diff options
author | mtklein <mtklein@chromium.org> | 2015-05-12 06:11:21 -0700 |
---|---|---|
committer | Commit bot <commit-bot@chromium.org> | 2015-05-12 06:11:21 -0700 |
commit | d2ffd36eb62e99abe2920369d1e040954cc2044f (patch) | |
tree | 9a1402e26a8b942643f4405f735b8cd38f8864cb /src/opts/Sk4px_SSE2.h | |
parent | 06f3647cad2fad8e60bd9215a1dcee9550586d77 (diff) |
Sk4px
Xfermode_SrcOver:
SSE: 2.08ms -> 2.03ms (~2% faster)
NEON: my N5 is noisy, but there appears to be no perf change
BUG=skia:
Review URL: https://codereview.chromium.org/1132273004
Diffstat (limited to 'src/opts/Sk4px_SSE2.h')
-rw-r--r-- | src/opts/Sk4px_SSE2.h | 39 |
1 files changed, 39 insertions, 0 deletions
diff --git a/src/opts/Sk4px_SSE2.h b/src/opts/Sk4px_SSE2.h new file mode 100644 index 0000000000..d036328c14 --- /dev/null +++ b/src/opts/Sk4px_SSE2.h @@ -0,0 +1,39 @@ +/* + * Copyright 2015 Google Inc. + * + * Use of this source code is governed by a BSD-style license that can be + * found in the LICENSE file. + */ + +inline Sk4px::Sk4px(SkPMColor px) : INHERITED(_mm_set1_epi32(px)) {} + +inline Sk4px Sk4px::Load4(const SkPMColor px[4]) { + return Sk16b(_mm_loadu_si128((const __m128i*)px)); +} +inline Sk4px Sk4px::Load2(const SkPMColor px[2]) { + return Sk16b(_mm_loadl_epi64((const __m128i*)px)); +} +inline Sk4px Sk4px::Load1(const SkPMColor px[1]) { return Sk16b(_mm_cvtsi32_si128(*px)); } + +inline void Sk4px::store4(SkPMColor px[4]) const { _mm_storeu_si128((__m128i*)px, this->fVec); } +inline void Sk4px::store2(SkPMColor px[2]) const { _mm_storel_epi64((__m128i*)px, this->fVec); } +inline void Sk4px::store1(SkPMColor px[1]) const { *px = _mm_cvtsi128_si32(this->fVec); } + +inline Sk4px::Wide Sk4px::widenLo() const { + return Sk16h(_mm_unpacklo_epi8(this->fVec, _mm_setzero_si128()), + _mm_unpackhi_epi8(this->fVec, _mm_setzero_si128())); +} + +inline Sk4px::Wide Sk4px::widenHi() const { + return Sk16h(_mm_unpacklo_epi8(_mm_setzero_si128(), this->fVec), + _mm_unpackhi_epi8(_mm_setzero_si128(), this->fVec)); +} + +inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const { + return this->widenLo() * Sk4px(other).widenLo(); +} + +inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const { + Sk4px::Wide r = (*this + other) >> 8; + return Sk4px(_mm_packus_epi16(r.fLo.fVec, r.fHi.fVec)); +} |