Mercurial > sdl-ios-xcode
annotate src/video/SDL_blit.h @ 1052:68f607298ca9
Some work on using accelerated alpha blits with hardware surfaces.
From Stephane Marchesin's fork, don't know who originally wrote it.
| author | Ryan C. Gordon <icculus@icculus.org> |
|---|---|
| date | Sun, 17 Apr 2005 10:40:41 +0000 |
| parents | ffaaf7ecf685 |
| children | 2651158f59b8 |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 SDL - Simple DirectMedia Layer | |
|
769
b8d311d90021
Updated copyright information for 2004 (Happy New Year!)
Sam Lantinga <slouken@libsdl.org>
parents:
689
diff
changeset
|
3 Copyright (C) 1997-2004 Sam Lantinga |
| 0 | 4 |
| 5 This library is free software; you can redistribute it and/or | |
| 6 modify it under the terms of the GNU Library General Public | |
| 7 License as published by the Free Software Foundation; either | |
| 8 version 2 of the License, or (at your option) any later version. | |
| 9 | |
| 10 This library is distributed in the hope that it will be useful, | |
| 11 but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 13 Library General Public License for more details. | |
| 14 | |
| 15 You should have received a copy of the GNU Library General Public | |
| 16 License along with this library; if not, write to the Free | |
| 17 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
| 18 | |
| 19 Sam Lantinga | |
|
252
e8157fcb3114
Updated the source with the correct e-mail address
Sam Lantinga <slouken@libsdl.org>
parents:
163
diff
changeset
|
20 slouken@libsdl.org |
| 0 | 21 */ |
| 22 | |
| 23 #ifdef SAVE_RCSID | |
| 24 static char rcsid = | |
| 25 "@(#) $Id$"; | |
| 26 #endif | |
| 27 | |
| 28 #ifndef _SDL_blit_h | |
| 29 #define _SDL_blit_h | |
| 30 | |
| 31 #include "SDL_endian.h" | |
| 32 | |
| 33 /* The structure passed to the low level blit functions */ | |
| 34 typedef struct { | |
| 35 Uint8 *s_pixels; | |
| 36 int s_width; | |
| 37 int s_height; | |
| 38 int s_skip; | |
| 39 Uint8 *d_pixels; | |
| 40 int d_width; | |
| 41 int d_height; | |
| 42 int d_skip; | |
| 43 void *aux_data; | |
| 44 SDL_PixelFormat *src; | |
| 45 Uint8 *table; | |
| 46 SDL_PixelFormat *dst; | |
| 47 } SDL_BlitInfo; | |
| 48 | |
| 49 /* The type definition for the low level blit functions */ | |
| 50 typedef void (*SDL_loblit)(SDL_BlitInfo *info); | |
| 51 | |
| 52 /* This is the private info structure for software accelerated blits */ | |
| 53 struct private_swaccel { | |
| 54 SDL_loblit blit; | |
| 55 void *aux_data; | |
| 56 }; | |
| 57 | |
| 58 /* Blit mapping definition */ | |
| 59 typedef struct SDL_BlitMap { | |
| 60 SDL_Surface *dst; | |
| 61 int identity; | |
| 62 Uint8 *table; | |
| 63 SDL_blit hw_blit; | |
| 64 SDL_blit sw_blit; | |
| 65 struct private_hwaccel *hw_data; | |
| 66 struct private_swaccel *sw_data; | |
