8* src_argb, int src_stride_argb, } } #endif - +#if defined(HAS_ARGBTOUVROW_NEON) + if (TestCpuFlag(kCpuHasNEON)) { + ARGBToUVRow = ARGBToUVRow_Any_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVRow = ARGBToUVRow_NEON; + } + } +#endif #if defined(HAS_I422TOUYVYROW_SSE2) if (TestCpuFlag(kCpuHasSSE2)) { I422ToUYVYRow = I422ToUYVYRow_Any_SSE2; @@ -674,7 +663,7 @@ int ARGBToUYVY(const uint8* src_argb, int src_stride_argb, uint8* row_v = row_u + ((width + 63) & ~63) / 2; for (y = 0; y < height; ++y) { - ARGBToUV422Row(src_argb, row_u, row_v, width); + ARGBToUVRow(src_argb, 0, row_u, row_v, width); ARGBToYRow(src_argb, row_y, width); I422ToUYVYRow(row_y, row_u, row_v, dst_uyvy, width); src_argb += src_stride_argb; @@ -1157,21 +1146,24 @@ int ARGBToJ420(const uint8* src_argb, int src_stride_argb, return 0; } -// ARGB little endian (bgra in memory) to J422 +// Convert ARGB to J422. (JPeg full range I422). LIBYUV_API int ARGBToJ422(const uint8* src_argb, int src_stride_argb, - uint8* dst_y, int dst_stride_y, + uint8* dst_yj, int dst_stride_yj, uint8* dst_u, int dst_stride_u, uint8* dst_v, int dst_stride_v, int width, int height) { int y; - void (*ARGBToUVJ422Row)(const uint8* src_argb, uint8* dst_u, uint8* dst_v, - int width) = ARGBToUVJ422Row_C; - void (*ARGBToYJRow)(const uint8* src_argb, uint8* dst_y, int width) = + void (*ARGBToUVJRow)(const uint8* src_argb0, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int width) = ARGBToUVJRow_C; + void (*ARGBToYJRow)(const uint8* src_argb, uint8* dst_yj, int width) = ARGBToYJRow_C; - if (!src_argb || !dst_y || !dst_u || !dst_v || width <= 0 || height == 0) { + if (!src_argb || + !dst_yj || !dst_u || !dst_v || + width <= 0 || height == 0) { return -1; } + // Negative height means invert the image. if (height < 0) { height = -height; src_argb = src_argb + (height - 1) * src_stride_argb; @@ -1179,34 +1171,19 @@ int ARGBToJ422(const uint8* src_argb, int src_stride_argb, } // Coalesce rows. if (src_stride_argb == width * 4 && - dst_stride_y == width && + dst_stride_yj == width && dst_stride_u * 2 == width && dst_stride_v * 2 == width) { width *= height; height = 1; - src_stride_argb = dst_stride_y = dst_stride_u = dst_stride_v = 0; - } -#if defined(HAS_ARGBTOUVJ422ROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3)) { - ARGBToUVJ422Row = ARGBToUVJ422Row_Any_SSSE3; - if (IS_ALIGNED(width, 16)) { - ARGBToUVJ422Row = ARGBToUVJ422Row_SSSE3; - } + src_stride_argb = dst_stride_yj = dst_stride_u = dst_stride_v = 0; } -#endif -#if defined(HAS_ARGBTOUVJ422ROW_NEON) - if (TestCpuFlag(kCpuHasNEON)) { - ARGBToUVJ422Row = ARGBToUVJ422Row_Any_NEON; - if (IS_ALIGNED(width, 16)) { - ARGBToUVJ422Row = ARGBToUVJ422Row_NEON; - } - } -#endif - -#if defined(HAS_ARGBTOYJROW_SSSE3) +#if defined(HAS_ARGBTOYJROW_SSSE3) && defined(HAS_ARGBTOUVJROW_SSSE3) if (TestCpuFlag(kCpuHasSSSE3)) { + ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3; ARGBToYJRow = ARGBToYJRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { + ARGBToUVJRow = ARGBToUVJRow_SSSE3; ARGBToYJRow = ARGBToYJRow_SSSE3; } } @@ -1227,12 +1204,20 @@ int ARGBToJ422(const uint8* src_argb, int src_stride_argb, } } #endif +#if defined(HAS_ARGBTOUVJROW_NEON) + if (TestCpuFlag(kCpuHasNEON)) { + ARGBToUVJRow = ARGBToUVJRow_Any_NEON; + if (IS_ALIGNED(width, 16)) { + ARGBToUVJRow = ARGBToUVJRow_NEON; + } + } +#endif for (y = 0; y < height; ++y) { - ARGBToUVJ422Row(src_argb, dst_u, dst_v, width); - ARGBToYJRow(src_argb, dst_y, width); + ARGBToUVJRow(src_argb, 0, dst_u, dst_v, width); + ARGBToYJRow(src_argb, dst_yj, width); src_argb += src_stride_argb; - dst_y += dst_stride_y; + dst_yj += dst_stride_yj; dst_u += dst_stride_u; dst_v += dst_stride_v; } diff --git a/chromium/third_party/libyuv/source/cpu_id.cc b/chromium/third_party/libyuv/source/cpu_id.cc index ff7bdbd92d0..d64d9d56572 100644 --- a/chromium/third_party/libyuv/source/cpu_id.cc +++ b/chromium/third_party/libyuv/source/cpu_id.cc @@ -10,12 +10,12 @@ #include "libyuv/cpu_id.h" -#if defined(_MSC_VER) && !defined(__clang__) +#if defined(_MSC_VER) #include <intrin.h> // For __cpuidex() #endif #if !defined(__pnacl__) && !defined(__CLR_VER) && \ !defined(__native_client__) && (defined(_M_IX86) || defined(_M_X64)) && \ - defined(_MSC_VER) && !defined(__clang__) && (_MSC_FULL_VER >= 160040219) + defined(_MSC_VER) && (_MSC_FULL_VER >= 160040219) #include <immintrin.h> // For _xgetbv() #endif @@ -48,7 +48,7 @@ extern "C" { !defined(__pnacl__) && !defined(__CLR_VER) LIBYUV_API void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) { -#if defined(_MSC_VER) && !defined(__clang__) +#if defined(_MSC_VER) // Visual C version uses intrinsic or inline x86 assembly. #if (_MSC_FULL_VER >= 160040219) __cpuidex((int*)(cpu_info), info_eax, info_ecx); @@ -71,7 +71,7 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) { } #endif // GCC version uses inline x86 assembly. -#else // defined(_MSC_VER) && !defined(__clang__) +#else // defined(_MSC_VER) uint32 info_ebx, info_edx; asm volatile ( #if defined( __i386__) && defined(__PIC__) @@ -89,7 +89,7 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) { cpu_info[1] = info_ebx; cpu_info[2] = info_ecx; cpu_info[3] = info_edx; -#endif // defined(_MSC_VER) && !defined(__clang__) +#endif // defined(_MSC_VER) } #else // (defined(_M_IX86) || defined(_M_X64) ... LIBYUV_API @@ -251,15 +251,11 @@ int InitCpuFlags(void) { #endif #if defined(__mips__) && defined(__linux__) #if defined(__mips_dspr2) - cpu_info |= kCpuHasMIPS_DSPR2; + cpu_info |= kCpuHasDSPR2; #endif cpu_info |= kCpuHasMIPS; - - if (getenv("LIBYUV_DISABLE_MIPS")) { - cpu_info &= ~kCpuHasMIPS; - } - if (getenv("LIBYUV_DISABLE_MIPS_DSPR2")) { - cpu_info &= ~kCpuHasMIPS_DSPR2; + if (getenv("LIBYUV_DISABLE_DSPR2")) { + cpu_info &= ~kCpuHasDSPR2; } #endif #if defined(__arm__) || defined(__aarch64__) diff --git a/chromium/third_party/libyuv/source/planar_functions.cc b/chromium/third_party/libyuv/source/planar_functions.cc index df3041e4956..851c0fea911 100644 --- a/chromium/third_party/libyuv/source/planar_functions.cc +++ b/chromium/third_party/libyuv/source/planar_functions.cc @@ -255,11 +255,11 @@ void MirrorPlane(const uint8* src_y, int src_stride_y, } #endif // TODO(fbarchard): Mirror on mips handle unaligned memory. -#if defined(HAS_MIRRORROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_MIRRORROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(dst_y, 4) && IS_ALIGNED(dst_stride_y, 4)) { - MirrorRow = MirrorRow_MIPS_DSPR2; + MirrorRow = MirrorRow_DSPR2; } #endif @@ -986,13 +986,13 @@ static int I422ToRGBAMatrix(const uint8* src_y, int src_stride_y, } } #endif -#if defined(HAS_I422TORGBAROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 4) && +#if defined(HAS_I422TORGBAROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 4) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) && IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) && IS_ALIGNED(dst_rgba, 4) && IS_ALIGNED(dst_stride_rgba, 4)) { - I422ToRGBARow = I422ToRGBARow_MIPS_DSPR2; + I422ToRGBARow = I422ToRGBARow_DSPR2; } #endif @@ -1906,13 +1906,13 @@ int InterpolatePlane(const uint8* src0, int src_stride0, } } #endif -#if defined(HAS_INTERPOLATEROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_INTERPOLATEROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src0, 