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 SkColorPriv_DEFINED 9#define SkColorPriv_DEFINED 10 11// turn this own for extra debug checking when blending onto 565 12#ifdef SK_DEBUG 13 #define CHECK_FOR_565_OVERFLOW 14#endif 15 16#include "SkColor.h" 17#include "SkMath.h" 18 19////////////////////////////////////////////////////////////////////////////// 20 21#define SkASSERT_IS_BYTE(x) SkASSERT(0 == ((x) & ~0xFF)) 22 23/* 24 * Skia's 32bit backend only supports 1 sizzle order at a time (compile-time). 25 * This is specified by 4 defines SK_A32_SHIFT, SK_R32_SHIFT, ... for G and B. 26 * 27 * For easier compatibility with Skia's GPU backend, we further restrict these 28 * to either (in memory-byte-order) RGBA or BGRA. Note that this "order" does 29 * not directly correspond to the same shift-order, since we have to take endianess 30 * into account. 31 * 32 * Here we enforce this constraint. 33 */ 34 35#ifdef SK_CPU_BENDIAN 36 #define SK_RGBA_R32_SHIFT 24 37 #define SK_RGBA_G32_SHIFT 16 38 #define SK_RGBA_B32_SHIFT 8 39 #define SK_RGBA_A32_SHIFT 0 40 41 #define SK_BGRA_B32_SHIFT 24 42 #define SK_BGRA_G32_SHIFT 16 43 #define SK_BGRA_R32_SHIFT 8 44 #define SK_BGRA_A32_SHIFT 0 45#else 46 #define SK_RGBA_R32_SHIFT 0 47 #define SK_RGBA_G32_SHIFT 8 48 #define SK_RGBA_B32_SHIFT 16 49 #define SK_RGBA_A32_SHIFT 24 50 51 #define SK_BGRA_B32_SHIFT 0 52 #define SK_BGRA_G32_SHIFT 8 53 #define SK_BGRA_R32_SHIFT 16 54 #define SK_BGRA_A32_SHIFT 24 55#endif 56 57#if defined(SK_PMCOLOR_IS_RGBA) && defined(SK_PMCOLOR_IS_BGRA) 58 #error "can't define PMCOLOR to be RGBA and BGRA" 59#endif 60 61#define LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA \ 62 (SK_A32_SHIFT == SK_RGBA_A32_SHIFT && \ 63 SK_R32_SHIFT == SK_RGBA_R32_SHIFT && \ 64 SK_G32_SHIFT == SK_RGBA_G32_SHIFT && \ 65 SK_B32_SHIFT == SK_RGBA_B32_SHIFT) 66 67#define LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA \ 68 (SK_A32_SHIFT == SK_BGRA_A32_SHIFT && \ 69 SK_R32_SHIFT == SK_BGRA_R32_SHIFT && \ 70 SK_G32_SHIFT == SK_BGRA_G32_SHIFT && \ 71 SK_B32_SHIFT == SK_BGRA_B32_SHIFT) 72 73 74#if defined(SK_PMCOLOR_IS_RGBA) && !LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA 75 #error "SK_PMCOLOR_IS_RGBA does not match SK_*32_SHIFT values" 76#endif 77 78#if defined(SK_PMCOLOR_IS_BGRA) && !LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA 79 #error "SK_PMCOLOR_IS_BGRA does not match SK_*32_SHIFT values" 80#endif 81 82#if !defined(SK_PMCOLOR_IS_RGBA) && !defined(SK_PMCOLOR_IS_RGBA) 83 // deduce which to define from the _SHIFT defines 84 85 #if LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA 86 #define SK_PMCOLOR_IS_RGBA 87 #elif LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA 88 #define SK_PMCOLOR_IS_BGRA 89 #else 90 #error "need 32bit packing to be either RGBA or BGRA" 91 #endif 92#endif 93 94// hide these now that we're done 95#undef LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA 96#undef LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA 97 98////////////////////////////////////////////////////////////////////////////// 99 100// Reverse the bytes coorsponding to RED and BLUE in a packed pixels. Note the 101// pair of them are in the same 2 slots in both RGBA and BGRA, thus there is 102// no need to pass in the colortype to this function. 103static inline uint32_t SkSwizzle_RB(uint32_t c) { 104 static const uint32_t kRBMask = (0xFF << SK_R32_SHIFT) | (0xFF << SK_B32_SHIFT); 105 106 unsigned c0 = (c >> SK_R32_SHIFT) & 0xFF; 107 unsigned c1 = (c >> SK_B32_SHIFT) & 0xFF; 108 return (c & ~kRBMask) | (c0 << SK_B32_SHIFT) | (c1 << SK_R32_SHIFT); 109} 110 111static inline uint32_t SkPackARGB_as_RGBA(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { 112 SkASSERT_IS_BYTE(a); 113 SkASSERT_IS_BYTE(r); 114 SkASSERT_IS_BYTE(g); 115 SkASSERT_IS_BYTE(b); 116 return (a << SK_RGBA_A32_SHIFT) | (r << SK_RGBA_R32_SHIFT) | 117 (g << SK_RGBA_G32_SHIFT) | (b << SK_RGBA_B32_SHIFT); 118} 119 120static inline uint32_t SkPackARGB_as_BGRA(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { 121 SkASSERT_IS_BYTE(a); 122 SkASSERT_IS_BYTE(r); 123 SkASSERT_IS_BYTE(g); 124 SkASSERT_IS_BYTE(b); 125 return (a << SK_BGRA_A32_SHIFT) | (r << SK_BGRA_R32_SHIFT) | 126 (g << SK_BGRA_G32_SHIFT) | (b << SK_BGRA_B32_SHIFT); 127} 128 129static inline SkPMColor SkSwizzle_RGBA_to_PMColor(uint32_t c) { 130#ifdef SK_PMCOLOR_IS_RGBA 131 return c; 132#else 133 return SkSwizzle_RB(c); 134#endif 135} 136 137static inline SkPMColor SkSwizzle_BGRA_to_PMColor(uint32_t c) { 138#ifdef SK_PMCOLOR_IS_BGRA 139 return c; 140#else 141 return SkSwizzle_RB(c); 142#endif 143} 144 145////////////////////////////////////////////////////////////////////////////// 146 147///@{ 148/** See ITU-R Recommendation BT.709 at http://www.itu.int/rec/R-REC-BT.709/ .