OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2015 Google Inc. | 2 * Copyright 2015 Google Inc. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
6 */ | 6 */ |
7 | 7 |
8 namespace { // See Sk4px.h | 8 namespace { // See Sk4px.h |
9 | 9 |
10 inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b((uint8x16_t)vdupq_n_
u32(px)); } | 10 inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b((uint8x16_t)vdupq_n_
u32(px)); } |
(...skipping 71 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
82 | 82 |
83 inline Sk4px Sk4px::zeroColors() const { | 83 inline Sk4px Sk4px::zeroColors() const { |
84 return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); | 84 return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); |
85 } | 85 } |
86 | 86 |
87 inline Sk4px Sk4px::zeroAlphas() const { | 87 inline Sk4px Sk4px::zeroAlphas() const { |
88 // vbic(a,b) == a & ~b | 88 // vbic(a,b) == a & ~b |
89 return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); | 89 return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); |
90 } | 90 } |
91 | 91 |
| 92 static inline uint8x16_t widen_to_8888(uint16x4_t v) { |
| 93 // RGB565 format: |R....|G.....|B....| |
| 94 // Bit: 16 11 5 0 |
| 95 |
| 96 // First get each pixel into its own 32-bit lane. |
| 97 // v == rgb3 rgb2 rgb1 rgb0 |
| 98 // spread == 0000 rgb3 0000 rgb2 0000 rgb1 0000 rgb0 |
| 99 uint32x4_t spread = vmovl_u16(v); |
| 100 |
| 101 // Get each color independently, still in 565 precison but down at bit 0. |
| 102 auto r5 = vshrq_n_u32(spread, 11), |
| 103 g6 = vandq_u32(vdupq_n_u32(63), vshrq_n_u32(spread, 5)), |
| 104 b5 = vandq_u32(vdupq_n_u32(31), spread); |
| 105 |
| 106 // Scale 565 precision up to 8-bit each, filling low 323 bits with high bits
of each component. |
| 107 auto r8 = vorrq_u32(vshlq_n_u32(r5, 3), vshrq_n_u32(r5, 2)), |
| 108 g8 = vorrq_u32(vshlq_n_u32(g6, 2), vshrq_n_u32(g6, 4)), |
| 109 b8 = vorrq_u32(vshlq_n_u32(b5, 3), vshrq_n_u32(b5, 2)); |
| 110 |
| 111 // Now put all the 8-bit components into SkPMColor order. |
| 112 return (uint8x16_t)vorrq_u32(vshlq_n_u32(r8, SK_R32_SHIFT), // TODO: one s
hift is zero... |
| 113 vorrq_u32(vshlq_n_u32(g8, SK_G32_SHIFT), |
| 114 vorrq_u32(vshlq_n_u32(b8, SK_B32_SHIFT), |
| 115 vdupq_n_u32(0xFF << SK_A32_SHIFT)))); |
| 116 } |
| 117 |
| 118 static inline uint16x4_t narrow_to_565(uint8x16_t w8x16) { |
| 119 uint32x4_t w = (uint32x4_t)w8x16; |
| 120 |
| 121 // Extract out top RGB 565 bits of each pixel, with no rounding. |
| 122 auto r5 = vandq_u32(vdupq_n_u32(31), vshrq_n_u32(w, SK_R32_SHIFT + 3)), |
| 123 g6 = vandq_u32(vdupq_n_u32(63), vshrq_n_u32(w, SK_G32_SHIFT + 2)), |
| 124 b5 = vandq_u32(vdupq_n_u32(31), vshrq_n_u32(w, SK_B32_SHIFT + 3)); |
| 125 |
| 126 // Now put the bits in place in the low 16-bits of each 32-bit lane. |
| 127 auto spread = vorrq_u32(vshlq_n_u32(r5, 11), |
| 128 vorrq_u32(vshlq_n_u32(g6, 5), |
| 129 b5)); |
| 130 |
| 131 // Pack the low 16-bits of our 128-bit register down into a 64-bit register. |
| 132 // spread == 0000 rgb3 0000 rgb2 0000 rgb1 0000 rgb0 |
| 133 // v == rgb3 rgb2 rgb1 rgb0 |
| 134 auto v = vmovn_u32(spread); |
| 135 return v; |
| 136 } |
| 137 |
| 138 |
| 139 inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) { |
| 140 return Sk16b(widen_to_8888(vld1_u16(src))); |
| 141 } |
| 142 inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) { |
| 143 auto src2 = ((uint32_t)src[0] ) |
| 144 | ((uint32_t)src[1] << 16); |
| 145 return Sk16b(widen_to_8888(vcreate_u16(src2))); |
| 146 } |
| 147 inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) { |
| 148 return Sk16b(widen_to_8888(vcreate_u16(src[0]))); |
| 149 } |
| 150 |
| 151 inline void Sk4px::store4(SkPMColor16 dst[4]) const { |
| 152 vst1_u16(dst, narrow_to_565(this->fVec)); |
| 153 } |
| 154 inline void Sk4px::store2(SkPMColor16 dst[2]) const { |
| 155 auto v = narrow_to_565(this->fVec); |
| 156 dst[0] = vget_lane_u16(v, 0); |
| 157 dst[1] = vget_lane_u16(v, 1); |
| 158 } |
| 159 inline void Sk4px::store1(SkPMColor16 dst[1]) const { |
| 160 dst[0] = vget_lane_u16(narrow_to_565(this->fVec), 0); |
| 161 } |
| 162 |
92 } // namespace | 163 } // namespace |
93 | 164 |
OLD | NEW |