diff options
| author |  mtklein <mtklein@chromium.org> | 2016-02-09 15:41:36 -0800 |
|---|---|---|
| committer |  Commit bot <commit-bot@chromium.org> | 2016-02-09 15:41:36 -0800 |
| commit | 126626e52310b39429ad5eaa8e68e0caa24e6202 (patch) | |
| tree | f3037609300c28c3adf4cc3da61057dbb2435c08 /src/opts/SkNx_sse.h | |
| parent | c023210893da7ca407bad8c9f07c8339ee81854c (diff) | |
Sk4f: add floor()
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1685773002
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
Committed: https://skia.googlesource.com/skia/+/86c6c4935171a1d2d6a9ffbff37ec6dac1326614
CQ_EXTRA_TRYBOTS=client.skia.android:Test-Android-GCC-Nexus7-GPU-Tegra3-Arm7-Release-Trybot,Test-Android-GCC-Nexus9-GPU-TegraK1-Arm64-Release-Trybot
Review URL: https://codereview.chromium.org/1685773002
Diffstat (limited to 'src/opts/SkNx_sse.h')
| -rw-r--r-- | src/opts/SkNx_sse.h | 12 |
1 files changed, 12 insertions, 0 deletions
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h index 889253b9d8..def7ba1719 100644 --- a/src/opts/SkNx_sse.h +++ b/src/opts/SkNx_sse.h @@ -13,6 +13,17 @@ #define SKNX_IS_FAST +// SSE 4.1 has _mm_floor_ps to floor 4 floats. We emulate it: +// - round by adding (1<<23) with our sign, then subtracting it; +// - if that rounded value is bigger than our input, subtract 1. +static inline __m128 sse2_mm_floor_ps(__m128 v) { + __m128 sign = _mm_and_ps(v, _mm_set1_ps(-0.0f)); + __m128 bias = _mm_or_ps(sign, _mm_set1_ps(1<<23)); + __m128 rounded = _mm_sub_ps(_mm_add_ps(v, bias), bias); + __m128 too_big = _mm_cmpgt_ps(rounded, v); + return _mm_sub_ps(rounded, _mm_and_ps(too_big, _mm_set1_ps(1.0f))); +} + template <> class SkNx<2, float> { public: @@ -92,6 +103,7 @@ public: static SkNx Max(const SkNx& l, const SkNx& r) { return _mm_max_ps(l.fVec, r.fVec); } SkNx abs() const { return _mm_andnot_ps(_mm_set1_ps(-0.0f), fVec); } + SkNx floor() const { return sse2_mm_floor_ps(fVec); } SkNx sqrt () const { return _mm_sqrt_ps (fVec); } SkNx rsqrt0() const { return _mm_rsqrt_ps(fVec); } |