| 67 | |
| 68 /* the version count matches the destination; mismatch indicates | |
| 69 an invalid mapping */ | |
| 70 unsigned int format_version; | |
| 71 } SDL_BlitMap; | |
| 72 | |
| 73 | |
| 74 /* Functions found in SDL_blit.c */ | |
| 75 extern int SDL_CalculateBlit(SDL_Surface *surface); | |
| 76 | |
| 77 /* Functions found in SDL_blit_{0,1,N,A}.c */ | |
| 78 extern SDL_loblit SDL_CalculateBlit0(SDL_Surface *surface, int complex); | |
| 79 extern SDL_loblit SDL_CalculateBlit1(SDL_Surface *surface, int complex); | |
| 80 extern SDL_loblit SDL_CalculateBlitN(SDL_Surface *surface, int complex); | |
| 81 extern SDL_loblit SDL_CalculateAlphaBlit(SDL_Surface *surface, int complex); | |
| 82 | |
| 83 /* | |
| 84 * Useful macros for blitting routines | |
| 85 */ | |
| 86 | |
| 87 #define FORMAT_EQUAL(A, B) \ | |
| 88 ((A)->BitsPerPixel == (B)->BitsPerPixel \ | |
| 89 && ((A)->Rmask == (B)->Rmask) && ((A)->Amask == (B)->Amask)) | |
| 90 | |
| 91 /* Load pixel of the specified format from a buffer and get its R-G-B values */ | |
| 92 /* FIXME: rescale values to 0..255 here? */ | |
| 93 #define RGB_FROM_PIXEL(pixel, fmt, r, g, b) \ | |
| 94 { \ | |
| 95 r = (((pixel&fmt->Rmask)>>fmt->Rshift)<<fmt->Rloss); \ | |
| 96 g = (((pixel&fmt->Gmask)>>fmt->Gshift)<<fmt->Gloss); \ | |
| 97 b = (((pixel&fmt->Bmask)>>fmt->Bshift)<<fmt->Bloss); \ | |
| 98 } | |
| 99 #define RGB_FROM_RGB565(pixel, r, g, b) \ | |
| 100 { \ | |
| 101 r = (((pixel&0xF800)>>11)<<3); \ | |
| 102 g = (((pixel&0x07E0)>>5)<<2); \ | |
| 103 b = ((pixel&0x001F)<<3); \ | |
| 104 } | |
| 105 #define RGB_FROM_RGB555(pixel, r, g, b) \ | |
| 106 { \ | |
| 107 r = (((pixel&0x7C00)>>10)<<3); \ | |
| 108 g = (((pixel&0x03E0)>>5)<<3); \ | |
| 109 b = ((pixel&0x001F)<<3); \ | |
| 110 } | |
| 111 #define RGB_FROM_RGB888(pixel, r, g, b) \ | |
| 112 { \ | |
| 113 r = ((pixel&0xFF0000)>>16); \ | |
| 114 g = ((pixel&0xFF00)>>8); \ | |
| 115 b = (pixel&0xFF); \ | |
| 116 } | |
| 117 #define RETRIEVE_RGB_PIXEL(buf, bpp, pixel) \ | |
| 118 do { \ | |
| 119 switch (bpp) { \ | |
| 120 case 2: \ | |
| 121 pixel = *((Uint16 *)(buf)); \ | |
| 122 break; \ | |
| 123 \ | |
| 124 case 3: { \ | |
| 125 Uint8 *B = (Uint8 *)(buf); \ | |
| 126 if(SDL_BYTEORDER == SDL_LIL_ENDIAN) { \ | |
| 127 pixel = B[0] + (B[1] << 8) + (B[2] << 16); \ | |
| 128 } else { \ | |
| 129 pixel = (B[0] << 16) + (B[1] << 8) + B[2]; \ | |
| 130 } \ | |
| 131 } \ | |
| 132 break; \ | |
| 133 \ | |
| 134 case 4: \ | |
| 135 pixel = *((Uint32 *)(buf)); \ | |
| 136 break; \ | |
| 137 \ | |
| 138 default: \ | |
| 139 pixel = 0; /* appease gcc */ \ | |
| 140 break; \ | |
| 141 } \ | |
| 142 } while(0) | |
| 143 | |
| 144 #define DISEMBLE_RGB(buf, bpp, fmt, pixel, r, g, b) \ | |
| 145 do { \ | |
| 146 switch (bpp) { \ | |
| 147 case 2: \ | |
