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// 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 "SkColorPriv.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_BGRA_B32_SHIFT 24 37 #define SK_BGRA_G32_SHIFT 16 38 #define SK_BGRA_R32_SHIFT 8 39 #define SK_BGRA_A32_SHIFT 0 40#else 41 #define SK_BGRA_B32_SHIFT 0 42 #define SK_BGRA_G32_SHIFT 8 43 #define SK_BGRA_R32_SHIFT 16 44 #define SK_BGRA_A32_SHIFT 24 45#endif 46 47#if defined(SK_PMCOLOR_IS_RGBA) && defined(SK_PMCOLOR_IS_BGRA) 48 #error "can't define PMCOLOR to be RGBA and BGRA" 49#endif 50 51#define LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA \ 52 (SK_A32_SHIFT == SK_RGBA_A32_SHIFT && \ 53 SK_R32_SHIFT == SK_RGBA_R32_SHIFT && \ 54 SK_G32_SHIFT == SK_RGBA_G32_SHIFT && \ 55 SK_B32_SHIFT == SK_RGBA_B32_SHIFT) 56 57#define LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA \ 58 (SK_A32_SHIFT == SK_BGRA_A32_SHIFT && \ 59 SK_R32_SHIFT == SK_BGRA_R32_SHIFT && \ 60 SK_G32_SHIFT == SK_BGRA_G32_SHIFT && \ 61 SK_B32_SHIFT == SK_BGRA_B32_SHIFT) 62 63 64#define SK_A_INDEX (SK_A32_SHIFT/8) 65#define SK_R_INDEX (SK_R32_SHIFT/8) 66#define SK_G_INDEX (SK_G32_SHIFT/8) 67#define SK_B_INDEX (SK_B32_SHIFT/8) 68 69#if defined(SK_PMCOLOR_IS_RGBA) && !LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA 70 #error "SK_PMCOLOR_IS_RGBA does not match SK_*32_SHIFT values" 71#endif 72 73#if defined(SK_PMCOLOR_IS_BGRA) && !LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA 74 #error "SK_PMCOLOR_IS_BGRA does not match SK_*32_SHIFT values" 75#endif 76 77#if !defined(SK_PMCOLOR_IS_RGBA) && !defined(SK_PMCOLOR_IS_BGRA) 78 // deduce which to define from the _SHIFT defines 79 80 #if LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA 81 #define SK_PMCOLOR_IS_RGBA 82 #elif LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA 83 #define SK_PMCOLOR_IS_BGRA 84 #else 85 #error "need 32bit packing to be either RGBA or BGRA" 86 #endif 87#endif 88 89// hide these now that we're done 90#undef LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_RGBA 91#undef LOCAL_PMCOLOR_SHIFTS_EQUIVALENT_TO_BGRA 92 93////////////////////////////////////////////////////////////////////////////// 94 95// Reverse the bytes coorsponding to RED and BLUE in a packed pixels. Note the 96// pair of them are in the same 2 slots in both RGBA and BGRA, thus there is 97// no need to pass in the colortype to this function. 98static inline uint32_t SkSwizzle_RB(uint32_t c) { 99 static const uint32_t kRBMask = (0xFF << SK_R32_SHIFT) | (0xFF << SK_B32_SHIFT); 100 101 unsigned c0 = (c >> SK_R32_SHIFT) & 0xFF; 102 unsigned c1 = (c >> SK_B32_SHIFT) & 0xFF; 103 return (c & ~kRBMask) | (c0 << SK_B32_SHIFT) | (c1 << SK_R32_SHIFT); 104} 105 106static inline uint32_t SkPackARGB_as_RGBA(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { 107 SkASSERT_IS_BYTE(a); 108 SkASSERT_IS_BYTE(r); 109 SkASSERT_IS_BYTE(g); 110 SkASSERT_IS_BYTE(b); 111 return (a << SK_RGBA_A32_SHIFT) | (r << SK_RGBA_R32_SHIFT) | 112 (g << SK_RGBA_G32_SHIFT) | (b << SK_RGBA_B32_SHIFT); 113} 114 115static inline uint32_t SkPackARGB_as_BGRA(U8CPU a, U8CPU r, U8CPU g, U8CPU b) { 116 SkASSERT_IS_BYTE(a); 117 SkASSERT_IS_BYTE(r); 118 SkASSERT_IS_BYTE(g); 119 SkASSERT_IS_BYTE(b); 120 return (a << SK_BGRA_A32_SHIFT) | (r << SK_BGRA_R32_SHIFT) | 121 (g << SK_BGRA_G32_SHIFT) | (b << SK_BGRA_B32_SHIFT); 122} 123 124static inline SkPMColor SkSwizzle_RGBA_to_PMColor(uint32_t c) { 125#ifdef SK_PMCOLOR_IS_RGBA 126 return c; 127#else 128 return SkSwizzle_RB(c); 129#endif 130} 131 132static inline SkPMColor SkSwizzle_BGRA_to_PMColor(uint32_t c) { 133#ifdef SK_PMCOLOR_IS_BGRA 134 return c; 135#else 136 return SkSwizzle_RB(c); 137#endif 138} 139 140////////////////////////////////////////////////////////////////////////////// 141 142///@{ 143/** See ITU-R Recommendation BT.709 at http://www.itu.int/rec/R-REC-BT.709/ .