diff --git a/include/libyuv/row.h b/include/libyuv/row.h index 885f8de38..f5abf1eba 100644 --- a/include/libyuv/row.h +++ b/include/libyuv/row.h @@ -655,13 +655,6 @@ void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int pix); void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix); void RGB24ToYRow_SSSE3(const uint8* src_rgb24, uint8* dst_y, int pix); void RAWToYRow_SSSE3(const uint8* src_raw, uint8* dst_y, int pix); -void ARGBToYRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix); -void ARGBToYJRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix); -void BGRAToYRow_Unaligned_SSSE3(const uint8* src_bgra, uint8* dst_y, int pix); -void ABGRToYRow_Unaligned_SSSE3(const uint8* src_abgr, uint8* dst_y, int pix); -void RGBAToYRow_Unaligned_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix); -void RGB24ToYRow_Unaligned_SSSE3(const uint8* src_rgb24, uint8* dst_y, int pix); -void RAWToYRow_Unaligned_SSSE3(const uint8* src_raw, uint8* dst_y, int pix); void ARGBToYRow_NEON(const uint8* src_argb, uint8* dst_y, int pix); void ARGBToYJRow_NEON(const uint8* src_argb, uint8* dst_y, int pix); void ARGBToUV444Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v, @@ -740,16 +733,6 @@ void ABGRToUVRow_SSSE3(const uint8* src_abgr, int src_stride_abgr, uint8* dst_u, uint8* dst_v, int width); void RGBAToUVRow_SSSE3(const uint8* src_rgba, int src_stride_rgba, uint8* dst_u, uint8* dst_v, int width); -void ARGBToUVRow_Unaligned_SSSE3(const uint8* src_argb, int src_stride_argb, - uint8* dst_u, uint8* dst_v, int width); -void ARGBToUVJRow_Unaligned_SSSE3(const uint8* src_argb, int src_stride_argb, - uint8* dst_u, uint8* dst_v, int width); -void BGRAToUVRow_Unaligned_SSSE3(const uint8* src_bgra, int src_stride_bgra, - uint8* dst_u, uint8* dst_v, int width); -void ABGRToUVRow_Unaligned_SSSE3(const uint8* src_abgr, int src_stride_abgr, - uint8* dst_u, uint8* dst_v, int width); -void RGBAToUVRow_Unaligned_SSSE3(const uint8* src_rgba, int src_stride_rgba, - uint8* dst_u, uint8* dst_v, int width); void ARGBToUVRow_Any_SSSE3(const uint8* src_argb, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width); void ARGBToUVJRow_Any_SSSE3(const uint8* src_argb, int src_stride_argb, @@ -811,15 +794,11 @@ void ARGB4444ToUVRow_C(const uint8* src_argb4444, int src_stride_argb4444, void ARGBToUV444Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width); -void ARGBToUV444Row_Unaligned_SSSE3(const uint8* src_argb, - uint8* dst_u, uint8* dst_v, int width); void ARGBToUV444Row_Any_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width); void ARGBToUV422Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width); -void ARGBToUV422Row_Unaligned_SSSE3(const uint8* src_argb, - uint8* dst_u, uint8* dst_v, int width); void ARGBToUV422Row_Any_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width); @@ -857,10 +836,6 @@ void SplitUVRow_AVX2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix); void SplitUVRow_NEON(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix); void SplitUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix); -void SplitUVRow_Unaligned_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, - int pix); -void SplitUVRow_Unaligned_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, - uint8* dst_v, int pix); void SplitUVRow_Any_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix); void SplitUVRow_Any_AVX2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, @@ -878,8 +853,6 @@ void MergeUVRow_AVX2(const uint8* src_u, const uint8* src_v, uint8* dst_uv, int width); void MergeUVRow_NEON(const uint8* src_u, const uint8* src_v, uint8* dst_uv, int width); -void MergeUVRow_Unaligned_SSE2(const uint8* src_u, const uint8* src_v, - uint8* dst_uv, int width); void MergeUVRow_Any_SSE2(const uint8* src_u, const uint8* src_v, uint8* dst_uv, int width); void MergeUVRow_Any_AVX2(const uint8* src_u, const uint8* src_v, uint8* dst_uv, @@ -926,8 +899,6 @@ void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb, const uint8* shuffler, int pix); void ARGBShuffleRow_NEON(const uint8* src_argb, uint8* dst_argb, const uint8* shuffler, int pix); -void ARGBShuffleRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_argb, - const uint8* shuffler, int pix); void ARGBShuffleRow_Any_SSE2(const uint8* src_argb, uint8* dst_argb, const uint8* shuffler, int pix); void ARGBShuffleRow_Any_SSSE3(const uint8* src_argb, uint8* dst_argb, @@ -994,7 +965,6 @@ void ARGBToARGB1555Row_C(const uint8* src_argb, uint8* dst_rgb, int pix); void ARGBToARGB4444Row_C(const uint8* src_argb, uint8* dst_rgb, int pix); void I400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix); -void I400ToARGBRow_Unaligned_SSE2(const uint8* src_y, uint8* dst_argb, int pix); void I400ToARGBRow_NEON(const uint8* src_y, uint8* dst_argb, int pix); void I400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int pix); void I400ToARGBRow_Any_SSE2(const uint8* src_y, uint8* dst_argb, int pix); @@ -1152,7 +1122,6 @@ void I422ToRGB565Row_SSSE3(const uint8* src_y, const uint8* src_v, uint8* dst_argb, int width); -// RGB24/RAW are unaligned. void I422ToRGB24Row_SSSE3(const uint8* src_y, const uint8* src_u, const uint8* src_v, @@ -1163,51 +1132,6 @@ void I422ToRAWRow_SSSE3(const uint8* src_y, const uint8* src_v, uint8* dst_raw, int width); - -void I444ToARGBRow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_u, - const uint8* src_v, - uint8* dst_argb, - int width); -void I422ToARGBRow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_u, - const uint8* src_v, - uint8* dst_argb, - int width); -void I411ToARGBRow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_u, - const uint8* src_v, - uint8* dst_argb, - int width); -void NV12ToARGBRow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_uv, - uint8* dst_argb, - int width); -void NV21ToARGBRow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_vu, - uint8* dst_argb, - int width); -void YUY2ToARGBRow_Unaligned_SSSE3(const uint8* src_yuy2, - uint8* dst_argb, - int width); -void UYVYToARGBRow_Unaligned_SSSE3(const uint8* src_uyvy, - uint8* dst_argb, - int width); -void I422ToBGRARow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_u, - const uint8* src_v, - uint8* dst_bgra, - int width); -void I422ToABGRRow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_u, - const uint8* src_v, - uint8* dst_abgr, - int width); -void I422ToRGBARow_Unaligned_SSSE3(const uint8* src_y, - const uint8* src_u, - const uint8* src_v, - uint8* dst_rgba, - int width); void I422ToARGBRow_Any_AVX2(const uint8* src_y, const uint8* src_u, const uint8* src_v, @@ -1280,7 +1204,6 @@ void I422ToRGB565Row_Any_SSSE3(const uint8* src_y, const uint8* src_v, uint8* dst_rgba, int width); -// RGB24/RAW are unaligned. void I422ToRGB24Row_Any_SSSE3(const uint8* src_y, const uint8* src_u, const uint8* src_v, @@ -1494,12 +1417,6 @@ void YUY2ToUVRow_SSE2(const uint8* src_yuy2, int stride_yuy2, uint8* dst_u, uint8* dst_v, int pix); void YUY2ToUV422Row_SSE2(const uint8* src_yuy2, uint8* dst_u, uint8* dst_v, int pix); -void YUY2ToYRow_Unaligned_SSE2(const uint8* src_yuy2, - uint8* dst_y, int pix); -void YUY2ToUVRow_Unaligned_SSE2(const uint8* src_yuy2, int stride_yuy2, - uint8* dst_u, uint8* dst_v, int pix); -void YUY2ToUV422Row_Unaligned_SSE2(const uint8* src_yuy2, - uint8* dst_u, uint8* dst_v, int pix); void YUY2ToYRow_NEON(const uint8* src_yuy2, uint8* dst_y, int pix); void YUY2ToUVRow_NEON(const uint8* src_yuy2, int stride_yuy2, uint8* dst_u, uint8* dst_v, int pix); @@ -1535,12 +1452,6 @@ void UYVYToUVRow_SSE2(const uint8* src_uyvy, int stride_uyvy, uint8* dst_u, uint8* dst_v, int pix); void UYVYToUV422Row_SSE2(const uint8* src_uyvy, uint8* dst_u, uint8* dst_v, int pix); -void UYVYToYRow_Unaligned_SSE2(const uint8* src_uyvy, - uint8* dst_y, int pix); -void UYVYToUVRow_Unaligned_SSE2(const uint8* src_uyvy, int stride_uyvy, - uint8* dst_u, uint8* dst_v, int pix); -void UYVYToUV422Row_Unaligned_SSE2(const uint8* src_uyvy, - uint8* dst_u, uint8* dst_v, int pix); void UYVYToYRow_AVX2(const uint8* src_uyvy, uint8* dst_y, int pix); void UYVYToUVRow_AVX2(const uint8* src_uyvy, int stride_uyvy, uint8* dst_u, uint8* dst_v, int pix); diff --git a/source/convert.cc b/source/convert.cc index f20514338..19ad65926 100644 --- a/source/convert.cc +++ b/source/convert.cc @@ -194,10 +194,7 @@ static void CopyPlane2(const uint8* src, int src_stride_0, int src_stride_1, } #endif #if defined(HAS_COPYROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32) && - IS_ALIGNED(src, 16) && - IS_ALIGNED(src_stride_0, 16) && IS_ALIGNED(src_stride_1, 16) && - IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32)) { CopyRow = CopyRow_SSE2; } #endif @@ -291,12 +288,7 @@ static int X420ToI420(const uint8* src_y, if (TestCpuFlag(kCpuHasSSE2) && halfwidth >= 16) { SplitUVRow = SplitUVRow_Any_SSE2; if (IS_ALIGNED(halfwidth, 16)) { - SplitUVRow = SplitUVRow_Unaligned_SSE2; - if (IS_ALIGNED(src_uv, 16) && IS_ALIGNED(src_stride_uv, 16) && - IS_ALIGNED(dst_u, 16) && IS_ALIGNED(dst_stride_u, 16) && - IS_ALIGNED(dst_v, 16) && IS_ALIGNED(dst_stride_v, 16)) { - SplitUVRow = SplitUVRow_SSE2; - } + SplitUVRow = SplitUVRow_SSE2; } } #endif @@ -317,15 +309,13 @@ static int X420ToI420(const uint8* src_y, } #endif #if defined(HAS_SPLITUVROW_MIPS_DSPR2) - if (TestCpuFlag(kCpuHasMIPS_DSPR2) && halfwidth >= 16) { + if (TestCpuFlag(kCpuHasMIPS_DSPR2) && halfwidth >= 16 && + 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; if (IS_ALIGNED(halfwidth, 16)) { - SplitUVRow = SplitUVRow_Unaligned_MIPS_DSPR2; - if (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_MIPS_DSPR2; - } + SplitUVRow = SplitUVRow_MIPS_DSPR2; } } #endif @@ -440,9 +430,7 @@ int Q420ToI420(const uint8* src_y, int src_stride_y, } #endif #if defined(HAS_COPYROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32) && - IS_ALIGNED(src_y, 16) && IS_ALIGNED(src_stride_y, 16) && - IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32)) { CopyRow = CopyRow_SSE2; } #endif @@ -467,14 +455,8 @@ int Q420ToI420(const uint8* src_y, int src_stride_y, YUY2ToUV422Row = YUY2ToUV422Row_Any_SSE2; YUY2ToYRow = YUY2ToYRow_Any_SSE2; if (IS_ALIGNED(width, 16)) { - YUY2ToUV422Row = YUY2ToUV422Row_Unaligned_SSE2; - YUY2ToYRow = YUY2ToYRow_Unaligned_SSE2; - if (IS_ALIGNED(src_yuy2, 16) && IS_ALIGNED(src_stride_yuy2, 16)) { - YUY2ToUV422Row = YUY2ToUV422Row_SSE2; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - YUY2ToYRow = YUY2ToYRow_SSE2; - } - } + YUY2ToUV422Row = YUY2ToUV422Row_SSE2; + YUY2ToYRow = YUY2ToYRow_SSE2; } } #endif @@ -543,14 +525,8 @@ int YUY2ToI420(const uint8* src_yuy2, int src_stride_yuy2, YUY2ToUVRow = YUY2ToUVRow_Any_SSE2; YUY2ToYRow = YUY2ToYRow_Any_SSE2; if (IS_ALIGNED(width, 16)) { - YUY2ToUVRow = YUY2ToUVRow_Unaligned_SSE2; - YUY2ToYRow = YUY2ToYRow_Unaligned_SSE2; - if (IS_ALIGNED(src_yuy2, 16) && IS_ALIGNED(src_stride_yuy2, 16)) { - YUY2ToUVRow = YUY2ToUVRow_SSE2; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - YUY2ToYRow = YUY2ToYRow_SSE2; - } - } + YUY2ToUVRow = YUY2ToUVRow_SSE2; + YUY2ToYRow = YUY2ToYRow_SSE2; } } #endif @@ -616,14 +592,8 @@ int UYVYToI420(const uint8* src_uyvy, int src_stride_uyvy, UYVYToUVRow = UYVYToUVRow_Any_SSE2; UYVYToYRow = UYVYToYRow_Any_SSE2; if (IS_ALIGNED(width, 16)) { - UYVYToUVRow = UYVYToUVRow_Unaligned_SSE2; - UYVYToYRow = UYVYToYRow_Unaligned_SSE2; - if (IS_ALIGNED(src_uyvy, 16) && IS_ALIGNED(src_stride_uyvy, 16)) { - UYVYToUVRow = UYVYToUVRow_SSE2; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - UYVYToYRow = UYVYToYRow_SSE2; - } - } + UYVYToUVRow = UYVYToUVRow_SSE2; + UYVYToYRow = UYVYToYRow_SSE2; } } #endif @@ -694,14 +664,8 @@ int ARGBToI420(const uint8* src_argb, int src_stride_argb, ARGBToUVRow = ARGBToUVRow_Any_SSSE3; ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUVRow = ARGBToUVRow_Unaligned_SSSE3; - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUVRow = ARGBToUVRow_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } - } + ARGBToUVRow = ARGBToUVRow_SSSE3; + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif @@ -776,14 +740,8 @@ int BGRAToI420(const uint8* src_bgra, int src_stride_bgra, BGRAToUVRow = BGRAToUVRow_Any_SSSE3; BGRAToYRow = BGRAToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - BGRAToUVRow = BGRAToUVRow_Unaligned_SSSE3; - BGRAToYRow = BGRAToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_bgra, 16) && IS_ALIGNED(src_stride_bgra, 16)) { - BGRAToUVRow = BGRAToUVRow_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - BGRAToYRow = BGRAToYRow_SSSE3; - } - } + BGRAToUVRow = BGRAToUVRow_SSSE3; + BGRAToYRow = BGRAToYRow_SSSE3; } } #elif defined(HAS_BGRATOYROW_NEON) @@ -847,14 +805,8 @@ int ABGRToI420(const uint8* src_abgr, int src_stride_abgr, ABGRToUVRow = ABGRToUVRow_Any_SSSE3; ABGRToYRow = ABGRToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ABGRToUVRow = ABGRToUVRow_Unaligned_SSSE3; - ABGRToYRow = ABGRToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_abgr, 16) && IS_ALIGNED(src_stride_abgr, 16)) { - ABGRToUVRow = ABGRToUVRow_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ABGRToYRow = ABGRToYRow_SSSE3; - } - } + ABGRToUVRow = ABGRToUVRow_SSSE3; + ABGRToYRow = ABGRToYRow_SSSE3; } } #elif defined(HAS_ABGRTOYROW_NEON) @@ -918,14 +870,8 @@ int RGBAToI420(const uint8* src_rgba, int src_stride_rgba, RGBAToUVRow = RGBAToUVRow_Any_SSSE3; RGBAToYRow = RGBAToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - RGBAToUVRow = RGBAToUVRow_Unaligned_SSSE3; - RGBAToYRow = RGBAToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_rgba, 16) && IS_ALIGNED(src_stride_rgba, 16)) { - RGBAToUVRow = RGBAToUVRow_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - RGBAToYRow = RGBAToYRow_SSSE3; - } - } + RGBAToUVRow = RGBAToUVRow_SSSE3; + RGBAToYRow = RGBAToYRow_SSSE3; } } #elif defined(HAS_RGBATOYROW_NEON) @@ -1029,10 +975,7 @@ int RGB24ToI420(const uint8* src_rgb24, int src_stride_rgb24, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif // HAS_ARGBTOUVROW_SSSE3 @@ -1146,10 +1089,7 @@ int RAWToI420(const uint8* src_raw, int src_stride_raw, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif // HAS_ARGBTOUVROW_SSSE3 @@ -1260,10 +1200,7 @@ int RGB565ToI420(const uint8* src_rgb565, int