4) && IS_ALIGNED(src_stride0, 4) && IS_ALIGNED(src1, 4) && IS_ALIGNED(src_stride1, 4) && IS_ALIGNED(dst, 4) && IS_ALIGNED(dst_stride, 4) && IS_ALIGNED(width, 4)) { - InterpolateRow = InterpolateRow_MIPS_DSPR2; + InterpolateRow = InterpolateRow_DSPR2; } #endif @@ -2424,6 +2424,9 @@ int ARGBCopyYToAlpha(const uint8* src_y, int src_stride_y, return 0; } +// TODO(fbarchard): Consider if width is even Y channel can be split +// directly. A SplitUVRow_Odd function could copy the remaining chroma. + LIBYUV_API int YUY2ToNV12(const uint8* src_yuy2, int src_stride_yuy2, uint8* dst_y, int dst_stride_y, @@ -2498,22 +2501,24 @@ int YUY2ToNV12(const uint8* src_yuy2, int src_stride_yuy2, { int awidth = halfwidth * 2; - // 2 rows of uv - align_buffer_64(rows, awidth * 2); + // row of y and 2 rows of uv + align_buffer_64(rows, awidth * 3); for (y = 0; y < height - 1; y += 2) { // Split Y from UV. - SplitUVRow(src_yuy2, dst_y, rows, awidth); - SplitUVRow(src_yuy2 + src_stride_yuy2, dst_y + dst_stride_y, - rows + awidth, awidth); - InterpolateRow(dst_uv, rows, awidth, awidth, 128); + SplitUVRow(src_yuy2, rows, rows + awidth, awidth); + memcpy(dst_y, rows, width); + SplitUVRow(src_yuy2 + src_stride_yuy2, rows, rows + awidth * 2, awidth); + memcpy(dst_y + dst_stride_y, rows, width); + InterpolateRow(dst_uv, rows + awidth, awidth, awidth, 128); src_yuy2 += src_stride_yuy2 * 2; dst_y += dst_stride_y * 2; dst_uv += dst_stride_uv; } if (height & 1) { // Split Y from UV. - SplitUVRow(src_yuy2, dst_y, dst_uv, awidth); + SplitUVRow(src_yuy2, rows, dst_uv, awidth); + memcpy(dst_y, rows, width); } free_aligned_buffer_64(rows); } @@ -2594,22 +2599,24 @@ int UYVYToNV12(const uint8* src_uyvy, int src_stride_uyvy, { int awidth = halfwidth * 2; - // 2 rows of uv - align_buffer_64(rows, awidth * 2); + // row of y and 2 rows of uv + align_buffer_64(rows, awidth * 3); for (y = 0; y < height - 1; y += 2) { // Split Y from UV. - SplitUVRow(src_uyvy, rows, dst_y, awidth); - SplitUVRow(src_uyvy + src_stride_uyvy, rows + awidth, - dst_y + dst_stride_y, awidth); - InterpolateRow(dst_uv, rows, awidth, awidth, 128); + SplitUVRow(src_uyvy, rows + awidth, rows, awidth); + memcpy(dst_y, rows, width); + SplitUVRow(src_uyvy + src_stride_uyvy, rows + awidth * 2, rows, awidth); + memcpy(dst_y + dst_stride_y, rows, width); + InterpolateRow(dst_uv, rows + awidth, awidth, awidth, 128); src_uyvy += src_stride_uyvy * 2; dst_y += dst_stride_y * 2; dst_uv += dst_stride_uv; } if (height & 1) { // Split Y from UV. - SplitUVRow(src_uyvy, dst_uv, dst_y, awidth); + SplitUVRow(src_uyvy, dst_uv, rows, awidth); + memcpy(dst_y, rows, width); } free_aligned_buffer_64(rows); } diff --git a/chromium/third_party/libyuv/source/rotate.cc b/chromium/third_party/libyuv/source/rotate.cc index 31e04af9c0b..01ea5c40744 100644 --- a/chromium/third_party/libyuv/source/rotate.cc +++ b/chromium/third_party/libyuv/source/rotate.cc @@ -49,13 +49,13 @@ void TransposePlane(const uint8* src, int src_stride, } } #endif -#if defined(HAS_TRANSPOSEWX8_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2)) { +#if defined(HAS_TRANSPOSEWX8_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { if (IS_ALIGNED(width, 4) && IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) { - TransposeWx8 = TransposeWx8_Fast_MIPS_DSPR2; + TransposeWx8 = TransposeWx8_Fast_DSPR2; } else { - TransposeWx8 = TransposeWx8_MIPS_DSPR2; + TransposeWx8 = TransposeWx8_DSPR2; } } #endif @@ -134,11 +134,11 @@ void RotatePlane180(const uint8* src, int src_stride, } #endif // TODO(fbarchard): Mirror on mips handle unaligned memory. -#if defined(HAS_MIRRORROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_MIRRORROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst, 4) && IS_ALIGNED(dst_stride, 4)) { - MirrorRow = MirrorRow_MIPS_DSPR2; + MirrorRow = MirrorRow_DSPR2; } #endif #if defined(HAS_COPYROW_SSE2) @@ -203,10 +203,10 @@ void TransposeUV(const uint8* src, int src_stride, } } #endif -#if defined(HAS_TRANSPOSEUVWX8_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 2) && +#if defined(HAS_TRANSPOSEUVWX8_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 2) && IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) { - TransposeUVWx8 = TransposeUVWx8_MIPS_DSPR2; + TransposeUVWx8 = TransposeUVWx8_DSPR2; } #endif @@ -267,22 +267,22 @@ void RotateUV180(const uint8* src, int src_stride, uint8* dst_b, int dst_stride_b, int width, int height) { int i; - void (*MirrorRowUV)(const uint8* src, uint8* dst_u, uint8* dst_v, int width) = + void (*MirrorUVRow)(const uint8* src, uint8* dst_u, uint8* dst_v, int width) = MirrorUVRow_C; #if defined(HAS_MIRRORUVROW_NEON) if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) { - MirrorRowUV = MirrorUVRow_NEON; + MirrorUVRow = MirrorUVRow_NEON; } #endif -#if defined(HAS_MIRRORROW_UV_SSSE3) +#if defined(HAS_MIRRORUVROW_SSSE3) if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 16)) { - MirrorRowUV = MirrorUVRow_SSSE3; + MirrorUVRow = MirrorUVRow_SSSE3; } #endif -#if defined(HAS_MIRRORUVROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_MIRRORUVROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) { - MirrorRowUV = MirrorUVRow_MIPS_DSPR2; + MirrorUVRow = MirrorUVRow_DSPR2; } #endif @@ -290,7 +290,7 @@ void RotateUV180(const uint8* src, int src_stride, dst_b += dst_stride_b * (height - 1); for (i = 0; i < height; ++i) { - MirrorRowUV(src, dst_a, dst_b, width); + MirrorUVRow(src, dst_a, dst_b, width); src += src_stride; dst_a -= dst_stride_a; dst_b -= dst_stride_b; diff --git a/chromium/third_party/libyuv/source/rotate_any.cc b/chromium/third_party/libyuv/source/rotate_any.cc index d12bad5dc77..31a74c31555 100644 --- a/chromium/third_party/libyuv/source/rotate_any.cc +++ b/chromium/third_party/libyuv/source/rotate_any.cc @@ -38,8 +38,8 @@ TANY(TransposeWx8_Any_SSSE3, TransposeWx8_SSSE3, 7) #ifdef HAS_TRANSPOSEWX8_FAST_SSSE3 TANY(TransposeWx8_Fast_Any_SSSE3, TransposeWx8_Fast_SSSE3, 15) #endif -#ifdef HAS_TRANSPOSEWX8_MIPS_DSPR2 -TANY(TransposeWx8_Any_MIPS_DSPR2, TransposeWx8_MIPS_DSPR2, 7) +#ifdef HAS_TRANSPOSEWX8_DSPR2 +TANY(TransposeWx8_Any_DSPR2, TransposeWx8_DSPR2, 7) #endif #undef TANY @@ -64,8 +64,8 @@ TUVANY(TransposeUVWx8_Any_NEON, TransposeUVWx8_NEON, 7) #ifdef HAS_TRANSPOSEUVWX8_SSE2 TUVANY(TransposeUVWx8_Any_SSE2, TransposeUVWx8_SSE2, 7) #endif -#ifdef HAS_TRANSPOSEUVWX8_MIPS_DSPR2 -TUVANY(TransposeUVWx8_Any_MIPS_DSPR2, TransposeUVWx8_MIPS_DSPR2, 7) +#ifdef HAS_TRANSPOSEUVWX8_DSPR2 +TUVANY(TransposeUVWx8_Any_DSPR2, TransposeUVWx8_DSPR2, 7) #endif #undef TUVANY diff --git a/chromium/third_party/libyuv/source/rotate_mips.cc b/chromium/third_party/libyuv/source/rotate_mips.cc index efe6bd909e1..23e89fbad4c 100644 --- a/chromium/third_party/libyuv/source/rotate_mips.cc +++ b/chromium/third_party/libyuv/source/rotate_mips.cc @@ -22,7 +22,7 @@ extern "C" { defined(__mips_dsp) && (__mips_dsp_rev >= 2) && \ (_MIPS_SIM == _MIPS_SIM_ABI32) -void TransposeWx8_MIPS_DSPR2(const uint8* src, int src_stride, +void TransposeWx8_DSPR2(const uint8* src, int src_stride, uint8* dst, int dst_stride, int width) { __asm__ __volatile__ ( ".set push \n" @@ -106,7 +106,7 @@ void TransposeWx8_MIPS_DSPR2(const uint8* src, int