src/video/SDL_RLEaccel.c
author Sam Lantinga <slouken@libsdl.org>
Wed, 01 Mar 2006 09:43:47 +0000
changeset 1456 84de7511f79f
parent 1428 5f52867ba65c
child 1662 782fd950bd46
child 1895 c121d94672cb
child 4159 a1b03ba2fcd0
permissions -rw-r--r--
Fixed a bunch of 64-bit compatibility problems
     1 /*
     2     SDL - Simple DirectMedia Layer
     3     Copyright (C) 1997-2006 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 #if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__)) && SDL_ASSEMBLY_ROUTINES
    95 #define MMX_ASMBLIT
    96 #endif
    97 
    98 #ifdef MMX_ASMBLIT
    99 #include "mmx.h"
   100 #include "SDL_cpuinfo.h"
   101 #endif
   102 
   103 #ifndef MAX
   104 #define MAX(a, b) ((a) > (b) ? (a) : (b))
   105 #endif
   106 #ifndef MIN
   107 #define MIN(a, b) ((a) < (b) ? (a) : (b))
   108 #endif
   109 
   110 #define PIXEL_COPY(to, from, len, bpp)			\
   111 do {							\
   112     if(bpp == 4) {					\
   113 	SDL_memcpy4(to, from, (size_t)(len));		\
   114     } else {						\
   115 	SDL_memcpy(to, from, (size_t)(len) * (bpp));	\
   116     }							\
   117 } while(0)
   118 
   119 /*
   120  * Various colorkey blit methods, for opaque and per-surface alpha
   121  */
   122 
   123 #define OPAQUE_BLIT(to, from, length, bpp, alpha)	\
   124     PIXEL_COPY(to, from, length, bpp)
   125 
   126 #ifdef MMX_ASMBLIT
   127 
   128 #define ALPHA_BLIT32_888MMX(to, from, length, bpp, alpha)	\
   129     do {							\
   130 	Uint32 *srcp = (Uint32 *)(from);			\
   131 	Uint32 *dstp = (Uint32 *)(to);				\
   132         int i = 0x00FF00FF;					\
   133         movd_m2r(*(&i), mm3);					\
   134         punpckldq_r2r(mm3, mm3);				\
   135         i = 0xFF000000;						\
   136         movd_m2r(*(&i), mm7);					\
   137         punpckldq_r2r(mm7, mm7);				\
   138         i = alpha | alpha << 16;				\
   139         movd_m2r(*(&i), mm4);					\
   140         punpckldq_r2r(mm4, mm4);				\
   141 	pcmpeqd_r2r(mm5,mm5); /* set mm5 to "1" */		\
   142 	pxor_r2r(mm7, mm5); /* make clear alpha mask */		\
   143         i = length;						\
   144 	if(i & 1) {						\
   145           movd_m2r((*srcp), mm1); /* src -> mm1 */		\
   146           punpcklbw_r2r(mm1, mm1);				\
   147           pand_r2r(mm3, mm1);					\
   148 	  movd_m2r((*dstp), mm2); /* dst -> mm2 */		\
   149           punpcklbw_r2r(mm2, mm2);				\
   150           pand_r2r(mm3, mm2);					\
   151 	  psubw_r2r(mm2, mm1);					\
   152 	  pmullw_r2r(mm4, mm1);					\
   153 	  psrlw_i2r(8, mm1);					\
   154 	  paddw_r2r(mm1, mm2);					\
   155 	  pand_r2r(mm3, mm2);					\
   156 	  packuswb_r2r(mm2, mm2);				\
   157 	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
   158 	  movd_r2m(mm2, *dstp);					\
   159 	  ++srcp;						\
   160 	  ++dstp;						\
   161 	  i--;							\
   162 	}							\
   163 	for(; i > 0; --i) {					\
   164           movq_m2r((*srcp), mm0);				\
   165 	  movq_r2r(mm0, mm1);					\
   166           punpcklbw_r2r(mm0, mm0);				\
   167 	  movq_m2r((*dstp), mm2);				\
   168 	  punpckhbw_r2r(mm1, mm1);				\
   169 	  movq_r2r(mm2, mm6);					\
   170           pand_r2r(mm3, mm0);					\
   171           punpcklbw_r2r(mm2, mm2);				\
   172 	  pand_r2r(mm3, mm1);					\
   173 	  punpckhbw_r2r(mm6, mm6);				\
   174           pand_r2r(mm3, mm2);					\
   175 	  psubw_r2r(mm2, mm0);					\
   176 	  pmullw_r2r(mm4, mm0);					\
   177 	  pand_r2r(mm3, mm6);					\
   178 	  psubw_r2r(mm6, mm1);					\
   179 	  pmullw_r2r(mm4, mm1);					\
   180 	  psrlw_i2r(8, mm0);					\
   181 	  paddw_r2r(mm0, mm2);					\
   182 	  psrlw_i2r(8, mm1);					\
   183 	  paddw_r2r(mm1, mm6);					\
   184 	  pand_r2r(mm3, mm2);					\
   185 	  pand_r2r(mm3, mm6);					\
   186 	  packuswb_r2r(mm2, mm2);				\
   187 	  packuswb_r2r(mm6, mm6);				\
   188 	  psrlq_i2r(32, mm2);					\
   189 	  psllq_i2r(32, mm6);					\
   190 	  por_r2r(mm6, mm2);					\
   191 	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
   192          movq_r2m(mm2, *dstp);					\
   193 	  srcp += 2;						\
   194 	  dstp += 2;						\
   195 	  i--;							\
   196 	}							\
   197 	emms();							\
   198     } while(0)
   199 
   200 #define ALPHA_BLIT16_565MMX(to, from, length, bpp, alpha)	\
   201     do {						\
   202         int i, n = 0;					\
   203 	Uint16 *srcp = (Uint16 *)(from);		\
   204 	Uint16 *dstp = (Uint16 *)(to);			\
   205         Uint32 ALPHA = 0xF800;				\
   206 	movd_m2r(*(&ALPHA), mm1);			\
   207         punpcklwd_r2r(mm1, mm1);			\
   208         punpcklwd_r2r(mm1, mm1);			\
   209 	ALPHA = 0x07E0;					\
   210 	movd_m2r(*(&ALPHA), mm4);			\
   211         punpcklwd_r2r(mm4, mm4);			\
   212         punpcklwd_r2r(mm4, mm4);			\
   213 	ALPHA = 0x001F;					\
   214 	movd_m2r(*(&ALPHA), mm7);			\
   215         punpcklwd_r2r(mm7, mm7);			\
   216         punpcklwd_r2r(mm7, mm7);			\
   217 	alpha &= ~(1+2+4);				\
   218         i = (Uint32)alpha | (Uint32)alpha << 16;	\
   219         movd_m2r(*(&i), mm0);				\
   220         punpckldq_r2r(mm0, mm0);			\
   221         ALPHA = alpha >> 3;				\
   222         i = ((int)(length) & 3);			\
   223 	for(; i > 0; --i) {				\
   224 	    Uint32 s = *srcp++;				\
   225 	    Uint32 d = *dstp;				\
   226 	    s = (s | s << 16) & 0x07e0f81f;		\
   227 	    d = (d | d << 16) & 0x07e0f81f;		\
   228 	    d += (s - d) * ALPHA >> 5;			\
   229 	    d &= 0x07e0f81f;				\
   230 	    *dstp++ = d | d >> 16;			\
   231 	    n++;					\
   232 	}						\
   233 	i = (int)(length) - n;				\
   234 	for(; i > 0; --i) {				\
   235 	  movq_m2r((*dstp), mm3);			\
   236 	  movq_m2r((*srcp), mm2);			\
   237 	  movq_r2r(mm2, mm5);				\
   238 	  pand_r2r(mm1 , mm5);				\
   239 	  psrlq_i2r(11, mm5);				\
   240 	  movq_r2r(mm3, mm6);				\
   241 	  pand_r2r(mm1 , mm6);				\
   242 	  psrlq_i2r(11, mm6);				\
   243 	  psubw_r2r(mm6, mm5);				\
   244 	  pmullw_r2r(mm0, mm5);				\
   245 	  psrlw_i2r(8, mm5);				\
   246 	  paddw_r2r(mm5, mm6);				\
   247 	  psllq_i2r(11, mm6);				\
   248 	  pand_r2r(mm1, mm6);				\
   249 	  movq_r2r(mm4, mm5);				\
   250 	  por_r2r(mm7, mm5);				\
   251 	  pand_r2r(mm5, mm3);				\
   252 	  por_r2r(mm6, mm3);				\
   253 	  movq_r2r(mm2, mm5);				\
   254 	  pand_r2r(mm4 , mm5);				\
   255 	  psrlq_i2r(5, mm5);				\
   256 	  movq_r2r(mm3, mm6);				\
   257 	  pand_r2r(mm4 , mm6);				\
   258 	  psrlq_i2r(5, mm6);				\
   259 	  psubw_r2r(mm6, mm5);				\
   260 	  pmullw_r2r(mm0, mm5);				\
   261 	  psrlw_i2r(8, mm5);				\
   262 	  paddw_r2r(mm5, mm6);				\
   263 	  psllq_i2r(5, mm6);				\
   264 	  pand_r2r(mm4, mm6);				\
   265 	  movq_r2r(mm1, mm5);				\
   266 	  por_r2r(mm7, mm5);				\
   267 	  pand_r2r(mm5, mm3);				\
   268 	  por_r2r(mm6, mm3);				\
   269 	  movq_r2r(mm2, mm5);				\
   270 	  pand_r2r(mm7 , mm5);				\
   271           movq_r2r(mm3, mm6);				\
   272 	  pand_r2r(mm7 , mm6);				\
   273 	  psubw_r2r(mm6, mm5);				\
   274 	  pmullw_r2r(mm0, mm5);				\
   275 	  psrlw_i2r(8, mm5);				\
   276 	  paddw_r2r(mm5, mm6);				\
   277 	  pand_r2r(mm7, mm6);				\
   278 	  movq_r2r(mm1, mm5);				\
   279 	  por_r2r(mm4, mm5);				\
   280 	  pand_r2r(mm5, mm3);				\
   281 	  por_r2r(mm6, mm3);				\
   282 	  movq_r2m(mm3, *dstp);				\
   283 	  srcp += 4;					\
   284 	  dstp += 4;					\
   285 	  i -= 3;					\
   286 	}						\
   287 	emms();						\
   288     } while(0)
   289 
   290 #define ALPHA_BLIT16_555MMX(to, from, length, bpp, alpha)	\
   291     do {						\
   292         int i, n = 0;					\
   293 	Uint16 *srcp = (Uint16 *)(from);		\
   294 	Uint16 *dstp = (Uint16 *)(to);			\
   295         Uint32 ALPHA = 0x7C00;				\
   296 	movd_m2r(*(&ALPHA), mm1);			\
   297         punpcklwd_r2r(mm1, mm1);			\
   298         punpcklwd_r2r(mm1, mm1);			\
   299 	ALPHA = 0x03E0;					\
   300         movd_m2r(*(&ALPHA), mm4);			\
   301         punpcklwd_r2r(mm4, mm4);			\
   302         punpcklwd_r2r(mm4, mm4);			\
   303 	ALPHA = 0x001F;					\
   304 	movd_m2r(*(&ALPHA), mm7);			\
   305         punpcklwd_r2r(mm7, mm7);			\
   306         punpcklwd_r2r(mm7, mm7);			\
   307 	alpha &= ~(1+2+4);				\
   308         i = (Uint32)alpha | (Uint32)alpha << 16;	\
   309         movd_m2r(*(&i), mm0);				\
   310         punpckldq_r2r(mm0, mm0);			\
   311         i = ((int)(length) & 3);				\
   312         ALPHA = alpha >> 3;				\
   313 	for(; i > 0; --i) {				\
   314 	    Uint32 s = *srcp++;				\
   315 	    Uint32 d = *dstp;				\
   316 	    s = (s | s << 16) & 0x03e07c1f;		\
   317 	    d = (d | d << 16) & 0x03e07c1f;		\
   318 	    d += (s - d) * ALPHA >> 5;			\
   319 	    d &= 0x03e07c1f;				\
