1 /* x86_convert.c - convert RImage to XImage with x86 optimizations
3 * Raster graphics library
5 * Copyright (c) 2000 Alfredo K. Kojima
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Library General Public
9 * License as published by the Free Software Foundation; either
10 * version 2 of the License, or (at your option) any later version.
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Library General Public License for more details.
17 * You should have received a copy of the GNU Library General Public
18 * License along with this library; if not, write to the Free
19 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
32 static int result
= -1;
42 ("pushfl \n" // check whether cpuid supported
44 "movl %%eax, %%ebx \n"
45 "xorl 1<<21, %%eax \n"
50 "xorl %%eax, %%ebx \n"
51 "andl 1<<21, %%eax \n"
53 "xorl %%eax, %%eax \n"
57 "test 1<<23, %%edx \n"
73 * 32/8 24/8 32/16 24/16 32/24 24/24
78 * - try to align stack (local variable space) into quadword boundary
85 x86_mmx_TrueColor_32_to_16(unsigned char *image
, // 8
86 unsigned short *ximage
, // 12
95 unsigned int roffs
, // 48
96 unsigned int goffs
, // 52
97 unsigned int boffs
, // 56
100 int line_offset
) // 68
104 long long rrggbbaa;// -16
105 long long pixel; //-24
106 short *tmp_err; //-32
107 short *tmp_nerr; //-36
112 "subl $128, %esp \n" // alloc some more stack
116 // pack dr, dg and db into mm6
117 "movl 36(%ebp), %eax \n"
118 "movl 40(%ebp), %ebx \n"
119 "movw %ax, -16(%ebp) \n"
121 "movw %bx, -14(%ebp) \n"
122 "movl 44(%ebp), %eax \n"
123 "movw $0, -10(%ebp) \n"
124 "movw %ax, -12(%ebp) \n"
126 "movq -16(%ebp), %mm6 \n" // dr dg db 0
128 // pack 4|4|4|4 into mm7, for shifting (/16)
129 "movl $0x00040004, -16(%ebp) \n"
130 "movl $0x00040004, -12(%ebp) \n"
131 "movq -16(%ebp), %mm7 \n"
133 // store constant values for using with mmx when dithering
134 "movl $0x00070007, -16(%ebp) \n"
135 "movl $0x00070007, -12(%ebp) \n"
136 "movq -16(%ebp), %mm5 \n"
138 "movl $0x00050005, -16(%ebp) \n"
139 "movl $0x00050005, -12(%ebp) \n"
140 "movq -16(%ebp), %mm4 \n"
142 "movl $0x00030003, -16(%ebp) \n"
143 "movl $0x00030003, -12(%ebp) \n"
144 "movq -16(%ebp), %mm3 \n"
146 // process 1 pixel / cycle, each component treated as 16bit
147 "movl 8(%ebp), %esi \n" // esi = image->data
150 "movl 60(%ebp), %eax \n"
151 "movl %eax, -4(%ebp) \n" // x = width
153 "movl 64(%ebp), %eax \n"
154 "decl %eax \n" // y--
155 "movl %eax, 64(%ebp) \n"
156 "js .Enda \n" // if y < 0, goto end
158 "jz .LoopY_1a \n" // if (y&1) goto LoopY_1
162 "movl 16(%ebp), %ebx \n" // ebx = err
163 "movl %ebx, -36(%ebp) \n" // [-36] = err
164 "movl 20(%ebp), %eax \n" //
165 "movl %eax, -32(%ebp) \n" // [-32] = nerr
171 "movl 20(%ebp), %ebx \n" // ebx = nerr
172 "movl %ebx, -36(%ebp) \n" // [-36] = nerr
173 "movl 16(%ebp), %eax \n" //
174 "movl %eax, -32(%ebp) \n" // [-32] = eerr
179 // calculate errors and pixel components
181 // depend on ebx, esi, mm6
182 "movq (%ebx), %mm1 \n" // mm1 = error[0..3]
183 "punpcklbw (%esi), %mm0 \n" // mm0 = image->data[0..3]
184 "psrlw $8, %mm0 \n" // fixup mm0
185 "paddusb %mm1, %mm0 \n" // mm0 = mm0 + mm1 (sat. to 255)
186 "movq %mm0, -24(%ebp) \n" // save the pixel
188 "movzwl -24(%ebp), %ecx \n" // ecx = pixel.red
189 "movl 24(%ebp), %edi \n" // edi = rtable
190 "leal (%edi, %ecx, 2), %eax \n" // eax = &rtable[pixel.red]
191 "movl (%eax), %edx \n" // edx = rtable[pixel.red]
192 "movw %dx, -16(%ebp) \n" // save rr
194 "movzwl -22(%ebp), %ecx \n" // ecx = pixel.green
195 "movl 28(%ebp), %edi \n" // edi = gtable
196 "leal (%edi, %ecx, 2), %eax \n" // eax = >able[pixel.green]
197 "movl (%eax), %edx \n" // ebx = gtable[pixel.green]
198 "movw %dx, -14(%ebp) \n" // save gg
200 "movzwl -20(%ebp), %ecx \n" // ecx = pixel.blue
201 "movl 32(%ebp), %edi \n" // ebx = btable
202 "leal (%edi, %ecx, 2), %eax \n" // eax = &btable[pixel.blue]
203 "movl (%eax), %edx \n" // ecx = btable[pixel.blue]
204 "movw %dx, -12(%ebp) \n" // save bb
206 "movw $0, -10(%ebp) \n" // save dummy aa
208 "movq -16(%ebp), %mm1 \n" // load mm1 with rrggbbaa
209 "pmullw %mm6, %mm1 \n" // mm1 = rr*dr|...
