OLD | NEW |
1 #include "SkColorPriv.h" | 1 inline SkPMFloat& SkPMFloat::operator=(const SkPMFloat& that) { |
2 #include <arm_neon.h> | 2 fColors = that.fColors; |
| 3 return *this; |
| 4 } |
3 | 5 |
4 // For SkPMFloat(SkPMFColor), we widen our 8 bit components (fix8) to 8-bit comp
onents in 16 bits | 6 // For SkPMFloat(SkPMFColor), we widen our 8 bit components (fix8) to 8-bit comp
onents in 16 bits |
5 // (fix8_16), then widen those to 8-bit-in-32-bits (fix8_32), and finally conver
t those to floats. | 7 // (fix8_16), then widen those to 8-bit-in-32-bits (fix8_32), and finally conver
t those to floats. |
6 | 8 |
7 // get() and clamped() do the opposite, working from floats to 8-bit-in-32-bit, | 9 // 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. | 10 // 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. | 11 // clamped() uses vqmovn to clamp while narrowing instead of just narrowing with
vmovn. |
10 | 12 |
11 inline SkPMFloat::SkPMFloat(SkPMColor c) { | 13 inline SkPMFloat::SkPMFloat(SkPMColor c) { |
12 SkPMColorAssert(c); | 14 SkPMColorAssert(c); |
13 uint8x8_t fix8 = (uint8x8_t)vdup_n_u32(c); | 15 uint8x8_t fix8 = (uint8x8_t)vdup_n_u32(c); |
14 uint16x8_t fix8_16 = vmovl_u8(fix8); | 16 uint16x8_t fix8_16 = vmovl_u8(fix8); |
15 uint32x4_t fix8_32 = vmovl_u16(vget_low_u16(fix8_16)); | 17 uint32x4_t fix8_32 = vmovl_u16(vget_low_u16(fix8_16)); |
16 vst1q_f32(fColor, vcvtq_f32_u32(fix8_32)); | 18 fColors = vcvtq_f32_u32(fix8_32); |
17 SkASSERT(this->isValid()); | 19 SkASSERT(this->isValid()); |
18 } | 20 } |
19 | 21 |
20 inline SkPMColor SkPMFloat::get() const { | 22 inline SkPMColor SkPMFloat::get() const { |
21 SkASSERT(this->isValid()); | 23 SkASSERT(this->isValid()); |
22 float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f)); | 24 float32x4_t add_half = vaddq_f32(fColors, vdupq_n_f32(0.5f)); |
23 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates,
so round manually | 25 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates,
so round manually |
24 uint16x4_t fix8_16 = vmovn_u32(fix8_32); | 26 uint16x4_t fix8_16 = vmovn_u32(fix8_32); |
25 uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); | 27 uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); |
26 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); | 28 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); |
27 SkPMColorAssert(c); | 29 SkPMColorAssert(c); |
28 return c; | 30 return c; |
29 } | 31 } |
30 | 32 |
31 inline SkPMColor SkPMFloat::clamped() const { | 33 inline SkPMColor SkPMFloat::clamped() const { |
32 float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f)); | 34 float32x4_t add_half = vaddq_f32(fColors, vdupq_n_f32(0.5f)); |
33 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates,
so round manually | 35 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates,
so round manually |
34 uint16x4_t fix8_16 = vqmovn_u32(fix8_32); | 36 uint16x4_t fix8_16 = vqmovn_u32(fix8_32); |
35 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); | 37 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); |
36 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); | 38 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); |
37 SkPMColorAssert(c); | 39 SkPMColorAssert(c); |
38 return c; | 40 return c; |
39 } | 41 } |
40 | 42 |
41 // TODO: we should be able to beat these loops on all three methods. | 43 // TODO: we should be able to beat these loops on all three methods. |
42 inline void SkPMFloat::From4PMColors(SkPMFloat floats[4], const SkPMColor colors
[4]) { | 44 inline void SkPMFloat::From4PMColors(SkPMFloat floats[4], const SkPMColor colors
[4]) { |
43 for (int i = 0; i < 4; i++) { floats[i] = FromPMColor(colors[i]); } | 45 for (int i = 0; i < 4; i++) { floats[i] = FromPMColor(colors[i]); } |
44 } | 46 } |
45 | 47 |
46 inline void SkPMFloat::To4PMColors(SkPMColor colors[4], const SkPMFloat floats[4
]) { | 48 inline void SkPMFloat::To4PMColors(SkPMColor colors[4], const SkPMFloat floats[4
]) { |
47 for (int i = 0; i < 4; i++) { colors[i] = floats[i].get(); } | 49 for (int i = 0; i < 4; i++) { colors[i] = floats[i].get(); } |
48 } | 50 } |
49 | 51 |
50 inline void SkPMFloat::ClampTo4PMColors(SkPMColor colors[4], const SkPMFloat flo
ats[4]) { | 52 inline void SkPMFloat::ClampTo4PMColors(SkPMColor colors[4], const SkPMFloat flo
ats[4]) { |
51 for (int i = 0; i < 4; i++) { colors[i] = floats[i].clamped(); } | 53 for (int i = 0; i < 4; i++) { colors[i] = floats[i].clamped(); } |
52 } | 54 } |
OLD | NEW |