| 148 pixel = *((Uint16 *)(buf)); \ | |
| 149 break; \ | |
| 150 \ | |
| 151 case 3: { \ | |
| 152 Uint8 *B = (Uint8 *)buf; \ | |
| 153 if(SDL_BYTEORDER == SDL_LIL_ENDIAN) { \ | |
| 154 pixel = B[0] + (B[1] << 8) + (B[2] << 16); \ | |
| 155 } else { \ | |
| 156 pixel = (B[0] << 16) + (B[1] << 8) + B[2]; \ | |
| 157 } \ | |
| 158 } \ | |
| 159 break; \ | |
| 160 \ | |
| 161 case 4: \ | |
| 162 pixel = *((Uint32 *)(buf)); \ | |
| 163 break; \ | |
| 164 \ | |
| 165 default: \ | |
| 166 pixel = 0; /* prevent gcc from complaining */ \ | |
| 167 break; \ | |
| 168 } \ | |
| 169 RGB_FROM_PIXEL(pixel, fmt, r, g, b); \ | |
| 170 } while(0) | |
| 171 | |
| 172 /* Assemble R-G-B values into a specified pixel format and store them */ | |
| 173 #define PIXEL_FROM_RGB(pixel, fmt, r, g, b) \ | |
| 174 { \ | |
| 175 pixel = ((r>>fmt->Rloss)<<fmt->Rshift)| \ | |
| 176 ((g>>fmt->Gloss)<<fmt->Gshift)| \ | |
| 177 ((b>>fmt->Bloss)<<fmt->Bshift); \ | |
| 178 } | |
| 179 #define RGB565_FROM_RGB(pixel, r, g, b) \ | |
| 180 { \ | |
| 181 pixel = ((r>>3)<<11)|((g>>2)<<5)|(b>>3); \ | |
| 182 } | |
| 183 #define RGB555_FROM_RGB(pixel, r, g, b) \ | |
| 184 { \ | |
| 185 pixel = ((r>>3)<<10)|((g>>3)<<5)|(b>>3); \ | |
| 186 } | |
| 187 #define RGB888_FROM_RGB(pixel, r, g, b) \ | |
| 188 { \ | |
| 189 pixel = (r<<16)|(g<<8)|b; \ | |
| 190 } | |
| 191 #define ASSEMBLE_RGB(buf, bpp, fmt, r, g, b) \ | |
| 192 { \ | |
| 193 switch (bpp) { \ | |
| 194 case 2: { \ | |
| 195 Uint16 pixel; \ | |
| 196 \ | |
| 197 PIXEL_FROM_RGB(pixel, fmt, r, g, b); \ | |
| 198 *((Uint16 *)(buf)) = pixel; \ | |
| 199 } \ | |
| 200 break; \ | |
| 201 \ | |
| 202 case 3: { \ | |
| 203 if(SDL_BYTEORDER == SDL_LIL_ENDIAN) { \ | |
| 204 *((buf)+fmt->Rshift/8) = r; \ | |
| 205 *((buf)+fmt->Gshift/8) = g; \ | |
| 206 *((buf)+fmt->Bshift/8) = b; \ | |
| 207 } else { \ | |
| 208 *((buf)+2-fmt->Rshift/8) = r; \ | |
| 209 *((buf)+2-fmt->Gshift/8) = g; \ | |
| 210 *((buf)+2-fmt->Bshift/8) = b; \ | |
| 211 } \ | |
| 212 } \ | |
| 213 break; \ | |
| 214 \ | |
| 215 case 4: { \ | |
| 216 Uint32 pixel; \ | |
| 217 \ | |
| 218 PIXEL_FROM_RGB(pixel, fmt, r, g, b); \ | |
| 219 *((Uint32 *)(buf)) = pixel; \ | |
| 220 } \ | |
| 221 break; \ | |
| 222 } \ | |
| 223 } | |
| 224 #define ASSEMBLE_RGB_AMASK(buf, bpp, fmt, r, g, b, Amask) \ | |
| 225 { \ | |
| 226 switch (bpp) { \ | |
| 227 case 2: { \ | |
| 228 Uint16 *bufp; \ | |
| 229 Uint16 pixel; \ | |
| 230 \ | |
| 231 bufp = (Uint16 *)buf; \ | |
| 232 PIXEL_FROM_RGB(pixel, fmt, r, g, b); \ | |
| 233 *bufp = pixel | (*bufp & Amask); \ | |
| 234 } \ | |
| 235 break; \ | |
| 236 \ | |
| 237 case 3: { \ | |
| 238 if(SDL_BYTEORDER == SDL_LIL_ENDIAN) { \ | |
| 239 *((buf)+fmt->Rshift/8) = r; \ | |
| 240 *((buf)+fmt->Gshift/8) = g; \ | |
| 241 *((buf)+fmt->Bshift/8) = b; \ | |