*/ 144#define SK_ITU_BT709_LUM_COEFF_R (0.2126f) 145#define SK_ITU_BT709_LUM_COEFF_G (0.7152f) 146#define SK_ITU_BT709_LUM_COEFF_B (0.0722f) 147///@} 148 149///@{ 150/** A float value which specifies this channel's contribution to luminance. */ 151#define SK_LUM_COEFF_R SK_ITU_BT709_LUM_COEFF_R 152#define SK_LUM_COEFF_G SK_ITU_BT709_LUM_COEFF_G 153#define SK_LUM_COEFF_B SK_ITU_BT709_LUM_COEFF_B 154///@} 155 156/** Computes the luminance from the given r, g, and b in accordance with 157 SK_LUM_COEFF_X. For correct results, r, g, and b should be in linear space. 158*/ 159static inline U8CPU SkComputeLuminance(U8CPU r, U8CPU g, U8CPU b) { 160 //The following is 161 //r * SK_LUM_COEFF_R + g * SK_LUM_COEFF_G + b * SK_LUM_COEFF_B 162 //with SK_LUM_COEFF_X in 1.8 fixed point (rounding adjusted to sum to 256). 163 return (r * 54 + g * 183 + b * 19) >> 8; 164} 165 166/** 167 * Turn a 0..255 value into a 0..256 value, rounding up if the value is >= 0x80. 168 * This is slightly more accurate than SkAlpha255To256. 169 */ 170static inline unsigned Sk255To256(U8CPU value) { 171 SkASSERT(SkToU8(value) == value); 172 return value + (value >> 7); 173} 174 175/** Calculates 256 - (value * alpha256) / 255 in range [0,256], 176 * for [0,255] value and [0,256] alpha256. 177 */ 178static inline U16CPU SkAlphaMulInv256(U16CPU value, U16CPU alpha256) { 179 unsigned prod = 0xFFFF - value * alpha256; 180 return (prod + (prod >> 8)) >> 8; 181} 182 183// The caller may want negative values, so keep all params signed (int) 184// so we don't accidentally slip into unsigned math and lose the sign 185// extension when we shift (in SkAlphaMul) 186static inline int SkAlphaBlend(int src, int dst, int scale256) { 187 SkASSERT((unsigned)scale256 <= 256); 188 return dst + SkAlphaMul(src - dst, scale256); 189} 190 191/** 192 * Returns (src * alpha + dst * (255 - alpha)) / 255 193 * 194 * This is more accurate than SkAlphaBlend, but slightly slower 195 */ 196static inline int SkAlphaBlend255(S16CPU src, S16CPU dst, U8CPU alpha) { 197 SkASSERT((int16_t)src == src); 198 SkASSERT((int16_t)dst == dst); 199 SkASSERT((uint8_t)alpha == alpha); 200 201 int prod = (src - dst) * alpha + 128; 202 prod = (prod + (prod >> 8)) >> 8; 203 return dst + prod; 204} 205 206#define SkR16Assert(r) SkASSERT((unsigned)(r) <= SK_R16_MASK) 207#define SkG16Assert(g) SkASSERT((unsigned)(g) <= SK_G16_MASK) 208#define SkB16Assert(b) SkASSERT((unsigned)(b) <= SK_B16_MASK) 209 210static inline uint16_t SkPackRGB16(unsigned r, unsigned g, unsigned b) { 211 SkASSERT(r <= SK_R16_MASK); 212 SkASSERT(g <= SK_G16_MASK); 213 SkASSERT(b <= SK_B16_MASK); 214 215 return SkToU16((r << SK_R16_SHIFT) | (g << SK_G16_SHIFT) | (b << SK_B16_SHIFT)); 216} 217 218#define SK_R16_MASK_IN_PLACE (SK_R16_MASK << SK_R16_SHIFT) 219#define SK_G16_MASK_IN_PLACE (SK_G16_MASK << SK_G16_SHIFT) 220#define SK_B16_MASK_IN_PLACE (SK_B16_MASK << SK_B16_SHIFT) 221 222/** Expand the 16bit color into a 32bit value that can be scaled all at once 223 by a value up to 32. Used in conjunction with SkCompact_rgb_16. 224*/ 225static inline uint32_t SkExpand_rgb_16(U16CPU c) { 226 SkASSERT(c == (uint16_t)c); 227 228 return ((c & SK_G16_MASK_IN_PLACE) << 16) | (c & ~SK_G16_MASK_IN_PLACE); 229} 230 231/** Compress an expanded value (from SkExpand_rgb_16) back down to a 16bit 232 color value. The computation yields only 16bits of valid data, but we claim 233 to return 32bits, so that the compiler won't generate extra instructions to 234 "clean" the top 16bits. However, the top 16 can contain garbage, so it is 235 up to the caller to safely ignore them. 236*/ 237static inline U16CPU SkCompact_rgb_16(uint32_t c) { 238 return ((c >> 16) & SK_G16_MASK_IN_PLACE) | (c & ~SK_G16_MASK_IN_PLACE); 239} 240 241/** Scale the 16bit color value by the 0..256 scale parameter. 242 The computation yields only 16bits of valid data, but we claim 243 to return 32bits, so that the compiler won't generate extra instructions to 244 "clean" the top 16bits. 245*/ 246static inline U16CPU SkAlphaMulRGB16(U16CPU c, unsigned scale) { 247 return SkCompact_rgb_16(SkExpand_rgb_16(c) * (scale >> 3) >> 5); 248} 249 250// this helper explicitly returns a clean 16bit value (but slower) 251#define SkAlphaMulRGB16_ToU16(c, s) (uint16_t)SkAlphaMulRGB16(c, s) 252 253/** Blend pre-expanded RGB32 with 16bit color value by the 0..32 scale parameter. 