src/video/SDL_RLEaccel.c
author Ozkan Sezer <sezeroz@gmail.com>
Sun, 14 Oct 2018 15:25:04 +0300
branchSDL-1.2
changeset 12325 c4f2eeda176f
parent 6227 e11fd9609d10
permissions -rw-r--r--
remove symlink for libSDL-1.0.so.0 from the rpm spec file.

also fix a changelog date.
     1 /*
     2     SDL - Simple DirectMedia Layer
     3     Copyright (C) 1997-2012 Sam Lantinga
     4 
     5     This library is free software; you can redistribute it and/or
     6     modify it under the terms of the GNU Lesser General Public
     7     License as published by the Free Software Foundation; either
     8     version 2.1 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     Lesser General Public License for more details.
    14 
    15     You should have received a copy of the GNU Lesser General Public
    16     License along with this library; if not, write to the Free Software
    17     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
    18 
    19     Sam Lantinga
    20     slouken@libsdl.org
    21 */
    22 #include "SDL_config.h"
    23 
    24 /*
    25  * RLE encoding for software colorkey and alpha-channel acceleration
    26  *
    27  * Original version by Sam Lantinga
    28  *
    29  * Mattias Engdegård (Yorick): Rewrite. New encoding format, encoder and
    30  * decoder. Added per-surface alpha blitter. Added per-pixel alpha
    31  * format, encoder and blitter.
    32  *
    33  * Many thanks to Xark and johns for hints, benchmarks and useful comments
    34  * leading to this code.
    35  *
    36  * Welcome to Macro Mayhem.
    37  */
    38 
    39 /*
    40  * The encoding translates the image data to a stream of segments of the form
    41  *
    42  * <skip> <run> <data>
    43  *
    44  * where <skip> is the number of transparent pixels to skip,
    45  *       <run>  is the number of opaque pixels to blit,
    46  * and   <data> are the pixels themselves.
    47  *
    48  * This basic structure is used both for colorkeyed surfaces, used for simple
    49  * binary transparency and for per-surface alpha blending, and for surfaces
    50  * with per-pixel alpha. The details differ, however:
    51  *
    52  * Encoding of colorkeyed surfaces:
    53  *
    54  *   Encoded pixels always have the same format as the target surface.
    55  *   <skip> and <run> are unsigned 8 bit integers, except for 32 bit depth
    56  *   where they are 16 bit. This makes the pixel data aligned at all times.
    57  *   Segments never wrap around from one scan line to the next.
    58  *
    59  *   The end of the sequence is marked by a zero <skip>,<run> pair at the *
    60  *   beginning of a line.
    61  *
    62  * Encoding of surfaces with per-pixel alpha:
    63  *
    64  *   The sequence begins with a struct RLEDestFormat describing the target
    65  *   pixel format, to provide reliable un-encoding.
    66  *
    67  *   Each scan line is encoded twice: First all completely opaque pixels,
    68  *   encoded in the target format as described above, and then all
    69  *   partially transparent (translucent) pixels (where 1 <= alpha <= 254),
    70  *   in the following 32-bit format:
    71  *
    72  *   For 32-bit targets, each pixel has the target RGB format but with
    73  *   the alpha value occupying the highest 8 bits. The <skip> and <run>
    74  *   counts are 16 bit.
    75  * 
    76  *   For 16-bit targets, each pixel has the target RGB format, but with
    77  *   the middle component (usually green) shifted 16 steps to the left,
    78  *   and the hole filled with the 5 most significant bits of the alpha value.
    79  *   i.e. if the target has the format         rrrrrggggggbbbbb,
    80  *   the encoded pixel will be 00000gggggg00000rrrrr0aaaaabbbbb.
    81  *   The <skip> and <run> counts are 8 bit for the opaque lines, 16 bit
    82  *   for the translucent lines. Two padding bytes may be inserted
    83  *   before each translucent line to keep them 32-bit aligned.
    84  *
    85  *   The end of the sequence is marked by a zero <skip>,<run> pair at the
    86  *   beginning of an opaque line.
    87  */
    88 
    89 #include "SDL_video.h"
    90 #include "SDL_sysvideo.h"
    91 #include "SDL_blit.h"
    92 #include "SDL_RLEaccel_c.h"
    93 
    94 /* Force MMX to 0; this blows up on almost every major compiler now. --ryan. */
    95 #if 0 && defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__)) && SDL_ASSEMBLY_ROUTINES
    96 #define MMX_ASMBLIT
    97 #endif
    98 
    99 #ifdef MMX_ASMBLIT
   100 #include "mmx.h"
   101 #include "SDL_cpuinfo.h"
   102 #endif
   103 
   104 #ifndef MAX
   105 #define MAX(a, b) ((a) > (b) ? (a) : (b))
   106 #endif
   107 #ifndef MIN
   108 #define MIN(a, b) ((a) < (b) ? (a) : (b))
   109 #endif
   110 
   111 #define PIXEL_COPY(to, from, len, bpp)			\
   112 do {							\
   113     if(bpp == 4) {					\
   114 	SDL_memcpy4(to, from, (size_t)(len));		\
   115     } else {						\
   116 	SDL_memcpy(to, from, (size_t)(len) * (bpp));	\
   117     }							\
   118 } while(0)
   119 
   120 /*
   121  * Various colorkey blit methods, for opaque and per-surface alpha
   122  */
   123 
   124 #define OPAQUE_BLIT(to, from, length, bpp, alpha)	\
   125     PIXEL_COPY(to, from, length, bpp)
   126 
   127 #ifdef MMX_ASMBLIT
   128 
   129 #define ALPHA_BLIT32_888MMX(to, from, length, bpp, alpha)	\
   130     do {							\
   131 	Uint32 *srcp = (Uint32 *)(from);			\
   132 	Uint32 *dstp = (Uint32 *)(to);				\
   133         int i = 0x00FF00FF;					\
   134         movd_m2r(*(&i), mm3);					\
   135         punpckldq_r2r(mm3, mm3);				\
   136         i = 0xFF000000;						\
   137         movd_m2r(*(&i), mm7);					\
   138         punpckldq_r2r(mm7, mm7);				\
   139         i = alpha | alpha << 16;				\
   140         movd_m2r(*(&i), mm4);					\
   141         punpckldq_r2r(mm4, mm4);				\
   142 	pcmpeqd_r2r(mm5,mm5); /* set mm5 to "1" */		\
   143 	pxor_r2r(mm7, mm5); /* make clear alpha mask */		\
   144         i = length;						\
   145 	if(i & 1) {						\
   146           movd_m2r((*srcp), mm1); /* src -> mm1 */		\
   147           punpcklbw_r2r(mm1, mm1);				\
   148           pand_r2r(mm3, mm1);					\
   149 	  movd_m2r((*dstp), mm2); /* dst -> mm2 */		\
   150           punpcklbw_r2r(mm2, mm2);				\
   151           pand_r2r(mm3, mm2);					\
   152 	  psubw_r2r(mm2, mm1);					\
   153 	  pmullw_r2r(mm4, mm1);					\
   154 	  psrlw_i2r(8, mm1);					\
   155 	  paddw_r2r(mm1, mm2);					\
   156 	  pand_r2r(mm3, mm2);					\
   157 	  packuswb_r2r(mm2, mm2);				\
   158 	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
   159 	  movd_r2m(mm2, *dstp);					\
   160 	  ++srcp;						\
   161 	  ++dstp;						\
   162 	  i--;							\
   163 	}							\
   164 	for(; i > 0; --i) {					\
   165           movq_m2r((*srcp), mm0);				\
   166 	  movq_r2r(mm0, mm1);					\
   167           punpcklbw_r2r(mm0, mm0);				\
   168 	  movq_m2r((*dstp), mm2);				\
   169 	  punpckhbw_r2r(mm1, mm1);				\
   170 	  movq_r2r(mm2, mm6);					\
   171           pand_r2r(mm3, mm0);					\
   172           punpcklbw_r2r(mm2, mm2);				\
   173 	  pand_r2r(mm3, mm1);					\
   174 	  punpckhbw_r2r(mm6, mm6);				\
   175           pand_r2r(mm3, mm2);					\
   176 	  psubw_r2r(mm2, mm0);					\
   177 	  pmullw_r2r(mm4, mm0);					\
   178 	  pand_r2r(mm3, mm6);					\
   179 	  psubw_r2r(mm6, mm1);					\
   180 	  pmullw_r2r(mm4, mm1);					\
   181 	  psrlw_i2r(8, mm0);					\
   182 	  paddw_r2r(mm0, mm2);					\
   183 	  psrlw_i2r(8, mm1);					\
   184 	  paddw_r2r(mm1, mm6);					\
   185 	  pand_r2r(mm3, mm2);					\
   186 	  pand_r2r(mm3, mm6);					\
   187 	  packuswb_r2r(mm2, mm2);				\
   188 	  packuswb_r2r(mm6, mm6);				\
   189 	  psrlq_i2r(32, mm2);					\
   190 	  psllq_i2r(32, mm6);					\
   191 	  por_r2r(mm6, mm2);					\
   192 	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
   193          movq_r2m(mm2, *dstp);					\
   194 	  srcp += 2;						\
