1 | /* |
2 | * Copyright 2006 The Android Open Source Project |
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 | #ifndef SkColorData_DEFINED |
9 | #define SkColorData_DEFINED |
10 | |
11 | #include "include/core/SkColor.h" |
12 | #include "include/core/SkColorPriv.h" |
13 | #include "include/private/SkNx.h" |
14 | #include "include/private/SkTo.h" |
15 | |
16 | //////////////////////////////////////////////////////////////////////////////////////////// |
17 | // Convert a 16bit pixel to a 32bit pixel |
18 | |
19 | #define SK_R16_BITS 5 |
20 | #define SK_G16_BITS 6 |
21 | #define SK_B16_BITS 5 |
22 | |
23 | #define SK_R16_SHIFT (SK_B16_BITS + SK_G16_BITS) |
24 | #define SK_G16_SHIFT (SK_B16_BITS) |
25 | #define SK_B16_SHIFT 0 |
26 | |
27 | #define SK_R16_MASK ((1 << SK_R16_BITS) - 1) |
28 | #define SK_G16_MASK ((1 << SK_G16_BITS) - 1) |
29 | #define SK_B16_MASK ((1 << SK_B16_BITS) - 1) |
30 | |
31 | #define SkGetPackedR16(color) (((unsigned)(color) >> SK_R16_SHIFT) & SK_R16_MASK) |
32 | #define SkGetPackedG16(color) (((unsigned)(color) >> SK_G16_SHIFT) & SK_G16_MASK) |
33 | #define SkGetPackedB16(color) (((unsigned)(color) >> SK_B16_SHIFT) & SK_B16_MASK) |
34 | |
35 | static inline unsigned SkR16ToR32(unsigned r) { |
36 | return (r << (8 - SK_R16_BITS)) | (r >> (2 * SK_R16_BITS - 8)); |
37 | } |
38 | |
39 | static inline unsigned SkG16ToG32(unsigned g) { |
40 | return (g << (8 - SK_G16_BITS)) | (g >> (2 * SK_G16_BITS - 8)); |
41 | } |
42 | |
43 | static inline unsigned SkB16ToB32(unsigned b) { |
44 | return (b << (8 - SK_B16_BITS)) | (b >> (2 * SK_B16_BITS - 8)); |
45 | } |
46 | |
47 | #define SkPacked16ToR32(c) SkR16ToR32(SkGetPackedR16(c)) |
48 | #define SkPacked16ToG32(c) SkG16ToG32(SkGetPackedG16(c)) |
49 | #define SkPacked16ToB32(c) SkB16ToB32(SkGetPackedB16(c)) |
50 | |
51 | ////////////////////////////////////////////////////////////////////////////// |
52 | |
53 | #define SkASSERT_IS_BYTE(x) SkASSERT(0 == ((x) & ~0xFF)) |
54 | |
55 | // Reverse the bytes coorsponding to RED and BLUE in a packed pixels. Note the |
56 | // pair of them are in the same 2 slots in both RGBA and BGRA, thus there is |
57 | // no need to pass in the colortype to this function. |
58 | static inline uint32_t SkSwizzle_RB(uint32_t c) { |
59 | static const uint32_t kRBMask = (0xFF << SK_R32_SHIFT) | (0xFF << SK_B32_SHIFT); |
60 | |
61 | unsigned c0 = (c >> SK_R32_SHIFT) & 0xFF; |
62 | unsigned c1 = (c >> SK_B32_SHIFT) & 0xFF; |
63 | return (c & ~kRBMask) | (c0 << SK_B32_SHIFT) | (c1 << SK_R32_SHIFT); |
64 | } |
65 | |
66 | static inline uint32_t SkPackARGB_as_RGBA(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { |
67 | SkASSERT_IS_BYTE(a); |
68 | SkASSERT_IS_BYTE(r); |
69 | SkASSERT_IS_BYTE(g); |
70 | SkASSERT_IS_BYTE(b); |
71 | return (a << SK_RGBA_A32_SHIFT) | (r << SK_RGBA_R32_SHIFT) | |
72 | (g << SK_RGBA_G32_SHIFT) | (b << SK_RGBA_B32_SHIFT); |
73 | } |
74 | |
