diff options
author | mtklein <mtklein@chromium.org> | 2015-04-20 10:52:26 -0700 |
---|---|---|
committer | Commit bot <commit-bot@chromium.org> | 2015-04-20 10:52:26 -0700 |
commit | 61221e7f87a99765b0e034020e06bb018e2a08c2 (patch) | |
tree | deca5e11fdb2ffa348e39ee8eb1f5b60d961c437 /src/opts/SkBlitRow_opts_SSE2.cpp | |
parent | 49124378913f3467eb67e653b3b48f80899a3f37 (diff) |
Convert Color32 code to perfect blend.
Before we commit to blend_256_round_alt, let's make sure blend_perfect is
really slower in practice (i.e. regresses on perf.skia.org).
blend_perfect is really the most desirable algorithm if we can afford it. Not
only is it correct, but it's easy to think about and break into correct pieces:
for instance, its div255() doesn't require any coordination with the multiply.
This looks like a 30% hit according to microbenches. That said, microbenches
said my previous change would be a 20-25% perf improvement, but it didn't end
up showing a significant effect at a high level.
As for correctness, I see a bunch of off-by-1 compared to blend_256_round_alt
(exactly what we'd expect), and one off-by-3 in a GM that looks like it has a
bunch of overdraw.
BUG=skia:
Review URL: https://codereview.chromium.org/1098913002
Diffstat (limited to 'src/opts/SkBlitRow_opts_SSE2.cpp')
-rw-r--r-- | src/opts/SkBlitRow_opts_SSE2.cpp | 43 |
1 files changed, 16 insertions, 27 deletions
diff --git a/src/opts/SkBlitRow_opts_SSE2.cpp b/src/opts/SkBlitRow_opts_SSE2.cpp index 59375f1831..3fcb9e0e14 100644 --- a/src/opts/SkBlitRow_opts_SSE2.cpp +++ b/src/opts/SkBlitRow_opts_SSE2.cpp @@ -234,41 +234,30 @@ void S32A_Blend_BlitRow32_SSE2(SkPMColor* SK_RESTRICT dst, #define SK_SUPPORT_LEGACY_COLOR32_MATHx -/* SSE2 version of Color32() - * portable version is in core/SkBlitRow_D32.cpp - */ -// Color32 and its SIMD specializations use the blend_256_round_alt algorithm -// from tests/BlendTest.cpp. It's not quite perfect, but it's never wrong in the -// interesting edge cases, and it's quite a bit faster than blend_perfect. -// -// blend_256_round_alt is our currently blessed algorithm. Please use it or an analogous one. +/* SSE2 version of Color32(), portable version is in core/SkBlitRow_D32.cpp */ +// Color32 and its SIMD specializations use the blend_perfect algorithm from tests/BlendTest.cpp. +// An acceptable alternative is blend_256_round_alt, which is faster but not quite perfect. void Color32_SSE2(SkPMColor dst[], const SkPMColor src[], int count, SkPMColor color) { switch (SkGetPackedA32(color)) { case 0: memmove(dst, src, count * sizeof(SkPMColor)); return; case 255: sk_memset32(dst, color, count); return; } - __m128i colorHigh = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_set1_epi32(color)); -#ifdef SK_SUPPORT_LEGACY_COLOR32_MATH // blend_256_plus1_trunc, busted - __m128i colorAndRound = colorHigh; -#else // blend_256_round_alt, good - __m128i colorAndRound = _mm_add_epi16(colorHigh, _mm_set1_epi16(128)); -#endif - - unsigned invA = 255 - SkGetPackedA32(color); -#ifdef SK_SUPPORT_LEGACY_COLOR32_MATH // blend_256_plus1_trunc, busted - __m128i invA16 = _mm_set1_epi16(invA); -#else // blend_256_round_alt, good - SkASSERT(invA + (invA >> 7) < 256); // We should still fit in the low byte here. - __m128i invA16 = _mm_set1_epi16(invA + (invA >> 7)); -#endif + __m128i color_2x_high = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_set1_epi32(color)), + invA_8x = _mm_set1_epi16(255 - SkGetPackedA32(color)); // Does the core work of blending color onto 4 pixels, returning the resulting 4 pixels. - auto kernel = [&](const __m128i& src4) -> __m128i { - __m128i lo = _mm_mullo_epi16(invA16, _mm_unpacklo_epi8(src4, _mm_setzero_si128())), - hi = _mm_mullo_epi16(invA16, _mm_unpackhi_epi8(src4, _mm_setzero_si128())); - return _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(colorAndRound, lo), 8), - _mm_srli_epi16(_mm_add_epi16(colorAndRound, hi), 8)); + auto kernel = [&](const __m128i& src_4x) -> __m128i { + __m128i lo = _mm_mullo_epi16(invA_8x, _mm_unpacklo_epi8(src_4x, _mm_setzero_si128())), + hi = _mm_mullo_epi16(invA_8x, _mm_unpackhi_epi8(src_4x, _mm_setzero_si128())); + #ifndef SK_SUPPORT_LEGACY_COLOR32_MATH + lo = _mm_add_epi16(lo, _mm_set1_epi16(128)); + hi = _mm_add_epi16(hi, _mm_set1_epi16(128)); + lo = _mm_add_epi16(lo, _mm_srli_epi16(lo, 8)); + hi = _mm_add_epi16(hi, _mm_srli_epi16(hi, 8)); + #endif + return _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(color_2x_high, lo), 8), + _mm_srli_epi16(_mm_add_epi16(color_2x_high, hi), 8)); }; while (count >= 8) { |