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 77 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
88 | 88 |
89 inline Sk4px Sk4px::zeroColors() const { | 89 inline Sk4px Sk4px::zeroColors() const { |
90 return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); | 90 return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); |
91 } | 91 } |
92 | 92 |
93 inline Sk4px Sk4px::zeroAlphas() const { | 93 inline Sk4px Sk4px::zeroAlphas() const { |
94 // vbic(a,b) == a & ~b | 94 // vbic(a,b) == a & ~b |
95 return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); | 95 return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHI
FT))); |
96 } | 96 } |
97 | 97 |
98 static inline uint8x16_t widen_to_8888(uint16x4_t v) { | |
99 // RGB565 format: |R....|G.....|B....| | |
100 // Bit: 16 11 5 0 | |
101 | |
102 // First get each pixel into its own 32-bit lane. | |
103 // v == rgb3 rgb2 rgb1 rgb0 | |
104 // spread == 0000 rgb3 0000 rgb2 0000 rgb1 0000 rgb0 | |
105 uint32x4_t spread = vmovl_u16(v); | |
106 | |
107 // Get each color independently, still in 565 precison but down at bit 0. | |
108 auto r5 = vshrq_n_u32(spread, 11), | |
109 g6 = vandq_u32(vdupq_n_u32(63), vshrq_n_u32(spread, 5)), | |
110 b5 = vandq_u32(vdupq_n_u32(31), spread); | |
111 | |
112 // Scale 565 precision up to 8-bit each, filling low 323 bits with high bits
of each component. | |
113 auto r8 = vorrq_u32(vshlq_n_u32(r5, 3), vshrq_n_u32(r5, 2)), | |
114 g8 = vorrq_u32(vshlq_n_u32(g6, 2), vshrq_n_u32(g6, 4)), | |
115 b8 = vorrq_u32(vshlq_n_u32(b5, 3), vshrq_n_u32(b5, 2)); | |
116 | |
117 // Now put all the 8-bit components into SkPMColor order. | |
118 return (uint8x16_t)vorrq_u32(vshlq_n_u32(r8, SK_R32_SHIFT), // TODO: one s
hift is zero... | |
119 vorrq_u32(vshlq_n_u32(g8, SK_G32_SHIFT), | |
120 vorrq_u32(vshlq_n_u32(b8, SK_B32_SHIFT), | |
121 vdupq_n_u32(0xFF << SK_A32_SHIFT)))); | |
122 } | |
123 | |
124 static inline uint16x4_t narrow_to_565(uint8x16_t w8x16) { | |
125 uint32x4_t w = (uint32x4_t)w8x16; | |
126 | |
127 // Extract out top RGB 565 bits of each pixel, with no rounding. | |
128 auto r5 = vandq_u32(vdupq_n_u32(31), vshrq_n_u32(w, SK_R32_SHIFT + 3)), | |
129 g6 = vandq_u32(vdupq_n_u32(63), vshrq_n_u32(w, SK_G32_SHIFT + 2)), | |
130 b5 = vandq_u32(vdupq_n_u32(31), vshrq_n_u32(w, SK_B32_SHIFT + 3)); | |
131 | |
132 // Now put the bits in place in the low 16-bits of each 32-bit lane. | |
133 auto spread = vorrq_u32(vshlq_n_u32(r5, 11), | |
134 vorrq_u32(vshlq_n_u32(g6, 5), | |
135 b5)); | |
136 | |
137 // Pack the low 16-bits of our 128-bit register down into a 64-bit register. | |
138 // spread == 0000 rgb3 0000 rgb2 0000 rgb1 0000 rgb0 | |
139 // v == rgb3 rgb2 rgb1 rgb0 | |
140 auto v = vmovn_u32(spread); | |
141 return v; | |
142 } | |
143 | |
144 | |
145 inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) { | |
146 return Sk16b(widen_to_8888(vld1_u16(src))); | |
147 } | |
148 inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) { | |
149 auto src2 = ((uint32_t)src[0] ) | |
150 | ((uint32_t)src[1] << 16); | |
151 return Sk16b(widen_to_8888(vcreate_u16(src2))); | |
152 } | |
153 inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) { | |
154 return Sk16b(widen_to_8888(vcreate_u16(src[0]))); | |
155 } | |
156 | |
157 inline void Sk4px::store4(SkPMColor16 dst[4]) const { | |
158 vst1_u16(dst, narrow_to_565(this->fVec)); | |
159 } | |
160 inline void Sk4px::store2(SkPMColor16 dst[2]) const { | |
161 auto v = narrow_to_565(this->fVec); | |
162 dst[0] = vget_lane_u16(v, 0); | |
163 dst[1] = vget_lane_u16(v, 1); | |
164 } | |
165 inline void Sk4px::store1(SkPMColor16 dst[1]) const { | |
166 dst[0] = vget_lane_u16(narrow_to_565(this->fVec), 0); | |
167 } | |
168 | |
169 } // namespace | 98 } // namespace |
170 | 99 |
OLD | NEW |