75 | static inline uint32_t SkPackARGB_as_BGRA(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { |
76 | SkASSERT_IS_BYTE(a); |
77 | SkASSERT_IS_BYTE(r); |
78 | SkASSERT_IS_BYTE(g); |
79 | SkASSERT_IS_BYTE(b); |
80 | return (a << SK_BGRA_A32_SHIFT) | (r << SK_BGRA_R32_SHIFT) | |
81 | (g << SK_BGRA_G32_SHIFT) | (b << SK_BGRA_B32_SHIFT); |
82 | } |
83 | |
84 | static inline SkPMColor SkSwizzle_RGBA_to_PMColor(uint32_t c) { |
85 | #ifdef SK_PMCOLOR_IS_RGBA |
86 | return c; |
87 | #else |
88 | return SkSwizzle_RB(c); |
89 | #endif |
90 | } |
91 | |
92 | static inline SkPMColor SkSwizzle_BGRA_to_PMColor(uint32_t c) { |
93 | #ifdef SK_PMCOLOR_IS_BGRA |
94 | return c; |
95 | #else |
96 | return SkSwizzle_RB(c); |
97 | #endif |
98 | } |
99 | |
100 | ////////////////////////////////////////////////////////////////////////////// |
101 | |
102 | ///@{ |
103 | /** See ITU-R Recommendation BT.709 at http://www.itu.int/rec/R-REC-BT.709/ .*/ |
104 | #define SK_ITU_BT709_LUM_COEFF_R (0.2126f) |
105 | #define SK_ITU_BT709_LUM_COEFF_G (0.7152f) |
106 | #define SK_ITU_BT709_LUM_COEFF_B (0.0722f) |
107 | ///@} |
108 | |
109 | ///@{ |
110 | /** A float value which specifies this channel's contribution to luminance. */ |
111 | #define SK_LUM_COEFF_R SK_ITU_BT709_LUM_COEFF_R |
112 | #define SK_LUM_COEFF_G SK_ITU_BT709_LUM_COEFF_G |
113 | #define SK_LUM_COEFF_B SK_ITU_BT709_LUM_COEFF_B |
114 | ///@} |
115 | |
116 | /** Computes the luminance from the given r, g, and b in accordance with |
117 | SK_LUM_COEFF_X. For correct results, r, g, and b should be in linear space. |
118 | */ |
119 | static inline U8CPU SkComputeLuminance(U8CPU r, U8CPU g, U8CPU b) { |
120 | //The following is |
121 | //r * SK_LUM_COEFF_R + g * SK_LUM_COEFF_G + b * SK_LUM_COEFF_B |
122 | //with SK_LUM_COEFF_X in 1.8 fixed point (rounding adjusted to sum to 256). |
123 | return (r * 54 + g * 183 + b * 19) >> 8; |
124 | } |
125 | |
126 | /** Calculates 256 - (value * alpha256) / 255 in range [0,256], |
127 | * for [0,255] value and [0,256] alpha256. |
128 | */ |
129 | static inline U16CPU SkAlphaMulInv256(U16CPU value, U16CPU alpha256) { |
130 | unsigned prod = 0xFFFF - value * alpha256; |
131 | return (prod + (prod >> 8)) >> 8; |
132 | } |
133 | |
134 | // The caller may want negative values, so keep all params signed (int) |
135 | // so we don't accidentally slip into unsigned math and lose the sign |
136 | // extension when we shift (in SkAlphaMul) |
137 | static inline int SkAlphaBlend(int src, int dst, int scale256) { |
138 | SkASSERT((unsigned)scale256 <= 256); |
139 | return dst + SkAlphaMul(src - dst, scale256); |
140 | } |
141 | |
142 | static inline uint16_t SkPackRGB16(unsigned r, unsigned g, unsigned b) { |
143 | SkASSERT(r <= SK_R16_MASK); |
144 | SkASSERT(g <= SK_G16_MASK); |
145 | SkASSERT(b <= SK_B16_MASK); |
146 | |
147 | return SkToU16((r << SK_R16_SHIFT) | (g << SK_G16_SHIFT) | (b << SK_B16_SHIFT)); |
148 | } |
