blob: c8abff24b16446257bb745673299f5be6bd365e5 [file] [log] [blame]
mtkleind2ffd362015-05-12 06:11:21 -07001/*
2 * Copyright 2015 Google Inc.
3 *
4 * Use of this source code is governed by a BSD-style license that can be
5 * found in the LICENSE file.
6 */
7
mtklein082e3292015-08-12 11:56:43 -07008namespace { // See Sk4px.h
mtkleinaa999cb2015-05-22 17:18:21 -07009
mtklein082e3292015-08-12 11:56:43 -070010inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b((uint8x16_t)vdupq_n_u32(px)); }
11
12inline Sk4px Sk4px::Load4(const SkPMColor px[4]) {
mtkleind2ffd362015-05-12 06:11:21 -070013 return Sk16b((uint8x16_t)vld1q_u32(px));
14}
mtklein082e3292015-08-12 11:56:43 -070015inline Sk4px Sk4px::Load2(const SkPMColor px[2]) {
mtkleind2ffd362015-05-12 06:11:21 -070016 uint32x2_t px2 = vld1_u32(px);
17 return Sk16b((uint8x16_t)vcombine_u32(px2, px2));
18}
mtklein082e3292015-08-12 11:56:43 -070019inline Sk4px Sk4px::Load1(const SkPMColor px[1]) {
mtkleind2ffd362015-05-12 06:11:21 -070020 return Sk16b((uint8x16_t)vdupq_n_u32(*px));
21}
22
mtklein082e3292015-08-12 11:56:43 -070023inline void Sk4px::store4(SkPMColor px[4]) const {
mtkleind2ffd362015-05-12 06:11:21 -070024 vst1q_u32(px, (uint32x4_t)this->fVec);
25}
mtklein082e3292015-08-12 11:56:43 -070026inline void Sk4px::store2(SkPMColor px[2]) const {
mtkleind2ffd362015-05-12 06:11:21 -070027 vst1_u32(px, (uint32x2_t)vget_low_u8(this->fVec));
28}
mtklein082e3292015-08-12 11:56:43 -070029inline void Sk4px::store1(SkPMColor px[1]) const {
mtkleind2ffd362015-05-12 06:11:21 -070030 vst1q_lane_u32(px, (uint32x4_t)this->fVec, 0);
31}
32
mtklein082e3292015-08-12 11:56:43 -070033inline Sk4px::Wide Sk4px::widenLo() const {
mtkleind2ffd362015-05-12 06:11:21 -070034 return Sk16h(vmovl_u8(vget_low_u8 (this->fVec)),
35 vmovl_u8(vget_high_u8(this->fVec)));
36}
37
mtklein082e3292015-08-12 11:56:43 -070038inline Sk4px::Wide Sk4px::widenHi() const {
mtkleind2ffd362015-05-12 06:11:21 -070039 return Sk16h(vshll_n_u8(vget_low_u8 (this->fVec), 8),
40 vshll_n_u8(vget_high_u8(this->fVec), 8));
41}
42
mtklein082e3292015-08-12 11:56:43 -070043inline Sk4px::Wide Sk4px::widenLoHi() const {
mtklein4be181e2015-07-14 10:54:19 -070044 auto zipped = vzipq_u8(this->fVec, this->fVec);
45 return Sk16h((uint16x8_t)zipped.val[0],
46 (uint16x8_t)zipped.val[1]);
47}
48
mtklein082e3292015-08-12 11:56:43 -070049inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
mtkleind2ffd362015-05-12 06:11:21 -070050 return Sk16h(vmull_u8(vget_low_u8 (this->fVec), vget_low_u8 (other.fVec)),
51 vmull_u8(vget_high_u8(this->fVec), vget_high_u8(other.fVec)));
52}
53
mtklein082e3292015-08-12 11:56:43 -070054inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
mtkleind2ffd362015-05-12 06:11:21 -070055 const Sk4px::Wide o(other); // Should be no code, but allows us to access fLo, fHi.
56 return Sk16b(vcombine_u8(vaddhn_u16(this->fLo.fVec, o.fLo.fVec),
57 vaddhn_u16(this->fHi.fVec, o.fHi.fVec)));
58}
mtklein8a90edc2015-05-13 12:19:42 -070059
mtkleincbf4fba2015-11-17 14:19:52 -080060inline Sk4px Sk4px::Wide::div255() const {
mtklein9d344062015-12-07 08:21:11 -080061 // Calculated as (x + (x+128)>>8 +128) >> 8. The 'r' in each instruction provides each +128.
