src/video/SDL_RLEaccel.c
author Sam Lantinga <slouken@libsdl.org>
Wed, 01 Feb 2006 06:32:25 +0000
changeset 1312 c9b51268668f
parent 1155 91569ec25acd
child 1330 450721ad5436
permissions -rw-r--r--
Updated copyright information and removed rcs id lines (problematic in branch merges)
I batch edited these files, so please let me know if I've accidentally removed anybody's
credit here.
     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 
    23 /*
    24  * RLE encoding for software colorkey and alpha-channel acceleration
    25  *
    26  * Original version by Sam Lantinga
    27  *
    28  * Mattias Engdegård (Yorick): Rewrite. New encoding format, encoder and
    29  * decoder. Added per-surface alpha blitter. Added per-pixel alpha
    30  * format, encoder and blitter.
    31  *
    32  * Many thanks to Xark and johns for hints, benchmarks and useful comments
    33  * leading to this code.
    34  *
    35  * Welcome to Macro Mayhem.
    36  */
    37 
    38 /*
    39  * The encoding translates the image data to a stream of segments of the form
    40  *
    41  * <skip> <run> <data>
    42  *
    43  * where <skip> is the number of transparent pixels to skip,
    44  *       <run>  is the number of opaque pixels to blit,
    45  * and   <data> are the pixels themselves.
    46  *
    47  * This basic structure is used both for colorkeyed surfaces, used for simple
    48  * binary transparency and for per-surface alpha blending, and for surfaces
    49  * with per-pixel alpha. The details differ, however:
    50  *
    51  * Encoding of colorkeyed surfaces:
    52  *
    53  *   Encoded pixels always have the same format as the target surface.
    54  *   <skip> and <run> are unsigned 8 bit integers, except for 32 bit depth
    55  *   where they are 16 bit. This makes the pixel data aligned at all times.
    56  *   Segments never wrap around from one scan line to the next.
    57  *
    58  *   The end of the sequence is marked by a zero <skip>,<run> pair at the *
    59  *   beginning of a line.
    60  *
    61  * Encoding of surfaces with per-pixel alpha:
    62  *
    63  *   The sequence begins with a struct RLEDestFormat describing the target
    64  *   pixel format, to provide reliable un-encoding.
    65  *
    66  *   Each scan line is encoded twice: First all completely opaque pixels,
    67  *   encoded in the target format as described above, and then all
    68  *   partially transparent (translucent) pixels (where 1 <= alpha <= 254),
    69  *   in the following 32-bit format:
    70  *
    71  *   For 32-bit targets, each pixel has the target RGB format but with
    72  *   the alpha value occupying the highest 8 bits. The <skip> and <run>
    73  *   counts are 16 bit.
    74  * 
    75  *   For 16-bit targets, each pixel has the target RGB format, but with
    76  *   the middle component (usually green) shifted 16 steps to the left,
    77  *   and the hole filled with the 5 most significant bits of the alpha value.
    78  *   i.e. if the target has the format         rrrrrggggggbbbbb,
    79  *   the encoded pixel will be 00000gggggg00000rrrrr0aaaaabbbbb.
    80  *   The <skip> and <run> counts are 8 bit for the opaque lines, 16 bit
    81  *   for the translucent lines. Two padding bytes may be inserted
    82  *   before each translucent line to keep them 32-bit aligned.
    83  *
    84  *   The end of the sequence is marked by a zero <skip>,<run> pair at the
    85  *   beginning of an opaque line.
    86  */
    87 
    88 #include <stdio.h>
    89 #include <stdlib.h>
    90 #include <string.h>
    91 
    92 #include "SDL_types.h"
    93 #include "SDL_video.h"
    94 #include "SDL_error.h"
    95 #include "SDL_sysvideo.h"
    96 #include "SDL_blit.h"
    97 #include "SDL_memops.h"
    98 #include "SDL_RLEaccel_c.h"
    99 
   100 #if (defined(i386) || defined(__x86_64__)) && defined(__GNUC__) && defined(USE_ASMBLIT)
   101 #define MMX_ASMBLIT
   102 #endif
   103 
   104 #ifdef MMX_ASMBLIT
   105 #include "mmx.h"
   106 #include "SDL_cpuinfo.h"
   107 #endif
   108 
   109 #ifndef MAX
   110 #define MAX(a, b) ((a) > (b) ? (a) : (b))
   111 #endif
   112 #ifndef MIN
   113 #define MIN(a, b) ((a) < (b) ? (a) : (b))
   114 #endif
   115 
   116 #define PIXEL_COPY(to, from, len, bpp)			\
   117 do {							\
   118     if(bpp == 4) {					\
   119 	SDL_memcpy4(to, from, (unsigned)(len));		\
   120     } else {						\
   121 	SDL_memcpy(to, from, (unsigned)(len) * (bpp));	\
   122     }							\
   123 } while(0)
   124 
   125 /*
   126  * Various colorkey blit methods, for opaque and per-surface alpha
   127  */
   128 
   129 #define OPAQUE_BLIT(to, from, length, bpp, alpha)	\
   130     PIXEL_COPY(to, from, length, bpp)
   131 
   132 #ifdef MMX_ASMBLIT
   133 
   134 #define ALPHA_BLIT32_888MMX(to, from, length, bpp, alpha)	\
   135     do {							\
   136 	Uint32 *srcp = (Uint32 *)(from);			\
   137 	Uint32 *dstp = (Uint32 *)(to);				\
   138         int i = 0x00FF00FF;					\
   139         movd_m2r(*(&i), mm3);					\
   140         punpckldq_r2r(mm3, mm3);				\
   141         i = 0xFF000000;						\
   142         movd_m2r(*(&i), mm7);					\
   143         punpckldq_r2r(mm7, mm7);				\
   144         i = alpha | alpha << 16;				\
   145         movd_m2r(*(&i), mm4);					\
   146         punpckldq_r2r(mm4, mm4);				\
   147 	pcmpeqd_r2r(mm5,mm5); /* set mm5 to "1" */		\
   148 	pxor_r2r(mm7, mm5); /* make clear alpha mask */		\
   149         i = length;						\
   150 	if(i & 1) {						\
   151           movd_m2r((*srcp), mm1); /* src -> mm1 */		\
   152           punpcklbw_r2r(mm1, mm1);				\
   153           pand_r2r(mm3, mm1);					\
   154 	  movd_m2r((*dstp), mm2); /* dst -> mm2 */		\
   155           punpcklbw_r2r(mm2, mm2);				\
   156           pand_r2r(mm3, mm2);					\
   157 	  psubw_r2r(mm2, mm1);					\
   158 	  pmullw_r2r(mm4, mm1);					\
   159 	  psrlw_i2r(8, mm1);					\
   160 	  paddw_r2r(mm1, mm2);					\
   161 	  pand_r2r(mm3, mm2);					\
   162 	  packuswb_r2r(mm2, mm2);				\
   163 	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
   164 	  movd_r2m(mm2, *dstp);					\
   165 	  ++srcp;						\
   166 	  ++dstp;						\
   167 	  i--;							\
   168 	}							\
   169 	for(; i > 0; --i) {					\
   170           movq_m2r((*srcp), mm0);				\
   171 	  movq_r2r(mm0, mm1);					\
   172           punpcklbw_r2r(mm0, mm0);				\
   173 	  movq_m2r((*dstp), mm2);				\
   174 	  punpckhbw_r2r(mm1, mm1);				\
   175 	  movq_r2r(mm2, mm6);					\
   176           pand_r2r(mm3, mm0);					\
   177           punpcklbw_r2r(mm2, mm2);				\
   178 	  pand_r2r(mm3, mm1);					\
   179 	  punpckhbw_r2r(mm6, mm6);				\
   180           pand_r2r(mm3, mm2);					\
   181 	  psubw_r2r(mm2, mm0);					\
   182 	  pmullw_r2r(mm4, mm0);					\
   183 	  pand_r2r(mm3, mm6);					\
   184 	  psubw_r2r(mm6, mm1);					\
   185 	  pmullw_r2r(mm4, mm1);					\
   186 	  psrlw_i2r(8, mm0);					\
   187 	  paddw_r2r(mm0, mm2);					\
   188 	  psrlw_i2r(8, mm1);					\
   189 	  paddw_r2r(mm1, mm6);					\
   190 	  pand_r2r(mm3, mm2);					\
   191 	  pand_r2r(mm3, mm6);					\
   192 	  packuswb_r2r(mm2, mm2);				\
   193 	  packuswb_r2r(mm6, mm6);				\
   194 	  psrlq_i2r(32, mm2);					\
   195 	  psllq_i2r(32, mm6);					\
   196 	  por_r2r(mm6, mm2);					\
   197 	  pand_r2r(mm5, mm2); /* 00000RGB -> mm2 */		\
   198          movq_r2m(mm2, *dstp);					\
   199 	  srcp += 2;						\
   200 	  dstp += 2;						\
   201 	  i--;							\
   202 	}							\
   203 	emms();							\
   204     } while(0)
   205 
   206 #define ALPHA_BLIT16_565MMX(to, from, length, bpp, alpha)	\
   207     do {						\
   208         int i, n = 0;					\
