diff options
author | mtklein <mtklein@chromium.org> | 2015-07-14 10:54:19 -0700 |
---|---|---|
committer | Commit bot <commit-bot@chromium.org> | 2015-07-14 10:54:19 -0700 |
commit | 4be181e304d2b280c6801bd13369cfba236d1a66 (patch) | |
tree | ae0510f8a6504c3333582fa004e961a8771a2d99 /src/opts/Sk4px_NEON.h | |
parent | a5517e2b190a8083b38964972b031c13e99f1012 (diff) |
3-15% speedup to HardLight / Overlay xfermodes.
While investigating my bug (skia:4052) I saw this TODO and figured
it'd make me feel better about an otherwise unsuccessful investigation.
This speeds up HardLight and Overlay (same code) by about 15% with SSE, mostly
by rewriting the logic from 1 cheap comparison and 2 expensive div255() calls
to 2 cheap comparisons and 1 expensive div255().
NEON speeds up by a more modest ~3%.
BUG=skia:
Review URL: https://codereview.chromium.org/1230663005
Diffstat (limited to 'src/opts/Sk4px_NEON.h')
-rw-r--r-- | src/opts/Sk4px_NEON.h | 6 |
1 files changed, 6 insertions, 0 deletions
diff --git a/src/opts/Sk4px_NEON.h b/src/opts/Sk4px_NEON.h index 9401864697..cd6dea9979 100644 --- a/src/opts/Sk4px_NEON.h +++ b/src/opts/Sk4px_NEON.h @@ -40,6 +40,12 @@ inline Sk4px::Wide Sk4px::widenHi() const { vshll_n_u8(vget_high_u8(this->fVec), 8)); } +inline Sk4px::Wide Sk4px::widenLoHi() const { + auto zipped = vzipq_u8(this->fVec, this->fVec); + return Sk16h((uint16x8_t)zipped.val[0], + (uint16x8_t)zipped.val[1]); +} + inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const { return Sk16h(vmull_u8(vget_low_u8 (this->fVec), vget_low_u8 (other.fVec)), vmull_u8(vget_high_u8(this->fVec), vget_high_u8(other.fVec))); |