OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2015 Google Inc. | 2 * Copyright 2015 Google Inc. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
6 */ | 6 */ |
7 | 7 |
8 #ifndef Sk4px_DEFINED | 8 #ifndef Sk4px_DEFINED |
9 #define Sk4px_DEFINED | 9 #define Sk4px_DEFINED |
10 | 10 |
11 #include "SkNx.h" | 11 #include "SkNx.h" |
12 #include "SkColor.h" | 12 #include "SkColor.h" |
| 13 #include "SkColorPriv.h" |
13 | 14 |
14 // This file may be included multiple times by .cpp files with different flags,
leading | 15 // This file may be included multiple times by .cpp files with different flags,
leading |
15 // to different definitions. Usually that doesn't matter because it's all inlin
ed, but | 16 // to different definitions. Usually that doesn't matter because it's all inlin
ed, but |
16 // in Debug modes the compilers may not inline everything. So wrap everything i
n an | 17 // in Debug modes the compilers may not inline everything. So wrap everything i
n an |
17 // anonymous namespace to give each includer their own silo of this code (or the
linker | 18 // anonymous namespace to give each includer their own silo of this code (or the
linker |
18 // will probably pick one randomly for us, which is rarely correct). | 19 // will probably pick one randomly for us, which is rarely correct). |
19 namespace { | 20 namespace { |
20 | 21 |
21 // 1, 2 or 4 SkPMColors, generally vectorized. | 22 // 1, 2 or 4 SkPMColors, generally vectorized. |
22 class Sk4px : public Sk16b { | 23 class Sk4px : public Sk16b { |
(...skipping 17 matching lines...) Expand all Loading... |
40 static Sk4px Load1(const SkPMColor[1]); // PMColor[1] -> ARGB ???? ???? ???
? | 41 static Sk4px Load1(const SkPMColor[1]); // PMColor[1] -> ARGB ???? ???? ???
? |
41 | 42 |
42 // Ditto for Alphas... Load2Alphas fills the low two lanes of Sk4px. | 43 // Ditto for Alphas... Load2Alphas fills the low two lanes of Sk4px. |
43 static Sk4px Load4Alphas(const SkAlpha[4]); // AaXx -> AAAA aaaa XXXX xxxx | 44 static Sk4px Load4Alphas(const SkAlpha[4]); // AaXx -> AAAA aaaa XXXX xxxx |
44 static Sk4px Load2Alphas(const SkAlpha[2]); // Aa -> AAAA aaaa ???? ???? | 45 static Sk4px Load2Alphas(const SkAlpha[2]); // Aa -> AAAA aaaa ???? ???? |
45 | 46 |
46 void store4(SkPMColor[4]) const; | 47 void store4(SkPMColor[4]) const; |
47 void store2(SkPMColor[2]) const; | 48 void store2(SkPMColor[2]) const; |
48 void store1(SkPMColor[1]) const; | 49 void store1(SkPMColor[1]) const; |
49 | 50 |
| 51 // Same as above for 565. |
| 52 static Sk4px Load4(const SkPMColor16 src[4]); |
| 53 static Sk4px Load2(const SkPMColor16 src[2]); |
| 54 static Sk4px Load1(const SkPMColor16 src[1]); |
| 55 void store4(SkPMColor16 dst[4]) const; |
| 56 void store2(SkPMColor16 dst[2]) const; |
| 57 void store1(SkPMColor16 dst[1]) const; |
| 58 |
50 // 1, 2, or 4 SkPMColors with 16-bit components. | 59 // 1, 2, or 4 SkPMColors with 16-bit components. |
51 // This is most useful as the result of a multiply, e.g. from mulWiden(). | 60 // This is most useful as the result of a multiply, e.g. from mulWiden(). |
52 class Wide : public Sk16h { | 61 class Wide : public Sk16h { |
53 public: | 62 public: |
54 Wide(const Sk16h& v) : Sk16h(v) {} | 63 Wide(const Sk16h& v) : Sk16h(v) {} |
55 | 64 |