   209 	Uint16 *srcp = (Uint16 *)(from);		\
   210 	Uint16 *dstp = (Uint16 *)(to);			\
   211         Uint32 ALPHA = 0xF800;				\
   212 	movd_m2r(*(&ALPHA), mm1);			\
   213         punpcklwd_r2r(mm1, mm1);			\
   214         punpcklwd_r2r(mm1, mm1);			\
   215 	ALPHA = 0x07E0;					\
   216 	movd_m2r(*(&ALPHA), mm4);			\
   217         punpcklwd_r2r(mm4, mm4);			\
   218         punpcklwd_r2r(mm4, mm4);			\
   219 	ALPHA = 0x001F;					\
   220 	movd_m2r(*(&ALPHA), mm7);			\
   221         punpcklwd_r2r(mm7, mm7);			\
   222         punpcklwd_r2r(mm7, mm7);			\
   223 	alpha &= ~(1+2+4);				\
   224         i = (Uint32)alpha | (Uint32)alpha << 16;	\
   225         movd_m2r(*(&i), mm0);				\
   226         punpckldq_r2r(mm0, mm0);			\
   227         ALPHA = alpha >> 3;				\
   228         i = ((int)(length) & 3);			\
   229 	for(; i > 0; --i) {				\
   230 	    Uint32 s = *srcp++;				\
   231 	    Uint32 d = *dstp;				\
   232 	    s = (s | s << 16) & 0x07e0f81f;		\
   233 	    d = (d | d << 16) & 0x07e0f81f;		\
   234 	    d += (s - d) * ALPHA >> 5;			\
   235 	    d &= 0x07e0f81f;				\
   236 	    *dstp++ = d | d >> 16;			\
   237 	    n++;					\
   238 	}						\
   239 	i = (int)(length) - n;				\
   240 	for(; i > 0; --i) {				\
   241 	  movq_m2r((*dstp), mm3);			\
   242 	  movq_m2r((*srcp), mm2);			\
   243 	  movq_r2r(mm2, mm5);				\
   244 	  pand_r2r(mm1 , mm5);				\
   245 	  psrlq_i2r(11, mm5);				\
   246 	  movq_r2r(mm3, mm6);				\
   247 	  pand_r2r(mm1 , mm6);				\
   248 	  psrlq_i2r(11, mm6);				\
   249 	  psubw_r2r(mm6, mm5);				\
   250 	  pmullw_r2r(mm0, mm5);				\
   251 	  psrlw_i2r(8, mm5);				\
   252 	  paddw_r2r(mm5, mm6);				\
   253 	  psllq_i2r(11, mm6);				\
   254 	  pand_r2r(mm1, mm6);				\
   255 	  movq_r2r(mm4, mm5);				\
   256 	  por_r2r(mm7, mm5);				\
   257 	  pand_r2r(mm5, mm3);				\
   258 	  por_r2r(mm6, mm3);				\
   259 	  movq_r2r(mm2, mm5);				\
   260 	  pand_r2r(mm4 , mm5);				\
   261 	  psrlq_i2r(5, mm5);				\
   262 	  movq_r2r(mm3, mm6);				\
   263 	  pand_r2r(mm4 , mm6);				\
   264 	  psrlq_i2r(5, mm6);				\
   265 	  psubw_r2r(mm6, mm5);				\
   266 	  pmullw_r2r(mm0, mm5);				\
   267 	  psrlw_i2r(8, mm5);				\
   268 	  paddw_r2r(mm5, mm6);				\
   269 	  psllq_i2r(5, mm6);				\
   270 	  pand_r2r(mm4, mm6);				\
   271 	  movq_r2r(mm1, mm5);				\
   272 	  por_r2r(mm7, mm5);				\
   273 	  pand_r2r(mm5, mm3);				\
   274 	  por_r2r(mm6, mm3);				\
   275 	  movq_r2r(mm2, mm5);				\
   276 	  pand_r2r(mm7 , mm5);				\
   277           movq_r2r(mm3, mm6);				\
   278 	  pand_r2r(mm7 , mm6);				\
   279 	  psubw_r2r(mm6, mm5);				\
   280 	  pmullw_r2r(mm0, mm5);				\
   281 	  psrlw_i2r(8, mm5);				\
   282 	  paddw_r2r(mm5, mm6);				\
   283 	  pand_r2r(mm7, mm6);				\
   284 	  movq_r2r(mm1, mm5);				\
   285 	  por_r2r(mm4, mm5);				\
   286 	  pand_r2r(mm5, mm3);				\
   287 	  por_r2r(mm6, mm3);				\
   288 	  movq_r2m(mm3, *dstp);				\
   289 	  srcp += 4;					\
   290 	  dstp += 4;					\
   291 	  i -= 3;					\
   292 	}						\
   293 	emms();						\
   294     } while(0)
   295 
   296 #define ALPHA_BLIT16_555MMX(to, from, length, bpp, alpha)	\
   297     do {						\
   298         int i, n = 0;					\
   299 	Uint16 *srcp = (Uint16 *)(from);		\
   300 	Uint16 *dstp = (Uint16 *)(to);			\
   301         Uint32 ALPHA = 0x7C00;				\
   302 	movd_m2r(*(&ALPHA), mm1);			\
   303         punpcklwd_r2r(mm1, mm1);			\
   304         punpcklwd_r2r(mm1, mm1);			\
   305 	ALPHA = 0x03E0;					\
   306         movd_m2r(*(&ALPHA), mm4);			\
   307         punpcklwd_r2r(mm4, mm4);			\
   308         punpcklwd_r2r(mm4, mm4);			\
   309 	ALPHA = 0x001F;					\
   310 	movd_m2r(*(&ALPHA), mm7);			\
   311         punpcklwd_r2r(mm7, mm7);			\
   312         punpcklwd_r2r(mm7, mm7);			\
   313 	alpha &= ~(1+2+4);				\
   314         i = (Uint32)alpha | (Uint32)alpha << 16;	\
   315         movd_m2r(*(&i), mm0);				\
   316         punpckldq_r2r(mm0, mm0);			\
   317         i = ((int)(length) & 3);				\
   318         ALPHA = alpha >> 3;				\
   319 	for(; i > 0; --i) {				\
   320 	    Uint32 s = *srcp++;				\
   321 	    Uint32 d = *dstp;				\
   322 	    s = (s | s << 16) & 0x03e07c1f;		\
   323 	    d = (d | d << 16) & 0x03e07c1f;		\
   324 	    d += (s - d) * ALPHA >> 5;			\
   325 	    d &= 0x03e07c1f;				\
   326 	    *dstp++ = d | d >> 16;			\
   327 	    n++;					\
   328 	}						\
   329 	i = (int)(length) - n;				\
   330 	for(; i > 0; --i) {				\
   331 	  movq_m2r((*dstp), mm3);			\
   332 	  movq_m2r((*srcp), mm2);			\
   333 	  movq_r2r(mm2, mm5);				\
   334 	  pand_r2r(mm1 , mm5);				\
   335 	  psrlq_i2r(10, mm5);				\
   336 	  movq_r2r(mm3, mm6);				\
   337 	  pand_r2r(mm1 , mm6);				\
   338 	  psrlq_i2r(10, mm6);				\
   339 	  psubw_r2r(mm6, mm5);				\
   340 	  pmullw_r2r(mm0, mm5);				\
   341 	  psrlw_i2r(8, mm5);				\
   342 	  paddw_r2r(mm5, mm6);				\
   343 	  psllq_i2r(10, mm6);				\
   344 	  pand_r2r(mm1, mm6);				\
   345 	  movq_r2r(mm4, mm5);				\
   346 	  por_r2r(mm7, mm5);				\
   347 	  pand_r2r(mm5, mm3);				\
   348 	  por_r2r(mm6, mm3);				\
   349 	  movq_r2r(mm2, mm5);				\
   350 	  pand_r2r(mm4 , mm5);				\
   351 	  psrlq_i2r(5, mm5);				\
   352 	  movq_r2r(mm3, mm6);				\
   353 	  pand_r2r(mm4 , mm6);				\
   354 	  psrlq_i2r(5, mm6);				\
   355 	  psubw_r2r(mm6, mm5);				\
   356 	  pmullw_r2r(mm0, mm5);				\
   357 	  psrlw_i2r(8, mm5);				\
   358 	  paddw_r2r(mm5, mm6);				\
   359 	  psllq_i2r(5, mm6);				\
   360 	  pand_r2r(mm4, mm6);				\
   361 	  movq_r2r(mm1, mm5);				\
   362 	  por_r2r(mm7, mm5);				\
   363 	  pand_r2r(mm5, mm3);				\
   364 	  por_r2r(mm6, mm3);				\
   365 	  movq_r2r(mm2, mm5);				\
   366 	  pand_r2r(mm7 , mm5);				\
   367           movq_r2r(mm3, mm6);				\
   368 	  pand_r2r(mm7 , mm6);				\
   369 	  psubw_r2r(mm6, mm5);				\
   370 	  pmullw_r2r(mm0, mm5);				\
   371 	  psrlw_i2r(8, mm5);				\
   372 	  paddw_r2r(mm5, mm6);				\
   373 	  pand_r2r(mm7, mm6);				\
   374 	  movq_r2r(mm1, mm5);				\
   375 	  por_r2r(mm4, mm5);				\
   376 	  pand_r2r(mm5, mm3);				\
   377 	  por_r2r(mm6, mm3);				\
   378 	  movq_r2m(mm3, *dstp);				\
   379 	  srcp += 4;					\
   380 	  dstp += 4;					\
   381 	  i -= 3;					\
   382 	}						\
   383 	emms();						\
   384     } while(0)
   385 
   386 #endif
   387 
   388 /*
   389  * For 32bpp pixels on the form 0x00rrggbb:
   390  * If we treat the middle component separately, we can process the two
   391  * remaining in parallel. This is safe to do because of the gap to the left
   392  * of each component, so the bits from the multiplication don't collide.
   393  * This can be used for any RGB permutation of course.
   394  */
   395 #define ALPHA_BLIT32_888(to, from, length, bpp, alpha)		\
   396     do {							\
   397         int i;							\
   398 	Uint32 *src = (Uint32 *)(from);				\
   399 	Uint32 *dst = (Uint32 *)(to);				\
   400 	for(i = 0; i < (int)(length); i++) {			\
   401 	    Uint32 s = *src++;					\
   402 	    Uint32 d = *dst;					\
   403 	    Uint32 s1 = s & 0xff00ff;				\
   404 	    Uint32 d1 = d & 0xff00ff;				\
   405 	    d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;	\
   406 	    s &= 0xff00;					\
   407 	    d &= 0xff00;					\
   408 	    d = (d + ((s - d) * alpha >> 8)) & 0xff00;		\
   409 	    *dst++ = d1 | d;					\
   410 	}							\
   411     } while(0)
   412 
   413 /*
   414  * For 16bpp pixels we can go a step further: put the middle component
   415  * in the high 16 bits of a 32 bit word, and process all three RGB
   416  * components at the same time. Since the smallest gap is here just
   417  * 5 bits, we have to scale alpha down to 5 bits as well.