254 The computation yields only 16bits of valid data, but we claim to return 255 32bits, so that the compiler won't generate extra instructions to "clean" 256 the top 16bits. 257*/ 258static inline U16CPU SkBlend32_RGB16(uint32_t src_expand, uint16_t dst, unsigned scale) { 259 uint32_t dst_expand = SkExpand_rgb_16(dst) * scale; 260 return SkCompact_rgb_16((src_expand + dst_expand) >> 5); 261} 262 263/** Blend src and dst 16bit colors by the 0..256 scale parameter. 264 The computation yields only 16bits of valid data, but we claim 265 to return 32bits, so that the compiler won't generate extra instructions to 266 "clean" the top 16bits. 267*/ 268static inline U16CPU SkBlendRGB16(U16CPU src, U16CPU dst, int srcScale) { 269 SkASSERT((unsigned)srcScale <= 256); 270 271 srcScale >>= 3; 272 273 uint32_t src32 = SkExpand_rgb_16(src); 274 uint32_t dst32 = SkExpand_rgb_16(dst); 275 return SkCompact_rgb_16(dst32 + ((src32 - dst32) * srcScale >> 5)); 276} 277 278static inline void SkBlendRGB16(const uint16_t src[], uint16_t dst[], 279 int srcScale, int count) { 280 SkASSERT(count > 0); 281 SkASSERT((unsigned)srcScale <= 256); 282 283 srcScale >>= 3; 284 285 do { 286 uint32_t src32 = SkExpand_rgb_16(*src++); 287 uint32_t dst32 = SkExpand_rgb_16(*dst); 288 *dst++ = static_cast<uint16_t>( 289 SkCompact_rgb_16(dst32 + ((src32 - dst32) * srcScale >> 5))); 290 } while (--count > 0); 291} 292 293#ifdef SK_DEBUG 294 static inline U16CPU SkRGB16Add(U16CPU a, U16CPU b) { 295 SkASSERT(SkGetPackedR16(a) + SkGetPackedR16(b) <= SK_R16_MASK); 296 SkASSERT(SkGetPackedG16(a) + SkGetPackedG16(b) <= SK_G16_MASK); 297 SkASSERT(SkGetPackedB16(a) + SkGetPackedB16(b) <= SK_B16_MASK); 298 299 return a + b; 300 } 301#else 302 #define SkRGB16Add(a, b) ((a) + (b)) 303#endif 304 305/////////////////////////////////////////////////////////////////////////////// 306 307#ifdef SK_DEBUG 308 #define SkPMColorAssert(color_value) \ 309 do { \ 310 SkPMColor pm_color_value = (color_value); \ 311 uint32_t alpha_color_value = SkGetPackedA32(pm_color_value); \ 312 SkA32Assert(alpha_color_value); \ 313 SkASSERT(SkGetPackedR32(pm_color_value) <= alpha_color_value); \ 314 SkASSERT(SkGetPackedG32(pm_color_value) <= alpha_color_value); \ 315 SkASSERT(SkGetPackedB32(pm_color_value) <= alpha_color_value); \ 316 } while (false) 317#else 318 #define SkPMColorAssert(c) 319#endif 320 321static inline bool SkPMColorValid(SkPMColor c) { 322 auto a = SkGetPackedA32(c); 323 bool valid = a <= SK_A32_MASK 324 && SkGetPackedR32(c) <= a 325 && SkGetPackedG32(c) <= a 326 && SkGetPackedB32(c) <= a; 327 if (valid) { 328 SkPMColorAssert(c); // Make sure we're consistent when it counts. 329 } 330 return valid; 331} 332 333static inline uint32_t SkPackPMColor_as_RGBA(SkPMColor c) { 334 return SkPackARGB_as_RGBA(SkGetPackedA32(c), SkGetPackedR32(c), 335 SkGetPackedG32(c), SkGetPackedB32(c)); 336} 337 338static inline uint32_t SkPackPMColor_as_BGRA(SkPMColor c) { 339 return SkPackARGB_as_BGRA(SkGetPackedA32(c), SkGetPackedR32(c), 340 SkGetPackedG32(c), SkGetPackedB32(c)); 341} 342 343/** 344 * Abstract 4-byte interpolation, implemented on top of SkPMColor 345 * utility functions. Third parameter controls blending of the first two: 346 * (src, dst, 0) returns dst 347 * (src, dst, 0xFF) returns src 348 * srcWeight is [0..256], unlike SkFourByteInterp which takes [0..255] 349 */ 350static inline SkPMColor SkFourByteInterp256(SkPMColor src, SkPMColor dst, 351 unsigned scale) { 352 unsigned a = SkAlphaBlend(SkGetPackedA32(src), SkGetPackedA32(dst), scale); 353 unsigned r = SkAlphaBlend(SkGetPackedR32(src), SkGetPackedR32(dst), scale); 354 unsigned g = SkAlphaBlend(SkGetPackedG32(src), SkGetPackedG32(dst), scale); 355 unsigned b = SkAlphaBlend(SkGetPackedB32(src), SkGetPackedB32(dst), scale); 356 357 return SkPackARGB32(a, r, g, b); 358} 359 360/** 361 * Abstract 4-byte interpolation, implemented on top of SkPMColor 362 * utility functions. Third parameter controls blending of the first two: 363 * (src, dst, 0) returns dst 364 * (src, dst, 0xFF) returns src 365 */ 366static inline SkPMColor SkFourByteInterp(SkPMColor src, SkPMColor dst, 367 U8CPU srcWeight) { 368 unsigned scale = SkAlpha255To256(srcWeight); 369 return SkFourByteInterp256(src, dst, scale); 370} 371 372/** 373 * 0xAARRGGBB -> 0x00AA00GG, 0x00RR00BB 374 */ 375static inline void SkSplay(uint32_t color, uint32_t* ag, uint32_t* rb) { 376 const uint32_t mask = 0x00FF00FF; 377 *ag = (color >> 8) & mask; 378 *rb = color & mask; 379} 380 381/** 382 * 0xAARRGGBB -> 0x00AA00GG00RR00BB 383 * (note, ARGB -> AGRB) 384 */ 385static inline uint64_t SkSplay(uint32_t color) { 386 const uint32_t mask = 0x00FF00FF; 387 uint64_t agrb = (color >> 8) & mask; // 0x0000000000AA00GG 388 agrb <<= 32; // 0x00AA00GG00000000 389 agrb |= color & mask; // 0x00AA00GG00RR00BB 390 return agrb; 391} 392 393/** 394 * 0xAAxxGGxx, 0xRRxxBBxx-> 0xAARRGGBB 395 */ 396static inline uint32_t SkUnsplay(uint32_t ag, uint32_t rb) { 397 const uint32_t mask = 0xFF00FF00; 398 return (ag & mask) | ((rb & mask) >> 8); 399} 400 401/** 402 * 0xAAxxGGxxRRxxBBxx -> 0xAARRGGBB 403 * (note, AGRB -> ARGB) 404 */ 405static inline uint32_t SkUnsplay(uint64_t agrb) { 406 const uint32_t mask = 0xFF00FF00; 407 return SkPMColor( 408 ((agrb & mask) >> 8) | // 0x00RR00BB 409 ((agrb >> 32) & mask)); // 0xAARRGGBB 410} 411 412static inline SkPMColor SkFastFourByteInterp256_32(SkPMColor src, SkPMColor dst, unsigned scale) { 413 SkASSERT(scale <= 256); 414 415 // Two 8-bit blends per two 32-bit registers, with space to make sure the math doesn't collide. 416 uint32_t src_ag, src_rb, dst_ag, dst_rb; 417 SkSplay(src, &src_ag, &src_rb); 418 SkSplay(dst, &dst_ag, &dst_rb); 419 420 const uint32_t ret_ag = src_ag * scale + (256 - scale) * dst_ag; 421 const uint32_t ret_rb = src_rb * scale + (256 - scale) * dst_rb; 422 423 return SkUnsplay(ret_ag, ret_rb); 424} 425 426static inline SkPMColor SkFastFourByteInterp256_64(SkPMColor src, SkPMColor dst, unsigned scale) { 427 SkASSERT(scale <= 256); 428 // Four 8-bit blends in one 64-bit register, with space to make sure the math doesn't collide. 429 return SkUnsplay(SkSplay(src) * scale + (256-scale) * SkSplay(dst)); 430} 431 432// TODO(mtklein): Replace slow versions with fast versions, using scale + (scale>>7) everywhere. 433 434/** 435 * Same as SkFourByteInterp256, but faster. 436 */ 437static inline SkPMColor SkFastFourByteInterp256(SkPMColor src, SkPMColor dst, unsigned scale) { 438 // On a 64-bit machine, _64 is about 10% faster than _32, but ~40% slower on a 32-bit machine. 439 if (sizeof(void*) == 4) { 440 return SkFastFourByteInterp256_32(src, dst, scale); 441 } else { 442 return SkFastFourByteInterp256_64(src, dst, scale); 443 } 444} 445 446/** 447 * Nearly the same as SkFourByteInterp, but faster and a touch more accurate, due to better 448 * srcWeight scaling to [0, 256]. 449 */ 450static inline SkPMColor SkFastFourByteInterp(SkPMColor src, 451 SkPMColor dst, 452 U8CPU srcWeight) { 453 SkASSERT(srcWeight <= 255); 454 // scale = srcWeight + (srcWeight >> 7) is more accurate than 455 // scale = srcWeight + 1, but 7% slower 456 return SkFastFourByteInterp256(src, dst, srcWeight + (srcWeight >> 7)); 457} 458 459/** 460 * Interpolates between colors src and dst using [0,256] scale. 461 */ 462static inline SkPMColor SkPMLerp(SkPMColor src, SkPMColor dst, unsigned scale) { 463 return SkFastFourByteInterp256(src, dst, scale); 464} 465 466static inline SkPMColor SkBlendARGB32(SkPMColor src, SkPMColor dst, U8CPU aa) { 467 SkASSERT((unsigned)aa <= 255); 468 469 unsigned src_scale = SkAlpha255To256(aa); 470 unsigned dst_scale = SkAlphaMulInv256(SkGetPackedA32(src), src_scale); 471 472 const uint32_t mask = 0xFF00FF; 473 474 uint32_t src_rb = (src & mask) * src_scale; 475 uint32_t src_ag = ((src >> 8) & mask) * src_scale; 476 477 uint32_t dst_rb = (dst & mask) * dst_scale; 478 uint32_t dst_ag = ((dst >> 8) & mask) * dst_scale; 479 480 return (((src_rb + dst_rb) >> 8) & mask) | ((src_ag + dst_ag) & ~mask); 481} 482 483//////////////////////////////////////////////////////////////////////////////////////////// 