   195 	  dstp += 2;						\
   196 	  i--;							\
   197 	}							\
   198 	emms();							\
   199     } while(0)
   200 
   201 #define ALPHA_BLIT16_565MMX(to, from, length, bpp, alpha)	\
   202     do {						\
   203         int i, n = 0;					\
   204 	Uint16 *srcp = (Uint16 *)(from);		\
   205 	Uint16 *dstp = (Uint16 *)(to);			\
   206         Uint32 ALPHA = 0xF800;				\
   207 	movd_m2r(*(&ALPHA), mm1);			\
   208         punpcklwd_r2r(mm1, mm1);			\
   209         punpcklwd_r2r(mm1, mm1);			\
   210 	ALPHA = 0x07E0;					\
   211 	movd_m2r(*(&ALPHA), mm4);			\
   212         punpcklwd_r2r(mm4, mm4);			\
   213         punpcklwd_r2r(mm4, mm4);			\
   214 	ALPHA = 0x001F;					\
   215 	movd_m2r(*(&ALPHA), mm7);			\
   216         punpcklwd_r2r(mm7, mm7);			\
   217         punpcklwd_r2r(mm7, mm7);			\
   218 	alpha &= ~(1+2+4);				\
   219         i = (Uint32)alpha | (Uint32)alpha << 16;	\
   220         movd_m2r(*(&i), mm0);				\
   221         punpckldq_r2r(mm0, mm0);			\
   222         ALPHA = alpha >> 3;				\
   223         i = ((int)(length) & 3);			\
   224 	for(; i > 0; --i) {				\
   225 	    Uint32 s = *srcp++;				\
   226 	    Uint32 d = *dstp;				\
   227 	    s = (s | s << 16) & 0x07e0f81f;		\
   228 	    d = (d | d << 16) & 0x07e0f81f;		\
   229 	    d += (s - d) * ALPHA >> 5;			\
   230 	    d &= 0x07e0f81f;				\
   231 	    *dstp++ = d | d >> 16;			\
   232 	    n++;					\
   233 	}						\
   234 	i = (int)(length) - n;				\
   235 	for(; i > 0; --i) {				\
   236 	  movq_m2r((*dstp), mm3);			\
   237 	  movq_m2r((*srcp), mm2);			\
   238 	  movq_r2r(mm2, mm5);				\
   239 	  pand_r2r(mm1 , mm5);				\
   240 	  psrlq_i2r(11, mm5);				\
   241 	  movq_r2r(mm3, mm6);				\
   242 	  pand_r2r(mm1 , mm6);				\
   243 	  psrlq_i2r(11, mm6);				\
   244 	  psubw_r2r(mm6, mm5);				\
   245 	  pmullw_r2r(mm0, mm5);				\
   246 	  psrlw_i2r(8, mm5);				\
   247 	  paddw_r2r(mm5, mm6);				\
   248 	  psllq_i2r(11, mm6);				\
   249 	  pand_r2r(mm1, mm6);				\
   250 	  movq_r2r(mm4, mm5);				\
   251 	  por_r2r(mm7, mm5);				\
   252 	  pand_r2r(mm5, mm3);				\
   253 	  por_r2r(mm6, mm3);				\
   254 	  movq_r2r(mm2, mm5);				\
   255 	  pand_r2r(mm4 , mm5);				\
   256 	  psrlq_i2r(5, mm5);				\
   257 	  movq_r2r(mm3, mm6);				\
   258 	  pand_r2r(mm4 , mm6);				\
   259 	  psrlq_i2r(5, mm6);				\
   260 	  psubw_r2r(mm6, mm5);				\
   261 	  pmullw_r2r(mm0, mm5);				\
   262 	  psrlw_i2r(8, mm5);				\
   263 	  paddw_r2r(mm5, mm6);				\
   264 	  psllq_i2r(5, mm6);				\
   265 	  pand_r2r(mm4, mm6);				\
   266 	  movq_r2r(mm1, mm5);				\
   267 	  por_r2r(mm7, mm5);				\
   268 	  pand_r2r(mm5, mm3);				\
   269 	  por_r2r(mm6, mm3);				\
   270 	  movq_r2r(mm2, mm5);				\
   271 	  pand_r2r(mm7 , mm5);				\
   272           movq_r2r(mm3, mm6);				\
   273 	  pand_r2r(mm7 , mm6);				\
   274 	  psubw_r2r(mm6, mm5);				\
   275 	  pmullw_r2r(mm0, mm5);				\
   276 	  psrlw_i2r(8, mm5);				\
   277 	  paddw_r2r(mm5, mm6);				\
   278 	  pand_r2r(mm7, mm6);				\
   279 	  movq_r2r(mm1, mm5);				\
   280 	  por_r2r(mm4, mm5);				\
   281 	  pand_r2r(mm5, mm3);				\
   282 	  por_r2r(mm6, mm3);				\
   283 	  movq_r2m(mm3, *dstp);				\
   284 	  srcp += 4;					\
   285 	  dstp += 4;					\
   286 	  i -= 3;					\
   287 	}						\
   288 	emms();						\
   289     } while(0)
   290 
   291 #define ALPHA_BLIT16_555MMX(to, from, length, bpp, alpha)	\
   292     do {						\
   293         int i, n = 0;					\
   294 	Uint16 *srcp = (Uint16 *)(from);		\
   295 	Uint16 *dstp = (Uint16 *)(to);			\
   296         Uint32 ALPHA = 0x7C00;				\
   297 	movd_m2r(*(&ALPHA), mm1);			\
   298         punpcklwd_r2r(mm1, mm1);			\
   299         punpcklwd_r2r(mm1, mm1);			\
   300 	ALPHA = 0x03E0;					\
   301         movd_m2r(*(&ALPHA), mm4);			\
   302         punpcklwd_r2r(mm4, mm4);			\
   303         punpcklwd_r2r(mm4, mm4);			\
   304 	ALPHA = 0x001F;					\
   305 	movd_m2r(*(&ALPHA), mm7);			\
   306         punpcklwd_r2r(mm7, mm7);			\
   307         punpcklwd_r2r(mm7, mm7);			\
   308 	alpha &= ~(1+2+4);				\
   309         i = (Uint32)alpha | (Uint32)alpha << 16;	\
   310         movd_m2r(*(&i), mm0);				\
   311         punpckldq_r2r(mm0, mm0);			\
   312         i = ((int)(length) & 3);				\
   313         ALPHA = alpha >> 3;				\
   314 	for(; i > 0; --i) {				\
   315 	    Uint32 s = *srcp++;				\
   316 	    Uint32 d = *dstp;				\
   317 	    s = (s | s << 16) & 0x03e07c1f;		\
   318 	    d = (d | d << 16) & 0x03e07c1f;		\
   319 	    d += (s - d) * ALPHA >> 5;			\
   320 	    d &= 0x03e07c1f;				\
   321 	    *dstp++ = d | d >> 16;			\
   322 	    n++;					\
   323 	}						\
   324 	i = (int)(length) - n;				\
   325 	for(; i > 0; --i) {				\
   326 	  movq_m2r((*dstp), mm3);			\
   327 	  movq_m2r((*srcp), mm2);			\
   328 	  movq_r2r(mm2, mm5);				\
   329 	  pand_r2r(mm1 , mm5);				\
   330 	  psrlq_i2r(10, mm5);				\
   331 	  movq_r2r(mm3, mm6);				\
   332 	  pand_r2r(mm1 , mm6);				\
   333 	  psrlq_i2r(10, mm6);				\
   334 	  psubw_r2r(mm6, mm5);				\
   335 	  pmullw_r2r(mm0, mm5);				\
   336 	  psrlw_i2r(8, mm5);				\
   337 	  paddw_r2r(mm5, mm6);				\
   338 	  psllq_i2r(10, mm6);				\
   339 	  pand_r2r(mm1, mm6);				\
   340 	  movq_r2r(mm4, mm5);				\
   341 	  por_r2r(mm7, mm5);				\
   342 	  pand_r2r(mm5, mm3);				\
   343 	  por_r2r(mm6, mm3);				\
   344 	  movq_r2r(mm2, mm5);				\
   345 	  pand_r2r(mm4 , mm5);				\
   346 	  psrlq_i2r(5, mm5);				\
   347 	  movq_r2r(mm3, mm6);				\
   348 	  pand_r2r(mm4 , mm6);				\
   349 	  psrlq_i2r(5, mm6);				\
   350 	  psubw_r2r(mm6, mm5);				\
   351 	  pmullw_r2r(mm0, mm5);				\
   352 	  psrlw_i2r(8, mm5);				\
   353 	  paddw_r2r(mm5, mm6);				\
   354 	  psllq_i2r(5, mm6);				\
   355 	  pand_r2r(mm4, mm6);				\
   356 	  movq_r2r(mm1, mm5);				\
   357 	  por_r2r(mm7, mm5);				\
   358 	  pand_r2r(mm5, mm3);				\
   359 	  por_r2r(mm6, mm3);				\
   360 	  movq_r2r(mm2, mm5);				\
   361 	  pand_r2r(mm7 , mm5);				\
   362           movq_r2r(mm3, mm6);				\
   363 	  pand_r2r(mm7 , mm6);				\
   364 	  psubw_r2r(mm6, mm5);				\
   365 	  pmullw_r2r(mm0, mm5);				\
   366 	  psrlw_i2r(8, mm5);				\
   367 	  paddw_r2r(mm5, mm6);				\
   368 	  pand_r2r(mm7, mm6);				\
   369 	  movq_r2r(mm1, mm5);				\
   370 	  por_r2r(mm4, mm5);				\
   371 	  pand_r2r(mm5, mm3);				\
   372 	  por_r2r(mm6, mm3);				\
   373 	  movq_r2m(mm3, *dstp);				\
   374 	  srcp += 4;					\
   375 	  dstp += 4;					\
   376 	  i -= 3;					\
   377 	}						\
   378 	emms();						\
   379     } while(0)
   380 
   381 #endif
   382 
   383 /*
   384  * For 32bpp pixels on the form 0x00rrggbb:
   385  * If we treat the middle component separately, we can process the two
   386  * remaining in parallel. This is safe to do because of the gap to the left
   387  * of each component, so the bits from the multiplication don't collide.
   388  * This can be used for any RGB permutation of course.
   389  */
   390 #define ALPHA_BLIT32_888(to, from, length, bpp, alpha)		\
   391     do {							\
   392         int i;							\
   393 	Uint32 *src = (Uint32 *)(from);				\
   394 	Uint32 *dst = (Uint32 *)(to);				\
   395 	for(i = 0; i < (int)(length); i++) {			\
   396 	    Uint32 s = *src++;					\
   397 	    Uint32 d = *dst;					\
   398 	    Uint32 s1 = s & 0xff00ff;				\
   399 	    Uint32 d1 = d & 0xff00ff;				\
   400 	    d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;	\
   401 	    s &= 0xff00;					\
   402 	    d &= 0xff00;					\
   403 	    d = (d + ((s - d) * alpha >> 8)) & 0xff00;		\
   404 	    *dst++ = d1 | d;					\
   405 	}							\
   406     } while(0)
   407 
   408 /*
   409  * For 16bpp pixels we can go a step further: put the middle component
   410  * in the high 16 bits of a 32 bit word, and process all three RGB
   411  * components at the same time. Since the smallest gap is here just
   412  * 5 bits, we have to scale alpha down to 5 bits as well.