   320 	    *dstp++ = d | d >> 16;			\
   321 	    n++;					\
   322 	}						\
   323 	i = (int)(length) - n;				\
   324 	for(; i > 0; --i) {				\
   325 	  movq_m2r((*dstp), mm3);			\
   326 	  movq_m2r((*srcp), mm2);			\
   327 	  movq_r2r(mm2, mm5);				\
   328 	  pand_r2r(mm1 , mm5);				\
   329 	  psrlq_i2r(10, mm5);				\
   330 	  movq_r2r(mm3, mm6);				\
   331 	  pand_r2r(mm1 , mm6);				\
   332 	  psrlq_i2r(10, mm6);				\
   333 	  psubw_r2r(mm6, mm5);				\
   334 	  pmullw_r2r(mm0, mm5);				\
   335 	  psrlw_i2r(8, mm5);				\
   336 	  paddw_r2r(mm5, mm6);				\
   337 	  psllq_i2r(10, mm6);				\
   338 	  pand_r2r(mm1, mm6);				\
   339 	  movq_r2r(mm4, mm5);				\
   340 	  por_r2r(mm7, mm5);				\
   341 	  pand_r2r(mm5, mm3);				\
   342 	  por_r2r(mm6, mm3);				\
   343 	  movq_r2r(mm2, mm5);				\
   344 	  pand_r2r(mm4 , mm5);				\
   345 	  psrlq_i2r(5, mm5);				\
   346 	  movq_r2r(mm3, mm6);				\
   347 	  pand_r2r(mm4 , mm6);				\
   348 	  psrlq_i2r(5, mm6);				\
   349 	  psubw_r2r(mm6, mm5);				\
   350 	  pmullw_r2r(mm0, mm5);				\
   351 	  psrlw_i2r(8, mm5);				\
   352 	  paddw_r2r(mm5, mm6);				\
   353 	  psllq_i2r(5, mm6);				\
   354 	  pand_r2r(mm4, mm6);				\
   355 	  movq_r2r(mm1, mm5);				\
   356 	  por_r2r(mm7, mm5);				\
   357 	  pand_r2r(mm5, mm3);				\
   358 	  por_r2r(mm6, mm3);				\
   359 	  movq_r2r(mm2, mm5);				\
   360 	  pand_r2r(mm7 , mm5);				\
   361           movq_r2r(mm3, mm6);				\
   362 	  pand_r2r(mm7 , mm6);				\
   363 	  psubw_r2r(mm6, mm5);				\
   364 	  pmullw_r2r(mm0, mm5);				\
   365 	  psrlw_i2r(8, mm5);				\
   366 	  paddw_r2r(mm5, mm6);				\
   367 	  pand_r2r(mm7, mm6);				\
   368 	  movq_r2r(mm1, mm5);				\
   369 	  por_r2r(mm4, mm5);				\
   370 	  pand_r2r(mm5, mm3);				\
   371 	  por_r2r(mm6, mm3);				\
   372 	  movq_r2m(mm3, *dstp);				\
   373 	  srcp += 4;					\
   374 	  dstp += 4;					\
   375 	  i -= 3;					\
   376 	}						\
   377 	emms();						\
   378     } while(0)
   379 
   380 #endif
   381 
   382 /*
   383  * For 32bpp pixels on the form 0x00rrggbb:
   384  * If we treat the middle component separately, we can process the two
   385  * remaining in parallel. This is safe to do because of the gap to the left
   386  * of each component, so the bits from the multiplication don't collide.
   387  * This can be used for any RGB permutation of course.
   388  */
   389 #define ALPHA_BLIT32_888(to, from, length, bpp, alpha)		\
   390     do {							\
   391         int i;							\
   392 	Uint32 *src = (Uint32 *)(from);				\
   393 	Uint32 *dst = (Uint32 *)(to);				\
   394 	for(i = 0; i < (int)(length); i++) {			\
   395 	    Uint32 s = *src++;					\
   396 	    Uint32 d = *dst;					\
   397 	    Uint32 s1 = s & 0xff00ff;				\
   398 	    Uint32 d1 = d & 0xff00ff;				\
   399 	    d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;	\
   400 	    s &= 0xff00;					\
   401 	    d &= 0xff00;					\
   402 	    d = (d + ((s - d) * alpha >> 8)) & 0xff00;		\
   403 	    *dst++ = d1 | d;					\
   404 	}							\
   405     } while(0)
   406 
   407 /*
   408  * For 16bpp pixels we can go a step further: put the middle component
   409  * in the high 16 bits of a 32 bit word, and process all three RGB
   410  * components at the same time. Since the smallest gap is here just
   411  * 5 bits, we have to scale alpha down to 5 bits as well.
   412  */
   413 #define ALPHA_BLIT16_565(to, from, length, bpp, alpha)	\
   414     do {						\
   415         int i;						\
   416 	Uint16 *src = (Uint16 *)(from);			\
   417 	Uint16 *dst = (Uint16 *)(to);			\
   418 	Uint32 ALPHA = alpha >> 3;			\
   419 	for(i = 0; i < (int)(length); i++) {		\
   420 	    Uint32 s = *src++;				\
   421 	    Uint32 d = *dst;				\
   422 	    s = (s | s << 16) & 0x07e0f81f;		\
   423 	    d = (d | d << 16) & 0x07e0f81f;		\
   424 	    d += (s - d) * ALPHA >> 5;			\
   425 	    d &= 0x07e0f81f;				\
   426 	    *dst++ = (Uint16)(d | d >> 16);			\
   427 	}						\
   428     } while(0)
   429 
   430 #define ALPHA_BLIT16_555(to, from, length, bpp, alpha)	\
   431     do {						\
   432         int i;						\
   433 	Uint16 *src = (Uint16 *)(from);			\
   434 	Uint16 *dst = (Uint16 *)(to);			\
   435 	Uint32 ALPHA = alpha >> 3;			\
   436 	for(i = 0; i < (int)(length); i++) {		\
   437 	    Uint32 s = *src++;				\
   438 	    Uint32 d = *dst;				\
   439 	    s = (s | s << 16) & 0x03e07c1f;		\
   440 	    d = (d | d << 16) & 0x03e07c1f;		\
   441 	    d += (s - d) * ALPHA >> 5;			\
   442 	    d &= 0x03e07c1f;				\
   443 	    *dst++ = (Uint16)(d | d >> 16);			\
   444 	}						\
   445     } while(0)
   446 
   447 /*
   448  * The general slow catch-all function, for remaining depths and formats
   449  */
   450 #define ALPHA_BLIT_ANY(to, from, length, bpp, alpha)			\
   451     do {								\
   452         int i;								\
   453 	Uint8 *src = from;						\
   454 	Uint8 *dst = to;						\
   455 	for(i = 0; i < (int)(length); i++) {				\
   456 	    Uint32 s, d;						\
   457 	    unsigned rs, gs, bs, rd, gd, bd;				\
   458 	    switch(bpp) {						\
   459 	    case 2:							\
   460 		s = *(Uint16 *)src;					\
   461 		d = *(Uint16 *)dst;					\
   462 		break;							\
   463 	    case 3:							\
   464 		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
   465 		    s = (src[0] << 16) | (src[1] << 8) | src[2];	\
   466 		    d = (dst[0] << 16) | (dst[1] << 8) | dst[2];	\
   467 		} else {						\
   468 		    s = (src[2] << 16) | (src[1] << 8) | src[0];	\
   469 		    d = (dst[2] << 16) | (dst[1] << 8) | dst[0];	\
   470 		}							\
   471 		break;							\
   472 	    case 4:							\
   473 		s = *(Uint32 *)src;					\
   474 		d = *(Uint32 *)dst;					\
   475 		break;							\
   476 	    }								\
   477 	    RGB_FROM_PIXEL(s, fmt, rs, gs, bs);				\
   478 	    RGB_FROM_PIXEL(d, fmt, rd, gd, bd);				\
   479 	    rd += (rs - rd) * alpha >> 8;				\
   480 	    gd += (gs - gd) * alpha >> 8;				\
   481 	    bd += (bs - bd) * alpha >> 8;				\
   482 	    PIXEL_FROM_RGB(d, fmt, rd, gd, bd);				\
   483 	    switch(bpp) {						\
   484 	    case 2:							\
   485 		*(Uint16 *)dst = (Uint16)d;					\
   486 		break;							\
   487 	    case 3:							\
   488 		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
   489 		    dst[0] = (Uint8)(d >> 16);					\
   490 		    dst[1] = (Uint8)(d >> 8);					\
   491 		    dst[2] = (Uint8)(d);						\
   492 		} else {						\
   493 		    dst[0] = (Uint8)d;						\
   494 		    dst[1] = (Uint8)(d >> 8);					\
   495 		    dst[2] = (Uint8)(d >> 16);					\
   496 		}							\
   497 		break;							\
   498 	    case 4:							\
   499 		*(Uint32 *)dst = d;					\
   500 		break;							\
   501 	    }								\
   502 	    src += bpp;							\
   503 	    dst += bpp;							\
   504 	}								\
   505     } while(0)
   506 
   507 #ifdef MMX_ASMBLIT
   508 
   509 #define ALPHA_BLIT32_888_50MMX(to, from, length, bpp, alpha)		\
   510     do {								\
   511 	Uint32 *srcp = (Uint32 *)(from);				\
   512 	Uint32 *dstp = (Uint32 *)(to);					\
   513         int i = 0x00fefefe;						\
   514         movd_m2r(*(&i), mm4);						\
   515         punpckldq_r2r(mm4, mm4);					\
   516         i = 0x00010101;							\
   517         movd_m2r(*(&i), mm3);						\
   518         punpckldq_r2r(mm3, mm3);					\
   519         i = (int)(length);						\
   520         if( i & 1 ) {							\
   521 	  Uint32 s = *srcp++;						\
   522 	  Uint32 d = *dstp;						\
   523 	  *dstp++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
   524 		     + (s & d & 0x00010101);				\
   525 	  i--;								\
   526 	}								\
   527 	for(; i > 0; --i) {						\
   528 	    movq_m2r((*dstp), mm2); /* dst -> mm2 */			\
   529 	    movq_r2r(mm2, mm6);	/* dst -> mm6 */			\
   530 	    movq_m2r((*srcp), mm1); /* src -> mm1 */			\
   531 	    movq_r2r(mm1, mm5);	/* src -> mm5 */			\
   532 	    pand_r2r(mm4, mm6);	/* dst & 0x00fefefe -> mm6 */		\
   533 	    pand_r2r(mm4, mm5); /* src & 0x00fefefe -> mm5 */		\
   534 	    paddd_r2r(mm6, mm5); /* (dst & 0x00fefefe) + (dst & 0x00fefefe) -> mm5 */	\
   535 	    psrld_i2r(1, mm5);						\
   536 	    pand_r2r(mm1, mm2);	/* s & d -> mm2 */			\
   537 	    pand_r2r(mm3, mm2);	/* s & d & 0x00010101 -> mm2 */		\
   538 	    paddd_r2r(mm5, mm2);					\
   539 	    movq_r2m(mm2, (*dstp));					\
   540 	    dstp += 2;							\
   541 	    srcp += 2;							\
   542 	    i--;							\
   543 	}								\
   544 	emms();								\
   545     } while(0)
   546 
   547 #endif
   548     
   549 /*
   550  * Special case: 50% alpha (alpha=128)