484// Convert a 32bit pixel to a 16bit pixel (no dither) 485 486#define SkR32ToR16_MACRO(r) ((unsigned)(r) >> (SK_R32_BITS - SK_R16_BITS)) 487#define SkG32ToG16_MACRO(g) ((unsigned)(g) >> (SK_G32_BITS - SK_G16_BITS)) 488#define SkB32ToB16_MACRO(b) ((unsigned)(b) >> (SK_B32_BITS - SK_B16_BITS)) 489 490#ifdef SK_DEBUG 491 static inline unsigned SkR32ToR16(unsigned r) { 492 SkR32Assert(r); 493 return SkR32ToR16_MACRO(r); 494 } 495 static inline unsigned SkG32ToG16(unsigned g) { 496 SkG32Assert(g); 497 return SkG32ToG16_MACRO(g); 498 } 499 static inline unsigned SkB32ToB16(unsigned b) { 500 SkB32Assert(b); 501 return SkB32ToB16_MACRO(b); 502 } 503#else 504 #define SkR32ToR16(r) SkR32ToR16_MACRO(r) 505 #define SkG32ToG16(g) SkG32ToG16_MACRO(g) 506 #define SkB32ToB16(b) SkB32ToB16_MACRO(b) 507#endif 508 509#define SkPacked32ToR16(c) (((unsigned)(c) >> (SK_R32_SHIFT + SK_R32_BITS - SK_R16_BITS)) & SK_R16_MASK) 510#define SkPacked32ToG16(c) (((unsigned)(c) >> (SK_G32_SHIFT + SK_G32_BITS - SK_G16_BITS)) & SK_G16_MASK) 511#define SkPacked32ToB16(c) (((unsigned)(c) >> (SK_B32_SHIFT + SK_B32_BITS - SK_B16_BITS)) & SK_B16_MASK) 512 513static inline U16CPU SkPixel32ToPixel16(SkPMColor c) { 514 unsigned r = ((c >> (SK_R32_SHIFT + (8 - SK_R16_BITS))) & SK_R16_MASK) << SK_R16_SHIFT; 515 unsigned g = ((c >> (SK_G32_SHIFT + (8 - SK_G16_BITS))) & SK_G16_MASK) << SK_G16_SHIFT; 516 unsigned b = ((c >> (SK_B32_SHIFT + (8 - SK_B16_BITS))) & SK_B16_MASK) << SK_B16_SHIFT; 517 return r | g | b; 518} 519 520static inline U16CPU SkPack888ToRGB16(U8CPU r, U8CPU g, U8CPU b) { 521 return (SkR32ToR16(r) << SK_R16_SHIFT) | 522 (SkG32ToG16(g) << SK_G16_SHIFT) | 523 (SkB32ToB16(b) << SK_B16_SHIFT); 524} 525 526#define SkPixel32ToPixel16_ToU16(src) SkToU16(SkPixel32ToPixel16(src)) 527 528///////////////////////////////////////////////////////////////////////////////////////// 529// Fast dither from 32->16 530 531#define SkShouldDitherXY(x, y) (((x) ^ (y)) & 1) 532 533static inline uint16_t SkDitherPack888ToRGB16(U8CPU r, U8CPU g, U8CPU b) { 534 r = ((r << 1) - ((r >> (8 - SK_R16_BITS) << (8 - SK_R16_BITS)) | (r >> SK_R16_BITS))) >> (8 - SK_R16_BITS); 535 g = ((g << 1) - ((g >> (8 - SK_G16_BITS) << (8 - SK_G16_BITS)) | (g >> SK_G16_BITS))) >> (8 - SK_G16_BITS); 536 b = ((b << 1) - ((b >> (8 - SK_B16_BITS) << (8 - SK_B16_BITS)) | (b >> SK_B16_BITS))) >> (8 - SK_B16_BITS); 537 538 return SkPackRGB16(r, g, b); 539} 540 541static inline uint16_t SkDitherPixel32ToPixel16(SkPMColor c) { 542 return SkDitherPack888ToRGB16(SkGetPackedR32(c), SkGetPackedG32(c), SkGetPackedB32(c)); 543} 544 545/* Return c in expanded_rgb_16 format, but also scaled up by 32 (5 bits) 546 It is now suitable for combining with a scaled expanded_rgb_16 color 547 as in SkSrcOver32To16(). 548 We must do this 565 high-bit replication, in order for the subsequent add 549 to saturate properly (and not overflow). If we take the 8 bits as is, it is 550 possible to overflow. 551*/ 552static inline uint32_t SkPMColorToExpanded16x5(SkPMColor c) { 553 unsigned sr = SkPacked32ToR16(c); 554 unsigned sg = SkPacked32ToG16(c); 555 unsigned sb = SkPacked32ToB16(c); 556 557 sr = (sr << 5) | sr; 558 sg = (sg << 5) | (sg >> 1); 559 sb = (sb << 5) | sb; 560 return (sr << 11) | (sg << 21) | (sb << 0); 561} 562 563/* SrcOver the 32bit src color with the 16bit dst, returning a 16bit value 564 (with dirt in the high 16bits, so caller beware). 565*/ 566static inline U16CPU SkSrcOver32To16(SkPMColor src, uint16_t dst) { 567 unsigned sr = SkGetPackedR32(src); 568 unsigned sg = SkGetPackedG32(src); 569 unsigned sb = SkGetPackedB32(src); 570 571 unsigned dr = SkGetPackedR16(dst); 572 unsigned dg = SkGetPackedG16(dst); 573 unsigned db = SkGetPackedB16(dst); 574 575 unsigned isa = 255 - SkGetPackedA32(src); 576 577 dr = (sr + SkMul16ShiftRound(dr, isa, SK_R16_BITS)) >> (8 - SK_R16_BITS); 578 dg = (sg + SkMul16ShiftRound(dg, isa, SK_G16_BITS)) >> (8 - SK_G16_BITS); 579 db = (sb + SkMul16ShiftRound(db, isa, SK_B16_BITS)) >> (8 - SK_B16_BITS); 580 581 return SkPackRGB16(dr, dg, db); 582} 583 584static