   413  */
   414 #define ALPHA_BLIT16_565(to, from, length, bpp, alpha)	\
   415     do {						\
   416         int i;						\
   417 	Uint16 *src = (Uint16 *)(from);			\
   418 	Uint16 *dst = (Uint16 *)(to);			\
   419 	Uint32 ALPHA = alpha >> 3;			\
   420 	for(i = 0; i < (int)(length); i++) {		\
   421 	    Uint32 s = *src++;				\
   422 	    Uint32 d = *dst;				\
   423 	    s = (s | s << 16) & 0x07e0f81f;		\
   424 	    d = (d | d << 16) & 0x07e0f81f;		\
   425 	    d += (s - d) * ALPHA >> 5;			\
   426 	    d &= 0x07e0f81f;				\
   427 	    *dst++ = (Uint16)(d | d >> 16);			\
   428 	}						\
   429     } while(0)
   430 
   431 #define ALPHA_BLIT16_555(to, from, length, bpp, alpha)	\
   432     do {						\
   433         int i;						\
   434 	Uint16 *src = (Uint16 *)(from);			\
   435 	Uint16 *dst = (Uint16 *)(to);			\
   436 	Uint32 ALPHA = alpha >> 3;			\
   437 	for(i = 0; i < (int)(length); i++) {		\
   438 	    Uint32 s = *src++;				\
   439 	    Uint32 d = *dst;				\
   440 	    s = (s | s << 16) & 0x03e07c1f;		\
   441 	    d = (d | d << 16) & 0x03e07c1f;		\
   442 	    d += (s - d) * ALPHA >> 5;			\
   443 	    d &= 0x03e07c1f;				\
   444 	    *dst++ = (Uint16)(d | d >> 16);			\
   445 	}						\
   446     } while(0)
   447 
   448 /*
   449  * The general slow catch-all function, for remaining depths and formats
   450  */
   451 #define ALPHA_BLIT_ANY(to, from, length, bpp, alpha)			\
   452     do {								\
   453         int i;								\
   454 	Uint8 *src = from;						\
   455 	Uint8 *dst = to;						\
   456 	for(i = 0; i < (int)(length); i++) {				\
   457 	    Uint32 s, d;						\
   458 	    unsigned rs, gs, bs, rd, gd, bd;				\
   459 	    switch(bpp) {						\
   460 	    case 2:							\
   461 		s = *(Uint16 *)src;					\
   462 		d = *(Uint16 *)dst;					\
   463 		break;							\
   464 	    case 3:							\
   465 		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
   466 		    s = (src[0] << 16) | (src[1] << 8) | src[2];	\
   467 		    d = (dst[0] << 16) | (dst[1] << 8) | dst[2];	\
   468 		} else {						\
   469 		    s = (src[2] << 16) | (src[1] << 8) | src[0];	\
   470 		    d = (dst[2] << 16) | (dst[1] << 8) | dst[0];	\
   471 		}							\
   472 		break;							\
   473 	    case 4:							\
   474 		s = *(Uint32 *)src;					\
   475 		d = *(Uint32 *)dst;					\
   476 		break;							\
   477 	    }								\
   478 	    RGB_FROM_PIXEL(s, fmt, rs, gs, bs);				\
   479 	    RGB_FROM_PIXEL(d, fmt, rd, gd, bd);				\
   480 	    rd += (rs - rd) * alpha >> 8;				\
   481 	    gd += (gs - gd) * alpha >> 8;				\
   482 	    bd += (bs - bd) * alpha >> 8;				\
   483 	    PIXEL_FROM_RGB(d, fmt, rd, gd, bd);				\
   484 	    switch(bpp) {						\
   485 	    case 2:							\
   486 		*(Uint16 *)dst = (Uint16)d;					\
   487 		break;							\
   488 	    case 3:							\
   489 		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
   490 		    dst[0] = (Uint8)(d >> 16);					\
   491 		    dst[1] = (Uint8)(d >> 8);					\
   492 		    dst[2] = (Uint8)(d);						\
   493 		} else {						\
   494 		    dst[0] = (Uint8)d;						\
   495 		    dst[1] = (Uint8)(d >> 8);					\
   496 		    dst[2] = (Uint8)(d >> 16);					\
   497 		}							\
   498 		break;							\
   499 	    case 4:							\
   500 		*(Uint32 *)dst = d;					\
   501 		break;							\
   502 	    }								\
   503 	    src += bpp;							\
   504 	    dst += bpp;							\
   505 	}								\
   506     } while(0)
   507 
   508 #ifdef MMX_ASMBLIT
   509 
   510 #define ALPHA_BLIT32_888_50MMX(to, from, length, bpp, alpha)		\
   511     do {								\
   512 	Uint32 *srcp = (Uint32 *)(from);				\
   513 	Uint32 *dstp = (Uint32 *)(to);					\
   514         int i = 0x00fefefe;						\
   515         movd_m2r(*(&i), mm4);						\
   516         punpckldq_r2r(mm4, mm4);					\
   517         i = 0x00010101;							\
   518         movd_m2r(*(&i), mm3);						\
   519         punpckldq_r2r(mm3, mm3);					\
   520         i = (int)(length);						\
   521         if( i & 1 ) {							\
   522 	  Uint32 s = *srcp++;						\
   523 	  Uint32 d = *dstp;						\
   524 	  *dstp++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
   525 		     + (s & d & 0x00010101);				\
   526 	  i--;								\
   527 	}								\
   528 	for(; i > 0; --i) {						\
   529 	    movq_m2r((*dstp), mm2); /* dst -> mm2 */			\
   530 	    movq_r2r(mm2, mm6);	/* dst -> mm6 */			\
   531 	    movq_m2r((*srcp), mm1); /* src -> mm1 */			\
   532 	    movq_r2r(mm1, mm5);	/* src -> mm5 */			\
   533 	    pand_r2r(mm4, mm6);	/* dst & 0x00fefefe -> mm6 */		\
   534 	    pand_r2r(mm4, mm5); /* src & 0x00fefefe -> mm5 */		\
   535 	    paddd_r2r(mm6, mm5); /* (dst & 0x00fefefe) + (dst & 0x00fefefe) -> mm5 */	\
   536 	    psrld_i2r(1, mm5);						\
   537 	    pand_r2r(mm1, mm2);	/* s & d -> mm2 */			\
   538 	    pand_r2r(mm3, mm2);	/* s & d & 0x00010101 -> mm2 */		\
   539 	    paddd_r2r(mm5, mm2);					\
   540 	    movq_r2m(mm2, (*dstp));					\
   541 	    dstp += 2;							\
   542 	    srcp += 2;							\
   543 	    i--;							\
   544 	}								\
   545 	emms();								\
   546     } while(0)
   547 
   548 #endif
   549     
   550 /*
   551  * Special case: 50% alpha (alpha=128)
   552  * This is treated specially because it can be optimized very well, and
   553  * since it is good for many cases of semi-translucency.
   554  * The theory is to do all three components at the same time:
   555  * First zero the lowest bit of each component, which gives us room to
   556  * add them. Then shift right and add the sum of the lowest bits.
   557  */
   558 #define ALPHA_BLIT32_888_50(to, from, length, bpp, alpha)		\
   559     do {								\
   560         int i;								\
   561 	Uint32 *src = (Uint32 *)(from);					\
   562 	Uint32 *dst = (Uint32 *)(to);					\
   563 	for(i = 0; i < (int)(length); i++) {				\
   564 	    Uint32 s = *src++;						\
   565 	    Uint32 d = *dst;						\
   566 	    *dst++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
   567 		     + (s & d & 0x00010101);				\
   568 	}								\
   569     } while(0)
   570 
   571 /*
   572  * For 16bpp, we can actually blend two pixels in parallel, if we take
   573  * care to shift before we add, not after.
   574  */
   575 
   576 /* helper: blend a single 16 bit pixel at 50% */
   577 #define BLEND16_50(dst, src, mask)			\
   578     do {						\
   579 	Uint32 s = *src++;				\
   580 	Uint32 d = *dst;				\
   581 	*dst++ = (Uint16)((((s & mask) + (d & mask)) >> 1) +	\
   582 	                  (s & d & (~mask & 0xffff)));		\
   583     } while(0)
   584 
   585 /* basic 16bpp blender. mask is the pixels to keep when adding. */
   586 #define ALPHA_BLIT16_50(to, from, length, bpp, alpha, mask)		\
   587     do {								\
   588 	unsigned n = (length);						\
   589 	Uint16 *src = (Uint16 *)(from);					\
   590 	Uint16 *dst = (Uint16 *)(to);					\
   591 	if(((uintptr_t)src ^ (uintptr_t)dst) & 3) {			\
   592 	    /* source and destination not in phase, blit one by one */	\
   593 	    while(n--)							\
   594 		BLEND16_50(dst, src, mask);				\
   595 	} else {							\
   596 	    if((uintptr_t)src & 3) {					\
   597 		/* first odd pixel */					\
   598 		BLEND16_50(dst, src, mask);				\
   599 		n--;							\
   600 	    }								\
   601 	    for(; n > 1; n -= 2) {					\
   602 		Uint32 s = *(Uint32 *)src;				\
   603 		Uint32 d = *(Uint32 *)dst;				\
   604 		*(Uint32 *)dst = ((s & (mask | mask << 16)) >> 1)	\
   605 		               + ((d & (mask | mask << 16)) >> 1)	\
   606 		               + (s & d & (~(mask | mask << 16)));	\
   607 		src += 2;						\
   608 		dst += 2;						\
   609 	    }								\
   610 	    if(n)							\
   611 		BLEND16_50(dst, src, mask); /* last odd pixel */	\
   612 	}								\
   613     } while(0)
   614 
   615 #define ALPHA_BLIT16_565_50(to, from, length, bpp, alpha)	\
   616     ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xf7de)
   617 
   618 #define ALPHA_BLIT16_555_50(to, from, length, bpp, alpha)	\
   619     ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xfbde)
   620 
   621 #ifdef MMX_ASMBLIT
   622 
   623 #define CHOOSE_BLIT(blitter, alpha, fmt)				\
   624     do {								\
   625         if(alpha == 255) {						\
   626 	    switch(fmt->BytesPerPixel) {				\
   627 	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
   628 	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
   629 	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
   630 	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
   631 	    }								\
   632 	} else {							\
   633 	    switch(fmt->BytesPerPixel) {				\
   634 	    case 1:							\
   635 		/* No 8bpp alpha blitting */				\
   636 		break;							\
   637 									\
   638 	    case 2:							\
   639 		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
   640 		case 0xffff:						\
   641 		    if(fmt->Gmask == 0x07e0				\
   642 		       || fmt->Rmask == 0x07e0				\
   643 		       || fmt->Bmask == 0x07e0) {			\
   644 			if(alpha == 128)				\
   645 			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
   646 			else {						\
   647 			    if(SDL_HasMMX())				\
   648 				blitter(2, Uint8, ALPHA_BLIT16_565MMX);	\
   649 			    else					\
   650 				blitter(2, Uint8, ALPHA_BLIT16_565);	\
   651 			}						\
   652 		    } else						\
   653 			goto general16;					\
   654 		    break;						\
   655 									\
   656 		case 0x7fff:						\
   657 		    if(fmt->Gmask == 0x03e0				\
   658 		       || fmt->Rmask == 0x03e0				\
   659 		       || fmt->Bmask == 0x03e0) {			\
   660 			if(alpha == 128)				\
   661 			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
   662 			else {						\
   663 			    if(SDL_HasMMX())				\
   664 				blitter(2, Uint8, ALPHA_BLIT16_555MMX);	\
   665 			    else					\
   666 				blitter(2, Uint8, ALPHA_BLIT16_555);	\
   667 			}						\
   668 			break;						\
   669 		    }							\
   670 		    /* fallthrough */					\
   671 									\
   672 		default:						\
   673 		general16:						\