   551  * This is treated specially because it can be optimized very well, and
   552  * since it is good for many cases of semi-translucency.
   553  * The theory is to do all three components at the same time:
   554  * First zero the lowest bit of each component, which gives us room to
   555  * add them. Then shift right and add the sum of the lowest bits.
   556  */
   557 #define ALPHA_BLIT32_888_50(to, from, length, bpp, alpha)		\
   558     do {								\
   559         int i;								\
   560 	Uint32 *src = (Uint32 *)(from);					\
   561 	Uint32 *dst = (Uint32 *)(to);					\
   562 	for(i = 0; i < (int)(length); i++) {				\
   563 	    Uint32 s = *src++;						\
   564 	    Uint32 d = *dst;						\
   565 	    *dst++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
   566 		     + (s & d & 0x00010101);				\
   567 	}								\
   568     } while(0)
   569 
   570 /*
   571  * For 16bpp, we can actually blend two pixels in parallel, if we take
   572  * care to shift before we add, not after.
   573  */
   574 
   575 /* helper: blend a single 16 bit pixel at 50% */
   576 #define BLEND16_50(dst, src, mask)			\
   577     do {						\
   578 	Uint32 s = *src++;				\
   579 	Uint32 d = *dst;				\
   580 	*dst++ = (Uint16)((((s & mask) + (d & mask)) >> 1) +	\
   581 	                  (s & d & (~mask & 0xffff)));		\
   582     } while(0)
   583 
   584 /* basic 16bpp blender. mask is the pixels to keep when adding. */
   585 #define ALPHA_BLIT16_50(to, from, length, bpp, alpha, mask)		\
   586     do {								\
   587 	unsigned n = (length);						\
   588 	Uint16 *src = (Uint16 *)(from);					\
   589 	Uint16 *dst = (Uint16 *)(to);					\
   590 	if(((uintptr_t)src ^ (uintptr_t)dst) & 3) {			\
   591 	    /* source and destination not in phase, blit one by one */	\
   592 	    while(n--)							\
   593 		BLEND16_50(dst, src, mask);				\
   594 	} else {							\
   595 	    if((uintptr_t)src & 3) {					\
   596 		/* first odd pixel */					\
   597 		BLEND16_50(dst, src, mask);				\
   598 		n--;							\
   599 	    }								\
   600 	    for(; n > 1; n -= 2) {					\
   601 		Uint32 s = *(Uint32 *)src;				\
   602 		Uint32 d = *(Uint32 *)dst;				\
   603 		*(Uint32 *)dst = ((s & (mask | mask << 16)) >> 1)	\
   604 		               + ((d & (mask | mask << 16)) >> 1)	\
   605 		               + (s & d & (~(mask | mask << 16)));	\
   606 		src += 2;						\
   607 		dst += 2;						\
   608 	    }								\
   609 	    if(n)							\
   610 		BLEND16_50(dst, src, mask); /* last odd pixel */	\
   611 	}								\
   612     } while(0)
   613 
   614 #define ALPHA_BLIT16_565_50(to, from, length, bpp, alpha)	\
   615     ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xf7de)
   616 
   617 #define ALPHA_BLIT16_555_50(to, from, length, bpp, alpha)	\
   618     ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xfbde)
   619 
   620 #ifdef MMX_ASMBLIT
   621 
   622 #define CHOOSE_BLIT(blitter, alpha, fmt)				\
   623     do {								\
   624         if(alpha == 255) {						\
   625 	    switch(fmt->BytesPerPixel) {				\
   626 	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
   627 	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
   628 	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
   629 	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
   630 	    }								\
   631 	} else {							\
   632 	    switch(fmt->BytesPerPixel) {				\
   633 	    case 1:							\
   634 		/* No 8bpp alpha blitting */				\
   635 		break;							\
   636 									\
   637 	    case 2:							\
   638 		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
   639 		case 0xffff:						\
   640 		    if(fmt->Gmask == 0x07e0				\
   641 		       || fmt->Rmask == 0x07e0				\
   642 		       || fmt->Bmask == 0x07e0) {			\
   643 			if(alpha == 128)				\
   644 			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
   645 			else {						\
   646 			    if(SDL_HasMMX())				\
   647 				blitter(2, Uint8, ALPHA_BLIT16_565MMX);	\
   648 			    else					\
   649 				blitter(2, Uint8, ALPHA_BLIT16_565);	\
   650 			}						\
   651 		    } else						\
   652 			goto general16;					\
   653 		    break;						\
   654 									\
   655 		case 0x7fff:						\
   656 		    if(fmt->Gmask == 0x03e0				\
   657 		       || fmt->Rmask == 0x03e0				\
   658 		       || fmt->Bmask == 0x03e0) {			\
   659 			if(alpha == 128)				\
   660 			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
   661 			else {						\
   662 			    if(SDL_HasMMX())				\
   663 				blitter(2, Uint8, ALPHA_BLIT16_555MMX);	\
   664 			    else					\
   665 				blitter(2, Uint8, ALPHA_BLIT16_555);	\
   666 			}						\
   667 			break;						\
   668 		    }							\
   669 		    /* fallthrough */					\
   670 									\
   671 		default:						\
   672 		general16:						\
   673 		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
   674 		}							\
   675 		break;							\
   676 									\
   677 	    case 3:							\
   678 		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
   679 		break;							\
   680 									\
   681 	    case 4:							\
   682 		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
   683 		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
   684 		       || fmt->Bmask == 0xff00)) {			\
   685 		    if(alpha == 128)					\
   686 		    {							\
   687 			if(SDL_HasMMX())				\
   688 				blitter(4, Uint16, ALPHA_BLIT32_888_50MMX);\
   689 			else						\
   690 				blitter(4, Uint16, ALPHA_BLIT32_888_50);\
   691 		    }							\
   692 		    else						\
   693 		    {							\
   694 			if(SDL_HasMMX())				\
   695 				blitter(4, Uint16, ALPHA_BLIT32_888MMX);\
   696 			else						\
   697 				blitter(4, Uint16, ALPHA_BLIT32_888);	\
   698 		    }							\
   699 		} else							\
   700 		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
   701 		break;							\
   702 	    }								\
   703 	}								\
   704     } while(0)
   705 
   706 #else
   707 	
   708 #define CHOOSE_BLIT(blitter, alpha, fmt)				\
   709     do {								\
   710         if(alpha == 255) {						\
   711 	    switch(fmt->BytesPerPixel) {				\
   712 	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
   713 	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
   714 	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
   715 	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
   716 	    }								\
   717 	} else {							\
   718 	    switch(fmt->BytesPerPixel) {				\
   719 	    case 1:							\
   720 		/* No 8bpp alpha blitting */				\
   721 		break;							\
   722 									\
   723 	    case 2:							\
   724 		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
   725 		case 0xffff:						\
   726 		    if(fmt->Gmask == 0x07e0				\
   727 		       || fmt->Rmask == 0x07e0				\
   728 		       || fmt->Bmask == 0x07e0) {			\
   729 			if(alpha == 128)				\
   730 			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
   731 			else {						\
   732 			    blitter(2, Uint8, ALPHA_BLIT16_565);	\
   733 			}						\
   734 		    } else						\
   735 			goto general16;					\
   736 		    break;						\
   737 									\
   738 		case 0x7fff:						\
   739 		    if(fmt->Gmask == 0x03e0				\
   740 		       || fmt->Rmask == 0x03e0				\
   741 		       || fmt->Bmask == 0x03e0) {			\
   742 			if(alpha == 128)				\
   743 			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
   744 			else {						\
   745 			    blitter(2, Uint8, ALPHA_BLIT16_555);	\
   746 			}						\
   747 			break;						\
   748 		    }							\
   749 		    /* fallthrough */					\
   750 									\
   751 		default:						\
   752 		general16:						\
   753 		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
   754 		}							\
   755 		break;							\
   756 									\
   757 	    case 3:							\
   758 		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
   759 		break;							\
   760 									\
   761 	    case 4:							\
   762 		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
   763 		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
   764 		       || fmt->Bmask == 0xff00)) {			\
   765 		    if(alpha == 128)					\
   766 			blitter(4, Uint16, ALPHA_BLIT32_888_50);	\
   767 		    else						\
   768 			blitter(4, Uint16, ALPHA_BLIT32_888);		\
   769 		} else							\
   770 		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
   771 		break;							\
   772 	    }								\
   773 	}								\
   774     } while(0)
   775 
   776 #endif
   777 
   778 /*
   779  * This takes care of the case when the surface is clipped on the left and/or
   780  * right. Top clipping has already been taken care of.