src_stride_rgb565, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif // HAS_ARGBTOUVROW_SSSE3 @@ -1377,10 +1314,7 @@ int ARGB1555ToI420(const uint8* src_argb1555, int src_stride_argb1555, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif // HAS_ARGBTOUVROW_SSSE3 @@ -1495,10 +1429,7 @@ int ARGB4444ToI420(const uint8* src_argb4444, int src_stride_argb4444, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif // HAS_ARGBTOUVROW_SSSE3 diff --git a/source/convert_argb.cc b/source/convert_argb.cc index ac0bc3d15..3d8250c3f 100644 --- a/source/convert_argb.cc +++ b/source/convert_argb.cc @@ -82,10 +82,7 @@ int I444ToARGB(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { I444ToARGBRow = I444ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - I444ToARGBRow = I444ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - I444ToARGBRow = I444ToARGBRow_SSSE3; - } + I444ToARGBRow = I444ToARGBRow_SSSE3; } } #elif defined(HAS_I444TOARGBROW_NEON) @@ -144,10 +141,7 @@ int I422ToARGB(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { I422ToARGBRow = I422ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - I422ToARGBRow = I422ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - I422ToARGBRow = I422ToARGBRow_SSSE3; - } + I422ToARGBRow = I422ToARGBRow_SSSE3; } } #endif @@ -224,10 +218,7 @@ int I411ToARGB(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { I411ToARGBRow = I411ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - I411ToARGBRow = I411ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - I411ToARGBRow = I411ToARGBRow_SSSE3; - } + I411ToARGBRow = I411ToARGBRow_SSSE3; } } #elif defined(HAS_I411TOARGBROW_NEON) @@ -276,8 +267,7 @@ int I400ToARGB_Reference(const uint8* src_y, int src_stride_y, src_stride_y = dst_stride_argb = 0; } #if defined(HAS_YTOARGBROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && width >= 8 && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && width >= 8) { YToARGBRow = YToARGBRow_Any_SSE2; if (IS_ALIGNED(width, 8)) { YToARGBRow = YToARGBRow_SSE2; @@ -329,10 +319,7 @@ int I400ToARGB(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSE2) && width >= 8) { I400ToARGBRow = I400ToARGBRow_Any_SSE2; if (IS_ALIGNED(width, 8)) { - I400ToARGBRow = I400ToARGBRow_Unaligned_SSE2; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - I400ToARGBRow = I400ToARGBRow_SSE2; - } + I400ToARGBRow = I400ToARGBRow_SSE2; } } #elif defined(HAS_I400TOARGBROW_NEON) @@ -447,8 +434,7 @@ int RGB24ToARGB(const uint8* src_rgb24, int src_stride_rgb24, src_stride_rgb24 = dst_stride_argb = 0; } #if defined(HAS_RGB24TOARGBROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && width >= 16 && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { RGB24ToARGBRow = RGB24ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { RGB24ToARGBRow = RGB24ToARGBRow_SSSE3; @@ -497,8 +483,7 @@ int RAWToARGB(const uint8* src_raw, int src_stride_raw, src_stride_raw = dst_stride_argb = 0; } #if defined(HAS_RAWTOARGBROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && width >= 16 && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { RAWToARGBRow = RAWToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { RAWToARGBRow = RAWToARGBRow_SSSE3; @@ -547,8 +532,7 @@ int RGB565ToARGB(const uint8* src_rgb565, int src_stride_rgb565, src_stride_rgb565 = dst_stride_argb = 0; } #if defined(HAS_RGB565TOARGBROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && width >= 8 && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && width >= 8) { RGB565ToARGBRow = RGB565ToARGBRow_Any_SSE2; if (IS_ALIGNED(width, 8)) { RGB565ToARGBRow = RGB565ToARGBRow_SSE2; @@ -597,8 +581,7 @@ int ARGB1555ToARGB(const uint8* src_argb1555, int src_stride_argb1555, src_stride_argb1555 = dst_stride_argb = 0; } #if defined(HAS_ARGB1555TOARGBROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && width >= 8 && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && width >= 8) { ARGB1555ToARGBRow = ARGB1555ToARGBRow_Any_SSE2; if (IS_ALIGNED(width, 8)) { ARGB1555ToARGBRow = ARGB1555ToARGBRow_SSE2; @@ -647,8 +630,7 @@ int ARGB4444ToARGB(const uint8* src_argb4444, int src_stride_argb4444, src_stride_argb4444 = dst_stride_argb = 0; } #if defined(HAS_ARGB4444TOARGBROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && width >= 8 && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && width >= 8) { ARGB4444ToARGBRow = ARGB4444ToARGBRow_Any_SSE2; if (IS_ALIGNED(width, 8)) { ARGB4444ToARGBRow = ARGB4444ToARGBRow_SSE2; @@ -696,10 +678,7 @@ int NV12ToARGB(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { NV12ToARGBRow = NV12ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - NV12ToARGBRow = NV12ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - NV12ToARGBRow = NV12ToARGBRow_SSSE3; - } + NV12ToARGBRow = NV12ToARGBRow_SSSE3; } } #elif defined(HAS_NV12TOARGBROW_NEON) @@ -747,10 +726,7 @@ int NV21ToARGB(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { NV21ToARGBRow = NV21ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - NV21ToARGBRow = NV21ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - NV21ToARGBRow = NV21ToARGBRow_SSSE3; - } + NV21ToARGBRow = NV21ToARGBRow_SSSE3; } } #endif @@ -798,10 +774,7 @@ int M420ToARGB(const uint8* src_m420, int src_stride_m420, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { NV12ToARGBRow = NV12ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - NV12ToARGBRow = NV12ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - NV12ToARGBRow = NV12ToARGBRow_SSSE3; - } + NV12ToARGBRow = NV12ToARGBRow_SSSE3; } } #elif defined(HAS_NV12TOARGBROW_NEON) @@ -856,11 +829,7 @@ int YUY2ToARGB(const uint8* src_yuy2, int src_stride_yuy2, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { YUY2ToARGBRow = YUY2ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - YUY2ToARGBRow = YUY2ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_yuy2, 16) && IS_ALIGNED(src_stride_yuy2, 16) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - YUY2ToARGBRow = YUY2ToARGBRow_SSSE3; - } + YUY2ToARGBRow = YUY2ToARGBRow_SSSE3; } } #elif defined(HAS_YUY2TOARGBROW_NEON) @@ -909,11 +878,7 @@ int UYVYToARGB(const uint8* src_uyvy, int src_stride_uyvy, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { UYVYToARGBRow = UYVYToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - UYVYToARGBRow = UYVYToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_uyvy, 16) && IS_ALIGNED(src_stride_uyvy, 16) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - UYVYToARGBRow = UYVYToARGBRow_SSSE3; - } + UYVYToARGBRow = UYVYToARGBRow_SSSE3; } } #elif defined(HAS_UYVYTOARGBROW_NEON) diff --git a/source/convert_from.cc b/source/convert_from.cc index c1a2f62f0..068f14c4e 100644 --- a/source/convert_from.cc +++ b/source/convert_from.cc @@ -400,12 +400,7 @@ int I420ToNV12(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSE2) && halfwidth >= 16) { MergeUVRow_ = MergeUVRow_Any_SSE2; if (IS_ALIGNED(halfwidth, 16)) { - MergeUVRow_ = MergeUVRow_Unaligned_SSE2; - if (IS_ALIGNED(src_u, 16) && IS_ALIGNED(src_stride_u, 16) && - IS_ALIGNED(src_v, 16) && IS_ALIGNED(src_stride_v, 16) && - IS_ALIGNED(dst_uv, 16) && IS_ALIGNED(dst_stride_uv, 16)) { - MergeUVRow_ = MergeUVRow_SSE2; - } + MergeUVRow_ = MergeUVRow_SSE2; } } #endif @@ -479,10 +474,7 @@ int I420ToARGB(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { I422ToARGBRow = I422ToARGBRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - I422ToARGBRow = I422ToARGBRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { - I422ToARGBRow = I422ToARGBRow_SSSE3; - } + I422ToARGBRow = I422ToARGBRow_SSSE3; } } #endif @@ -551,10 +543,7 @@ int I420ToBGRA(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { I422ToBGRARow = I422ToBGRARow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - I422ToBGRARow = I422ToBGRARow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_bgra, 16) && IS_ALIGNED(dst_stride_bgra, 16)) { - I422ToBGRARow = I422ToBGRARow_SSSE3; - } + I422ToBGRARow = I422ToBGRARow_SSSE3; } } #elif defined(HAS_I422TOBGRAROW_NEON) @@ -613,10 +602,7 @@ int I420ToABGR(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { I422ToABGRRow = I422ToABGRRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - I422ToABGRRow = I422ToABGRRow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_abgr, 16) && IS_ALIGNED(dst_stride_abgr, 16)) { - I422ToABGRRow = I422ToABGRRow_SSSE3; - } + I422ToABGRRow = I422ToABGRRow_SSSE3; } } #elif defined(HAS_I422TOABGRROW_NEON) @@ -667,10 +653,7 @@ int I420ToRGBA(const uint8* src_y, int src_stride_y, if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { I422ToRGBARow = I422ToRGBARow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { - I422ToRGBARow = I422ToRGBARow_Unaligned_SSSE3; - if (IS_ALIGNED(dst_rgba, 16) && IS_ALIGNED(dst_stride_rgba, 16)) { - I422ToRGBARow = I422ToRGBARow_SSSE3; - } + I422ToRGBARow = I422ToRGBARow_SSSE3; } } #elif defined(HAS_I422TORGBAROW_NEON) diff --git a/source/convert_from_argb.cc b/source/convert_from_argb.cc index de461ddb0..b0f9b1eff 100644 --- a/source/convert_from_argb.cc +++ b/source/convert_from_argb.cc @@ -54,10 +54,7 @@ int ARGBToI444(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToUV444Row = ARGBToUV444Row_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUV444Row = ARGBToUV444Row_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUV444Row = ARGBToUV444Row_SSSE3; - } + ARGBToUV444Row = ARGBToUV444Row_SSSE3; } } #elif defined(HAS_ARGBTOUV444ROW_NEON) @@ -72,11 +69,7 @@ int ARGBToI444(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) && - IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } @@ -133,10 +126,7 @@ int ARGBToI422(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToUV422Row = ARGBToUV422Row_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUV422Row = ARGBToUV422Row_SSSE3; - } + ARGBToUV422Row = ARGBToUV422Row_SSSE3; } } #elif defined(HAS_ARGBTOUV422ROW_NEON) @@ -152,11 +142,7 @@ int ARGBToI422(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) && - IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #elif defined(HAS_ARGBTOYROW_NEON) @@ -212,11 +198,7 @@ int ARGBToI411(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) && - IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif @@ -285,14 +267,8 @@ int ARGBToNV12(const uint8* src_argb, int src_stride_argb, ARGBToUVRow = ARGBToUVRow_Any_SSSE3; ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUVRow = ARGBToUVRow_Unaligned_SSSE3; - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUVRow = ARGBToUVRow_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } - } + ARGBToUVRow = ARGBToUVRow_SSSE3; + ARGBToYRow = ARGBToYRow_SSSE3; } } #elif defined(HAS_ARGBTOYROW_NEON) @@ -315,10 +291,7 @@ int ARGBToNV12(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSE2) && halfwidth >= 16) { MergeUVRow_ = MergeUVRow_Any_SSE2; if (IS_ALIGNED(halfwidth, 16)) { - MergeUVRow_ = MergeUVRow_Unaligned_SSE2; - if (IS_ALIGNED(dst_uv, 16) && IS_ALIGNED(dst_stride_uv, 16)) { - MergeUVRow_ = MergeUVRow_SSE2; - } + MergeUVRow_ = MergeUVRow_SSE2; } } #endif @@ -392,14 +365,8 @@ int ARGBToNV21(const uint8* src_argb, int src_stride_argb, ARGBToUVRow = ARGBToUVRow_Any_SSSE3; ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUVRow = ARGBToUVRow_Unaligned_SSSE3; - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUVRow = ARGBToUVRow_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } - } + ARGBToUVRow = ARGBToUVRow_SSSE3; + ARGBToYRow = ARGBToYRow_SSSE3; } } #elif defined(HAS_ARGBTOYROW_NEON) @@ -422,10 +389,7 @@ int ARGBToNV21(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSE2) && halfwidth >= 16) { MergeUVRow_ = MergeUVRow_Any_SSE2; if (IS_ALIGNED(halfwidth, 16)) { - MergeUVRow_ = MergeUVRow_Unaligned_SSE2; - if (IS_ALIGNED(dst_uv, 16) && IS_ALIGNED(dst_stride_uv, 16)) { - MergeUVRow_ = MergeUVRow_SSE2; - } + MergeUVRow_ = MergeUVRow_SSE2; } } #endif @@ -503,10 +467,7 @@ int ARGBToYUY2(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToUV422Row = ARGBToUV422Row_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUV422Row = ARGBToUV422Row_SSSE3; - } + ARGBToUV422Row = ARGBToUV422Row_SSSE3; } } #elif defined(HAS_ARGBTOUV422ROW_NEON) @@ -521,10 +482,7 @@ int ARGBToYUY2(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #elif defined(HAS_ARGBTOYROW_NEON) @@ -605,10 +563,7 @@ int ARGBToUYVY(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToUV422Row = ARGBToUV422Row_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUV422Row = ARGBToUV422Row_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUV422Row = ARGBToUV422Row_SSSE3; - } + ARGBToUV422Row = ARGBToUV422Row_SSSE3; } } #elif defined(HAS_ARGBTOUV422ROW_NEON) @@ -623,10 +578,7 @@ int ARGBToUYVY(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #elif defined(HAS_ARGBTOYROW_NEON) @@ -700,11 +652,7 @@ int ARGBToI400(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) && - IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } + ARGBToYRow = ARGBToYRow_SSSE3; } } #endif @@ -867,8 +815,7 @@ int ARGBToRGB565(const uint8* src_argb, int src_stride_argb, src_stride_argb = dst_stride_rgb565 = 0; } #if defined(HAS_ARGBTORGB565ROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && width >= 4 && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && width >= 4) { ARGBToRGB565Row = ARGBToRGB565Row_Any_SSE2; if (IS_ALIGNED(width, 4)) { ARGBToRGB565Row = ARGBToRGB565Row_SSE2; @@ -915,8 +862,7 @@ int ARGBToARGB1555(const uint8* src_argb, int src_stride_argb, src_stride_argb = dst_stride_argb1555 = 0; } #if defined(HAS_ARGBTOARGB1555ROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && width >= 4 && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && width >= 4) { ARGBToARGB1555Row = ARGBToARGB1555Row_Any_SSE2; if (IS_ALIGNED(width, 4)) { ARGBToARGB1555Row = ARGBToARGB1555Row_SSE2; @@ -963,8 +909,7 @@ int ARGBToARGB4444(const uint8* src_argb, int src_stride_argb, src_stride_argb = dst_stride_argb4444 = 0; } #if defined(HAS_ARGBTOARGB4444ROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && width >= 4 && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && width >= 4) { ARGBToARGB4444Row = ARGBToARGB4444Row_Any_SSE2; if (IS_ALIGNED(width, 4)) { ARGBToARGB4444Row = ARGBToARGB4444Row_SSE2; @@ -1015,14 +960,8 @@ int ARGBToJ420(const uint8* src_argb, int src_stride_argb, ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3; ARGBToYJRow = ARGBToYJRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToUVJRow = ARGBToUVJRow_Unaligned_SSSE3; - ARGBToYJRow = ARGBToYJRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { - ARGBToUVJRow = ARGBToUVJRow_SSSE3; - if (IS_ALIGNED(dst_yj, 16) && IS_ALIGNED(dst_stride_yj, 16)) { - ARGBToYJRow = ARGBToYJRow_SSSE3; - } - } + ARGBToUVJRow = ARGBToUVJRow_SSSE3; + ARGBToYJRow = ARGBToYJRow_SSSE3; } } #endif @@ -1094,11 +1033,7 @@ int ARGBToJ400(const uint8* src_argb, int src_stride_argb, if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) { ARGBToYJRow = ARGBToYJRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYJRow = ARGBToYJRow_Unaligned_SSSE3; - if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) && - IS_ALIGNED(dst_yj, 16) && IS_ALIGNED(dst_stride_yj, 16)) { - ARGBToYJRow = ARGBToYJRow_SSSE3; - } + ARGBToYJRow = ARGBToYJRow_SSSE3; } } #endif diff --git a/source/format_conversion.cc b/source/format_conversion.cc index 3c1737153..53d15c095 100644 --- a/source/format_conversion.cc +++ b/source/format_conversion.cc @@ -77,8 +77,7 @@ int ARGBToBayer(const uint8* src_argb, int src_stride_argb, src_stride_argb = -src_stride_argb; } #if defined(HAS_ARGBTOBAYERROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && width >= 8 && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { ARGBToBayerRow = ARGBToBayerRow_Any_SSSE3; if (IS_ALIGNED(width, 8)) { ARGBToBayerRow = ARGBToBayerRow_SSSE3; @@ -319,11 +318,8 @@ int BayerToI420(const uint8* src_bayer, int src_stride_bayer, ARGBToUVRow = ARGBToUVRow_Any_SSSE3; ARGBToYRow = ARGBToYRow_Any_SSSE3; if (IS_ALIGNED(width, 16)) { - ARGBToYRow = ARGBToYRow_Unaligned_SSSE3; + ARGBToYRow = ARGBToYRow_SSSE3; ARGBToUVRow = ARGBToUVRow_SSSE3; - if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) { - ARGBToYRow = ARGBToYRow_SSSE3; - } } } #elif defined(HAS_ARGBTOYROW_NEON) diff --git a/source/rotate_argb.cc b/source/rotate_argb.cc index a8d7fc2e7..5951f57b2 100644 --- a/source/rotate_argb.cc +++ b/source/rotate_argb.cc @@ -50,13 +50,11 @@ static void ARGBTranspose(const uint8* src, int src_stride, void (*ScaleARGBRowDownEven)(const uint8* src_ptr, int src_stride, int src_step, uint8* dst_ptr, int dst_width) = ScaleARGBRowDownEven_C; #if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(height, 4) && // Width of dest. - IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(height, 4)) { // Width of dest. ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2; } #elif defined(HAS_SCALEARGBROWDOWNEVEN_NEON) - if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(height, 4) && // Width of dest. - IS_ALIGNED(src, 4)) { + if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(height, 4)) { // Width of dest. ScaleARGBRowDownEven = ScaleARGBRowDownEven_NEON; } #endif @@ -103,9 +101,7 @@ void ARGBRotate180(const uint8* src, int src_stride, ARGBMirrorRow_C; void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C; #if defined(HAS_ARGBMIRRORROW_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 4) && - IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 4)) { ARGBMirrorRow = ARGBMirrorRow_SSSE3; } #endif @@ -130,9 +126,7 @@ void ARGBRotate180(const uint8* src, int src_stride, } #endif #if defined(HAS_COPYROW_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width * 4, 32) && - IS_ALIGNED(src, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width * 4, 32)) { CopyRow = CopyRow_SSE2; } #endif diff --git a/source/row_any.cc b/source/row_any.cc index 31ab08a99..0d58f5df8 100644 --- a/source/row_any.cc +++ b/source/row_any.cc @@ -35,19 +35,19 @@ extern "C" { } #ifdef HAS_I422TOARGBROW_SSSE3 -YANY(I422ToARGBRow_Any_SSSE3, I422ToARGBRow_Unaligned_SSSE3, I422ToARGBRow_C, +YANY(I422ToARGBRow_Any_SSSE3, I422ToARGBRow_SSSE3, I422ToARGBRow_C, 1, 4, 7) #endif // HAS_I422TOARGBROW_SSSE3 #ifdef HAS_I444TOARGBROW_SSSE3 -YANY(I444ToARGBRow_Any_SSSE3, I444ToARGBRow_Unaligned_SSSE3, I444ToARGBRow_C, +YANY(I444ToARGBRow_Any_SSSE3, I444ToARGBRow_SSSE3, I444ToARGBRow_C, 0, 4, 7) -YANY(I411ToARGBRow_Any_SSSE3, I411ToARGBRow_Unaligned_SSSE3, I411ToARGBRow_C, +YANY(I411ToARGBRow_Any_SSSE3, I411ToARGBRow_SSSE3, I411ToARGBRow_C, 2, 4, 7) -YANY(I422ToBGRARow_Any_SSSE3, I422ToBGRARow_Unaligned_SSSE3, I422ToBGRARow_C, +YANY(I422ToBGRARow_Any_SSSE3, I422ToBGRARow_SSSE3, I422ToBGRARow_C, 1, 4, 7) -YANY(I422ToABGRRow_Any_SSSE3, I422ToABGRRow_Unaligned_SSSE3, I422ToABGRRow_C, +YANY(I422ToABGRRow_Any_SSSE3, I422ToABGRRow_SSSE3, I422ToABGRRow_C, 1, 4, 7) -YANY(I422ToRGBARow_Any_SSSE3, I422ToRGBARow_Unaligned_SSSE3, I422ToRGBARow_C, +YANY(I422ToRGBARow_Any_SSSE3, I422ToRGBARow_SSSE3, I422ToRGBARow_C, 1, 4, 7) // I422ToRGB565Row_SSSE3 is unaligned. YANY(I422ToARGB4444Row_Any_SSSE3, I422ToARGB4444Row_SSSE3, I422ToARGB4444Row_C, @@ -102,9 +102,9 @@ YANY(I422ToUYVYRow_Any_NEON, I422ToUYVYRow_NEON, I422ToUYVYRow_C, 1, 2, 15) } #ifdef HAS_NV12TOARGBROW_SSSE3 -NV2NY(NV12ToARGBRow_Any_SSSE3, NV12ToARGBRow_Unaligned_SSSE3, NV12ToARGBRow_C, +NV2NY(NV12ToARGBRow_Any_SSSE3, NV12ToARGBRow_SSSE3, NV12ToARGBRow_C, 0, 4) -NV2NY(NV21ToARGBRow_Any_SSSE3, NV21ToARGBRow_Unaligned_SSSE3, NV21ToARGBRow_C, +NV2NY(NV21ToARGBRow_Any_SSSE3, NV21ToARGBRow_SSSE3, NV21ToARGBRow_C, 0, 4) #endif // HAS_NV12TOARGBROW_SSSE3 #ifdef HAS_NV12TOARGBROW_NEON @@ -145,15 +145,15 @@ RGBANY(ARGBToARGB4444Row_Any_SSE2, ARGBToARGB4444Row_SSE2, ARGBToARGB4444Row_C, 3, 4, 2) #endif #if defined(HAS_I400TOARGBROW_SSE2) -RGBANY(I400ToARGBRow_Any_SSE2, I400ToARGBRow_Unaligned_SSE2, I400ToARGBRow_C, +RGBANY(I400ToARGBRow_Any_SSE2, I400ToARGBRow_SSE2, I400ToARGBRow_C, 7, 1, 4) #endif #if defined(HAS_YTOARGBROW_SSE2) RGBANY(YToARGBRow_Any_SSE2, YToARGBRow_SSE2, YToARGBRow_C, 7, 1, 4) -RGBANY(YUY2ToARGBRow_Any_SSSE3, YUY2ToARGBRow_Unaligned_SSSE3, YUY2ToARGBRow_C, +RGBANY(YUY2ToARGBRow_Any_SSSE3, YUY2ToARGBRow_SSSE3, YUY2ToARGBRow_C, 15, 2, 4) -RGBANY(UYVYToARGBRow_Any_SSSE3, UYVYToARGBRow_Unaligned_SSSE3, UYVYToARGBRow_C, +RGBANY(UYVYToARGBRow_Any_SSSE3, UYVYToARGBRow_SSSE3, UYVYToARGBRow_C, 15, 2, 4) // These require alignment on ARGB, so C is used for remainder. RGBANY(RGB24ToARGBRow_Any_SSSE3, RGB24ToARGBRow_SSSE3, RGB24ToARGBRow_C, @@ -231,17 +231,17 @@ YANY(YUY2ToYRow_Any_AVX2, YUY2ToYRow_AVX2, 2, 1, 32) YANY(UYVYToYRow_Any_AVX2, UYVYToYRow_AVX2, 2, 1, 32) #endif #ifdef HAS_ARGBTOYROW_SSSE3 -YANY(ARGBToYRow_Any_SSSE3, ARGBToYRow_Unaligned_SSSE3, 4, 1, 16) +YANY(ARGBToYRow_Any_SSSE3, ARGBToYRow_SSSE3, 4, 1, 16) #endif #ifdef HAS_BGRATOYROW_SSSE3 -YANY(BGRAToYRow_Any_SSSE3, BGRAToYRow_Unaligned_SSSE3, 4, 1, 16) -YANY(ABGRToYRow_Any_SSSE3, ABGRToYRow_Unaligned_SSSE3, 4, 1, 16) -YANY(RGBAToYRow_Any_SSSE3, RGBAToYRow_Unaligned_SSSE3, 4, 1, 16) -YANY(YUY2ToYRow_Any_SSE2, YUY2ToYRow_Unaligned_SSE2, 2, 1, 16) -YANY(UYVYToYRow_Any_SSE2, UYVYToYRow_Unaligned_SSE2, 2, 1, 16) +YANY(BGRAToYRow_Any_SSSE3, BGRAToYRow_SSSE3, 4, 1, 16) +YANY(ABGRToYRow_Any_SSSE3, ABGRToYRow_SSSE3, 4, 1, 16) +YANY(RGBAToYRow_Any_SSSE3, RGBAToYRow_SSSE3, 4, 1, 16) +YANY(YUY2ToYRow_Any_SSE2, YUY2ToYRow_SSE2, 2, 1, 16) +YANY(UYVYToYRow_Any_SSE2, UYVYToYRow_SSE2, 2, 1, 16) #endif #ifdef HAS_ARGBTOYJROW_SSSE3 -YANY(ARGBToYJRow_Any_SSSE3, ARGBToYJRow_Unaligned_SSSE3, 4, 1, 16) +YANY(ARGBToYJRow_Any_SSSE3, ARGBToYJRow_SSSE3, 4, 1, 16) #endif #ifdef HAS_ARGBTOYROW_NEON YANY(ARGBToYRow_Any_NEON, ARGBToYRow_NEON, 4, 1, 8) @@ -349,14 +349,14 @@ UVANY(YUY2ToUVRow_Any_AVX2, YUY2ToUVRow_AVX2, YUY2ToUVRow_C, 2, 31) UVANY(UYVYToUVRow_Any_AVX2, UYVYToUVRow_AVX2, UYVYToUVRow_C, 2, 31) #endif #ifdef HAS_ARGBTOUVROW_SSSE3 -UVANY(ARGBToUVRow_Any_SSSE3, ARGBToUVRow_Unaligned_SSSE3, ARGBToUVRow_C, 4, 15) -UVANY(ARGBToUVJRow_Any_SSSE3, ARGBToUVJRow_Unaligned_SSSE3, ARGBToUVJRow_C, +UVANY(ARGBToUVRow_Any_SSSE3, ARGBToUVRow_SSSE3, ARGBToUVRow_C, 4, 15) +UVANY(ARGBToUVJRow_Any_SSSE3, ARGBToUVJRow_SSSE3, ARGBToUVJRow_C, 4, 15) -UVANY(BGRAToUVRow_Any_SSSE3, BGRAToUVRow_Unaligned_SSSE3, BGRAToUVRow_C, 4, 15) -UVANY(ABGRToUVRow_Any_SSSE3, ABGRToUVRow_Unaligned_SSSE3, ABGRToUVRow_C, 4, 15) -UVANY(RGBAToUVRow_Any_SSSE3, RGBAToUVRow_Unaligned_SSSE3, RGBAToUVRow_C, 4, 15) -UVANY(YUY2ToUVRow_Any_SSE2, YUY2ToUVRow_Unaligned_SSE2, YUY2ToUVRow_C, 2, 15) -UVANY(UYVYToUVRow_Any_SSE2, UYVYToUVRow_Unaligned_SSE2, UYVYToUVRow_C, 2, 15) +UVANY(BGRAToUVRow_Any_SSSE3, BGRAToUVRow_SSSE3, BGRAToUVRow_C, 4, 15) +UVANY(ABGRToUVRow_Any_SSSE3, ABGRToUVRow_SSSE3, ABGRToUVRow_C, 4, 15) +UVANY(RGBAToUVRow_Any_SSSE3, RGBAToUVRow_SSSE3, RGBAToUVRow_C, 4, 15) +UVANY(YUY2ToUVRow_Any_SSE2, YUY2ToUVRow_SSE2, YUY2ToUVRow_C, 2, 15) +UVANY(UYVYToUVRow_Any_SSE2, UYVYToUVRow_SSE2, UYVYToUVRow_C, 2, 15) #endif #ifdef HAS_ARGBTOUVROW_NEON UVANY(ARGBToUVRow_Any_NEON, ARGBToUVRow_NEON, ARGBToUVRow_C, 4, 15) @@ -408,7 +408,7 @@ UVANY(UYVYToUVRow_Any_NEON, UYVYToUVRow_NEON, UYVYToUVRow_C, 2, 15) } #ifdef HAS_ARGBTOUV444ROW_SSSE3 -UV422ANY(ARGBToUV444Row_Any_SSSE3, ARGBToUV444Row_Unaligned_SSSE3, +UV422ANY(ARGBToUV444Row_Any_SSSE3, ARGBToUV444Row_SSSE3, ARGBToUV444Row_C, 4, 15, 0) #endif #ifdef HAS_YUY2TOUV422ROW_AVX2 @@ -418,11 +418,11 @@ UV422ANY(UYVYToUV422Row_Any_AVX2, UYVYToUV422Row_AVX2, UYVYToUV422Row_C, 2, 31, 1) #endif #ifdef HAS_ARGBTOUVROW_SSSE3 -UV422ANY(ARGBToUV422Row_Any_SSSE3, ARGBToUV422Row_Unaligned_SSSE3, +UV422ANY(ARGBToUV422Row_Any_SSSE3, ARGBToUV422Row_SSSE3, ARGBToUV422Row_C, 4, 15, 1) -UV422ANY(YUY2ToUV422Row_Any_SSE2, YUY2ToUV422Row_Unaligned_SSE2, +UV422ANY(YUY2ToUV422Row_Any_SSE2, YUY2ToUV422Row_SSE2, YUY2ToUV422Row_C, 2, 15, 1) -UV422ANY(UYVYToUV422Row_Any_SSE2, UYVYToUV422Row_Unaligned_SSE2, +UV422ANY(UYVYToUV422Row_Any_SSE2, UYVYToUV422Row_SSE2, UYVYToUV422Row_C, 2, 15, 1) #endif #ifdef HAS_YUY2TOUV422ROW_NEON @@ -451,7 +451,7 @@ UV422ANY(UYVYToUV422Row_Any_NEON, UYVYToUV422Row_NEON, } #ifdef HAS_SPLITUVROW_SSE2 -SPLITUVROWANY(SplitUVRow_Any_SSE2, SplitUVRow_Unaligned_SSE2, SplitUVRow_C, 15) +SPLITUVROWANY(SplitUVRow_Any_SSE2, SplitUVRow_SSE2, SplitUVRow_C, 15) #endif #ifdef HAS_SPLITUVROW_AVX2 SPLITUVROWANY(SplitUVRow_Any_AVX2, SplitUVRow_AVX2, SplitUVRow_C, 31) @@ -460,7 +460,7 @@ SPLITUVROWANY(SplitUVRow_Any_AVX2, SplitUVRow_AVX2, SplitUVRow_C, 31) SPLITUVROWANY(SplitUVRow_Any_NEON, SplitUVRow_NEON, SplitUVRow_C, 15) #endif #ifdef HAS_SPLITUVROW_MIPS_DSPR2 -SPLITUVROWANY(SplitUVRow_Any_MIPS_DSPR2, SplitUVRow_Unaligned_MIPS_DSPR2, +SPLITUVROWANY(SplitUVRow_Any_MIPS_DSPR2, SplitUVRow_MIPS_DSPR2, SplitUVRow_C, 15) #endif #undef SPLITUVROWANY @@ -477,7 +477,7 @@ SPLITUVROWANY(SplitUVRow_Any_MIPS_DSPR2, SplitUVRow_Unaligned_MIPS_DSPR2, } #ifdef HAS_MERGEUVROW_SSE2 -MERGEUVROW_ANY(MergeUVRow_Any_SSE2, MergeUVRow_Unaligned_SSE2, MergeUVRow_C, 15) +MERGEUVROW_ANY(MergeUVRow_Any_SSE2, MergeUVRow_SSE2, MergeUVRow_C, 15) #endif #ifdef HAS_MERGEUVROW_AVX2 MERGEUVROW_ANY(MergeUVRow_Any_AVX2, MergeUVRow_AVX2, MergeUVRow_C, 31) @@ -548,7 +548,7 @@ YANY(ARGBShuffleRow_Any_SSE2, ARGBShuffleRow_SSE2, ARGBShuffleRow_C, 4, 4, 3) #endif #ifdef HAS_ARGBSHUFFLEROW_SSSE3 -YANY(ARGBShuffleRow_Any_SSSE3, ARGBShuffleRow_Unaligned_SSSE3, +YANY(ARGBShuffleRow_Any_SSSE3, ARGBShuffleRow_SSSE3, ARGBShuffleRow_C, 4, 4, 7) #endif #ifdef HAS_ARGBSHUFFLEROW_AVX2 diff --git a/source/row_common.cc b/source/row_common.cc index 40a826171..afc74c05c 100644 --- a/source/row_common.cc +++ b/source/row_common.cc @@ -2137,19 +2137,6 @@ void YUY2ToARGBRow_SSSE3(const uint8* src_yuy2, free_aligned_buffer_64(row_y); } -void YUY2ToARGBRow_Unaligned_SSSE3(const uint8* src_yuy2, - uint8* dst_argb, - int width) { - // Allocate a rows of yuv. - align_buffer_64(row_y, ((width + 63) & ~63) * 2); - uint8* row_u = row_y + ((width + 63) & ~63); - uint8* row_v = row_u + ((width + 63) & ~63) / 2; - YUY2ToUV422Row_Unaligned_SSE2(src_yuy2, row_u, row_v, width); - YUY2ToYRow_Unaligned_SSE2(src_yuy2, row_y, width); - I422ToARGBRow_Unaligned_SSSE3(row_y, row_u, row_v, dst_argb, width); - free_aligned_buffer_64(row_y); -} - void UYVYToARGBRow_SSSE3(const uint8* src_uyvy, uint8* dst_argb, int width) { @@ -2163,19 +2150,6 @@ void UYVYToARGBRow_SSSE3(const uint8* src_uyvy, free_aligned_buffer_64(row_y); } -void UYVYToARGBRow_Unaligned_SSSE3(const uint8* src_uyvy, - uint8* dst_argb, - int width) { - // Allocate a rows of yuv. - align_buffer_64(row_y, ((width + 63) & ~63) * 2); - uint8* row_u = row_y + ((width + 63) & ~63); - uint8* row_v = row_u + ((width + 63) & ~63) / 2; - UYVYToUV422Row_Unaligned_SSE2(src_uyvy, row_u, row_v, width); - UYVYToYRow_Unaligned_SSE2(src_uyvy, row_y, width); - I422ToARGBRow_Unaligned_SSSE3(row_y, row_u, row_v, dst_argb, width); - free_aligned_buffer_64(row_y); -} - #endif // defined(_M_IX86) || defined(__x86_64__) || defined(__i386__) #endif // !defined(LIBYUV_DISABLE_X86) diff --git a/source/row_mips.cc b/source/row_mips.cc index da7183bc1..d71332115 100644 --- a/source/row_mips.cc +++ b/source/row_mips.cc @@ -447,89 +447,6 @@ void SplitUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, ); } -void SplitUVRow_Unaligned_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, - uint8* dst_v, int width) { - __asm__ __volatile__ ( - ".set push \n" - ".set noreorder \n" - "srl $t4, %[width], 4 \n" // multiplies of 16 - "blez $t4, 2f \n" - " andi %[width], %[width], 0xf \n" // residual - - ".p2align 2 \n" - "1: \n" - "addiu $t4, $t4, -1 \n" - "lwr $t0, 0(%[src_uv]) \n" - "lwl $t0, 3(%[src_uv]) \n" // V1 | U1 | V0 | U0 - "lwr $t1, 4(%[src_uv]) \n" - "lwl $t1, 7(%[src_uv]) \n" // V3 | U3 | V2 | U2 - "lwr $t2, 8(%[src_uv]) \n" - "lwl $t2, 11(%[src_uv]) \n" // V5 | U5 | V4 | U4 - "lwr $t3, 12(%[src_uv]) \n" - "lwl $t3, 15(%[src_uv]) \n" // V7 | U7 | V6 | U6 - "lwr $t5, 16(%[src_uv]) \n" - "lwl $t5, 19(%[src_uv]) \n" // V9 | U9 | V8 | U8 - "lwr $t6, 20(%[src_uv]) \n" - "lwl $t6, 23(%[src_uv]) \n" // V11 | U11 | V10 | U10 - "lwr $t7, 24(%[src_uv]) \n" - "lwl $t7, 27(%[src_uv]) \n" // V13 | U13 | V12 | U12 - "lwr $t8, 28(%[src_uv]) \n" - "lwl $t8, 31(%[src_uv]) \n" // V15 | U15 | V14 | U14 - "precrq.qb.ph $t9, $t1, $t0 \n" // V3 | V2 | V1 | V0 - "precr.qb.ph $t0, $t1, $t0 \n" // U3 | U2 | U1 | U0 - "precrq.qb.ph $t1, $t3, $t2 \n" // V7 | V6 | V5 | V4 - "precr.qb.ph $t2, $t3, $t2 \n" // U7 | U6 | U5 | U4 - "precrq.qb.ph $t3, $t6, $t5 \n" // V11 | V10 | V9 | V8 - "precr.qb.ph $t5, $t6, $t5 \n" // U11 | U10 | U9 | U8 - "precrq.qb.ph $t6, $t8, $t7 \n" // V15 | V14 | V13 | V12 - "precr.qb.ph $t7, $t8, $t7 \n" // U15 | U14 | U13 | U12 - "addiu %[src_uv], %[src_uv], 32 \n" - "swr $t9, 0(%[dst_v]) \n" - "swl $t9, 3(%[dst_v]) \n" - "swr $t0, 0(%[dst_u]) \n" - "swl $t0, 3(%[dst_u]) \n" - "swr $t1, 4(%[dst_v]) \n" - "swl $t1, 7(%[dst_v]) \n" - "swr $t2, 4(%[dst_u]) \n" - "swl $t2, 7(%[dst_u]) \n" - "swr $t3, 8(%[dst_v]) \n" - "swl $t3, 11(%[dst_v]) \n" - "swr $t5, 8(%[dst_u]) \n" - "swl $t5, 11(%[dst_u]) \n" - "swr $t6, 12(%[dst_v]) \n" - "swl $t6, 15(%[dst_v]) \n" - "swr $t7, 12(%[dst_u]) \n" - "swl $t7, 15(%[dst_u]) \n" - "addiu %[dst_u], %[dst_u], 16 \n" - "bgtz $t4, 1b \n" - " addiu %[dst_v], %[dst_v], 16 \n" - - "beqz %[width], 3f \n" - " nop \n" - - "2: \n" - "lbu $t0, 0(%[src_uv]) \n" - "lbu $t1, 1(%[src_uv]) \n" - "addiu %[src_uv], %[src_uv], 2 \n" - "addiu %[width], %[width], -1 \n" - "sb $t0, 0(%[dst_u]) \n" - "sb $t1, 0(%[dst_v]) \n" - "addiu %[dst_u], %[dst_u], 1 \n" - "bgtz %[width], 2b \n" - " addiu %[dst_v], %[dst_v], 1 \n" - - "3: \n" - ".set pop \n" - : [src_uv] "+r" (src_uv), - [width] "+r" (width), - [dst_u] "+r" (dst_u), - [dst_v] "+r" (dst_v) - : - : "t0", "t1", "t2", "t3", - "t4", "t5", "t6", "t7", "t8", "t9" - ); -} - void MirrorRow_MIPS_DSPR2(const uint8* src, uint8* dst, int width) { __asm__ __volatile__ ( ".set push \n" diff --git a/source/row_posix.cc b/source/row_posix.cc index e92c9f5f2..94b24a6f3 100644 --- a/source/row_posix.cc +++ b/source/row_posix.cc @@ -267,37 +267,6 @@ void I400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix) { #endif ); } - -void I400ToARGBRow_Unaligned_SSE2(const uint8* src_y, uint8* dst_argb, - int pix) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "pslld $0x18,%%xmm5 \n" - LABELALIGN - "1: \n" - "movq " MEMACCESS(0) ",%%xmm0 \n" - "lea " MEMLEA(0x8,0) ",%0 \n" - "punpcklbw %%xmm0,%%xmm0 \n" - "movdqa %%xmm0,%%xmm1 \n" - "punpcklwd %%xmm0,%%xmm0 \n" - "punpckhwd %%xmm1,%%xmm1 \n" - "por %%xmm5,%%xmm0 \n" - "por %%xmm5,%%xmm1 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - "movdqu %%xmm1," MEMACCESS2(0x10,1) " \n" - "lea " MEMLEA(0x20,1) ",%1 \n" - "sub $0x8,%2 \n" - "jg 1b \n" - : "+r"(src_y), // %0 - "+r"(dst_argb), // %1 - "+r"(pix) // %2 - : - : "memory", "cc" -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm5" -#endif - ); -} #endif // HAS_I400TOARGBROW_SSE2 #ifdef HAS_RGB24TOARGBROW_SSSE3 @@ -775,7 +744,7 @@ void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { ); } -void ARGBToYRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { +void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" "movdqa %3,%%xmm4 \n" @@ -851,44 +820,6 @@ void ARGBToYJRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { #endif ); } - -void ARGBToYJRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { - asm volatile ( - "movdqa %3,%%xmm4 \n" - "movdqa %4,%%xmm5 \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) ",%%xmm3 \n" - "pmaddubsw %%xmm4,%%xmm0 \n" - "pmaddubsw %%xmm4,%%xmm1 \n" - "pmaddubsw %%xmm4,%%xmm2 \n" - "pmaddubsw %%xmm4,%%xmm3 \n" - "lea " MEMLEA(0x40,0) ",%0 \n" - "phaddw %%xmm1,%%xmm0 \n" - "phaddw %%xmm3,%%xmm2 \n" - "paddw %%xmm5,%%xmm0 \n" - "paddw %%xmm5,%%xmm2 \n" - "psrlw $0x7,%%xmm0 \n" - "psrlw $0x7,%%xmm2 \n" - "packuswb %%xmm2,%%xmm0 \n" - "sub $0x10,%2 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_argb), // %0 - "+r"(dst_y), // %1 - "+r"(pix) // %2 - : "m"(kARGBToYJ), // %3 - "m"(kAddYJ64) // %4 - : "memory", "cc" -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} #endif // HAS_ARGBTOYJROW_SSSE3 #ifdef HAS_ARGBTOUVROW_SSSE3 @@ -1031,147 +962,6 @@ void ARGBToUVJRow_SSSE3(const uint8* src_argb0, int src_stride_argb, ); } -void ARGBToUVRow_Unaligned_SSSE3(const uint8* src_argb0, int src_stride_argb, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "movdqa %0,%%xmm4 \n" - "movdqa %1,%%xmm3 \n" - "movdqa %2,%%xmm5 \n" - : - : "m"(kARGBToU), // %0 - "m"(kARGBToV), // %1 - "m"(kAddUV128) // %2 - ); - asm volatile ( - "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" - BUNDLEALIGN - MEMOPREG(movdqu,0x00,0,4,1,xmm7) // movdqu (%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm0 \n" - MEMOPREG(movdqu,0x10,0,4,1,xmm7) // movdqu 0x10(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm1 \n" - MEMOPREG(movdqu,0x20,0,4,1,xmm7) // movdqu 0x20(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm2 \n" - MEMOPREG(movdqu,0x30,0,4,1,xmm7) // movdqu 0x30(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%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" - "sub $0x10,%3 \n" - "movlps %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) - "lea " MEMLEA(0x8,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_argb0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : "r"((intptr_t)(src_stride_argb)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" -#endif - ); -} - -void ARGBToUVJRow_Unaligned_SSSE3(const uint8* src_argb0, int src_stride_argb, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "movdqa %0,%%xmm4 \n" - "movdqa %1,%%xmm3 \n" - "movdqa %2,%%xmm5 \n" - : - : "m"(kARGBToUJ), // %0 - "m"(kARGBToVJ), // %1 - "m"(kAddUVJ128) // %2 - ); - asm volatile ( - "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" - BUNDLEALIGN - MEMOPREG(movdqu,0x00,0,4,1,xmm7) // movdqu (%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm0 \n" - MEMOPREG(movdqu,0x10,0,4,1,xmm7) // movdqu 0x10(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm1 \n" - MEMOPREG(movdqu,0x20,0,4,1,xmm7) // movdqu 0x20(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm2 \n" - MEMOPREG(movdqu,0x30,0,4,1,xmm7) // movdqu 0x30(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%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" - "paddw %%xmm5,%%xmm0 \n" - "paddw %%xmm5,%%xmm1 \n" - "psraw $0x8,%%xmm0 \n" - "psraw $0x8,%%xmm1 \n" - "packsswb %%xmm1,%%xmm0 \n" - "sub $0x10,%3 \n" - "movlps %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) - "lea " MEMLEA(0x8,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_argb0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : "r"((intptr_t)(src_stride_argb)) - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" -#endif - ); -} - void ARGBToUV444Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width) { asm volatile ( @@ -1237,7 +1027,7 @@ void ARGBToUV444Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, ); } -void ARGBToUV444Row_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_u, +void ARGBToUV444Row_SSSE3(const uint8* src_argb, uint8* dst_u, uint8* dst_v, int width) { asm volatile ( "movdqa %0,%%xmm4 \n" @@ -1363,67 +1153,6 @@ void ARGBToUV422Row_SSSE3(const uint8* src_argb0, ); } -void ARGBToUV422Row_Unaligned_SSSE3(const uint8* src_argb0, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "movdqa %0,%%xmm4 \n" - "movdqa %1,%%xmm3 \n" - "movdqa %2,%%xmm5 \n" - : - : "m"(kARGBToU), // %0 - "m"(kARGBToV), // %1 - "m"(kAddUV128) // %2 - ); - asm volatile ( - "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" - "sub $0x10,%3 \n" - "movlps %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) - "lea " MEMLEA(0x8,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_argb0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" -#endif - ); -} - void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" @@ -1461,43 +1190,6 @@ void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int pix) { ); } -void BGRAToYRow_Unaligned_SSSE3(const uint8* src_bgra, uint8* dst_y, int pix) { - asm volatile ( - "movdqa %4,%%xmm5 \n" - "movdqa %3,%%xmm4 \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) ",%%xmm3 \n" - "pmaddubsw %%xmm4,%%xmm0 \n" - "pmaddubsw %%xmm4,%%xmm1 \n" - "pmaddubsw %%xmm4,%%xmm2 \n" - "pmaddubsw %%xmm4,%%xmm3 \n" - "lea " MEMLEA(0x40,0) ",%0 \n" - "phaddw %%xmm1,%%xmm0 \n" - "phaddw %%xmm3,%%xmm2 \n" - "psrlw $0x7,%%xmm0 \n" - "psrlw $0x7,%%xmm2 \n" - "packuswb %%xmm2,%%xmm0 \n" - "paddb %%xmm5,%%xmm0 \n" - "sub $0x10,%2 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_bgra), // %0 - "+r"(dst_y), // %1 - "+r"(pix) // %2 - : "m"(kBGRAToY), // %3 - "m"(kAddY16) // %4 - : "memory", "cc" -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - void BGRAToUVRow_SSSE3(const uint8* src_bgra0, int src_stride_bgra, uint8* dst_u, uint8* dst_v, int width) { asm volatile ( @@ -1564,76 +1256,6 @@ void BGRAToUVRow_SSSE3(const uint8* src_bgra0, int src_stride_bgra, ); } -void BGRAToUVRow_Unaligned_SSSE3(const uint8* src_bgra0, int src_stride_bgra, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "movdqa %0,%%xmm4 \n" - "movdqa %1,%%xmm3 \n" - "movdqa %2,%%xmm5 \n" - : - : "m"(kBGRAToU), // %0 - "m"(kBGRAToV), // %1 - "m"(kAddUV128) // %2 - ); - asm volatile ( - "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" - BUNDLEALIGN - MEMOPREG(movdqu,0x00,0,4,1,xmm7) // movdqu (%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm0 \n" - MEMOPREG(movdqu,0x10,0,4,1,xmm7) // movdqu 0x10(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm1 \n" - MEMOPREG(movdqu,0x20,0,4,1,xmm7) // movdqu 0x20(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm2 \n" - MEMOPREG(movdqu,0x30,0,4,1,xmm7) // movdqu 0x30(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%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" - "sub $0x10,%3 \n" - "movlps %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) - "lea " MEMLEA(0x8,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_bgra0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : "r"((intptr_t)(src_stride_bgra)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" -#endif - ); -} - void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" @@ -1671,43 +1293,6 @@ void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int pix) { ); } -void ABGRToYRow_Unaligned_SSSE3(const uint8* src_abgr, uint8* dst_y, int pix) { - asm volatile ( - "movdqa %4,%%xmm5 \n" - "movdqa %3,%%xmm4 \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) ",%%xmm3 \n" - "pmaddubsw %%xmm4,%%xmm0 \n" - "pmaddubsw %%xmm4,%%xmm1 \n" - "pmaddubsw %%xmm4,%%xmm2 \n" - "pmaddubsw %%xmm4,%%xmm3 \n" - "lea " MEMLEA(0x40,0) ",%0 \n" - "phaddw %%xmm1,%%xmm0 \n" - "phaddw %%xmm3,%%xmm2 \n" - "psrlw $0x7,%%xmm0 \n" - "psrlw $0x7,%%xmm2 \n" - "packuswb %%xmm2,%%xmm0 \n" - "paddb %%xmm5,%%xmm0 \n" - "sub $0x10,%2 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_abgr), // %0 - "+r"(dst_y), // %1 - "+r"(pix) // %2 - : "m"(kABGRToY), // %3 - "m"(kAddY16) // %4 - : "memory", "cc" -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" @@ -1745,7 +1330,7 @@ void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix) { ); } -void RGBAToYRow_Unaligned_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix) { +void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix) { asm volatile ( "movdqa %4,%%xmm5 \n" "movdqa %3,%%xmm4 \n" @@ -1848,76 +1433,6 @@ void ABGRToUVRow_SSSE3(const uint8* src_abgr0, int src_stride_abgr, ); } -void ABGRToUVRow_Unaligned_SSSE3(const uint8* src_abgr0, int src_stride_abgr, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "movdqa %0,%%xmm4 \n" - "movdqa %1,%%xmm3 \n" - "movdqa %2,%%xmm5 \n" - : - : "m"(kABGRToU), // %0 - "m"(kABGRToV), // %1 - "m"(kAddUV128) // %2 - ); - asm volatile ( - "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" - BUNDLEALIGN - MEMOPREG(movdqu,0x00,0,4,1,xmm7) // movdqu (%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm0 \n" - MEMOPREG(movdqu,0x10,0,4,1,xmm7) // movdqu 0x10(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm1 \n" - MEMOPREG(movdqu,0x20,0,4,1,xmm7) // movdqu 0x20(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm2 \n" - MEMOPREG(movdqu,0x30,0,4,1,xmm7) // movdqu 0x30(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%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" - "sub $0x10,%3 \n" - "movlps %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) - "lea " MEMLEA(0x8,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_abgr0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : "r"((intptr_t)(src_stride_abgr)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" -#endif - ); -} - void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, uint8* dst_u, uint8* dst_v, int width) { asm volatile ( @@ -1983,76 +1498,6 @@ void RGBAToUVRow_SSSE3(const uint8* src_rgba0, int src_stride_rgba, #endif ); } - -void RGBAToUVRow_Unaligned_SSSE3(const uint8* src_rgba0, int src_stride_rgba, - uint8* dst_u, uint8* dst_v, int width) { - asm volatile ( - "movdqa %0,%%xmm4 \n" - "movdqa %1,%%xmm3 \n" - "movdqa %2,%%xmm5 \n" - : - : "m"(kRGBAToU), // %0 - "m"(kRGBAToV), // %1 - "m"(kAddUV128) // %2 - ); - asm volatile ( - "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" - BUNDLEALIGN - MEMOPREG(movdqu,0x00,0,4,1,xmm7) // movdqu (%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm0 \n" - MEMOPREG(movdqu,0x10,0,4,1,xmm7) // movdqu 0x10(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm1 \n" - MEMOPREG(movdqu,0x20,0,4,1,xmm7) // movdqu 0x20(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%xmm2 \n" - MEMOPREG(movdqu,0x30,0,4,1,xmm7) // movdqu 0x30(%0,%4,1),%%xmm7 - "pavgb %%xmm7,%%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" - "sub $0x10,%3 \n" - "movlps %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movhps,xmm0,0x00,1,2,1) // movhps %%xmm0,(%1,%2,1) - "lea " MEMLEA(0x8,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_rgba0), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+rm"(width) // %3 - : "r"((intptr_t)(src_stride_rgba)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm6", "xmm7" -#endif - ); -} #endif // HAS_ARGBTOUVROW_SSSE3 #ifdef HAS_I422TOARGBROW_SSSE3 @@ -2482,191 +1927,6 @@ void OMITFP NV21ToARGBRow_SSSE3(const uint8* y_buf, ); } -void OMITFP I444ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_argb, - int width) { - asm volatile ( - "sub %[u_buf],%[v_buf] \n" - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READYUV444 - YUVTORGB - "punpcklbw %%xmm1,%%xmm0 \n" - "punpcklbw %%xmm5,%%xmm2 \n" - "movdqa %%xmm0,%%xmm1 \n" - "punpcklwd %%xmm2,%%xmm0 \n" - "punpckhwd %%xmm2,%%xmm1 \n" - "movdqu %%xmm0," MEMACCESS([dst_argb]) "\n" - "movdqu %%xmm1," MEMACCESS2(0x10,[dst_argb]) "\n" - "lea " MEMLEA(0x20,[dst_argb]) ",%[dst_argb] \n" - "sub $0x8,%[width] \n" - "jg 1b \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] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - -void OMITFP I422ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_argb, - int width) { - asm volatile ( - "sub %[u_buf],%[v_buf] \n" - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READYUV422 - YUVTORGB - "punpcklbw %%xmm1,%%xmm0 \n" - "punpcklbw %%xmm5,%%xmm2 \n" - "movdqa %%xmm0,%%xmm1 \n" - "punpcklwd %%xmm2,%%xmm0 \n" - "punpckhwd %%xmm2,%%xmm1 \n" - "movdqu %%xmm0," MEMACCESS([dst_argb]) "\n" - "movdqu %%xmm1," MEMACCESS2(0x10,[dst_argb]) "\n" - "lea " MEMLEA(0x20,[dst_argb]) ",%[dst_argb] \n" - "sub $0x8,%[width] \n" - "jg 1b \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] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - -void OMITFP I411ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_argb, - int width) { - asm volatile ( - "sub %[u_buf],%[v_buf] \n" - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READYUV411 - YUVTORGB - "punpcklbw %%xmm1,%%xmm0 \n" - "punpcklbw %%xmm5,%%xmm2 \n" - "movdqa %%xmm0,%%xmm1 \n" - "punpcklwd %%xmm2,%%xmm0 \n" - "punpckhwd %%xmm2,%%xmm1 \n" - "movdqu %%xmm0," MEMACCESS([dst_argb]) "\n" - "movdqu %%xmm1," MEMACCESS2(0x10,[dst_argb]) "\n" - "lea " MEMLEA(0x20,[dst_argb]) ",%[dst_argb] \n" - "sub $0x8,%[width] \n" - "jg 1b \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] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - -void OMITFP NV12ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* uv_buf, - uint8* dst_argb, - int width) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READNV12 - YUVTORGB - "punpcklbw %%xmm1,%%xmm0 \n" - "punpcklbw %%xmm5,%%xmm2 \n" - "movdqa %%xmm0,%%xmm1 \n" - "punpcklwd %%xmm2,%%xmm0 \n" - "punpckhwd %%xmm2,%%xmm1 \n" - "movdqu %%xmm0," MEMACCESS([dst_argb]) "\n" - "movdqu %%xmm1," MEMACCESS2(0x10,[dst_argb]) "\n" - "lea " MEMLEA(0x20,[dst_argb]) ",%[dst_argb] \n" - "sub $0x8,%[width] \n" - "jg 1b \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [uv_buf]"+r"(uv_buf), // %[uv_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] - [width]"+rm"(width) // %[width] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" - // Does not use r14. -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - -void OMITFP NV21ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* uv_buf, - uint8* dst_argb, - int width) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READNV12 - YVUTORGB - "punpcklbw %%xmm1,%%xmm0 \n" - "punpcklbw %%xmm5,%%xmm2 \n" - "movdqa %%xmm0,%%xmm1 \n" - "punpcklwd %%xmm2,%%xmm0 \n" - "punpckhwd %%xmm2,%%xmm1 \n" - "movdqu %%xmm0," MEMACCESS([dst_argb]) "\n" - "movdqu %%xmm1," MEMACCESS2(0x10,[dst_argb]) "\n" - "lea " MEMLEA(0x20,[dst_argb]) ",%[dst_argb] \n" - "sub $0x8,%[width] \n" - "jg 1b \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [uv_buf]"+r"(uv_buf), // %[uv_buf] - [dst_argb]"+r"(dst_argb), // %[dst_argb] - [width]"+rm"(width) // %[width] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" - // Does not use r14. -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - void OMITFP I422ToBGRARow_SSSE3(const uint8* y_buf, const uint8* u_buf, const uint8* v_buf, @@ -2786,125 +2046,6 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8* y_buf, ); } -void OMITFP I422ToBGRARow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_bgra, - int width) { - asm volatile ( - "sub %[u_buf],%[v_buf] \n" - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READYUV422 - YUVTORGB - "pcmpeqb %%xmm5,%%xmm5 \n" - "punpcklbw %%xmm0,%%xmm1 \n" - "punpcklbw %%xmm2,%%xmm5 \n" - "movdqa %%xmm5,%%xmm0 \n" - "punpcklwd %%xmm1,%%xmm5 \n" - "punpckhwd %%xmm1,%%xmm0 \n" - "movdqu %%xmm5," MEMACCESS([dst_bgra]) "\n" - "movdqu %%xmm0," MEMACCESS2(0x10,[dst_bgra]) "\n" - "lea " MEMLEA(0x20,[dst_bgra]) ",%[dst_bgra] \n" - "sub $0x8,%[width] \n" - "jg 1b \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [u_buf]"+r"(u_buf), // %[u_buf] - [v_buf]"+r"(v_buf), // %[v_buf] - [dst_bgra]"+r"(dst_bgra), // %[dst_bgra] - [width]"+rm"(width) // %[width] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - -void OMITFP I422ToABGRRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_abgr, - int width) { - asm volatile ( - "sub %[u_buf],%[v_buf] \n" - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READYUV422 - YUVTORGB - "punpcklbw %%xmm1,%%xmm2 \n" - "punpcklbw %%xmm5,%%xmm0 \n" - "movdqa %%xmm2,%%xmm1 \n" - "punpcklwd %%xmm0,%%xmm2 \n" - "punpckhwd %%xmm0,%%xmm1 \n" - "movdqu %%xmm2," MEMACCESS([dst_abgr]) "\n" - "movdqu %%xmm1," MEMACCESS2(0x10,[dst_abgr]) "\n" - "lea " MEMLEA(0x20,[dst_abgr]) ",%[dst_abgr] \n" - "sub $0x8,%[width] \n" - "jg 1b \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [u_buf]"+r"(u_buf), // %[u_buf] - [v_buf]"+r"(v_buf), // %[v_buf] - [dst_abgr]"+r"(dst_abgr), // %[dst_abgr] - [width]"+rm"(width) // %[width] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - -void OMITFP I422ToRGBARow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_rgba, - int width) { - asm volatile ( - "sub %[u_buf],%[v_buf] \n" - "pcmpeqb %%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - LABELALIGN - "1: \n" - READYUV422 - YUVTORGB - "pcmpeqb %%xmm5,%%xmm5 \n" - "punpcklbw %%xmm2,%%xmm1 \n" - "punpcklbw %%xmm0,%%xmm5 \n" - "movdqa %%xmm5,%%xmm0 \n" - "punpcklwd %%xmm1,%%xmm5 \n" - "punpckhwd %%xmm1,%%xmm0 \n" - "movdqu %%xmm5," MEMACCESS([dst_rgba]) "\n" - "movdqu %%xmm0," MEMACCESS2(0x10,[dst_rgba]) "\n" - "lea " MEMLEA(0x20,[dst_rgba]) ",%[dst_rgba] \n" - "sub $0x8,%[width] \n" - "jg 1b \n" - : [y_buf]"+r"(y_buf), // %[y_buf] - [u_buf]"+r"(u_buf), // %[u_buf] - [v_buf]"+r"(v_buf), // %[v_buf] - [dst_rgba]"+r"(dst_rgba), // %[dst_rgba] - [width]"+rm"(width) // %[width] - : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants] - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} - #endif // HAS_I422TOARGBROW_SSSE3 #ifdef HAS_YTOARGBROW_SSE2 @@ -3134,45 +2275,6 @@ void SplitUVRow_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix) { #endif ); } - -void SplitUVRow_Unaligned_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, - int pix) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "psrlw $0x8,%%xmm5 \n" - "sub %1,%2 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "lea " MEMLEA(0x20,0) ",%0 \n" - "movdqa %%xmm0,%%xmm2 \n" - "movdqa %%xmm1,%%xmm3 \n" - "pand %%xmm5,%%xmm0 \n" - "pand %%xmm5,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "psrlw $0x8,%%xmm2 \n" - "psrlw $0x8,%%xmm3 \n" - "packuswb %%xmm3,%%xmm2 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - MEMOPMEM(movdqu,xmm2,0x00,1,2,1) // movdqu %%xmm2,(%1,%2) - "lea " MEMLEA(0x10,1) ",%1 \n" - "sub $0x10,%3 \n" - "jg 1b \n" - : "+r"(src_uv), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(pix) // %3 - : - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" -#endif - ); -} #endif // HAS_SPLITUVROW_SSE2 #ifdef HAS_MERGEUVROW_SSE2 @@ -3207,38 +2309,6 @@ void MergeUVRow_SSE2(const uint8* src_u, const uint8* src_v, uint8* dst_uv, #endif ); } - -void MergeUVRow_Unaligned_SSE2(const uint8* src_u, const uint8* src_v, - uint8* dst_uv, int width) { - asm volatile ( - "sub %0,%1 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - MEMOPREG(movdqu,0x00,0,1,1,xmm1) // movdqu (%0,%1,1),%%xmm1 - "lea " MEMLEA(0x10,0) ",%0 \n" - "movdqa %%xmm0,%%xmm2 \n" - "punpcklbw %%xmm1,%%xmm0 \n" - "punpckhbw %%xmm1,%%xmm2 \n" - "movdqu %%xmm0," MEMACCESS(2) " \n" - "movdqu %%xmm2," MEMACCESS2(0x10,2) " \n" - "lea " MEMLEA(0x20,2) ",%2 \n" - "sub $0x10,%3 \n" - "jg 1b \n" - : "+r"(src_u), // %0 - "+r"(src_v), // %1 - "+r"(dst_uv), // %2 - "+r"(width) // %3 - : - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2" -#endif - ); -} #endif // HAS_MERGEUVROW_SSE2 #ifdef HAS_COPYROW_SSE2 @@ -3307,7 +2377,7 @@ void CopyRow_X86(const uint8* src, uint8* dst, int width) { #endif // HAS_COPYROW_X86 #ifdef HAS_COPYROW_ERMS -// Unaligned Multiple of 1. +// Multiple of 1. void CopyRow_ERMS(const uint8* src, uint8* dst, int width) { size_t width_tmp = (size_t)(width); asm volatile ( @@ -3604,120 +2674,6 @@ void YUY2ToUV422Row_SSE2(const uint8* src_yuy2, ); } -void YUY2ToYRow_Unaligned_SSE2(const uint8* src_yuy2, - uint8* dst_y, int pix) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "psrlw $0x8,%%xmm5 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "lea " MEMLEA(0x20,0) ",%0 \n" - "pand %%xmm5,%%xmm0 \n" - "pand %%xmm5,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_yuy2), // %0 - "+r"(dst_y), // %1 - "+r"(pix) // %2 - : - : "memory", "cc" -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm5" -#endif - ); -} - -void YUY2ToUVRow_Unaligned_SSE2(const uint8* src_yuy2, - int stride_yuy2, - uint8* dst_u, uint8* dst_v, int pix) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "psrlw $0x8,%%xmm5 \n" - "sub %1,%2 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - BUNDLEALIGN - MEMOPREG(movdqu,0x00,0,4,1,xmm2) // movdqu (%0,%4,1),%%xmm2 - MEMOPREG(movdqu,0x10,0,4,1,xmm3) // movdqu 0x10(%0,%4,1),%%xmm3 - "lea " MEMLEA(0x20,0) ",%0 \n" - "pavgb %%xmm2,%%xmm0 \n" - "pavgb %%xmm3,%%xmm1 \n" - "psrlw $0x8,%%xmm0 \n" - "psrlw $0x8,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "movdqa %%xmm0,%%xmm1 \n" - "pand %%xmm5,%%xmm0 \n" - "packuswb %%xmm0,%%xmm0 \n" - "psrlw $0x8,%%xmm1 \n" - "packuswb %%xmm1,%%xmm1 \n" - "movq %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movq,xmm1,0x00,1,2,1) // movq %%xmm1,(%1,%2) - "lea " MEMLEA(0x8,1) ",%1 \n" - "sub $0x10,%3 \n" - "jg 1b \n" - : "+r"(src_yuy2), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(pix) // %3 - : "r"((intptr_t)(stride_yuy2)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" -#endif - ); -} - -void YUY2ToUV422Row_Unaligned_SSE2(const uint8* src_yuy2, - uint8* dst_u, uint8* dst_v, int pix) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "psrlw $0x8,%%xmm5 \n" - "sub %1,%2 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "lea " MEMLEA(0x20,0) ",%0 \n" - "psrlw $0x8,%%xmm0 \n" - "psrlw $0x8,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "movdqa %%xmm0,%%xmm1 \n" - "pand %%xmm5,%%xmm0 \n" - "packuswb %%xmm0,%%xmm0 \n" - "psrlw $0x8,%%xmm1 \n" - "packuswb %%xmm1,%%xmm1 \n" - "movq %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movq,xmm1,0x00,1,2,1) // movq %%xmm1,(%1,%2) - "lea " MEMLEA(0x8,1) ",%1 \n" - "sub $0x10,%3 \n" - "jg 1b \n" - : "+r"(src_yuy2), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(pix) // %3 - : - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm5" -#endif - ); -} - void UYVYToYRow_SSE2(const uint8* src_uyvy, uint8* dst_y, int pix) { asm volatile ( LABELALIGN @@ -3827,117 +2783,6 @@ void UYVYToUV422Row_SSE2(const uint8* src_uyvy, #endif ); } - -void UYVYToYRow_Unaligned_SSE2(const uint8* src_uyvy, - uint8* dst_y, int pix) { - asm volatile ( - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "lea " MEMLEA(0x20,0) ",%0 \n" - "psrlw $0x8,%%xmm0 \n" - "psrlw $0x8,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_uyvy), // %0 - "+r"(dst_y), // %1 - "+r"(pix) // %2 - : - : "memory", "cc" -#if defined(__SSE2__) - , "xmm0", "xmm1" -#endif - ); -} - -void UYVYToUVRow_Unaligned_SSE2(const uint8* src_uyvy, int stride_uyvy, - uint8* dst_u, uint8* dst_v, int pix) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "psrlw $0x8,%%xmm5 \n" - "sub %1,%2 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - BUNDLEALIGN - MEMOPREG(movdqu,0x00,0,4,1,xmm2) // movdqu (%0,%4,1),%%xmm2 - MEMOPREG(movdqu,0x10,0,4,1,xmm3) // movdqu 0x10(%0,%4,1),%%xmm3 - "lea " MEMLEA(0x20,0) ",%0 \n" - "pavgb %%xmm2,%%xmm0 \n" - "pavgb %%xmm3,%%xmm1 \n" - "pand %%xmm5,%%xmm0 \n" - "pand %%xmm5,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "movdqa %%xmm0,%%xmm1 \n" - "pand %%xmm5,%%xmm0 \n" - "packuswb %%xmm0,%%xmm0 \n" - "psrlw $0x8,%%xmm1 \n" - "packuswb %%xmm1,%%xmm1 \n" - "movq %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movq,xmm1,0x00,1,2,1) // movq %%xmm1,(%1,%2) - "lea " MEMLEA(0x8,1) ",%1 \n" - "sub $0x10,%3 \n" - "jg 1b \n" - : "+r"(src_uyvy), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(pix) // %3 - : "r"((intptr_t)(stride_uyvy)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5" -#endif - ); -} - -void UYVYToUV422Row_Unaligned_SSE2(const uint8* src_uyvy, - uint8* dst_u, uint8* dst_v, int pix) { - asm volatile ( - "pcmpeqb %%xmm5,%%xmm5 \n" - "psrlw $0x8,%%xmm5 \n" - "sub %1,%2 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "lea " MEMLEA(0x20,0) ",%0 \n" - "pand %%xmm5,%%xmm0 \n" - "pand %%xmm5,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "movdqa %%xmm0,%%xmm1 \n" - "pand %%xmm5,%%xmm0 \n" - "packuswb %%xmm0,%%xmm0 \n" - "psrlw $0x8,%%xmm1 \n" - "packuswb %%xmm1,%%xmm1 \n" - "movq %%xmm0," MEMACCESS(1) " \n" - BUNDLEALIGN - MEMOPMEM(movq,xmm1,0x00,1,2,1) // movq %%xmm1,(%1,%2) - "lea " MEMLEA(0x8,1) ",%1 \n" - "sub $0x10,%3 \n" - "jg 1b \n" - : "+r"(src_uyvy), // %0 - "+r"(dst_u), // %1 - "+r"(dst_v), // %2 - "+r"(pix) // %3 - : - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm5" -#endif - ); -} #endif // HAS_YUY2TOYROW_SSE2 #ifdef HAS_ARGBBLENDROW_SSE2 @@ -5581,238 +4426,6 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr, } #endif // HAS_INTERPOLATEROW_SSE2 -#ifdef HAS_INTERPOLATEROW_SSSE3 -// Bilinear filter 16x2 -> 16x1 -void InterpolateRow_Unaligned_SSSE3(uint8* dst_ptr, const uint8* src_ptr, - ptrdiff_t src_stride, int dst_width, - int source_y_fraction) { - asm volatile ( - "sub %1,%0 \n" - "shr %3 \n" - "cmp $0x0,%3 \n" - "je 100f \n" - "cmp $0x20,%3 \n" - "je 75f \n" - "cmp $0x40,%3 \n" - "je 50f \n" - "cmp $0x60,%3 \n" - "je 25f \n" - - "movd %3,%%xmm0 \n" - "neg %3 \n" - "add $0x80,%3 \n" - "movd %3,%%xmm5 \n" - "punpcklbw %%xmm0,%%xmm5 \n" - "punpcklwd %%xmm5,%%xmm5 \n" - "pshufd $0x0,%%xmm5,%%xmm5 \n" - - // General purpose row blend. - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm2) - "movdqu %%xmm0,%%xmm1 \n" - "punpcklbw %%xmm2,%%xmm0 \n" - "punpckhbw %%xmm2,%%xmm1 \n" - "pmaddubsw %%xmm5,%%xmm0 \n" - "pmaddubsw %%xmm5,%%xmm1 \n" - "psrlw $0x7,%%xmm0 \n" - "psrlw $0x7,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 1b \n" - "jmp 99f \n" - - // Blend 25 / 75. - LABELALIGN - "25: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm1) - "pavgb %%xmm1,%%xmm0 \n" - "pavgb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 25b \n" - "jmp 99f \n" - - // Blend 50 / 50. - LABELALIGN - "50: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm1) - "pavgb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 50b \n" - "jmp 99f \n" - - // Blend 75 / 25. - LABELALIGN - "75: \n" - "movdqu " MEMACCESS(1) ",%%xmm1 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm0) - "pavgb %%xmm1,%%xmm0 \n" - "pavgb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 75b \n" - "jmp 99f \n" - - // Blend 100 / 0 - Copy row unchanged. - LABELALIGN - "100: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - "sub $0x10,%2 \n" - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 100b \n" - - "99: \n" - : "+r"(dst_ptr), // %0 - "+r"(src_ptr), // %1 - "+r"(dst_width), // %2 - "+r"(source_y_fraction) // %3 - : "r"((intptr_t)(src_stride)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm5" -#endif - ); -} -#endif // HAS_INTERPOLATEROW_SSSE3 - -#ifdef HAS_INTERPOLATEROW_SSE2 -// Bilinear filter 16x2 -> 16x1 -void InterpolateRow_Unaligned_SSE2(uint8* dst_ptr, const uint8* src_ptr, - ptrdiff_t src_stride, int dst_width, - int source_y_fraction) { - asm volatile ( - "sub %1,%0 \n" - "shr %3 \n" - "cmp $0x0,%3 \n" - "je 100f \n" - "cmp $0x20,%3 \n" - "je 75f \n" - "cmp $0x40,%3 \n" - "je 50f \n" - "cmp $0x60,%3 \n" - "je 25f \n" - - "movd %3,%%xmm0 \n" - "neg %3 \n" - "add $0x80,%3 \n" - "movd %3,%%xmm5 \n" - "punpcklbw %%xmm0,%%xmm5 \n" - "punpcklwd %%xmm5,%%xmm5 \n" - "pshufd $0x0,%%xmm5,%%xmm5 \n" - "pxor %%xmm4,%%xmm4 \n" - - // General purpose row blend. - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm2) // movdqu (%1,%4,1),%%xmm2 - "movdqu %%xmm0,%%xmm1 \n" - "movdqu %%xmm2,%%xmm3 \n" - "punpcklbw %%xmm4,%%xmm2 \n" - "punpckhbw %%xmm4,%%xmm3 \n" - "punpcklbw %%xmm4,%%xmm0 \n" - "punpckhbw %%xmm4,%%xmm1 \n" - "psubw %%xmm0,%%xmm2 \n" - "psubw %%xmm1,%%xmm3 \n" - "paddw %%xmm2,%%xmm2 \n" - "paddw %%xmm3,%%xmm3 \n" - "pmulhw %%xmm5,%%xmm2 \n" - "pmulhw %%xmm5,%%xmm3 \n" - "paddw %%xmm2,%%xmm0 \n" - "paddw %%xmm3,%%xmm1 \n" - "packuswb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 1b \n" - "jmp 99f \n" - - // Blend 25 / 75. - LABELALIGN - "25: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm1) // movdqu (%1,%4,1),%%xmm1 - "pavgb %%xmm1,%%xmm0 \n" - "pavgb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 25b \n" - "jmp 99f \n" - - // Blend 50 / 50. - LABELALIGN - "50: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm1) // movdqu (%1,%4,1),%%xmm1 - "pavgb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 50b \n" - "jmp 99f \n" - - // Blend 75 / 25. - LABELALIGN - "75: \n" - "movdqu " MEMACCESS(1) ",%%xmm1 \n" - MEMOPREG(movdqu,0x00,1,4,1,xmm0) // movdqu (%1,%4,1),%%xmm0 - "pavgb %%xmm1,%%xmm0 \n" - "pavgb %%xmm1,%%xmm0 \n" - "sub $0x10,%2 \n" - BUNDLEALIGN - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 75b \n" - "jmp 99f \n" - - // Blend 100 / 0 - Copy row unchanged. - LABELALIGN - "100: \n" - "movdqu " MEMACCESS(1) ",%%xmm0 \n" - "sub $0x10,%2 \n" - MEMOPMEM(movdqu,xmm0,0x00,1,0,1) // movdqu %%xmm0,(%1,%0,1) - "lea " MEMLEA(0x10,1) ",%1 \n" - "jg 100b \n" - - "99: \n" - : "+r"(dst_ptr), // %0 - "+r"(src_ptr), // %1 - "+r"(dst_width), // %2 - "+r"(source_y_fraction) // %3 - : "r"((intptr_t)(src_stride)) // %4 - : "memory", "cc" -#if defined(__native_client__) && defined(__x86_64__) - , "r14" -#endif -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5" -#endif - ); -} -#endif // HAS_INTERPOLATEROW_SSE2 - #ifdef HAS_ARGBTOBAYERROW_SSSE3 void ARGBToBayerRow_SSSE3(const uint8* src_argb, uint8* dst_bayer, uint32 selector, int pix) { @@ -5905,33 +4518,6 @@ void ARGBShuffleRow_SSSE3(const uint8* src_argb, uint8* dst_argb, #endif ); } - -void ARGBShuffleRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_argb, - const uint8* shuffler, int pix) { - asm volatile ( - "movdqu " MEMACCESS(3) ",%%xmm5 \n" - LABELALIGN - "1: \n" - "movdqu " MEMACCESS(0) ",%%xmm0 \n" - "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n" - "lea " MEMLEA(0x20,0) ",%0 \n" - "pshufb %%xmm5,%%xmm0 \n" - "pshufb %%xmm5,%%xmm1 \n" - "sub $0x8,%2 \n" - "movdqu %%xmm0," MEMACCESS(1) " \n" - "movdqu %%xmm1," MEMACCESS2(0x10,1) " \n" - "lea " MEMLEA(0x20,1) ",%1 \n" - "jg 1b \n" - : "+r"(src_argb), // %0 - "+r"(dst_argb), // %1 - "+r"(pix) // %2 - : "r"(shuffler) // %3 - : "memory", "cc" -#if defined(__SSE2__) - , "xmm0", "xmm1", "xmm5" -#endif - ); -} #endif // HAS_ARGBSHUFFLEROW_SSSE3 #ifdef HAS_ARGBSHUFFLEROW_AVX2 diff --git a/source/row_win.cc b/source/row_win.cc index 9996ccd61..a5b99d7b0 100644 --- a/source/row_win.cc +++ b/source/row_win.cc @@ -126,58 +126,6 @@ void I422ToARGBRow_SSSE3(const uint8* y_buf, } } -// Unaligned destination version. -void I422ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_argb, - int width) { - __m128i xmm0, xmm1, xmm2, xmm3; - const __m128i xmm5 = _mm_set1_epi8(-1); - const __m128i xmm4 = _mm_setzero_si128(); - const ptrdiff_t offset = (uint8*)v_buf - (uint8*)u_buf; - - while (width > 0) { - xmm0 = _mm_cvtsi32_si128(*(uint32*)u_buf); - xmm1 = _mm_cvtsi32_si128(*(uint32*)(u_buf + offset)); - xmm0 = _mm_unpacklo_epi8(xmm0, xmm1); - xmm0 = _mm_unpacklo_epi16(xmm0, xmm0); - xmm1 = _mm_load_si128(&xmm0); - xmm2 = _mm_load_si128(&xmm0); - xmm0 = _mm_maddubs_epi16(xmm0, *(__m128i*)kUVToB); - xmm1 = _mm_maddubs_epi16(xmm1, *(__m128i*)kUVToG); - xmm2 = _mm_maddubs_epi16(xmm2, *(__m128i*)kUVToR); - xmm0 = _mm_sub_epi16(xmm0, *(__m128i*)kUVBiasB); - xmm1 = _mm_sub_epi16(xmm1, *(__m128i*)kUVBiasG); - xmm2 = _mm_sub_epi16(xmm2, *(__m128i*)kUVBiasR); - xmm3 = _mm_loadl_epi64((__m128i*)y_buf); - xmm3 = _mm_unpacklo_epi8(xmm3, xmm4); - xmm3 = _mm_subs_epi16(xmm3, *(__m128i*)kYSub16); - xmm3 = _mm_mullo_epi16(xmm3, *(__m128i*)kYToRgb); - xmm0 = _mm_adds_epi16(xmm0, xmm3); - xmm1 = _mm_adds_epi16(xmm1, xmm3); - xmm2 = _mm_adds_epi16(xmm2, xmm3); - xmm0 = _mm_srai_epi16(xmm0, 6); - xmm1 = _mm_srai_epi16(xmm1, 6); - xmm2 = _mm_srai_epi16(xmm2, 6); - xmm0 = _mm_packus_epi16(xmm0, xmm0); - xmm1 = _mm_packus_epi16(xmm1, xmm1); - xmm2 = _mm_packus_epi16(xmm2, xmm2); - xmm0 = _mm_unpacklo_epi8(xmm0, xmm1); - xmm2 = _mm_unpacklo_epi8(xmm2, xmm5); - xmm1 = _mm_load_si128(&xmm0); - xmm0 = _mm_unpacklo_epi16(xmm0, xmm2); - xmm1 = _mm_unpackhi_epi16(xmm1, xmm2); - - _mm_storeu_si128((__m128i *)dst_argb, xmm0); - _mm_storeu_si128((__m128i *)(dst_argb + 16), xmm1); - - y_buf += 8; - u_buf += 4; - dst_argb += 32; - width -= 8; - } -} // 32 bit #else // defined(_M_X64) @@ -335,35 +283,6 @@ void I400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix) { } } -__declspec(naked) __declspec(align(16)) -void I400ToARGBRow_Unaligned_SSE2(const uint8* src_y, uint8* dst_argb, - int pix) { - __asm { - mov eax, [esp + 4] // src_y - mov edx, [esp + 8] // dst_argb - mov ecx, [esp + 12] // pix - pcmpeqb xmm5, xmm5 // generate mask 0xff000000 - pslld xmm5, 24 - - align 4 - convertloop: - movq xmm0, qword ptr [eax] - lea eax, [eax + 8] - punpcklbw xmm0, xmm0 - movdqa xmm1, xmm0 - punpcklwd xmm0, xmm0 - punpckhwd xmm1, xmm1 - por xmm0, xmm5 - por xmm1, xmm5 - movdqu [edx], xmm0 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - ret - } -} - __declspec(naked) __declspec(align(16)) void RGB24ToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int pix) { __asm { @@ -936,75 +855,6 @@ void ARGBToYJRow_AVX2(const uint8* src_argb, uint8* dst_y, int pix) { } #endif // HAS_ARGBTOYJROW_AVX2 -__declspec(naked) __declspec(align(16)) -void ARGBToYRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { - __asm { - mov eax, [esp + 4] /* src_argb */ - mov edx, [esp + 8] /* dst_y */ - mov ecx, [esp + 12] /* pix */ - movdqa xmm5, kAddY16 - movdqa xmm4, kARGBToY - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - pmaddubsw xmm0, xmm4 - pmaddubsw xmm1, xmm4 - pmaddubsw xmm2, xmm4 - pmaddubsw xmm3, xmm4 - lea eax, [eax + 64] - phaddw xmm0, xmm1 - phaddw xmm2, xmm3 - psrlw xmm0, 7 - psrlw xmm2, 7 - packuswb xmm0, xmm2 - paddb xmm0, xmm5 - sub ecx, 16 - movdqu [edx], xmm0 - lea edx, [edx + 16] - jg convertloop - ret - } -} - -__declspec(naked) __declspec(align(16)) -void ARGBToYJRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { - __asm { - mov eax, [esp + 4] /* src_argb */ - mov edx, [esp + 8] /* dst_y */ - mov ecx, [esp + 12] /* pix */ - movdqa xmm4, kARGBToYJ - movdqa xmm5, kAddYJ64 - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - pmaddubsw xmm0, xmm4 - pmaddubsw xmm1, xmm4 - pmaddubsw xmm2, xmm4 - pmaddubsw xmm3, xmm4 - lea eax, [eax + 64] - phaddw xmm0, xmm1 - phaddw xmm2, xmm3 - paddw xmm0, xmm5 - paddw xmm2, xmm5 - psrlw xmm0, 7 - psrlw xmm2, 7 - packuswb xmm0, xmm2 - sub ecx, 16 - movdqu [edx], xmm0 - lea edx, [edx + 16] - jg convertloop - ret - } -} - __declspec(naked) __declspec(align(16)) void BGRAToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { __asm { @@ -1039,40 +889,6 @@ void BGRAToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { } } -__declspec(naked) __declspec(align(16)) -void BGRAToYRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { - __asm { - mov eax, [esp + 4] /* src_argb */ - mov edx, [esp + 8] /* dst_y */ - mov ecx, [esp + 12] /* pix */ - movdqa xmm5, kAddY16 - movdqa xmm4, kBGRAToY - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - pmaddubsw xmm0, xmm4 - pmaddubsw xmm1, xmm4 - pmaddubsw xmm2, xmm4 - pmaddubsw xmm3, xmm4 - lea eax, [eax + 64] - phaddw xmm0, xmm1 - phaddw xmm2, xmm3 - psrlw xmm0, 7 - psrlw xmm2, 7 - packuswb xmm0, xmm2 - paddb xmm0, xmm5 - sub ecx, 16 - movdqu [edx], xmm0 - lea edx, [edx + 16] - jg convertloop - ret - } -} - __declspec(naked) __declspec(align(16)) void ABGRToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { __asm { @@ -1107,40 +923,6 @@ void ABGRToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { } } -__declspec(naked) __declspec(align(16)) -void ABGRToYRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { - __asm { - mov eax, [esp + 4] /* src_argb */ - mov edx, [esp + 8] /* dst_y */ - mov ecx, [esp + 12] /* pix */ - movdqa xmm5, kAddY16 - movdqa xmm4, kABGRToY - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - pmaddubsw xmm0, xmm4 - pmaddubsw xmm1, xmm4 - pmaddubsw xmm2, xmm4 - pmaddubsw xmm3, xmm4 - lea eax, [eax + 64] - phaddw xmm0, xmm1 - phaddw xmm2, xmm3 - psrlw xmm0, 7 - psrlw xmm2, 7 - packuswb xmm0, xmm2 - paddb xmm0, xmm5 - sub ecx, 16 - movdqu [edx], xmm0 - lea edx, [edx + 16] - jg convertloop - ret - } -} - __declspec(naked) __declspec(align(16)) void RGBAToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { __asm { @@ -1175,40 +957,6 @@ void RGBAToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { } } -__declspec(naked) __declspec(align(16)) -void RGBAToYRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) { - __asm { - mov eax, [esp + 4] /* src_argb */ - mov edx, [esp + 8] /* dst_y */ - mov ecx, [esp + 12] /* pix */ - movdqa xmm5, kAddY16 - movdqa xmm4, kRGBAToY - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - pmaddubsw xmm0, xmm4 - pmaddubsw xmm1, xmm4 - pmaddubsw xmm2, xmm4 - pmaddubsw xmm3, xmm4 - lea eax, [eax + 64] - phaddw xmm0, xmm1 - phaddw xmm2, xmm3 - psrlw xmm0, 7 - psrlw xmm2, 7 - packuswb xmm0, xmm2 - paddb xmm0, xmm5 - sub ecx, 16 - movdqu [edx], xmm0 - lea edx, [edx + 16] - jg convertloop - ret - } -} - __declspec(naked) __declspec(align(16)) void ARGBToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width) { @@ -1409,147 +1157,6 @@ void ARGBToUVRow_AVX2(const uint8* src_argb0, int src_stride_argb, } #endif // HAS_ARGBTOUVROW_AVX2 -__declspec(naked) __declspec(align(16)) -void ARGBToUVRow_Unaligned_SSSE3(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] // pix - movdqa xmm7, kARGBToU - movdqa xmm6, kARGBToV - movdqa xmm5, kAddUV128 - sub edi, edx // stride from u to v - - align 4 - convertloop: - /* step 1 - subsample 16x2 argb pixels to 8x1 */ - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - movdqu xmm4, [eax + esi] - pavgb xmm0, xmm4 - movdqu xmm4, [eax + esi + 16] - pavgb xmm1, xmm4 - movdqu xmm4, [eax + esi + 32] - pavgb xmm2, xmm4 - movdqu xmm4, [eax + esi + 48] - pavgb xmm3, xmm4 - 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 - sub ecx, 16 - movlps qword ptr [edx], xmm0 // U - movhps qword ptr [edx + edi], xmm0 // V - lea edx, [edx + 8] - jg convertloop - - pop edi - pop esi - ret - } -} - -__declspec(naked) __declspec(align(16)) -void ARGBToUVJRow_Unaligned_SSSE3(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] // pix - movdqa xmm7, kARGBToUJ - movdqa xmm6, kARGBToVJ - movdqa xmm5, kAddUVJ128 - sub edi, edx // stride from u to v - - align 4 - convertloop: - /* step 1 - subsample 16x2 argb pixels to 8x1 */ - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - movdqu xmm4, [eax + esi] - pavgb xmm0, xmm4 - movdqu xmm4, [eax + esi + 16] - pavgb xmm1, xmm4 - movdqu xmm4, [eax + esi + 32] - pavgb xmm2, xmm4 - movdqu xmm4, [eax + esi + 48] - pavgb xmm3, xmm4 - 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 - paddw xmm0, xmm5 // +.5 rounding -> unsigned - paddw xmm1, xmm5 - psraw xmm0, 8 - psraw xmm1, 8 - packsswb xmm0, xmm1 - - // step 3 - store 8 U and 8 V values - sub ecx, 16 - movlps qword ptr [edx], xmm0 // U - movhps qword ptr [edx + edi], xmm0 // V - lea edx, [edx + 8] - jg convertloop - - pop edi - pop esi - ret - } -} - __declspec(naked) __declspec(align(16)) void ARGBToUV444Row_SSSE3(const uint8* src_argb0, uint8* dst_u, uint8* dst_v, int width) { @@ -1608,64 +1215,6 @@ void ARGBToUV444Row_SSSE3(const uint8* src_argb0, } } -__declspec(naked) __declspec(align(16)) -void ARGBToUV444Row_Unaligned_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] // pix - movdqa xmm7, kARGBToU - movdqa xmm6, kARGBToV - movdqa xmm5, kAddUV128 - sub edi, edx // stride from u to v - - align 4 - convertloop: - /* convert to U and V */ - movdqu xmm0, [eax] // U - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - pmaddubsw