   674 		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
   675 		}							\
   676 		break;							\
   677 									\
   678 	    case 3:							\
   679 		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
   680 		break;							\
   681 									\
   682 	    case 4:							\
   683 		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
   684 		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
   685 		       || fmt->Bmask == 0xff00)) {			\
   686 		    if(alpha == 128)					\
   687 		    {							\
   688 			if(SDL_HasMMX())				\
   689 				blitter(4, Uint16, ALPHA_BLIT32_888_50MMX);\
   690 			else						\
   691 				blitter(4, Uint16, ALPHA_BLIT32_888_50);\
   692 		    }							\
   693 		    else						\
   694 		    {							\
   695 			if(SDL_HasMMX())				\
   696 				blitter(4, Uint16, ALPHA_BLIT32_888MMX);\
   697 			else						\
   698 				blitter(4, Uint16, ALPHA_BLIT32_888);	\
   699 		    }							\
   700 		} else							\
   701 		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
   702 		break;							\
   703 	    }								\
   704 	}								\
   705     } while(0)
   706 
   707 #else
   708 	
   709 #define CHOOSE_BLIT(blitter, alpha, fmt)				\
   710     do {								\
   711         if(alpha == 255) {						\
   712 	    switch(fmt->BytesPerPixel) {				\
   713 	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
   714 	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
   715 	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
   716 	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
   717 	    }								\
   718 	} else {							\
   719 	    switch(fmt->BytesPerPixel) {				\
   720 	    case 1:							\
   721 		/* No 8bpp alpha blitting */				\
   722 		break;							\
   723 									\
   724 	    case 2:							\
   725 		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
   726 		case 0xffff:						\
   727 		    if(fmt->Gmask == 0x07e0				\
   728 		       || fmt->Rmask == 0x07e0				\
   729 		       || fmt->Bmask == 0x07e0) {			\
   730 			if(alpha == 128)				\
   731 			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
   732 			else {						\
   733 			    blitter(2, Uint8, ALPHA_BLIT16_565);	\
   734 			}						\
   735 		    } else						\
   736 			goto general16;					\
   737 		    break;						\
   738 									\
   739 		case 0x7fff:						\
   740 		    if(fmt->Gmask == 0x03e0				\
   741 		       || fmt->Rmask == 0x03e0				\
   742 		       || fmt->Bmask == 0x03e0) {			\
   743 			if(alpha == 128)				\
   744 			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
   745 			else {						\
   746 			    blitter(2, Uint8, ALPHA_BLIT16_555);	\
   747 			}						\
   748 			break;						\
   749 		    }							\
   750 		    /* fallthrough */					\
   751 									\
   752 		default:						\
   753 		general16:						\
   754 		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
   755 		}							\
   756 		break;							\
   757 									\
   758 	    case 3:							\
   759 		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
   760 		break;							\
   761 									\
   762 	    case 4:							\
   763 		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
   764 		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
   765 		       || fmt->Bmask == 0xff00)) {			\
   766 		    if(alpha == 128)					\
   767 			blitter(4, Uint16, ALPHA_BLIT32_888_50);	\
   768 		    else						\
   769 			blitter(4, Uint16, ALPHA_BLIT32_888);		\
   770 		} else							\
   771 		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
   772 		break;							\
   773 	    }								\
   774 	}								\
   775     } while(0)
   776 
   777 #endif
   778 
   779 /*
   780  * This takes care of the case when the surface is clipped on the left and/or
   781  * right. Top clipping has already been taken care of.
   782  */
   783 static void RLEClipBlit(int w, Uint8 *srcbuf, SDL_Surface *dst,
   784 			Uint8 *dstbuf, SDL_Rect *srcrect, unsigned alpha)
   785 {
   786     SDL_PixelFormat *fmt = dst->format;
   787 
   788 #define RLECLIPBLIT(bpp, Type, do_blit)					   \
   789     do {								   \
   790 	int linecount = srcrect->h;					   \
   791 	int ofs = 0;							   \
   792 	int left = srcrect->x;						   \
   793 	int right = left + srcrect->w;					   \
   794 	dstbuf -= left * bpp;						   \
   795 	for(;;) {							   \
   796 	    int run;							   \
   797 	    ofs += *(Type *)srcbuf;					   \
   798 	    run = ((Type *)srcbuf)[1];					   \
   799 	    srcbuf += 2 * sizeof(Type);					   \
   800 	    if(run) {							   \
   801 		/* clip to left and right borders */			   \
   802 		if(ofs < right) {					   \
   803 		    int start = 0;					   \
   804 		    int len = run;					   \
   805 		    int startcol;					   \
   806 		    if(left - ofs > 0) {				   \
   807 			start = left - ofs;				   \
   808 			len -= start;					   \
   809 			if(len <= 0)					   \
   810 			    goto nocopy ## bpp ## do_blit;		   \
   811 		    }							   \
   812 		    startcol = ofs + start;				   \
   813 		    if(len > right - startcol)				   \
   814 			len = right - startcol;				   \
   815 		    do_blit(dstbuf + startcol * bpp, srcbuf + start * bpp, \
   816 			    len, bpp, alpha);				   \
   817 		}							   \
   818 	    nocopy ## bpp ## do_blit:					   \
   819 		srcbuf += run * bpp;					   \
   820 		ofs += run;						   \
   821 	    } else if(!ofs)						   \
   822 		break;							   \
   823 	    if(ofs == w) {						   \
   824 		ofs = 0;						   \
   825 		dstbuf += dst->pitch;					   \
   826 		if(!--linecount)					   \
   827 		    break;						   \
   828 	    }								   \
   829 	}								   \
   830     } while(0)
   831 
   832     CHOOSE_BLIT(RLECLIPBLIT, alpha, fmt);
   833 
   834 #undef RLECLIPBLIT
   835 
   836 }
   837 
   838 
   839 /* blit a colorkeyed RLE surface */
   840 int SDL_RLEBlit(SDL_Surface *src, SDL_Rect *srcrect,
   841 		SDL_Surface *dst, SDL_Rect *dstrect)
   842 {
   843 	Uint8 *dstbuf;
   844 	Uint8 *srcbuf;
   845 	int x, y;
   846 	int w = src->w;
   847 	unsigned alpha;
   848 
   849 	/* Lock the destination if necessary */
   850 	if ( SDL_MUSTLOCK(dst) ) {
   851 		if ( SDL_LockSurface(dst) < 0 ) {
   852 			return(-1);
   853 		}
   854 	}
   855 
   856 	/* Set up the source and destination pointers */
   857 	x = dstrect->x;
   858 	y = dstrect->y;
   859 	dstbuf = (Uint8 *)dst->pixels
   860 	         + y * dst->pitch + x * src->format->BytesPerPixel;
   861 	srcbuf = (Uint8 *)src->map->sw_data->aux_data;
   862 
   863 	{
   864 	    /* skip lines at the top if neccessary */
   865 	    int vskip = srcrect->y;
   866 	    int ofs = 0;
   867 	    if(vskip) {
   868 
   869 #define RLESKIP(bpp, Type)			\
   870 		for(;;) {			\
   871 		    int run;			\
   872 		    ofs += *(Type *)srcbuf;	\
   873 		    run = ((Type *)srcbuf)[1];	\
   874 		    srcbuf += sizeof(Type) * 2;	\
   875 		    if(run) {			\
   876 			srcbuf += run * bpp;	\
   877 			ofs += run;		\
   878 		    } else if(!ofs)		\
   879 			goto done;		\
   880 		    if(ofs == w) {		\
   881 			ofs = 0;		\
   882 			if(!--vskip)		\
   883 			    break;		\
   884 		    }				\
   885 		}
   886 
   887 		switch(src->format->BytesPerPixel) {
   888 		case 1: RLESKIP(1, Uint8); break;
   889 		case 2: RLESKIP(2, Uint8); break;
   890 		case 3: RLESKIP(3, Uint8); break;
   891 		case 4: RLESKIP(4, Uint16); break;
   892 		}
   893 
   894 #undef RLESKIP
   895 
   896 	    }
   897 	}
   898 
   899 	alpha = (src->flags & SDL_SRCALPHA) == SDL_SRCALPHA
   900 	        ? src->format->alpha : 255;
   901 	/* if left or right edge clipping needed, call clip blit */
   902 	if ( srcrect->x || srcrect->w != src->w ) {
   903 	    RLEClipBlit(w, srcbuf, dst, dstbuf, srcrect, alpha);
   904 	} else {
   905 	    SDL_PixelFormat *fmt = src->format;
   906 
   907 #define RLEBLIT(bpp, Type, do_blit)					      \
   908 	    do {							      \
   909 		int linecount = srcrect->h;				      \
   910 		int ofs = 0;						      \
   911 		for(;;) {						      \
   912 		    unsigned run;					      \
   913 		    ofs += *(Type *)srcbuf;				      \
   914 		    run = ((Type *)srcbuf)[1];				      \
   915 		    srcbuf += 2 * sizeof(Type);				      \
   916 		    if(run) {						      \
   917 			do_blit(dstbuf + ofs * bpp, srcbuf, run, bpp, alpha); \
   918 			srcbuf += run * bpp;				      \
   919 			ofs += run;					      \
   920 		    } else if(!ofs)					      \
   921 			break;						      \
   922 		    if(ofs == w) {					      \
   923 			ofs = 0;					      \
   924 			dstbuf += dst->pitch;				      \
   925 			if(!--linecount)				      \
   926 			    break;					      \
   927 		    }							      \
   928 		}							      \
   929 	    } while(0)
   930 
   931 	    CHOOSE_BLIT(RLEBLIT, alpha, fmt);
   932 
   933 #undef RLEBLIT
   934 	}
   935 
   936 done:
   937 	/* Unlock the destination if necessary */
   938 	if ( SDL_MUSTLOCK(dst) ) {
   939 		SDL_UnlockSurface(dst);
   940 	}
   941 	return(0);
   942 }
   943 
   944 #undef OPAQUE_BLIT
   945 
   946 /*
   947  * Per-pixel blitting macros for translucent pixels:
   948  * These use the same techniques as the per-surface blitting macros
   949  */
   950 
   951 /*
   952  * For 32bpp pixels, we have made sure the alpha is stored in the top
   953  * 8 bits, so proceed as usual
   954  */
   955 #define BLIT_TRANSL_888(src, dst)				\
   956     do {							\
   957         Uint32 s = src;						\
   958 	Uint32 d = dst;						\
   959 	unsigned alpha = s >> 24;				\
   960 	Uint32 s1 = s & 0xff00ff;				\
   961 	Uint32 d1 = d & 0xff00ff;				\
   962 	d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;	\
   963 	s &= 0xff00;						\
   964 	d &= 0xff00;						\
   965 	d = (d + ((s - d) * alpha >> 8)) & 0xff00;		\
   966 	dst = d1 | d;						\
   967     } while(0)
   968 
   969 /*
   970  * For 16bpp pixels, we have stored the 5 most significant alpha bits in
   971  * bits 5-10. As before, we can process all 3 RGB components at the same time.