   418  */
   419 #define ALPHA_BLIT16_565(to, from, length, bpp, alpha)	\
   420     do {						\
   421         int i;						\
   422 	Uint16 *src = (Uint16 *)(from);			\
   423 	Uint16 *dst = (Uint16 *)(to);			\
   424 	Uint32 ALPHA = alpha >> 3;			\
   425 	for(i = 0; i < (int)(length); i++) {		\
   426 	    Uint32 s = *src++;				\
   427 	    Uint32 d = *dst;				\
   428 	    s = (s | s << 16) & 0x07e0f81f;		\
   429 	    d = (d | d << 16) & 0x07e0f81f;		\
   430 	    d += (s - d) * ALPHA >> 5;			\
   431 	    d &= 0x07e0f81f;				\
   432 	    *dst++ = d | d >> 16;			\
   433 	}						\
   434     } while(0)
   435 
   436 #define ALPHA_BLIT16_555(to, from, length, bpp, alpha)	\
   437     do {						\
   438         int i;						\
   439 	Uint16 *src = (Uint16 *)(from);			\
   440 	Uint16 *dst = (Uint16 *)(to);			\
   441 	Uint32 ALPHA = alpha >> 3;			\
   442 	for(i = 0; i < (int)(length); i++) {		\
   443 	    Uint32 s = *src++;				\
   444 	    Uint32 d = *dst;				\
   445 	    s = (s | s << 16) & 0x03e07c1f;		\
   446 	    d = (d | d << 16) & 0x03e07c1f;		\
   447 	    d += (s - d) * ALPHA >> 5;			\
   448 	    d &= 0x03e07c1f;				\
   449 	    *dst++ = d | d >> 16;			\
   450 	}						\
   451     } while(0)
   452 
   453 /*
   454  * The general slow catch-all function, for remaining depths and formats
   455  */
   456 #define ALPHA_BLIT_ANY(to, from, length, bpp, alpha)			\
   457     do {								\
   458         int i;								\
   459 	Uint8 *src = from;						\
   460 	Uint8 *dst = to;						\
   461 	for(i = 0; i < (int)(length); i++) {				\
   462 	    Uint32 s, d;						\
   463 	    unsigned rs, gs, bs, rd, gd, bd;				\
   464 	    switch(bpp) {						\
   465 	    case 2:							\
   466 		s = *(Uint16 *)src;					\
   467 		d = *(Uint16 *)dst;					\
   468 		break;							\
   469 	    case 3:							\
   470 		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
   471 		    s = (src[0] << 16) | (src[1] << 8) | src[2];	\
   472 		    d = (dst[0] << 16) | (dst[1] << 8) | dst[2];	\
   473 		} else {						\
   474 		    s = (src[2] << 16) | (src[1] << 8) | src[0];	\
   475 		    d = (dst[2] << 16) | (dst[1] << 8) | dst[0];	\
   476 		}							\
   477 		break;							\
   478 	    case 4:							\
   479 		s = *(Uint32 *)src;					\
   480 		d = *(Uint32 *)dst;					\
   481 		break;							\
   482 	    }								\
   483 	    RGB_FROM_PIXEL(s, fmt, rs, gs, bs);				\
   484 	    RGB_FROM_PIXEL(d, fmt, rd, gd, bd);				\
   485 	    rd += (rs - rd) * alpha >> 8;				\
   486 	    gd += (gs - gd) * alpha >> 8;				\
   487 	    bd += (bs - bd) * alpha >> 8;				\
   488 	    PIXEL_FROM_RGB(d, fmt, rd, gd, bd);				\
   489 	    switch(bpp) {						\
   490 	    case 2:							\
   491 		*(Uint16 *)dst = d;					\
   492 		break;							\
   493 	    case 3:							\
   494 		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
   495 		    dst[0] = d >> 16;					\
   496 		    dst[1] = d >> 8;					\
   497 		    dst[2] = d;						\
   498 		} else {						\
   499 		    dst[0] = d;						\
   500 		    dst[1] = d >> 8;					\
   501 		    dst[2] = d >> 16;					\
   502 		}							\
   503 		break;							\
   504 	    case 4:							\
   505 		*(Uint32 *)dst = d;					\
   506 		break;							\
   507 	    }								\
   508 	    src += bpp;							\
   509 	    dst += bpp;							\
   510 	}								\
   511     } while(0)
   512 
   513 #ifdef MMX_ASMBLIT
   514 
   515 #define ALPHA_BLIT32_888_50MMX(to, from, length, bpp, alpha)		\
   516     do {								\
   517 	Uint32 *srcp = (Uint32 *)(from);				\
   518 	Uint32 *dstp = (Uint32 *)(to);					\
   519         int i = 0x00fefefe;						\
   520         movd_m2r(*(&i), mm4);						\
   521         punpckldq_r2r(mm4, mm4);					\
   522         i = 0x00010101;							\
   523         movd_m2r(*(&i), mm3);						\
   524         punpckldq_r2r(mm3, mm3);					\
   525         i = (int)(length);						\
   526         if( i & 1 ) {							\
   527 	  Uint32 s = *srcp++;						\
   528 	  Uint32 d = *dstp;						\
   529 	  *dstp++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
   530 		     + (s & d & 0x00010101);				\
   531 	  i--;								\
   532 	}								\
   533 	for(; i > 0; --i) {						\
   534 	    movq_m2r((*dstp), mm2); /* dst -> mm2 */			\
   535 	    movq_r2r(mm2, mm6);	/* dst -> mm6 */			\
   536 	    movq_m2r((*srcp), mm1); /* src -> mm1 */			\
   537 	    movq_r2r(mm1, mm5);	/* src -> mm5 */			\
   538 	    pand_r2r(mm4, mm6);	/* dst & 0x00fefefe -> mm6 */		\
   539 	    pand_r2r(mm4, mm5); /* src & 0x00fefefe -> mm5 */		\
   540 	    paddd_r2r(mm6, mm5); /* (dst & 0x00fefefe) + (dst & 0x00fefefe) -> mm5 */	\
   541 	    psrld_i2r(1, mm5);						\
   542 	    pand_r2r(mm1, mm2);	/* s & d -> mm2 */			\
   543 	    pand_r2r(mm3, mm2);	/* s & d & 0x00010101 -> mm2 */		\
   544 	    paddd_r2r(mm5, mm2);					\
   545 	    movq_r2m(mm2, (*dstp));					\
   546 	    dstp += 2;							\
   547 	    srcp += 2;							\
   548 	    i--;							\
   549 	}								\
   550 	emms();								\
   551     } while(0)
   552 
   553 #endif
   554     
   555 /*
   556  * Special case: 50% alpha (alpha=128)
   557  * This is treated specially because it can be optimized very well, and
   558  * since it is good for many cases of semi-translucency.
   559  * The theory is to do all three components at the same time:
   560  * First zero the lowest bit of each component, which gives us room to
   561  * add them. Then shift right and add the sum of the lowest bits.
   562  */
   563 #define ALPHA_BLIT32_888_50(to, from, length, bpp, alpha)		\
   564     do {								\
   565         int i;								\
   566 	Uint32 *src = (Uint32 *)(from);					\
   567 	Uint32 *dst = (Uint32 *)(to);					\
   568 	for(i = 0; i < (int)(length); i++) {				\
   569 	    Uint32 s = *src++;						\
   570 	    Uint32 d = *dst;						\
   571 	    *dst++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
   572 		     + (s & d & 0x00010101);				\
   573 	}								\
   574     } while(0)
   575 
   576 /*
   577  * For 16bpp, we can actually blend two pixels in parallel, if we take
   578  * care to shift before we add, not after.
   579  */
   580 
   581 /* helper: blend a single 16 bit pixel at 50% */
   582 #define BLEND16_50(dst, src, mask)			\
   583     do {						\
   584         Uint32 s = *src++;				\
   585 	Uint32 d = *dst;				\
   586 	*dst++ = (((s & mask) + (d & mask)) >> 1)	\
   587 	         + (s & d & (~mask & 0xffff));		\
   588     } while(0)
   589 
   590 /* basic 16bpp blender. mask is the pixels to keep when adding. */
   591 #define ALPHA_BLIT16_50(to, from, length, bpp, alpha, mask)		\
   592     do {								\
   593 	unsigned n = (length);						\
   594 	Uint16 *src = (Uint16 *)(from);					\
   595 	Uint16 *dst = (Uint16 *)(to);					\
   596 	if(((unsigned long)src ^ (unsigned long)dst) & 3) {		\
   597 	    /* source and destination not in phase, blit one by one */	\
   598 	    while(n--)							\
   599 		BLEND16_50(dst, src, mask);				\
   600 	} else {							\
   601 	    if((unsigned long)src & 3) {				\
   602 		/* first odd pixel */					\
   603 		BLEND16_50(dst, src, mask);				\
   604 		n--;							\
   605 	    }								\
   606 	    for(; n > 1; n -= 2) {					\
   607 		Uint32 s = *(Uint32 *)src;				\
   608 		Uint32 d = *(Uint32 *)dst;				\
   609 		*(Uint32 *)dst = ((s & (mask | mask << 16)) >> 1)	\
   610 		               + ((d & (mask | mask << 16)) >> 1)	\
   611 		               + (s & d & (~(mask | mask << 16)));	\
   612 		src += 2;						\
   613 		dst += 2;						\
   614 	    }								\
   615 	    if(n)							\
   616 		BLEND16_50(dst, src, mask); /* last odd pixel */	\
   617 	}								\
   618     } while(0)
   619 
   620 #define ALPHA_BLIT16_565_50(to, from, length, bpp, alpha)	\
   621     ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xf7de)
   622 
   623 #define ALPHA_BLIT16_555_50(to, from, length, bpp, alpha)	\
   624     ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xfbde)
   625 
   626 #ifdef MMX_ASMBLIT
   627 
   628 #define CHOOSE_BLIT(blitter, alpha, fmt)				\
   629     do {								\
   630         if(alpha == 255) {						\
   631 	    switch(fmt->BytesPerPixel) {				\
   632 	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
   633 	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
   634 	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
   635 	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
   636 	    }								\
   637 	} else {							\
   638 	    switch(fmt->BytesPerPixel) {				\
   639 	    case 1:							\
   640 		/* No 8bpp alpha blitting */				\
   641 		break;							\
   642 									\
   643 	    case 2:							\
   644 		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
   645 		case 0xffff:						\
   646 		    if(fmt->Gmask == 0x07e0				\
   647 		       || fmt->Rmask == 0x07e0				\
   648 		       || fmt->Bmask == 0x07e0) {			\
   649 			if(alpha == 128)				\
   650 			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
   651 			else {						\
   652 			    if(SDL_HasMMX())				\
   653 				blitter(2, Uint8, ALPHA_BLIT16_565MMX);	\
   654 			    else					\
   655 				blitter(2, Uint8, ALPHA_BLIT16_565);	\
   656 			}						\
   657 		    } else						\
   658 			goto general16;					\
   659 		    break;						\
   660 									\
   661 		case 0x7fff:						\
   662 		    if(fmt->Gmask == 0x03e0				\
   663 		       || fmt->Rmask == 0x03e0				\
   664 		       || fmt->Bmask == 0x03e0) {			\
   665 			if(alpha == 128)				\
   666 			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
   667 			else {						\
   668 			    if(SDL_HasMMX())				\
   669 				blitter(2, Uint8, ALPHA_BLIT16_555MMX);	\
   670 			    else					\
   671 				blitter(2, Uint8, ALPHA_BLIT16_555);	\
   672 			}						\
   673 			break;						\
   674 		    }							\
   675 		    /* fallthrough */					\
   676 									\
   677 		default:						\
   678 		general16:						\
   679 		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