src_stride, ); } -void TransposeWx8_Fast_MIPS_DSPR2(const uint8* src, int src_stride, +void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride, uint8* dst, int dst_stride, int width) { __asm__ __volatile__ ( ".set noat \n" @@ -308,7 +308,7 @@ void TransposeWx8_Fast_MIPS_DSPR2(const uint8* src, int src_stride, ); } -void TransposeUVWx8_MIPS_DSPR2(const uint8* src, int src_stride, +void TransposeUVWx8_DSPR2(const uint8* src, int src_stride, uint8* dst_a, int dst_stride_a, uint8* dst_b, int dst_stride_b, int width) { diff --git a/chromium/third_party/libyuv/source/row_any.cc b/chromium/third_party/libyuv/source/row_any.cc index 5e5f435a6fe..29b7a343d53 100644 --- a/chromium/third_party/libyuv/source/row_any.cc +++ b/chromium/third_party/libyuv/source/row_any.cc @@ -596,8 +596,8 @@ ANY11T(InterpolateRow_Any_SSSE3, InterpolateRow_SSSE3, 1, 1, 15) #ifdef HAS_INTERPOLATEROW_NEON ANY11T(InterpolateRow_Any_NEON, InterpolateRow_NEON, 1, 1, 15) #endif -#ifdef HAS_INTERPOLATEROW_MIPS_DSPR2 -ANY11T(InterpolateRow_Any_MIPS_DSPR2, InterpolateRow_MIPS_DSPR2, 1, 1, 3) +#ifdef HAS_INTERPOLATEROW_DSPR2 +ANY11T(InterpolateRow_Any_DSPR2, InterpolateRow_DSPR2, 1, 1, 3) #endif #undef ANY11T @@ -705,8 +705,8 @@ ANY12(SplitUVRow_Any_AVX2, SplitUVRow_AVX2, 0, 2, 0, 31) #ifdef HAS_SPLITUVROW_NEON ANY12(SplitUVRow_Any_NEON, SplitUVRow_NEON, 0, 2, 0, 15) #endif -#ifdef HAS_SPLITUVROW_MIPS_DSPR2 -ANY12(SplitUVRow_Any_MIPS_DSPR2, SplitUVRow_MIPS_DSPR2, 0, 2, 0, 15) +#ifdef HAS_SPLITUVROW_DSPR2 +ANY12(SplitUVRow_Any_DSPR2, SplitUVRow_DSPR2, 0, 2, 0, 15) #endif #ifdef HAS_ARGBTOUV444ROW_SSSE3 ANY12(ARGBToUV444Row_Any_SSSE3, ARGBToUV444Row_SSSE3, 0, 4, 0, 15) @@ -715,16 +715,12 @@ ANY12(ARGBToUV444Row_Any_SSSE3, ARGBToUV444Row_SSSE3, 0, 4, 0, 15) ANY12(YUY2ToUV422Row_Any_AVX2, YUY2ToUV422Row_AVX2, 1, 4, 1, 31) ANY12(UYVYToUV422Row_Any_AVX2, UYVYToUV422Row_AVX2, 1, 4, 1, 31) #endif -#ifdef HAS_ARGBTOUV422ROW_SSSE3 -ANY12(ARGBToUV422Row_Any_SSSE3, ARGBToUV422Row_SSSE3, 0, 4, 1, 15) -#endif #ifdef HAS_YUY2TOUV422ROW_SSE2 ANY12(YUY2ToUV422Row_Any_SSE2, YUY2ToUV422Row_SSE2, 1, 4, 1, 15) ANY12(UYVYToUV422Row_Any_SSE2, UYVYToUV422Row_SSE2, 1, 4, 1, 15) #endif #ifdef HAS_YUY2TOUV422ROW_NEON ANY12(ARGBToUV444Row_Any_NEON, ARGBToUV444Row_NEON, 0, 4, 0, 7) -ANY12(ARGBToUV422Row_Any_NEON, ARGBToUV422Row_NEON, 0, 4, 1, 15) ANY12(ARGBToUV411Row_Any_NEON, ARGBToUV411Row_NEON, 0, 4, 2, 31) ANY12(YUY2ToUV422Row_Any_NEON, YUY2ToUV422Row_NEON, 1, 4, 1, 15) ANY12(UYVYToUV422Row_Any_NEON, UYVYToUV422Row_NEON, 1, 4, 1, 15) @@ -760,6 +756,9 @@ ANY12(UYVYToUV422Row_Any_NEON, UYVYToUV422Row_NEON, 1, 4, 1, 15) #ifdef HAS_ARGBTOUVROW_AVX2 ANY12S(ARGBToUVRow_Any_AVX2, ARGBToUVRow_AVX2, 0, 4, 31) #endif +#ifdef HAS_ARGBTOUVJROW_AVX2 +ANY12S(ARGBToUVJRow_Any_AVX2, ARGBToUVJRow_AVX2, 0, 4, 31) +#endif #ifdef HAS_ARGBTOUVROW_SSSE3 ANY12S(ARGBToUVRow_Any_SSSE3, ARGBToUVRow_SSSE3, 0, 4, 15) ANY12S(ARGBToUVJRow_Any_SSSE3, ARGBToUVJRow_SSSE3, 0, 4, 15) diff --git a/chromium/third_party/libyuv/source/row_common.cc b/chromium/third_party/libyuv/source/row_common.cc index c820cdf1f87..2b80d074cea 100644 --- a/chromium/third_party/libyuv/source/row_common.cc +++ b/chromium/third_party/libyuv/source/row_common.cc @@ -433,28 +433,6 @@ void NAME ## ToUVJRow_C(const uint8* src_rgb0, int src_stride_rgb, \ MAKEROWYJ(ARGB, 2, 1, 0, 4) #undef MAKEROWYJ -void ARGBToUVJ422Row_C(const uint8* src_argb, - uint8* dst_u, uint8* dst_v, int width) { - int x; - for (x = 0; x < width - 1; x += 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] = RGBToUJ(ar, ag, ab); - dst_v[0] = RGBToVJ(ar, ag, ab); - src_argb += 8; - dst_u += 1; - dst_v += 1; - } - if (width & 1) { - uint8 ab = src_argb[0]; - uint8 ag = src_argb[1]; - uint8 ar = src_argb[2]; - dst_u[0] = RGBToUJ(ar, ag, ab); - dst_v[0] = RGBToVJ(ar, ag, ab); - } -} - void RGB565ToYRow_C(const uint8* src_rgb565, uint8* dst_y, int width) { int x; for (x = 0; x < width; ++x) { @@ -658,28 +636,6 @@ void ARGBToUV444Row_C(const uint8* src_argb, } } -void ARGBToUV422Row_C(const uint8* src_argb, - uint8* dst_u, uint8* dst_v, int width) { - int x; - for (x = 0; x < width - 1; x += 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); - src_argb += 8; - dst_u += 1; - dst_v += 1; - } - if (width & 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 ARGBToUV411Row_C(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width) { int x; @@ -2539,7 +2495,11 @@ void I422ToRGB565Row_AVX2(const uint8* src_y, 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; @@ -2561,7 +2521,11 @@ void I422ToARGB1555Row_AVX2(const uint8* src_y, 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; @@ -2583,7 +2547,11 @@ void I422ToARGB4444Row_AVX2(const uint8* src_y, 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; @@ -2627,7 +2595,11 @@ void NV12ToRGB565Row_AVX2(const uint8* src_y, 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; diff --git a/chromium/third_party/libyuv/source/row_gcc.cc b/chromium/third_party/libyuv/source/row_gcc.cc index 80b2a95aa10..d5174516e71 100644 --- a/chromium/third_party/libyuv/source/row_gcc.cc +++ b/chromium/third_party/libyuv/source/row_gcc.cc @@ -1023,6 +1023,67 @@ void ARGBToUVRow_AVX2(const uint8* src_argb0, int src_stride_argb, } #endif // HAS_ARGBTOUVROW_AVX2 +#ifdef HAS_ARGBTOUVJROW_AVX2 +void ARGBToUVJRow_AVX2(const uint8* src_argb0, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int width) { + asm volatile ( + "vbroadcastf128 %5,%%ymm5 \n" + "vbroadcastf128 %6,%%ymm6 \n" + "vbroadcastf128 %7,%%ymm7 \n" + "sub %1,%2 \n" + LABELALIGN + "1: \n" + "vmovdqu " MEMACCESS(0) ",%%ymm0 \n" + "vmovdqu " MEMACCESS2(0x20,0) ",%%ymm1 \n" + "vmovdqu " MEMACCESS2(0x40,0) ",%%ymm2 \n" + "vmovdqu " MEMACCESS2(0x60,0) ",%%ymm3 \n" + VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0) // vpavgb (%0,%4,1),%%ymm0,%%ymm0 + VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1) + VMEMOPREG(vpavgb,0x40,0,4,1,ymm2,ymm2) + VMEMOPREG(vpavgb,0x60,0,4,1,ymm3,ymm3) + "lea " MEMLEA(0x80,0) ",%0 \n" + "vshufps $0x88,%%ymm1,%%ymm0,%%ymm4 \n" + "vshufps $0xdd,%%ymm1,%%ymm0,%%ymm0 \n" + "vpavgb %%ymm4,%%ymm0,%%ymm0 \n" + "vshufps $0x88,%%ymm3,%%ymm2,%%ymm4 \n" + "vshufps $0xdd,%%ymm3,%%ymm2,%%ymm2 \n" + "vpavgb %%ymm4,%%ymm2,%%ymm2 \n" + + "vpmaddubsw %%ymm7,%%ymm0,%%ymm1 \n" + "vpmaddubsw %%ymm7,%%ymm2,%%ymm3 \n" + "vpmaddubsw %%ymm6,%%ymm0,%%ymm0 \n" + "vpmaddubsw %%ymm6,%%ymm2,%%ymm2 \n" + "vphaddw %%ymm3,%%ymm1,%%ymm1 \n" + "vphaddw %%ymm2,%%ymm0,%%ymm0 \n" + "vpaddw %%ymm5,%%ymm0,%%ymm0 \n" + "vpaddw %%ymm5,%%ymm1,%%ymm1 \n" + "vpsraw $0x8,%%ymm1,%%ymm1 \n" + "vpsraw $0x8,%%ymm0,%%ymm0 \n" + "vpacksswb %%ymm0,%%ymm1,%%ymm0 \n" + "vpermq $0xd8,%%ymm0,%%ymm0 \n" + "vpshufb %8,%%ymm0,%%ymm0 \n" + + "vextractf128 $0x0,%%ymm0," MEMACCESS(1) " \n" + VEXTOPMEM(vextractf128,1,ymm0,0x0,1,2,1) // vextractf128 $1,%%ymm0,(%1,%2,1) + "lea " MEMLEA(0x10,1) ",%1 \n" + "sub $0x20,%3 \n" + "jg 1b \n" + "vzeroupper \n" + : "+r"(src_argb0), // %0 + "+r"(dst_u), // %1 + "+r"(dst_v), // %2 + "+rm"(width) // %3 + : "r"((intptr_t)(src_stride_argb)), // %4 + "m"(kAddUVJ128), // %5 + "m"(kARGBToVJ), // %6 + "m"(kARGBToUJ), // %7 + "m"(kShufARGBToUV_AVX) // %8 + : "memory", "cc", NACL_R14 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" + ); +} +#endif // HAS_ARGBTOUVJROW_AVX2 + #ifdef HAS_ARGBTOUVJROW_SSSE3 void ARGBToUVJRow_SSSE3(const uint8* src_argb0, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width) { @@ -1144,59 +1205,6 @@ void ARGBToUV444Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, } #endif // HAS_ARGBTOUV444ROW_SSSE3 -#ifdef HAS_ARGBTOUV422ROW_SSSE3 -void ARGBToUV422Row_SSSE3(const uint8* src_argb0, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "movdqa %4,%%xmm3 \n" - "movdqa %5,%%xmm4 \n" - "movdqa %6,%%xmm5 \n" - "sub %1,%2 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "movdqu " MEMACCESS2(0x20,0) ",%%xmm2 \n" - "movdqu " MEMACCESS2(0x30,0) ",%%xmm6 \n" - "lea " MEMLEA(0x40,0) ",%0 \n" - "movdqa %%xmm0,%%xmm7 \n" - "shufps $0x88,%%xmm1,%%xmm0 \n" - "shufps $0xdd,%%xmm1,%%xmm7 \n" - "pavgb %%xmm7,%%xmm0 \n" - "movdqa %%xmm2,%%xmm7 \n" - "shufps $0x88,%%xmm6,%%xmm2 \n" - "shufps $0xdd,%%xmm6,%%xmm7 \n" - "pavgb %%xmm7,%%xmm2 \n" - "movdqa %%xmm0,%%xmm1 \n" - "movdqa %%xmm2,%%xmm6 \n" - "pmaddubsw %%xmm4,%%xmm0 \n" - "pmaddubsw %%xmm4,%%xmm2 \n" - "pmaddubsw %%xmm3,%%xmm1 \n" - "pmaddubsw %%xmm3,%%xmm6 \n" - "phaddw %%xmm2,%%xmm0 \n" - "phaddw %%xmm6,%%xmm1 \n" - "psraw $0x8,%%xmm0 \n" - "psraw $0x8,%%xmm1 \n" - "packsswb %%xmm1,%%xmm0 \n" - "paddb %%xmm5,%%xmm0 \n" - "movlps %%xmm0," MEMACCESS(1) " \n" - MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) - "lea " MEMLEA(0x8,1) ",%1 \n" - "sub $0x10,%3 \n" - "jg 1b \n" - : "+r"(src_argb0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : "m"(kARGBToV), // %4 - "m"(kARGBToU), // %5 - "m"(kAddUV128) // %6 - : "memory", "cc", NACL_R14 - "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" - ); -} -#endif // HAS_ARGBTOUV422ROW_SSSE3 - void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int width) { asm volatile ( "movdqa %4,%%xmm5 \n" @@ -1484,7 +1492,7 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, #if defined(HAS_I422TOARGBROW_SSSE3) || defined(HAS_I422TOARGBROW_AVX2) -// Read 8 UV from 411 +// Read 8 UV from 444 #define READYUV444 \ "movq " MEMACCESS([u_buf]) ",%%xmm0 \n" \ MEMOPREG(movq, 0x00, [u_buf], [v_buf], 1, xmm1) \ @@ -1528,7 +1536,7 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, #define READYUV411_TEMP \ "movzwl " MEMACCESS([u_buf]) ",%[temp] \n" \ "movd %[temp],%%xmm0 \n" \ - MEMOPARG(movzwl,0x00,[u_buf],[v_buf],1,[temp]) " \n" \ + MEMOPARG(movzwl, 0x00, [u_buf], [v_buf], 1, [temp]) " \n" \ "movd %[temp],%%xmm1 \n" \ "lea " MEMLEA(0x2, [u_buf]) ",%[u_buf] \n" \ "punpcklbw %%xmm1,%%xmm0 \n" \ @@ -2005,6 +2013,20 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8* y_buf, "vpermq $0xd8,%%ymm5,%%ymm5 \n" \ "lea " MEMLEA(0x10, [a_buf]) ",%[a_buf] \n" +// Read 4 UV from 411, upsample to 16 UV. +#define READYUV411_AVX2 \ + "vmovd " MEMACCESS([u_buf]) ",%%xmm0 \n" \ + MEMOPREG(vmovd, 0x00, [u_buf], [v_buf], 1, xmm1) \ + "lea " MEMLEA(0x4, [u_buf]) ",%[u_buf] \n" \ + "vpunpcklbw %%ymm1,%%ymm0,%%ymm0 \n" \ + "vpunpcklwd %%ymm0,%%ymm0,%%ymm0 \n" \ + "vpermq $0xd8,%%ymm0,%%ymm0 \n" \ + "vpunpckldq %%ymm0,%%ymm0,%%ymm0 \n" \ + "vmovdqu " MEMACCESS([y_buf]) ",%%xmm4 \n" \ + "vpermq $0xd8,%%ymm4,%%ymm4 \n" \ + "vpunpcklbw %%ymm4,%%ymm4,%%ymm4 \n" \ + "lea " MEMLEA(0x10, [y_buf]) ",%[y_buf] \n" + // Read 8 UV from NV12, upsample to 16 UV. #define READNV12_AVX2 \ "vmovdqu " MEMACCESS([uv_buf]) ",%%xmm0 \n" \ @@ -2071,7 +2093,7 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8* y_buf, "vpackuswb %%ymm2,%%ymm2,%%ymm2 \n" #define YUVTORGB_REGS_AVX2 \ "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", -#else// Convert 16 pixels: 16 UV and 16 Y. +#else // Convert 16 pixels: 16 UV and 16 Y. #define YUVTORGB_SETUP_AVX2(yuvconstants) #define YUVTORGB_AVX2(yuvconstants) \ "vpmaddubsw " MEMACCESS2(64, [yuvconstants]) ",%%ymm0,%%ymm2 \n" \ @@ -2120,7 +2142,7 @@ void OMITFP I444ToARGBRow_AVX2(const uint8* y_buf, asm volatile ( YUVTORGB_SETUP_AVX2(yuvconstants) "sub %[u_buf],%[v_buf] \n" - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" + "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" LABELALIGN "1: \n" READYUV444_AVX2 @@ -2141,6 +2163,39 @@ void OMITFP I444ToARGBRow_AVX2(const uint8* y_buf, } #endif // HAS_I444TOARGBROW_AVX2 +#ifdef HAS_I411TOARGBROW_AVX2 +// 16 pixels +// 4 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes). +void OMITFP I411ToARGBRow_AVX2(const uint8* y_buf, + const uint8* u_buf, + const uint8* v_buf, + uint8* dst_argb, + const struct YuvConstants* yuvconstants, + int width) { + asm volatile ( + YUVTORGB_SETUP_AVX2(yuvconstants) + "sub %[u_buf],%[v_buf] \n" + "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" + LABELALIGN + "1: \n" + READYUV411_AVX2 + YUVTORGB_AVX2(yuvconstants) + STOREARGB_AVX2 + "sub $0x10,%[width] \n" + "jg 1b \n" + "vzeroupper \n" + : [y_buf]"+r"(y_buf), // %[y_buf] + [u_buf]"+r"(u_buf), // %[u_buf] + [v_buf]"+r"(v_buf), // %[v_buf] + [dst_argb]"+r"(dst_argb), // %[dst_argb] + [width]"+rm"(width) // %[width] + : [yuvconstants]"r"(yuvconstants) // %[yuvconstants] + : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2 + "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" + ); +} +#endif // HAS_I411TOARGBROW_AVX2 + #if defined(HAS_I422TOARGBROW_AVX2) // 16 pixels // 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes). @@ -2153,7 +2208,7 @@ void OMITFP I422ToARGBRow_AVX2(const uint8* y_buf, asm volatile ( YUVTORGB_SETUP_AVX2(yuvconstants) "sub %[u_buf],%[v_buf] \n" - "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" + "vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" LABELALIGN "1: \n" READYUV422_AVX2 @@ -2521,7 +2576,7 @@ void MirrorRow_AVX2(const uint8* src, uint8* dst, int width) { } #endif // HAS_MIRRORROW_AVX2 -#ifdef HAS_MIRRORROW_UV_SSSE3 +#ifdef HAS_MIRRORUVROW_SSSE3 // Shuffle table for reversing the bytes of UV channels. static uvec8 kShuffleMirrorUV = { 14u, 12u, 10u, 8u, 6u, 4u, 2u, 0u, 15u, 13u, 11u, 9u, 7u, 5u, 3u, 1u @@ -2552,7 +2607,7 @@ void MirrorUVRow_SSSE3(const uint8* src, uint8* dst_u, uint8* dst_v, "xmm0", "xmm1" ); } -#endif // HAS_MIRRORROW_UV_SSSE3 +#endif // HAS_MIRRORUVROW_SSSE3 #ifdef HAS_ARGBMIRRORROW_SSE2 @@ -2953,7 +3008,7 @@ void ARGBCopyYToAlphaRow_AVX2(const uint8* src, uint8* dst, int width) { #ifdef HAS_SETROW_X86 void SetRow_X86(uint8* dst, uint8 v8, int width) { size_t width_tmp = (size_t)(width >> 2); - const uint32 v32 = v8 * 0x01010101; // Duplicate byte to all bytes. + const uint32 v32 = v8 * 0x01010101u; // Duplicate byte to all bytes. asm volatile ( "rep stosl " MEMSTORESTRING(eax,0) " \n" : "+D"(dst), // %0 diff --git a/chromium/third_party/libyuv/source/row_mips.cc b/chromium/third_party/libyuv/source/row_mips.cc index d12cf6ab790..2c55b786b2a 100644 --- a/chromium/third_party/libyuv/source/row_mips.cc +++ b/chromium/third_party/libyuv/source/row_mips.cc @@ -375,12 +375,12 @@ void CopyRow_MIPS(const uint8* src, uint8* dst, int count) { } #endif // HAS_COPYROW_MIPS -// MIPS DSPR2 functions +// DSPR2 functions #if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips_dsp) && \ (__mips_dsp_rev >= 2) && \ (_MIPS_SIM == _MIPS_SIM_ABI32) && (__mips_isa_rev < 6) -void SplitUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, +void SplitUVRow_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { __asm__ __volatile__ ( ".set push \n" @@ -446,7 +446,7 @@ void SplitUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, ); } -void MirrorRow_MIPS_DSPR2(const uint8* src, uint8* dst, int width) { +void MirrorRow_DSPR2(const uint8* src, uint8* dst, int width) { __asm__ __volatile__ ( ".set push \n" ".set noreorder \n" @@ -496,7 +496,7 @@ void MirrorRow_MIPS_DSPR2(const uint8* src, uint8* dst, int width) { ); } -void MirrorUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, +void MirrorUVRow_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) { int x = 0; int y = 0; @@ -653,7 +653,7 @@ void MirrorUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, "addu.ph $t1, $t1, $s5 \n" // TODO(fbarchard): accept yuv conversion constants. -void I422ToARGBRow_MIPS_DSPR2(const uint8* y_buf, +void I422ToARGBRow_DSPR2(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, uint8* rgb_buf, @@ -716,7 +716,7 @@ void I422ToARGBRow_MIPS_DSPR2(const uint8* y_buf, } // Bilinear filter 8x2 -> 8x1 -void InterpolateRow_MIPS_DSPR2(uint8* dst_ptr, const uint8* src_ptr, +void InterpolateRow_DSPR2(uint8* dst_ptr, const uint8* src_ptr, ptrdiff_t src_stride, int dst_width, int source_y_fraction) { int y0_fraction = 256 - source_y_fraction; diff --git a/chromium/third_party/libyuv/source/row_neon.cc b/chromium/third_party/libyuv/source/row_neon.cc index 5b4ff3b5a88..91d6aa857b6 100644 --- a/chromium/third_party/libyuv/source/row_neon.cc +++ b/chromium/third_party/libyuv/source/row_neon.cc @@ -317,16 +317,11 @@ void I422ToRGB24Row_NEON(const uint8* src_y, } #define ARGBTORGB565 \ - "vshr.u8 d20, d20, #3 \n" /* B */ \ - "vshr.u8 d21, d21, #2 \n" /* G */ \ - "vshr.u8 d22, d22, #3 \n" /* R */ \ - "vmovl.u8 q8, d20 \n" /* B */ \ - "vmovl.u8 q9, d21 \n" /* G */ \ - "vmovl.u8 q10, d22 \n" /* R */ \ - "vshl.u16 q9, q9, #5 \n" /* G */ \ - "vshl.u16 q10, q10, #11 \n" /* R */ \ - "vorr q0, q8, q9 \n" /* BG */ \ - "vorr q0, q0, q10 \n" /* BGR */ + "vshll.u8 q0, d22, #8 \n" /* R */ \ + "vshll.u8 q8, d21, #8 \n" /* G */ \ + "vshll.u8 q9, d20, #8 \n" /* B */ \ + "vsri.16 q0, q8, #5 \n" /* RG */ \ + "vsri.16 q0, q9, #11 \n" /* RGB */ void I422ToRGB565Row_NEON(const uint8* src_y, const uint8* src_u, @@ -359,19 +354,13 @@ void I422ToRGB565Row_NEON(const uint8* src_y, } #define ARGBTOARGB1555 \ - "vshr.u8 q10, q10, #3 \n" /* B */ \ - "vshr.u8 d22, d22, #3 \n" /* R */ \ - "vshr.u8 d23, d23, #7 \n" /* A */ \ - "vmovl.u8 q8, d20 \n" /* B */ \ - "vmovl.u8 q9, d21 \n" /* G */ \ - "vmovl.u8 q10, d22 \n" /* R */ \ - "vmovl.u8 q11, d23 \n" /* A */ \ - "vshl.u16 q9, q9, #5 \n" /* G */ \ - "vshl.u16 q10, q10, #10 \n" /* R */ \ - "vshl.u16 q11, q11, #15 \n" /* A */ \ - "vorr q0, q8, q9 \n" /* BG */ \ - "vorr q1, q10, q11 \n" /* RA */ \ - "vorr q0, q0, q1 \n" /* BGRA */ + "vshll.u8 q0, d23, #8 \n" /* A */ \ + "vshll.u8 q8, d22, #8 \n" /* R */ \ + "vshll.u8 q9, d21, #8 \n" /* G */ \ + "vshll.u8 q10, d20, #8 \n" /* B */ \ + "vsri.16 q0, q8, #1 \n" /* AR */ \ + "vsri.16 q0, q9, #6 \n" /* ARG */ \ + "vsri.16 q0, q10, #11 \n" /* ARGB */ void I422ToARGB1555Row_NEON(const uint8* src_y, const uint8* src_u, @@ -1374,55 +1363,6 @@ void ARGBToUV444Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, ); } -// 16x1 pixels -> 8x1. width is number of argb pixels. e.g. 16. -void ARGBToUV422Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, - int width) { - asm volatile ( - "vmov.s16 q10, #112 / 2 \n" // UB / VR 0.875 coefficient - "vmov.s16 q11, #74 / 2 \n" // UG -0.5781 coefficient - "vmov.s16 q12, #38 / 2 \n" // UR -0.2969 coefficient - "vmov.s16 q13, #18 / 2 \n" // VB -0.1406 coefficient - "vmov.s16 q14, #94 / 2 \n" // VG -0.7344 coefficient - "vmov.u16 q15, #0x8080 \n" // 128.5 - "1: \n" - MEMACCESS(0) - "vld4.8 {d0, d2, d4, d6}, [%0]! \n" // load 8 ARGB pixels. - MEMACCESS(0) - "vld4.8 {d1, d3, d5, d7}, [%0]! \n" // load next 8 ARGB pixels. - - "vpaddl.u8 q0, q0 \n" // B 16 bytes -> 8 shorts. - "vpaddl.u8 q1, q1 \n" // G 16 bytes -> 8 shorts. - "vpaddl.u8 q2, q2 \n" // R 16 bytes -> 8 shorts. - - "subs %3, %3, #16 \n" // 16 processed per loop. - "vmul.s16 q8, q0, q10 \n" // B - "vmls.s16 q8, q1, q11 \n" // G - "vmls.s16 q8, q2, q12 \n" // R - "vadd.u16 q8, q8, q15 \n" // +128 -> unsigned - - "vmul.s16 q9, q2, q10 \n" // R - "vmls.s16 q9, q1, q14 \n" // G - "vmls.s16 q9, q0, q13 \n" // B - "vadd.u16 q9, q9, q15 \n" // +128 -> unsigned - - "vqshrn.u16 d0, q8, #8 \n" // 16 bit to 8 bit U - "vqshrn.u16 d1, q9, #8 \n" // 16 bit to 8 bit V - - MEMACCESS(1) - "vst1.8 {d0}, [%1]! \n" // store 8 pixels U. - MEMACCESS(2) - "vst1.8 {d1}, [%2]! \n" // store 8 pixels V. - "bgt 1b \n" - : "+r"(src_argb), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(width) // %3 - : - : "cc", "memory", "q0", "q1", "q2", "q3", - "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15" - ); -} - // 32x1 pixels -> 8x1. width is number of argb pixels. e.g. 32. void ARGBToUV411Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width) { diff --git a/chromium/third_party/libyuv/source/row_neon64.cc b/chromium/third_party/libyuv/source/row_neon64.cc index 6fe5a108030..ee42af12e37 100644 --- a/chromium/third_party/libyuv/source/row_neon64.cc +++ b/chromium/third_party/libyuv/source/row_neon64.cc @@ -323,8 +323,8 @@ void I422ToRGB24Row_NEON(const uint8* src_y, #define ARGBTORGB565 \ "shll v0.8h, v22.8b, #8 \n" /* R */ \ - "shll v20.8h, v20.8b, #8 \n" /* B */ \ "shll v21.8h, v21.8b, #8 \n" /* G */ \ + "shll v20.8h, v20.8b, #8 \n" /* B */ \ "sri v0.8h, v21.8h, #5 \n" /* RG */ \ "sri v0.8h, v20.8h, #11 \n" /* RGB */ @@ -363,8 +363,8 @@ void I422ToRGB565Row_NEON(const uint8* src_y, #define ARGBTOARGB1555 \ "shll v0.8h, v23.8b, #8 \n" /* A */ \ "shll v22.8h, v22.8b, #8 \n" /* R */ \ - "shll v20.8h, v20.8b, #8 \n" /* B */ \ "shll v21.8h, v21.8b, #8 \n" /* G */ \ + "shll v20.8h, v20.8b, #8 \n" /* B */ \ "sri v0.8h, v22.8h, #1 \n" /* AR */ \ "sri v0.8h, v21.8h, #6 \n" /* ARG */ \ "sri v0.8h, v20.8h, #11 \n" /* ARGB */ @@ -1477,50 +1477,6 @@ void ARGBToUV444Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, "movi v24.8h, #47, lsl #0 \n" /* VG coefficient (-0.7344) / 2 */ \ "movi v25.16b, #0x80 \n" /* 128.5 (0x8080 in 16-bit) */ -// 16x1 pixels -> 8x1. width is number of argb pixels. e.g. 16. -#ifdef HAS_ARGBTOUV422ROW_NEON -void ARGBToUV422Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, - int width) { - asm volatile ( - RGBTOUV_SETUP_REG - "1: \n" - MEMACCESS(0) - "ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 16 pixels. - - "uaddlp v0.8h, v0.16b \n" // B 16 bytes -> 8 shorts. - "uaddlp v1.8h, v1.16b \n" // G 16 bytes -> 8 shorts. - "uaddlp v2.8h, v2.16b \n" // R 16 bytes -> 8 shorts. - - "subs %w3, %w3, #16 \n" // 16 processed per loop. - "mul v3.8h, v0.8h, v20.8h \n" // B - "mls v3.8h, v1.8h, v21.8h \n" // G - "mls v3.8h, v2.8h, v22.8h \n" // R - "add v3.8h, v3.8h, v25.8h \n" // +128 -> unsigned - - "mul v4.8h, v2.8h, v20.8h \n" // R - "mls v4.8h, v1.8h, v24.8h \n" // G - "mls v4.8h, v0.8h, v23.8h \n" // B - "add v4.8h, v4.8h, v25.8h \n" // +128 -> unsigned - - "uqshrn v0.8b, v3.8h, #8 \n" // 16 bit to 8 bit U - "uqshrn v1.8b, v4.8h, #8 \n" // 16 bit to 8 bit V - - MEMACCESS(1) - "st1 {v0.8b}, [%1], #8 \n" // store 8 pixels U. - MEMACCESS(2) - "st1 {v1.8b}, [%2], #8 \n" // store 8 