*/ 149#define SK_ITU_BT709_LUM_COEFF_R (0.2126f) 150#define SK_ITU_BT709_LUM_COEFF_G (0.7152f) 151#define SK_ITU_BT709_LUM_COEFF_B (0.0722f) 152///@} 153 154///@{ 155/** A float value which specifies this channel's contribution to luminance. */ 156#define SK_LUM_COEFF_R SK_ITU_BT709_LUM_COEFF_R 157#define SK_LUM_COEFF_G SK_ITU_BT709_LUM_COEFF_G 158#define SK_LUM_COEFF_B SK_ITU_BT709_LUM_COEFF_B 159///@} 160 161/** Computes the luminance from the given r, g, and b in accordance with 162 SK_LUM_COEFF_X. For correct results, r, g, and b should be in linear space. 163*/ 164static inline U8CPU SkComputeLuminance(U8CPU r, U8CPU g, U8CPU b) { 165 //The following is 166 //r * SK_LUM_COEFF_R + g * SK_LUM_COEFF_G + b * SK_LUM_COEFF_B 167 //with SK_LUM_COEFF_X in 1.8 fixed point (rounding adjusted to sum to 256). 168 return (r * 54 + g * 183 + b * 19) >> 8; 169} 170 171/** Turn 0..255 into 0..256 by adding 1 at the half-way point. Used to turn a 172 byte into a scale value, so that we can say scale * value >> 8 instead of 173 alpha * value / 255. 174 175 In debugging, asserts that alpha is 0..255 176*/ 177static inline unsigned SkAlpha255To256(U8CPU alpha) { 178 SkASSERT(SkToU8(alpha) == alpha); 179 // this one assues that blending on top of an opaque dst keeps it that way 180 // even though it is less accurate than a+(a>>7) for non-opaque dsts 181 return alpha + 1; 182} 183 184/** 185 * Turn a 0..255 value into a 0..256 value, rounding up if the value is >= 0x80. 186 * This is slightly more accurate than SkAlpha255To256. 187 */ 188static inline unsigned Sk255To256(U8CPU value) { 189 SkASSERT(SkToU8(value) == value); 190 return value + (value >> 7); 191} 192 193/** Multiplify value by 0..256, and shift the result down 8 194 (i.e. return (value * alpha256) >> 8) 195 */ 196#define SkAlphaMul(value, alpha256) (SkMulS16(value, alpha256) >> 8) 197 198// The caller may want negative values, so keep all params signed (int) 199// so we don't accidentally slip into unsigned math and lose the sign 200// extension when we shift (in SkAlphaMul) 201static inline int SkAlphaBlend(int src, int dst, int scale256) { 202 SkASSERT((unsigned)scale256 <= 256); 203 return dst + SkAlphaMul(src - dst, scale256); 204} 205 206/** 207 * Returns (src * alpha + dst * (255 - alpha)) / 255 208 * 209 * This is more accurate than SkAlphaBlend, but slightly slower 210 */ 211static inline int SkAlphaBlend255(S16CPU src, S16CPU dst, U8CPU alpha) { 212 SkASSERT((int16_t)src == src); 213 SkASSERT((int16_t)dst == dst); 214 SkASSERT((uint8_t)alpha == alpha); 215 216 int prod = SkMulS16(src - dst, alpha) + 128; 217 prod = (prod + (prod >> 8)) >> 8; 218 return dst + prod; 219} 220 221#define SK_R16_BITS 5 222#define SK_G16_BITS 6 223#define SK_B16_BITS 5 224 225#define SK_R16_SHIFT (SK_B16_BITS + SK_G16_BITS) 226#define SK_G16_SHIFT (SK_B16_BITS) 227#define SK_B16_SHIFT 0 228 229#define SK_R16_MASK ((1 << SK_R16_BITS) - 1) 230#define SK_G16_MASK ((1 << SK_G16_BITS) - 1) 231#define SK_B16_MASK ((1 << SK_B16_BITS) - 1) 232 233#define SkGetPackedR16(color) (((unsigned)(color) >> SK_R16_SHIFT) & SK_R16_MASK) 234#define SkGetPackedG16(color) (((unsigned)(color) >> SK_G16_SHIFT) & SK_G16_MASK) 235#define SkGetPackedB16(color) (((unsigned)(color) >> SK_B16_SHIFT) & SK_B16_MASK) 236 237#define SkR16Assert(r) SkASSERT((unsigned)(r) <= SK_R16_MASK) 238#define SkG16Assert(g) SkASSERT((unsigned)(g) <= SK_G16_MASK) 239#define SkB16Assert(b) SkASSERT((unsigned)(b) <= SK_B16_MASK) 240 241static inline uint16_t SkPackRGB16(unsigned r, unsigned g, unsigned b) { 242 SkASSERT(r <= SK_R16_MASK); 243 SkASSERT(g <= SK_G16_MASK); 244 SkASSERT(b <= SK_B16_MASK); 245 246 return SkToU16((r << SK_R16_SHIFT) | (g << SK_G16_SHIFT) | (b << SK_B16_SHIFT)); 247} 248 249#define SK_R16_MASK_IN_PLACE (SK_R16_MASK << SK_R16_SHIFT) 250#define SK_G16_MASK_IN_PLACE (SK_G16_MASK << SK_G16_SHIFT) 251#define SK_B16_MASK_IN_PLACE (SK_B16_MASK << SK_B16_SHIFT) 252 253/** Expand the 16bit color into a 32bit value that can be scaled all at once 254 by a value up to 32. Used in conjunction with SkCompact_rgb_16. 255*/ 256static inline uint32_t SkExpand_rgb_16(U16CPU c) { 257 SkASSERT(c == (uint16_t)c); 258 259 return ((c & SK_G16_MASK_IN_PLACE) << 16) | (c & ~SK_G16_MASK_IN_PLACE); 260} 261 262/** Compress an expanded value (from SkExpand_rgb_16) back down to a 16bit 263 color value. The computation yields only 16bits of valid data, but we claim 264 to return 32bits, so that the compiler won't generate extra instructions to 265 "clean" the top 16bits. However, the top 16 can contain garbage, so it is 266 up to the caller to safely ignore them. 267*/ 268static inline U16CPU SkCompact_rgb_16(uint32_t c) { 269 return ((c >> 16) & SK_G16_MASK_IN_PLACE) | (c & ~SK_G16_MASK_IN_PLACE); 270} 271 272/** Scale the 16bit color value by the 0..256 scale parameter. 