   781  */
   782 static void RLEClipBlit(int w, Uint8 *srcbuf, SDL_Surface *dst,
   783 			Uint8 *dstbuf, SDL_Rect *srcrect, unsigned alpha)
   784 {
   785     SDL_PixelFormat *fmt = dst->format;
   786 
   787 #define RLECLIPBLIT(bpp, Type, do_blit)					   \
   788     do {								   \
   789 	int linecount = srcrect->h;					   \
   790 	int ofs = 0;							   \
   791 	int left = srcrect->x;						   \
   792 	int right = left + srcrect->w;					   \
   793 	dstbuf -= left * bpp;						   \
   794 	for(;;) {							   \
   795 	    int run;							   \
   796 	    ofs += *(Type *)srcbuf;					   \
   797 	    run = ((Type *)srcbuf)[1];					   \
   798 	    srcbuf += 2 * sizeof(Type);					   \
   799 	    if(run) {							   \
   800 		/* clip to left and right borders */			   \
   801 		if(ofs < right) {					   \
   802 		    int start = 0;					   \
   803 		    int len = run;					   \
   804 		    int startcol;					   \
   805 		    if(left - ofs > 0) {				   \
   806 			start = left - ofs;				   \
   807 			len -= start;					   \
   808 			if(len <= 0)					   \
   809 			    goto nocopy ## bpp ## do_blit;		   \
   810 		    }							   \
   811 		    startcol = ofs + start;				   \
   812 		    if(len > right - startcol)				   \
   813 			len = right - startcol;				   \
   814 		    do_blit(dstbuf + startcol * bpp, srcbuf + start * bpp, \
   815 			    len, bpp, alpha);				   \
   816 		}							   \
   817 	    nocopy ## bpp ## do_blit:					   \
   818 		srcbuf += run * bpp;					   \
   819 		ofs += run;						   \
   820 	    } else if(!ofs)						   \
   821 		break;							   \
   822 	    if(ofs == w) {						   \
   823 		ofs = 0;						   \
   824 		dstbuf += dst->pitch;					   \
   825 		if(!--linecount)					   \
   826 		    break;						   \
   827 	    }								   \
   828 	}								   \
   829     } while(0)
   830 
   831     CHOOSE_BLIT(RLECLIPBLIT, alpha, fmt);
   832 
   833 #undef RLECLIPBLIT
   834 
   835 }
   836 
   837 
   838 /* blit a colorkeyed RLE surface */
   839 int SDL_RLEBlit(SDL_Surface *src, SDL_Rect *srcrect,
   840 		SDL_Surface *dst, SDL_Rect *dstrect)
   841 {
   842 	Uint8 *dstbuf;
   843 	Uint8 *srcbuf;
   844 	int x, y;
   845 	int w = src->w;
   846 	unsigned alpha;
   847 
   848 	/* Lock the destination if necessary */
   849 	if ( SDL_MUSTLOCK(dst) ) {
   850 		if ( SDL_LockSurface(dst) < 0 ) {
   851 			return(-1);
   852 		}
   853 	}
   854 
   855 	/* Set up the source and destination pointers */
   856 	x = dstrect->x;
   857 	y = dstrect->y;
   858 	dstbuf = (Uint8 *)dst->pixels
   859 	         + y * dst->pitch + x * src->format->BytesPerPixel;
   860 	srcbuf = (Uint8 *)src->map->sw_data->aux_data;
   861 
   862 	{
   863 	    /* skip lines at the top if neccessary */
   864 	    int vskip = srcrect->y;
   865 	    int ofs = 0;
   866 	    if(vskip) {
   867 
   868 #define RLESKIP(bpp, Type)			\
   869 		for(;;) {			\
   870 		    int run;			\
   871 		    ofs += *(Type *)srcbuf;	\
   872 		    run = ((Type *)srcbuf)[1];	\
   873 		    srcbuf += sizeof(Type) * 2;	\
   874 		    if(run) {			\
   875 			srcbuf += run * bpp;	\
   876 			ofs += run;		\
   877 		    } else if(!ofs)		\
   878 			goto done;		\
   879 		    if(ofs == w) {		\
   880 			ofs = 0;		\
   881 			if(!--vskip)		\
   882 			    break;		\
   883 		    }				\
   884 		}
   885 
   886 		switch(src->format->BytesPerPixel) {
   887 		case 1: RLESKIP(1, Uint8); break;
   888 		case 2: RLESKIP(2, Uint8); break;
   889 		case 3: RLESKIP(3, Uint8); break;
   890 		case 4: RLESKIP(4, Uint16); break;
   891 		}
   892 
   893 #undef RLESKIP
   894 
   895 	    }
   896 	}
   897 
   898 	alpha = (src->flags & SDL_SRCALPHA) == SDL_SRCALPHA
   899 	        ? src->format->alpha : 255;
   900 	/* if left or right edge clipping needed, call clip blit */
   901 	if ( srcrect->x || srcrect->w != src->w ) {
   902 	    RLEClipBlit(w, srcbuf, dst, dstbuf, srcrect, alpha);
   903 	} else {
   904 	    SDL_PixelFormat *fmt = src->format;
   905 
   906 #define RLEBLIT(bpp, Type, do_blit)					      \
   907 	    do {							      \
   908 		int linecount = srcrect->h;				      \
   909 		int ofs = 0;						      \
   910 		for(;;) {						      \
   911 		    unsigned run;					      \
   912 		    ofs += *(Type *)srcbuf;				      \
   913 		    run = ((Type *)srcbuf)[1];				      \
   914 		    srcbuf += 2 * sizeof(Type);				      \
   915 		    if(run) {						      \
   916 			do_blit(dstbuf + ofs * bpp, srcbuf, run, bpp, alpha); \
   917 			srcbuf += run * bpp;				      \
   918 			ofs += run;					      \
   919 		    } else if(!ofs)					      \
   920 			break;						      \
   921 		    if(ofs == w) {					      \
   922 			ofs = 0;					      \
   923 			dstbuf += dst->pitch;				      \
   924 			if(!--linecount)				      \
   925 			    break;					      \
   926 		    }							      \
   927 		}							      \
   928 	    } while(0)
   929 
   930 	    CHOOSE_BLIT(RLEBLIT, alpha, fmt);
   931 
   932 #undef RLEBLIT
   933 	}
   934 
   935 done:
   936 	/* Unlock the destination if necessary */
   937 	if ( SDL_MUSTLOCK(dst) ) {
   938 		SDL_UnlockSurface(dst);
   939 	}
   940 	return(0);
   941 }
   942 
   943 #undef OPAQUE_BLIT
   944 
   945 /*
   946  * Per-pixel blitting macros for translucent pixels:
   947  * These use the same techniques as the per-surface blitting macros
   948  */
   949 
   950 /*
   951  * For 32bpp pixels, we have made sure the alpha is stored in the top
   952  * 8 bits, so proceed as usual
   953  */
   954 #define BLIT_TRANSL_888(src, dst)				\
   955     do {							\
   956         Uint32 s = src;						\
   957 	Uint32 d = dst;						\
   958 	unsigned alpha = s >> 24;				\
   959 	Uint32 s1 = s & 0xff00ff;				\
   960 	Uint32 d1 = d & 0xff00ff;				\
   961 	d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;	\
   962 	s &= 0xff00;						\
   963 	d &= 0xff00;						\
   964 	d = (d + ((s - d) * alpha >> 8)) & 0xff00;		\
   965 	dst = d1 | d;						\
   966     } while(0)
   967 
   968 /*
   969  * For 16bpp pixels, we have stored the 5 most significant alpha bits in
   970  * bits 5-10. As before, we can process all 3 RGB components at the same time.
