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
8namespace { // See Sk4px.h
9
10inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b((uint8x16_t)vdupq_n_u32(px)); }
11
12inline Sk4px Sk4px::Load4(const SkPMColor px[4]) {
13    return Sk16b((uint8x16_t)vld1q_u32(px));
14}
15inline Sk4px Sk4px::Load2(const SkPMColor px[2]) {
16    uint32x2_t px2 = vld1_u32(px);
17    return Sk16b((uint8x16_t)vcombine_u32(px2, px2));
18}
19inline Sk4px Sk4px::Load1(const SkPMColor px[1]) {
20    return Sk16b((uint8x16_t)vdupq_n_u32(*px));
21}
22
23inline void Sk4px::store4(SkPMColor px[4]) const {
24    vst1q_u32(px, (uint32x4_t)this->fVec);
25}
26inline void Sk4px::store2(SkPMColor px[2]) const {
27    vst1_u32(px, (uint32x2_t)vget_low_u8(this->fVec));
28}
29inline void Sk4px::store1(SkPMColor px[1]) const {
30    vst1q_lane_u32(px, (uint32x4_t)this->fVec, 0);
31}
32
33inline Sk4px::Wide Sk4px::widenLo() const {
34    return Sk16h(vmovl_u8(vget_low_u8 (this->fVec)),
35                 vmovl_u8(vget_high_u8(this->fVec)));
36}
37
38inline Sk4px::Wide Sk4px::widenHi() const {
39    return Sk16h(vshll_n_u8(vget_low_u8 (this->fVec), 8),
40                 vshll_n_u8(vget_high_u8(this->fVec), 8));
41}
42
43inline Sk4px::Wide Sk4px::widenLoHi() const {
44    auto zipped = vzipq_u8(this->fVec, this->fVec);
45    return Sk16h((uint16x8_t)zipped.val[0],
46                 (uint16x8_t)zipped.val[1]);
47}
48
49inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
50    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
54inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
55    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}
59
60inline Sk4px Sk4px::Wide::div255() const {
61    // 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))));
64}
65
66inline Sk4px Sk4px::alphas() const {
67    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
69}
70
71inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
72    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
79}
80
81inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
82    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
87}
88
89inline Sk4px Sk4px::zeroColors() const {
90    return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
91}
92
93inline Sk4px Sk4px::zeroAlphas() const {
94    // vbic(a,b) == a & ~b
95    return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
96}
97
98} // namespace
99
100