149 | |
150 | #define SK_R16_MASK_IN_PLACE (SK_R16_MASK << SK_R16_SHIFT) |
151 | #define SK_G16_MASK_IN_PLACE (SK_G16_MASK << SK_G16_SHIFT) |
152 | #define SK_B16_MASK_IN_PLACE (SK_B16_MASK << SK_B16_SHIFT) |
153 | |
154 | /////////////////////////////////////////////////////////////////////////////// |
155 | |
156 | /** |
157 | * Abstract 4-byte interpolation, implemented on top of SkPMColor |
158 | * utility functions. Third parameter controls blending of the first two: |
159 | * (src, dst, 0) returns dst |
160 | * (src, dst, 0xFF) returns src |
161 | * srcWeight is [0..256], unlike SkFourByteInterp which takes [0..255] |
162 | */ |
163 | static inline SkPMColor SkFourByteInterp256(SkPMColor src, SkPMColor dst, |
164 | unsigned scale) { |
165 | unsigned a = SkAlphaBlend(SkGetPackedA32(src), SkGetPackedA32(dst), scale); |
166 | unsigned r = SkAlphaBlend(SkGetPackedR32(src), SkGetPackedR32(dst), scale); |
167 | unsigned g = SkAlphaBlend(SkGetPackedG32(src), SkGetPackedG32(dst), scale); |
168 | unsigned b = SkAlphaBlend(SkGetPackedB32(src), SkGetPackedB32(dst), scale); |
169 | |
170 | return SkPackARGB32(a, r, g, b); |
171 | } |
172 | |
173 | /** |
174 | * Abstract 4-byte interpolation, implemented on top of SkPMColor |
175 | * utility functions. Third parameter controls blending of the first two: |
176 | * (src, dst, 0) returns dst |
177 | * (src, dst, 0xFF) returns src |
178 | */ |
179 | static inline SkPMColor SkFourByteInterp(SkPMColor src, SkPMColor dst, |
180 | U8CPU srcWeight) { |
181 | unsigned scale = SkAlpha255To256(srcWeight); |
182 | return SkFourByteInterp256(src, dst, scale); |
183 | } |
184 | |
185 | /** |
186 | * 0xAARRGGBB -> 0x00AA00GG, 0x00RR00BB |
187 | */ |
188 | static inline void SkSplay(uint32_t color, uint32_t* ag, uint32_t* rb) { |
189 | const uint32_t mask = 0x00FF00FF; |
190 | *ag = (color >> 8) & mask; |
191 | *rb = color & mask; |
192 | } |
193 | |
194 | /** |
195 | * 0xAARRGGBB -> 0x00AA00GG00RR00BB |
196 | * (note, ARGB -> AGRB) |
197 | */ |
198 | static inline uint64_t SkSplay(uint32_t color) { |
199 | const uint32_t mask = 0x00FF00FF; |
200 | uint64_t agrb = (color >> 8) & mask; // 0x0000000000AA00GG |
201 | agrb <<= 32; // 0x00AA00GG00000000 |
202 | agrb |= color & mask; // 0x00AA00GG00RR00BB |
203 | return agrb; |
204 | } |
205 | |
206 | /** |
207 | * 0xAAxxGGxx, 0xRRxxBBxx-> 0xAARRGGBB |
208 | */ |
209 | static inline uint32_t SkUnsplay(uint32_t ag, uint32_t rb) { |
210 | const uint32_t mask = 0xFF00FF00; |
211 | return (ag & mask) | ((rb & mask) >> 8); |
212 | } |
213 | |
214 | /** |
215 | * 0xAAxxGGxxRRxxBBxx -> 0xAARRGGBB |
216 | * (note, AGRB -> ARGB) |
217 | */ |
218 | static inline uint32_t SkUnsplay(uint64_t agrb) { |
219 | const uint32_t mask = 0xFF00FF00; |
220 | return SkPMColor( |
221 | ((agrb & mask) >> 8) | // 0x00RR00BB |
222 | ((agrb >> 32) & mask)); // 0xAARRGGBB |
223 | } |
224 | |