273 The computation yields only 16bits of valid data, but we claim 274 to return 32bits, so that the compiler won't generate extra instructions to 275 "clean" the top 16bits. 276*/ 277static inline U16CPU SkAlphaMulRGB16(U16CPU c, unsigned scale) { 278 return SkCompact_rgb_16(SkExpand_rgb_16(c) * (scale >> 3) >> 5); 279} 280 281// this helper explicitly returns a clean 16bit value (but slower) 282#define SkAlphaMulRGB16_ToU16(c, s) (uint16_t)SkAlphaMulRGB16(c, s) 283 284/** Blend src and dst 16bit colors by the 0..256 scale parameter. 285 The computation yields only 16bits of valid data, but we claim 286 to return 32bits, so that the compiler won't generate extra instructions to 287 "clean" the top 16bits. 288*/ 289static inline U16CPU SkBlendRGB16(U16CPU src, U16CPU dst, int srcScale) { 290 SkASSERT((unsigned)srcScale <= 256); 291 292 srcScale >>= 3; 293 294 uint32_t src32 = SkExpand_rgb_16(src); 295 uint32_t dst32 = SkExpand_rgb_16(dst); 296 return SkCompact_rgb_16(dst32 + ((src32 - dst32) * srcScale >> 5)); 297} 298 299static inline void SkBlendRGB16(const uint16_t src[], uint16_t dst[], 300 int srcScale, int count) { 301 SkASSERT(count > 0); 302 SkASSERT((unsigned)srcScale <= 256); 303 304 srcScale >>= 3; 305 306 do { 307 uint32_t src32 = SkExpand_rgb_16(*src++); 308 uint32_t dst32 = SkExpand_rgb_16(*dst); 309 *dst++ = SkCompact_rgb_16(dst32 + ((src32 - dst32) * srcScale >> 5)); 310 } while (--count > 0); 311} 312 313#ifdef SK_DEBUG 314 static inline U16CPU SkRGB16Add(U16CPU a, U16CPU b) { 315 SkASSERT(SkGetPackedR16(a) + SkGetPackedR16(b) <= SK_R16_MASK); 316 SkASSERT(SkGetPackedG16(a) + SkGetPackedG16(b) <= SK_G16_MASK); 317 SkASSERT(SkGetPackedB16(a) + SkGetPackedB16(b) <= SK_B16_MASK); 318 319 return a + b; 320 } 321#else 322 #define SkRGB16Add(a, b) ((a) + (b)) 323#endif 324 325/////////////////////////////////////////////////////////////////////////////// 326 327#define SK_A32_BITS 8 328#define SK_R32_BITS 8 329#define SK_G32_BITS 8 330#define SK_B32_BITS 8 331 332#define SK_A32_MASK ((1 << SK_A32_BITS) - 1) 333#define SK_R32_MASK ((1 << SK_R32_BITS) - 1) 334#define SK_G32_MASK ((1 << SK_G32_BITS) - 1) 335#define SK_B32_MASK ((1 << SK_B32_BITS) - 1) 336 337#define SkGetPackedA32(packed) ((uint32_t)((packed) << (24 - SK_A32_SHIFT)) >> 24) 338#define SkGetPackedR32(packed) ((uint32_t)((packed) << (24 - SK_R32_SHIFT)) >> 24) 339#define SkGetPackedG32(packed) ((uint32_t)((packed) << (24 - SK_G32_SHIFT)) >> 24) 340#define SkGetPackedB32(packed) ((uint32_t)((packed) << (24 - SK_B32_SHIFT)) >> 24) 341 342#define SkA32Assert(a) SkASSERT((unsigned)(a) <= SK_A32_MASK) 343#define SkR32Assert(r) SkASSERT((unsigned)(r) <= SK_R32_MASK) 344#define SkG32Assert(g) SkASSERT((unsigned)(g) <= SK_G32_MASK) 345#define SkB32Assert(b) SkASSERT((unsigned)(b) <= SK_B32_MASK) 346 347#ifdef SK_DEBUG 348 static inline void SkPMColorAssert(SkPMColor c) { 349 unsigned a = SkGetPackedA32(c); 350 unsigned r = SkGetPackedR32(c); 351 unsigned g = SkGetPackedG32(c); 352 unsigned b = SkGetPackedB32(c); 353 354 SkA32Assert(a); 355 SkASSERT(r <= a); 356 SkASSERT(g <= a); 357 SkASSERT(b <= a); 358 } 359#else 360 #define SkPMColorAssert(c) 361#endif 362 363/** 364 * Pack the components into a SkPMColor, checking (in the debug version) that 365 * the components are 0..255, and are already premultiplied (i.e. alpha >= color) 366 */ 367static inline SkPMColor SkPackARGB32(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { 368 SkA32Assert(a); 369 SkASSERT(r <= a); 370 SkASSERT(g <= a); 371 SkASSERT(b <= a); 372 373 return (a << SK_A32_SHIFT) | (r << SK_R32_SHIFT) | 374 (g << SK_G32_SHIFT) | (b << SK_B32_SHIFT); 375} 376 377static inline uint32_t SkPackPMColor_as_RGBA(SkPMColor c) { 378 return SkPackARGB_as_RGBA(SkGetPackedA32(c), SkGetPackedR32(c), 379 SkGetPackedG32(c), SkGetPackedB32(c)); 380} 381 382static inline uint32_t SkPackPMColor_as_BGRA(SkPMColor c) { 383 return SkPackARGB_as_BGRA(SkGetPackedA32(c), SkGetPackedR32(c), 384 SkGetPackedG32(c), SkGetPackedB32(c)); 385} 386 387/** 388 * Abstract 4-byte interpolation, implemented on top of SkPMColor 389 * utility functions. Third parameter controls blending of the first two: 390 * (src, dst, 0) returns dst 391 * (src, dst, 0xFF) returns src 392 * srcWeight is [0..256], unlike SkFourByteInterp which takes [0..255] 393 */ 394static inline SkPMColor SkFourByteInterp256(SkPMColor src, SkPMColor dst, 395 unsigned scale) { 396 unsigned a = SkAlphaBlend(SkGetPackedA32(src), SkGetPackedA32(dst), scale); 397 unsigned