| 242 } else { \ | |
| 243 *((buf)+2-fmt->Rshift/8) = r; \ | |
| 244 *((buf)+2-fmt->Gshift/8) = g; \ | |
| 245 *((buf)+2-fmt->Bshift/8) = b; \ | |
| 246 } \ | |
| 247 } \ | |
| 248 break; \ | |
| 249 \ | |
| 250 case 4: { \ | |
| 251 Uint32 *bufp; \ | |
| 252 Uint32 pixel; \ | |
| 253 \ | |
| 254 bufp = (Uint32 *)buf; \ | |
| 255 PIXEL_FROM_RGB(pixel, fmt, r, g, b); \ | |
| 256 *bufp = pixel | (*bufp & Amask); \ | |
| 257 } \ | |
| 258 break; \ | |
| 259 } \ | |
| 260 } | |
| 261 | |
| 262 /* FIXME: Should we rescale alpha into 0..255 here? */ | |
| 263 #define RGBA_FROM_PIXEL(pixel, fmt, r, g, b, a) \ | |
| 264 { \ | |
| 265 r = ((pixel&fmt->Rmask)>>fmt->Rshift)<<fmt->Rloss; \ | |
| 266 g = ((pixel&fmt->Gmask)>>fmt->Gshift)<<fmt->Gloss; \ | |
| 267 b = ((pixel&fmt->Bmask)>>fmt->Bshift)<<fmt->Bloss; \ | |
| 268 a = ((pixel&fmt->Amask)>>fmt->Ashift)<<fmt->Aloss; \ | |
| 269 } | |
| 270 #define RGBA_FROM_8888(pixel, fmt, r, g, b, a) \ | |
| 271 { \ | |
| 272 r = (pixel&fmt->Rmask)>>fmt->Rshift; \ | |
| 273 g = (pixel&fmt->Gmask)>>fmt->Gshift; \ | |
| 274 b = (pixel&fmt->Bmask)>>fmt->Bshift; \ | |
| 275 a = (pixel&fmt->Amask)>>fmt->Ashift; \ | |
| 276 } | |
| 277 #define RGBA_FROM_RGBA8888(pixel, r, g, b, a) \ | |
| 278 { \ | |
| 279 r = (pixel>>24); \ | |
| 280 g = ((pixel>>16)&0xFF); \ | |
| 281 b = ((pixel>>8)&0xFF); \ | |
| 282 a = (pixel&0xFF); \ | |
| 283 } | |
| 284 #define RGBA_FROM_ARGB8888(pixel, r, g, b, a) \ | |
| 285 { \ | |
| 286 r = ((pixel>>16)&0xFF); \ | |
| 287 g = ((pixel>>8)&0xFF); \ | |
| 288 b = (pixel&0xFF); \ | |
| 289 a = (pixel>>24); \ | |
| 290 } | |
| 291 #define RGBA_FROM_ABGR8888(pixel, r, g, b, a) \ | |
| 292 { \ | |
| 293 r = (pixel&0xFF); \ | |
| 294 g = ((pixel>>8)&0xFF); \ | |
| 295 b = ((pixel>>16)&0xFF); \ | |
| 296 a = (pixel>>24); \ | |
| 297 } | |
| 298 #define DISEMBLE_RGBA(buf, bpp, fmt, pixel, r, g, b, a) \ | |
| 299 do { \ | |
| 300 switch (bpp) { \ | |
| 301 case 2: \ | |
| 302 pixel = *((Uint16 *)(buf)); \ | |
| 303 break; \ | |
| 304 \ | |
| 305 case 3: {/* FIXME: broken code (no alpha) */ \ | |
| 306 Uint8 *b = (Uint8 *)buf; \ | |
| 307 if(SDL_BYTEORDER == SDL_LIL_ENDIAN) { \ | |
| 308 pixel = b[0] + (b[1] << 8) + (b[2] << 16); \ | |
| 309 } else { \ | |
| 310 pixel = (b[0] << 16) + (b[1] << 8) + b[2]; \ | |
| 311 } \ | |
| 312 } \ | |
| 313 break; \ | |
| 314 \ | |
| 315 case 4: \ | |
| 316 pixel = *((Uint32 *)(buf)); \ | |
| 317 break; \ | |
| 318 \ | |
| 319 default: \ | |
| 320 pixel = 0; /* stop gcc complaints */ \ | |
| 321 break; \ | |
| 322 } \ | |
| 323 RGBA_FROM_PIXEL(pixel, fmt, r, g, b, a); \ | |
| 324 pixel &= ~fmt->Amask; \ | |
| 325 } while(0) | |
| 326 | |
| 327 /* FIXME: this isn't correct, especially for Alpha (maximum != 255) */ | |
| 328 #define PIXEL_FROM_RGBA(pixel, fmt, r, g, b, a) \ | |