pixels V. - "b.gt 1b \n" - : "+r"(src_argb), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(width) // %3 - : - : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", - "v20", "v21", "v22", "v23", "v24", "v25" - ); -} -#endif // HAS_ARGBTOUV422ROW_NEON - // 32x1 pixels -> 8x1. width is number of argb pixels. e.g. 32. #ifdef HAS_ARGBTOUV411ROW_NEON void ARGBToUV411Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, diff --git a/chromium/third_party/libyuv/source/row_win.cc b/chromium/third_party/libyuv/source/row_win.cc index 5cb5d1e4f5d..a8c16c3c1ef 100644 --- a/chromium/third_party/libyuv/source/row_win.cc +++ b/chromium/third_party/libyuv/source/row_win.cc @@ -1505,7 +1505,7 @@ void ARGBToUVJRow_SSSE3(const uint8* src_argb0, int src_stride_argb, pmaddubsw xmm3, xmm6 phaddw xmm0, xmm2 phaddw xmm1, xmm3 - paddw xmm0, xmm5 // +.5 rounding -> unsigned + paddw xmm0, xmm5 // +.5 rounding -> unsigned paddw xmm1, xmm5 psraw xmm0, 8 psraw xmm1, 8 @@ -1590,6 +1590,73 @@ void ARGBToUVRow_AVX2(const uint8* src_argb0, int src_stride_argb, } #endif // HAS_ARGBTOUVROW_AVX2 +#ifdef HAS_ARGBTOUVJROW_AVX2 +__declspec(naked) +void ARGBToUVJRow_AVX2(const uint8* src_argb0, int src_stride_argb, + uint8* dst_u, uint8* dst_v, int width) { + __asm { + push esi + push edi + mov eax, [esp + 8 + 4] // src_argb + mov esi, [esp + 8 + 8] // src_stride_argb + mov edx, [esp + 8 + 12] // dst_u + mov edi, [esp + 8 + 16] // dst_v + mov ecx, [esp + 8 + 20] // width + vbroadcastf128 ymm5, xmmword ptr kAddUV128 + vbroadcastf128 ymm6, xmmword ptr kARGBToV + vbroadcastf128 ymm7, xmmword ptr kARGBToU + sub edi, edx // stride from u to v + + convertloop: + /* step 1 - subsample 32x2 argb pixels to 16x1 */ + vmovdqu ymm0, [eax] + vmovdqu ymm1, [eax + 32] + vmovdqu ymm2, [eax + 64] + vmovdqu ymm3, [eax + 96] + vpavgb ymm0, ymm0, [eax + esi] + vpavgb ymm1, ymm1, [eax + esi + 32] + vpavgb ymm2, ymm2, [eax + esi + 64] + vpavgb ymm3, ymm3, [eax + esi + 96] + lea eax, [eax + 128] + vshufps ymm4, ymm0, ymm1, 0x88 + vshufps ymm0, ymm0, ymm1, 0xdd + vpavgb ymm0, ymm0, ymm4 // mutated by vshufps + vshufps ymm4, ymm2, ymm3, 0x88 + vshufps ymm2, ymm2, ymm3, 0xdd + vpavgb ymm2, ymm2, ymm4 // mutated by vshufps + + // step 2 - convert to U and V + // from here down is very similar to Y code except + // instead of 32 different pixels, its 16 pixels of U and 16 of V + vpmaddubsw ymm1, ymm0, ymm7 // U + vpmaddubsw ymm3, ymm2, ymm7 + vpmaddubsw ymm0, ymm0, ymm6 // V + vpmaddubsw ymm2, ymm2, ymm6 + vphaddw ymm1, ymm1, ymm3 // mutates + vphaddw ymm0, ymm0, ymm2 + vpaddw ymm1, ymm1, ymm5 // +.5 rounding -> unsigned + vpaddw ymm0, ymm0, ymm5 + vpsraw ymm1, ymm1, 8 + vpsraw ymm0, ymm0, 8 + vpacksswb ymm0, ymm1, ymm0 // mutates + vpermq ymm0, ymm0, 0xd8 // For vpacksswb + vpshufb ymm0, ymm0, ymmword ptr kShufARGBToUV_AVX // for vshufps/vphaddw + + // step 3 - store 16 U and 16 V values + vextractf128 [edx], ymm0, 0 // U + vextractf128 [edx + edi], ymm0, 1 // V + lea edx, [edx + 16] + sub ecx, 32 + jg convertloop + + pop edi + pop esi + vzeroupper + ret + } +} +#endif // HAS_ARGBTOUVJROW_AVX2 + __declspec(naked) void ARGBToUV444Row_SSSE3(const uint8* src_argb0, uint8* dst_u, uint8* dst_v, int width) { @@ -1648,64 +1715,6 @@ void ARGBToUV444Row_SSSE3(const uint8* src_argb0, } __declspec(naked) -void ARGBToUV422Row_SSSE3(const uint8* src_argb0, - uint8* dst_u, uint8* dst_v, int width) { - __asm { - push edi - mov eax, [esp + 4 + 4] // src_argb - mov edx, [esp + 4 + 8] // dst_u - mov edi, [esp + 4 + 12] // dst_v - mov ecx, [esp + 4 + 16] // width - movdqa xmm5, xmmword ptr kAddUV128 - movdqa xmm6, xmmword ptr kARGBToV - movdqa xmm7, xmmword ptr kARGBToU - sub edi, edx // stride from u to v - - convertloop: - /* step 1 - subsample 16x2 argb pixels to 8x1 */ - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - lea eax, [eax + 64] - movdqa xmm4, xmm0 - shufps xmm0, xmm1, 0x88 - shufps xmm4, xmm1, 0xdd - pavgb xmm0, xmm4 - movdqa xmm4, xmm2 - shufps xmm2, xmm3, 0x88 - shufps xmm4, xmm3, 0xdd - pavgb xmm2, xmm4 - - // step 2 - convert to U and V - // from here down is very similar to Y code except - // instead of 16 different pixels, its 8 pixels of U and 8 of V - movdqa xmm1, xmm0 - movdqa xmm3, xmm2 - pmaddubsw xmm0, xmm7 // U - pmaddubsw xmm2, xmm7 - pmaddubsw xmm1, xmm6 // V - pmaddubsw xmm3, xmm6 - phaddw xmm0, xmm2 - phaddw xmm1, xmm3 - psraw xmm0, 8 - psraw xmm1, 8 - packsswb xmm0, xmm1 - paddb xmm0, xmm5 // -> unsigned - - // step 3 - store 8 U and 8 V values - movlps qword ptr [edx], xmm0 // U - movhps qword ptr [edx + edi], xmm0 // V - lea edx, [edx + 8] - sub ecx, 16 - jg convertloop - - pop edi - ret - } -} - -__declspec(naked) void BGRAToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width) { __asm { @@ -3154,7 +3163,7 @@ void MirrorRow_AVX2(const uint8* src, uint8* dst, int width) { } #endif // HAS_MIRRORROW_AVX2 -#ifdef HAS_MIRRORROW_UV_SSSE3 +#ifdef HAS_MIRRORUVROW_SSSE3 // Shuffle table for reversing the bytes of UV channels. static const uvec8 kShuffleMirrorUV = { 14u, 12u, 10u, 8u, 6u, 4u, 2u, 0u, 15u, 13u, 11u, 9u, 7u, 5u, 3u, 1u @@ -3187,7 +3196,7 @@ void MirrorUVRow_SSSE3(const uint8* src, uint8* dst_u, uint8* dst_v, ret } } -#endif // HAS_MIRRORROW_UV_SSSE3 +#endif // HAS_MIRRORUVROW_SSSE3 #ifdef HAS_ARGBMIRRORROW_SSE2 __declspec(naked) diff --git a/chromium/third_party/libyuv/source/scale.cc b/chromium/third_party/libyuv/source/scale.cc index 595314f35cb..36e3fe52813 100644 --- a/chromium/third_party/libyuv/source/scale.cc +++ b/chromium/third_party/libyuv/source/scale.cc @@ -85,12 +85,12 @@ static void ScalePlaneDown2(int src_width, int src_height, } } #endif -#if defined(HAS_SCALEROWDOWN2_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(src_ptr, 4) && +#if defined(HAS_SCALEROWDOWN2_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(row_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { ScaleRowDown2 = filtering ? - ScaleRowDown2Box_MIPS_DSPR2 : ScaleRowDown2_MIPS_DSPR2; + ScaleRowDown2Box_DSPR2 : ScaleRowDown2_DSPR2; } #endif @@ -135,12 +135,12 @@ static void ScalePlaneDown2_16(int src_width, int src_height, ScaleRowDown2Box_16_SSE2); } #endif -#if defined(HAS_SCALEROWDOWN2_16_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(src_ptr, 4) && +#if defined(HAS_SCALEROWDOWN2_16_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(row_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { ScaleRowDown2 = filtering ? - ScaleRowDown2Box_16_MIPS_DSPR2 : ScaleRowDown2_16_MIPS_DSPR2; + ScaleRowDown2Box_16_DSPR2 : ScaleRowDown2_16_DSPR2; } #endif @@ -200,12 +200,12 @@ static void ScalePlaneDown4(int src_width, int src_height, } } #endif -#if defined(HAS_SCALEROWDOWN4_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(row_stride, 4) && +#if defined(HAS_SCALEROWDOWN4_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(row_stride, 4) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { ScaleRowDown4 = filtering ? - ScaleRowDown4Box_MIPS_DSPR2 : ScaleRowDown4_MIPS_DSPR2; + ScaleRowDown4Box_DSPR2 : ScaleRowDown4_DSPR2; } #endif @@ -245,12 +245,12 @@ static void ScalePlaneDown4_16(int src_width, int src_height, ScaleRowDown4_16_SSE2; } #endif -#if defined(HAS_SCALEROWDOWN4_16_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(row_stride, 