r = SkAlphaBlend(SkGetPackedR32(src), SkGetPackedR32(dst), scale); 398 unsigned g = SkAlphaBlend(SkGetPackedG32(src), SkGetPackedG32(dst), scale); 399 unsigned b = SkAlphaBlend(SkGetPackedB32(src), SkGetPackedB32(dst), scale); 400 401 return SkPackARGB32(a, r, g, b); 402} 403 404/** 405 * Abstract 4-byte interpolation, implemented on top of SkPMColor 406 * utility functions. Third parameter controls blending of the first two: 407 * (src, dst, 0) returns dst 408 * (src, dst, 0xFF) returns src 409 */ 410static inline SkPMColor SkFourByteInterp(SkPMColor src, SkPMColor dst, 411 U8CPU srcWeight) { 412 unsigned scale = SkAlpha255To256(srcWeight); 413 return SkFourByteInterp256(src, dst, scale); 414} 415 416/** 417 * 0xAARRGGBB -> 0x00AA00GG, 0x00RR00BB 418 */ 419static inline void SkSplay(uint32_t color, uint32_t* ag, uint32_t* rb) { 420 const uint32_t mask = 0x00FF00FF; 421 *ag = (color >> 8) & mask; 422 *rb = color & mask; 423} 424 425/** 426 * 0xAARRGGBB -> 0x00AA00GG00RR00BB 427 * (note, ARGB -> AGRB) 428 */ 429static inline uint64_t SkSplay(uint32_t color) { 430 const uint32_t mask = 0x00FF00FF; 431 uint64_t agrb = (color >> 8) & mask; // 0x0000000000AA00GG 432 agrb <<= 32; // 0x00AA00GG00000000 433 agrb |= color & mask; // 0x00AA00GG00RR00BB 434 return agrb; 435} 436 437/** 438 * 0xAAxxGGxx, 0xRRxxBBxx-> 0xAARRGGBB 439 */ 440static inline uint32_t SkUnsplay(uint32_t ag, uint32_t rb) { 441 const uint32_t mask = 0xFF00FF00; 442 return (ag & mask) | ((rb & mask) >> 8); 443} 444 445/** 446 * 0xAAxxGGxxRRxxBBxx -> 0xAARRGGBB 447 * (note, AGRB -> ARGB) 448 */ 449static inline uint32_t SkUnsplay(uint64_t agrb) { 450 const uint32_t mask = 0xFF00FF00; 451 return SkPMColor( 452 ((agrb & mask) >> 8) | // 0x00RR00BB 453 ((agrb >> 32) & mask)); // 0xAARRGGBB 454} 455 456static inline SkPMColor SkFastFourByteInterp256_32(SkPMColor src, SkPMColor dst, unsigned scale) { 457 SkASSERT(scale <= 256); 458 459 // Two 8-bit blends per two 32-bit registers, with space to make sure the math doesn't collide. 460 uint32_t src_ag, src_rb, dst_ag, dst_rb; 461 SkSplay(src, &src_ag, &src_rb); 462 SkSplay(dst, &dst_ag, &dst_rb); 463 464 const uint32_t ret_ag = src_ag * scale + (256 - scale) * dst_ag; 465 const uint32_t ret_rb = src_rb * scale + (256 - scale) * dst_rb; 466 467 return SkUnsplay(ret_ag, ret_rb); 468} 469 470static inline SkPMColor SkFastFourByteInterp256_64(SkPMColor src, SkPMColor dst, unsigned scale) { 471 SkASSERT(scale <= 256); 472 // Four 8-bit blends in one 64-bit register, with space to make sure the math doesn't collide. 473 return SkUnsplay(SkSplay(src) * scale + (256-scale) * SkSplay(dst)); 474} 475 476// TODO(mtklein): Replace slow versions with fast versions, using scale + (scale>>7) everywhere. 477 478/** 479 * Same as SkFourByteInterp256, but faster. 480 */ 481static inline SkPMColor SkFastFourByteInterp256(SkPMColor src, SkPMColor dst, unsigned scale) { 482 // On a 64-bit machine, _64 is about 10% faster than _32, but ~40% slower on a 32-bit machine. 483 if (sizeof(void*) == 4) { 484 return SkFastFourByteInterp256_32(src, dst, scale); 485 } else { 486 return SkFastFourByteInterp256_64(src, dst, scale); 487 } 488} 489 490/** 491 * Nearly the same as SkFourByteInterp, but faster and a touch more accurate, due to better 492 * srcWeight scaling to [0, 256]. 493 */ 494static inline SkPMColor SkFastFourByteInterp(SkPMColor src, 495 SkPMColor dst, 496 U8CPU srcWeight) { 497 SkASSERT(srcWeight <= 255); 498 // scale = srcWeight + (srcWeight >> 7) is more accurate than 499 // scale = srcWeight + 1, but 7% slower 500 return SkFastFourByteInterp256(src, dst, srcWeight + (srcWeight >> 7)); 501} 502 503/** 504 * Same as SkPackARGB32, but this version guarantees to not check that the 505 * values are premultiplied in the debug version. 506 */ 507static inline SkPMColor SkPackARGB32NoCheck(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { 508 return (a << SK_A32_SHIFT) | (r << SK_R32_SHIFT) | 509 (g << SK_G32_SHIFT) | (b << SK_B32_SHIFT); 510} 511 512static inline 513SkPMColor SkPremultiplyARGBInline(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { 514 SkA32Assert(a); 515 SkR32Assert(r); 516 SkG32Assert(g); 517 SkB32Assert(b); 518 519 if (a != 255) { 520 r = SkMulDiv255Round(r, a); 521 g = SkMulDiv255Round(g, a); 522 b = SkMulDiv255Round(b, a); 523 } 524 return SkPackARGB32(a, r, g, b); 525} 526 527// When Android is compiled optimizing for size, SkAlphaMulQ doesn't get 528// inlined; forcing inlining significantly improves performance. 