225 | static inline SkPMColor SkFastFourByteInterp256_32(SkPMColor src, SkPMColor dst, unsigned scale) { |
226 | SkASSERT(scale <= 256); |
227 | |
228 | // Two 8-bit blends per two 32-bit registers, with space to make sure the math doesn't collide. |
229 | uint32_t src_ag, src_rb, dst_ag, dst_rb; |
230 | SkSplay(src, &src_ag, &src_rb); |
231 | SkSplay(dst, &dst_ag, &dst_rb); |
232 | |
233 | const uint32_t ret_ag = src_ag * scale + (256 - scale) * dst_ag; |
234 | const uint32_t ret_rb = src_rb * scale + (256 - scale) * dst_rb; |
235 | |
236 | return SkUnsplay(ret_ag, ret_rb); |
237 | } |
238 | |
239 | static inline SkPMColor SkFastFourByteInterp256_64(SkPMColor src, SkPMColor dst, unsigned scale) { |
240 | SkASSERT(scale <= 256); |
241 | // Four 8-bit blends in one 64-bit register, with space to make sure the math doesn't collide. |
242 | return SkUnsplay(SkSplay(src) * scale + (256-scale) * SkSplay(dst)); |
243 | } |
244 | |
245 | // TODO(mtklein): Replace slow versions with fast versions, using scale + (scale>>7) everywhere. |
246 | |
247 | /** |
248 | * Same as SkFourByteInterp256, but faster. |
249 | */ |
250 | static inline SkPMColor SkFastFourByteInterp256(SkPMColor src, SkPMColor dst, unsigned scale) { |
251 | // On a 64-bit machine, _64 is about 10% faster than _32, but ~40% slower on a 32-bit machine. |
252 | if (sizeof(void*) == 4) { |
253 | return SkFastFourByteInterp256_32(src, dst, scale); |
254 | } else { |
255 | return SkFastFourByteInterp256_64(src, dst, scale); |
256 | } |
257 | } |
258 | |
259 | /** |
260 | * Nearly the same as SkFourByteInterp, but faster and a touch more accurate, due to better |
261 | * srcWeight scaling to [0, 256]. |
262 | */ |
263 | static inline SkPMColor SkFastFourByteInterp(SkPMColor src, |
264 | SkPMColor dst, |
265 | U8CPU srcWeight) { |
266 | SkASSERT(srcWeight <= 255); |
267 | // scale = srcWeight + (srcWeight >> 7) is more accurate than |
268 | // scale = srcWeight + 1, but 7% slower |
269 | return SkFastFourByteInterp256(src, dst, srcWeight + (srcWeight >> 7)); |
270 | } |
271 | |
272 | /** |
273 | * Interpolates between colors src and dst using [0,256] scale. |
274 | */ |
275 | static inline SkPMColor SkPMLerp(SkPMColor src, SkPMColor dst, unsigned scale) { |
276 | return SkFastFourByteInterp256(src, dst, scale); |
277 | } |
278 | |
279 | static inline SkPMColor SkBlendARGB32(SkPMColor src, SkPMColor dst, U8CPU aa) { |
280 | SkASSERT((unsigned)aa <= 255); |
281 | |
282 | unsigned src_scale = SkAlpha255To256(aa); |
283 | unsigned dst_scale = SkAlphaMulInv256(SkGetPackedA32(src), src_scale); |
284 | |
285 | const uint32_t mask = 0xFF00FF; |
286 | |
287 | uint32_t src_rb = (src & mask) * src_scale; |
288 | uint32_t src_ag = ((src >> 8) & mask) * src_scale; |
289 | |
290 | uint32_t dst_rb = (dst & mask) * dst_scale; |
291 | uint32_t dst_ag = ((dst >> 8) & mask) * dst_scale; |
292 | |
293 | return (((src_rb + dst_rb) >> 8) & mask) | ((src_ag + dst_ag) & ~mask); |