xmm0, xmm7 - pmaddubsw xmm1, xmm7 - pmaddubsw xmm2, xmm7 - pmaddubsw xmm3, xmm7 - phaddw xmm0, xmm1 - phaddw xmm2, xmm3 - psraw xmm0, 8 - psraw xmm2, 8 - packsswb xmm0, xmm2 - paddb xmm0, xmm5 - sub ecx, 16 - movdqu [edx], xmm0 - - movdqu xmm0, [eax] // V - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - pmaddubsw xmm0, xmm6 - pmaddubsw xmm1, xmm6 - pmaddubsw xmm2, xmm6 - pmaddubsw xmm3, xmm6 - phaddw xmm0, xmm1 - phaddw xmm2, xmm3 - psraw xmm0, 8 - psraw xmm2, 8 - packsswb xmm0, xmm2 - paddb xmm0, xmm5 - lea eax, [eax + 64] - movdqu [edx + edi], xmm0 - lea edx, [edx + 16] - jg convertloop - - pop edi - ret - } -} - __declspec(naked) __declspec(align(16)) void ARGBToUV422Row_SSSE3(const uint8* src_argb0, uint8* dst_u, uint8* dst_v, int width) { @@ -1725,65 +1274,6 @@ void ARGBToUV422Row_SSSE3(const uint8* src_argb0, } } -__declspec(naked) __declspec(align(16)) -void ARGBToUV422Row_Unaligned_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] // pix - movdqa xmm7, kARGBToU - movdqa xmm6, kARGBToV - movdqa xmm5, kAddUV128 - sub edi, edx // stride from u to v - - align 4 - 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 - sub ecx, 16 - movlps qword ptr [edx], xmm0 // U - movhps qword ptr [edx + edi], xmm0 // V - lea edx, [edx + 8] - jg convertloop - - pop edi - ret - } -} - __declspec(naked) __declspec(align(16)) void BGRAToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width) { @@ -1850,76 +1340,6 @@ void BGRAToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, } } -__declspec(naked) __declspec(align(16)) -void BGRAToUVRow_Unaligned_SSSE3(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] // pix - movdqa xmm7, kBGRAToU - movdqa xmm6, kBGRAToV - movdqa xmm5, kAddUV128 - sub edi, edx // stride from u to v - - align 4 - convertloop: - /* step 1 - subsample 16x2 argb pixels to 8x1 */ - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - movdqu xmm4, [eax + esi] - pavgb xmm0, xmm4 - movdqu xmm4, [eax + esi + 16] - pavgb xmm1, xmm4 - movdqu xmm4, [eax + esi + 32] - pavgb xmm2, xmm4 - movdqu xmm4, [eax + esi + 48] - pavgb xmm3, xmm4 - 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 - sub ecx, 16 - movlps qword ptr [edx], xmm0 // U - movhps qword ptr [edx + edi], xmm0 // V - lea edx, [edx + 8] - jg convertloop - - pop edi - pop esi - ret - } -} - __declspec(naked) __declspec(align(16)) void ABGRToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width) { @@ -1986,76 +1406,6 @@ void ABGRToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, } } -__declspec(naked) __declspec(align(16)) -void ABGRToUVRow_Unaligned_SSSE3(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] // pix - movdqa xmm7, kABGRToU - movdqa xmm6, kABGRToV - movdqa xmm5, kAddUV128 - sub edi, edx // stride from u to v - - align 4 - convertloop: - /* step 1 - subsample 16x2 argb pixels to 8x1 */ - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - movdqu xmm4, [eax + esi] - pavgb xmm0, xmm4 - movdqu xmm4, [eax + esi + 16] - pavgb xmm1, xmm4 - movdqu xmm4, [eax + esi + 32] - pavgb xmm2, xmm4 - movdqu xmm4, [eax + esi + 48] - pavgb xmm3, xmm4 - 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 - sub ecx, 16 - movlps qword ptr [edx], xmm0 // U - movhps qword ptr [edx + edi], xmm0 // V - lea edx, [edx + 8] - jg convertloop - - pop edi - pop esi - ret - } -} - __declspec(naked) __declspec(align(16)) void RGBAToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, uint8* dst_u, uint8* dst_v, int width) { @@ -2122,75 +1472,6 @@ void RGBAToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb, } } -__declspec(naked) __declspec(align(16)) -void RGBAToUVRow_Unaligned_SSSE3(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] // pix - movdqa xmm7, kRGBAToU - movdqa xmm6, kRGBAToV - movdqa xmm5, kAddUV128 - sub edi, edx // stride from u to v - - align 4 - convertloop: - /* step 1 - subsample 16x2 argb pixels to 8x1 */ - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + 32] - movdqu xmm3, [eax + 48] - movdqu xmm4, [eax + esi] - pavgb xmm0, xmm4 - movdqu xmm4, [eax + esi + 16] - pavgb xmm1, xmm4 - movdqu xmm4, [eax + esi + 32] - pavgb xmm2, xmm4 - movdqu xmm4, [eax + esi + 48] - pavgb xmm3, xmm4 - 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 - sub ecx, 16 - movlps qword ptr [edx], xmm0 // U - movhps qword ptr [edx + edi], xmm0 // V - lea edx, [edx + 8] - jg convertloop - - pop edi - pop esi - ret - } -} #endif // HAS_ARGBTOYROW_SSSE3 #ifdef HAS_I422TOARGBROW_AVX2 @@ -2529,7 +1810,7 @@ void I422ToRAWRow_SSSE3(const uint8* y_buf, } } -// 8 pixels, dest unaligned. +// 8 pixels // 4 UV values upsampled to 8 UV, mixed with 8 Y producing 8 ARGB (32 bytes). __declspec(naked) __declspec(align(16)) void I422ToRGB565Row_SSSE3(const uint8* y_buf, @@ -2767,214 +2048,6 @@ void NV21ToARGBRow_SSSE3(const uint8* y_buf, } } -// 8 pixels, unaligned. -// 8 UV values, mixed with 8 Y producing 8 ARGB (32 bytes). -__declspec(naked) __declspec(align(16)) -void I444ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_argb, - int width) { - __asm { - push esi - push edi - mov eax, [esp + 8 + 4] // Y - mov esi, [esp + 8 + 8] // U - mov edi, [esp + 8 + 12] // V - mov edx, [esp + 8 + 16] // argb - mov ecx, [esp + 8 + 20] // width - sub edi, esi - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - pxor xmm4, xmm4 - - align 4 - convertloop: - READYUV444 - YUVTORGB - - // Step 3: Weave into ARGB - punpcklbw xmm0, xmm1 // BG - punpcklbw xmm2, xmm5 // RA - movdqa xmm1, xmm0 - punpcklwd xmm0, xmm2 // BGRA first 4 pixels - punpckhwd xmm1, xmm2 // BGRA next 4 pixels - movdqu [edx], xmm0 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop edi - pop esi - ret - } -} - -// 8 pixels, unaligned. -// 4 UV values upsampled to 8 UV, mixed with 8 Y producing 8 ARGB (32 bytes). -__declspec(naked) __declspec(align(16)) -void I422ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_argb, - int width) { - __asm { - push esi - push edi - mov eax, [esp + 8 + 4] // Y - mov esi, [esp + 8 + 8] // U - mov edi, [esp + 8 + 12] // V - mov edx, [esp + 8 + 16] // argb - mov ecx, [esp + 8 + 20] // width - sub edi, esi - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - pxor xmm4, xmm4 - - align 4 - convertloop: - READYUV422 - YUVTORGB - - // Step 3: Weave into ARGB - punpcklbw xmm0, xmm1 // BG - punpcklbw xmm2, xmm5 // RA - movdqa xmm1, xmm0 - punpcklwd xmm0, xmm2 // BGRA first 4 pixels - punpckhwd xmm1, xmm2 // BGRA next 4 pixels - movdqu [edx], xmm0 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop edi - pop esi - ret - } -} - -// 8 pixels, unaligned. -// 2 UV values upsampled to 8 UV, mixed with 8 Y producing 8 ARGB (32 bytes). -// Similar to I420 but duplicate UV once more. -__declspec(naked) __declspec(align(16)) -void I411ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_argb, - int width) { - __asm { - push ebx - push esi - push edi - mov eax, [esp + 12 + 4] // Y - mov esi, [esp + 12 + 8] // U - mov edi, [esp + 12 + 12] // V - mov edx, [esp + 12 + 16] // argb - mov ecx, [esp + 12 + 20] // width - sub edi, esi - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - pxor xmm4, xmm4 - - align 4 - convertloop: - READYUV411 // modifies EBX - YUVTORGB - - // Step 3: Weave into ARGB - punpcklbw xmm0, xmm1 // BG - punpcklbw xmm2, xmm5 // RA - movdqa xmm1, xmm0 - punpcklwd xmm0, xmm2 // BGRA first 4 pixels - punpckhwd xmm1, xmm2 // BGRA next 4 pixels - movdqu [edx], xmm0 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop edi - pop esi - pop ebx - ret - } -} - -// 8 pixels, dest aligned 16. -// 4 UV values upsampled to 8 UV, mixed with 8 Y producing 8 ARGB (32 bytes). -__declspec(naked) __declspec(align(16)) -void NV12ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* uv_buf, - uint8* dst_argb, - int width) { - __asm { - push esi - mov eax, [esp + 4 + 4] // Y - mov esi, [esp + 4 + 8] // UV - mov edx, [esp + 4 + 12] // argb - mov ecx, [esp + 4 + 16] // width - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - pxor xmm4, xmm4 - - align 4 - convertloop: - READNV12 - YUVTORGB - - // Step 3: Weave into ARGB - punpcklbw xmm0, xmm1 // BG - punpcklbw xmm2, xmm5 // RA - movdqa xmm1, xmm0 - punpcklwd xmm0, xmm2 // BGRA first 4 pixels - punpckhwd xmm1, xmm2 // BGRA next 4 pixels - movdqu [edx], xmm0 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop esi - ret - } -} - -// 8 pixels, dest aligned 16. -// 4 UV values upsampled to 8 UV, mixed with 8 Y producing 8 ARGB (32 bytes). -__declspec(naked) __declspec(align(16)) -void NV21ToARGBRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* uv_buf, - uint8* dst_argb, - int width) { - __asm { - push esi - mov eax, [esp + 4 + 4] // Y - mov esi, [esp + 4 + 8] // VU - mov edx, [esp + 4 + 12] // argb - mov ecx, [esp + 4 + 16] // width - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - pxor xmm4, xmm4 - - align 4 - convertloop: - READNV12 - YVUTORGB - - // Step 3: Weave into ARGB - punpcklbw xmm0, xmm1 // BG - punpcklbw xmm2, xmm5 // RA - movdqa xmm1, xmm0 - punpcklwd xmm0, xmm2 // BGRA first 4 pixels - punpckhwd xmm1, xmm2 // BGRA next 4 pixels - movdqu [edx], xmm0 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop esi - ret - } -} - __declspec(naked) __declspec(align(16)) void I422ToBGRARow_SSSE3(const uint8* y_buf, const uint8* u_buf, @@ -3016,47 +2089,6 @@ void I422ToBGRARow_SSSE3(const uint8* y_buf, } } -__declspec(naked) __declspec(align(16)) -void I422ToBGRARow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_bgra, - int width) { - __asm { - push esi - push edi - mov eax, [esp + 8 + 4] // Y - mov esi, [esp + 8 + 8] // U - mov edi, [esp + 8 + 12] // V - mov edx, [esp + 8 + 16] // bgra - mov ecx, [esp + 8 + 20] // width - sub edi, esi - pxor xmm4, xmm4 - - align 4 - convertloop: - READYUV422 - YUVTORGB - - // Step 3: Weave into BGRA - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - punpcklbw xmm1, xmm0 // GB - punpcklbw xmm5, xmm2 // AR - movdqa xmm0, xmm5 - punpcklwd xmm5, xmm1 // BGRA first 4 pixels - punpckhwd xmm0, xmm1 // BGRA next 4 pixels - movdqu [edx], xmm5 - movdqu [edx + 16], xmm0 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop edi - pop esi - ret - } -} - __declspec(naked) __declspec(align(16)) void I422ToABGRRow_SSSE3(const uint8* y_buf, const uint8* u_buf, @@ -3098,47 +2130,6 @@ void I422ToABGRRow_SSSE3(const uint8* y_buf, } } -__declspec(naked) __declspec(align(16)) -void I422ToABGRRow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_abgr, - int width) { - __asm { - push esi - push edi - mov eax, [esp + 8 + 4] // Y - mov esi, [esp + 8 + 8] // U - mov edi, [esp + 8 + 12] // V - mov edx, [esp + 8 + 16] // abgr - mov ecx, [esp + 8 + 20] // width - sub edi, esi - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - pxor xmm4, xmm4 - - align 4 - convertloop: - READYUV422 - YUVTORGB - - // Step 3: Weave into ARGB - punpcklbw xmm2, xmm1 // RG - punpcklbw xmm0, xmm5 // BA - movdqa xmm1, xmm2 - punpcklwd xmm2, xmm0 // RGBA first 4 pixels - punpckhwd xmm1, xmm0 // RGBA next 4 pixels - movdqu [edx], xmm2 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop edi - pop esi - ret - } -} - __declspec(naked) __declspec(align(16)) void I422ToRGBARow_SSSE3(const uint8* y_buf, const uint8* u_buf, @@ -3180,47 +2171,6 @@ void I422ToRGBARow_SSSE3(const uint8* y_buf, } } -__declspec(naked) __declspec(align(16)) -void I422ToRGBARow_Unaligned_SSSE3(const uint8* y_buf, - const uint8* u_buf, - const uint8* v_buf, - uint8* dst_rgba, - int width) { - __asm { - push esi - push edi - mov eax, [esp + 8 + 4] // Y - mov esi, [esp + 8 + 8] // U - mov edi, [esp + 8 + 12] // V - mov edx, [esp + 8 + 16] // rgba - mov ecx, [esp + 8 + 20] // width - sub edi, esi - pxor xmm4, xmm4 - - align 4 - convertloop: - READYUV422 - YUVTORGB - - // Step 3: Weave into RGBA - pcmpeqb xmm5, xmm5 // generate 0xffffffff for alpha - punpcklbw xmm1, xmm2 // GR - punpcklbw xmm5, xmm0 // AB - movdqa xmm0, xmm5 - punpcklwd xmm5, xmm1 // RGBA first 4 pixels - punpckhwd xmm0, xmm1 // RGBA next 4 pixels - movdqu [edx], xmm5 - movdqu [edx + 16], xmm0 - lea edx, [edx + 32] - sub ecx, 8 - jg convertloop - - pop edi - pop esi - ret - } -} - #endif // HAS_I422TOARGBROW_SSSE3 #ifdef HAS_YTOARGBROW_SSE2 @@ -3331,8 +2281,6 @@ void MirrorRow_AVX2(const uint8* src, uint8* dst, int width) { #endif // HAS_MIRRORROW_AVX2 #ifdef HAS_MIRRORROW_SSE2 -// SSE2 version has movdqu so it can be used on unaligned buffers when SSSE3 -// version can not. __declspec(naked) __declspec(align(16)) void MirrorRow_SSE2(const uint8* src, uint8* dst, int width) { __asm { @@ -3490,42 +2438,6 @@ void SplitUVRow_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix) { } } -__declspec(naked) __declspec(align(16)) -void SplitUVRow_Unaligned_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, - int pix) { - __asm { - push edi - mov eax, [esp + 4 + 4] // src_uv - mov edx, [esp + 4 + 8] // dst_u - mov edi, [esp + 4 + 12] // dst_v - mov ecx, [esp + 4 + 16] // pix - pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff - psrlw xmm5, 8 - sub edi, edx - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - lea eax, [eax + 32] - movdqa xmm2, xmm0 - movdqa xmm3, xmm1 - pand xmm0, xmm5 // even bytes - pand xmm1, xmm5 - packuswb xmm0, xmm1 - psrlw xmm2, 8 // odd bytes - psrlw xmm3, 8 - packuswb xmm2, xmm3 - movdqu [edx], xmm0 - movdqu [edx + edi], xmm2 - lea edx, [edx + 16] - sub ecx, 16 - jg convertloop - - pop edi - ret - } -} #endif // HAS_SPLITUVROW_SSE2 #ifdef HAS_SPLITUVROW_AVX2 @@ -3597,36 +2509,6 @@ void MergeUVRow_SSE2(const uint8* src_u, const uint8* src_v, uint8* dst_uv, ret } } - -__declspec(naked) __declspec(align(16)) -void MergeUVRow_Unaligned_SSE2(const uint8* src_u, const uint8* src_v, - uint8* dst_uv, int width) { - __asm { - push edi - mov eax, [esp + 4 + 4] // src_u - mov edx, [esp + 4 + 8] // src_v - mov edi, [esp + 4 + 12] // dst_uv - mov ecx, [esp + 4 + 16] // width - sub edx, eax - - align 4 - convertloop: - movdqu xmm0, [eax] // read 16 U's - movdqu xmm1, [eax + edx] // and 16 V's - lea eax, [eax + 16] - movdqa xmm2, xmm0 - punpcklbw xmm0, xmm1 // first 8 UV pairs - punpckhbw xmm2, xmm1 // next 8 UV pairs - movdqu [edi], xmm0 - movdqu [edi + 16], xmm2 - lea edi, [edi + 32] - sub ecx, 16 - jg convertloop - - pop edi - ret - } -} #endif // HAS_MERGEUVROW_SSE2 #ifdef HAS_MERGEUVROW_AVX2 @@ -3713,7 +2595,7 @@ void CopyRow_AVX(const uint8* src, uint8* dst, int count) { } #endif // HAS_COPYROW_AVX -// Unaligned Multiple of 1. +// Multiple of 1. __declspec(naked) __declspec(align(16)) void CopyRow_ERMS(const uint8* src, uint8* dst, int count) { __asm { @@ -3730,7 +2612,7 @@ void CopyRow_ERMS(const uint8* src, uint8* dst, int count) { } #ifdef HAS_COPYROW_X86 -// Unaligned Multiple of 4. +// Multiple of 4. __declspec(naked) __declspec(align(16)) void CopyRow_X86(const uint8* src, uint8* dst, int count) { __asm { @@ -4268,113 +3150,6 @@ void YUY2ToUV422Row_SSE2(const uint8* src_yuy2, } } -__declspec(naked) __declspec(align(16)) -void YUY2ToYRow_Unaligned_SSE2(const uint8* src_yuy2, - uint8* dst_y, int pix) { - __asm { - mov eax, [esp + 4] // src_yuy2 - mov edx, [esp + 8] // dst_y - mov ecx, [esp + 12] // pix - pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff - psrlw xmm5, 8 - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - lea eax, [eax + 32] - pand xmm0, xmm5 // even bytes are Y - pand xmm1, xmm5 - packuswb xmm0, xmm1 - sub ecx, 16 - movdqu [edx], xmm0 - lea edx, [edx + 16] - jg convertloop - ret - } -} - -__declspec(naked) __declspec(align(16)) -void YUY2ToUVRow_Unaligned_SSE2(const uint8* src_yuy2, int stride_yuy2, - uint8* dst_u, uint8* dst_v, int pix) { - __asm { - push esi - push edi - mov eax, [esp + 8 + 4] // src_yuy2 - mov esi, [esp + 8 + 8] // stride_yuy2 - mov edx, [esp + 8 + 12] // dst_u - mov edi, [esp + 8 + 16] // dst_v - mov ecx, [esp + 8 + 20] // pix - pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff - psrlw xmm5, 8 - sub edi, edx - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + esi] - movdqu xmm3, [eax + esi + 16] - lea eax, [eax + 32] - pavgb xmm0, xmm2 - pavgb xmm1, xmm3 - psrlw xmm0, 8 // YUYV -> UVUV - psrlw xmm1, 8 - packuswb xmm0, xmm1 - movdqa xmm1, xmm0 - pand xmm0, xmm5 // U - packuswb xmm0, xmm0 - psrlw xmm1, 8 // V - packuswb xmm1, xmm1 - movq qword ptr [edx], xmm0 - movq qword ptr [edx + edi], xmm1 - lea edx, [edx + 8] - sub ecx, 16 - jg convertloop - - pop edi - pop esi - ret - } -} - -__declspec(naked) __declspec(align(16)) -void YUY2ToUV422Row_Unaligned_SSE2(const uint8* src_yuy2, - uint8* dst_u, uint8* dst_v, int pix) { - __asm { - push edi - mov eax, [esp + 4 + 4] // src_yuy2 - mov edx, [esp + 4 + 8] // dst_u - mov edi, [esp + 4 + 12] // dst_v - mov ecx, [esp + 4 + 16] // pix - pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff - psrlw xmm5, 8 - sub edi, edx - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - lea eax, [eax + 32] - psrlw xmm0, 8 // YUYV -> UVUV - psrlw xmm1, 8 - packuswb xmm0, xmm1 - movdqa xmm1, xmm0 - pand xmm0, xmm5 // U - packuswb xmm0, xmm0 - psrlw xmm1, 8 // V - packuswb xmm1, xmm1 - movq qword ptr [edx], xmm0 - movq qword ptr [edx + edi], xmm1 - lea edx, [edx + 8] - sub ecx, 16 - jg convertloop - - pop edi - ret - } -} - __declspec(naked) __declspec(align(16)) void UYVYToYRow_SSE2(const uint8* src_uyvy, uint8* dst_y, int pix) { @@ -4479,111 +3254,6 @@ void UYVYToUV422Row_SSE2(const uint8* src_uyvy, ret } } - -__declspec(naked) __declspec(align(16)) -void UYVYToYRow_Unaligned_SSE2(const uint8* src_uyvy, - uint8* dst_y, int pix) { - __asm { - mov eax, [esp + 4] // src_uyvy - mov edx, [esp + 8] // dst_y - mov ecx, [esp + 12] // pix - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - lea eax, [eax + 32] - psrlw xmm0, 8 // odd bytes are Y - psrlw xmm1, 8 - packuswb xmm0, xmm1 - sub ecx, 16 - movdqu [edx], xmm0 - lea edx, [edx + 16] - jg convertloop - ret - } -} - -__declspec(naked) __declspec(align(16)) -void UYVYToUVRow_Unaligned_SSE2(const uint8* src_uyvy, int stride_uyvy, - uint8* dst_u, uint8* dst_v, int pix) { - __asm { - push esi - push edi - mov eax, [esp + 8 + 4] // src_yuy2 - mov esi, [esp + 8 + 8] // stride_yuy2 - mov edx, [esp + 8 + 12] // dst_u - mov edi, [esp + 8 + 16] // dst_v - mov ecx, [esp + 8 + 20] // pix - pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff - psrlw xmm5, 8 - sub edi, edx - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - movdqu xmm2, [eax + esi] - movdqu xmm3, [eax + esi + 16] - lea eax, [eax + 32] - pavgb xmm0, xmm2 - pavgb xmm1, xmm3 - pand xmm0, xmm5 // UYVY -> UVUV - pand xmm1, xmm5 - packuswb xmm0, xmm1 - movdqa xmm1, xmm0 - pand xmm0, xmm5 // U - packuswb xmm0, xmm0 - psrlw xmm1, 8 // V - packuswb xmm1, xmm1 - movq qword ptr [edx], xmm0 - movq qword ptr [edx + edi], xmm1 - lea edx, [edx + 8] - sub ecx, 16 - jg convertloop - - pop edi - pop esi - ret - } -} - -__declspec(naked) __declspec(align(16)) -void UYVYToUV422Row_Unaligned_SSE2(const uint8* src_uyvy, - uint8* dst_u, uint8* dst_v, int pix) { - __asm { - push edi - mov eax, [esp + 4 + 4] // src_yuy2 - mov edx, [esp + 4 + 8] // dst_u - mov edi, [esp + 4 + 12] // dst_v - mov ecx, [esp + 4 + 16] // pix - pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff - psrlw xmm5, 8 - sub edi, edx - - align 4 - convertloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - lea eax, [eax + 32] - pand xmm0, xmm5 // UYVY -> UVUV - pand xmm1, xmm5 - packuswb xmm0, xmm1 - movdqa xmm1, xmm0 - pand xmm0, xmm5 // U - packuswb xmm0, xmm0 - psrlw xmm1, 8 // V - packuswb xmm1, xmm1 - movq qword ptr [edx], xmm0 - movq qword ptr [edx + edi], xmm1 - lea edx, [edx + 8] - sub ecx, 16 - jg convertloop - - pop edi - ret - } -} #endif // HAS_YUY2TOYROW_SSE2 #ifdef HAS_ARGBBLENDROW_SSE2 @@ -6624,32 +5294,6 @@ void ARGBShuffleRow_SSSE3(const uint8* src_argb, uint8* dst_argb, } } -__declspec(naked) __declspec(align(16)) -void ARGBShuffleRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_argb, - const uint8* shuffler, int pix) { - __asm { - mov eax, [esp + 4] // src_argb - mov edx, [esp + 8] // dst_argb - mov ecx, [esp + 12] // shuffler - movdqu xmm5, [ecx] - mov ecx, [esp + 16] // pix - - align 4 - wloop: - movdqu xmm0, [eax] - movdqu xmm1, [eax + 16] - lea eax, [eax + 32] - pshufb xmm0, xmm5 - pshufb xmm1, xmm5 - sub ecx, 8 - movdqu [edx], xmm0 - movdqu [edx + 16], xmm1 - lea edx, [edx + 32] - jg wloop - ret - } -} - #ifdef HAS_ARGBSHUFFLEROW_AVX2 __declspec(naked) __declspec(align(16)) void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb, diff --git a/source/scale.cc b/source/scale.cc index f37da7ef6..3083a8391 100644 --- a/source/scale.cc +++ b/source/scale.cc @@ -154,9 +154,7 @@ static void ScalePlaneDown4(int src_width, int src_height, ScaleRowDown4 = filtering ? ScaleRowDown4Box_NEON : ScaleRowDown4_NEON; } #elif defined(HAS_SCALEROWDOWN4_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && - IS_ALIGNED(dst_width, 8) && IS_ALIGNED(row_stride, 16) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleRowDown4 = filtering ? ScaleRowDown4Box_SSE2 : ScaleRowDown4_SSE2; } #elif defined(HAS_SCALEROWDOWN4_MIPS_DSPR2) @@ -198,9 +196,7 @@ static void ScalePlaneDown4_16(int src_width, int src_height, ScaleRowDown4_16_NEON; } #elif defined(HAS_SCALEROWDOWN4_16_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && - IS_ALIGNED(dst_width, 8) && IS_ALIGNED(row_stride, 16) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleRowDown4 = filtering ? ScaleRowDown4Box_16_SSE2 : ScaleRowDown4_16_SSE2; } @@ -256,8 +252,7 @@ static void ScalePlaneDown34(int src_width, int src_height, } #endif #if defined(HAS_SCALEROWDOWN34_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0)) { if (!filtering) { ScaleRowDown34_0 = ScaleRowDown34_SSSE3; ScaleRowDown34_1 = ScaleRowDown34_SSSE3; @@ -336,8 +331,7 @@ static void ScalePlaneDown34_16(int src_width, int src_height, } #endif #if defined(HAS_SCALEROWDOWN34_16_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0)) { if (!filtering) { ScaleRowDown34_0 = ScaleRowDown34_16_SSSE3; ScaleRowDown34_1 = ScaleRowDown34_16_SSSE3; @@ -431,8 +425,7 @@ static void ScalePlaneDown38(int src_width, int src_height, } } #elif defined(HAS_SCALEROWDOWN38_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0)) { if (!filtering) { ScaleRowDown38_3 = ScaleRowDown38_SSSE3; ScaleRowDown38_2 = ScaleRowDown38_SSSE3; @@ -508,8 +501,7 @@ static void ScalePlaneDown38_16(int src_width, int src_height, } } #elif defined(HAS_SCALEROWDOWN38_16_SSSE3) - if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + if (TestCpuFlag(kCpuHasSSSE3) && (dst_width % 24 == 0)) { if (!filtering) { ScaleRowDown38_3 = ScaleRowDown38_16_SSSE3; ScaleRowDown38_2 = ScaleRowDown38_16_SSSE3; @@ -743,11 +735,11 @@ static void ScalePlaneBox(int src_width, int src_height, uint16* dst_ptr, int src_width, int src_height) = ScaleAddRows_C; #if defined(HAS_SCALEADDROWS_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && + if (TestCpuFlag(kCpuHasSSE2) #ifdef AVOID_OVERREAD - IS_ALIGNED(src_width, 16) && + && IS_ALIGNED(src_width, 16) #endif - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + ) { ScaleAddRows = ScaleAddRows_SSE2; } #endif @@ -815,11 +807,11 @@ static void ScalePlaneBox_16(int src_width, int src_height, uint32* dst_ptr, int src_width, int src_height) = ScaleAddRows_16_C; #if defined(HAS_SCALEADDROWS_16_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && + if (TestCpuFlag(kCpuHasSSE2) #ifdef AVOID_OVERREAD - IS_ALIGNED(src_width, 16) && + && IS_ALIGNED(src_width, 16) #endif - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16)) { + ) { ScaleAddRows = ScaleAddRows_16_SSE2; } #endif @@ -1111,9 +1103,7 @@ void ScalePlaneBilinearUp(int src_width, int src_height, if (!filtering && src_width * 2 == dst_width && x < 0x8000) { ScaleFilterCols = ScaleColsUp2_C; #if defined(HAS_SCALECOLS_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst_ptr, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleFilterCols = ScaleColsUp2_SSE2; } #endif @@ -1244,9 +1234,7 @@ void ScalePlaneBilinearUp_16(int src_width, int src_height, if (!filtering && src_width * 2 == dst_width && x < 0x8000) { ScaleFilterCols = ScaleColsUp2_16_C; #if defined(HAS_SCALECOLS_16_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst_ptr, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleFilterCols = ScaleColsUp2_16_SSE2; } #endif @@ -1327,9 +1315,7 @@ static void ScalePlaneSimple(int src_width, int src_height, if (src_width * 2 == dst_width && x < 0x8000) { ScaleCols = ScaleColsUp2_C; #if defined(HAS_SCALECOLS_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst_ptr, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleCols = ScaleColsUp2_SSE2; } #endif @@ -1362,9 +1348,7 @@ static void ScalePlaneSimple_16(int src_width, int src_height, if (src_width * 2 == dst_width && x < 0x8000) { ScaleCols = ScaleColsUp2_16_C; #if defined(HAS_SCALECOLS_16_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_ptr, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst_ptr, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleCols = ScaleColsUp2_16_SSE2; } #endif diff --git a/source/scale_argb.cc b/source/scale_argb.cc index b6d512984..7e081ff07 100644 --- a/source/scale_argb.cc +++ b/source/scale_argb.cc @@ -53,16 +53,13 @@ static void ScaleARGBDown2(int src_width, int src_height, } #if defined(HAS_SCALEARGBROWDOWN2_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4) && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(row_stride, 16) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4)) { ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_SSE2 : (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_SSE2 : ScaleARGBRowDown2Box_SSE2); } #elif defined(HAS_SCALEARGBROWDOWN2_NEON) - if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_argb, 4) && IS_ALIGNED(row_stride, 4)) { + if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 8)) { ScaleARGBRowDown2 = filtering ? ScaleARGBRowDown2Box_NEON : ScaleARGBRowDown2_NEON; } @@ -98,14 +95,11 @@ static void ScaleARGBDown4Box(int src_width, int src_height, assert(dx == 65536 * 4); // Test scale factor of 4. assert((dy & 0x3ffff) == 0); // Test vertical scale is multiple of 4. #if defined(HAS_SCALEARGBROWDOWN2_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4) && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(row_stride, 16) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4)) { ScaleARGBRowDown2 = ScaleARGBRowDown2Box_SSE2; } #elif defined(HAS_SCALEARGBROWDOWN2_NEON) - if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_argb, 4) && IS_ALIGNED(row_stride, 4)) { + if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 8)) { ScaleARGBRowDown2 = ScaleARGBRowDown2Box_NEON; } #endif @@ -139,14 +133,12 @@ static void ScaleARGBDownEven(int src_width, int src_height, assert(IS_ALIGNED(src_height, 2)); src_argb += (y >> 16) * src_stride + (x >> 16) * 4; #if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4)) { ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_SSE2 : ScaleARGBRowDownEven_SSE2; } #elif defined(HAS_SCALEARGBROWDOWNEVEN_NEON) - if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 4) && - IS_ALIGNED(src_argb, 4)) { + if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 4)) { ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_NEON : ScaleARGBRowDownEven_NEON; } @@ -334,9 +326,7 @@ static void ScaleARGBBilinearUp(int src_width, int src_height, if (!filtering && src_width * 2 == dst_width && x < 0x8000) { ScaleARGBFilterCols = ScaleARGBColsUp2_C; #if defined(HAS_SCALEARGBCOLSUP2_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleARGBFilterCols = ScaleARGBColsUp2_SSE2; } #endif @@ -510,9 +500,7 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height, if (!filtering && src_width * 2 == dst_width && x < 0x8000) { ScaleARGBFilterCols = ScaleARGBColsUp2_C; #if defined(HAS_SCALEARGBCOLSUP2_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleARGBFilterCols = ScaleARGBColsUp2_SSE2; } #endif @@ -619,9 +607,7 @@ static void ScaleARGBSimple(int src_width, int src_height, if (src_width * 2 == dst_width && x < 0x8000) { ScaleARGBCols = ScaleARGBColsUp2_C; #if defined(HAS_SCALEARGBCOLSUP2_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) && - IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16) && - IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) { + if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) { ScaleARGBCols = ScaleARGBColsUp2_SSE2; } #endif