| 329 { \ | |
| 330 pixel = ((r>>fmt->Rloss)<<fmt->Rshift)| \ | |
| 331 ((g>>fmt->Gloss)<<fmt->Gshift)| \ | |
| 332 ((b>>fmt->Bloss)<<fmt->Bshift)| \ | |
|
535
917cc5c56176
Fixed alpha blending bug (thanks Glenn!)
Sam Lantinga <slouken@libsdl.org>
parents:
297
diff
changeset
|
333 ((a>>fmt->Aloss)<<fmt->Ashift); \ |
| 0 | 334 } |
| 335 #define ASSEMBLE_RGBA(buf, bpp, fmt, r, g, b, a) \ | |
| 336 { \ | |
| 337 switch (bpp) { \ | |
| 338 case 2: { \ | |
| 339 Uint16 pixel; \ | |
| 340 \ | |
| 341 PIXEL_FROM_RGBA(pixel, fmt, r, g, b, a); \ | |
| 342 *((Uint16 *)(buf)) = pixel; \ | |
| 343 } \ | |
| 344 break; \ | |
| 345 \ | |
| 346 case 3: { /* FIXME: broken code (no alpha) */ \ | |
| 347 if(SDL_BYTEORDER == SDL_LIL_ENDIAN) { \ | |
| 348 *((buf)+fmt->Rshift/8) = r; \ | |
| 349 *((buf)+fmt->Gshift/8) = g; \ | |
| 350 *((buf)+fmt->Bshift/8) = b; \ | |
| 351 } else { \ | |
| 352 *((buf)+2-fmt->Rshift/8) = r; \ | |
| 353 *((buf)+2-fmt->Gshift/8) = g; \ | |
| 354 *((buf)+2-fmt->Bshift/8) = b; \ | |
| 355 } \ | |
| 356 } \ | |
| 357 break; \ | |
| 358 \ | |
| 359 case 4: { \ | |
| 360 Uint32 pixel; \ | |
| 361 \ | |
| 362 PIXEL_FROM_RGBA(pixel, fmt, r, g, b, a); \ | |
| 363 *((Uint32 *)(buf)) = pixel; \ | |
| 364 } \ | |
| 365 break; \ | |
| 366 } \ | |
| 367 } | |
| 368 | |
| 369 /* Blend the RGB values of two pixels based on a source alpha value */ | |
| 370 #define ALPHA_BLEND(sR, sG, sB, A, dR, dG, dB) \ | |
| 371 do { \ | |
| 372 dR = (((sR-dR)*(A))>>8)+dR; \ | |
| 373 dG = (((sG-dG)*(A))>>8)+dG; \ | |
| 374 dB = (((sB-dB)*(A))>>8)+dB; \ | |
| 375 } while(0) | |
| 376 | |
|
1047
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
377 /* Blend the RGB values of two pixels based on a source alpha value */ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
378 #define ACCURATE_ALPHA_BLEND(sR, sG, sB, sA, dR, dG, dB) \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
379 do { \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
380 unsigned tR, tG, tB, tA; \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
381 tA = 255 - sA; \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
382 tR = 1 + (sR * sA) + (dR * tA); \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
383 dR = (tR + (tR >> 8)) >> 8; \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
384 tG = 1 + (sG * sA) + (dG * tA); \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
385 dG = (tG + (tG >> 8)) >> 8; \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
386 tB = 1 + (sB * sA) + (dB * tA); \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
387 dB = (tB + (tB >> 8)) >> 8; \ |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
388 } while(0) |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
389 |
|
ffaaf7ecf685
Altivec-optimized blitters!