529static SK_ALWAYS_INLINE uint32_t SkAlphaMulQ(uint32_t c, unsigned scale) { 530 uint32_t mask = 0xFF00FF; 531 532 uint32_t rb = ((c & mask) * scale) >> 8; 533 uint32_t ag = ((c >> 8) & mask) * scale; 534 return (rb & mask) | (ag & ~mask); 535} 536 537static inline SkPMColor SkPMSrcOver(SkPMColor src, SkPMColor dst) { 538 return src + SkAlphaMulQ(dst, SkAlpha255To256(255 - SkGetPackedA32(src))); 539} 540 541static inline SkPMColor SkBlendARGB32(SkPMColor src, SkPMColor dst, U8CPU aa) { 542 SkASSERT((unsigned)aa <= 255); 543 544 unsigned src_scale = SkAlpha255To256(aa); 545 unsigned dst_scale = SkAlpha255To256(255 - SkAlphaMul(SkGetPackedA32(src), src_scale)); 546 547 return SkAlphaMulQ(src, src_scale) + SkAlphaMulQ(dst, dst_scale); 548} 549 550//////////////////////////////////////////////////////////////////////////////////////////// 551// Convert a 32bit pixel to a 16bit pixel (no dither) 552 553#define SkR32ToR16_MACRO(r) ((unsigned)(r) >> (SK_R32_BITS - SK_R16_BITS)) 554#define SkG32ToG16_MACRO(g) ((unsigned)(g) >> (SK_G32_BITS - SK_G16_BITS)) 555#define SkB32ToB16_MACRO(b) ((unsigned)(b) >> (SK_B32_BITS - SK_B16_BITS)) 556 557#ifdef SK_DEBUG 558 static inline unsigned SkR32ToR16(unsigned r) { 559 SkR32Assert(r); 560 return SkR32ToR16_MACRO(r); 561 } 562 static inline unsigned SkG32ToG16(unsigned g) { 563 SkG32Assert(g); 564 return SkG32ToG16_MACRO(g); 565 } 566 static inline unsigned SkB32ToB16(unsigned b) { 567 SkB32Assert(b); 568 return SkB32ToB16_MACRO(b); 569 } 570#else 571 #define SkR32ToR16(r) SkR32ToR16_MACRO(r) 572 #define SkG32ToG16(g) SkG32ToG16_MACRO(g) 573 #define SkB32ToB16(b) SkB32ToB16_MACRO(b) 574#endif 575 576#define SkPacked32ToR16(c) (((unsigned)(c) >> (SK_R32_SHIFT + SK_R32_BITS - SK_R16_BITS)) & SK_R16_MASK) 577#define SkPacked32ToG16(c) (((unsigned)(c) >> (SK_G32_SHIFT + SK_G32_BITS - SK_G16_BITS)) & SK_G16_MASK) 578#define SkPacked32ToB16(c) (((unsigned)(c) >> (SK_B32_SHIFT + SK_B32_BITS - SK_B16_BITS)) & SK_B16_MASK) 579 580static inline U16CPU SkPixel32ToPixel16(SkPMColor c) { 581 unsigned r = ((c >> (SK_R32_SHIFT + (8 - SK_R16_BITS))) & SK_R16_MASK) << SK_R16_SHIFT; 582 unsigned g = ((c >> (SK_G32_SHIFT + (8 - SK_G16_BITS))) & SK_G16_MASK) << SK_G16_SHIFT; 583 unsigned b = ((c >> (SK_B32_SHIFT + (8 - SK_B16_BITS))) & SK_B16_MASK) << SK_B16_SHIFT; 584 return r | g | b; 585} 586 587static inline U16CPU SkPack888ToRGB16(U8CPU r, U8CPU g, U8CPU b) { 588 return (SkR32ToR16(r) << SK_R16_SHIFT) | 589 (SkG32ToG16(g) << SK_G16_SHIFT) | 590 (SkB32ToB16(b) << SK_B16_SHIFT); 591} 592 593#define SkPixel32ToPixel16_ToU16(src) SkToU16(SkPixel32ToPixel16(src)) 594 595///////////////////////////////////////////////////////////////////////////////////////// 596// Fast dither from 32->16 597 598#define SkShouldDitherXY(x, y) (((x) ^ (y)) & 1) 599 600static inline uint16_t SkDitherPack888ToRGB16(U8CPU r, U8CPU g, U8CPU b) { 601 r = ((r << 1) - ((r >> (8 - SK_R16_BITS) << (8 - SK_R16_BITS)) | (r >> SK_R16_BITS))) >> (8 - SK_R16_BITS); 602 g = ((g << 1) - ((g >> (8 - SK_G16_BITS) << (8 - SK_G16_BITS)) | (g >> SK_G16_BITS))) >> (8 - SK_G16_BITS); 603 b = ((b << 1) - ((b >> (8 - SK_B16_BITS) << (8 - SK_B16_BITS)) | (b >> SK_B16_BITS))) >> (8 - SK_B16_BITS); 604 605 return SkPackRGB16(r, g, b); 606} 607 608static inline uint16_t SkDitherPixel32ToPixel16(SkPMColor c) { 609 return SkDitherPack888ToRGB16(SkGetPackedR32(c), SkGetPackedG32(c), SkGetPackedB32(c)); 610} 611 612/* Return c in expanded_rgb_16 format, but also scaled up by 32 (5 bits) 613 It is now suitable for combining with a scaled expanded_rgb_16 color 614 as in SkSrcOver32To16(). 615 We must do this 565 high-bit replication, in order for the subsequent add 616 to saturate properly (and not overflow). If we take the 8 bits as is, it is 617 possible to overflow. 618*/ 619static inline uint32_t SkPMColorToExpanded16x5(SkPMColor c) { 620 unsigned sr = SkPacked32ToR16(c); 621 unsigned sg = SkPacked32ToG16(c); 622 unsigned sb = SkPacked32ToB16(c); 623 624 sr = (sr << 5) | sr; 625 sg = (sg << 5) | (sg >> 1); 626 sb = (sb << 5) | sb; 627 return (sr << 11) | (sg << 21) | (sb << 0); 628} 629 630/* SrcOver the 32bit src color with the 16bit dst, returning a 16bit value 631 (with dirt in the high 16bits, so caller beware). 