   972  */
   973 #define BLIT_TRANSL_565(src, dst)		\
   974     do {					\
   975 	Uint32 s = src;				\
   976 	Uint32 d = dst;				\
   977 	unsigned alpha = (s & 0x3e0) >> 5;	\
   978 	s &= 0x07e0f81f;			\
   979 	d = (d | d << 16) & 0x07e0f81f;		\
   980 	d += (s - d) * alpha >> 5;		\
   981 	d &= 0x07e0f81f;			\
   982 	dst = (Uint16)(d | d >> 16);			\
   983     } while(0)
   984 
   985 #define BLIT_TRANSL_555(src, dst)		\
   986     do {					\
   987 	Uint32 s = src;				\
   988 	Uint32 d = dst;				\
   989 	unsigned alpha = (s & 0x3e0) >> 5;	\
   990 	s &= 0x03e07c1f;			\
   991 	d = (d | d << 16) & 0x03e07c1f;		\
   992 	d += (s - d) * alpha >> 5;		\
   993 	d &= 0x03e07c1f;			\
   994 	dst = (Uint16)(d | d >> 16);			\
   995     } while(0)
   996 
   997 /* used to save the destination format in the encoding. Designed to be
   998    macro-compatible with SDL_PixelFormat but without the unneeded fields */
   999 typedef struct {
  1000 	Uint8  BytesPerPixel;
  1001 	Uint8  Rloss;
  1002 	Uint8  Gloss;
  1003 	Uint8  Bloss;
  1004 	Uint8  Rshift;
  1005 	Uint8  Gshift;
  1006 	Uint8  Bshift;
  1007 	Uint8  Ashift;
  1008 	Uint32 Rmask;
  1009 	Uint32 Gmask;
  1010 	Uint32 Bmask;
  1011 	Uint32 Amask;
  1012 } RLEDestFormat;
  1013 
  1014 /* blit a pixel-alpha RLE surface clipped at the right and/or left edges */
  1015 static void RLEAlphaClipBlit(int w, Uint8 *srcbuf, SDL_Surface *dst,
  1016 			     Uint8 *dstbuf, SDL_Rect *srcrect)
  1017 {
  1018     SDL_PixelFormat *df = dst->format;
  1019     /*
  1020      * clipped blitter: Ptype is the destination pixel type,
  1021      * Ctype the translucent count type, and do_blend the macro
  1022      * to blend one pixel.
  1023      */
  1024 #define RLEALPHACLIPBLIT(Ptype, Ctype, do_blend)			  \
  1025     do {								  \
  1026 	int linecount = srcrect->h;					  \
  1027 	int left = srcrect->x;						  \
  1028 	int right = left + srcrect->w;					  \
  1029 	dstbuf -= left * sizeof(Ptype);					  \
  1030 	do {								  \
  1031 	    int ofs = 0;						  \
  1032 	    /* blit opaque pixels on one line */			  \
  1033 	    do {							  \
  1034 		unsigned run;						  \
  1035 		ofs += ((Ctype *)srcbuf)[0];				  \
  1036 		run = ((Ctype *)srcbuf)[1];				  \
  1037 		srcbuf += 2 * sizeof(Ctype);				  \
  1038 		if(run) {						  \
  1039 		    /* clip to left and right borders */		  \
  1040 		    int cofs = ofs;					  \
  1041 		    int crun = run;					  \
  1042 		    if(left - cofs > 0) {				  \
  1043 			crun -= left - cofs;				  \
  1044 			cofs = left;					  \
  1045 		    }							  \
  1046 		    if(crun > right - cofs)				  \
  1047 			crun = right - cofs;				  \
  1048 		    if(crun > 0)					  \
  1049 			PIXEL_COPY(dstbuf + cofs * sizeof(Ptype),	  \
  1050 				   srcbuf + (cofs - ofs) * sizeof(Ptype), \
  1051 				   (unsigned)crun, sizeof(Ptype));	  \
  1052 		    srcbuf += run * sizeof(Ptype);			  \
  1053 		    ofs += run;						  \
  1054 		} else if(!ofs)						  \
  1055 		    return;						  \
  1056 	    } while(ofs < w);						  \
  1057 	    /* skip padding if necessary */				  \
  1058 	    if(sizeof(Ptype) == 2)					  \
  1059 		srcbuf += (uintptr_t)srcbuf & 2;			  \
  1060 	    /* blit translucent pixels on the same line */		  \
  1061 	    ofs = 0;							  \
  1062 	    do {							  \
  1063 		unsigned run;						  \
  1064 		ofs += ((Uint16 *)srcbuf)[0];				  \
  1065 		run = ((Uint16 *)srcbuf)[1];				  \
  1066 		srcbuf += 4;						  \
  1067 		if(run) {						  \
  1068 		    /* clip to left and right borders */		  \
  1069 		    int cofs = ofs;					  \
  1070 		    int crun = run;					  \
  1071 		    if(left - cofs > 0) {				  \
  1072 			crun -= left - cofs;				  \
  1073 			cofs = left;					  \
  1074 		    }							  \
  1075 		    if(crun > right - cofs)				  \
  1076 			crun = right - cofs;				  \
  1077 		    if(crun > 0) {					  \
  1078 			Ptype *dst = (Ptype *)dstbuf + cofs;		  \
  1079 			Uint32 *src = (Uint32 *)srcbuf + (cofs - ofs);	  \
  1080 			int i;						  \
  1081 			for(i = 0; i < crun; i++)			  \
  1082 			    do_blend(src[i], dst[i]);			  \
  1083 		    }							  \
  1084 		    srcbuf += run * 4;					  \
  1085 		    ofs += run;						  \
  1086 		}							  \
  1087 	    } while(ofs < w);						  \
  1088 	    dstbuf += dst->pitch;					  \
  1089 	} while(--linecount);						  \
  1090     } while(0)
  1091 
  1092     switch(df->BytesPerPixel) {
  1093     case 2:
  1094 	if(df->Gmask == 0x07e0 || df->Rmask == 0x07e0
  1095 	   || df->Bmask == 0x07e0)
  1096 	    RLEALPHACLIPBLIT(Uint16, Uint8, BLIT_TRANSL_565);
  1097 	else
  1098 	    RLEALPHACLIPBLIT(Uint16, Uint8, BLIT_TRANSL_555);
  1099 	break;
  1100     case 4:
  1101 	RLEALPHACLIPBLIT(Uint32, Uint16, BLIT_TRANSL_888);
  1102 	break;
  1103     }
  1104 }
  1105 
  1106 /* blit a pixel-alpha RLE surface */
  1107 int SDL_RLEAlphaBlit(SDL_Surface *src, SDL_Rect *srcrect,
  1108 		     SDL_Surface *dst, SDL_Rect *dstrect)
  1109 {
  1110     int x, y;
  1111     int w = src->w;
  1112     Uint8 *srcbuf, *dstbuf;
  1113     SDL_PixelFormat *df = dst->format;
  1114 
  1115     /* Lock the destination if necessary */
  1116     if ( SDL_MUSTLOCK(dst) ) {
  1117 	if ( SDL_LockSurface(dst) < 0 ) {
  1118 	    return -1;
  1119 	}
  1120     }
  1121 
  1122     x = dstrect->x;
  1123     y = dstrect->y;
  1124     dstbuf = (Uint8 *)dst->pixels
  1125 	     + y * dst->pitch + x * df->BytesPerPixel;
  1126     srcbuf = (Uint8 *)src->map->sw_data->aux_data + sizeof(RLEDestFormat);
  1127 
  1128     {
  1129 	/* skip lines at the top if necessary */
  1130 	int vskip = srcrect->y;
  1131 	if(vskip) {
  1132 	    int ofs;
  1133 	    if(df->BytesPerPixel == 2) {
  1134 		/* the 16/32 interleaved format */
  1135 		do {
  1136 		    /* skip opaque line */
  1137 		    ofs = 0;
  1138 		    do {
  1139 			int run;
  1140 			ofs += srcbuf[0];
  1141 			run = srcbuf[1];
  1142 			srcbuf += 2;
  1143 			if(run) {
  1144 			    srcbuf += 2 * run;
  1145 			    ofs += run;
  1146 			} else if(!ofs)
  1147 			    goto done;
  1148 		    } while(ofs < w);
  1149 
  1150 		    /* skip padding */
  1151 		    srcbuf += (uintptr_t)srcbuf & 2;
  1152 
  1153 		    /* skip translucent line */
  1154 		    ofs = 0;
  1155 		    do {
  1156 			int run;
  1157 			ofs += ((Uint16 *)srcbuf)[0];
  1158 			run = ((Uint16 *)srcbuf)[1];
  1159 			srcbuf += 4 * (run + 1);
  1160 			ofs += run;
  1161 		    } while(ofs < w);
  1162 		} while(--vskip);
  1163 	    } else {
  1164 		/* the 32/32 interleaved format */
  1165 		vskip <<= 1;	/* opaque and translucent have same format */
  1166 		do {
  1167 		    ofs = 0;
  1168 		    do {
  1169 			int run;
  1170 			ofs += ((Uint16 *)srcbuf)[0];
  1171 			run = ((Uint16 *)srcbuf)[1];
  1172 			srcbuf += 4;
  1173 			if(run) {
  1174 			    srcbuf += 4 * run;
  1175 			    ofs += run;
  1176 			} else if(!ofs)
  1177 			    goto done;
  1178 		    } while(ofs < w);
  1179 		} while(--vskip);
  1180 	    }
  1181 	}
  1182     }
  1183 
  1184     /* if left or right edge clipping needed, call clip blit */
  1185     if(srcrect->x || srcrect->w != src->w) {
  1186 	RLEAlphaClipBlit(w, srcbuf, dst, dstbuf, srcrect);
  1187     } else {
  1188 
  1189 	/*
  1190 	 * non-clipped blitter. Ptype is the destination pixel type,
  1191 	 * Ctype the translucent count type, and do_blend the
  1192 	 * macro to blend one pixel.