inline SkPMColor SkPixel16ToPixel32(U16CPU src) { 585 SkASSERT(src == SkToU16(src)); 586 587 unsigned r = SkPacked16ToR32(src); 588 unsigned g = SkPacked16ToG32(src); 589 unsigned b = SkPacked16ToB32(src); 590 591 SkASSERT((r >> (8 - SK_R16_BITS)) == SkGetPackedR16(src)); 592 SkASSERT((g >> (8 - SK_G16_BITS)) == SkGetPackedG16(src)); 593 SkASSERT((b >> (8 - SK_B16_BITS)) == SkGetPackedB16(src)); 594 595 return SkPackARGB32(0xFF, r, g, b); 596} 597 598// similar to SkPixel16ToPixel32, but returns SkColor instead of SkPMColor 599static inline SkColor SkPixel16ToColor(U16CPU src) { 600 SkASSERT(src == SkToU16(src)); 601 602 unsigned r = SkPacked16ToR32(src); 603 unsigned g = SkPacked16ToG32(src); 604 unsigned b = SkPacked16ToB32(src); 605 606 SkASSERT((r >> (8 - SK_R16_BITS)) == SkGetPackedR16(src)); 607 SkASSERT((g >> (8 - SK_G16_BITS)) == SkGetPackedG16(src)); 608 SkASSERT((b >> (8 - SK_B16_BITS)) == SkGetPackedB16(src)); 609 610 return SkColorSetRGB(r, g, b); 611} 612 613/////////////////////////////////////////////////////////////////////////////// 614 615typedef uint16_t SkPMColor16; 616 617// Put in OpenGL order (r g b a) 618#define SK_A4444_SHIFT 0 619#define SK_R4444_SHIFT 12 620#define SK_G4444_SHIFT 8 621#define SK_B4444_SHIFT 4 622 623#define SkA32To4444(a) ((unsigned)(a) >> 4) 624#define SkR32To4444(r) ((unsigned)(r) >> 4) 625#define SkG32To4444(g) ((unsigned)(g) >> 4) 626#define SkB32To4444(b) ((unsigned)(b) >> 4) 627 628static inline U8CPU SkReplicateNibble(unsigned nib) { 629 SkASSERT(nib <= 0xF); 630 return (nib << 4) | nib; 631} 632 633#define SkA4444ToA32(a) SkReplicateNibble(a) 634#define SkR4444ToR32(r) SkReplicateNibble(r) 635#define SkG4444ToG32(g) SkReplicateNibble(g) 636#define SkB4444ToB32(b) SkReplicateNibble(b) 637 638#define SkGetPackedA4444(c) (((unsigned)(c) >> SK_A4444_SHIFT) & 0xF) 639#define SkGetPackedR4444(c) (((unsigned)(c) >> SK_R4444_SHIFT) & 0xF) 640#define SkGetPackedG4444(c) (((unsigned)(c) >> SK_G4444_SHIFT) & 0xF) 641#define SkGetPackedB4444(c) (((unsigned)(c) >> SK_B4444_SHIFT) & 0xF) 642 643#define SkPacked4444ToA32(c) SkReplicateNibble(SkGetPackedA4444(c)) 644#define SkPacked4444ToR32(c) SkReplicateNibble(SkGetPackedR4444(c)) 645#define SkPacked4444ToG32(c) SkReplicateNibble(SkGetPackedG4444(c)) 646#define SkPacked4444ToB32(c) SkReplicateNibble(SkGetPackedB4444(c)) 647 648#ifdef SK_DEBUG 649static inline void SkPMColor16Assert(U16CPU c) { 650 unsigned a = SkGetPackedA4444(c); 651 unsigned r = SkGetPackedR4444(c); 652 unsigned g = SkGetPackedG4444(c); 653 unsigned b = SkGetPackedB4444(c); 654 655 SkASSERT(a <= 0xF); 656 SkASSERT(r <= a); 657 SkASSERT(g <= a); 658 SkASSERT(b <= a); 659} 660#else 661#define SkPMColor16Assert(c) 662#endif 663 664static inline unsigned SkAlpha15To16(unsigned a) { 665 SkASSERT(a <= 0xF); 666 return a + (a >> 3); 667} 668 669#ifdef SK_DEBUG 670 static inline int SkAlphaMul4(int value, int scale) { 671 SkASSERT((unsigned)scale <= 0x10); 672 return value * scale >> 4; 673 } 674#else 675 #define SkAlphaMul4(value, scale) ((value) * (scale) >> 4) 676#endif 677 678static inline unsigned SkR4444ToR565(unsigned r) { 679 SkASSERT(r <= 0xF); 680 return (r << (SK_R16_BITS - 4)) | (r >> (8 - SK_R16_BITS)); 681} 682 683static inline unsigned SkG4444ToG565(unsigned g) { 684 SkASSERT(g <= 0xF); 685 return (g << (SK_G16_BITS - 4)) | (g >> (8 - SK_G16_BITS)); 686} 687 688static inline unsigned SkB4444ToB565(unsigned b) { 689 SkASSERT(b <= 0xF); 690 return (b << (SK_B16_BITS - 4)) | (b >> (8 - SK_B16_BITS)); 691} 692 693static inline SkPMColor16 SkPackARGB4444(unsigned a, unsigned r, 694 unsigned g, unsigned b) { 695 SkASSERT(a <= 0xF); 696 SkASSERT(r <= a); 697 SkASSERT(g <= a); 698 SkASSERT(b <= a); 699 700 return (SkPMColor16)((a << SK_A4444_SHIFT) | (r << SK_R4444_SHIFT) | 701 (g << SK_G4444_SHIFT) | (b << SK_B4444_SHIFT)); 702} 703 704static inline SkPMColor16 SkAlphaMulQ4(SkPMColor16 c, int scale) { 705 SkASSERT(scale <= 16); 706 707 const unsigned mask = 0xF0F; //gMask_0F0F; 708 709#if 0 710 unsigned rb = ((c & mask) * scale) >> 4; 711 unsigned ag = ((c >> 4) & mask) * scale; 712 return (rb & mask) | (ag & ~mask); 713#else 714 unsigned expanded_c = (c & mask) | ((c & (mask << 4)) << 12); 715 unsigned scaled_c = (expanded_c * scale) >> 4; 716 return (scaled_c & mask) | ((scaled_c >> 12) & (mask << 4)); 717#endif 718} 719 720/** Expand the SkPMColor16 color into a 32bit value that can be scaled all at 721 once by a value up to 16. 