reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2016 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 | |
| 8 | #include "SkHalf.h" |
| 9 | #include "SkPM4fPriv.h" |
| 10 | #include "SkUtils.h" |
| 11 | #include "SkXfermode.h" |
| 12 | |
| 13 | static Sk4f lerp_by_coverage(const Sk4f& src, const Sk4f& dst, uint8_t srcCoverage) { |
| 14 | return dst + (src - dst) * Sk4f(srcCoverage * (1/255.0f)); |
| 15 | } |
| 16 | |
| 17 | /////////////////////////////////////////////////////////////////////////////////////////////////// |
| 18 | |
| 19 | static void xfer_1(const SkXfermode* xfer, uint64_t dst[], const SkPM4f* src, int count, |
| 20 | const SkAlpha aa[]) { |
| 21 | SkXfermodeProc4f proc = xfer->getProc4f(); |
| 22 | SkPM4f d; |
| 23 | if (aa) { |
| 24 | for (int i = 0; i < count; ++i) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 25 | Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 26 | d4.store(d.fVec); |
| 27 | Sk4f r4 = Sk4f::Load(proc(*src, d).fVec); |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 28 | SkFloatToHalf_finite_ftz(lerp_by_coverage(r4, d4, aa[i])).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 29 | } |
| 30 | } else { |
| 31 | for (int i = 0; i < count; ++i) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 32 | SkHalfToFloat_finite_ftz(dst[i]).store(d.fVec); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 33 | Sk4f r4 = Sk4f::Load(proc(*src, d).fVec); |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 34 | SkFloatToHalf_finite_ftz(r4).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 35 | } |
| 36 | } |
| 37 | } |
| 38 | |
| 39 | static void xfer_n(const SkXfermode* xfer, uint64_t dst[], const SkPM4f src[], int count, |
| 40 | const SkAlpha aa[]) { |
| 41 | SkXfermodeProc4f proc = xfer->getProc4f(); |
| 42 | SkPM4f d; |
| 43 | if (aa) { |
| 44 | for (int i = 0; i < count; ++i) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 45 | Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 46 | d4.store(d.fVec); |
| 47 | Sk4f r4 = Sk4f::Load(proc(src[i], d).fVec); |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 48 | SkFloatToHalf_finite_ftz(lerp_by_coverage(r4, d4, aa[i])).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 49 | } |
| 50 | } else { |
| 51 | for (int i = 0; i < count; ++i) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 52 | SkHalfToFloat_finite_ftz(dst[i]).store(d.fVec); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 53 | Sk4f r4 = Sk4f::Load(proc(src[i], d).fVec); |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 54 | SkFloatToHalf_finite_ftz(r4).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 55 | } |
| 56 | } |
| 57 | } |
| 58 | |
| 59 | const SkXfermode::F16Proc gProcs_General[] = { xfer_n, xfer_n, xfer_1, xfer_1 }; |
| 60 | |
| 61 | /////////////////////////////////////////////////////////////////////////////////////////////////// |
| 62 | |
| 63 | static void clear(const SkXfermode*, uint64_t dst[], const SkPM4f*, int count, const SkAlpha aa[]) { |
| 64 | if (aa) { |
| 65 | for (int i = 0; i < count; ++i) { |
| 66 | if (aa[i]) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 67 | const Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]); |
| 68 | SkFloatToHalf_finite_ftz(d4 * Sk4f((255 - aa[i]) * 1.0f/255)).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 69 | } |
| 70 | } |
| 71 | } else { |
| 72 | sk_memset64(dst, 0, count); |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | const SkXfermode::F16Proc gProcs_Clear[] = { clear, clear, clear, clear }; |
| 77 | |
| 78 | /////////////////////////////////////////////////////////////////////////////////////////////////// |
| 79 | |
| 80 | static void src_1(const SkXfermode*, uint64_t dst[], const SkPM4f* src, int count, |
| 81 | const SkAlpha aa[]) { |
| 82 | const Sk4f s4 = Sk4f::Load(src->fVec); |
| 83 | if (aa) { |
| 84 | for (int i = 0; i < count; ++i) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 85 | const Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]); |
| 86 | SkFloatToHalf_finite_ftz(lerp_by_coverage(s4, d4, aa[i])).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 87 | } |
| 88 | } else { |
msarett | 6bdbf44 | 2016-07-19 09:07:55 -0700 | [diff] [blame] | 89 | uint64_t s4h; |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 90 | SkFloatToHalf_finite_ftz(s4).store(&s4h); |
msarett | 6bdbf44 | 2016-07-19 09:07:55 -0700 | [diff] [blame] | 91 | sk_memset64(dst, s4h, count); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 92 | } |