632*/ 633static inline U16CPU SkSrcOver32To16(SkPMColor src, uint16_t dst) { 634 unsigned sr = SkGetPackedR32(src); 635 unsigned sg = SkGetPackedG32(src); 636 unsigned sb = SkGetPackedB32(src); 637 638 unsigned dr = SkGetPackedR16(dst); 639 unsigned dg = SkGetPackedG16(dst); 640 unsigned db = SkGetPackedB16(dst); 641 642 unsigned isa = 255 - SkGetPackedA32(src); 643 644 dr = (sr + SkMul16ShiftRound(dr, isa, SK_R16_BITS)) >> (8 - SK_R16_BITS); 645 dg = (sg + SkMul16ShiftRound(dg, isa, SK_G16_BITS)) >> (8 - SK_G16_BITS); 646 db = (sb + SkMul16ShiftRound(db, isa, SK_B16_BITS)) >> (8 - SK_B16_BITS); 647 648 return SkPackRGB16(dr, dg, db); 649} 650 651//////////////////////////////////////////////////////////////////////////////////////////// 652// Convert a 16bit pixel to a 32bit pixel 653 654static inline unsigned SkR16ToR32(unsigned r) { 655 return (r << (8 - SK_R16_BITS)) | (r >> (2 * SK_R16_BITS - 8)); 656} 657 658static inline unsigned SkG16ToG32(unsigned g) { 659 return (g << (8 - SK_G16_BITS)) | (g >> (2 * SK_G16_BITS - 8)); 660} 661 662static inline unsigned SkB16ToB32(unsigned b) { 663 return (b << (8 - SK_B16_BITS)) | (b >> (2 * SK_B16_BITS - 8)); 664} 665 666#define SkPacked16ToR32(c) SkR16ToR32(SkGetPackedR16(c)) 667#define SkPacked16ToG32(c) SkG16ToG32(SkGetPackedG16(c)) 668#define SkPacked16ToB32(c) SkB16ToB32(SkGetPackedB16(c)) 669 670static inline SkPMColor SkPixel16ToPixel32(U16CPU src) { 671 SkASSERT(src == SkToU16(src)); 672 673 unsigned r = SkPacked16ToR32(src); 674 unsigned g = SkPacked16ToG32(src); 675 unsigned b = SkPacked16ToB32(src); 676 677 SkASSERT((r >> (8 - SK_R16_BITS)) == SkGetPackedR16(src)); 678 SkASSERT((g >> (8 - SK_G16_BITS)) == SkGetPackedG16(src)); 679 SkASSERT((b >> (8 - SK_B16_BITS)) == SkGetPackedB16(src)); 680 681 return SkPackARGB32(0xFF, r, g, b); 682} 683 684// similar to SkPixel16ToPixel32, but returns SkColor instead of SkPMColor 685static inline SkColor SkPixel16ToColor(U16CPU src) { 686 SkASSERT(src == SkToU16(src)); 687 688 unsigned r = SkPacked16ToR32(src); 689 unsigned g = SkPacked16ToG32(src); 690 unsigned b = SkPacked16ToB32(src); 691 692 SkASSERT((r >> (8 - SK_R16_BITS)) == SkGetPackedR16(src)); 693 SkASSERT((g >> (8 - SK_G16_BITS)) == SkGetPackedG16(src)); 694 SkASSERT((b >> (8 - SK_B16_BITS)) == SkGetPackedB16(src)); 695 696 return SkColorSetRGB(r, g, b); 697} 698 699/////////////////////////////////////////////////////////////////////////////// 700 701typedef uint16_t SkPMColor16; 702 703// Put in OpenGL order (r g b a) 704#define SK_A4444_SHIFT 0 705#define SK_R4444_SHIFT 12 706#define SK_G4444_SHIFT 8 707#define SK_B4444_SHIFT 4 708 709#define SkA32To4444(a) ((unsigned)(a) >> 4) 710#define SkR32To4444(r) ((unsigned)(r) >> 4) 711#define SkG32To4444(g) ((unsigned)(g) >> 4) 712#define SkB32To4444(b) ((unsigned)(b) >> 4) 713 714static inline U8CPU SkReplicateNibble(unsigned nib) { 715 SkASSERT(nib <= 0xF); 716 return (nib << 4) | nib; 717} 718 719#define SkA4444ToA32(a) SkReplicateNibble(a) 720#define SkR4444ToR32(r) SkReplicateNibble(r) 721#define SkG4444ToG32(g) SkReplicateNibble(g) 722#define SkB4444ToB32(b) SkReplicateNibble(b) 723 724#define SkGetPackedA4444(c) (((unsigned)(c) >> SK_A4444_SHIFT) & 0xF) 725#define SkGetPackedR4444(c) (((unsigned)(c) >> SK_R4444_SHIFT) & 0xF) 726#define SkGetPackedG4444(c) (((unsigned)(c) >> SK_G4444_SHIFT) & 0xF) 727#define SkGetPackedB4444(c) (((unsigned)(c) >> SK_B4444_SHIFT) & 0xF) 728 729#define SkPacked4444ToA32(c) SkReplicateNibble(SkGetPackedA4444(c)) 730#define SkPacked4444ToR32(c) SkReplicateNibble(SkGetPackedR4444(c)) 731#define SkPacked4444ToG32(c) SkReplicateNibble(SkGetPackedG4444(c)) 732#define SkPacked4444ToB32(c) SkReplicateNibble(SkGetPackedB4444(c)) 733 734#ifdef SK_DEBUG 735static inline void SkPMColor16Assert(U16CPU c) { 736 unsigned a = SkGetPackedA4444(c); 737 unsigned r = SkGetPackedR4444(c); 738 unsigned g = SkGetPackedG4444(c); 739 unsigned b = SkGetPackedB4444(c); 740 741 SkASSERT(a <= 0xF); 742 SkASSERT(r <= a); 743 SkASSERT(g <= a); 744 SkASSERT(b <= a); 745} 746#else 747#define SkPMColor16Assert(c) 748#endif 749 750static inline unsigned SkAlpha15To16(unsigned a) { 751 SkASSERT(a <= 0xF); 752 return a + (a >> 3); 753} 754 755#ifdef SK_DEBUG 756 static inline int SkAlphaMul4(int value, int scale) { 757 SkASSERT((unsigned)scale <= 0x10); 758 return value * scale >> 4; 759 } 760#else 761 #define SkAlphaMul4(value, scale) ((value) * (scale) >> 4) 762#endif 763 764static inline unsigned SkR4444ToR565(unsigned r) { 765 SkASSERT(r <= 0xF); 766 return (r << (SK_R16_BITS - 4)) | (r >> (8 - SK_R16_BITS)); 767} 768 769static inline unsigned SkG4444ToG565(unsigned g) { 770 SkASSERT(g <= 0xF); 771 return (g << (SK_G16_BITS - 4)) | (g >> (8 - SK_G16_BITS)); 772} 773 774static inline unsigned SkB4444ToB565(unsigned b) { 775 SkASSERT(b <= 0xF); 776 return (b << (SK_B16_BITS - 4)) | (b >> (8 - SK_B16_BITS)); 777} 778 779static inline SkPMColor16 SkPackARGB4444(unsigned a, unsigned r, 780 unsigned g, unsigned b) { 781 SkASSERT(a <= 0xF); 782 SkASSERT(r <= a); 783 SkASSERT(g <= a); 784 SkASSERT(b <= a); 785 786 return (SkPMColor16)((a << SK_A4444_SHIFT) | (r << SK_R4444_SHIFT) | 787 (g << SK_G4444_SHIFT) | (b << SK_B4444_SHIFT)); 788} 789 790static inline U16CPU SkAlphaMulQ4(U16CPU c, unsigned scale) { 791 SkASSERT(scale <= 16); 792 793 const unsigned mask = 0xF0F; //gMask_0F0F; 794 795#if 0 796 unsigned rb = ((c & mask) * scale) >> 4; 797 unsigned ag = ((c >> 4) & mask) * scale; 798 return (rb & mask) | (ag & ~mask); 799#else 800 c = (c & mask) | ((c & (mask << 4)) << 12); 801 c = c * scale >> 4; 802 return (c & mask) | ((c >> 12) & (mask << 4)); 803#endif 804} 805 806/** Expand the SkPMColor16 color into a 32bit value that can be scaled all at 807 once by a value up to 16. Used in conjunction with SkCompact_4444. 