diff options
| author |  mtklein <mtklein@chromium.org> | 2015-12-01 07:10:21 -0800 |
|---|---|---|
| committer |  Commit bot <commit-bot@chromium.org> | 2015-12-01 07:10:21 -0800 |
| commit | 9db43ac4ee1a83a4f7b332fe6c00f592b6237349 (patch) | |
| tree | ce3b9eedc76bf142d31325d041295097774fee5c /src/opts | |
| parent | eeebdb538d476c1bfc8b63a946094ca1b505ecd1 (diff) | |
Add Sk4f::ToBytes(uint8_t[16], Sk4f, Sk4f, Sk4f, Sk4f)
This is a big speedup for float -> byte. E.g. gradient_linear_clamp_3color:
x86-64 147µs -> 103µs (Broadwell MBP)
arm64 2.03ms -> 648µs (Galaxy S6)
armv7 1.12ms -> 489µs (Galaxy S6, same device!)
BUG=skia:
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.android:Test-Android-GCC-Nexus9-CPU-Denver-Arm64-Debug-Trybot
Review URL: https://codereview.chromium.org/1483953002
Diffstat (limited to 'src/opts')
| -rw-r--r-- | src/opts/SkNx_neon.h | 8 | ||||
| -rw-r--r-- | src/opts/SkNx_sse.h | 9 |
2 files changed, 17 insertions, 0 deletions
diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h index a03f0be674..6fe6137e5f 100644 --- a/src/opts/SkNx_neon.h +++ b/src/opts/SkNx_neon.h @@ -165,6 +165,14 @@ public: vst1_lane_u32((uint32_t*)bytes, (uint32x2_t)fix8, 0); } + static void ToBytes(uint8_t bytes[16], + const SkNx& a, const SkNx& b, const SkNx& c, const SkNx& d) { + vst1q_u8(bytes, vuzpq_u8(vuzpq_u8((uint8x16_t)vcvtq_u32_f32(a.fVec), + (uint8x16_t)vcvtq_u32_f32(b.fVec)).val[0], + vuzpq_u8((uint8x16_t)vcvtq_u32_f32(c.fVec), + (uint8x16_t)vcvtq_u32_f32(d.fVec)).val[0]).val[0]); + } + SkNx approxInvert() const { float32x4_t est0 = vrecpeq_f32(fVec), est1 = vmulq_f32(vrecpsq_f32(est0, fVec), est0); diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h index f0ccd3f7b5..a4f8656536 100644 --- a/src/opts/SkNx_sse.h +++ b/src/opts/SkNx_sse.h @@ -132,6 +132,15 @@ public: *(int*)bytes = _mm_cvtsi128_si32(fix8); } + static void ToBytes(uint8_t bytes[16], + const SkNx& a, const SkNx& b, const SkNx& c, const SkNx& d) { + _mm_storeu_si128((__m128i*)bytes, + _mm_packus_epi16(_mm_packus_epi16(_mm_cvttps_epi32(a.fVec), + _mm_cvttps_epi32(b.fVec)), + _mm_packus_epi16(_mm_cvttps_epi32(c.fVec), + _mm_cvttps_epi32(d.fVec)))); + } + SkNx operator + (const SkNx& o) const { return _mm_add_ps(fVec, o.fVec); } SkNx operator - (const SkNx& o) const { return _mm_sub_ps(fVec, o.fVec); } SkNx operator * (const SkNx& o) const { return _mm_mul_ps(fVec, o.fVec); } |