   971  */
   972 #define BLIT_TRANSL_565(src, dst)		\
   973     do {					\
   974 	Uint32 s = src;				\
   975 	Uint32 d = dst;				\
   976 	unsigned alpha = (s & 0x3e0) >> 5;	\
   977 	s &= 0x07e0f81f;			\
   978 	d = (d | d << 16) & 0x07e0f81f;		\
   979 	d += (s - d) * alpha >> 5;		\
   980 	d &= 0x07e0f81f;			\
   981 	dst = (Uint16)(d | d >> 16);			\
   982     } while(0)
   983 
   984 #define BLIT_TRANSL_555(src, dst)		\
   985     do {					\
   986 	Uint32 s = src;				\
   987 	Uint32 d = dst;				\
   988 	unsigned alpha = (s & 0x3e0) >> 5;	\
   989 	s &= 0x03e07c1f;			\
   990 	d = (d | d << 16) & 0x03e07c1f;		\
   991 	d += (s - d) * alpha >> 5;		\
   992 	d &= 0x03e07c1f;			\
   993 	dst = (Uint16)(d | d >> 16);			\
   994     } while(0)
   995 
   996 /* used to save the destination format in the encoding. Designed to be
   997    macro-compatible with SDL_PixelFormat but without the unneeded fields */
   998 typedef struct {
   999 	Uint8  BytesPerPixel;
  1000 	Uint8  Rloss;
  1001 	Uint8  Gloss;
  1002 	Uint8  Bloss;
  1003 	Uint8  Rshift;
  1004 	Uint8  Gshift;
  1005 	Uint8  Bshift;
  1006 	Uint8  Ashift;
  1007 	Uint32 Rmask;
  1008 	Uint32 Gmask;
  1009 	Uint32 Bmask;
  1010 	Uint32 Amask;
  1011 } RLEDestFormat;
  1012 
  1013 /* blit a pixel-alpha RLE surface clipped at the right and/or left edges */
  1014 static void RLEAlphaClipBlit(int w, Uint8 *srcbuf, SDL_Surface *dst,
  1015 			     Uint8 *dstbuf, SDL_Rect *srcrect)
  1016 {
  1017     SDL_PixelFormat *df = dst->format;
  1018     /*
  1019      * clipped blitter: Ptype is the destination pixel type,
  1020      * Ctype the translucent count type, and do_blend the macro
  1021      * to blend one pixel.
  1022      */
  1023 #define RLEALPHACLIPBLIT(Ptype, Ctype, do_blend)			  \
  1024     do {								  \
  1025 	int linecount = srcrect->h;					  \
  1026 	int left = srcrect->x;						  \
  1027 	int right = left + srcrect->w;					  \
  1028 	dstbuf -= left * sizeof(Ptype);					  \
  1029 	do {								  \
  1030 	    int ofs = 0;						  \
  1031 	    /* blit opaque pixels on one line */			  \
  1032 	    do {							  \
  1033 		unsigned run;						  \
  1034 		ofs += ((Ctype *)srcbuf)[0];				  \
  1035 		run = ((Ctype *)srcbuf)[1];				  \
  1036 		srcbuf += 2 * sizeof(Ctype);				  \
  1037 		if(run) {						  \
  1038 		    /* clip to left and right borders */		  \
  1039 		    int cofs = ofs;					  \
  1040 		    int crun = run;					  \
  1041 		    if(left - cofs > 0) {				  \
  1042 			crun -= left - cofs;				  \
  1043 			cofs = left;					  \
  1044 		    }							  \
  1045 		    if(crun > right - cofs)				  \
  1046 			crun = right - cofs;				  \
  1047 		    if(crun > 0)					  \
  1048 			PIXEL_COPY(dstbuf + cofs * sizeof(Ptype),	  \
  1049 				   srcbuf + (cofs - ofs) * sizeof(Ptype), \
  1050 				   (unsigned)crun, sizeof(Ptype));	  \
  1051 		    srcbuf += run * sizeof(Ptype);			  \
  1052 		    ofs += run;						  \
  1053 		} else if(!ofs)						  \
  1054 		    return;						  \
  1055 	    } while(ofs < w);						  \
  1056 	    /* skip padding if necessary */				  \
  1057 	    if(sizeof(Ptype) == 2)					  \
  1058 		srcbuf += (uintptr_t)srcbuf & 2;			  \
  1059 	    /* blit translucent pixels on the same line */		  \
  1060 	    ofs = 0;							  \
  1061 	    do {							  \
  1062 		unsigned run;						  \
  1063 		ofs += ((Uint16 *)srcbuf)[0];				  \
  1064 		run = ((Uint16 *)srcbuf)[1];				  \
  1065 		srcbuf += 4;						  \
  1066 		if(run) {						  \
  1067 		    /* clip to left and right borders */		  \
  1068 		    int cofs = ofs;					  \
  1069 		    int crun = run;					  \
  1070 		    if(left - cofs > 0) {				  \
  1071 			crun -= left - cofs;				  \
  1072 			cofs = left;					  \
  1073 		    }							  \
  1074 		    if(crun > right - cofs)				  \
  1075 			crun = right - cofs;				  \
  1076 		    if(crun > 0) {					  \
  1077 			Ptype *dst = (Ptype *)dstbuf + cofs;		  \
  1078 			Uint32 *src = (Uint32 *)srcbuf + (cofs - ofs);	  \
  1079 			int i;						  \
  1080 			for(i = 0; i < crun; i++)			  \
  1081 			    do_blend(src[i], dst[i]);			  \
  1082 		    }							  \
  1083 		    srcbuf += run * 4;					  \
  1084 		    ofs += run;						  \
  1085 		}							  \
  1086 	    } while(ofs < w);						  \
  1087 	    dstbuf += dst->pitch;					  \
  1088 	} while(--linecount);						  \
  1089     } while(0)
  1090 
  1091     switch(df->BytesPerPixel) {
  1092     case 2:
  1093 	if(df->Gmask == 0x07e0 || df->Rmask == 0x07e0
  1094 	   || df->Bmask == 0x07e0)
  1095 	    RLEALPHACLIPBLIT(Uint16, Uint8, BLIT_TRANSL_565);
  1096 	else
  1097 	    RLEALPHACLIPBLIT(Uint16, Uint8, BLIT_TRANSL_555);
  1098 	break;
  1099     case 4:
  1100 	RLEALPHACLIPBLIT(Uint32, Uint16, BLIT_TRANSL_888);
  1101 	break;
  1102     }
  1103 }
  1104 
  1105 /* blit a pixel-alpha RLE surface */
  1106 int SDL_RLEAlphaBlit(SDL_Surface *src, SDL_Rect *srcrect,
  1107 		     SDL_Surface *dst, SDL_Rect *dstrect)
  1108 {
  1109     int x, y;
  1110     int w = src->w;
  1111     Uint8 *srcbuf, *dstbuf;
  1112     SDL_PixelFormat *df = dst->format;
  1113 
  1114     /* Lock the destination if necessary */
  1115     if ( SDL_MUSTLOCK(dst) ) {
  1116 	if ( SDL_LockSurface(dst) < 0 ) {
  1117 	    return -1;
  1118 	}
  1119     }
  1120 
  1121     x = dstrect->x;
  1122     y = dstrect->y;
  1123     dstbuf = (Uint8 *)dst->pixels
  1124 	     + y * dst->pitch + x * df->BytesPerPixel;
  1125     srcbuf = (Uint8 *)src->map->sw_data->aux_data + sizeof(RLEDestFormat);
  1126 
  1127     {
  1128 	/* skip lines at the top if necessary */
  1129 	int vskip = srcrect->y;
  1130 	if(vskip) {
  1131 	    int ofs;
  1132 	    if(df->BytesPerPixel == 2) {
  1133 		/* the 16/32 interleaved format */
  1134 		do {
  1135 		    /* skip opaque line */
  1136 		    ofs = 0;
  1137 		    do {
  1138 			int run;
  1139 			ofs += srcbuf[0];
  1140 			run = srcbuf[1];
  1141 			srcbuf += 2;
  1142 			if(run) {
  1143 			    srcbuf += 2 * run;
  1144 			    ofs += run;
  1145 			} else if(!ofs)
  1146 			    goto done;
  1147 		    } while(ofs < w);
  1148 
  1149 		    /* skip padding */
  1150 		    srcbuf += (uintptr_t)srcbuf & 2;
  1151 
  1152 		    /* skip translucent line */
  1153 		    ofs = 0;
  1154 		    do {
  1155 			int run;
  1156 			ofs += ((Uint16 *)srcbuf)[0];
  1157 			run = ((Uint16 *)srcbuf)[1];
  1158 			srcbuf += 4 * (run + 1);
  1159 			ofs += run;
  1160 		    } while(ofs < w);
  1161 		} while(--vskip);
  1162 	    } else {
  1163 		/* the 32/32 interleaved format */
  1164 		vskip <<= 1;	/* opaque and translucent have same format */
  1165 		do {
  1166 		    ofs = 0;
  1167 		    do {
  1168 			int run;
  1169 			ofs += ((Uint16 *)srcbuf)[0];
  1170 			run = ((Uint16 *)srcbuf)[1];
  1171 			srcbuf += 4;
  1172 			if(run) {
  1173 			    srcbuf += 4 * run;
  1174 			    ofs += run;
  1175 			} else if(!ofs)
  1176 			    goto done;
  1177 		    } while(ofs < w);
  1178 		} while(--vskip);
  1179 	    }
  1180 	}
  1181     }
  1182 
  1183     /* if left or right edge clipping needed, call clip blit */
  1184     if(srcrect->x || srcrect->w != src->w) {
  1185 	RLEAlphaClipBlit(w, srcbuf, dst, dstbuf, srcrect);
  1186     } else {
  1187 
  1188 	/*
  1189 	 * non-clipped blitter. Ptype is the destination pixel type,
  1190 	 * Ctype the translucent count type, and do_blend the
  1191 	 * macro to blend one pixel.
