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

exact divide by 255 with NEON

Browse Source 
Change-Id: Ib121eb0d5af1f22f48f517fe909112a77d92032e
Reviewed-on: https://skia-review.googlesource.com/113666
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Herb Derby <herb@google.com>
This commit is contained in:
Mike Klein 2018-03-10 11:34:53 -05:00 committed by Skia Commit-Bot
parent ba321b6017
commit d8853ec0fd
2 changed files with 25 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/opts/SkRasterPipeline_opts.h
View File

@ -2487,6 +2487,11 @@ static void start_pipeline(const size_t x0, const size_t y0,
SI U16 div255(U16 v) {
#if 0
return (v+127)/255; // The ideal rounding divide by 255.
#elif 1 && defined(__ARM_NEON)
// With NEON we can compute (v+127)/255 as (v + ((v+128)>>8) + 128)>>8
// just as fast as we can do the approximation below, so might as well be correct!
// First we compute v + ((v+128)>>8), then one more round of (...+128)>>8 to finish up.
return vrshrq_n_u16(vrsraq_n_u16(v, v, 8), 8);
#else
return (v+255)/256; // A good approximation of (v+127)/255.
#endif

20
tests/MathTest.cpp
View File

@ -750,3 +750,23 @@ DEF_TEST(DoubleSaturate32, reporter) {
REPORTER_ASSERT(reporter, r.fExpectedInt == i);
}
}
#if defined(__ARM_NEON)
#include <arm_neon.h>
DEF_TEST(NeonU16Div255, r) {
for (int v = 0; v <= 255*255; v++) {
int want = (v + 127)/255;
uint16x8_t V = vdupq_n_u16(v);
int got = vrshrq_n_u16(vrsraq_n_u16(V, V, 8), 8)[0];
if (got != want) {
SkDebugf("%d -> %d, want %d\n", v, got, want);
}
REPORTER_ASSERT(r, got == want);
}
}
#endif