294 | } |
295 | |
296 | //////////////////////////////////////////////////////////////////////////////////////////// |
297 | // Convert a 32bit pixel to a 16bit pixel (no dither) |
298 | |
299 | #define SkR32ToR16_MACRO(r) ((unsigned)(r) >> (SK_R32_BITS - SK_R16_BITS)) |
300 | #define SkG32ToG16_MACRO(g) ((unsigned)(g) >> (SK_G32_BITS - SK_G16_BITS)) |
301 | #define SkB32ToB16_MACRO(b) ((unsigned)(b) >> (SK_B32_BITS - SK_B16_BITS)) |
302 | |
303 | #ifdef SK_DEBUG |
304 | static inline unsigned SkR32ToR16(unsigned r) { |
305 | SkR32Assert(r); |
306 | return SkR32ToR16_MACRO(r); |
307 | } |
308 | static inline unsigned SkG32ToG16(unsigned g) { |
309 | SkG32Assert(g); |
310 | return SkG32ToG16_MACRO(g); |
311 | } |
312 | static inline unsigned SkB32ToB16(unsigned b) { |
313 | SkB32Assert(b); |
314 | return SkB32ToB16_MACRO(b); |
315 | } |
316 | #else |
317 | #define SkR32ToR16(r) SkR32ToR16_MACRO(r) |
318 | #define SkG32ToG16(g) SkG32ToG16_MACRO(g) |
319 | #define SkB32ToB16(b) SkB32ToB16_MACRO(b) |
320 | #endif |
321 | |
322 | static inline U16CPU SkPixel32ToPixel16(SkPMColor c) { |
323 | unsigned r = ((c >> (SK_R32_SHIFT + (8 - SK_R16_BITS))) & SK_R16_MASK) << SK_R16_SHIFT; |
324 | unsigned g = ((c >> (SK_G32_SHIFT + (8 - SK_G16_BITS))) & SK_G16_MASK) << SK_G16_SHIFT; |
325 | unsigned b = ((c >> (SK_B32_SHIFT + (8 - SK_B16_BITS))) & SK_B16_MASK) << SK_B16_SHIFT; |
326 | return r | g | b; |
327 | } |
328 | |
329 | static inline U16CPU SkPack888ToRGB16(U8CPU r, U8CPU g, U8CPU b) { |
330 | return (SkR32ToR16(r) << SK_R16_SHIFT) | |
331 | (SkG32ToG16(g) << SK_G16_SHIFT) | |
332 | (SkB32ToB16(b) << SK_B16_SHIFT); |
333 | } |
334 | |
335 | ///////////////////////////////////////////////////////////////////////////////////////// |
336 | |
337 | /* SrcOver the 32bit src color with the 16bit dst, returning a 16bit value |
338 | (with dirt in the high 16bits, so caller beware). |
339 | */ |
340 | static inline U16CPU SkSrcOver32To16(SkPMColor src, uint16_t dst) { |
341 | unsigned sr = SkGetPackedR32(src); |
342 | unsigned sg = SkGetPackedG32(src); |
343 | unsigned sb = SkGetPackedB32(src); |
344 | |
345 | unsigned dr = SkGetPackedR16(dst); |
346 | unsigned dg = SkGetPackedG16(dst); |
347 | unsigned db = SkGetPackedB16(dst); |
348 | |
349 | unsigned isa = 255 - SkGetPackedA32(src); |
350 | |
351 | dr = (sr + SkMul16ShiftRound(dr, isa, SK_R16_BITS)) >> (8 - SK_R16_BITS); |
352 | dg = (sg + SkMul16ShiftRound(dg, isa, SK_G16_BITS)) >> (8 - SK_G16_BITS); |
353 | db = (sb + SkMul16ShiftRound(db, isa, SK_B16_BITS)) >> (8 - SK_B16_BITS); |
354 | |
355 | return SkPackRGB16(dr, dg, db); |
356 | } |
357 | |
358 | static inline SkColor SkPixel16ToColor(U16CPU src) { |
359 | SkASSERT(src == SkToU16(src)); |
360 | |
361 | unsigned r = SkPacked16ToR32(src); |
362 | unsigned g = SkPacked16ToG32(src); |
363 | unsigned b = SkPacked16ToB32(src); |
364 | |
365 | SkASSERT((r >> (8 - SK_R16_BITS)) == SkGetPackedR16(src)); |