   680 		}							\
   681 		break;							\
   682 									\
   683 	    case 3:							\
   684 		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
   685 		break;							\
   686 									\
   687 	    case 4:							\
   688 		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
   689 		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
   690 		       || fmt->Bmask == 0xff00)) {			\
   691 		    if(alpha == 128)					\
   692 		    {							\
   693 			if(SDL_HasMMX())				\
   694 				blitter(4, Uint16, ALPHA_BLIT32_888_50MMX);\
   695 			else						\
   696 				blitter(4, Uint16, ALPHA_BLIT32_888_50);\
   697 		    }							\
   698 		    else						\
   699 		    {							\
   700 			if(SDL_HasMMX())				\
   701 				blitter(4, Uint16, ALPHA_BLIT32_888MMX);\
   702 			else						\
   703 				blitter(4, Uint16, ALPHA_BLIT32_888);	\
   704 		    }							\
   705 		} else							\
   706 		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
   707 		break;							\
   708 	    }								\
   709 	}								\
   710     } while(0)
   711 
   712 #else
   713 	
   714 #define CHOOSE_BLIT(blitter, alpha, fmt)				\
   715     do {								\
   716         if(alpha == 255) {						\
   717 	    switch(fmt->BytesPerPixel) {				\
   718 	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
   719 	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
   720 	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
   721 	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
   722 	    }								\
   723 	} else {							\
   724 	    switch(fmt->BytesPerPixel) {				\
   725 	    case 1:							\
   726 		/* No 8bpp alpha blitting */				\
   727 		break;							\
   728 									\
   729 	    case 2:							\
   730 		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
   731 		case 0xffff:						\
   732 		    if(fmt->Gmask == 0x07e0				\
   733 		       || fmt->Rmask == 0x07e0				\
   734 		       || fmt->Bmask == 0x07e0) {			\
   735 			if(alpha == 128)				\
   736 			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
   737 			else {						\
   738 			    blitter(2, Uint8, ALPHA_BLIT16_565);	\
   739 			}						\
   740 		    } else						\
   741 			goto general16;					\
   742 		    break;						\
   743 									\
   744 		case 0x7fff:						\
   745 		    if(fmt->Gmask == 0x03e0				\
   746 		       || fmt->Rmask == 0x03e0				\
   747 		       || fmt->Bmask == 0x03e0) {			\
   748 			if(alpha == 128)				\
   749 			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
   750 			else {						\
   751 			    blitter(2, Uint8, ALPHA_BLIT16_555);	\
   752 			}						\
   753 			break;						\
   754 		    }							\
   755 		    /* fallthrough */					\
   756 									\
   757 		default:						\
   758 		general16:						\
   759 		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
   760 		}							\
   761 		break;							\
   762 									\
   763 	    case 3:							\
   764 		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
   765 		break;							\
   766 									\
   767 	    case 4:							\
   768 		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
   769 		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
   770 		       || fmt->Bmask == 0xff00)) {			\
   771 		    if(alpha == 128)					\
   772 			blitter(4, Uint16, ALPHA_BLIT32_888_50);	\
   773 		    else						\
   774 			blitter(4, Uint16, ALPHA_BLIT32_888);		\
   775 		} else							\
   776 		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
   777 		break;							\
   778 	    }								\
   779 	}								\
   780     } while(0)
   781 
   782 #endif
   783 
   784 /*
   785  * This takes care of the case when the surface is clipped on the left and/or
   786  * right. Top clipping has already been taken care of.
   787  */
   788 static void RLEClipBlit(int w, Uint8 *srcbuf, SDL_Surface *dst,
   789 			Uint8 *dstbuf, SDL_Rect *srcrect, unsigned alpha)
   790 {
   791     SDL_PixelFormat *fmt = dst->format;
   792 
   793 #define RLECLIPBLIT(bpp, Type, do_blit)					   \
   794     do {								   \
   795 	int linecount = srcrect->h;					   \
   796 	int ofs = 0;							   \
   797 	int left = srcrect->x;						   \
   798 	int right = left + srcrect->w;					   \
   799 	dstbuf -= left * bpp;						   \
   800 	for(;;) {							   \
   801 	    int run;							   \
   802 	    ofs += *(Type *)srcbuf;					   \
   803 	    run = ((Type *)srcbuf)[1];					   \
   804 	    srcbuf += 2 * sizeof(Type);					   \
   805 	    if(run) {							   \
   806 		/* clip to left and right borders */			   \
   807 		if(ofs < right) {					   \
   808 		    int start = 0;					   \
   809 		    int len = run;					   \
   810 		    int startcol;					   \
   811 		    if(left - ofs > 0) {				   \
   812 			start = left - ofs;				   \
   813 			len -= start;					   \
   814 			if(len <= 0)					   \
   815 			    goto nocopy ## bpp ## do_blit;		   \
   816 		    }							   \
   817 		    startcol = ofs + start;				   \
   818 		    if(len > right - startcol)				   \
   819 			len = right - startcol;				   \
   820 		    do_blit(dstbuf + startcol * bpp, srcbuf + start * bpp, \
   821 			    len, bpp, alpha);				   \
   822 		}							   \
   823 	    nocopy ## bpp ## do_blit:					   \
   824 		srcbuf += run * bpp;					   \
   825 		ofs += run;						   \
   826 	    } else if(!ofs)						   \
   827 		break;							   \
   828 	    if(ofs == w) {						   \
   829 		ofs = 0;						   \
   830 		dstbuf += dst->pitch;					   \
   831 		if(!--linecount)					   \
   832 		    break;						   \
   833 	    }								   \
   834 	}								   \
   835     } while(0)
   836 
   837     CHOOSE_BLIT(RLECLIPBLIT, alpha, fmt);
   838 
   839 #undef RLECLIPBLIT
   840 
   841 }
   842 
   843 
   844 /* blit a colorkeyed RLE surface */
   845 int SDL_RLEBlit(SDL_Surface *src, SDL_Rect *srcrect,
   846 		SDL_Surface *dst, SDL_Rect *dstrect)
   847 {
   848 	Uint8 *dstbuf;
   849 	Uint8 *srcbuf;
   850 	int x, y;
   851 	int w = src->w;
   852 	unsigned alpha;
   853 
   854 	/* Lock the destination if necessary */
   855 	if ( SDL_MUSTLOCK(dst) ) {
   856 		if ( SDL_LockSurface(dst) < 0 ) {
   857 			return(-1);
   858 		}
   859 	}
   860 
   861 	/* Set up the source and destination pointers */
   862 	x = dstrect->x;
   863 	y = dstrect->y;
   864 	dstbuf = (Uint8 *)dst->pixels
   865 	         + y * dst->pitch + x * src->format->BytesPerPixel;
   866 	srcbuf = (Uint8 *)src->map->sw_data->aux_data;
   867 
   868 	{
   869 	    /* skip lines at the top if neccessary */
   870 	    int vskip = srcrect->y;
   871 	    int ofs = 0;
   872 	    if(vskip) {
   873 
   874 #define RLESKIP(bpp, Type)			\
   875 		for(;;) {			\
   876 		    int run;			\
   877 		    ofs += *(Type *)srcbuf;	\
   878 		    run = ((Type *)srcbuf)[1];	\
   879 		    srcbuf += sizeof(Type) * 2;	\
   880 		    if(run) {			\
   881 			srcbuf += run * bpp;	\
   882 			ofs += run;		\
   883 		    } else if(!ofs)		\
   884 			goto done;		\
   885 		    if(ofs == w) {		\
   886 			ofs = 0;		\
   887 			if(!--vskip)		\
   888 			    break;		\
   889 		    }				\
   890 		}
   891 
   892 		switch(src->format->BytesPerPixel) {
   893 		case 1: RLESKIP(1, Uint8); break;
   894 		case 2: RLESKIP(2, Uint8); break;
   895 		case 3: RLESKIP(3, Uint8); break;
   896 		case 4: RLESKIP(4, Uint16); break;
   897 		}
   898 
   899 #undef RLESKIP
   900 
   901 	    }
   902 	}
   903 
   904 	alpha = (src->flags & SDL_SRCALPHA) == SDL_SRCALPHA
   905 	        ? src->format->alpha : 255;
   906 	/* if left or right edge clipping needed, call clip blit */
   907 	if ( srcrect->x || srcrect->w != src->w ) {
   908 	    RLEClipBlit(w, srcbuf, dst, dstbuf, srcrect, alpha);
   909 	} else {
   910 	    SDL_PixelFormat *fmt = src->format;
   911 
   912 #define RLEBLIT(bpp, Type, do_blit)					      \
   913 	    do {							      \
   914 		int linecount = srcrect->h;				      \
   915 		int ofs = 0;						      \
   916 		for(;;) {						      \
   917 		    unsigned run;					      \
   918 		    ofs += *(Type *)srcbuf;				      \
   919 		    run = ((Type *)srcbuf)[1];				      \
   920 		    srcbuf += 2 * sizeof(Type);				      \
   921 		    if(run) {						      \
   922 			do_blit(dstbuf + ofs * bpp, srcbuf, run, bpp, alpha); \
   923 			srcbuf += run * bpp;				      \
   924 			ofs += run;					      \
   925 		    } else if(!ofs)					      \
   926 			break;						      \
   927 		    if(ofs == w) {					      \
   928 			ofs = 0;					      \
   929 			dstbuf += dst->pitch;				      \
   930 			if(!--linecount)				      \
   931 			    break;					      \
   932 		    }							      \
   933 		}							      \
   934 	    } while(0)
   935 
   936 	    CHOOSE_BLIT(RLEBLIT, alpha, fmt);
   937 
   938 #undef RLEBLIT
   939 	}
   940 
   941 done:
   942 	/* Unlock the destination if necessary */
   943 	if ( SDL_MUSTLOCK(dst) ) {
   944 		SDL_UnlockSurface(dst);
   945 	}
   946 	return(0);
   947 }
   948 
   949 #undef OPAQUE_BLIT
   950 
   951 /*
   952  * Per-pixel blitting macros for translucent pixels:
   953  * These use the same techniques as the per-surface blitting macros
   954  */
   955 
   956 /*
   957  * For 32bpp pixels, we have made sure the alpha is stored in the top
   958  * 8 bits, so proceed as usual
   959  */
   960 #define BLIT_TRANSL_888(src, dst)				\
   961     do {							\
   962         Uint32 s = src;						\
   963 	Uint32 d = dst;						\
   964 	unsigned alpha = s >> 24;				\
   965 	Uint32 s1 = s & 0xff00ff;				\
   966 	Uint32 d1 = d & 0xff00ff;				\
   967 	d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;	\
   968 	s &= 0xff00;						\
   969 	d &= 0xff00;						\
   970 	d = (d + ((s - d) * alpha >> 8)) & 0xff00;		\
   971 	dst = d1 | d;						\
   972     } while(0)
   973 
   974 /*
   975  * For 16bpp pixels, we have stored the 5 most significant alpha bits in
   976  * bits 5-10. As before, we can process all 3 RGB components at the same time.
