diff options
| author |  Mike Klein <mtklein@chromium.org> | 2016-10-20 18:30:11 -0400 |
|---|---|---|
| committer |  Skia Commit-Bot <skia-commit-bot@chromium.org> | 2016-10-20 23:06:24 +0000 |
| commit | 3ae677c8f85fe0a4703d1df0b5941d59282d7ff7 (patch) | |
| tree | ab35ffc371a92488517fbb92c9297b99c1ecc91a /src/opts/SkRasterPipeline_opts.h | |
| parent | 49372e69d3ad4497ee2d46f1ab363d10d3432471 (diff) | |
Zero tail stack buffers.
MSAN and the no-SIMD bots have noticed the top lanes of the tail vectors are not initialized.
As they were written it was faster to leave them unintialized, but as re-written here it's equal speed and now safe.
CQ_INCLUDE_TRYBOTS=master.client.skia:Perf-Ubuntu-Clang-GCE-CPU-AVX2-x86_64-Debug-MSAN-Trybot,Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=3790
Change-Id: Icd41ba14ae6baf9947eb361a366f1ce19ad8aa67
Reviewed-on: https://skia-review.googlesource.com/3790
Reviewed-by: Mike Klein <mtklein@chromium.org>
Commit-Queue: Mike Klein <mtklein@chromium.org>
Diffstat (limited to 'src/opts/SkRasterPipeline_opts.h')
| -rw-r--r-- | src/opts/SkRasterPipeline_opts.h | 22 |
1 files changed, 12 insertions, 10 deletions
diff --git a/src/opts/SkRasterPipeline_opts.h b/src/opts/SkRasterPipeline_opts.h index 0759229608..f072f843bd 100644 --- a/src/opts/SkRasterPipeline_opts.h +++ b/src/opts/SkRasterPipeline_opts.h @@ -166,8 +166,8 @@ namespace SK_OPTS_NS { SI SkNx<N,T> load(size_t tail, const T* src) { SkASSERT(kIsTail == (tail > 0)); // TODO: maskload for 32- and 64-bit T - T buf[8]; if (kIsTail) { + T buf[8] = {0}; switch (tail & (N-1)) { case 7: buf[6] = src[6]; case 6: buf[5] = src[5]; @@ -177,7 +177,7 @@ namespace SK_OPTS_NS { case 2: buf[1] = src[1]; } buf[0] = src[0]; - src = buf; + return SkNx<N,T>::Load(buf); } return SkNx<N,T>::Load(src); } @@ -298,8 +298,9 @@ namespace SK_OPTS_NS { STAGE(load_d_f16, true) { auto ptr = (const uint64_t*)ctx + x; - uint64_t buf[8]; + SkNh rh, gh, bh, ah; if (kIsTail) { + uint64_t buf[8] = {0}; switch (tail & (N-1)) { case 7: buf[6] = ptr[6]; case 6: buf[5] = ptr[5]; @@ -309,11 +310,11 @@ namespace SK_OPTS_NS { case 2: buf[1] = ptr[1]; } buf[0] = ptr[0]; - ptr = buf; + SkNh::Load4(buf, &rh, &gh, &bh, &ah); + } else { + SkNh::Load4(ptr, &rh, &gh, &bh, &ah); } - SkNh rh, gh, bh, ah; - SkNh::Load4(ptr, &rh, &gh, &bh, &ah); dr = SkHalfToFloat_finite_ftz(rh); dg = SkHalfToFloat_finite_ftz(gh); db = SkHalfToFloat_finite_ftz(bh); @@ -323,8 +324,9 @@ namespace SK_OPTS_NS { STAGE(load_s_f16, true) { auto ptr = (const uint64_t*)ctx + x; - uint64_t buf[8]; + SkNh rh, gh, bh, ah; if (kIsTail) { + uint64_t buf[8] = {0}; switch (tail & (N-1)) { case 7: buf[6] = ptr[6]; case 6: buf[5] = ptr[5]; @@ -334,11 +336,11 @@ namespace SK_OPTS_NS { case 2: buf[1] = ptr[1]; } buf[0] = ptr[0]; - ptr = buf; + SkNh::Load4(buf, &rh, &gh, &bh, &ah); + } else { + SkNh::Load4(ptr, &rh, &gh, &bh, &ah); } - SkNh rh, gh, bh, ah; - SkNh::Load4(ptr, &rh, &gh, &bh, &ah); r = SkHalfToFloat_finite_ftz(rh); g = SkHalfToFloat_finite_ftz(gh); b = SkHalfToFloat_finite_ftz(bh); |