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
sk_memset64(uint64_t dst[],uint64_t value,int count)13 static void sk_memset64(uint64_t dst[], uint64_t value, int count) {
14 for (int i = 0; i < count; ++i) {
15 dst[i] = value;
16 }
17 }
18
19 enum DstType {
20 kU16_Dst,
21 kF16_Dst,
22 };
23
lerp_by_coverage(const Sk4f & src,const Sk4f & dst,uint8_t srcCoverage)24 static Sk4f lerp_by_coverage(const Sk4f& src, const Sk4f& dst, uint8_t srcCoverage) {
25 return dst + (src - dst) * Sk4f(srcCoverage * (1/255.0f));
26 }
27
unit_to_bias(const Sk4f & x4)28 template <DstType D> Sk4f unit_to_bias(const Sk4f& x4) {
29 return (D == kU16_Dst) ? x4 * Sk4f(65535) : x4;
30 }
31
bias_to_unit(const Sk4f & x4)32 template <DstType D> Sk4f bias_to_unit(const Sk4f& x4) {
33 return (D == kU16_Dst) ? x4 * Sk4f(1.0f/65535) : x4;
34 }
35
36 // returns value already biased by 65535
load_from_u16(uint64_t value)37 static Sk4f load_from_u16(uint64_t value) {
38 return SkNx_cast<float>(Sk4h::Load(&value));
39 }
40
41 // takes floats already biased by 65535
store_to_u16(const Sk4f & x4)42 static uint64_t store_to_u16(const Sk4f& x4) {
43 uint64_t value;
44 SkNx_cast<uint16_t>(x4 + Sk4f(0.5f)).store(&value);
45 return value;
46 }
47
48 // Returns dst in its "natural" bias (either unit-float or 16bit int)
49 //
load_from_dst(uint64_t dst)50 template <DstType D> Sk4f load_from_dst(uint64_t dst) {
51 return (D == kU16_Dst) ? load_from_u16(dst) : SkHalfToFloat_01(dst);
52 }
53
54 // Assumes x4 is already in the "natural" bias (either unit-float or 16bit int)
store_to_dst(const Sk4f & x4)55 template <DstType D> uint64_t store_to_dst(const Sk4f& x4) {
56 return (D == kU16_Dst) ? store_to_u16(x4) : SkFloatToHalf_01(x4);
57 }
58
pm_to_rgba_order(const Sk4f & x)59 static inline Sk4f pm_to_rgba_order(const Sk4f& x) {
60 if (SkPM4f::R == 0) {
61 return x; // we're already RGBA
62 } else {
63 // we're BGRA, so swap R and B
64 return SkNx_shuffle<2, 1, 0, 3>(x);
65 }
66 }
67
68 ///////////////////////////////////////////////////////////////////////////////////////////////////
69
xfer_u64_1(const SkXfermode * xfer,uint64_t dst[],const SkPM4f * src,int count,const SkAlpha aa[])70 template <DstType D> void xfer_u64_1(const SkXfermode* xfer, uint64_t dst[],
71 const SkPM4f* src, int count, const SkAlpha aa[]) {
72 SkXfermodeProc4f proc = xfer->getProc4f();
73 SkPM4f d;
74 if (aa) {
75 for (int i = 0; i < count; ++i) {
76 Sk4f d4 = bias_to_unit<D>(load_from_dst<D>(dst[i]));
77 d4.store(d.fVec);
78 Sk4f r4 = unit_to_bias<D>(Sk4f::Load(proc(*src, d).fVec));
79 dst[i] = store_to_dst<D>(lerp_by_coverage(r4, d4, aa[i]));
80 }
81 } else {
82 for (int i = 0; i < count; ++i) {
83 bias_to_unit<D>(load_from_dst<D>(dst[i])).store(d.fVec);
84 Sk4f r4 = unit_to_bias<D>(Sk4f::Load(proc(*src, d).fVec));
85 dst[i] = store_to_dst<D>(r4);
86 }
87 }
88 }
89
xfer_u64_n(const SkXfermode * xfer,uint64_t dst[],const SkPM4f src[],int count,const SkAlpha aa[])90 template <DstType D> void xfer_u64_n(const SkXfermode* xfer, uint64_t dst[],
91 const SkPM4f src[], int count, const SkAlpha aa[]) {
92 SkXfermodeProc4f proc = xfer->getProc4f();