  1192 	 */
  1193 #define RLEALPHABLIT(Ptype, Ctype, do_blend)				 \
  1194 	do {								 \
  1195 	    int linecount = srcrect->h;					 \
  1196 	    do {							 \
  1197 		int ofs = 0;						 \
  1198 		/* blit opaque pixels on one line */			 \
  1199 		do {							 \
  1200 		    unsigned run;					 \
  1201 		    ofs += ((Ctype *)srcbuf)[0];			 \
  1202 		    run = ((Ctype *)srcbuf)[1];				 \
  1203 		    srcbuf += 2 * sizeof(Ctype);			 \
  1204 		    if(run) {						 \
  1205 			PIXEL_COPY(dstbuf + ofs * sizeof(Ptype), srcbuf, \
  1206 				   run, sizeof(Ptype));			 \
  1207 			srcbuf += run * sizeof(Ptype);			 \
  1208 			ofs += run;					 \
  1209 		    } else if(!ofs)					 \
  1210 			goto done;					 \
  1211 		} while(ofs < w);					 \
  1212 		/* skip padding if necessary */				 \
  1213 		if(sizeof(Ptype) == 2)					 \
  1214 		    srcbuf += (uintptr_t)srcbuf & 2;		 	 \
  1215 		/* blit translucent pixels on the same line */		 \
  1216 		ofs = 0;						 \
  1217 		do {							 \
  1218 		    unsigned run;					 \
  1219 		    ofs += ((Uint16 *)srcbuf)[0];			 \
  1220 		    run = ((Uint16 *)srcbuf)[1];			 \
  1221 		    srcbuf += 4;					 \
  1222 		    if(run) {						 \
  1223 			Ptype *dst = (Ptype *)dstbuf + ofs;		 \
  1224 			unsigned i;					 \
  1225 			for(i = 0; i < run; i++) {			 \
  1226 			    Uint32 src = *(Uint32 *)srcbuf;		 \
  1227 			    do_blend(src, *dst);			 \
  1228 			    srcbuf += 4;				 \
  1229 			    dst++;					 \
  1230 			}						 \
  1231 			ofs += run;					 \
  1232 		    }							 \
  1233 		} while(ofs < w);					 \
  1234 		dstbuf += dst->pitch;					 \
  1235 	    } while(--linecount);					 \
  1236 	} while(0)
  1237 
  1238 	switch(df->BytesPerPixel) {
  1239 	case 2:
  1240 	    if(df->Gmask == 0x07e0 || df->Rmask == 0x07e0
  1241 	       || df->Bmask == 0x07e0)
  1242 		RLEALPHABLIT(Uint16, Uint8, BLIT_TRANSL_565);
  1243 	    else
  1244 		RLEALPHABLIT(Uint16, Uint8, BLIT_TRANSL_555);
  1245 	    break;
  1246 	case 4:
  1247 	    RLEALPHABLIT(Uint32, Uint16, BLIT_TRANSL_888);
  1248 	    break;
  1249 	}
  1250     }
  1251 
  1252  done:
  1253     /* Unlock the destination if necessary */
  1254     if ( SDL_MUSTLOCK(dst) ) {
  1255 	SDL_UnlockSurface(dst);
  1256     }
  1257     return 0;
  1258 }
  1259 
  1260 /*
  1261  * Auxiliary functions:
  1262  * The encoding functions take 32bpp rgb + a, and
  1263  * return the number of bytes copied to the destination.
  1264  * The decoding functions copy to 32bpp rgb + a, and
  1265  * return the number of bytes copied from the source.
  1266  * These are only used in the encoder and un-RLE code and are therefore not
  1267  * highly optimised.
  1268  */
  1269 
  1270 /* encode 32bpp rgb + a into 16bpp rgb, losing alpha */
  1271 static int copy_opaque_16(void *dst, Uint32 *src, int n,
  1272 			  SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1273 {
  1274     int i;
  1275     Uint16 *d = dst;
  1276     for(i = 0; i < n; i++) {
  1277 	unsigned r, g, b;
  1278 	RGB_FROM_PIXEL(*src, sfmt, r, g, b);
  1279 	PIXEL_FROM_RGB(*d, dfmt, r, g, b);
  1280 	src++;
  1281 	d++;
  1282     }
  1283     return n * 2;
  1284 }
  1285 
  1286 /* decode opaque pixels from 16bpp to 32bpp rgb + a */
  1287 static int uncopy_opaque_16(Uint32 *dst, void *src, int n,
  1288 			    RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1289 {
  1290     int i;
  1291     Uint16 *s = src;
  1292     unsigned alpha = dfmt->Amask ? 255 : 0;
  1293     for(i = 0; i < n; i++) {
  1294 	unsigned r, g, b;
  1295 	RGB_FROM_PIXEL(*s, sfmt, r, g, b);
  1296 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, alpha);
  1297 	s++;
  1298 	dst++;
  1299     }
  1300     return n * 2;
  1301 }
  1302 
  1303 
  1304 
  1305 /* encode 32bpp rgb + a into 32bpp G0RAB format for blitting into 565 */
  1306 static int copy_transl_565(void *dst, Uint32 *src, int n,
  1307 			   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1308 {
  1309     int i;
  1310     Uint32 *d = dst;
  1311     for(i = 0; i < n; i++) {
  1312 	unsigned r, g, b, a;
  1313 	Uint16 pix;
  1314 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1315 	PIXEL_FROM_RGB(pix, dfmt, r, g, b);
  1316 	*d = ((pix & 0x7e0) << 16) | (pix & 0xf81f) | ((a << 2) & 0x7e0);
  1317 	src++;
  1318 	d++;
  1319     }
  1320     return n * 4;
  1321 }
  1322 
  1323 /* encode 32bpp rgb + a into 32bpp G0RAB format for blitting into 555 */
  1324 static int copy_transl_555(void *dst, Uint32 *src, int n,
  1325 			   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1326 {
  1327     int i;
  1328     Uint32 *d = dst;
  1329     for(i = 0; i < n; i++) {
  1330 	unsigned r, g, b, a;
  1331 	Uint16 pix;
  1332 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1333 	PIXEL_FROM_RGB(pix, dfmt, r, g, b);
  1334 	*d = ((pix & 0x3e0) << 16) | (pix & 0xfc1f) | ((a << 2) & 0x3e0);
  1335 	src++;
  1336 	d++;
  1337     }
  1338     return n * 4;
  1339 }
  1340 
  1341 /* decode translucent pixels from 32bpp GORAB to 32bpp rgb + a */
  1342 static int uncopy_transl_16(Uint32 *dst, void *src, int n,
  1343 			    RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1344 {
  1345     int i;
  1346     Uint32 *s = src;
  1347     for(i = 0; i < n; i++) {
  1348 	unsigned r, g, b, a;
  1349 	Uint32 pix = *s++;
  1350 	a = (pix & 0x3e0) >> 2;
  1351 	pix = (pix & ~0x3e0) | pix >> 16;
  1352 	RGB_FROM_PIXEL(pix, sfmt, r, g, b);
  1353 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, a);
  1354 	dst++;
  1355     }
  1356     return n * 4;
  1357 }
  1358 
  1359 /* encode 32bpp rgba into 32bpp rgba, keeping alpha (dual purpose) */
  1360 static int copy_32(void *dst, Uint32 *src, int n,
  1361 		   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1362 {
  1363     int i;
  1364     Uint32 *d = dst;
  1365     for(i = 0; i < n; i++) {
  1366 	unsigned r, g, b, a;
  1367 	Uint32 pixel;
  1368 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1369 	PIXEL_FROM_RGB(pixel, dfmt, r, g, b);
  1370 	*d++ = pixel | a << 24;
  1371 	src++;
  1372     }
  1373     return n * 4;
  1374 }
  1375 
  1376 /* decode 32bpp rgba into 32bpp rgba, keeping alpha (dual purpose) */
  1377 static int uncopy_32(Uint32 *dst, void *src, int n,
  1378 		     RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1379 {
  1380     int i;
  1381     Uint32 *s = src;
  1382     for(i = 0; i < n; i++) {
  1383 	unsigned r, g, b, a;
  1384 	Uint32 pixel = *s++;
  1385 	RGB_FROM_PIXEL(pixel, sfmt, r, g, b);
  1386 	a = pixel >> 24;
  1387 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, a);
  1388 	dst++;
  1389     }
  1390     return n * 4;
  1391 }
  1392 
  1393 #define ISOPAQUE(pixel, fmt) ((((pixel) & fmt->Amask) >> fmt->Ashift) == 255)
  1394 
  1395 #define ISTRANSL(pixel, fmt)	\
  1396     ((unsigned)((((pixel) & fmt->Amask) >> fmt->Ashift) - 1U) < 254U)
  1397 
  1398 /* convert surface to be quickly alpha-blittable onto dest, if possible */
  1399 static int RLEAlphaSurface(SDL_Surface *surface)
  1400 {
  1401     SDL_Surface *dest;
  1402     SDL_PixelFormat *df;
  1403     int maxsize = 0;
  1404     int max_opaque_run;
  1405     int max_transl_run = 65535;
  1406     unsigned masksum;
  1407     Uint8 *rlebuf, *dst;
  1408     int (*copy_opaque)(void *, Uint32 *, int,
  1409 		       SDL_PixelFormat *, SDL_PixelFormat *);
  1410     int (*copy_transl)(void *, Uint32 *, int,
  1411 		       SDL_PixelFormat *, SDL_PixelFormat *);
  1412 
  1413     dest = surface->map->dst;
  1414     if(!dest)
  1415 	return -1;
  1416     df = dest->format;
  1417     if(surface->format->BitsPerPixel != 32)
  1418 	return -1;		/* only 32bpp source supported */
  1419 
  1420     /* find out whether the destination is one we support,