Ryan C. Gordon <icculus@icculus.org>
parents:
769
diff
changeset
|
390 |
| 0 | 391 /* This is a very useful loop for optimizing blitters */ |
|
553
417f8709e648
There's a bug in the VC7 optimizer relating to the duff loop optimization
Sam Lantinga <slouken@libsdl.org>
parents:
535
diff
changeset
|
392 #if defined(_MSC_VER) && (_MSC_VER == 1300) |
|
417f8709e648
There's a bug in the VC7 optimizer relating to the duff loop optimization
Sam Lantinga <slouken@libsdl.org>
parents:
535
diff
changeset
|
393 /* There's a bug in the Visual C++ 7 optimizer when compiling this code */ |
|
417f8709e648
There's a bug in the VC7 optimizer relating to the duff loop optimization
Sam Lantinga <slouken@libsdl.org>
parents:
535
diff
changeset
|
394 #else |
| 0 | 395 #define USE_DUFFS_LOOP |
|
553
417f8709e648
There's a bug in the VC7 optimizer relating to the duff loop optimization
Sam Lantinga <slouken@libsdl.org>
parents:
535
diff
changeset
|
396 #endif |
| 0 | 397 #ifdef USE_DUFFS_LOOP |
| 398 | |
| 399 /* 8-times unrolled loop */ | |
| 400 #define DUFFS_LOOP8(pixel_copy_increment, width) \ | |
| 401 { int n = (width+7)/8; \ | |
|
91
e85e03f195b4
From: "Markus F.X.J. Oberhumer"
Sam Lantinga <slouken@lokigames.com>
parents:
0
diff
changeset
|
402 switch (width & 7) { \ |
| 0 | 403 case 0: do { pixel_copy_increment; \ |
| 404 case 7: pixel_copy_increment; \ | |
| 405 case 6: pixel_copy_increment; \ | |
| 406 case 5: pixel_copy_increment; \ | |
| 407 case 4: pixel_copy_increment; \ | |
| 408 case 3: pixel_copy_increment; \ | |
| 409 case 2: pixel_copy_increment; \ | |
| 410 case 1: pixel_copy_increment; \ | |
| 411 } while ( --n > 0 ); \ | |
| 412 } \ | |
| 413 } | |
| 414 | |
| 415 /* 4-times unrolled loop */ | |
| 416 #define DUFFS_LOOP4(pixel_copy_increment, width) \ | |
| 417 { int n = (width+3)/4; \ | |
|
91
e85e03f195b4
From: "Markus F.X.J. Oberhumer"
Sam Lantinga <slouken@lokigames.com>
parents:
0
diff
changeset
|
418 switch (width & 3) { \ |
| 0 | 419 case 0: do { pixel_copy_increment; \ |
| 420 case 3: pixel_copy_increment; \ | |
| 421 case 2: pixel_copy_increment; \ | |
| 422 case 1: pixel_copy_increment; \ | |
| 423 } while ( --n > 0 ); \ | |
| 424 } \ | |
| 425 } | |
| 426 | |
|
689
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
427 /* 2 - times unrolled loop */ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
428 #define DUFFS_LOOP_DOUBLE2(pixel_copy_increment, \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
429 double_pixel_copy_increment, width) \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
430 { int n, w = width; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
431 if( w & 1 ) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
432 pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
433 w--; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
434 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
435 if ( w > 0 ) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
436 n = ( w + 2) / 4; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
437 switch( w & 2 ) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
438 case 0: do { double_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
439 case 2: double_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
440 } while ( --n > 0 ); \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
441 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
442 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
443 } |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
444 |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
445 /* 2 - times unrolled loop 4 pixels */ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
446 #define DUFFS_LOOP_QUATRO2(pixel_copy_increment, \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
447 double_pixel_copy_increment, \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
448 quatro_pixel_copy_increment, width) \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
449 { int n, w = width; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
450 if(w & 1) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
451 pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
452 w--; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
453 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