366 | SkASSERT((g >> (8 - SK_G16_BITS)) == SkGetPackedG16(src)); |
367 | SkASSERT((b >> (8 - SK_B16_BITS)) == SkGetPackedB16(src)); |
368 | |
369 | return SkColorSetRGB(r, g, b); |
370 | } |
371 | |
372 | /////////////////////////////////////////////////////////////////////////////// |
373 | |
374 | typedef uint16_t SkPMColor16; |
375 | |
376 | // Put in OpenGL order (r g b a) |
377 | #define SK_A4444_SHIFT 0 |
378 | #define SK_R4444_SHIFT 12 |
379 | #define SK_G4444_SHIFT 8 |
380 | #define SK_B4444_SHIFT 4 |
381 | |
382 | static inline U8CPU SkReplicateNibble(unsigned nib) { |
383 | SkASSERT(nib <= 0xF); |
384 | return (nib << 4) | nib; |
385 | } |
386 | |
387 | #define SkGetPackedA4444(c) (((unsigned)(c) >> SK_A4444_SHIFT) & 0xF) |
388 | #define SkGetPackedR4444(c) (((unsigned)(c) >> SK_R4444_SHIFT) & 0xF) |
389 | #define SkGetPackedG4444(c) (((unsigned)(c) >> SK_G4444_SHIFT) & 0xF) |
390 | #define SkGetPackedB4444(c) (((unsigned)(c) >> SK_B4444_SHIFT) & 0xF) |
391 | |
392 | #define SkPacked4444ToA32(c) SkReplicateNibble(SkGetPackedA4444(c)) |
393 | |
394 | static inline SkPMColor SkPixel4444ToPixel32(U16CPU c) { |
395 | uint32_t d = (SkGetPackedA4444(c) << SK_A32_SHIFT) | |
396 | (SkGetPackedR4444(c) << SK_R32_SHIFT) | |
397 | (SkGetPackedG4444(c) << SK_G32_SHIFT) | |
398 | (SkGetPackedB4444(c) << SK_B32_SHIFT); |
399 | return d | (d << 4); |
400 | } |
401 | |
402 | static inline Sk4f swizzle_rb(const Sk4f& x) { |
403 | return SkNx_shuffle<2, 1, 0, 3>(x); |
404 | } |
405 | |
406 | static inline Sk4f swizzle_rb_if_bgra(const Sk4f& x) { |
407 | #ifdef SK_PMCOLOR_IS_BGRA |
408 | return swizzle_rb(x); |
409 | #else |
410 | return x; |
411 | #endif |
412 | } |
413 | |
414 | static inline Sk4f Sk4f_fromL32(uint32_t px) { |
415 | return SkNx_cast<float>(Sk4b::Load(&px)) * (1 / 255.0f); |
416 | } |
417 | |
418 | static inline uint32_t Sk4f_toL32(const Sk4f& px) { |
419 | Sk4f v = px; |
420 | |
421 | #if !defined(SKNX_NO_SIMD) && SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2 |
422 | // SkNx_cast<uint8_t, int32_t>() pins, and we don't anticipate giant floats |
423 | #elif !defined(SKNX_NO_SIMD) && defined(SK_ARM_HAS_NEON) |
424 | // SkNx_cast<uint8_t, int32_t>() pins, and so does Sk4f_round(). |
425 | #else |
426 | // No guarantee of a pin. |
427 | v = Sk4f::Max(0, Sk4f::Min(v, 1)); |
428 | #endif |
429 | |
430 | uint32_t l32; |
431 | SkNx_cast<uint8_t>(Sk4f_round(v * 255.0f)).store(&l32); |
432 | return l32; |
433 | } |
434 | |
435 | using SkPMColor4f = SkRGBA4f<kPremul_SkAlphaType>; |
436 | |
437 | constexpr SkPMColor4f SK_PMColor4fTRANSPARENT = { 0, 0, 0, 0 }; |
438 | constexpr SkPMColor4f SK_PMColor4fWHITE = { 1, 1, 1, 1 }; |
439 | constexpr SkPMColor4f SK_PMColor4fILLEGAL = { SK_FloatNegativeInfinity, |
440 | SK_FloatNegativeInfinity, |
441 | SK_FloatNegativeInfinity, |
442 | SK_FloatNegativeInfinity }; |
443 | |
444 | #endif |
445 | |