Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(207)

Side by Side Diff: src/opts/SkPMFloat_neon.h

Issue 974643002: Test and fix SkPMFloat rounding. (Closed) Base URL: https://skia.googlesource.com/skia.git@master
Patch Set: f Created 5 years, 9 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
« no previous file with comments | « src/opts/SkPMFloat_SSE2.h ('k') | src/opts/SkPMFloat_none.h » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 #include "SkColorPriv.h" 1 #include "SkColorPriv.h"
2 #include <arm_neon.h> 2 #include <arm_neon.h>
3 3
4 // For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bit s (fix8_16), 4 // For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bit s (fix8_16),
5 // then widen those to 8-bit-in-32-bits (fix8_32), and finally convert those to floats. 5 // then widen those to 8-bit-in-32-bits (fix8_32), and finally convert those to floats.
6 6
7 // get() and clamped() do the opposite, working from floats to 8-bit-in-32-bit, 7 // get() and clamped() do the opposite, working from floats to 8-bit-in-32-bit,
8 // to 8-bit-in-16-bit, back down to 8-bit components. 8 // to 8-bit-in-16-bit, back down to 8-bit components.
9 // clamped() uses vqmovn to clamp while narrowing instead of just narrowing with vmovn. 9 // clamped() uses vqmovn to clamp while narrowing instead of just narrowing with vmovn.
10 10
11 inline void SkPMFloat::set(SkPMColor c) { 11 inline void SkPMFloat::set(SkPMColor c) {
12 SkPMColorAssert(c); 12 SkPMColorAssert(c);
13 uint8x8_t fix8 = (uint8x8_t)vdup_n_u32(c); 13 uint8x8_t fix8 = (uint8x8_t)vdup_n_u32(c);
14 uint16x8_t fix8_16 = vmovl_u8(fix8); 14 uint16x8_t fix8_16 = vmovl_u8(fix8);
15 uint32x4_t fix8_32 = vmovl_u16(vget_low_u16(fix8_16)); 15 uint32x4_t fix8_32 = vmovl_u16(vget_low_u16(fix8_16));
16 vst1q_f32(fColor, vcvtq_f32_u32(fix8_32)); 16 vst1q_f32(fColor, vcvtq_f32_u32(fix8_32));
17 SkASSERT(this->isValid()); 17 SkASSERT(this->isValid());
18 } 18 }
19 19
20 inline SkPMColor SkPMFloat::get() const { 20 inline SkPMColor SkPMFloat::get() const {
21 SkASSERT(this->isValid()); 21 SkASSERT(this->isValid());
22 uint32x4_t fix8_32 = vcvtq_u32_f32(vld1q_f32(fColor)); 22 float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f));
23 uint16x4_t fix8_16 = vmovn_u32(fix8_32); 23 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates, so round manually
24 uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); 24 uint16x4_t fix8_16 = vmovn_u32(fix8_32);
25 uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
25 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); 26 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0);
26 SkPMColorAssert(c); 27 SkPMColorAssert(c);
27 return c; 28 return c;
28 } 29 }
29 30
30 inline SkPMColor SkPMFloat::clamped() const { 31 inline SkPMColor SkPMFloat::clamped() const {
31 uint32x4_t fix8_32 = vcvtq_u32_f32(vld1q_f32(fColor)); 32 float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f));
32 uint16x4_t fix8_16 = vqmovn_u32(fix8_32); 33 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates, so round manually
33 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); 34 uint16x4_t fix8_16 = vqmovn_u32(fix8_32);
35 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
34 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); 36 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0);
35 SkPMColorAssert(c); 37 SkPMColorAssert(c);
36 return c; 38 return c;
37 } 39 }
OLDNEW
« no previous file with comments | « src/opts/SkPMFloat_SSE2.h ('k') | src/opts/SkPMFloat_none.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698