  1421        and determine the max size of the encoded result */
  1422     masksum = df->Rmask | df->Gmask | df->Bmask;
  1423     switch(df->BytesPerPixel) {
  1424     case 2:
  1425 	/* 16bpp: only support 565 and 555 formats */
  1426 	switch(masksum) {
  1427 	case 0xffff:
  1428 	    if(df->Gmask == 0x07e0
  1429 	       || df->Rmask == 0x07e0 || df->Bmask == 0x07e0) {
  1430 		copy_opaque = copy_opaque_16;
  1431 		copy_transl = copy_transl_565;
  1432 	    } else
  1433 		return -1;
  1434 	    break;
  1435 	case 0x7fff:
  1436 	    if(df->Gmask == 0x03e0
  1437 	       || df->Rmask == 0x03e0 || df->Bmask == 0x03e0) {
  1438 		copy_opaque = copy_opaque_16;
  1439 		copy_transl = copy_transl_555;
  1440 	    } else
  1441 		return -1;
  1442 	    break;
  1443 	default:
  1444 	    return -1;
  1445 	}
  1446 	max_opaque_run = 255;	/* runs stored as bytes */
  1447 
  1448 	/* worst case is alternating opaque and translucent pixels,
  1449 	   with room for alignment padding between lines */
  1450 	maxsize = surface->h * (2 + (4 + 2) * (surface->w + 1)) + 2;
  1451 	break;
  1452     case 4:
  1453 	if(masksum != 0x00ffffff)
  1454 	    return -1;		/* requires unused high byte */
  1455 	copy_opaque = copy_32;
  1456 	copy_transl = copy_32;
  1457 	max_opaque_run = 255;	/* runs stored as short ints */
  1458 
  1459 	/* worst case is alternating opaque and translucent pixels */
  1460 	maxsize = surface->h * 2 * 4 * (surface->w + 1) + 4;
  1461 	break;
  1462     default:
  1463 	return -1;		/* anything else unsupported right now */
  1464     }
  1465 
  1466     maxsize += sizeof(RLEDestFormat);
  1467     rlebuf = (Uint8 *)SDL_malloc(maxsize);
  1468     if(!rlebuf) {
  1469 	SDL_OutOfMemory();
  1470 	return -1;
  1471     }
  1472     {
  1473 	/* save the destination format so we can undo the encoding later */
  1474 	RLEDestFormat *r = (RLEDestFormat *)rlebuf;
  1475 	r->BytesPerPixel = df->BytesPerPixel;
  1476 	r->Rloss = df->Rloss;
  1477 	r->Gloss = df->Gloss;
  1478 	r->Bloss = df->Bloss;
  1479 	r->Rshift = df->Rshift;
  1480 	r->Gshift = df->Gshift;
  1481 	r->Bshift = df->Bshift;
  1482 	r->Ashift = df->Ashift;
  1483 	r->Rmask = df->Rmask;
  1484 	r->Gmask = df->Gmask;
  1485 	r->Bmask = df->Bmask;
  1486 	r->Amask = df->Amask;
  1487     }
  1488     dst = rlebuf + sizeof(RLEDestFormat);
  1489 
  1490     /* Do the actual encoding */
  1491     {
  1492 	int x, y;
  1493 	int h = surface->h, w = surface->w;
  1494 	SDL_PixelFormat *sf = surface->format;
  1495 	Uint32 *src = (Uint32 *)surface->pixels;
  1496 	Uint8 *lastline = dst;	/* end of last non-blank line */
  1497 
  1498 	/* opaque counts are 8 or 16 bits, depending on target depth */
  1499 #define ADD_OPAQUE_COUNTS(n, m)			\
  1500 	if(df->BytesPerPixel == 4) {		\
  1501 	    ((Uint16 *)dst)[0] = n;		\
  1502 	    ((Uint16 *)dst)[1] = m;		\
  1503 	    dst += 4;				\
  1504 	} else {				\
  1505 	    dst[0] = n;				\
  1506 	    dst[1] = m;				\
  1507 	    dst += 2;				\
  1508 	}
  1509 
  1510 	/* translucent counts are always 16 bit */
  1511 #define ADD_TRANSL_COUNTS(n, m)		\
  1512 	(((Uint16 *)dst)[0] = n, ((Uint16 *)dst)[1] = m, dst += 4)
  1513 
  1514 	for(y = 0; y < h; y++) {
  1515 	    int runstart, skipstart;
  1516 	    int blankline = 0;
  1517 	    /* First encode all opaque pixels of a scan line */
  1518 	    x = 0;
  1519 	    do {
  1520 		int run, skip, len;
  1521 		skipstart = x;
  1522 		while(x < w && !ISOPAQUE(src[x], sf))
  1523 		    x++;
  1524 		runstart = x;
  1525 		while(x < w && ISOPAQUE(src[x], sf))
  1526 		    x++;
  1527 		skip = runstart - skipstart;
  1528 		if(skip == w)
  1529 		    blankline = 1;
  1530 		run = x - runstart;
  1531 		while(skip > max_opaque_run) {
  1532 		    ADD_OPAQUE_COUNTS(max_opaque_run, 0);
  1533 		    skip -= max_opaque_run;
  1534 		}
  1535 		len = MIN(run, max_opaque_run);
  1536 		ADD_OPAQUE_COUNTS(skip, len);
  1537 		dst += copy_opaque(dst, src + runstart, len, sf, df);
  1538 		runstart += len;
  1539 		run -= len;
  1540 		while(run) {
  1541 		    len = MIN(run, max_opaque_run);
  1542 		    ADD_OPAQUE_COUNTS(0, len);
  1543 		    dst += copy_opaque(dst, src + runstart, len, sf, df);
  1544 		    runstart += len;
  1545 		    run -= len;
  1546 		}
  1547 	    } while(x < w);
  1548 
  1549 	    /* Make sure the next output address is 32-bit aligned */
  1550 	    dst += (uintptr_t)dst & 2;
  1551 
  1552 	    /* Next, encode all translucent pixels of the same scan line */
  1553 	    x = 0;
  1554 	    do {
  1555 		int run, skip, len;
  1556 		skipstart = x;
  1557 		while(x < w && !ISTRANSL(src[x], sf))
  1558 		    x++;
  1559 		runstart = x;
  1560 		while(x < w && ISTRANSL(src[x], sf))
  1561 		    x++;
  1562 		skip = runstart - skipstart;
  1563 		blankline &= (skip == w);
  1564 		run = x - runstart;
  1565 		while(skip > max_transl_run) {
  1566 		    ADD_TRANSL_COUNTS(max_transl_run, 0);
  1567 		    skip -= max_transl_run;
  1568 		}
  1569 		len = MIN(run, max_transl_run);
  1570 		ADD_TRANSL_COUNTS(skip, len);
  1571 		dst += copy_transl(dst, src + runstart, len, sf, df);
  1572 		runstart += len;
  1573 		run -= len;
  1574 		while(run) {
  1575 		    len = MIN(run, max_transl_run);
  1576 		    ADD_TRANSL_COUNTS(0, len);
  1577 		    dst += copy_transl(dst, src + runstart, len, sf, df);
  1578 		    runstart += len;
  1579 		    run -= len;
  1580 		}
  1581 		if(!blankline)
  1582 		    lastline = dst;
  1583 	    } while(x < w);
  1584 
  1585 	    src += surface->pitch >> 2;
  1586 	}
  1587 	dst = lastline;		/* back up past trailing blank lines */
  1588 	ADD_OPAQUE_COUNTS(0, 0);
  1589     }
  1590 
  1591 #undef ADD_OPAQUE_COUNTS
  1592 #undef ADD_TRANSL_COUNTS
  1593 
  1594     /* Now that we have it encoded, release the original pixels */
  1595     if((surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1596        && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1597 	SDL_free( surface->pixels );
  1598 	surface->pixels = NULL;
  1599     }
  1600 
  1601     /* realloc the buffer to release unused memory */
  1602     {
  1603 	Uint8 *p = SDL_realloc(rlebuf, dst - rlebuf);
  1604 	if(!p)
  1605 	    p = rlebuf;
  1606 	surface->map->sw_data->aux_data = p;
  1607     }
  1608 
  1609     return 0;
  1610 }
  1611 
  1612 static Uint32 getpix_8(Uint8 *srcbuf)
  1613 {
  1614     return *srcbuf;
  1615 }
  1616 
  1617 static Uint32 getpix_16(Uint8 *srcbuf)
  1618 {
  1619     return *(Uint16 *)srcbuf;
  1620 }
  1621 
  1622 static Uint32 getpix_24(Uint8 *srcbuf)
  1623 {
  1624 #if SDL_BYTEORDER == SDL_LIL_ENDIAN
  1625     return srcbuf[0] + (srcbuf[1] << 8) + (srcbuf[2] << 16);
  1626 #else
  1627     return (srcbuf[0] << 16) + (srcbuf[1] << 8) + srcbuf[2];
  1628 #endif
  1629 }
  1630 
  1631 static Uint32 getpix_32(Uint8 *srcbuf)
  1632 {
  1633     return *(Uint32 *)srcbuf;
  1634 }
  1635 
  1636 typedef Uint32 (*getpix_func)(Uint8 *);
  1637 
  1638 static getpix_func getpixes[4] = {
  1639     getpix_8, getpix_16, getpix_24, getpix_32
  1640 };
  1641 
  1642 static int RLEColorkeySurface(SDL_Surface *surface)
  1643 {
  1644         Uint8 *rlebuf, *dst;
  1645 	int maxn;
  1646 	int y;
  1647 	Uint8 *srcbuf, *curbuf, *lastline;
  1648 	int maxsize = 0;
  1649 	int skip, run;
  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 	curbuf = srcbuf;
  1684 	maxn = bpp == 4 ? 65535 : 255;
  1685 	skip = run = 0;
  1686 	dst = rlebuf;
  1687 	rgbmask = ~surface->format->Amask;
  1688 	ckey = surface->format->colorkey & rgbmask;
  1689 	lastline = dst;
  1690 	getpix = getpixes[bpp - 1];
  1691 	w = surface->w;
  1692 	h = surface->h;
  1693 
  1694 #define ADD_COUNTS(n, m)			\
  1695 	if(bpp == 4) {				\