808*/ 809static inline uint32_t SkExpand_4444(U16CPU c) { 810 SkASSERT(c == (uint16_t)c); 811 812 const unsigned mask = 0xF0F; //gMask_0F0F; 813 return (c & mask) | ((c & ~mask) << 12); 814} 815 816/** Compress an expanded value (from SkExpand_4444) back down to a SkPMColor16. 817 NOTE: this explicitly does not clean the top 16 bits (which may be garbage). 818 It does this for speed, since if it is being written directly to 16bits of 819 memory, the top 16bits will be ignored. Casting the result to uint16_t here 820 would add 2 more instructions, slow us down. It is up to the caller to 821 perform the cast if needed. 822*/ 823static inline U16CPU SkCompact_4444(uint32_t c) { 824 const unsigned mask = 0xF0F; //gMask_0F0F; 825 return (c & mask) | ((c >> 12) & ~mask); 826} 827 828static inline uint16_t SkSrcOver4444To16(SkPMColor16 s, uint16_t d) { 829 unsigned sa = SkGetPackedA4444(s); 830 unsigned sr = SkR4444ToR565(SkGetPackedR4444(s)); 831 unsigned sg = SkG4444ToG565(SkGetPackedG4444(s)); 832 unsigned sb = SkB4444ToB565(SkGetPackedB4444(s)); 833 834 // To avoid overflow, we have to clear the low bit of the synthetic sg 835 // if the src alpha is <= 7. 836 // to see why, try blending 0x4444 on top of 565-white and watch green 837 // overflow (sum == 64) 838 sg &= ~(~(sa >> 3) & 1); 839 840 unsigned scale = SkAlpha15To16(15 - sa); 841 unsigned dr = SkAlphaMul4(SkGetPackedR16(d), scale); 842 unsigned dg = SkAlphaMul4(SkGetPackedG16(d), scale); 843 unsigned db = SkAlphaMul4(SkGetPackedB16(d), scale); 844 845#if 0 846 if (sg + dg > 63) { 847 SkDebugf("---- SkSrcOver4444To16 src=%x dst=%x scale=%d, sg=%d dg=%d\n", s, d, scale, sg, dg); 848 } 849#endif 850 return SkPackRGB16(sr + dr, sg + dg, sb + db); 851} 852 853static inline uint16_t SkBlend4444To16(SkPMColor16 src, uint16_t dst, int scale16) { 854 SkASSERT((unsigned)scale16 <= 16); 855 856 return SkSrcOver4444To16(SkAlphaMulQ4(src, scale16), dst); 857} 858 859static inline uint16_t SkBlend4444(SkPMColor16 src, SkPMColor16 dst, int scale16) { 860 SkASSERT((unsigned)scale16 <= 16); 861 862 uint32_t src32 = SkExpand_4444(src) * scale16; 863 // the scaled srcAlpha is the bottom byte 864#ifdef SK_DEBUG 865 { 866 unsigned srcA = SkGetPackedA4444(src) * scale16; 867 SkASSERT(srcA == (src32 & 0xFF)); 868 } 869#endif 870 unsigned dstScale = SkAlpha255To256(255 - (src32 & 0xFF)) >> 4; 871 uint32_t dst32 = SkExpand_4444(dst) * dstScale; 872 return SkCompact_4444((src32 + dst32) >> 4); 873} 874 875static inline SkPMColor SkPixel4444ToPixel32(U16CPU c) { 876 uint32_t d = (SkGetPackedA4444(c) << SK_A32_SHIFT) | 877 (SkGetPackedR4444(c) << SK_R32_SHIFT) | 878 (SkGetPackedG4444(c) << SK_G32_SHIFT) | 879 (SkGetPackedB4444(c) << SK_B32_SHIFT); 880 return d | (d << 4); 881} 882 883static inline SkPMColor16 SkPixel32ToPixel4444(SkPMColor c) { 884 return (((c >> (SK_A32_SHIFT + 4)) & 0xF) << SK_A4444_SHIFT) | 885 (((c >> (SK_R32_SHIFT + 4)) & 0xF) << SK_R4444_SHIFT) | 886 (((c >> (SK_G32_SHIFT + 4)) & 0xF) << SK_G4444_SHIFT) | 887 (((c >> (SK_B32_SHIFT + 4)) & 0xF) << SK_B4444_SHIFT); 888} 889 890// cheap 2x2 dither 891static inline SkPMColor16 SkDitherARGB32To4444(U8CPU a, U8CPU r, 892 U8CPU g, U8CPU b) { 893 // to ensure that we stay a legal premultiplied color, we take the max() 894 // of the truncated and dithered alpha values. If we didn't, cases like 895 // SkDitherARGB32To4444(0x31, 0x2E, ...) would generate SkPackARGB4444(2, 3, ...) 896 // which is not legal premultiplied, since a < color 897 unsigned dithered_a = ((a << 1) - ((a >> 4 << 4) | (a >> 4))) >> 4; 898 a = SkMax32(a >> 4, dithered_a); 899 // these we just dither in place 900 r = ((r << 1) - ((r >> 4 << 4) | (r >> 4))) >> 4; 901 g = ((g << 1) - ((g >> 4 << 4) | (g >> 4))) >> 4; 902 b = ((b << 1) - ((b >> 4 << 4) | (b >> 4))) >> 4; 903 904 return SkPackARGB4444(a, r, g, b); 905} 906 907static inline SkPMColor16 SkDitherPixel32To4444(SkPMColor c) { 908 return SkDitherARGB32To4444(SkGetPackedA32(c), SkGetPackedR32(c), 909 SkGetPackedG32(c), SkGetPackedB32(c)); 910} 911 912/* Assumes 16bit is in standard RGBA order. 