h264 luma motion compensation in mmx2/3dnow
[libav.git] / libavcodec / i386 / dsputil_mmx_rnd.h
1 /*
2 * DSP utils mmx functions are compiled twice for rnd/no_rnd
3 * Copyright (c) 2000, 2001 Fabrice Bellard.
4 * Copyright (c) 2003-2004 Michael Niedermayer <michaelni@gmx.at>
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 *
20 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
21 * mostly rewritten by Michael Niedermayer <michaelni@gmx.at>
22 * and improved by Zdenek Kabelac <kabi@users.sf.net>
23 */
24
25 // put_pixels
26 static void DEF(put, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
27 {
28 MOVQ_BFE(mm6);
29 __asm __volatile(
30 "lea (%3, %3), %%eax \n\t"
31 ".balign 8 \n\t"
32 "1: \n\t"
33 "movq (%1), %%mm0 \n\t"
34 "movq 1(%1), %%mm1 \n\t"
35 "movq (%1, %3), %%mm2 \n\t"
36 "movq 1(%1, %3), %%mm3 \n\t"
37 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
38 "movq %%mm4, (%2) \n\t"
39 "movq %%mm5, (%2, %3) \n\t"
40 "addl %%eax, %1 \n\t"
41 "addl %%eax, %2 \n\t"
42 "movq (%1), %%mm0 \n\t"
43 "movq 1(%1), %%mm1 \n\t"
44 "movq (%1, %3), %%mm2 \n\t"
45 "movq 1(%1, %3), %%mm3 \n\t"
46 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
47 "movq %%mm4, (%2) \n\t"
48 "movq %%mm5, (%2, %3) \n\t"
49 "addl %%eax, %1 \n\t"
50 "addl %%eax, %2 \n\t"
51 "subl $4, %0 \n\t"
52 "jnz 1b \n\t"
53 :"+g"(h), "+S"(pixels), "+D"(block)
54 :"r"(line_size)
55 :"eax", "memory");
56 }
57
58 static void DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
59 {
60 MOVQ_BFE(mm6);
61 __asm __volatile(
62 "testl $1, %0 \n\t"
63 " jz 1f \n\t"
64 "movq (%1), %%mm0 \n\t"
65 "movq (%2), %%mm1 \n\t"
66 "addl %4, %1 \n\t"
67 "addl $8, %2 \n\t"
68 PAVGB(%%mm0, %%mm1, %%mm4, %%mm6)
69 "movq %%mm4, (%3) \n\t"
70 "addl %5, %3 \n\t"
71 "decl %0 \n\t"
72 ".balign 8 \n\t"
73 "1: \n\t"
74 "movq (%1), %%mm0 \n\t"
75 "movq (%2), %%mm1 \n\t"
76 "addl %4, %1 \n\t"
77 "movq (%1), %%mm2 \n\t"
78 "movq 8(%2), %%mm3 \n\t"
79 "addl %4, %1 \n\t"
80 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
81 "movq %%mm4, (%3) \n\t"
82 "addl %5, %3 \n\t"
83 "movq %%mm5, (%3) \n\t"
84 "addl %5, %3 \n\t"
85 "movq (%1), %%mm0 \n\t"
86 "movq 16(%2), %%mm1 \n\t"
87 "addl %4, %1 \n\t"
88 "movq (%1), %%mm2 \n\t"
89 "movq 24(%2), %%mm3 \n\t"
90 "addl %4, %1 \n\t"
91 "addl $32, %2 \n\t"
92 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
93 "movq %%mm4, (%3) \n\t"
94 "addl %5, %3 \n\t"
95 "movq %%mm5, (%3) \n\t"
96 "addl %5, %3 \n\t"
97 "subl $4, %0 \n\t"
98 "jnz 1b \n\t"
99 #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
100 :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
101 #else
102 :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
103 #endif
104 :"S"(src1Stride), "D"(dstStride)
105 :"memory");
106 }
107
108 static void DEF(put, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
109 {
110 MOVQ_BFE(mm6);
111 __asm __volatile(
112 "lea (%3, %3), %%eax \n\t"
113 ".balign 8 \n\t"
114 "1: \n\t"
115 "movq (%1), %%mm0 \n\t"
116 "movq 1(%1), %%mm1 \n\t"
117 "movq (%1, %3), %%mm2 \n\t"
118 "movq 1(%1, %3), %%mm3 \n\t"
119 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
120 "movq %%mm4, (%2) \n\t"
121 "movq %%mm5, (%2, %3) \n\t"
122 "movq 8(%1), %%mm0 \n\t"
123 "movq 9(%1), %%mm1 \n\t"
124 "movq 8(%1, %3), %%mm2 \n\t"
125 "movq 9(%1, %3), %%mm3 \n\t"
126 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
127 "movq %%mm4, 8(%2) \n\t"
128 "movq %%mm5, 8(%2, %3) \n\t"
129 "addl %%eax, %1 \n\t"
130 "addl %%eax, %2 \n\t"
131 "movq (%1), %%mm0 \n\t"
132 "movq 1(%1), %%mm1 \n\t"
133 "movq (%1, %3), %%mm2 \n\t"
134 "movq 1(%1, %3), %%mm3 \n\t"