   977  */
   978 #define BLIT_TRANSL_565(src, dst)		\
   979     do {					\
   980         Uint32 s = src;				\
   981 	Uint32 d = dst;				\
   982 	unsigned alpha = (s & 0x3e0) >> 5;	\
   983 	s &= 0x07e0f81f;			\
   984 	d = (d | d << 16) & 0x07e0f81f;		\
   985 	d += (s - d) * alpha >> 5;		\
   986 	d &= 0x07e0f81f;			\
   987 	dst = d | d >> 16;			\
   988     } while(0)
   989 
   990 #define BLIT_TRANSL_555(src, dst)		\
   991     do {					\
   992         Uint32 s = src;				\
   993 	Uint32 d = dst;				\
   994 	unsigned alpha = (s & 0x3e0) >> 5;	\
   995 	s &= 0x03e07c1f;			\
   996 	d = (d | d << 16) & 0x03e07c1f;		\
   997 	d += (s - d) * alpha >> 5;		\
   998 	d &= 0x03e07c1f;			\
   999 	dst = d | d >> 16;			\
  1000     } while(0)
  1001 
  1002 /* used to save the destination format in the encoding. Designed to be
  1003    macro-compatible with SDL_PixelFormat but without the unneeded fields */
  1004 typedef struct {
  1005     	Uint8  BytesPerPixel;
  1006 	Uint8  Rloss;
  1007 	Uint8  Gloss;
  1008 	Uint8  Bloss;
  1009 	Uint8  Rshift;
  1010 	Uint8  Gshift;
  1011 	Uint8  Bshift;
  1012 	Uint8  Ashift;
  1013 	Uint32 Rmask;
  1014 	Uint32 Gmask;
  1015 	Uint32 Bmask;
  1016 	Uint32 Amask;
  1017 } RLEDestFormat;
  1018 
  1019 /* blit a pixel-alpha RLE surface clipped at the right and/or left edges */
  1020 static void RLEAlphaClipBlit(int w, Uint8 *srcbuf, SDL_Surface *dst,
  1021 			     Uint8 *dstbuf, SDL_Rect *srcrect)
  1022 {
  1023     SDL_PixelFormat *df = dst->format;
  1024     /*
  1025      * clipped blitter: Ptype is the destination pixel type,
  1026      * Ctype the translucent count type, and do_blend the macro
  1027      * to blend one pixel.
  1028      */
  1029 #define RLEALPHACLIPBLIT(Ptype, Ctype, do_blend)			  \
  1030     do {								  \
  1031 	int linecount = srcrect->h;					  \
  1032 	int left = srcrect->x;						  \
  1033 	int right = left + srcrect->w;					  \
  1034 	dstbuf -= left * sizeof(Ptype);					  \
  1035 	do {								  \
  1036 	    int ofs = 0;						  \
  1037 	    /* blit opaque pixels on one line */			  \
  1038 	    do {							  \
  1039 		unsigned run;						  \
  1040 		ofs += ((Ctype *)srcbuf)[0];				  \
  1041 		run = ((Ctype *)srcbuf)[1];				  \
  1042 		srcbuf += 2 * sizeof(Ctype);				  \
  1043 		if(run) {						  \
  1044 		    /* clip to left and right borders */		  \
  1045 		    int cofs = ofs;					  \
  1046 		    int crun = run;					  \
  1047 		    if(left - cofs > 0) {				  \
  1048 			crun -= left - cofs;				  \
  1049 			cofs = left;					  \
  1050 		    }							  \
  1051 		    if(crun > right - cofs)				  \
  1052 			crun = right - cofs;				  \
  1053 		    if(crun > 0)					  \
  1054 			PIXEL_COPY(dstbuf + cofs * sizeof(Ptype),	  \
  1055 				   srcbuf + (cofs - ofs) * sizeof(Ptype), \
  1056 				   (unsigned)crun, sizeof(Ptype));	  \
  1057 		    srcbuf += run * sizeof(Ptype);			  \
  1058 		    ofs += run;						  \
  1059 		} else if(!ofs)						  \
  1060 		    return;						  \
  1061 	    } while(ofs < w);						  \
  1062 	    /* skip padding if necessary */				  \
  1063 	    if(sizeof(Ptype) == 2)					  \
  1064 		srcbuf += (unsigned long)srcbuf & 2;			  \
  1065 	    /* blit translucent pixels on the same line */		  \
  1066 	    ofs = 0;							  \
  1067 	    do {							  \
  1068 		unsigned run;						  \
  1069 		ofs += ((Uint16 *)srcbuf)[0];				  \
  1070 		run = ((Uint16 *)srcbuf)[1];				  \
  1071 		srcbuf += 4;						  \
  1072 		if(run) {						  \
  1073 		    /* clip to left and right borders */		  \
  1074 		    int cofs = ofs;					  \
  1075 		    int crun = run;					  \
  1076 		    if(left - cofs > 0) {				  \
  1077 			crun -= left - cofs;				  \
  1078 			cofs = left;					  \
  1079 		    }							  \
  1080 		    if(crun > right - cofs)				  \
  1081 			crun = right - cofs;				  \
  1082 		    if(crun > 0) {					  \
  1083 			Ptype *dst = (Ptype *)dstbuf + cofs;		  \
  1084 			Uint32 *src = (Uint32 *)srcbuf + (cofs - ofs);	  \
  1085 			int i;						  \
  1086 			for(i = 0; i < crun; i++)			  \
  1087 			    do_blend(src[i], dst[i]);			  \
  1088 		    }							  \
  1089 		    srcbuf += run * 4;					  \
  1090 		    ofs += run;						  \
  1091 		}							  \
  1092 	    } while(ofs < w);						  \
  1093 	    dstbuf += dst->pitch;					  \
  1094 	} while(--linecount);						  \
  1095     } while(0)
  1096 
  1097     switch(df->BytesPerPixel) {
  1098     case 2:
  1099 	if(df->Gmask == 0x07e0 || df->Rmask == 0x07e0
  1100 	   || df->Bmask == 0x07e0)
  1101 	    RLEALPHACLIPBLIT(Uint16, Uint8, BLIT_TRANSL_565);
  1102 	else
  1103 	    RLEALPHACLIPBLIT(Uint16, Uint8, BLIT_TRANSL_555);
  1104 	break;
  1105     case 4:
  1106 	RLEALPHACLIPBLIT(Uint32, Uint16, BLIT_TRANSL_888);
  1107 	break;
  1108     }
  1109 }
  1110 
  1111 /* blit a pixel-alpha RLE surface */
  1112 int SDL_RLEAlphaBlit(SDL_Surface *src, SDL_Rect *srcrect,
  1113 		     SDL_Surface *dst, SDL_Rect *dstrect)
  1114 {
  1115     int x, y;
  1116     int w = src->w;
  1117     Uint8 *srcbuf, *dstbuf;
  1118     SDL_PixelFormat *df = dst->format;
  1119 
  1120     /* Lock the destination if necessary */
  1121     if ( SDL_MUSTLOCK(dst) ) {
  1122 	if ( SDL_LockSurface(dst) < 0 ) {
  1123 	    return -1;
  1124 	}
  1125     }
  1126 
  1127     x = dstrect->x;
  1128     y = dstrect->y;
  1129     dstbuf = (Uint8 *)dst->pixels
  1130 	     + y * dst->pitch + x * df->BytesPerPixel;
  1131     srcbuf = (Uint8 *)src->map->sw_data->aux_data + sizeof(RLEDestFormat);
  1132 
  1133     {
  1134 	/* skip lines at the top if necessary */
  1135 	int vskip = srcrect->y;
  1136 	if(vskip) {
  1137 	    int ofs;
  1138 	    if(df->BytesPerPixel == 2) {
  1139 		/* the 16/32 interleaved format */
  1140 		do {
  1141 		    /* skip opaque line */
  1142 		    ofs = 0;
  1143 		    do {
  1144 			int run;
  1145 			ofs += srcbuf[0];
  1146 			run = srcbuf[1];
  1147 			srcbuf += 2;
  1148 			if(run) {
  1149 			    srcbuf += 2 * run;
  1150 			    ofs += run;
  1151 			} else if(!ofs)
  1152 			    goto done;
  1153 		    } while(ofs < w);
  1154 
  1155 		    /* skip padding */
  1156 		    srcbuf += (unsigned long)srcbuf & 2;
  1157 
  1158 		    /* skip translucent line */
  1159 		    ofs = 0;
  1160 		    do {
  1161 			int run;
  1162 			ofs += ((Uint16 *)srcbuf)[0];
  1163 			run = ((Uint16 *)srcbuf)[1];
  1164 			srcbuf += 4 * (run + 1);
  1165 			ofs += run;
  1166 		    } while(ofs < w);
  1167 		} while(--vskip);
  1168 	    } else {
  1169 		/* the 32/32 interleaved format */
  1170 		vskip <<= 1;	/* opaque and translucent have same format */
  1171 		do {
  1172 		    ofs = 0;
  1173 		    do {
  1174 			int run;
  1175 			ofs += ((Uint16 *)srcbuf)[0];
  1176 			run = ((Uint16 *)srcbuf)[1];
  1177 			srcbuf += 4;
  1178 			if(run) {
  1179 			    srcbuf += 4 * run;
  1180 			    ofs += run;
  1181 			} else if(!ofs)
  1182 			    goto done;
  1183 		    } while(ofs < w);
  1184 		} while(--vskip);
  1185 	    }
  1186 	}
  1187     }
  1188 
  1189     /* if left or right edge clipping needed, call clip blit */
  1190     if(srcrect->x || srcrect->w != src->w) {
  1191 	RLEAlphaClipBlit(w, srcbuf, dst, dstbuf, srcrect);
  1192     } else {
  1193 
  1194 	/*
  1195 	 * non-clipped blitter. Ptype is the destination pixel type,
  1196 	 * Ctype the translucent count type, and do_blend the
  1197 	 * macro to blend one pixel.