722*/ 723static inline uint32_t SkExpand_4444(U16CPU c) { 724 SkASSERT(c == (uint16_t)c); 725 726 const unsigned mask = 0xF0F; //gMask_0F0F; 727 return (c & mask) | ((c & ~mask) << 12); 728} 729 730static inline uint16_t SkSrcOver4444To16(SkPMColor16 s, uint16_t d) { 731 unsigned sa = SkGetPackedA4444(s); 732 unsigned sr = SkR4444ToR565(SkGetPackedR4444(s)); 733 unsigned sg = SkG4444ToG565(SkGetPackedG4444(s)); 734 unsigned sb = SkB4444ToB565(SkGetPackedB4444(s)); 735 736 // To avoid overflow, we have to clear the low bit of the synthetic sg 737 // if the src alpha is <= 7. 738 // to see why, try blending 0x4444 on top of 565-white and watch green 739 // overflow (sum == 64) 740 sg &= ~(~(sa >> 3) & 1); 741 742 unsigned scale = SkAlpha15To16(15 - sa); 743 unsigned dr = SkAlphaMul4(SkGetPackedR16(d), scale); 744 unsigned dg = SkAlphaMul4(SkGetPackedG16(d), scale); 745 unsigned db = SkAlphaMul4(SkGetPackedB16(d), scale); 746 747#if 0 748 if (sg + dg > 63) { 749 SkDebugf("---- SkSrcOver4444To16 src=%x dst=%x scale=%d, sg=%d dg=%d\n", s, d, scale, sg, dg); 750 } 751#endif 752 return SkPackRGB16(sr + dr, sg + dg, sb + db); 753} 754 755static inline uint16_t SkBlend4444To16(SkPMColor16 src, uint16_t dst, int scale16) { 756 SkASSERT((unsigned)scale16 <= 16); 757 758 return SkSrcOver4444To16(SkAlphaMulQ4(src, scale16), dst); 759} 760 761static inline SkPMColor SkPixel4444ToPixel32(U16CPU c) { 762 uint32_t d = (SkGetPackedA4444(c) << SK_A32_SHIFT) | 763 (SkGetPackedR4444(c) << SK_R32_SHIFT) | 764 (SkGetPackedG4444(c) << SK_G32_SHIFT) | 765 (SkGetPackedB4444(c) << SK_B32_SHIFT); 766 return d | (d << 4); 767} 768 769static inline SkPMColor16 SkPixel32ToPixel4444(SkPMColor c) { 770 return (((c >> (SK_A32_SHIFT + 4)) & 0xF) << SK_A4444_SHIFT) | 771 (((c >> (SK_R32_SHIFT + 4)) & 0xF) << SK_R4444_SHIFT) | 772 (((c >> (SK_G32_SHIFT + 4)) & 0xF) << SK_G4444_SHIFT) | 773 (((c >> (SK_B32_SHIFT + 4)) & 0xF) << SK_B4444_SHIFT); 774} 775 776// cheap 2x2 dither 777static inline SkPMColor16 SkDitherARGB32To4444(U8CPU a, U8CPU r, 778 U8CPU g, U8CPU b) { 779 // to ensure that we stay a legal premultiplied color, we take the max() 780 // of the truncated and dithered alpha values. If we didn't, cases like 781 // SkDitherARGB32To4444(0x31, 0x2E, ...) would generate SkPackARGB4444(2, 3, ...) 782 // which is not legal premultiplied, since a < color 783 unsigned dithered_a = ((a << 1) - ((a >> 4 << 4) | (a >> 4))) >> 4; 784 a = SkMax32(a >> 4, dithered_a); 785 // these we just dither in place 786 r = ((r << 1) - ((r >> 4 << 4) | (r >> 4))) >> 4; 787 g = ((g << 1) - ((g >> 4 << 4) | (g >> 4))) >> 4; 788 b = ((b << 1) - ((b >> 4 << 4) | (b >> 4))) >> 4; 789 790 return SkPackARGB4444(a, r, g, b); 791} 792 793static inline SkPMColor16 SkDitherPixel32To4444(SkPMColor c) { 794 return SkDitherARGB32To4444(SkGetPackedA32(c), SkGetPackedR32(c), 795 SkGetPackedG32(c), SkGetPackedB32(c)); 796} 797 798/* Assumes 16bit is in standard RGBA order. 799 Transforms a normal ARGB_8888 into the same byte order as 800 expanded ARGB_4444, but keeps each component 8bits 801*/ 802static inline uint32_t SkExpand_8888(SkPMColor c) { 803 return (((c >> SK_R32_SHIFT) & 0xFF) << 24) | 804 (((c >> SK_G32_SHIFT) & 0xFF) << 8) | 805 (((c >> SK_B32_SHIFT) & 0xFF) << 16) | 806 (((c >> SK_A32_SHIFT) & 0xFF) << 0); 807} 808 809/* Undo the operation of SkExpand_8888, turning the argument back into 810 a SkPMColor. 811*/ 812static inline SkPMColor SkCompact_8888(uint32_t c) { 813 return (((c >> 24) & 0xFF) << SK_R32_SHIFT) | 814 (((c >> 8) & 0xFF) << SK_G32_SHIFT) | 815 (((c >> 16) & 0xFF) << SK_B32_SHIFT) | 816 (((c >> 0) & 0xFF) << SK_A32_SHIFT); 817} 818 819/* Like SkExpand_8888, this transforms a pmcolor into the expanded 4444 format, 820 but this routine just keeps the high 4bits of each component in the low 821 4bits of the result (just like a newly expanded PMColor16). 