| 93 | } |
| 94 | |
| 95 | static void src_n(const SkXfermode*, uint64_t dst[], const SkPM4f src[], int count, |
| 96 | const SkAlpha aa[]) { |
| 97 | if (aa) { |
| 98 | for (int i = 0; i < count; ++i) { |
| 99 | const Sk4f s4 = Sk4f::Load(src[i].fVec); |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 100 | const Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]); |
| 101 | SkFloatToHalf_finite_ftz(lerp_by_coverage(s4, d4, aa[i])).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 102 | } |
| 103 | } else { |
| 104 | for (int i = 0; i < count; ++i) { |
| 105 | const Sk4f s4 = Sk4f::Load(src[i].fVec); |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 106 | SkFloatToHalf_finite_ftz(s4).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 107 | } |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | const SkXfermode::F16Proc gProcs_Src[] = { src_n, src_n, src_1, src_1 }; |
| 112 | |
| 113 | /////////////////////////////////////////////////////////////////////////////////////////////////// |
| 114 | |
| 115 | static void dst(const SkXfermode*, uint64_t*, const SkPM4f*, int count, const SkAlpha[]) {} |
| 116 | |
| 117 | const SkXfermode::F16Proc gProcs_Dst[] = { dst, dst, dst, dst }; |
| 118 | |
| 119 | /////////////////////////////////////////////////////////////////////////////////////////////////// |
| 120 | |
| 121 | static void srcover_1(const SkXfermode*, uint64_t dst[], const SkPM4f* src, int count, |
| 122 | const SkAlpha aa[]) { |
| 123 | const Sk4f s4 = Sk4f::Load(src->fVec); |
| 124 | const Sk4f dst_scale = Sk4f(1 - get_alpha(s4)); |
| 125 | for (int i = 0; i < count; ++i) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 126 | const Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 127 | const Sk4f r4 = s4 + d4 * dst_scale; |
| 128 | if (aa) { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 129 | SkFloatToHalf_finite_ftz(lerp_by_coverage(r4, d4, aa[i])).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 130 | } else { |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 131 | SkFloatToHalf_finite_ftz(r4).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 132 | } |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | static void srcover_n(const SkXfermode*, uint64_t dst[], const SkPM4f src[], int count, |
| 137 | const SkAlpha aa[]) { |
| 138 | for (int i = 0; i < count; ++i) { |
mtklein | 244a653 | 2016-04-19 14:21:30 -0700 | [diff] [blame] | 139 | Sk4f s = Sk4f::Load(src+i), |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 140 | d = SkHalfToFloat_finite_ftz(dst[i]), |
mtklein | 244a653 | 2016-04-19 14:21:30 -0700 | [diff] [blame] | 141 | r = s + d*(1.0f - SkNx_shuffle<3,3,3,3>(s)); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 142 | if (aa) { |
mtklein | 244a653 | 2016-04-19 14:21:30 -0700 | [diff] [blame] | 143 | r = lerp_by_coverage(r, d, aa[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 144 | } |
mtklein | 8ae991e | 2016-08-22 13:20:18 -0700 | [diff] [blame^] | 145 | SkFloatToHalf_finite_ftz(r).store(&dst[i]); |
reed | 3dc6aac | 2016-04-14 09:02:14 -0700 | [diff] [blame] | 146 | } |
| 147 | } |
| 148 | |
| 149 | const SkXfermode::F16Proc gProcs_SrcOver[] = { srcover_n, src_n, srcover_1, src_1 }; |
| 150 | |
| 151 | /////////////////////////////////////////////////////////////////////////////////////////////////// |
| 152 | |
| 153 | static SkXfermode::F16Proc find_proc(SkXfermode::Mode mode, uint32_t flags) { |
| 154 | SkASSERT(0 == (flags & ~3)); |
| 155 | flags &= 3; |
| 156 | |
| 157 | switch (mode) { |
| 158 | case SkXfermode::kClear_Mode: return gProcs_Clear[flags]; |
| 159 | case SkXfermode::kSrc_Mode: return gProcs_Src[flags]; |
| 160 | case SkXfermode::kDst_Mode: return gProcs_Dst[flags]; |
| 161 | case SkXfermode::kSrcOver_Mode: return gProcs_SrcOver[flags]; |
| 162 | default: |
| 163 | break; |
| 164 | } |
| 165 | return gProcs_General[flags]; |
| 166 | } |
| 167 | |
| 168 | SkXfermode::F16Proc SkXfermode::onGetF16Proc(uint32_t flags) const { |
| 169 | SkASSERT(0 == (flags & ~3)); |
| 170 | flags &= 3; |
| 171 | |
| 172 | Mode mode; |
| 173 | return this->asMode(&mode) ? find_proc(mode, flags) : gProcs_General[flags]; |
| 174 | } |
| 175 | |
| 176 | SkXfermode::F16Proc SkXfermode::GetF16Proc(SkXfermode* xfer, uint32_t flags) { |
| 177 | return xfer ? xfer->onGetF16Proc(flags) : find_proc(SkXfermode::kSrcOver_Mode, flags); |
| 178 | } |