diff options
| author |  mtklein <mtklein@chromium.org> | 2015-04-09 14:05:17 -0700 |
|---|---|---|
| committer |  Commit bot <commit-bot@chromium.org> | 2015-04-09 14:05:17 -0700 |
| commit | 9ff378b01be0b0a3fc35677a2155ba4ade286cc2 (patch) | |
| tree | 2d9221ef31e3632dde74bda9a1023bc1ce71b27c /src/core/SkUtils.cpp | |
| parent | a1e41c6d9a2029eb836c6120bb154ccd25e1588d (diff) | |
Rewrite memset benches, then use results to add a small-N optimization.
The benches for N <= 10 get around 2x faster on my N7 and N9. I believe this
is because of the reduced function-call-then-function-pointer-call overhead on
the N7, and additionally because it seems autovectorization beats our NEON code
for small N on the N9.
My desktop is unchanged, though that's probably because N=10 lies well within a
region where memset's performance is essentially constant: N=100 takes only
about 2x as long as N=1 and N=10, which perform nearly identically.
BUG=skia:
Review URL: https://codereview.chromium.org/1073863002
Diffstat (limited to 'src/core/SkUtils.cpp')
| -rw-r--r-- | src/core/SkUtils.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/core/SkUtils.cpp b/src/core/SkUtils.cpp index b063071932..33ea4db47c 100644 --- a/src/core/SkUtils.cpp +++ b/src/core/SkUtils.cpp @@ -134,12 +134,12 @@ SkMemcpy32Proc choose_memcpy32() { } // namespace -void sk_memset16(uint16_t dst[], uint16_t value, int count) { +void sk_memset16_large(uint16_t dst[], uint16_t value, int count) { SK_DECLARE_STATIC_LAZY_FN_PTR(SkMemset16Proc, proc, choose_memset16); proc.get()(dst, value, count); } -void sk_memset32(uint32_t dst[], uint32_t value, int count) { +void sk_memset32_large(uint32_t dst[], uint32_t value, int count) { SK_DECLARE_STATIC_LAZY_FN_PTR(SkMemset32Proc, proc, choose_memset32); proc.get()(dst, value, count); } |