AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

skip alpha scale if 256

Browse Source 
It's free performance!

Seriously, though, it's a perfectly predictable branch.
On opaque content I'm seeing a small speedup and two
extra instructions in the hot path:

      + cmpl $256, %regs
      + je <to write-it-out>
        pmullw %xmm, %xmm
        psrlw $8, %xmm
        pmullw %xmm, %xmm
        psrlw $8, %xmm
  write-it-out:

Change-Id: I1c19a9efac62df2bb901ca9eb2679c967729d863
Reviewed-on: https://skia-review.googlesource.com/c/170901
Auto-Submit: Mike Klein <mtklein@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Herb Derby <herb@google.com>
This commit is contained in:
Mike Klein 2018-11-13 19:37:14 -05:00 committed by Skia Commit-Bot
parent 2f9a5ea639
commit 9341952756
1 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/opts/SkBitmapProcState_opts_SSSE3.cpp
View File

@ -80,9 +80,12 @@ void S32_alpha_D32_filter_DX_SSSE3(const SkBitmapProcState& s,
SkASSERT(s.fFilterQuality != kNone_SkFilterQuality);
SkASSERT(kN32_SkColorType == s.fPixmap.colorType());
int alpha = s.fAlphaScale;
// Return (px * s.fAlphaScale) / 256. (s.fAlphaScale is in [0,256].)
auto scale_by_alpha = [&](const __m128i& px) {
return _mm_srli_epi16(_mm_mullo_epi16(px, _mm_set1_epi16(s.fAlphaScale)), 8);
auto scale_by_alpha = [alpha](const __m128i& px) {
return alpha == 256 ? px
: _mm_srli_epi16(_mm_mullo_epi16(px, _mm_set1_epi16(alpha)), 8);
};
// We're in _DX_ mode here, so we're only varying in X.