1d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein/* 2d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein * Copyright 2015 Google Inc. 3d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein * 4d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein * Use of this source code is governed by a BSD-style license that can be 5d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein * found in the LICENSE file. 6d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein */ 7d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein 8082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleinnamespace { // See Sk4px.h 9082e329887a8f1efe4e1020f0a0a6ea09961712dmtklein 10082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b((uint8x16_t)vdupq_n_u32(px)); } 11d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein 12082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::Load4(const SkPMColor px[4]) { 13d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein return Sk16b((uint8x16_t)vld1q_u32(px)); 14d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 15082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::Load2(const SkPMColor px[2]) { 16d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein uint32x2_t px2 = vld1_u32(px); 17d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein return Sk16b((uint8x16_t)vcombine_u32(px2, px2)); 18d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 19082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::Load1(const SkPMColor px[1]) { 20d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein return Sk16b((uint8x16_t)vdupq_n_u32(*px)); 21d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 22d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein 23082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline void Sk4px::store4(SkPMColor px[4]) const { 24d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein vst1q_u32(px, (uint32x4_t)this->fVec); 25d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 26082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline void Sk4px::store2(SkPMColor px[2]) const { 27d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein vst1_u32(px, (uint32x2_t)vget_low_u8(this->fVec)); 28d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 29082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline void Sk4px::store1(SkPMColor px[1]) const { 30d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein vst1q_lane_u32(px, (uint32x4_t)this->fVec, 0); 31d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 32d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein 33082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px::Wide Sk4px::widenLo() const { 34d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein return Sk16h(vmovl_u8(vget_low_u8 (this->fVec)), 35d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein vmovl_u8(vget_high_u8(this->fVec))); 36d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 37d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein 38082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px::Wide Sk4px::widenHi() const { 39d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein return Sk16h(vshll_n_u8(vget_low_u8 (this->fVec), 8), 40d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein vshll_n_u8(vget_high_u8(this->fVec), 8)); 41d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 42d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein 43082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px::Wide Sk4px::widenLoHi() const { 444be181e304d2b280c6801bd13369cfba236d1a66mtklein auto zipped = vzipq_u8(this->fVec, this->fVec); 454be181e304d2b280c6801bd13369cfba236d1a66mtklein return Sk16h((uint16x8_t)zipped.val[0], 464be181e304d2b280c6801bd13369cfba236d1a66mtklein (uint16x8_t)zipped.val[1]); 474be181e304d2b280c6801bd13369cfba236d1a66mtklein} 484be181e304d2b280c6801bd13369cfba236d1a66mtklein 49082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const { 50d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein return Sk16h(vmull_u8(vget_low_u8 (this->fVec), vget_low_u8 (other.fVec)), 51d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein vmull_u8(vget_high_u8(this->fVec), vget_high_u8(other.fVec))); 52d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 53d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein 54082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const { 55d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein const Sk4px::Wide o(other); // Should be no code, but allows us to access fLo, fHi. 56d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein return Sk16b(vcombine_u8(vaddhn_u16(this->fLo.fVec, o.fLo.fVec), 57d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein vaddhn_u16(this->fHi.fVec, o.fHi.fVec))); 58d2ffd36eb62e99abe2920369d1e040954cc2044fmtklein} 598a90edc2a58a4f8a4b4da73eb08e943be09538c0mtklein 60cbf4fba43933302a846872e4c5ce8f1adb8b325emtkleininline Sk4px Sk4px::Wide::div255() const { 619d344069c5d45a936823d3f84999898383a026cdmtklein // Calculated as (x + (x+128)>>8 +128) >> 8. The 'r' in each instruction provides each +128. 629d344069c5d45a936823d3f84999898383a026cdmtklein return Sk16b(vcombine_u8(vraddhn_u16(this->fLo.fVec, vrshrq_n_u16(this->fLo.fVec, 8)), 639d344069c5d45a936823d3f84999898383a026cdmtklein vraddhn_u16(this->fHi.fVec, vrshrq_n_u16(this->fHi.fVec, 8)))); 64cbf4fba43933302a846872e4c5ce8f1adb8b325emtklein} 65cbf4fba43933302a846872e4c5ce8f1adb8b325emtklein 66082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::alphas() const { 67343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein auto as = vshrq_n_u32((uint32x4_t)fVec, SK_A32_SHIFT); // ___3 ___2 ___1 ___0 68343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein return Sk16b((uint8x16_t)vmulq_n_u32(as, 0x01010101)); // 3333 2222 1111 0000 698a90edc2a58a4f8a4b4da73eb08e943be09538c0mtklein} 708a90edc2a58a4f8a4b4da73eb08e943be09538c0mtklein 71082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) { 72343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein uint8x16_t a8 = vdupq_n_u8(0); // ____ ____ ____ ____ 73343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein a8 = vld1q_lane_u8(a+0, a8, 0); // ____ ____ ____ ___0 74343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein a8 = vld1q_lane_u8(a+1, a8, 4); // ____ ____ ___1 ___0 75343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein a8 = vld1q_lane_u8(a+2, a8, 8); // ____ ___2 ___1 ___0 76343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein a8 = vld1q_lane_u8(a+3, a8, 12); // ___3 ___2 ___1 ___0 77343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein auto a32 = (uint32x4_t)a8; // 78343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101)); // 3333 2222 1111 0000 798a90edc2a58a4f8a4b4da73eb08e943be09538c0mtklein} 808a90edc2a58a4f8a4b4da73eb08e943be09538c0mtklein 81082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) { 82343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein uint8x16_t a8 = vdupq_n_u8(0); // ____ ____ ____ ____ 83343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein a8 = vld1q_lane_u8(a+0, a8, 0); // ____ ____ ____ ___0 84343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein a8 = vld1q_lane_u8(a+1, a8, 4); // ____ ____ ___1 ___0 85343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein auto a32 = (uint32x4_t)a8; // 86343c7d15c831a3c3d8a6984c1a1b524b414012dfmtklein return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101)); // ____ ____ 1111 0000 878a90edc2a58a4f8a4b4da73eb08e943be09538c0mtklein} 880135a41e095a433414e21e37b277dab7dcbec373mtklein 89082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::zeroColors() const { 900135a41e095a433414e21e37b277dab7dcbec373mtklein return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT))); 910135a41e095a433414e21e37b277dab7dcbec373mtklein} 920135a41e095a433414e21e37b277dab7dcbec373mtklein 93082e329887a8f1efe4e1020f0a0a6ea09961712dmtkleininline Sk4px Sk4px::zeroAlphas() const { 940135a41e095a433414e21e37b277dab7dcbec373mtklein // vbic(a,b) == a & ~b 950135a41e095a433414e21e37b277dab7dcbec373mtklein return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT))); 960135a41e095a433414e21e37b277dab7dcbec373mtklein} 970135a41e095a433414e21e37b277dab7dcbec373mtklein 98082e329887a8f1efe4e1020f0a0a6ea09961712dmtklein} // namespace 99082e329887a8f1efe4e1020f0a0a6ea09961712dmtklein 100