93 SkPM4f d;
94 if (aa) {
95 for (int i = 0; i < count; ++i) {
96 Sk4f d4 = bias_to_unit<D>(load_from_dst<D>(dst[i]));
97 d4.store(d.fVec);
98 Sk4f r4 = unit_to_bias<D>(Sk4f::Load(proc(src[i], d).fVec));
99 dst[i] = store_to_dst<D>(lerp_by_coverage(r4, d4, aa[i]));
100 }
101 } else {
102 for (int i = 0; i < count; ++i) {
103 bias_to_unit<D>(load_from_dst<D>(dst[i])).store(d.fVec);
104 Sk4f r4 = unit_to_bias<D>(Sk4f::Load(proc(src[i], d).fVec));
105 dst[i] = store_to_dst<D>(r4);
106 }
107 }
108 }
109
110 const SkXfermode::D64Proc gProcs_General[] = {
111 xfer_u64_n<kU16_Dst>, xfer_u64_n<kU16_Dst>,
112 xfer_u64_1<kU16_Dst>, xfer_u64_1<kU16_Dst>,
113 xfer_u64_n<kF16_Dst>, xfer_u64_n<kF16_Dst>,
114 xfer_u64_1<kF16_Dst>, xfer_u64_1<kF16_Dst>,
115 };
116
117 ///////////////////////////////////////////////////////////////////////////////////////////////////
118
clear(const SkXfermode *,uint64_t dst[],const SkPM4f *,int count,const SkAlpha aa[])119 template <DstType D> void clear(const SkXfermode*, uint64_t dst[],
120 const SkPM4f*, int count, const SkAlpha aa[]) {
121 if (aa) {
122 for (int i = 0; i < count; ++i) {
123 if (aa[i]) {
124 const Sk4f d4 = load_from_dst<D>(dst[i]);
125 dst[i] = store_to_dst<D>(d4 * Sk4f((255 - aa[i]) * 1.0f/255));
126 }
127 }
128 } else {
129 sk_memset64(dst, 0, count);
130 }
131 }
132
133 const SkXfermode::D64Proc gProcs_Clear[] = {
134 clear<kU16_Dst>, clear<kU16_Dst>,
135 clear<kU16_Dst>, clear<kU16_Dst>,
136 clear<kF16_Dst>, clear<kF16_Dst>,
137 clear<kF16_Dst>, clear<kF16_Dst>,
138 };
139
140 ///////////////////////////////////////////////////////////////////////////////////////////////////
141
src_1(const SkXfermode *,uint64_t dst[],const SkPM4f * src,int count,const SkAlpha aa[])142 template <DstType D> void src_1(const SkXfermode*, uint64_t dst[],
143 const SkPM4f* src, int count, const SkAlpha aa[]) {
144 const Sk4f s4 = pm_to_rgba_order(unit_to_bias<D>(Sk4f::Load(src->fVec)));
145 if (aa) {
146 for (int i = 0; i < count; ++i) {
147 const Sk4f d4 = load_from_dst<D>(dst[i]);
148 dst[i] = store_to_dst<D>(lerp_by_coverage(s4, d4, aa[i]));
149 }
150 } else {
151 sk_memset64(dst, store_to_dst<D>(s4), count);
152 }
153 }
154
src_n(const SkXfermode *,uint64_t dst[],const SkPM4f src[],int count,const SkAlpha aa[])155 template <DstType D> void src_n(const SkXfermode*, uint64_t dst[],
156 const SkPM4f src[], int count, const SkAlpha aa[]) {
157 if (aa) {
158 for (int i = 0; i < count; ++i) {
159 const Sk4f s4 = pm_to_rgba_order(unit_to_bias<D>(Sk4f::Load(src[i].fVec)));
160 const Sk4f d4 = load_from_dst<D>(dst[i]);
161 dst[i] = store_to_dst<D>(lerp_by_coverage(s4, d4, aa[i]));
162 }
163 } else {
164 for (int i = 0; i < count; ++i) {
165 const Sk4f s4 = pm_to_rgba_order(unit_to_bias<D>(Sk4f::Load(src[i].fVec)));
166 dst[i] = store_to_dst<D>(s4);
167 }
168 }
169 }
170
171 const SkXfermode::D64Proc gProcs_Src[] = {
172 src_n<kU16_Dst>, src_n<kU16_Dst>,
173 src_1<kU16_Dst>, src_1<kU16_Dst>,
174 src_n<kF16_Dst>, src_n<kF16_Dst>,
175 src_1<kF16_Dst>, src_1<kF16_Dst>,
176 };
177
178 ///////////////////////////////////////////////////////////////////////////////////////////////////