4) && +#if defined(HAS_SCALEROWDOWN4_16_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(row_stride, 4) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { ScaleRowDown4 = filtering ? - ScaleRowDown4Box_16_MIPS_DSPR2 : ScaleRowDown4_16_MIPS_DSPR2; + ScaleRowDown4Box_16_DSPR2 : ScaleRowDown4_16_DSPR2; } #endif @@ -325,16 +325,16 @@ static void ScalePlaneDown34(int src_width, int src_height, } } #endif -#if defined(HAS_SCALEROWDOWN34_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && (dst_width % 24 == 0) && +#if defined(HAS_SCALEROWDOWN34_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && (dst_width % 24 == 0) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { if (!filtering) { - ScaleRowDown34_0 = ScaleRowDown34_MIPS_DSPR2; - ScaleRowDown34_1 = ScaleRowDown34_MIPS_DSPR2; + ScaleRowDown34_0 = ScaleRowDown34_DSPR2; + ScaleRowDown34_1 = ScaleRowDown34_DSPR2; } else { - ScaleRowDown34_0 = ScaleRowDown34_0_Box_MIPS_DSPR2; - ScaleRowDown34_1 = ScaleRowDown34_1_Box_MIPS_DSPR2; + ScaleRowDown34_0 = ScaleRowDown34_0_Box_DSPR2; + ScaleRowDown34_1 = ScaleRowDown34_1_Box_DSPR2; } } #endif @@ -404,16 +404,16 @@ static void ScalePlaneDown34_16(int src_width, int src_height, } } #endif -#if defined(HAS_SCALEROWDOWN34_16_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && (dst_width % 24 == 0) && +#if defined(HAS_SCALEROWDOWN34_16_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && (dst_width % 24 == 0) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { if (!filtering) { - ScaleRowDown34_0 = ScaleRowDown34_16_MIPS_DSPR2; - ScaleRowDown34_1 = ScaleRowDown34_16_MIPS_DSPR2; + ScaleRowDown34_0 = ScaleRowDown34_16_DSPR2; + ScaleRowDown34_1 = ScaleRowDown34_16_DSPR2; } else { - ScaleRowDown34_0 = ScaleRowDown34_0_Box_16_MIPS_DSPR2; - ScaleRowDown34_1 = ScaleRowDown34_1_Box_16_MIPS_DSPR2; + ScaleRowDown34_0 = ScaleRowDown34_0_Box_16_DSPR2; + ScaleRowDown34_1 = ScaleRowDown34_1_Box_16_DSPR2; } } #endif @@ -517,16 +517,16 @@ static void ScalePlaneDown38(int src_width, int src_height, } } #endif -#if defined(HAS_SCALEROWDOWN38_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && (dst_width % 12 == 0) && +#if defined(HAS_SCALEROWDOWN38_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && (dst_width % 12 == 0) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { if (!filtering) { - ScaleRowDown38_3 = ScaleRowDown38_MIPS_DSPR2; - ScaleRowDown38_2 = ScaleRowDown38_MIPS_DSPR2; + ScaleRowDown38_3 = ScaleRowDown38_DSPR2; + ScaleRowDown38_2 = ScaleRowDown38_DSPR2; } else { - ScaleRowDown38_3 = ScaleRowDown38_3_Box_MIPS_DSPR2; - ScaleRowDown38_2 = ScaleRowDown38_2_Box_MIPS_DSPR2; + ScaleRowDown38_3 = ScaleRowDown38_3_Box_DSPR2; + ScaleRowDown38_2 = ScaleRowDown38_2_Box_DSPR2; } } #endif @@ -595,16 +595,16 @@ static void ScalePlaneDown38_16(int src_width, int src_height, } } #endif -#if defined(HAS_SCALEROWDOWN38_16_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && (dst_width % 12 == 0) && +#if defined(HAS_SCALEROWDOWN38_16_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && (dst_width % 12 == 0) && IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) { if (!filtering) { - ScaleRowDown38_3 = ScaleRowDown38_16_MIPS_DSPR2; - ScaleRowDown38_2 = ScaleRowDown38_16_MIPS_DSPR2; + ScaleRowDown38_3 = ScaleRowDown38_16_DSPR2; + ScaleRowDown38_2 = ScaleRowDown38_16_DSPR2; } else { - ScaleRowDown38_3 = ScaleRowDown38_3_Box_16_MIPS_DSPR2; - ScaleRowDown38_2 = ScaleRowDown38_2_Box_16_MIPS_DSPR2; + ScaleRowDown38_3 = ScaleRowDown38_3_Box_16_DSPR2; + ScaleRowDown38_2 = ScaleRowDown38_2_Box_16_DSPR2; } } #endif @@ -659,7 +659,6 @@ static void ScaleAddCols2_C(int dst_width, int boxheight, int x, int dx, int i; int scaletbl[2]; int minboxwidth = dx >> 16; - int* scaleptr = scaletbl - minboxwidth; int boxwidth; scaletbl[0] = 65536 / (MIN1(minboxwidth) * boxheight); scaletbl[1] = 65536 / (MIN1(minboxwidth + 1) * boxheight); @@ -667,7 +666,8 @@ static void ScaleAddCols2_C(int dst_width, int boxheight, int x, int dx, int ix = x >> 16; x += dx; boxwidth = MIN1((x >> 16) - ix); - *dst_ptr++ = SumPixels(boxwidth, src_ptr + ix) * scaleptr[boxwidth] >> 16; + *dst_ptr++ = SumPixels(boxwidth, src_ptr + ix) * + scaletbl[boxwidth - minboxwidth] >> 16; } } @@ -676,7 +676,6 @@ static void ScaleAddCols2_16_C(int dst_width, int boxheight, int x, int dx, int i; int scaletbl[2]; int minboxwidth = dx >> 16; - int* scaleptr = scaletbl - minboxwidth; int boxwidth; scaletbl[0] = 65536 / (MIN1(minboxwidth) * boxheight); scaletbl[1] = 65536 / (MIN1(minboxwidth + 1) * boxheight); @@ -684,8 +683,8 @@ static void ScaleAddCols2_16_C(int dst_width, int boxheight, int x, int dx, int ix = x >> 16; x += dx; boxwidth = MIN1((x >> 16) - ix); - *dst_ptr++ = - SumPixels_16(boxwidth, src_ptr + ix) * scaleptr[boxwidth] >> 16; + *dst_ptr++ = SumPixels_16(boxwidth, src_ptr + ix) * + scaletbl[boxwidth - minboxwidth] >> 16; } } @@ -899,11 +898,11 @@ void ScalePlaneBilinearDown(int src_width, int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2)) { - InterpolateRow = InterpolateRow_Any_MIPS_DSPR2; +#if defined(HAS_INTERPOLATEROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + InterpolateRow = InterpolateRow_Any_DSPR2; if (IS_ALIGNED(src_width, 4)) { - InterpolateRow = InterpolateRow_MIPS_DSPR2; + InterpolateRow = InterpolateRow_DSPR2; } } #endif @@ -1003,11 +1002,11 @@ void ScalePlaneBilinearDown_16(int src_width, int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_16_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2)) { - InterpolateRow = InterpolateRow_Any_16_MIPS_DSPR2; +#if defined(HAS_INTERPOLATEROW_16_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + InterpolateRow = InterpolateRow_Any_16_DSPR2; if (IS_ALIGNED(src_width, 4)) { - InterpolateRow = InterpolateRow_16_MIPS_DSPR2; + InterpolateRow = InterpolateRow_16_DSPR2; } } #endif @@ -1088,11 +1087,11 @@ void ScalePlaneBilinearUp(int src_width, int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2)) { - InterpolateRow = InterpolateRow_Any_MIPS_DSPR2; +#if defined(HAS_INTERPOLATEROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + InterpolateRow = InterpolateRow_Any_DSPR2; if (IS_ALIGNED(dst_width, 4)) { - InterpolateRow = InterpolateRow_MIPS_DSPR2; + InterpolateRow = InterpolateRow_DSPR2; } } #endif @@ -1227,11 +1226,11 @@ void ScalePlaneBilinearUp_16(int src_width, int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_16_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2)) { - InterpolateRow = InterpolateRow_Any_16_MIPS_DSPR2; +#if defined(HAS_INTERPOLATEROW_16_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2)) { + InterpolateRow = InterpolateRow_Any_16_DSPR2; if (IS_ALIGNED(dst_width, 4)) { - InterpolateRow = InterpolateRow_16_MIPS_DSPR2; + InterpolateRow = InterpolateRow_16_DSPR2; } } #endif diff --git a/chromium/third_party/libyuv/source/scale_argb.cc b/chromium/third_party/libyuv/source/scale_argb.cc index adddf9db5a4..17f51ae9bf8 100644 --- a/chromium/third_party/libyuv/source/scale_argb.cc +++ b/chromium/third_party/libyuv/source/scale_argb.cc @@ -234,12 +234,12 @@ static void ScaleARGBBilinearDown(int src_width, int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_INTERPOLATEROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4)) { - InterpolateRow = InterpolateRow_Any_MIPS_DSPR2; + InterpolateRow = InterpolateRow_Any_DSPR2; if (IS_ALIGNED(clip_src_width, 4)) { - InterpolateRow = InterpolateRow_MIPS_DSPR2; + InterpolateRow = InterpolateRow_DSPR2; } } #endif @@ -324,10 +324,10 @@ static void ScaleARGBBilinearUp(int src_width, int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_INTERPOLATEROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) { - InterpolateRow = InterpolateRow_MIPS_DSPR2; + InterpolateRow = InterpolateRow_DSPR2; } #endif if (src_width >= 32768) { @@ -465,13 +465,13 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height, } } #endif -#if defined(HAS_I422TOARGBROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(src_width, 4) && +#if defined(HAS_I422TOARGBROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_width, 4) && IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) && IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) && IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) { - I422ToARGBRow = I422ToARGBRow_MIPS_DSPR2; + I422ToARGBRow = I422ToARGBRow_DSPR2; } #endif @@ -502,10 +502,10 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_INTERPOLATEROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) { - InterpolateRow = InterpolateRow_MIPS_DSPR2; + InterpolateRow = InterpolateRow_DSPR2; } #endif @@ -835,7 +835,6 @@ int YUVToARGBScaleClip(const uint8* src_y, int src_stride_y, int dst_width, int dst_height, int clip_x, int clip_y, int clip_width, int clip_height, enum FilterMode filtering) { - uint8* argb_buffer = (uint8*)malloc(src_width * src_height * 4); int r; I420ToARGB(src_y, src_stride_y, diff --git a/chromium/third_party/libyuv/source/scale_common.cc b/chromium/third_party/libyuv/source/scale_common.cc index 30ff18c5075..d3992df2e6a 100644 --- a/chromium/third_party/libyuv/source/scale_common.cc +++ b/chromium/third_party/libyuv/source/scale_common.cc @@ -922,13 +922,13 @@ void ScalePlaneVertical(int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_INTERPOLATEROW_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) { - InterpolateRow = InterpolateRow_Any_MIPS_DSPR2; + InterpolateRow = InterpolateRow_Any_DSPR2; if (IS_ALIGNED(dst_width_bytes, 4)) { - InterpolateRow = InterpolateRow_MIPS_DSPR2; + InterpolateRow = InterpolateRow_DSPR2; } } #endif @@ -996,13 +996,13 @@ void ScalePlaneVertical_16(int src_height, } } #endif -#if defined(HAS_INTERPOLATEROW_16_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && +#if defined(HAS_INTERPOLATEROW_16_DSPR2) + if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) { - InterpolateRow = InterpolateRow_Any_16_MIPS_DSPR2; + InterpolateRow = InterpolateRow_Any_16_DSPR2; if (IS_ALIGNED(dst_width_bytes, 4)) { - InterpolateRow = InterpolateRow_16_MIPS_DSPR2; + InterpolateRow = InterpolateRow_16_DSPR2; } } #endif diff --git a/chromium/third_party/libyuv/source/scale_mips.cc b/chromium/third_party/libyuv/source/scale_mips.cc index 2298a74b956..ae953073fa8 100644 --- a/chromium/third_party/libyuv/source/scale_mips.cc +++ b/chromium/third_party/libyuv/source/scale_mips.cc @@ -21,8 +21,8 @@ extern "C" { defined(__mips_dsp) && (__mips_dsp_rev >= 2) && \ (_MIPS_SIM == _MIPS_SIM_ABI32) -void ScaleRowDown2_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* dst, int dst_width) { +void ScaleRowDown2_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* dst, int dst_width) { __asm__ __volatile__( ".set push \n" ".set noreorder \n" @@ -77,8 +77,8 @@ void ScaleRowDown2_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown2Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* dst, int dst_width) { +void ScaleRowDown2Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* dst, int dst_width) { const uint8* t = src_ptr + src_stride; __asm__ __volatile__ ( @@ -176,8 +176,8 @@ void ScaleRowDown2Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown4_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* dst, int dst_width) { +void ScaleRowDown4_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* dst, int dst_width) { __asm__ __volatile__ ( ".set push \n" ".set noreorder \n" @@ -231,8 +231,8 @@ void ScaleRowDown4_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown4Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* dst, int dst_width) { +void ScaleRowDown4Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* dst, int dst_width) { intptr_t stride = src_stride; const uint8* s1 = src_ptr + stride; const uint8* s2 = s1 + stride; @@ -310,8 +310,8 @@ void ScaleRowDown4Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown34_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* dst, int dst_width) { +void ScaleRowDown34_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* dst, int dst_width) { __asm__ __volatile__ ( ".set push \n" ".set noreorder \n" @@ -356,8 +356,8 @@ void ScaleRowDown34_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown34_0_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* d, int dst_width) { +void ScaleRowDown34_0_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* d, int dst_width) { __asm__ __volatile__ ( ".set push \n" ".set noreorder \n" @@ -412,8 +412,8 @@ void ScaleRowDown34_0_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown34_1_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* d, int dst_width) { +void ScaleRowDown34_1_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* d, int dst_width) { __asm__ __volatile__ ( ".set push \n" ".set noreorder \n" @@ -464,8 +464,8 @@ void ScaleRowDown34_1_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown38_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* dst, int dst_width) { +void ScaleRowDown38_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* dst, int dst_width) { __asm__ __volatile__ ( ".set push \n" ".set noreorder \n" @@ -510,8 +510,8 @@ void ScaleRowDown38_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown38_2_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, - uint8* dst_ptr, int dst_width) { +void ScaleRowDown38_2_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, + uint8* dst_ptr, int dst_width) { intptr_t stride = src_stride; const uint8* t = src_ptr + stride; const int c = 0x2AAA; @@ -563,9 +563,9 @@ void ScaleRowDown38_2_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride, ); } -void ScaleRowDown38_3_Box_MIPS_DSPR2(const uint8* src_ptr, - ptrdiff_t src_stride, - uint8* dst_ptr, int dst_width) { +void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr, + ptrdiff_t src_stride, + uint8* dst_ptr, int dst_width) { intptr_t stride = src_stride; const uint8* s1 = src_ptr + stride; stride += stride; diff --git a/chromium/third_party/libyuv/source/scale_win.cc b/chromium/third_party/libyuv/source/scale_win.cc index 5ab4fa0ccc2..21b1ed923fa 100644 --- a/chromium/third_party/libyuv/source/scale_win.cc +++ b/chromium/third_party/libyuv/source/scale_win.cc @@ -289,7 +289,7 @@ void ScaleRowDown2Box_AVX2(const uint8* src_ptr, ptrdiff_t src_stride, vpmaddubsw ymm3, ymm3, ymm4 vpaddw ymm0, ymm0, ymm2 // vertical add vpaddw ymm1, ymm1, ymm3 - vpsrlw ymm0, ymm0, 1 + vpsrlw ymm0, ymm0, 1 // (x + 2) / 4 = (x / 2 + 1) / 2 vpsrlw ymm1, ymm1, 1 vpavgw ymm0, ymm0, ymm5 // (x + 1) / 2 vpavgw ymm1, ymm1, ymm5 |