  1198 	 */
  1199 #define RLEALPHABLIT(Ptype, Ctype, do_blend)				 \
  1200 	do {								 \
  1201 	    int linecount = srcrect->h;					 \
  1202 	    do {							 \
  1203 		int ofs = 0;						 \
  1204 		/* blit opaque pixels on one line */			 \
  1205 		do {							 \
  1206 		    unsigned run;					 \
  1207 		    ofs += ((Ctype *)srcbuf)[0];			 \
  1208 		    run = ((Ctype *)srcbuf)[1];				 \
  1209 		    srcbuf += 2 * sizeof(Ctype);			 \
  1210 		    if(run) {						 \
  1211 			PIXEL_COPY(dstbuf + ofs * sizeof(Ptype), srcbuf, \
  1212 				   run, sizeof(Ptype));			 \
  1213 			srcbuf += run * sizeof(Ptype);			 \
  1214 			ofs += run;					 \
  1215 		    } else if(!ofs)					 \
  1216 			goto done;					 \
  1217 		} while(ofs < w);					 \
  1218 		/* skip padding if necessary */				 \
  1219 		if(sizeof(Ptype) == 2)					 \
  1220 		    srcbuf += (unsigned long)srcbuf & 2;		 \
  1221 		/* blit translucent pixels on the same line */		 \
  1222 		ofs = 0;						 \
  1223 		do {							 \
  1224 		    unsigned run;					 \
  1225 		    ofs += ((Uint16 *)srcbuf)[0];			 \
  1226 		    run = ((Uint16 *)srcbuf)[1];			 \
  1227 		    srcbuf += 4;					 \
  1228 		    if(run) {						 \
  1229 			Ptype *dst = (Ptype *)dstbuf + ofs;		 \
  1230 			unsigned i;					 \
  1231 			for(i = 0; i < run; i++) {			 \
  1232 			    Uint32 src = *(Uint32 *)srcbuf;		 \
  1233 			    do_blend(src, *dst);			 \
  1234 			    srcbuf += 4;				 \
  1235 			    dst++;					 \
  1236 			}						 \
  1237 			ofs += run;					 \
  1238 		    }							 \
  1239 		} while(ofs < w);					 \
  1240 		dstbuf += dst->pitch;					 \
  1241 	    } while(--linecount);					 \
  1242 	} while(0)
  1243 
  1244 	switch(df->BytesPerPixel) {
  1245 	case 2:
  1246 	    if(df->Gmask == 0x07e0 || df->Rmask == 0x07e0
  1247 	       || df->Bmask == 0x07e0)
  1248 		RLEALPHABLIT(Uint16, Uint8, BLIT_TRANSL_565);
  1249 	    else
  1250 		RLEALPHABLIT(Uint16, Uint8, BLIT_TRANSL_555);
  1251 	    break;
  1252 	case 4:
  1253 	    RLEALPHABLIT(Uint32, Uint16, BLIT_TRANSL_888);
  1254 	    break;
  1255 	}
  1256     }
  1257 
  1258  done:
  1259     /* Unlock the destination if necessary */
  1260     if ( SDL_MUSTLOCK(dst) ) {
  1261 	SDL_UnlockSurface(dst);
  1262     }
  1263     return 0;
  1264 }
  1265 
  1266 /*
  1267  * Auxiliary functions:
  1268  * The encoding functions take 32bpp rgb + a, and
  1269  * return the number of bytes copied to the destination.
  1270  * The decoding functions copy to 32bpp rgb + a, and
  1271  * return the number of bytes copied from the source.
  1272  * These are only used in the encoder and un-RLE code and are therefore not
  1273  * highly optimised.
  1274  */
  1275 
  1276 /* encode 32bpp rgb + a into 16bpp rgb, losing alpha */
  1277 static int copy_opaque_16(void *dst, Uint32 *src, int n,
  1278 			  SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1279 {
  1280     int i;
  1281     Uint16 *d = dst;
  1282     for(i = 0; i < n; i++) {
  1283 	unsigned r, g, b;
  1284 	RGB_FROM_PIXEL(*src, sfmt, r, g, b);
  1285 	PIXEL_FROM_RGB(*d, dfmt, r, g, b);
  1286 	src++;
  1287 	d++;
  1288     }
  1289     return n * 2;
  1290 }
  1291 
  1292 /* decode opaque pixels from 16bpp to 32bpp rgb + a */
  1293 static int uncopy_opaque_16(Uint32 *dst, void *src, int n,
  1294 			    RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1295 {
  1296     int i;
  1297     Uint16 *s = src;
  1298     unsigned alpha = dfmt->Amask ? 255 : 0;
  1299     for(i = 0; i < n; i++) {
  1300 	unsigned r, g, b;
  1301 	RGB_FROM_PIXEL(*s, sfmt, r, g, b);
  1302 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, alpha);
  1303 	s++;
  1304 	dst++;
  1305     }
  1306     return n * 2;
  1307 }
  1308 
  1309 
  1310 
  1311 /* encode 32bpp rgb + a into 32bpp G0RAB format for blitting into 565 */
  1312 static int copy_transl_565(void *dst, Uint32 *src, int n,
  1313 			   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1314 {
  1315     int i;
  1316     Uint32 *d = dst;
  1317     for(i = 0; i < n; i++) {
  1318 	unsigned r, g, b, a;
  1319 	Uint16 pix;
  1320 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1321 	PIXEL_FROM_RGB(pix, dfmt, r, g, b);
  1322 	*d = ((pix & 0x7e0) << 16) | (pix & 0xf81f) | ((a << 2) & 0x7e0);
  1323 	src++;
  1324 	d++;
  1325     }
  1326     return n * 4;
  1327 }
  1328 
  1329 /* encode 32bpp rgb + a into 32bpp G0RAB format for blitting into 555 */
  1330 static int copy_transl_555(void *dst, Uint32 *src, int n,
  1331 			   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1332 {
  1333     int i;
  1334     Uint32 *d = dst;
  1335     for(i = 0; i < n; i++) {
  1336 	unsigned r, g, b, a;
  1337 	Uint16 pix;
  1338 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1339 	PIXEL_FROM_RGB(pix, dfmt, r, g, b);
  1340 	*d = ((pix & 0x3e0) << 16) | (pix & 0xfc1f) | ((a << 2) & 0x3e0);
  1341 	src++;
  1342 	d++;
  1343     }
  1344     return n * 4;
  1345 }
  1346 
  1347 /* decode translucent pixels from 32bpp GORAB to 32bpp rgb + a */
  1348 static int uncopy_transl_16(Uint32 *dst, void *src, int n,
  1349 			    RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1350 {
  1351     int i;
  1352     Uint32 *s = src;
  1353     for(i = 0; i < n; i++) {
  1354 	unsigned r, g, b, a;
  1355 	Uint32 pix = *s++;
  1356 	a = (pix & 0x3e0) >> 2;
  1357 	pix = (pix & ~0x3e0) | pix >> 16;
  1358 	RGB_FROM_PIXEL(pix, sfmt, r, g, b);
  1359 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, a);
  1360 	dst++;
  1361     }
  1362     return n * 4;
  1363 }
  1364 
  1365 /* encode 32bpp rgba into 32bpp rgba, keeping alpha (dual purpose) */
  1366 static int copy_32(void *dst, Uint32 *src, int n,
  1367 		   SDL_PixelFormat *sfmt, SDL_PixelFormat *dfmt)
  1368 {
  1369     int i;
  1370     Uint32 *d = dst;
  1371     for(i = 0; i < n; i++) {
  1372 	unsigned r, g, b, a;
  1373 	Uint32 pixel;
  1374 	RGBA_FROM_8888(*src, sfmt, r, g, b, a);
  1375 	PIXEL_FROM_RGB(pixel, dfmt, r, g, b);
  1376 	*d++ = pixel | a << 24;
  1377 	src++;
  1378     }
  1379     return n * 4;
  1380 }
  1381 
  1382 /* decode 32bpp rgba into 32bpp rgba, keeping alpha (dual purpose) */
  1383 static int uncopy_32(Uint32 *dst, void *src, int n,
  1384 		     RLEDestFormat *sfmt, SDL_PixelFormat *dfmt)
  1385 {
  1386     int i;
  1387     Uint32 *s = src;
  1388     for(i = 0; i < n; i++) {
  1389 	unsigned r, g, b, a;
  1390 	Uint32 pixel = *s++;
  1391 	RGB_FROM_PIXEL(pixel, sfmt, r, g, b);
  1392 	a = pixel >> 24;
  1393 	PIXEL_FROM_RGBA(*dst, dfmt, r, g, b, a);
  1394 	dst++;
  1395     }
  1396     return n * 4;
  1397 }
  1398 
  1399 #define ISOPAQUE(pixel, fmt) ((((pixel) & fmt->Amask) >> fmt->Ashift) == 255)
  1400 
  1401 #define ISTRANSL(pixel, fmt)	\
  1402     ((unsigned)((((pixel) & fmt->Amask) >> fmt->Ashift) - 1U) < 254U)
  1403 
  1404 /* convert surface to be quickly alpha-blittable onto dest, if possible */
  1405 static int RLEAlphaSurface(SDL_Surface *surface)
  1406 {
  1407     SDL_Surface *dest;
  1408     SDL_PixelFormat *df;
  1409     int maxsize = 0;
  1410     int max_opaque_run;
  1411     int max_transl_run = 65535;
  1412     unsigned masksum;
  1413     Uint8 *rlebuf, *dst;
  1414     int (*copy_opaque)(void *, Uint32 *, int,
  1415 		       SDL_PixelFormat *, SDL_PixelFormat *);
  1416     int (*copy_transl)(void *, Uint32 *, int,
  1417 		       SDL_PixelFormat *, SDL_PixelFormat *);
  1418 
  1419     dest = surface->map->dst;
  1420     if(!dest)
  1421 	return -1;
  1422     df = dest->format;
  1423     if(surface->format->BitsPerPixel != 32)
  1424 	return -1;		/* only 32bpp source supported */
  1425 
  1426     /* find out whether the destination is one we support,
  1427        and determine the max size of the encoded result */
  1428     masksum = df->Rmask | df->Gmask | df->Bmask;
  1429     switch(df->BytesPerPixel) {
  1430     case 2:
  1431 	/* 16bpp: only support 565 and 555 formats */
  1432 	switch(masksum) {
  1433 	case 0xffff:
  1434 	    if(df->Gmask == 0x07e0
  1435 	       || df->Rmask == 0x07e0 || df->Bmask == 0x07e0) {
