OLD | NEW |
1 #include "SkColorPriv.h" | 1 #include "SkColorPriv.h" |
2 #include <arm_neon.h> | 2 #include <arm_neon.h> |
3 | 3 |
4 // For SkPMFloat(SkPMFColor), we widen our 8 bit components (fix8) to 8-bit comp
onents in 16 bits | 4 // 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. | 5 // (fix8_16), then widen those to 8-bit-in-32-bits (fix8_32), and finally conver
t 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 |
(...skipping 19 matching lines...) Expand all Loading... |
30 | 30 |
31 inline SkPMColor SkPMFloat::clamped() const { | 31 inline SkPMColor SkPMFloat::clamped() const { |
32 float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f)); | 32 float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f)); |
33 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates,
so round manually | 33 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); | 34 uint16x4_t fix8_16 = vqmovn_u32(fix8_32); |
35 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); | 35 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); |
36 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); | 36 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); |
37 SkPMColorAssert(c); | 37 SkPMColorAssert(c); |
38 return c; | 38 return c; |
39 } | 39 } |
| 40 |
| 41 // 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]) { |
| 43 for (int i = 0; i < 4; i++) { floats[i] = FromPMColor(colors[i]); } |
| 44 } |
| 45 |
| 46 inline void SkPMFloat::To4PMColors(SkPMColor colors[4], const SkPMFloat floats[4
]) { |
| 47 for (int i = 0; i < 4; i++) { colors[i] = floats[i].get(); } |
| 48 } |
| 49 |
| 50 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(); } |
| 52 } |
OLD | NEW |