diff options
| author |  commit-bot@chromium.org <commit-bot@chromium.org@2bbb7eff-a529-9590-31e7-b0007b416f81> | 2014-04-25 02:08:31 +0000 |
|---|---|---|
| committer | commit-bot@chromium.org <commit-bot@chromium.org@2bbb7eff-a529-9590-31e7-b0007b416f81> | 2014-04-25 02:08:31 +0000 |
| commit | 7bf10152b129e3b0cad76f2abd5136ccbc74a393 (patch) | |
| tree | 93d7cc2375457a42a3d0f2acb5f38d8c4d662b3f /src | |
| parent | b950c6fd7188d625ed64ee6a3ea4f66d6d52e10c (diff) | |
Xfermode: SSE2 implementation of overlay_modeproc
With SSE2 optimization, performance of Xfermode_Overlay will improve
about 35% on desktop i7-3770. Here are the data:
before:
Xfermode_Overlay 8888: cmsecs = 44.17 565: cmsecs = 59.27
after:
Xfermode_Overlay 8888: cmsecs = 28.30 565: cmsecs = 35.84
BUG=skia:
R=mtklein@google.com
Author: qiankun.miao@intel.com
Review URL: https://codereview.chromium.org/232783002
git-svn-id: http://skia.googlecode.com/svn/trunk@14370 2bbb7eff-a529-9590-31e7-b0007b416f81
Diffstat (limited to 'src')
| -rw-r--r-- | src/opts/SkColor_opts_SSE2.h | 15 | ||||
| -rw-r--r-- | src/opts/SkXfermode_opts_SSE2.cpp | 39 |
2 files changed, 53 insertions, 1 deletions
diff --git a/src/opts/SkColor_opts_SSE2.h b/src/opts/SkColor_opts_SSE2.h index 960c48a023..b06fe1a76f 100644 --- a/src/opts/SkColor_opts_SSE2.h +++ b/src/opts/SkColor_opts_SSE2.h @@ -10,6 +10,21 @@ #include <emmintrin.h> +// Because no _mm_mul_epi32() in SSE2, we emulate it here. +// Multiplies 4 32-bit integers from a by 4 32-bit intergers from b. +// The 4 multiplication results should be represented within 32-bit +// integers, otherwise they would be overflow. +static inline __m128i Multiply32_SSE2(const __m128i& a, const __m128i& b) { + // Calculate results of a0 * b0 and a2 * b2. + __m128i r1 = _mm_mul_epu32(a, b); + // Calculate results of a1 * b1 and a3 * b3. + __m128i r2 = _mm_mul_epu32(_mm_srli_si128(a, 4), _mm_srli_si128(b, 4)); + // Shuffle results to [63..0] and interleave the results. + __m128i r = _mm_unpacklo_epi32(_mm_shuffle_epi32(r1, _MM_SHUFFLE(0,0,2,0)), + _mm_shuffle_epi32(r2, _MM_SHUFFLE(0,0,2,0))); + return r; +} + static inline __m128i SkAlpha255To256_SSE2(const __m128i& alpha) { return _mm_add_epi32(alpha, _mm_set1_epi32(1)); } diff --git a/src/opts/SkXfermode_opts_SSE2.cpp b/src/opts/SkXfermode_opts_SSE2.cpp index d1a2632fdb..817f777bc4 100644 --- a/src/opts/SkXfermode_opts_SSE2.cpp +++ b/src/opts/SkXfermode_opts_SSE2.cpp @@ -229,6 +229,43 @@ static __m128i screen_modeproc_SSE2(const __m128i& src, const __m128i& dst) { return SkPackARGB32_SSE2(a, r, g, b); } +// Portable version overlay_byte() is in SkXfermode.cpp. +static inline __m128i overlay_byte_SSE2(const __m128i& sc, const __m128i& dc, + const __m128i& sa, const __m128i& da) { + __m128i ida = _mm_sub_epi32(_mm_set1_epi32(255), da); + __m128i tmp1 = _mm_mullo_epi16(sc, ida); + __m128i isa = _mm_sub_epi32(_mm_set1_epi32(255), sa); + __m128i tmp2 = _mm_mullo_epi16(dc, isa); + __m128i tmp = _mm_add_epi32(tmp1, tmp2); + + __m128i cmp = _mm_cmpgt_epi32(_mm_slli_epi32(dc, 1), da); + __m128i rc1 = _mm_slli_epi32(sc, 1); // 2 * sc + rc1 = Multiply32_SSE2(rc1, dc); // *dc + + __m128i rc2 = _mm_mullo_epi16(sa, da); // sa * da + __m128i tmp3 = _mm_slli_epi32(_mm_sub_epi32(da, dc), 1); // 2 * (da - dc) + tmp3 = Multiply32_SSE2(tmp3, _mm_sub_epi32(sa, sc)); // * (sa - sc) + rc2 = _mm_sub_epi32(rc2, tmp3); + + __m128i rc = _mm_or_si128(_mm_andnot_si128(cmp, rc1), + _mm_and_si128(cmp, rc2)); + return clamp_div255round_SSE2(_mm_add_epi32(rc, tmp)); +} + +static __m128i overlay_modeproc_SSE2(const __m128i& src, const __m128i& dst) { + __m128i sa = SkGetPackedA32_SSE2(src); + __m128i da = SkGetPackedA32_SSE2(dst); + + __m128i a = srcover_byte_SSE2(sa, da); + __m128i r = overlay_byte_SSE2(SkGetPackedR32_SSE2(src), + SkGetPackedR32_SSE2(dst), sa, da); + __m128i g = overlay_byte_SSE2(SkGetPackedG32_SSE2(src), + SkGetPackedG32_SSE2(dst), sa, da); + __m128i b = overlay_byte_SSE2(SkGetPackedB32_SSE2(src), + SkGetPackedB32_SSE2(dst), sa, da); + return SkPackARGB32_SSE2(a, r, g, b); +} + //////////////////////////////////////////////////////////////////////////////// typedef __m128i (*SkXfermodeProcSIMD)(const __m128i& src, const __m128i& dst); @@ -384,7 +421,7 @@ SkXfermodeProcSIMD gSSE2XfermodeProcs[] = { modulate_modeproc_SSE2, screen_modeproc_SSE2, - NULL, // kOverlay_Mode + overlay_modeproc_SSE2, NULL, // kDarken_Mode NULL, // kLighten_Mode NULL, // kColorDodge_Mode |