  1436 		copy_opaque = copy_opaque_16;
  1437 		copy_transl = copy_transl_565;
  1438 	    } else
  1439 		return -1;
  1440 	    break;
  1441 	case 0x7fff:
  1442 	    if(df->Gmask == 0x03e0
  1443 	       || df->Rmask == 0x03e0 || df->Bmask == 0x03e0) {
  1444 		copy_opaque = copy_opaque_16;
  1445 		copy_transl = copy_transl_555;
  1446 	    } else
  1447 		return -1;
  1448 	    break;
  1449 	default:
  1450 	    return -1;
  1451 	}
  1452 	max_opaque_run = 255;	/* runs stored as bytes */
  1453 
  1454 	/* worst case is alternating opaque and translucent pixels,
  1455 	   with room for alignment padding between lines */
  1456 	maxsize = surface->h * (2 + (4 + 2) * (surface->w + 1)) + 2;
  1457 	break;
  1458     case 4:
  1459 	if(masksum != 0x00ffffff)
  1460 	    return -1;		/* requires unused high byte */
  1461 	copy_opaque = copy_32;
  1462 	copy_transl = copy_32;
  1463 	max_opaque_run = 255;	/* runs stored as short ints */
  1464 
  1465 	/* worst case is alternating opaque and translucent pixels */
  1466 	maxsize = surface->h * 2 * 4 * (surface->w + 1) + 4;
  1467 	break;
  1468     default:
  1469 	return -1;		/* anything else unsupported right now */
  1470     }
  1471 
  1472     maxsize += sizeof(RLEDestFormat);
  1473     rlebuf = (Uint8 *)malloc(maxsize);
  1474     if(!rlebuf) {
  1475 	SDL_OutOfMemory();
  1476 	return -1;
  1477     }
  1478     {
  1479 	/* save the destination format so we can undo the encoding later */
  1480 	RLEDestFormat *r = (RLEDestFormat *)rlebuf;
  1481 	r->BytesPerPixel = df->BytesPerPixel;
  1482 	r->Rloss = df->Rloss;
  1483 	r->Gloss = df->Gloss;
  1484 	r->Bloss = df->Bloss;
  1485 	r->Rshift = df->Rshift;
  1486 	r->Gshift = df->Gshift;
  1487 	r->Bshift = df->Bshift;
  1488 	r->Ashift = df->Ashift;
  1489 	r->Rmask = df->Rmask;
  1490 	r->Gmask = df->Gmask;
  1491 	r->Bmask = df->Bmask;
  1492 	r->Amask = df->Amask;
  1493     }
  1494     dst = rlebuf + sizeof(RLEDestFormat);
  1495 
  1496     /* Do the actual encoding */
  1497     {
  1498 	int x, y;
  1499 	int h = surface->h, w = surface->w;
  1500 	SDL_PixelFormat *sf = surface->format;
  1501 	Uint32 *src = (Uint32 *)surface->pixels;
  1502 	Uint8 *lastline = dst;	/* end of last non-blank line */
  1503 
  1504 	/* opaque counts are 8 or 16 bits, depending on target depth */
  1505 #define ADD_OPAQUE_COUNTS(n, m)			\
  1506 	if(df->BytesPerPixel == 4) {		\
  1507 	    ((Uint16 *)dst)[0] = n;		\
  1508 	    ((Uint16 *)dst)[1] = m;		\
  1509 	    dst += 4;				\
  1510 	} else {				\
  1511 	    dst[0] = n;				\
  1512 	    dst[1] = m;				\
  1513 	    dst += 2;				\
  1514 	}
  1515 
  1516 	/* translucent counts are always 16 bit */
  1517 #define ADD_TRANSL_COUNTS(n, m)		\
  1518 	(((Uint16 *)dst)[0] = n, ((Uint16 *)dst)[1] = m, dst += 4)
  1519 
  1520 	for(y = 0; y < h; y++) {
  1521 	    int runstart, skipstart;
  1522 	    int blankline = 0;
  1523 	    /* First encode all opaque pixels of a scan line */
  1524 	    x = 0;
  1525 	    do {
  1526 		int run, skip, len;
  1527 		skipstart = x;
  1528 		while(x < w && !ISOPAQUE(src[x], sf))
  1529 		    x++;
  1530 		runstart = x;
  1531 		while(x < w && ISOPAQUE(src[x], sf))
  1532 		    x++;
  1533 		skip = runstart - skipstart;
  1534 		if(skip == w)
  1535 		    blankline = 1;
  1536 		run = x - runstart;
  1537 		while(skip > max_opaque_run) {
  1538 		    ADD_OPAQUE_COUNTS(max_opaque_run, 0);
  1539 		    skip -= max_opaque_run;
  1540 		}
  1541 		len = MIN(run, max_opaque_run);
  1542 		ADD_OPAQUE_COUNTS(skip, len);
  1543 		dst += copy_opaque(dst, src + runstart, len, sf, df);
  1544 		runstart += len;
  1545 		run -= len;
  1546 		while(run) {
  1547 		    len = MIN(run, max_opaque_run);
  1548 		    ADD_OPAQUE_COUNTS(0, len);
  1549 		    dst += copy_opaque(dst, src + runstart, len, sf, df);
  1550 		    runstart += len;
  1551 		    run -= len;
  1552 		}
  1553 	    } while(x < w);
  1554 
  1555 	    /* Make sure the next output address is 32-bit aligned */
  1556 	    dst += (unsigned long)dst & 2;
  1557 
  1558 	    /* Next, encode all translucent pixels of the same scan line */
  1559 	    x = 0;
  1560 	    do {
  1561 		int run, skip, len;
  1562 		skipstart = x;
  1563 		while(x < w && !ISTRANSL(src[x], sf))
  1564 		    x++;
  1565 		runstart = x;
  1566 		while(x < w && ISTRANSL(src[x], sf))
  1567 		    x++;
  1568 		skip = runstart - skipstart;
  1569 		blankline &= (skip == w);
  1570 		run = x - runstart;
  1571 		while(skip > max_transl_run) {
  1572 		    ADD_TRANSL_COUNTS(max_transl_run, 0);
  1573 		    skip -= max_transl_run;
  1574 		}
  1575 		len = MIN(run, max_transl_run);
  1576 		ADD_TRANSL_COUNTS(skip, len);
  1577 		dst += copy_transl(dst, src + runstart, len, sf, df);
  1578 		runstart += len;
  1579 		run -= len;
  1580 		while(run) {
  1581 		    len = MIN(run, max_transl_run);
  1582 		    ADD_TRANSL_COUNTS(0, len);
  1583 		    dst += copy_transl(dst, src + runstart, len, sf, df);
  1584 		    runstart += len;
  1585 		    run -= len;
  1586 		}
  1587 		if(!blankline)
  1588 		    lastline = dst;
  1589 	    } while(x < w);
  1590 
  1591 	    src += surface->pitch >> 2;
  1592 	}
  1593 	dst = lastline;		/* back up past trailing blank lines */
  1594 	ADD_OPAQUE_COUNTS(0, 0);
  1595     }
  1596 
  1597 #undef ADD_OPAQUE_COUNTS
  1598 #undef ADD_TRANSL_COUNTS
  1599 
  1600     /* Now that we have it encoded, release the original pixels */
  1601     if((surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1602        && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1603 	free( surface->pixels );
  1604 	surface->pixels = NULL;
  1605     }
  1606 
  1607     /* realloc the buffer to release unused memory */
  1608     {
  1609 	Uint8 *p = realloc(rlebuf, dst - rlebuf);
  1610 	if(!p)
  1611 	    p = rlebuf;
  1612 	surface->map->sw_data->aux_data = p;
  1613     }
  1614 
  1615     return 0;
  1616 }
  1617 
  1618 static Uint32 getpix_8(Uint8 *srcbuf)
  1619 {
  1620     return *srcbuf;
  1621 }
  1622 
  1623 static Uint32 getpix_16(Uint8 *srcbuf)
  1624 {
  1625     return *(Uint16 *)srcbuf;
  1626 }
  1627 
  1628 static Uint32 getpix_24(Uint8 *srcbuf)
  1629 {
  1630 #if SDL_BYTEORDER == SDL_LIL_ENDIAN
  1631     return srcbuf[0] + (srcbuf[1] << 8) + (srcbuf[2] << 16);
  1632 #else
  1633     return (srcbuf[0] << 16) + (srcbuf[1] << 8) + srcbuf[2];
  1634 #endif
  1635 }
  1636 
  1637 static Uint32 getpix_32(Uint8 *srcbuf)
  1638 {
  1639     return *(Uint32 *)srcbuf;
  1640 }
  1641 
  1642 typedef Uint32 (*getpix_func)(Uint8 *);
  1643 
  1644 static getpix_func getpixes[4] = {
  1645     getpix_8, getpix_16, getpix_24, getpix_32
  1646 };
  1647 
  1648 static int RLEColorkeySurface(SDL_Surface *surface)
  1649 {
  1650         Uint8 *rlebuf, *dst;
  1651 	int maxn;
  1652 	int y;
  1653 	Uint8 *srcbuf, *curbuf, *lastline;
  1654 	int maxsize = 0;
  1655 	int skip, run;
  1656 	int bpp = surface->format->BytesPerPixel;
  1657 	getpix_func getpix;
  1658 	Uint32 ckey, rgbmask;
  1659 	int w, h;
  1660 
  1661 	/* calculate the worst case size for the compressed surface */
  1662 	switch(bpp) {
  1663 	case 1:
  1664 	    /* worst case is alternating opaque and transparent pixels,
  1665 	       starting with an opaque pixel */
  1666 	    maxsize = surface->h * 3 * (surface->w / 2 + 1) + 2;
  1667 	    break;
  1668 	case 2:
  1669 	case 3:
  1670 	    /* worst case is solid runs, at most 255 pixels wide */
  1671 	    maxsize = surface->h * (2 * (surface->w / 255 + 1)
  1672 				    + surface->w * bpp) + 2;
  1673 	    break;
  1674 	case 4:
  1675 	    /* worst case is solid runs, at most 65535 pixels wide */
  1676 	    maxsize = surface->h * (4 * (surface->w / 65535 + 1)
  1677 				    + surface->w * 4) + 4;
  1678 	    break;
  1679 	}
  1680 
  1681 	rlebuf = (Uint8 *)malloc(maxsize);
  1682 	if ( rlebuf == NULL ) {
  1683 		SDL_OutOfMemory();
  1684 		return(-1);
  1685 	}
  1686 
  1687 	/* Set up the conversion */
  1688 	srcbuf = (Uint8 *)surface->pixels;
  1689 	curbuf = srcbuf;