913 Transforms a normal ARGB_8888 into the same byte order as 914 expanded ARGB_4444, but keeps each component 8bits 915*/ 916static inline uint32_t SkExpand_8888(SkPMColor c) { 917 return (((c >> SK_R32_SHIFT) & 0xFF) << 24) | 918 (((c >> SK_G32_SHIFT) & 0xFF) << 8) | 919 (((c >> SK_B32_SHIFT) & 0xFF) << 16) | 920 (((c >> SK_A32_SHIFT) & 0xFF) << 0); 921} 922 923/* Undo the operation of SkExpand_8888, turning the argument back into 924 a SkPMColor. 925*/ 926static inline SkPMColor SkCompact_8888(uint32_t c) { 927 return (((c >> 24) & 0xFF) << SK_R32_SHIFT) | 928 (((c >> 8) & 0xFF) << SK_G32_SHIFT) | 929 (((c >> 16) & 0xFF) << SK_B32_SHIFT) | 930 (((c >> 0) & 0xFF) << SK_A32_SHIFT); 931} 932 933/* Like SkExpand_8888, this transforms a pmcolor into the expanded 4444 format, 934 but this routine just keeps the high 4bits of each component in the low 935 4bits of the result (just like a newly expanded PMColor16). 936*/ 937static inline uint32_t SkExpand32_4444(SkPMColor c) { 938 return (((c >> (SK_R32_SHIFT + 4)) & 0xF) << 24) | 939 (((c >> (SK_G32_SHIFT + 4)) & 0xF) << 8) | 940 (((c >> (SK_B32_SHIFT + 4)) & 0xF) << 16) | 941 (((c >> (SK_A32_SHIFT + 4)) & 0xF) << 0); 942} 943 944// takes two values and alternamtes them as part of a memset16 945// used for cheap 2x2 dithering when the colors are opaque 946void sk_dither_memset16(uint16_t dst[], uint16_t value, uint16_t other, int n); 947 948/////////////////////////////////////////////////////////////////////////////// 949 950static inline int SkUpscale31To32(int value) { 951 SkASSERT((unsigned)value <= 31); 952 return value + (value >> 4); 953} 954 955static inline int SkBlend32(int src, int dst, int scale) { 956 SkASSERT((unsigned)src <= 0xFF); 957 SkASSERT((unsigned)dst <= 0xFF); 958 SkASSERT((unsigned)scale <= 32); 959 return dst + ((src - dst) * scale >> 5); 960} 961 962static inline SkPMColor SkBlendLCD16(int srcA, int srcR, int srcG, int srcB, 963 SkPMColor dst, uint16_t mask) { 964 if (mask == 0) { 965 return dst; 966 } 967 968 /* We want all of these in 5bits, hence the shifts in case one of them 969 * (green) is 6bits. 970 */ 971 int maskR = SkGetPackedR16(mask) >> (SK_R16_BITS - 5); 972 int maskG = SkGetPackedG16(mask) >> (SK_G16_BITS - 5); 973 int maskB = SkGetPackedB16(mask) >> (SK_B16_BITS - 5); 974 975 // Now upscale them to 0..32, so we can use blend32 976 maskR = SkUpscale31To32(maskR); 977 maskG = SkUpscale31To32(maskG); 978 maskB = SkUpscale31To32(maskB); 979 980 // srcA has been upscaled to 256 before passed into this function 981 maskR = maskR * srcA >> 8; 982 maskG = maskG * srcA >> 8; 983 maskB = maskB * srcA >> 8; 984 985 int dstR = SkGetPackedR32(dst); 986 int dstG = SkGetPackedG32(dst); 987 int dstB = SkGetPackedB32(dst); 988 989 // LCD blitting is only supported if the dst is known/required 990 // to be opaque 991 return SkPackARGB32(0xFF, 992 SkBlend32(srcR, dstR, maskR), 993 SkBlend32(srcG, dstG, maskG), 994 SkBlend32(srcB, dstB, maskB)); 995} 996 997static inline SkPMColor SkBlendLCD16Opaque(int srcR, int srcG, int srcB, 998 SkPMColor dst, uint16_t mask, 999 SkPMColor opaqueDst) { 1000 if (mask == 0) { 1001 return dst; 1002 } 1003 1004 if (0xFFFF == mask) { 1005 return opaqueDst; 1006 } 1007 1008 /* We want all of these in 5bits, hence the shifts in case one of them 1009 * (green) is 6bits. 1010 */ 1011 int maskR = SkGetPackedR16(mask) >> (SK_R16_BITS - 5); 1012 int maskG = SkGetPackedG16(mask) >> (SK_G16_BITS - 5); 1013 int maskB = SkGetPackedB16(mask) >> (SK_B16_BITS - 5); 1014 1015 // Now upscale them to 0..32, so we can use blend32 1016 maskR = SkUpscale31To32(maskR); 1017 maskG = SkUpscale31To32(maskG); 1018 maskB = SkUpscale31To32(maskB); 1019 1020 int dstR = SkGetPackedR32(dst); 1021 int dstG = SkGetPackedG32(dst); 1022 int dstB = SkGetPackedB32(dst); 1023 1024 // LCD blitting is only supported if the dst is known/required 1025 // to be opaque 1026 return SkPackARGB32(0xFF, 1027 SkBlend32(srcR, dstR, maskR), 1028 SkBlend32(srcG, dstG, maskG), 1029 SkBlend32(srcB, dstB, maskB)); 1030} 1031 1032static inline void SkBlitLCD16Row(SkPMColor dst[], const uint16_t mask[], 1033 SkColor src, int width, SkPMColor) { 1034 int srcA = SkColorGetA(src); 1035 int srcR = SkColorGetR(src); 1036 int srcG = SkColorGetG(src); 1037 int srcB = SkColorGetB(src); 1038 1039 srcA = SkAlpha255To256(srcA); 1040 1041 for (int i = 0; i < width; i++) { 1042 dst[i] = SkBlendLCD16(srcA, srcR, srcG, srcB, dst[i], mask[i]); 1043 } 1044} 1045 1046static inline void SkBlitLCD16OpaqueRow(SkPMColor dst[], const uint16_t mask[], 1047 SkColor src, int width, 1048 SkPMColor opaqueDst) { 1049 int srcR = SkColorGetR(src); 1050 int srcG = SkColorGetG(src); 1051 int srcB = SkColorGetB(src); 1052 1053 for (int i = 0; i < width; i++) { 1054 dst[i] = SkBlendLCD16Opaque(srcR, srcG, srcB, dst[i], mask[i], 1055 opaqueDst); 1056 } 1057} 1058 1059#endif 1060