210 "psubsw %mm1, %mm0 \n" // error = pixel - mm1
213 // distribute the error
215 // depend on mm0, mm7, mm3, mm4, mm5
217 "movl -36(%ebp), %ebx \n"
220 "pmullw %mm5, %mm1 \n" // mm1 = mm1*7
221 "psrlw %mm7, %mm1 \n" // mm1 = mm1/16
222 "paddw 8(%ebx), %mm1 \n"
223 "movq %mm1, 8(%ebx) \n" // err[x+1,y] = rer*7/16
226 "movl -32(%ebp), %ebx \n"
229 "pmullw %mm4, %mm1 \n" // mm1 = mm1*5
230 "psrlw %mm7, %mm1 \n" // mm1 = mm1/16
231 "paddw -8(%ebx), %mm1 \n"
232 "movq %mm1, -8(%ebx) \n" // err[x-1,y+1] += rer*3/16
235 "pmullw %mm3, %mm1 \n" // mm1 = mm1*3
236 "psrlw %mm7, %mm1 \n" // mm1 = mm1/16
237 "paddw 8(%ebx), %mm1 \n"
238 "movq %mm1, (%ebx) \n" // err[x,y+1] += rer*5/16
240 "psrlw %mm7, %mm0 \n" // mm0 = mm0/16
241 "movq %mm0, 8(%ebx) \n" // err[x+1,y+1] = rer/16
244 // calculate final pixel value and store
245 "movl 48(%ebp), %ecx \n"
246 "movw -16(%ebp), %ax \n"
247 "shlw %cl, %ax \n" //NP* ax = r<<roffs
249 "movl 52(%ebp), %ecx \n"
250 "movw -14(%ebp), %bx \n"
251 "shlw %cl, %bx \n" //NP*
254 "movl 56(%ebp), %ecx \n"
255 "movw -12(%ebp), %bx \n"
256 "shlw %cl, %bx \n" //NP*
259 "movl 12(%ebp), %edx \n"
260 "movw %ax, (%edx) \n"
261 "addl $2, %edx \n" // increment ximage
262 "movl %edx, 12(%ebp) \n"
264 // prepare for next iteration on X
266 "addl $8, -32(%ebp) \n" // nerr += 8
268 "movl -36(%ebp), %ebx \n"
270 "movl %ebx, -36(%ebp) \n" // ebx = err += 8
273 // Note: in the last pixel, this would cause an invalid memory access
274 // because, punpcklbw is used (which reads 8 bytes) and the last
275 // pixel is only 4 bytes. This is no problem because the image data
276 // was allocated with extra 4 bytes when created.
277 "addl $4, %esi \n" // image->data += 4
280 "decl -4(%ebp) \n" // x--
281 "jnz .LoopXa \n" // if x>0, goto .LoopX
285 "addl 68(%ebp), %edx \n" // add extra offset to ximage
286 "movl %edx, 12(%ebp) \n"
291 ".Enda: \n" // THE END
300 #endif /* ASM_X86_MMX */
304 /* convert and dither the image to XImage */
305 for (y
=0; y
<image
->height
; y
++) {
309 for (x
=0; x
<image
->width
*3; x
+=3, ptr
+=channels
) {
312 pixel
= *ptr
+ err
[x
];
313 if (pixel
<0) pixel
=0; else if (pixel
>0xff) pixel
=0xff;
319 pixel
= *(ptr
+1) + err
[x
+1];
320 if (pixel
<0) pixel
=0; else if (pixel
>0xff) pixel
=0xff;
326 pixel
= *(ptr
+2) + err
[x
+2];
327 if (pixel
<0) pixel
=0; else if (pixel
>0xff) pixel
=0xff;
332 *optr
++ = pixels
[r
*cpcpc
+ g
*cpc
+ b
];
334 /* distribute error */
348 nerr
[x
+1+3*1]=ger
-2*g
;
349 nerr
[x
+2+3*1]=ber
-2*b
;
351 /* skip to next line */
356 optr
+= ximg
->image
->bytes_per_line
- image
->width
;
363 x86_PseudoColor_32_to_8(unsigned char *image
, // 8
364 unsigned char *ximage
, // 12
373 unsigned long *pixels
, // 48
377 int line_offset
) // 64
382 "subl $128, %esp \n" // alloc some stack space
385 // process 1 pixel / cycle, each component treated as 16bit
386 "movl 8(%ebp), %esi \n" // esi = image->data
389 "movl 56(%ebp), %eax \n"
390 "movl %eax, -4(%ebp) \n" // x = width
392 "movl 60(%ebp), %eax \n"
393 "decl %eax \n" // y--
394 "movl %eax, 64(%ebp) \n"
395 "js .Endb \n" // if y < 0, goto end
397 "jz .LoopY_1b \n" // if (y&1) goto LoopY_1
401 "movl 16(%ebp), %ebx \n" // ebx = err
402 "movl %ebx, -36(%ebp) \n" // [-36] = err
403 "movl 20(%ebp), %eax \n" //
404 "movl %eax, -32(%ebp) \n" // [-32] = nerr
406 "movl $0, -32(%ebp) \n" // init error of nerr[0] to 0
412 "movl 20(%ebp), %ebx \n" // ebx = nerr
413 "movl %ebx, -36(%ebp) \n" // [-36] = nerr
414 "movl 16(%ebp), %eax \n" //
415 "movl %eax, -32(%ebp) \n" // [-32] = err
417 "movl $0, -32(%ebp) \n" // init error of nerr[0] to 0
421 "movl (%esi), %edx \n" // fetch a pixel