  1193 	 */
  1194 #define RLEALPHABLIT(Ptype, Ctype, do_blend)				 \
  1195 	do {								 \
  1196 	    int linecount = srcrect->h;					 \
  1197 	    do {							 \
  1198 		int ofs = 0;						 \
  1199 		/* blit opaque pixels on one line */			 \
  1200 		do {							 \
  1201 		    unsigned run;					 \
  1202 		    ofs += ((Ctype *)srcbuf)[0];			 \
  1203 		    run = ((Ctype *)srcbuf)[1];				 \
  1204 		    srcbuf += 2 * sizeof(Ctype);			 \
  1205 		    if(run) {						 \
  1206 			PIXEL_COPY(dstbuf + ofs * sizeof(Ptype), srcbuf, \
  1207 				   run, sizeof(Ptype));			 \
  1208 			srcbuf += run * sizeof(Ptype);			 \
  1209 			ofs += run;					 \
  1210 		    } else if(!ofs)					 \
  1211 			goto done;					 \
  1212 		} while(ofs < w);					 \
  1213 		/* skip padding if necessary */				 \
  1214 		if(sizeof(Ptype) == 2)					 \
  1215 		    srcbuf += (uintptr_t)srcbuf & 2;		 	 \
  1216 		/* blit translucent pixels on the same line */		 \
  1217 		ofs = 0;						 \
  1218 		do {							 \
  1219 		    unsigned run;					 \
  1220 		    ofs += ((Uint16 *)srcbuf)[0];			 \
  1221 		    run = ((Uint16 *)srcbuf)[1];			 \
  1222 		    srcbuf += 4;					 \
  1223 		    if(run) {						 \
  1224 			Ptype *dst = (Ptype *)dstbuf + ofs;		 \
  1225 			unsigned i;					 \
  1226 			for(i = 0; i < run; i++) {			 \
  1227 			    Uint32 src = *(Uint32 *)srcbuf;		 \
  1228 			    do_blend(src, *dst);			 \
  1229 			    srcbuf += 4;				 \
  1230 			    dst++;					 \
  1231 			}						 \
  1232 			ofs += run;					 \
  1233 		    }							 \
  1234 		} while(ofs < w);					 \
  1235 		dstbuf += dst->pitch;					 \
  1236 	    } while(--linecount);					 \
  1237 	} while(0)
  1238 
  1239 	switch(df->BytesPerPixel) {
  1240 	case 2:
  1241 	    if(df->Gmask == 0x07e0 || df->Rmask == 0x07e0
  1242 	       || df->Bmask == 0x07e0)
  1243 		RLEALPHABLIT(Uint16, Uint8, BLIT_TRANSL_565);
  1244 	    else
  1245 		RLEALPHABLIT(Uint16, Uint8, BLIT_TRANSL_555);
  1246 	    break;
  1247 	case 4:
  1248 	    RLEALPHABLIT(Uint32, Uint16, BLIT_TRANSL_888);
  1249 	    break;
  1250 	}
  1251     }
  1252 
  1253  done:
  1254     /* Unlock the destination if necessary */
  1255     if ( SDL_MUSTLOCK(dst) ) {
  1256 	SDL_UnlockSurface(dst);
  1257     }
  1258     return 0;
  1259 }
  1260 
  1261 /*
  1262  * Auxiliary functions:
  1263  * The encoding functions take 32bpp rgb + a, and
  1264  * return the number of bytes copied to the destination.
  1265  * The decoding functions copy to 32bpp rgb + a, and
  1266  * return the number of bytes copied from the source.
  1267  * These are only used in the encoder and un-RLE code and are therefore not
  1268  * highly optimised.
  1269  */
  1270 
  1271 /* encode 32bpp rgb + a into 16bpp rgb, losing alpha */
  1272 static int copy_opaque_16(void *dst, Uint32 *src, int n,
  1273 			  SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1274 {
  1275     int i;
  1276     Uint16 *d = dst;
  1277     for(i = 0; i < n; i++) {
  1278 	unsigned r, g, b;
  1279 	RGB_FROM_PIXEL(*src, sfmt, r, g, b);
  1280 	PIXEL_FROM_RGB(*d, dfmt, r, g, b);
  1281 	src++;
  1282 	d++;
  1283     }
  1284     return n * 2;
  1285 }
  1286 
  1287 /* decode opaque pixels from 16bpp to 32bpp rgb + a */
  1288 static int uncopy_opaque_16(Uint32 *dst, void *src, int n,
  1289 			    RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1290 {
  1291     int i;
  1292     Uint16 *s = src;
  1293     unsigned alpha = dfmt->Amask ? 255 : 0;
  1294     for(i = 0; i < n; i++) {
  1295 	unsigned r, g, b;
  1296 	RGB_FROM_PIXEL(*s, sfmt, r, g, b);
  1297 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, alpha);
  1298 	s++;
  1299 	dst++;
  1300     }
  1301     return n * 2;
  1302 }
  1303 
  1304 
  1305 
  1306 /* encode 32bpp rgb + a into 32bpp G0RAB format for blitting into 565 */
  1307 static int copy_transl_565(void *dst, Uint32 *src, int n,
  1308 			   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1309 {
  1310     int i;
  1311     Uint32 *d = dst;
  1312     for(i = 0; i < n; i++) {
  1313 	unsigned r, g, b, a;
  1314 	Uint16 pix;
  1315 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1316 	PIXEL_FROM_RGB(pix, dfmt, r, g, b);
  1317 	*d = ((pix & 0x7e0) << 16) | (pix & 0xf81f) | ((a << 2) & 0x7e0);
  1318 	src++;
  1319 	d++;
  1320     }
  1321     return n * 4;
  1322 }
  1323 
  1324 /* encode 32bpp rgb + a into 32bpp G0RAB format for blitting into 555 */
  1325 static int copy_transl_555(void *dst, Uint32 *src, int n,
  1326 			   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1327 {
  1328     int i;
  1329     Uint32 *d = dst;
  1330     for(i = 0; i < n; i++) {
  1331 	unsigned r, g, b, a;
  1332 	Uint16 pix;
  1333 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1334 	PIXEL_FROM_RGB(pix, dfmt, r, g, b);
  1335 	*d = ((pix & 0x3e0) << 16) | (pix & 0xfc1f) | ((a << 2) & 0x3e0);
  1336 	src++;
  1337 	d++;
  1338     }
  1339     return n * 4;
  1340 }
  1341 
  1342 /* decode translucent pixels from 32bpp GORAB to 32bpp rgb + a */
  1343 static int uncopy_transl_16(Uint32 *dst, void *src, int n,
  1344 			    RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1345 {
  1346     int i;
  1347     Uint32 *s = src;
  1348     for(i = 0; i < n; i++) {
  1349 	unsigned r, g, b, a;
  1350 	Uint32 pix = *s++;
  1351 	a = (pix & 0x3e0) >> 2;
  1352 	pix = (pix & ~0x3e0) | pix >> 16;
  1353 	RGB_FROM_PIXEL(pix, sfmt, r, g, b);
  1354 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, a);
  1355 	dst++;
  1356     }
  1357     return n * 4;
  1358 }
  1359 
  1360 /* encode 32bpp rgba into 32bpp rgba, keeping alpha (dual purpose) */
  1361 static int copy_32(void *dst, Uint32 *src, int n,
  1362 		   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1363 {
  1364     int i;
  1365     Uint32 *d = dst;
  1366     for(i = 0; i < n; i++) {
  1367 	unsigned r, g, b, a;
  1368 	Uint32 pixel;
  1369 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1370 	PIXEL_FROM_RGB(pixel, dfmt, r, g, b);
  1371 	*d++ = pixel | a << 24;
  1372 	src++;
  1373     }
  1374     return n * 4;
  1375 }
  1376 
  1377 /* decode 32bpp rgba into 32bpp rgba, keeping alpha (dual purpose) */
  1378 static int uncopy_32(Uint32 *dst, void *src, int n,
  1379 		     RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1380 {
  1381     int i;
  1382     Uint32 *s = src;
  1383     for(i = 0; i < n; i++) {
  1384 	unsigned r, g, b, a;
  1385 	Uint32 pixel = *s++;
  1386 	RGB_FROM_PIXEL(pixel, sfmt, r, g, b);
  1387 	a = pixel >> 24;
  1388 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, a);
  1389 	dst++;
  1390     }
  1391     return n * 4;
  1392 }
  1393 
  1394 #define ISOPAQUE(pixel, fmt) ((((pixel) & fmt->Amask) >> fmt->Ashift) == 255)
  1395 
  1396 #define ISTRANSL(pixel, fmt)	\
  1397     ((unsigned)((((pixel) & fmt->Amask) >> fmt->Ashift) - 1U) < 254U)
  1398 
  1399 /* convert surface to be quickly alpha-blittable onto dest, if possible */
  1400 static int RLEAlphaSurface(SDL_Surface *surface)
  1401 {
  1402     SDL_Surface *dest;
  1403     SDL_PixelFormat *df;
  1404     int maxsize = 0;
  1405     int max_opaque_run;
  1406     int max_transl_run = 65535;
  1407     unsigned masksum;
  1408     Uint8 *rlebuf, *dst;
  1409     int (*copy_opaque)(void *, Uint32 *, int,
  1410 		       SDL_PixelFormat *, SDL_PixelFormat *);
  1411     int (*copy_transl)(void *, Uint32 *, int,
  1412 		       SDL_PixelFormat *, SDL_PixelFormat *);
  1413 
  1414     dest = surface->map->dst;
  1415     if(!dest)
  1416 	return -1;
  1417     df = dest->format;
  1418     if(surface->format->BitsPerPixel != 32)
  1419 	return -1;		/* only 32bpp source supported */
  1420 
  1421     /* find out whether the destination is one we support,
  1422        and determine the max size of the encoded result */
  1423     masksum = df->Rmask | df->Gmask | df->Bmask;
  1424     switch(df->BytesPerPixel) {
  1425     case 2:
  1426 	/* 16bpp: only support 565 and 555 formats */
  1427 	switch(masksum) {
  1428 	case 0xffff:
  1429 	    if(df->Gmask == 0x07e0
