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