62 return Sk16b(vcombine_u8(vraddhn_u16(this->fLo.fVec, vrshrq_n_u16(this->fLo.fVec, 8)),
63 vraddhn_u16(this->fHi.fVec, vrshrq_n_u16(this->fHi.fVec, 8))));
mtkleincbf4fba2015-11-17 14:19:52 -080064}
65
mtklein082e3292015-08-12 11:56:43 -070066inline Sk4px Sk4px::alphas() const {
mtklein343c7d12015-06-22 11:00:47 -070067 auto as = vshrq_n_u32((uint32x4_t)fVec, SK_A32_SHIFT); // ___3 ___2 ___1 ___0
68 return Sk16b((uint8x16_t)vmulq_n_u32(as, 0x01010101)); // 3333 2222 1111 0000
mtklein8a90edc2015-05-13 12:19:42 -070069}
70
mtklein082e3292015-08-12 11:56:43 -070071inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
mtklein343c7d12015-06-22 11:00:47 -070072 uint8x16_t a8 = vdupq_n_u8(0); // ____ ____ ____ ____
73 a8 = vld1q_lane_u8(a+0, a8, 0); // ____ ____ ____ ___0
74 a8 = vld1q_lane_u8(a+1, a8, 4); // ____ ____ ___1 ___0
75 a8 = vld1q_lane_u8(a+2, a8, 8); // ____ ___2 ___1 ___0
76 a8 = vld1q_lane_u8(a+3, a8, 12); // ___3 ___2 ___1 ___0
77 auto a32 = (uint32x4_t)a8; //
78 return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101)); // 3333 2222 1111 0000
mtklein8a90edc2015-05-13 12:19:42 -070079}
80
mtklein082e3292015-08-12 11:56:43 -070081inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
mtklein343c7d12015-06-22 11:00:47 -070082 uint8x16_t a8 = vdupq_n_u8(0); // ____ ____ ____ ____
83 a8 = vld1q_lane_u8(a+0, a8, 0); // ____ ____ ____ ___0
84 a8 = vld1q_lane_u8(a+1, a8, 4); // ____ ____ ___1 ___0
85 auto a32 = (uint32x4_t)a8; //
86 return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101)); // ____ ____ 1111 0000
mtklein8a90edc2015-05-13 12:19:42 -070087}
mtklein0135a412015-05-15 10:36:21 -070088
mtklein082e3292015-08-12 11:56:43 -070089inline Sk4px Sk4px::zeroColors() const {
mtklein0135a412015-05-15 10:36:21 -070090 return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
91}
92
mtklein082e3292015-08-12 11:56:43 -070093inline Sk4px Sk4px::zeroAlphas() const {
mtklein0135a412015-05-15 10:36:21 -070094 // vbic(a,b) == a & ~b
95 return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
96}
97
mtklein082e3292015-08-12 11:56:43 -070098static inline uint8x16_t widen_to_8888(uint16x4_t v) {
mtkleinced15852015-07-22 10:52:53 -070099 // 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 shift 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
mtklein082e3292015-08-12 11:56:43 -0700124static inline uint16x4_t narrow_to_565(uint8x16_t w8x16) {
mtkleinced15852015-07-22 10:52:53 -0700125 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
mtklein082e3292015-08-12 11:56:43 -0700145inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
mtkleinced15852015-07-22 10:52:53 -0700146 return Sk16b(widen_to_8888(vld1_u16(src)));
147}
mtklein082e3292015-08-12 11:56:43 -0700148inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
mtkleinced15852015-07-22 10:52:53 -0700149 auto src2 = ((uint32_t)src[0] )
150 | ((uint32_t)src[1] << 16);
151 return Sk16b(widen_to_8888(vcreate_u16(src2)));
152}
mtklein082e3292015-08-12 11:56:43 -0700153inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
mtkleinced15852015-07-22 10:52:53 -0700154 return Sk16b(widen_to_8888(vcreate_u16(src[0])));
155}
156
mtklein082e3292015-08-12 11:56:43 -0700157inline void Sk4px::store4(SkPMColor16 dst[4]) const {
mtkleinced15852015-07-22 10:52:53 -0700158 vst1_u16(dst, narrow_to_565(this->fVec));
159}
mtklein082e3292015-08-12 11:56:43 -0700160inline void Sk4px::store2(SkPMColor16 dst[2]) const {
mtkleinced15852015-07-22 10:52:53 -0700161 auto v = narrow_to_565(this->fVec);
162 dst[0] = vget_lane_u16(v, 0);
163 dst[1] = vget_lane_u16(v, 1);
164}
mtklein082e3292015-08-12 11:56:43 -0700165inline void Sk4px::store1(SkPMColor16 dst[1]) const {
mtkleinced15852015-07-22 10:52:53 -0700166 dst[0] = vget_lane_u16(narrow_to_565(this->fVec), 0);
167}
168
mtklein082e3292015-08-12 11:56:43 -0700169} // namespace
170