56 // Pack the top byte of each component back down into 4 SkPMColors. | 65 // Pack the top byte of each component back down into 4 SkPMColors. |
57 Sk4px addNarrowHi(const Sk16h&) const; | 66 Sk4px addNarrowHi(const Sk16h&) const; |
58 | 67 |
59 // Rounds, i.e. (x+127) / 255. | 68 // Rounds, i.e. (x+127) / 255. |
(...skipping 32 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
92 | 101 |
93 // Generally faster than (*this * o).div255(). | 102 // Generally faster than (*this * o).div255(). |
94 // May be incorrect by +-1, but is always exactly correct when *this or o is
0 or 255. | 103 // May be incorrect by +-1, but is always exactly correct when *this or o is
0 or 255. |
95 Sk4px approxMulDiv255(const Sk16b& o) const { | 104 Sk4px approxMulDiv255(const Sk16b& o) const { |
96 // (x*y + x) / 256 meets these criteria. (As of course does (x*y + y) /
256 by symmetry.) | 105 // (x*y + x) / 256 meets these criteria. (As of course does (x*y + y) /
256 by symmetry.) |
97 return this->widenLo().addNarrowHi(*this * o); | 106 return this->widenLo().addNarrowHi(*this * o); |
98 } | 107 } |
99 | 108 |
100 // A generic driver that maps fn over a src array into a dst array. | 109 // A generic driver that maps fn over a src array into a dst array. |
101 // fn should take an Sk4px (4 src pixels) and return an Sk4px (4 dst pixels)
. | 110 // fn should take an Sk4px (4 src pixels) and return an Sk4px (4 dst pixels)
. |
102 template <typename Fn> | 111 template <typename Fn, typename Dst> |
103 static void MapSrc(int n, SkPMColor* dst, const SkPMColor* src, const Fn& fn
) { | 112 static void MapSrc(int n, Dst* dst, const SkPMColor* src, const Fn& fn) { |
104 // This looks a bit odd, but it helps loop-invariant hoisting across dif
ferent calls to fn. | 113 // This looks a bit odd, but it helps loop-invariant hoisting across dif
ferent calls to fn. |
105 // Basically, we need to make sure we keep things inside a single loop. | 114 // Basically, we need to make sure we keep things inside a single loop. |
106 while (n > 0) { | 115 while (n > 0) { |
107 if (n >= 8) { | 116 if (n >= 8) { |
108 Sk4px dst0 = fn(Load4(src+0)), | 117 Sk4px dst0 = fn(Load4(src+0)), |
109 dst4 = fn(Load4(src+4)); | 118 dst4 = fn(Load4(src+4)); |
110 dst0.store4(dst+0); | 119 dst0.store4(dst+0); |
111 dst4.store4(dst+4); | 120 dst4.store4(dst+4); |
112 dst += 8; src += 8; n -= 8; | 121 dst += 8; src += 8; n -= 8; |
113 continue; // Keep our stride at 8 pixels as long as possible. | 122 continue; // Keep our stride at 8 pixels as long as possible. |
114 } | 123 } |
115 SkASSERT(n <= 7); | 124 SkASSERT(n <= 7); |
116 if (n >= 4) { | 125 if (n >= 4) { |
117 fn(Load4(src)).store4(dst); | 126 fn(Load4(src)).store4(dst); |
118 dst += 4; src += 4; n -= 4; | 127 dst += 4; src += 4; n -= 4; |
119 } | 128 } |
120 if (n >= 2) { | 129 if (n >= 2) { |
121 fn(Load2(src)).store2(dst); | 130 fn(Load2(src)).store2(dst); |
122 dst += 2; src += 2; n -= 2; | 131 dst += 2; src += 2; n -= 2; |
123 } | 132 } |
124 if (n >= 1) { | 133 if (n >= 1) { |
125 fn(Load1(src)).store1(dst); | 134 fn(Load1(src)).store1(dst); |
126 } | 135 } |
127 break; | 136 break; |
128 } | 137 } |
129 } | 138 } |
130 | 139 |
131 // As above, but with dst4' = fn(dst4, src4). | 140 // As above, but with dst4' = fn(dst4, src4). |