454 if(w & 2) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
455 double_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
456 w -= 2; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
457 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
458 if ( w > 0 ) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
459 n = ( w + 7 ) / 8; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
460 switch( w & 4 ) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
461 case 0: do { quatro_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
462 case 4: quatro_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
463 } while ( --n > 0 ); \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
464 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
465 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
466 } |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
467 |
| 0 | 468 /* Use the 8-times version of the loop by default */ |
| 469 #define DUFFS_LOOP(pixel_copy_increment, width) \ | |
| 470 DUFFS_LOOP8(pixel_copy_increment, width) | |
| 471 | |
| 472 #else | |
| 473 | |
| 474 /* Don't use Duff's device to unroll loops */ | |
|
689
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
475 #define DUFFS_LOOP_DOUBLE2(pixel_copy_increment, \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
476 double_pixel_copy_increment, width) \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
477 { int n = width; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
478 if( n & 1 ) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
479 pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
480 n--; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
481 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
482 n=n>>1; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
483 for(; n > 0; --n) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
484 double_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
485 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
486 } |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
487 |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
488 /* Don't use Duff's device to unroll loops */ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
489 #define DUFFS_LOOP_QUATRO2(pixel_copy_increment, \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
490 double_pixel_copy_increment, \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
491 quatro_pixel_copy_increment, width) \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
492 { int n = width; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
493 if(n & 1) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
494 pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
495 n--; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
496 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
497 if(n & 2) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
498 double_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
499 n -= 2; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
500 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
501 n=n>>2; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
502 for(; n > 0; --n) { \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
503 quatro_pixel_copy_increment; \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
504 } \ |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
505 } |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
506 |
|
5bb080d35049
Date: Tue, 19 Aug 2003 17:57:00 +0200
Sam Lantinga <slouken@libsdl.org>
parents:
553
diff
changeset
|
507 /* Don't use Duff's device to unroll loops */ |
| 0 | 508 #define DUFFS_LOOP(pixel_copy_increment, width) \ |
| 509 { int n; \ | |
| 510 for ( n=width; n > 0; --n ) { \ | |
| 511 pixel_copy_increment; \ | |
| 512 } \ | |
| 513 } | |
| 514 #define DUFFS_LOOP8(pixel_copy_increment, width) \ | |
| 515 DUFFS_LOOP(pixel_copy_increment, width) | |
| 516 #define DUFFS_LOOP4(pixel_copy_increment, width) \ | |
| 517 DUFFS_LOOP(pixel_copy_increment, width) | |
| 518 | |
| 519 #endif /* USE_DUFFS_LOOP */ | |
| 520 | |
| 521 /* Prevent Visual C++ 6.0 from printing out stupid warnings */ | |
| 522 #if defined(_MSC_VER) && (_MSC_VER >= 600) | |
| 523 #pragma warning(disable: 4550) | |
| 524 #endif | |
| 525 | |
| 526 #endif /* _SDL_blit_h */ |