  1690 	maxn = bpp == 4 ? 65535 : 255;
  1691 	skip = run = 0;
  1692 	dst = rlebuf;
  1693 	rgbmask = ~surface->format->Amask;
  1694 	ckey = surface->format->colorkey & rgbmask;
  1695 	lastline = dst;
  1696 	getpix = getpixes[bpp - 1];
  1697 	w = surface->w;
  1698 	h = surface->h;
  1699 
  1700 #define ADD_COUNTS(n, m)			\
  1701 	if(bpp == 4) {				\
  1702 	    ((Uint16 *)dst)[0] = n;		\
  1703 	    ((Uint16 *)dst)[1] = m;		\
  1704 	    dst += 4;				\
  1705 	} else {				\
  1706 	    dst[0] = n;				\
  1707 	    dst[1] = m;				\
  1708 	    dst += 2;				\
  1709 	}
  1710 
  1711 	for(y = 0; y < h; y++) {
  1712 	    int x = 0;
  1713 	    int blankline = 0;
  1714 	    do {
  1715 		int run, skip, len;
  1716 		int runstart;
  1717 		int skipstart = x;
  1718 
  1719 		/* find run of transparent, then opaque pixels */
  1720 		while(x < w && (getpix(srcbuf + x * bpp) & rgbmask) == ckey)
  1721 		    x++;
  1722 		runstart = x;
  1723 		while(x < w && (getpix(srcbuf + x * bpp) & rgbmask) != ckey)
  1724 		    x++;
  1725 		skip = runstart - skipstart;
  1726 		if(skip == w)
  1727 		    blankline = 1;
  1728 		run = x - runstart;
  1729 
  1730 		/* encode segment */
  1731 		while(skip > maxn) {
  1732 		    ADD_COUNTS(maxn, 0);
  1733 		    skip -= maxn;
  1734 		}
  1735 		len = MIN(run, maxn);
  1736 		ADD_COUNTS(skip, len);
  1737 		memcpy(dst, srcbuf + runstart * bpp, len * bpp);
  1738 		dst += len * bpp;
  1739 		run -= len;
  1740 		runstart += len;
  1741 		while(run) {
  1742 		    len = MIN(run, maxn);
  1743 		    ADD_COUNTS(0, len);
  1744 		    memcpy(dst, srcbuf + runstart * bpp, len * bpp);
  1745 		    dst += len * bpp;
  1746 		    runstart += len;
  1747 		    run -= len;
  1748 		}
  1749 		if(!blankline)
  1750 		    lastline = dst;
  1751 	    } while(x < w);
  1752 
  1753 	    srcbuf += surface->pitch;
  1754 	}
  1755 	dst = lastline;		/* back up bast trailing blank lines */
  1756 	ADD_COUNTS(0, 0);
  1757 
  1758 #undef ADD_COUNTS
  1759 
  1760 	/* Now that we have it encoded, release the original pixels */
  1761 	if((surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1762 	   && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1763 	    free( surface->pixels );
  1764 	    surface->pixels = NULL;
  1765 	}
  1766 
  1767 	/* realloc the buffer to release unused memory */
  1768 	{
  1769 	    /* If realloc returns NULL, the original block is left intact */
  1770 	    Uint8 *p = realloc(rlebuf, dst - rlebuf);
  1771 	    if(!p)
  1772 		p = rlebuf;
  1773 	    surface->map->sw_data->aux_data = p;
  1774 	}
  1775 
  1776 	return(0);
  1777 }
  1778 
  1779 int SDL_RLESurface(SDL_Surface *surface)
  1780 {
  1781 	int retcode;
  1782 
  1783 	/* Clear any previous RLE conversion */
  1784 	if ( (surface->flags & SDL_RLEACCEL) == SDL_RLEACCEL ) {
  1785 		SDL_UnRLESurface(surface, 1);
  1786 	}
  1787 
  1788 	/* We don't support RLE encoding of bitmaps */
  1789 	if ( surface->format->BitsPerPixel < 8 ) {
  1790 		return(-1);
  1791 	}
  1792 
  1793 	/* Lock the surface if it's in hardware */
  1794 	if ( SDL_MUSTLOCK(surface) ) {
  1795 		if ( SDL_LockSurface(surface) < 0 ) {
  1796 			return(-1);
  1797 		}
  1798 	}
  1799 
  1800 	/* Encode */
  1801 	if((surface->flags & SDL_SRCCOLORKEY) == SDL_SRCCOLORKEY) {
  1802 	    retcode = RLEColorkeySurface(surface);
  1803 	} else {
  1804 	    if((surface->flags & SDL_SRCALPHA) == SDL_SRCALPHA
  1805 	       && surface->format->Amask != 0)
  1806 		retcode = RLEAlphaSurface(surface);
  1807 	    else
  1808 		retcode = -1;	/* no RLE for per-surface alpha sans ckey */
  1809 	}
  1810 
  1811 	/* Unlock the surface if it's in hardware */
  1812 	if ( SDL_MUSTLOCK(surface) ) {
  1813 		SDL_UnlockSurface(surface);
  1814 	}
  1815 
  1816 	if(retcode < 0)
  1817 	    return -1;
  1818 
  1819 	/* The surface is now accelerated */
  1820 	surface->flags |= SDL_RLEACCEL;
  1821 
  1822 	return(0);
  1823 }
  1824 
  1825 /*
  1826  * Un-RLE a surface with pixel alpha
  1827  * This may not give back exactly the image before RLE-encoding; all
  1828  * completely transparent pixels will be lost, and colour and alpha depth
  1829  * may have been reduced (when encoding for 16bpp targets).
  1830  */
  1831 static SDL_bool UnRLEAlpha(SDL_Surface *surface)
  1832 {
  1833     Uint8 *srcbuf;
  1834     Uint32 *dst;
  1835     SDL_PixelFormat *sf = surface->format;
  1836     RLEDestFormat *df = surface->map->sw_data->aux_data;
  1837     int (*uncopy_opaque)(Uint32 *, void *, int,
  1838 			 RLEDestFormat *, SDL_PixelFormat *);
  1839     int (*uncopy_transl)(Uint32 *, void *, int,
  1840 			 RLEDestFormat *, SDL_PixelFormat *);
  1841     int w = surface->w;
  1842     int bpp = df->BytesPerPixel;
  1843 
  1844     if(bpp == 2) {
  1845 	uncopy_opaque = uncopy_opaque_16;
  1846 	uncopy_transl = uncopy_transl_16;
  1847     } else {
  1848 	uncopy_opaque = uncopy_transl = uncopy_32;
  1849     }
  1850 
  1851     surface->pixels = malloc(surface->h * surface->pitch);
  1852     if ( !surface->pixels ) {
  1853         return(SDL_FALSE);
  1854     }
  1855     /* fill background with transparent pixels */
  1856     memset(surface->pixels, 0, surface->h * surface->pitch);
  1857 
  1858     dst = surface->pixels;
  1859     srcbuf = (Uint8 *)(df + 1);
  1860     for(;;) {
  1861 	/* copy opaque pixels */
  1862 	int ofs = 0;
  1863 	do {
  1864 	    unsigned run;
  1865 	    if(bpp == 2) {
  1866 		ofs += srcbuf[0];
  1867 		run = srcbuf[1];
  1868 		srcbuf += 2;
  1869 	    } else {
  1870 		ofs += ((Uint16 *)srcbuf)[0];
  1871 		run = ((Uint16 *)srcbuf)[1];
  1872 		srcbuf += 4;
  1873 	    }
  1874 	    if(run) {
  1875 		srcbuf += uncopy_opaque(dst + ofs, srcbuf, run, df, sf);
  1876 		ofs += run;
  1877 	    } else if(!ofs)
  1878 		return(SDL_TRUE);
  1879 	} while(ofs < w);
  1880 
  1881 	/* skip padding if needed */
  1882 	if(bpp == 2)
  1883 	    srcbuf += (unsigned long)srcbuf & 2;
  1884 	
  1885 	/* copy translucent pixels */
  1886 	ofs = 0;
  1887 	do {
  1888 	    unsigned run;
  1889 	    ofs += ((Uint16 *)srcbuf)[0];
  1890 	    run = ((Uint16 *)srcbuf)[1];
  1891 	    srcbuf += 4;
  1892 	    if(run) {
  1893 		srcbuf += uncopy_transl(dst + ofs, srcbuf, run, df, sf);
  1894 		ofs += run;
  1895 	    }
  1896 	} while(ofs < w);
  1897 	dst += surface->pitch >> 2;
  1898     }
  1899     /* Make the compiler happy */
  1900     return(SDL_TRUE);
  1901 }
  1902 
  1903 void SDL_UnRLESurface(SDL_Surface *surface, int recode)
  1904 {
  1905     if ( (surface->flags & SDL_RLEACCEL) == SDL_RLEACCEL ) {
  1906 	surface->flags &= ~SDL_RLEACCEL;
  1907 
  1908 	if(recode && (surface->flags & SDL_PREALLOC) != SDL_PREALLOC
  1909 	   && (surface->flags & SDL_HWSURFACE) != SDL_HWSURFACE) {
  1910 	    if((surface->flags & SDL_SRCCOLORKEY) == SDL_SRCCOLORKEY) {
  1911 		SDL_Rect full;
  1912 		unsigned alpha_flag;
  1913 
  1914 		/* re-create the original surface */
  1915 		surface->pixels = malloc(surface->h * surface->pitch);
  1916 		if ( !surface->pixels ) {
  1917 			/* Oh crap... */
  1918 			surface->flags |= SDL_RLEACCEL;
  1919 			return;
  1920 		}
  1921 
  1922 		/* fill it with the background colour */
  1923 		SDL_FillRect(surface, NULL, surface->format->colorkey);
  1924 
  1925 		/* now render the encoded surface */
  1926 		full.x = full.y = 0;
  1927 		full.w = surface->w;
  1928 		full.h = surface->h;
  1929 		alpha_flag = surface->flags & SDL_SRCALPHA;
  1930 		surface->flags &= ~SDL_SRCALPHA; /* opaque blit */
  1931 		SDL_RLEBlit(surface, &full, surface, &full);
  1932 		surface->flags |= alpha_flag;
  1933 	    } else {
  1934 		if ( !UnRLEAlpha(surface) ) {
  1935 		    /* Oh crap... */
  1936 		    surface->flags |= SDL_RLEACCEL;
  1937 		    return;
  1938 		}
  1939 	    }
  1940 	}
  1941 
  1942 	if ( surface->map && surface->map->sw_data->aux_data ) {
  1943 	    free(surface->map->sw_data->aux_data);
  1944 	    surface->map->sw_data->aux_data = NULL;
  1945 	}
  1946     }
  1947 }
  1948 
  1949