132 template <typename Fn> | 141 template <typename Fn, typename Dst> |
133 static void MapDstSrc(int n, SkPMColor* dst, const SkPMColor* src, const Fn&
fn) { | 142 static void MapDstSrc(int n, Dst* dst, const SkPMColor* src, const Fn& fn) { |
134 while (n > 0) { | 143 while (n > 0) { |
135 if (n >= 8) { | 144 if (n >= 8) { |
136 Sk4px dst0 = fn(Load4(dst+0), Load4(src+0)), | 145 Sk4px dst0 = fn(Load4(dst+0), Load4(src+0)), |
137 dst4 = fn(Load4(dst+4), Load4(src+4)); | 146 dst4 = fn(Load4(dst+4), Load4(src+4)); |
138 dst0.store4(dst+0); | 147 dst0.store4(dst+0); |
139 dst4.store4(dst+4); | 148 dst4.store4(dst+4); |
140 dst += 8; src += 8; n -= 8; | 149 dst += 8; src += 8; n -= 8; |
141 continue; // Keep our stride at 8 pixels as long as possible. | 150 continue; // Keep our stride at 8 pixels as long as possible. |
142 } | 151 } |
143 SkASSERT(n <= 7); | 152 SkASSERT(n <= 7); |
144 if (n >= 4) { | 153 if (n >= 4) { |
145 fn(Load4(dst), Load4(src)).store4(dst); | 154 fn(Load4(dst), Load4(src)).store4(dst); |
146 dst += 4; src += 4; n -= 4; | 155 dst += 4; src += 4; n -= 4; |
147 } | 156 } |
148 if (n >= 2) { | 157 if (n >= 2) { |
149 fn(Load2(dst), Load2(src)).store2(dst); | 158 fn(Load2(dst), Load2(src)).store2(dst); |
150 dst += 2; src += 2; n -= 2; | 159 dst += 2; src += 2; n -= 2; |
151 } | 160 } |
152 if (n >= 1) { | 161 if (n >= 1) { |
153 fn(Load1(dst), Load1(src)).store1(dst); | 162 fn(Load1(dst), Load1(src)).store1(dst); |
154 } | 163 } |
155 break; | 164 break; |
156 } | 165 } |
157 } | 166 } |
158 | 167 |
159 // As above, but with dst4' = fn(dst4, src4, alpha4). | 168 // As above, but with dst4' = fn(dst4, src4, alpha4). |
160 template <typename Fn> | 169 template <typename Fn, typename Dst> |
161 static void MapDstSrcAlpha(int n, SkPMColor* dst, const SkPMColor* src, cons
t SkAlpha* a, | 170 static void MapDstSrcAlpha(int n, Dst* dst, const SkPMColor* src, const SkAl
pha* a, |
162 const Fn& fn) { | 171 const Fn& fn) { |
163 while (n > 0) { | 172 while (n > 0) { |
164 if (n >= 8) { | 173 if (n >= 8) { |
165 Sk4px dst0 = fn(Load4(dst+0), Load4(src+0), Load4Alphas(a+0)), | 174 Sk4px dst0 = fn(Load4(dst+0), Load4(src+0), Load4Alphas(a+0)), |
166 dst4 = fn(Load4(dst+4), Load4(src+4), Load4Alphas(a+4)); | 175 dst4 = fn(Load4(dst+4), Load4(src+4), Load4Alphas(a+4)); |
167 dst0.store4(dst+0); | 176 dst0.store4(dst+0); |
168 dst4.store4(dst+4); | 177 dst4.store4(dst+4); |
169 dst += 8; src += 8; a += 8; n -= 8; | 178 dst += 8; src += 8; a += 8; n -= 8; |
170 continue; // Keep our stride at 8 pixels as long as possible. | 179 continue; // Keep our stride at 8 pixels as long as possible. |
171 } | 180 } |
(...skipping 25 matching lines...) Expand all Loading... |
197 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2 | 206 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2 |
198 #include "../opts/Sk4px_SSE2.h" | 207 #include "../opts/Sk4px_SSE2.h" |
199 #elif defined(SK_ARM_HAS_NEON) | 208 #elif defined(SK_ARM_HAS_NEON) |
200 #include "../opts/Sk4px_NEON.h" | 209 #include "../opts/Sk4px_NEON.h" |
201 #else | 210 #else |
202 #include "../opts/Sk4px_none.h" | 211 #include "../opts/Sk4px_none.h" |
203 #endif | 212 #endif |
204 #endif | 213 #endif |
205 | 214 |
206 #endif//Sk4px_DEFINED | 215 #endif//Sk4px_DEFINED |
OLD | NEW |