diff options
| author |  mtklein <mtklein@chromium.org> | 2015-03-03 08:57:07 -0800 |
|---|---|---|
| committer |  Commit bot <commit-bot@chromium.org> | 2015-03-03 08:57:07 -0800 |
| commit | 0aebf5d0d3a2aef38a71885c85303583fdeaad57 (patch) | |
| tree | 1ec6aeef90ecc654f5040e7244f5c7246ce74690 /src/opts/SkPMFloat_neon.h | |
| parent | c2574f3657b1359496a4eba5b191961974b3a64f (diff) | |
Test and fix SkPMFloat rounding.
SSE rounds for free (that was a happy accident: they also have a truncating version).
NEON does not, nor obviously the portable code, so they add 0.5 before truncating.
NOPRESUBMIT=true
BUG=skia:
Review URL: https://codereview.chromium.org/974643002
Diffstat (limited to 'src/opts/SkPMFloat_neon.h')
| -rw-r--r-- | src/opts/SkPMFloat_neon.h | 14 |
1 files changed, 8 insertions, 6 deletions
diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h index 15ba3a58e4..179c674550 100644 --- a/src/opts/SkPMFloat_neon.h +++ b/src/opts/SkPMFloat_neon.h @@ -19,18 +19,20 @@ inline void SkPMFloat::set(SkPMColor c) { inline SkPMColor SkPMFloat::get() const { SkASSERT(this->isValid()); - uint32x4_t fix8_32 = vcvtq_u32_f32(vld1q_f32(fColor)); - uint16x4_t fix8_16 = vmovn_u32(fix8_32); - uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); + float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f)); + uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates, so round manually + uint16x4_t fix8_16 = vmovn_u32(fix8_32); + uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); SkPMColorAssert(c); return c; } inline SkPMColor SkPMFloat::clamped() const { - uint32x4_t fix8_32 = vcvtq_u32_f32(vld1q_f32(fColor)); - uint16x4_t fix8_16 = vqmovn_u32(fix8_32); - uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); + float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f)); + uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates, so round manually + uint16x4_t fix8_16 = vqmovn_u32(fix8_32); + uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); SkPMColorAssert(c); return c; |