  1430 	       || df->Rmask == 0x07e0 || df->Bmask == 0x07e0) {
  1431 		copy_opaque = copy_opaque_16;
  1432 		copy_transl = copy_transl_565;
  1433 	    } else
  1434 		return -1;
  1435 	    break;
  1436 	case 0x7fff:
  1437 	    if(df->Gmask == 0x03e0
  1438 	       || df->Rmask == 0x03e0 || df->Bmask == 0x03e0) {
  1439 		copy_opaque = copy_opaque_16;
  1440 		copy_transl = copy_transl_555;
  1441 	    } else
  1442 		return -1;
  1443 	    break;
  1444 	default:
  1445 	    return -1;
  1446 	}
  1447 	max_opaque_run = 255;	/* runs stored as bytes */
  1448 
  1449 	/* worst case is alternating opaque and translucent pixels,
  1450 	   with room for alignment padding between lines */
  1451 	maxsize = surface->h * (2 + (4 + 2) * (surface->w + 1)) + 2;
  1452 	break;
  1453     case 4:
  1454 	if(masksum != 0x00ffffff)
  1455 	    return -1;		/* requires unused high byte */
  1456 	copy_opaque = copy_32;
  1457 	copy_transl = copy_32;
  1458 	max_opaque_run = 255;	/* runs stored as short ints */
  1459 
  1460 	/* worst case is alternating opaque and translucent pixels */
  1461 	maxsize = surface->h * 2 * 4 * (surface->w + 1) + 4;
  1462 	break;
  1463     default:
  1464 	return -1;		/* anything else unsupported right now */
  1465     }
  1466 
  1467     maxsize += sizeof(RLEDestFormat);
  1468     rlebuf = (Uint8 *)SDL_malloc(maxsize);
  1469     if(!rlebuf) {
  1470 	SDL_OutOfMemory();
  1471 	return -1;
  1472     }
  1473     {
  1474 	/* save the destination format so we can undo the encoding later */
  1475 	RLEDestFormat *r = (RLEDestFormat *)rlebuf;
  1476 	r->BytesPerPixel = df->BytesPerPixel;
  1477 	r->Rloss = df->Rloss;
  1478 	r->Gloss = df->Gloss;
  1479 	r->Bloss = df->Bloss;
  1480 	r->Rshift = df->Rshift;
  1481 	r->Gshift = df->Gshift;
  1482 	r->Bshift = df->Bshift;
  1483 	r->Ashift = df->Ashift;
  1484 	r->Rmask = df->Rmask;
  1485 	r->Gmask = df->Gmask;
  1486 	r->Bmask = df->Bmask;
  1487 	r->Amask = df->Amask;
  1488     }
  1489     dst = rlebuf + sizeof(RLEDestFormat);
  1490 
  1491     /* Do the actual encoding */
  1492     {
  1493 	int x, y;
  1494 	int h = surface->h, w = surface->w;
  1495 	SDL_PixelFormat *sf = surface->format;
  1496 	Uint32 *src = (Uint32 *)surface->pixels;
  1497 	Uint8 *lastline = dst;	/* end of last non-blank line */
  1498 
  1499 	/* opaque counts are 8 or 16 bits, depending on target depth */
  1500 #define ADD_OPAQUE_COUNTS(n, m)			\
  1501 	if(df->BytesPerPixel == 4) {		\
  1502 	    ((Uint16 *)dst)[0] = n;		\
  1503 	    ((Uint16 *)dst)[1] = m;		\
  1504 	    dst += 4;				\
  1505 	} else {				\
  1506 	    dst[0] = n;				\
  1507 	    dst[1] = m;				\
  1508 	    dst += 2;				\
  1509 	}
  1510 
  1511 	/* translucent counts are always 16 bit */
  1512 #define ADD_TRANSL_COUNTS(n, m)		\
  1513 	(((Uint16 *)dst)[0] = n, ((Uint16 *)dst)[1] = m, dst += 4)
  1514 
  1515 	for(y = 0; y < h; y++) {
  1516 	    int runstart, skipstart;
  1517 	    int blankline = 0;
  1518 	    /* First encode all opaque pixels of a scan line */
  1519 	    x = 0;
  1520 	    do {
  1521 		int run, skip, len;
  1522 		skipstart = x;
  1523 		while(x < w && !ISOPAQUE(src[x], sf))
  1524 		    x++;
  1525 		runstart = x;
  1526 		while(x < w && ISOPAQUE(src[x], sf))
  1527 		    x++;
  1528 		skip = runstart - skipstart;
  1529 		if(skip == w)
  1530 		    blankline = 1;
  1531 		run = x - runstart;
  1532 		while(skip > max_opaque_run) {
  1533 		    ADD_OPAQUE_COUNTS(max_opaque_run, 0);
  1534 		    skip -= max_opaque_run;
  1535 		}
  1536 		len = MIN(run, max_opaque_run);
  1537 		ADD_OPAQUE_COUNTS(skip, len);
  1538 		dst += copy_opaque(dst, src + runstart, len, sf, df);
  1539 		runstart += len;
  1540 		run -= len;
  1541 		while(run) {
  1542 		    len = MIN(run, max_opaque_run);
  1543 		    ADD_OPAQUE_COUNTS(0, len);
  1544 		    dst += copy_opaque(dst, src + runstart, len, sf, df);
  1545 		    runstart += len;
  1546 		    run -= len;
  1547 		}
  1548 	    } while(x < w);
  1549 
  1550 	    /* Make sure the next output address is 32-bit aligned */
  1551 	    dst += (uintptr_t)dst & 2;
  1552 
  1553 	    /* Next, encode all translucent pixels of the same scan line */
  1554 	    x = 0;
  1555 	    do {
  1556 		int run, skip, len;
  1557 		skipstart = x;
  1558 		while(x < w && !ISTRANSL(src[x], sf))
  1559 		    x++;
  1560 		runstart = x;
  1561 		while(x < w && ISTRANSL(src[x], sf))
  1562 		    x++;
  1563 		skip = runstart - skipstart;
  1564 		blankline &= (skip == w);
  1565 		run = x - runstart;
  1566 		while(skip > max_transl_run) {
  1567 		    ADD_TRANSL_COUNTS(max_transl_run, 0);
  1568 		    skip -= max_transl_run;
  1569 		}
  1570 		len = MIN(run, max_transl_run);
  1571 		ADD_TRANSL_COUNTS(skip, len);
  1572 		dst += copy_transl(dst, src + runstart, len, sf, df);
  1573 		runstart += len;
  1574 		run -= len;
  1575 		while(run) {
  1576 		    len = MIN(run, max_transl_run);
  1577 		    ADD_TRANSL_COUNTS(0, len);
  1578 		    dst += copy_transl(dst, src + runstart, len, sf, df);
  1579 		    runstart += len;
  1580 		    run -= len;
  1581 		}
  1582 		if(!blankline)
  1583 		    lastline = dst;
  1584 	    } while(x < w);
  1585 
  1586 	    src += surface->pitch >> 2;
  1587 	}
  1588 	dst = lastline;		/* back up past trailing blank lines */
  1589 	ADD_OPAQUE_COUNTS(0, 0);
  1590     }
  1591 
  1592 #undef ADD_OPAQUE_COUNTS
  1593 #undef ADD_TRANSL_COUNTS
  1594 
  1595     /* Now that we have it encoded, release the original pixels */
  1596     if((surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1597        && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1598 	SDL_free( surface->pixels );
  1599 	surface->pixels = NULL;
  1600     }
  1601 
  1602     /* realloc the buffer to release unused memory */
  1603     {
  1604 	Uint8 *p = SDL_realloc(rlebuf, dst - rlebuf);
  1605 	if(!p)
  1606 	    p = rlebuf;
  1607 	surface->map->sw_data->aux_data = p;
  1608     }
  1609 
  1610     return 0;
  1611 }
  1612 
  1613 static Uint32 getpix_8(Uint8 *srcbuf)
  1614 {
  1615     return *srcbuf;
  1616 }
  1617 
  1618 static Uint32 getpix_16(Uint8 *srcbuf)
  1619 {
  1620     return *(Uint16 *)srcbuf;
  1621 }
  1622 
  1623 static Uint32 getpix_24(Uint8 *srcbuf)
  1624 {
  1625 #if SDL_BYTEORDER == SDL_LIL_ENDIAN
  1626     return srcbuf[0] + (srcbuf[1] << 8) + (srcbuf[2] << 16);
  1627 #else
  1628     return (srcbuf[0] << 16) + (srcbuf[1] << 8) + srcbuf[2];
  1629 #endif
  1630 }
  1631 
  1632 static Uint32 getpix_32(Uint8 *srcbuf)
  1633 {
  1634     return *(Uint32 *)srcbuf;
  1635 }
  1636 
  1637 typedef Uint32 (*getpix_func)(Uint8 *);
  1638 
  1639 static getpix_func getpixes[4] = {
  1640     getpix_8, getpix_16, getpix_24, getpix_32
  1641 };
  1642 
  1643 static int RLEColorkeySurface(SDL_Surface *surface)
  1644 {
  1645         Uint8 *rlebuf, *dst;
  1646 	int maxn;
  1647 	int y;
  1648 	Uint8 *srcbuf, *lastline;
  1649 	int maxsize = 0;
  1650 	int bpp = surface->format->BytesPerPixel;
  1651 	getpix_func getpix;
  1652 	Uint32 ckey, rgbmask;
  1653 	int w, h;
  1654 
  1655 	/* calculate the worst case size for the compressed surface */
  1656 	switch(bpp) {
  1657 	case 1:
  1658 	    /* worst case is alternating opaque and transparent pixels,
  1659 	       starting with an opaque pixel */
  1660 	    maxsize = surface->h * 3 * (surface->w / 2 + 1) + 2;
  1661 	    break;
  1662 	case 2:
  1663 	case 3:
  1664 	    /* worst case is solid runs, at most 255 pixels wide */
  1665 	    maxsize = surface->h * (2 * (surface->w / 255 + 1)
  1666 				    + surface->w * bpp) + 2;
  1667 	    break;
  1668 	case 4:
  1669 	    /* worst case is solid runs, at most 65535 pixels wide */
  1670 	    maxsize = surface->h * (4 * (surface->w / 65535 + 1)
  1671 				    + surface->w * 4) + 4;
  1672 	    break;
  1673 	}
  1674 
  1675 	rlebuf = (Uint8 *)SDL_malloc(maxsize);
  1676 	if ( rlebuf == NULL ) {
  1677 		SDL_OutOfMemory();
  1678 		return(-1);
  1679 	}
  1680 
  1681 	/* Set up the conversion */
  1682 	srcbuf = (Uint8 *)surface->pixels;