179
dst(const SkXfermode *,uint64_t *,const SkPM4f *,int count,const SkAlpha[])180 static void dst(const SkXfermode*, uint64_t*, const SkPM4f*, int count, const SkAlpha[]) {}
181
182 const SkXfermode::D64Proc gProcs_Dst[] = {
183 dst, dst, dst, dst, dst, dst, dst, dst,
184 };
185
186 ///////////////////////////////////////////////////////////////////////////////////////////////////
187
srcover_1(const SkXfermode *,uint64_t dst[],const SkPM4f * src,int count,const SkAlpha aa[])188 template <DstType D> void srcover_1(const SkXfermode*, uint64_t dst[],
189 const SkPM4f* src, int count, const SkAlpha aa[]) {
190 const Sk4f s4 = pm_to_rgba_order(Sk4f::Load(src->fVec));
191 const Sk4f dst_scale = Sk4f(1 - get_alpha(s4));
192 const Sk4f s4bias = unit_to_bias<D>(s4);
193 for (int i = 0; i < count; ++i) {
194 const Sk4f d4bias = load_from_dst<D>(dst[i]);
195 const Sk4f r4bias = s4bias + d4bias * dst_scale;
196 if (aa) {
197 dst[i] = store_to_dst<D>(lerp_by_coverage(r4bias, d4bias, aa[i]));
198 } else {
199 dst[i] = store_to_dst<D>(r4bias);
200 }
201 }
202 }
203
srcover_n(const SkXfermode *,uint64_t dst[],const SkPM4f src[],int count,const SkAlpha aa[])204 template <DstType D> void srcover_n(const SkXfermode*, uint64_t dst[],
205 const SkPM4f src[], int count, const SkAlpha aa[]) {
206 for (int i = 0; i < count; ++i) {
207 const Sk4f s4 = pm_to_rgba_order(Sk4f::Load(src[i].fVec));
208 const Sk4f dst_scale = Sk4f(1 - get_alpha(s4));
209 const Sk4f s4bias = unit_to_bias<D>(s4);
210 const Sk4f d4bias = load_from_dst<D>(dst[i]);
211 const Sk4f r4bias = s4bias + d4bias * dst_scale;
212 if (aa) {
213 dst[i] = store_to_dst<D>(lerp_by_coverage(r4bias, d4bias, aa[i]));
214 } else {
215 dst[i] = store_to_dst<D>(r4bias);
216 }
217 }
218 }
219
220 const SkXfermode::D64Proc gProcs_SrcOver[] = {
221 srcover_n<kU16_Dst>, src_n<kU16_Dst>,
222 srcover_1<kU16_Dst>, src_1<kU16_Dst>,
223 srcover_n<kF16_Dst>, src_n<kF16_Dst>,
224 srcover_1<kF16_Dst>, src_1<kF16_Dst>,
225 };
226
227 ///////////////////////////////////////////////////////////////////////////////////////////////////
228
find_proc(SkXfermode::Mode mode,uint32_t flags)229 static SkXfermode::D64Proc find_proc(SkXfermode::Mode mode, uint32_t flags) {
230 SkASSERT(0 == (flags & ~7));
231 flags &= 7;
232
233 switch (mode) {
234 case SkXfermode::kClear_Mode: return gProcs_Clear[flags];
235 case SkXfermode::kSrc_Mode: return gProcs_Src[flags];
236 case SkXfermode::kDst_Mode: return gProcs_Dst[flags];
237 case SkXfermode::kSrcOver_Mode: return gProcs_SrcOver[flags];
238 default:
239 break;
240 }
241 return gProcs_General[flags];
242 }
243
onGetD64Proc(uint32_t flags) const244 SkXfermode::D64Proc SkXfermode::onGetD64Proc(uint32_t flags) const {
245 SkASSERT(0 == (flags & ~7));
246 flags &= 7;
247
248 Mode mode;
249 return this->asMode(&mode) ? find_proc(mode, flags) : gProcs_General[flags];
250 }
251
GetD64Proc(SkXfermode * xfer,uint32_t flags)252 SkXfermode::D64Proc SkXfermode::GetD64Proc(SkXfermode* xfer, uint32_t flags) {
253 return xfer ? xfer->onGetD64Proc(flags) : find_proc(SkXfermode::kSrcOver_Mode, flags);
254 }
255
256