diff options
author | 2014-12-02 18:26:09 -0800 | |
---|---|---|
committer | 2014-12-02 18:26:09 -0800 | |
commit | 45a05780867a06b9f8a8d5240cf6c5d5a2c15a35 (patch) | |
tree | 8000137dda38c6d723ddd2ea5519ae7ae8fedf9a /src/opts/SkBitmapProcState_opts_SSSE3.cpp | |
parent | 9503ac7a3cc98327a88eff59c71c451fbdd7e043 (diff) |
Add SSSE3 acceleration for S32_D16_filter_DXDY
With this CL, related nanobench can be improved for 565 config.
bitmap_BGRA_8888_scale_rotate_bilerp 115us -> 70.5us 0.61x
bitmap_BGRA_8888_update_volatile_scale_rotate_bilerp 115us -> 70.5us 0.61x
bitmap_BGRA_8888_update_scale_rotate_bilerp 112us -> 68us 0.6x
BUG=skia:
Review URL: https://codereview.chromium.org/773753002
Diffstat (limited to 'src/opts/SkBitmapProcState_opts_SSSE3.cpp')
-rw-r--r-- | src/opts/SkBitmapProcState_opts_SSSE3.cpp | 34 |
1 files changed, 26 insertions, 8 deletions
diff --git a/src/opts/SkBitmapProcState_opts_SSSE3.cpp b/src/opts/SkBitmapProcState_opts_SSSE3.cpp index 165f1f5ce0..5e7c7b4698 100644 --- a/src/opts/SkBitmapProcState_opts_SSSE3.cpp +++ b/src/opts/SkBitmapProcState_opts_SSSE3.cpp @@ -6,6 +6,7 @@ */ #include "SkBitmapProcState_opts_SSSE3.h" +#include "SkColorPriv.h" #include "SkPaint.h" #include "SkUtils.h" @@ -720,17 +721,28 @@ void S32_alpha_D32_filter_DX_SSSE3(const SkBitmapProcState& s, } void S32_opaque_D32_filter_DXDY_SSSE3(const SkBitmapProcState& s, - const uint32_t* xy, - int count, uint32_t* colors) { + const uint32_t* xy, + int count, uint32_t* colors) { S32_generic_D32_filter_DXDY_SSSE3<false>(s, xy, count, colors); } void S32_alpha_D32_filter_DXDY_SSSE3(const SkBitmapProcState& s, - const uint32_t* xy, - int count, uint32_t* colors) { + const uint32_t* xy, + int count, uint32_t* colors) { S32_generic_D32_filter_DXDY_SSSE3<true>(s, xy, count, colors); } +void S32_D16_filter_DXDY_SSSE3(const SkBitmapProcState& s, + const uint32_t* xy, + int count, uint16_t* colors) { + SkASSERT(64 >= count); + SkAutoSTMalloc<64, uint32_t> colors32(count); + S32_generic_D32_filter_DXDY_SSSE3<false>(s, xy, count, colors32); + for(int i = 0; i < count; i++) { + *colors++ = SkPixel32ToPixel16(colors32[i]); + } +} + #else // SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3 void S32_opaque_D32_filter_DX_SSSE3(const SkBitmapProcState& s, @@ -746,14 +758,20 @@ void S32_alpha_D32_filter_DX_SSSE3(const SkBitmapProcState& s, } void S32_opaque_D32_filter_DXDY_SSSE3(const SkBitmapProcState& s, - const uint32_t* xy, - int count, uint32_t* colors) { + const uint32_t* xy, + int count, uint32_t* colors) { sk_throw(); } void S32_alpha_D32_filter_DXDY_SSSE3(const SkBitmapProcState& s, - const uint32_t* xy, - int count, uint32_t* colors) { + const uint32_t* xy, + int count, uint32_t* colors) { + sk_throw(); +} + +void S32_D16_filter_DXDY_SSSE3(const SkBitmapProcState& s, + const uint32_t* xy, + int count, uint32_t* colors) { sk_throw(); } |