OLD | NEW |
(Empty) | |
| 1 #include "SkColorPriv.h" |
| 2 #include "SkPMFloat.h" |
| 3 #include <arm_neon.h> |
| 4 |
| 5 // For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bit
s (fix8_16), |
| 6 // then widen those to 8-bit-in-32-bits (fix8_32), convert those to floats (scal
ed), |
| 7 // then finally scale those down from [0.0f, 255.0f] to [0.0f, 1.0f] into fColor
. |
| 8 |
| 9 // get() and clamped() do the opposite, working from [0.0f, 1.0f] floats to [0.0
f, 255.0f], |
| 10 // to 8-bit-in-32-bit, to 8-bit-in-16-bit, back down to 8-bit components. |
| 11 // clamped() uses vqmovn to clamp while narrowing instead of just narrowing with
vmovn. |
| 12 |
| 13 inline void SkPMFloat::set(SkPMColor c) { |
| 14 SkPMColorAssert(c); |
| 15 uint8x8_t fix8 = (uint8x8_t)vdup_n_u32(c); |
| 16 uint16x8_t fix8_16 = vmovl_u8(fix8); |
| 17 uint32x4_t fix8_32 = vmovl_u16(vget_low_u16(fix8_16)); |
| 18 float32x4_t scaled = vcvtq_f32_u32(fix8_32); |
| 19 vst1q_f32(fColor, vmulq_f32(scaled, vdupq_n_f32(1.0f/255.0f))); |
| 20 SkASSERT(this->isValid()); |
| 21 } |
| 22 |
| 23 inline SkPMColor SkPMFloat::get() const { |
| 24 SkASSERT(this->isValid()); |
| 25 float32x4_t scaled = vmulq_f32(vld1q_f32(fColor), vdupq_n_f32(255.0f)); |
| 26 uint32x4_t fix8_32 = vcvtq_u32_f32(scaled); |
| 27 uint16x4_t fix8_16 = vmovn_u32(fix8_32); |
| 28 uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); |
| 29 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); |
| 30 SkPMColorAssert(c); |
| 31 return c; |
| 32 } |
| 33 |
| 34 inline SkPMColor SkPMFloat::clamped() const { |
| 35 float32x4_t scaled = vmulq_f32(vld1q_f32(fColor), vdupq_n_f32(255.0f)); |
| 36 uint32x4_t fix8_32 = vcvtq_u32_f32(scaled); |
| 37 uint16x4_t fix8_16 = vqmovn_u32(fix8_32); |
| 38 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); |
| 39 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0); |
| 40 SkPMColorAssert(c); |
| 41 return c; |
| 42 } |
OLD | NEW |