  1696 	    ((Uint16 *)dst)[0] = n;		\
  1697 	    ((Uint16 *)dst)[1] = m;		\
  1698 	    dst += 4;				\
  1699 	} else {				\
  1700 	    dst[0] = n;				\
  1701 	    dst[1] = m;				\
  1702 	    dst += 2;				\
  1703 	}
  1704 
  1705 	for(y = 0; y < h; y++) {
  1706 	    int x = 0;
  1707 	    int blankline = 0;
  1708 	    do {
  1709 		int run, skip, len;
  1710 		int runstart;
  1711 		int skipstart = x;
  1712 
  1713 		/* find run of transparent, then opaque pixels */
  1714 		while(x < w && (getpix(srcbuf + x * bpp) & rgbmask) == ckey)
  1715 		    x++;
  1716 		runstart = x;
  1717 		while(x < w && (getpix(srcbuf + x * bpp) & rgbmask) != ckey)
  1718 		    x++;
  1719 		skip = runstart - skipstart;
  1720 		if(skip == w)
  1721 		    blankline = 1;
  1722 		run = x - runstart;
  1723 
  1724 		/* encode segment */
  1725 		while(skip > maxn) {
  1726 		    ADD_COUNTS(maxn, 0);
  1727 		    skip -= maxn;
  1728 		}
  1729 		len = MIN(run, maxn);
  1730 		ADD_COUNTS(skip, len);
  1731 		SDL_memcpy(dst, srcbuf + runstart * bpp, len * bpp);
  1732 		dst += len * bpp;
  1733 		run -= len;
  1734 		runstart += len;
  1735 		while(run) {
  1736 		    len = MIN(run, maxn);
  1737 		    ADD_COUNTS(0, len);
  1738 		    SDL_memcpy(dst, srcbuf + runstart * bpp, len * bpp);
  1739 		    dst += len * bpp;
  1740 		    runstart += len;
  1741 		    run -= len;
  1742 		}
  1743 		if(!blankline)
  1744 		    lastline = dst;
  1745 	    } while(x < w);
  1746 
  1747 	    srcbuf += surface->pitch;
  1748 	}
  1749 	dst = lastline;		/* back up bast trailing blank lines */
  1750 	ADD_COUNTS(0, 0);
  1751 
  1752 #undef ADD_COUNTS
  1753 
  1754 	/* Now that we have it encoded, release the original pixels */
  1755 	if((surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1756 	   && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1757 	    SDL_free( surface->pixels );
  1758 	    surface->pixels = NULL;
  1759 	}
  1760 
  1761 	/* realloc the buffer to release unused memory */
  1762 	{
  1763 	    /* If realloc returns NULL, the original block is left intact */
  1764 	    Uint8 *p = SDL_realloc(rlebuf, dst - rlebuf);
  1765 	    if(!p)
  1766 		p = rlebuf;
  1767 	    surface->map->sw_data->aux_data = p;
  1768 	}
  1769 
  1770 	return(0);
  1771 }
  1772 
  1773 int SDL_RLESurface(SDL_Surface *surface)
  1774 {
  1775 	int retcode;
  1776 
  1777 	/* Clear any previous RLE conversion */
  1778 	if ( (surface->flags & SDL_RLEACCEL) == SDL_RLEACCEL ) {
  1779 		SDL_UnRLESurface(surface, 1);
  1780 	}
  1781 
  1782 	/* We don't support RLE encoding of bitmaps */
  1783 	if ( surface->format->BitsPerPixel < 8 ) {
  1784 		return(-1);
  1785 	}
  1786 
  1787 	/* Lock the surface if it's in hardware */
  1788 	if ( SDL_MUSTLOCK(surface) ) {
  1789 		if ( SDL_LockSurface(surface) < 0 ) {
  1790 			return(-1);
  1791 		}
  1792 	}
  1793 
  1794 	/* Encode */
  1795 	if((surface->flags & SDL_SRCCOLORKEY) == SDL_SRCCOLORKEY) {
  1796 	    retcode = RLEColorkeySurface(surface);
  1797 	} else {
  1798 	    if((surface->flags & SDL_SRCALPHA) == SDL_SRCALPHA
  1799 	       && surface->format->Amask != 0)
  1800 		retcode = RLEAlphaSurface(surface);
  1801 	    else
  1802 		retcode = -1;	/* no RLE for per-surface alpha sans ckey */
  1803 	}
  1804 
  1805 	/* Unlock the surface if it's in hardware */
  1806 	if ( SDL_MUSTLOCK(surface) ) {
  1807 		SDL_UnlockSurface(surface);
  1808 	}
  1809 
  1810 	if(retcode < 0)
  1811 	    return -1;
  1812 
  1813 	/* The surface is now accelerated */
  1814 	surface->flags |= SDL_RLEACCEL;
  1815 
  1816 	return(0);
  1817 }
  1818 
  1819 /*
  1820  * Un-RLE a surface with pixel alpha
  1821  * This may not give back exactly the image before RLE-encoding; all
  1822  * completely transparent pixels will be lost, and colour and alpha depth
  1823  * may have been reduced (when encoding for 16bpp targets).
  1824  */
  1825 static SDL_bool UnRLEAlpha(SDL_Surface *surface)
  1826 {
  1827     Uint8 *srcbuf;
  1828     Uint32 *dst;
  1829     SDL_PixelFormat *sf = surface->format;
  1830     RLEDestFormat *df = surface->map->sw_data->aux_data;
  1831     int (*uncopy_opaque)(Uint32 *, void *, int,
  1832 			 RLEDestFormat *, SDL_PixelFormat *);
  1833     int (*uncopy_transl)(Uint32 *, void *, int,
  1834 			 RLEDestFormat *, SDL_PixelFormat *);
  1835     int w = surface->w;
  1836     int bpp = df->BytesPerPixel;
  1837 
  1838     if(bpp == 2) {
  1839 	uncopy_opaque = uncopy_opaque_16;
  1840 	uncopy_transl = uncopy_transl_16;
  1841     } else {
  1842 	uncopy_opaque = uncopy_transl = uncopy_32;
  1843     }
  1844 
  1845     surface->pixels = SDL_malloc(surface->h * surface->pitch);
  1846     if ( !surface->pixels ) {
  1847         return(SDL_FALSE);
  1848     }
  1849     /* fill background with transparent pixels */
  1850     SDL_memset(surface->pixels, 0, surface->h * surface->pitch);
  1851 
  1852     dst = surface->pixels;
  1853     srcbuf = (Uint8 *)(df + 1);
  1854     for(;;) {
  1855 	/* copy opaque pixels */
  1856 	int ofs = 0;
  1857 	do {
  1858 	    unsigned run;
  1859 	    if(bpp == 2) {
  1860 		ofs += srcbuf[0];
  1861 		run = srcbuf[1];
  1862 		srcbuf += 2;
  1863 	    } else {
  1864 		ofs += ((Uint16 *)srcbuf)[0];
  1865 		run = ((Uint16 *)srcbuf)[1];
  1866 		srcbuf += 4;
  1867 	    }
  1868 	    if(run) {
  1869 		srcbuf += uncopy_opaque(dst + ofs, srcbuf, run, df, sf);
  1870 		ofs += run;
  1871 	    } else if(!ofs)
  1872 		return(SDL_TRUE);
  1873 	} while(ofs < w);
  1874 
  1875 	/* skip padding if needed */
  1876 	if(bpp == 2)
  1877 	    srcbuf += (uintptr_t)srcbuf & 2;
  1878 	
  1879 	/* copy translucent pixels */
  1880 	ofs = 0;
  1881 	do {
  1882 	    unsigned run;
  1883 	    ofs += ((Uint16 *)srcbuf)[0];
  1884 	    run = ((Uint16 *)srcbuf)[1];
  1885 	    srcbuf += 4;
  1886 	    if(run) {
  1887 		srcbuf += uncopy_transl(dst + ofs, srcbuf, run, df, sf);
  1888 		ofs += run;
  1889 	    }
  1890 	} while(ofs < w);
  1891 	dst += surface->pitch >> 2;
  1892     }
  1893     /* Make the compiler happy */
  1894     return(SDL_TRUE);
  1895 }
  1896 
  1897 void SDL_UnRLESurface(SDL_Surface *surface, int recode)
  1898 {
  1899     if ( (surface->flags & SDL_RLEACCEL) == SDL_RLEACCEL ) {
  1900 	surface->flags &= ~SDL_RLEACCEL;
  1901 
  1902 	if(recode && (surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1903 	   && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1904 	    if((surface->flags & SDL_SRCCOLORKEY) == SDL_SRCCOLORKEY) {
  1905 		SDL_Rect full;
  1906 		unsigned alpha_flag;
  1907 
  1908 		/* re-create the original surface */
  1909 		surface->pixels = SDL_malloc(surface->h * surface->pitch);
  1910 		if ( !surface->pixels ) {
  1911 			/* Oh crap... */
  1912 			surface->flags |= SDL_RLEACCEL;
  1913 			return;
  1914 		}
  1915 
  1916 		/* fill it with the background colour */
  1917 		SDL_FillRect(surface, NULL, surface->format->colorkey);
  1918 
  1919 		/* now render the encoded surface */
  1920 		full.x = full.y = 0;
  1921 		full.w = surface->w;
  1922 		full.h = surface->h;
  1923 		alpha_flag = surface->flags & SDL_SRCALPHA;
  1924 		surface->flags &= ~SDL_SRCALPHA; /* opaque blit */
  1925 		SDL_RLEBlit(surface, &full, surface, &full);
  1926 		surface->flags |= alpha_flag;
  1927 	    } else {
  1928 		if ( !UnRLEAlpha(surface) ) {
  1929 		    /* Oh crap... */
  1930 		    surface->flags |= SDL_RLEACCEL;
  1931 		    return;
  1932 		}
  1933 	    }
  1934 	}
  1935 
  1936 	if ( surface->map && surface->map->sw_data->aux_data ) {
  1937 	    SDL_free(surface->map->sw_data->aux_data);
  1938 	    surface->map->sw_data->aux_data = NULL;
  1939 	}
  1940     }
  1941 }
  1942 
  1943