822*/ 823static inline uint32_t SkExpand32_4444(SkPMColor c) { 824 return (((c >> (SK_R32_SHIFT + 4)) & 0xF) << 24) | 825 (((c >> (SK_G32_SHIFT + 4)) & 0xF) << 8) | 826 (((c >> (SK_B32_SHIFT + 4)) & 0xF) << 16) | 827 (((c >> (SK_A32_SHIFT + 4)) & 0xF) << 0); 828} 829 830// takes two values and alternamtes them as part of a memset16 831// used for cheap 2x2 dithering when the colors are opaque 832void sk_dither_memset16(uint16_t dst[], uint16_t value, uint16_t other, int n); 833 834/////////////////////////////////////////////////////////////////////////////// 835 836static inline int SkUpscale31To32(int value) { 837 SkASSERT((unsigned)value <= 31); 838 return value + (value >> 4); 839} 840 841static inline int SkBlend32(int src, int dst, int scale) { 842 SkASSERT((unsigned)src <= 0xFF); 843 SkASSERT((unsigned)dst <= 0xFF); 844 SkASSERT((unsigned)scale <= 32); 845 return dst + ((src - dst) * scale >> 5); 846} 847 848static inline SkPMColor SkBlendLCD16(int srcA, int srcR, int srcG, int srcB, 849 SkPMColor dst, uint16_t mask) { 850 if (mask == 0) { 851 return dst; 852 } 853 854 /* We want all of these in 5bits, hence the shifts in case one of them 855 * (green) is 6bits. 856 */ 857 int maskR = SkGetPackedR16(mask) >> (SK_R16_BITS - 5); 858 int maskG = SkGetPackedG16(mask) >> (SK_G16_BITS - 5); 859 int maskB = SkGetPackedB16(mask) >> (SK_B16_BITS - 5); 860 861 // Now upscale them to 0..32, so we can use blend32 862 maskR = SkUpscale31To32(maskR); 863 maskG = SkUpscale31To32(maskG); 864 maskB = SkUpscale31To32(maskB); 865 866 // srcA has been upscaled to 256 before passed into this function 867 maskR = maskR * srcA >> 8; 868 maskG = maskG * srcA >> 8; 869 maskB = maskB * srcA >> 8; 870 871 int dstR = SkGetPackedR32(dst); 872 int dstG = SkGetPackedG32(dst); 873 int dstB = SkGetPackedB32(dst); 874 875 // LCD blitting is only supported if the dst is known/required 876 // to be opaque 877 return SkPackARGB32(0xFF, 878 SkBlend32(srcR, dstR, maskR), 879 SkBlend32(srcG, dstG, maskG), 880 SkBlend32(srcB, dstB, maskB)); 881} 882 883static inline SkPMColor SkBlendLCD16Opaque(int srcR, int srcG, int srcB, 884 SkPMColor dst, uint16_t mask, 885 SkPMColor opaqueDst) { 886 if (mask == 0) { 887 return dst; 888 } 889 890 if (0xFFFF == mask) { 891 return opaqueDst; 892 } 893 894 /* We want all of these in 5bits, hence the shifts in case one of them 895 * (green) is 6bits. 896 */ 897 int maskR = SkGetPackedR16(mask) >> (SK_R16_BITS - 5); 898 int maskG = SkGetPackedG16(mask) >> (SK_G16_BITS - 5); 899 int maskB = SkGetPackedB16(mask) >> (SK_B16_BITS - 5); 900 901 // Now upscale them to 0..32, so we can use blend32 902 maskR = SkUpscale31To32(maskR); 903 maskG = SkUpscale31To32(maskG); 904 maskB = SkUpscale31To32(maskB); 905 906 int dstR = SkGetPackedR32(dst); 907 int dstG = SkGetPackedG32(dst); 908 int dstB = SkGetPackedB32(dst); 909 910 // LCD blitting is only supported if the dst is known/required 911 // to be opaque 912 return SkPackARGB32(0xFF, 913 SkBlend32(srcR, dstR, maskR), 914 SkBlend32(srcG, dstG, maskG), 915 SkBlend32(srcB, dstB, maskB)); 916} 917 918static inline void SkBlitLCD16Row(SkPMColor dst[], const uint16_t mask[], 919 SkColor src, int width, SkPMColor) { 920 int srcA = SkColorGetA(src); 921 int srcR = SkColorGetR(src); 922 int srcG = SkColorGetG(src); 923 int srcB = SkColorGetB(src); 924 925 srcA = SkAlpha255To256(srcA); 926 927 for (int i = 0; i < width; i++) { 928 dst[i] = SkBlendLCD16(srcA, srcR, srcG, srcB, dst[i], mask[i]); 929 } 930} 931 932static inline void SkBlitLCD16OpaqueRow(SkPMColor dst[], const uint16_t mask[], 933 SkColor src, int width, 934 SkPMColor opaqueDst) { 935 int srcR = SkColorGetR(src); 936 int srcG = SkColorGetG(src); 937 int srcB = SkColorGetB(src); 938 939 for (int i = 0; i < width; i++) { 940 dst[i] = SkBlendLCD16Opaque(srcR, srcG, srcB, dst[i], mask[i], 941 opaqueDst); 942 } 943} 944 945#endif 946