  1683 	maxn = bpp == 4 ? 65535 : 255;
  1684 	dst = rlebuf;
  1685 	rgbmask = ~surface->format->Amask;
  1686 	ckey = surface->format->colorkey & rgbmask;
  1687 	lastline = dst;
  1688 	getpix = getpixes[bpp - 1];
  1689 	w = surface->w;
  1690 	h = surface->h;
  1691 
  1692 #define ADD_COUNTS(n, m)			\
  1693 	if(bpp == 4) {				\
  1694 	    ((Uint16 *)dst)[0] = n;		\
  1695 	    ((Uint16 *)dst)[1] = m;		\
  1696 	    dst += 4;				\
  1697 	} else {				\
  1698 	    dst[0] = n;				\
  1699 	    dst[1] = m;				\
  1700 	    dst += 2;				\
  1701 	}
  1702 
  1703 	for(y = 0; y < h; y++) {
  1704 	    int x = 0;
  1705 	    int blankline = 0;
  1706 	    do {
  1707 		int run, skip, len;
  1708 		int runstart;
  1709 		int skipstart = x;
  1710 
  1711 		/* find run of transparent, then opaque pixels */
  1712 		while(x < w && (getpix(srcbuf + x * bpp) & rgbmask) == ckey)
  1713 		    x++;
  1714 		runstart = x;
  1715 		while(x < w && (getpix(srcbuf + x * bpp) & rgbmask) != ckey)
  1716 		    x++;
  1717 		skip = runstart - skipstart;
  1718 		if(skip == w)
  1719 		    blankline = 1;
  1720 		run = x - runstart;
  1721 
  1722 		/* encode segment */
  1723 		while(skip > maxn) {
  1724 		    ADD_COUNTS(maxn, 0);
  1725 		    skip -= maxn;
  1726 		}
  1727 		len = MIN(run, maxn);
  1728 		ADD_COUNTS(skip, len);
  1729 		SDL_memcpy(dst, srcbuf + runstart * bpp, len * bpp);
  1730 		dst += len * bpp;
  1731 		run -= len;
  1732 		runstart += len;
  1733 		while(run) {
  1734 		    len = MIN(run, maxn);
  1735 		    ADD_COUNTS(0, len);
  1736 		    SDL_memcpy(dst, srcbuf + runstart * bpp, len * bpp);
  1737 		    dst += len * bpp;
  1738 		    runstart += len;
  1739 		    run -= len;
  1740 		}
  1741 		if(!blankline)
  1742 		    lastline = dst;
  1743 	    } while(x < w);
  1744 
  1745 	    srcbuf += surface->pitch;
  1746 	}
  1747 	dst = lastline;		/* back up bast trailing blank lines */
  1748 	ADD_COUNTS(0, 0);
  1749 
  1750 #undef ADD_COUNTS
  1751 
  1752 	/* Now that we have it encoded, release the original pixels */
  1753 	if((surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1754 	   && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1755 	    SDL_free( surface->pixels );
  1756 	    surface->pixels = NULL;
  1757 	}
  1758 
  1759 	/* realloc the buffer to release unused memory */
  1760 	{
  1761 	    /* If realloc returns NULL, the original block is left intact */
  1762 	    Uint8 *p = SDL_realloc(rlebuf, dst - rlebuf);
  1763 	    if(!p)
  1764 		p = rlebuf;
  1765 	    surface->map->sw_data->aux_data = p;
  1766 	}
  1767 
  1768 	return(0);
  1769 }
  1770 
  1771 int SDL_RLESurface(SDL_Surface *surface)
  1772 {
  1773 	int retcode;
  1774 
  1775 	/* Clear any previous RLE conversion */
  1776 	if ( (surface->flags & SDL_RLEACCEL) == SDL_RLEACCEL ) {
  1777 		SDL_UnRLESurface(surface, 1);
  1778 	}
  1779 
  1780 	/* We don't support RLE encoding of bitmaps */
  1781 	if ( surface->format->BitsPerPixel < 8 ) {
  1782 		return(-1);
  1783 	}
  1784 
  1785 	/* Lock the surface if it's in hardware */
  1786 	if ( SDL_MUSTLOCK(surface) ) {
  1787 		if ( SDL_LockSurface(surface) < 0 ) {
  1788 			return(-1);
  1789 		}
  1790 	}
  1791 
  1792 	/* Encode */
  1793 	if((surface->flags & SDL_SRCCOLORKEY) == SDL_SRCCOLORKEY) {
  1794 	    retcode = RLEColorkeySurface(surface);
  1795 	} else {
  1796 	    if((surface->flags & SDL_SRCALPHA) == SDL_SRCALPHA
  1797 	       && surface->format->Amask != 0)
  1798 		retcode = RLEAlphaSurface(surface);
  1799 	    else
  1800 		retcode = -1;	/* no RLE for per-surface alpha sans ckey */
  1801 	}
  1802 
  1803 	/* Unlock the surface if it's in hardware */
  1804 	if ( SDL_MUSTLOCK(surface) ) {
  1805 		SDL_UnlockSurface(surface);
  1806 	}
  1807 
  1808 	if(retcode < 0)
  1809 	    return -1;
  1810 
  1811 	/* The surface is now accelerated */
  1812 	surface->flags |= SDL_RLEACCEL;
  1813 
  1814 	return(0);
  1815 }
  1816 
  1817 /*
  1818  * Un-RLE a surface with pixel alpha
  1819  * This may not give back exactly the image before RLE-encoding; all
  1820  * completely transparent pixels will be lost, and colour and alpha depth
  1821  * may have been reduced (when encoding for 16bpp targets).
  1822  */
  1823 static SDL_bool UnRLEAlpha(SDL_Surface *surface)
  1824 {
  1825     Uint8 *srcbuf;
  1826     Uint32 *dst;
  1827     SDL_PixelFormat *sf = surface->format;
  1828     RLEDestFormat *df = surface->map->sw_data->aux_data;
  1829     int (*uncopy_opaque)(Uint32 *, void *, int,
  1830 			 RLEDestFormat *, SDL_PixelFormat *);
  1831     int (*uncopy_transl)(Uint32 *, void *, int,
  1832 			 RLEDestFormat *, SDL_PixelFormat *);
  1833     int w = surface->w;
  1834     int bpp = df->BytesPerPixel;
  1835 
  1836     if(bpp == 2) {
  1837 	uncopy_opaque = uncopy_opaque_16;
  1838 	uncopy_transl = uncopy_transl_16;
  1839     } else {
  1840 	uncopy_opaque = uncopy_transl = uncopy_32;
  1841     }
  1842 
  1843     surface->pixels = SDL_malloc(surface->h * surface->pitch);
  1844     if ( !surface->pixels ) {
  1845         return(SDL_FALSE);
  1846     }
  1847     /* fill background with transparent pixels */
  1848     SDL_memset(surface->pixels, 0, surface->h * surface->pitch);
  1849 
  1850     dst = surface->pixels;
  1851     srcbuf = (Uint8 *)(df + 1);
  1852     for(;;) {
  1853 	/* copy opaque pixels */
  1854 	int ofs = 0;
  1855 	do {
  1856 	    unsigned run;
  1857 	    if(bpp == 2) {
  1858 		ofs += srcbuf[0];
  1859 		run = srcbuf[1];
  1860 		srcbuf += 2;
  1861 	    } else {
  1862 		ofs += ((Uint16 *)srcbuf)[0];
  1863 		run = ((Uint16 *)srcbuf)[1];
  1864 		srcbuf += 4;
  1865 	    }
  1866 	    if(run) {
  1867 		srcbuf += uncopy_opaque(dst + ofs, srcbuf, run, df, sf);
  1868 		ofs += run;
  1869 	    } else if(!ofs)
  1870 		return(SDL_TRUE);
  1871 	} while(ofs < w);
  1872 
  1873 	/* skip padding if needed */
  1874 	if(bpp == 2)
  1875 	    srcbuf += (uintptr_t)srcbuf & 2;
  1876 	
  1877 	/* copy translucent pixels */
  1878 	ofs = 0;
  1879 	do {
  1880 	    unsigned run;
  1881 	    ofs += ((Uint16 *)srcbuf)[0];
  1882 	    run = ((Uint16 *)srcbuf)[1];
  1883 	    srcbuf += 4;
  1884 	    if(run) {
  1885 		srcbuf += uncopy_transl(dst + ofs, srcbuf, run, df, sf);
  1886 		ofs += run;
  1887 	    }
  1888 	} while(ofs < w);
  1889 	dst += surface->pitch >> 2;
  1890     }
  1891     /* Make the compiler happy */
  1892     return(SDL_TRUE);
  1893 }
  1894 
  1895 void SDL_UnRLESurface(SDL_Surface *surface, int recode)
  1896 {
  1897     if ( (surface->flags & SDL_RLEACCEL) == SDL_RLEACCEL ) {
  1898 	surface->flags &= ~SDL_RLEACCEL;
  1899 
  1900 	if(recode && (surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1901 	   && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1902 	    if((surface->flags & SDL_SRCCOLORKEY) == SDL_SRCCOLORKEY) {
  1903 		SDL_Rect full;
  1904 		unsigned alpha_flag;
  1905 
  1906 		/* re-create the original surface */
  1907 		surface->pixels = SDL_malloc(surface->h * surface->pitch);
  1908 		if ( !surface->pixels ) {
  1909 			/* Oh crap... */
  1910 			surface->flags |= SDL_RLEACCEL;
  1911 			return;
  1912 		}
  1913 
  1914 		/* fill it with the background colour */
  1915 		SDL_FillRect(surface, NULL, surface->format->colorkey);
  1916 
  1917 		/* now render the encoded surface */
  1918 		full.x = full.y = 0;
  1919 		full.w = surface->w;
  1920 		full.h = surface->h;
  1921 		alpha_flag = surface->flags & SDL_SRCALPHA;
  1922 		surface->flags &= ~SDL_SRCALPHA; /* opaque blit */
  1923 		SDL_RLEBlit(surface, &full, surface, &full);
  1924 		surface->flags |= alpha_flag;
  1925 	    } else {
  1926 		if ( !UnRLEAlpha(surface) ) {
  1927 		    /* Oh crap... */
  1928 		    surface->flags |= SDL_RLEACCEL;
  1929 		    return;
  1930 		}
  1931 	    }
  1932 	}
  1933 
  1934 	if ( surface->map && surface->map->sw_data->aux_data ) {
  1935 	    SDL_free(surface->map->sw_data->aux_data);
  1936 	    surface->map->sw_data->aux_data = NULL;
  1937 	}
  1938     }
  1939 }
  1940 
  1941