1/*
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
8inline Sk4px::Sk4px(SkPMColor px) : INHERITED((uint8x16_t)vdupq_n_u32(px)) {}
9
10inline Sk4px Sk4px::Load4(const SkPMColor px[4]) {
11    return Sk16b((uint8x16_t)vld1q_u32(px));
12}
13inline Sk4px Sk4px::Load2(const SkPMColor px[2]) {
14    uint32x2_t px2 = vld1_u32(px);
15    return Sk16b((uint8x16_t)vcombine_u32(px2, px2));
16}
17inline Sk4px Sk4px::Load1(const SkPMColor px[1]) {
18    return Sk16b((uint8x16_t)vdupq_n_u32(*px));
19}
20
21inline void Sk4px::store4(SkPMColor px[4]) const {
22    vst1q_u32(px, (uint32x4_t)this->fVec);
23}
24inline void Sk4px::store2(SkPMColor px[2]) const {
25    vst1_u32(px, (uint32x2_t)vget_low_u8(this->fVec));
26}
27inline void Sk4px::store1(SkPMColor px[1]) const {
28    vst1q_lane_u32(px, (uint32x4_t)this->fVec, 0);
29}
30
31inline Sk4px::Wide Sk4px::widenLo() const {
32    return Sk16h(vmovl_u8(vget_low_u8 (this->fVec)),
33                 vmovl_u8(vget_high_u8(this->fVec)));
34}
35
36inline Sk4px::Wide Sk4px::widenHi() const {
37    return Sk16h(vshll_n_u8(vget_low_u8 (this->fVec), 8),
38                 vshll_n_u8(vget_high_u8(this->fVec), 8));
39}
40
41inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
42    return Sk16h(vmull_u8(vget_low_u8 (this->fVec), vget_low_u8 (other.fVec)),
43                 vmull_u8(vget_high_u8(this->fVec), vget_high_u8(other.fVec)));
44}
45
46inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
47    const Sk4px::Wide o(other);  // Should be no code, but allows us to access fLo, fHi.
48    return Sk16b(vcombine_u8(vaddhn_u16(this->fLo.fVec, o.fLo.fVec),
49                             vaddhn_u16(this->fHi.fVec, o.fHi.fVec)));
50}
51
52inline Sk4px Sk4px::alphas() const {
53    static_assert(SK_A32_SHIFT == 24, "This method assumes little-endian.");
54    auto as = vshrq_n_u32((uint32x4_t)this->fVec, 24);  // ___3 ___2 ___1 ___0
55    as = vorrq_u32(as, vshlq_n_u32(as,  8));            // __33 __22 __11 __11
56    as = vorrq_u32(as, vshlq_n_u32(as, 16));            // 3333 2222 1111 1111
57    return Sk16b((uint8x16_t)as);
58}
59
60inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
61    uint8x16_t a8 = vdupq_n_u8(0);                        // ____ ____ ____ ____
62    a8 = vld1q_lane_u8(a+0, a8,  0);                      // ____ ____ ____ ___0
63    a8 = vld1q_lane_u8(a+1, a8,  4);                      // ____ ____ ___1 ___0
64    a8 = vld1q_lane_u8(a+2, a8,  8);                      // ____ ___2 ___1 ___0
65    a8 = vld1q_lane_u8(a+3, a8, 12);                      // ___3 ___2 ___1 ___0
66    auto a32 = (uint32x4_t)a8;                            //
67    a32 = vorrq_u32(a32, vshlq_n_u32(a32,  8));           // __33 __22 __11 __00
68    a32 = vorrq_u32(a32, vshlq_n_u32(a32, 16));           // 3333 2222 1111 0000
69    return Sk16b((uint8x16_t)a32);
70}
71
72inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
73    uint8x16_t a8 = vdupq_n_u8(0);                        // ____ ____ ____ ____
74    a8 = vld1q_lane_u8(a+0, a8,  0);                      // ____ ____ ____ ___0
75    a8 = vld1q_lane_u8(a+1, a8,  4);                      // ____ ____ ___1 ___0
76    auto a32 = (uint32x4_t)a8;                            //
77    a32 = vorrq_u32(a32, vshlq_n_u32(a32,  8));           // ____ ____ __11 __00
78    a32 = vorrq_u32(a32, vshlq_n_u32(a32, 16));           // ____ ____ 1111 0000
79    return Sk16b((uint8x16_t)a32);
80}
81
82inline Sk4px Sk4px::zeroColors() const {
83    return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
84}
85
86inline Sk4px Sk4px::zeroAlphas() const {
87    // vbic(a,b) == a & ~b
88    return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
89}
90
91