135 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
136 "movq %%mm4, (%2) \n\t"
137 "movq %%mm5, (%2, %3) \n\t"
138 "movq 8(%1), %%mm0 \n\t"
139 "movq 9(%1), %%mm1 \n\t"
140 "movq 8(%1, %3), %%mm2 \n\t"
141 "movq 9(%1, %3), %%mm3 \n\t"
142 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
143 "movq %%mm4, 8(%2) \n\t"
144 "movq %%mm5, 8(%2, %3) \n\t"
145 "addl %%eax, %1 \n\t"
146 "addl %%eax, %2 \n\t"
147 "subl $4, %0 \n\t"
148 "jnz 1b \n\t"
149 :"+g"(h), "+S"(pixels), "+D"(block)
150 :"r"(line_size)
151 :"eax", "memory");
152 }
153
154 static void DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
155 {
156 MOVQ_BFE(mm6);
157 __asm __volatile(
158 "testl $1, %0 \n\t"
159 " jz 1f \n\t"
160 "movq (%1), %%mm0 \n\t"
161 "movq (%2), %%mm1 \n\t"
162 "movq 8(%1), %%mm2 \n\t"
163 "movq 8(%2), %%mm3 \n\t"
164 "addl %4, %1 \n\t"
165 "addl $16, %2 \n\t"
166 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
167 "movq %%mm4, (%3) \n\t"
168 "movq %%mm5, 8(%3) \n\t"
169 "addl %5, %3 \n\t"
170 "decl %0 \n\t"
171 ".balign 8 \n\t"
172 "1: \n\t"
173 "movq (%1), %%mm0 \n\t"
174 "movq (%2), %%mm1 \n\t"
175 "movq 8(%1), %%mm2 \n\t"
176 "movq 8(%2), %%mm3 \n\t"
177 "addl %4, %1 \n\t"
178 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
179 "movq %%mm4, (%3) \n\t"
180 "movq %%mm5, 8(%3) \n\t"
181 "addl %5, %3 \n\t"
182 "movq (%1), %%mm0 \n\t"
183 "movq 16(%2), %%mm1 \n\t"
184 "movq 8(%1), %%mm2 \n\t"
185 "movq 24(%2), %%mm3 \n\t"
186 "addl %4, %1 \n\t"
187 PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
188 "movq %%mm4, (%3) \n\t"
189 "movq %%mm5, 8(%3) \n\t"
190 "addl %5, %3 \n\t"
191 "addl $32, %2 \n\t"
192 "subl $2, %0 \n\t"
193 "jnz 1b \n\t"
194 #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
195 :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
196 #else
197 :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
198 #endif
199 :"S"(src1Stride), "D"(dstStride)
200 :"memory");
201 }
202
203 static void DEF(put, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
204 {
205 MOVQ_BFE(mm6);
206 __asm __volatile(
207 "lea (%3, %3), %%eax \n\t"
208 "movq (%1), %%mm0 \n\t"
209 ".balign 8 \n\t"
210 "1: \n\t"
211 "movq (%1, %3), %%mm1 \n\t"
212 "movq (%1, %%eax),%%mm2 \n\t"
213 PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
214 "movq %%mm4, (%2) \n\t"
215 "movq %%mm5, (%2, %3) \n\t"
216 "addl %%eax, %1 \n\t"
217 "addl %%eax, %2 \n\t"
218 "movq (%1, %3), %%mm1 \n\t"
219 "movq (%1, %%eax),%%mm0 \n\t"
220 PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
221 "movq %%mm4, (%2) \n\t"
222 "movq %%mm5, (%2, %3) \n\t"
223 "addl %%eax, %1 \n\t"
224 "addl %%eax, %2 \n\t"
225 "subl $4, %0 \n\t"
226 "jnz 1b \n\t"
227 :"+g"(h), "+S"(pixels), "+D"(block)
228 :"r"(line_size)
229 :"eax", "memory");
230 }
231
232 static void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
233 {
234 MOVQ_ZERO(mm7);
235 SET_RND(mm6); // =2 for rnd and =1 for no_rnd version
236 __asm __volatile(
237 "movq (%1), %%mm0 \n\t"
238 "movq 1(%1), %%mm4 \n\t"
239 "movq %%mm0, %%mm1 \n\t"
240 "movq %%mm4, %%mm5 \n\t"
241 "punpcklbw %%mm7, %%mm0 \n\t"
242 "punpcklbw %%mm7, %%mm4 \n\t"
243 "punpckhbw %%mm7, %%mm1 \n\t"
244 "punpckhbw %%mm7, %%mm5 \n\t"
245 "paddusw %%mm0, %%mm4 \n\t"
246 "paddusw %%mm1, %%mm5 \n\t"
247 "xorl %%eax, %%eax \n\t"
248 "addl %3, %1 \n\t"
249 ".balign 8 \n\t"
250 "1: \n\t"
251 "movq (%1, %%eax), %%mm0 \n\t"
252 "movq 1(%1, %%eax), %%mm2 \n\t"
253 "movq %%mm0, %%mm1 \n\t"
254 "movq %%mm2, %%mm3 \n\t"
255 "punpcklbw %%mm7, %%mm0 \n\t"
256 "punpcklbw %%mm7, %%mm2 \n\t"
257 "punpckhbw %%mm7, %%mm1 \n\t"
258 "punpckhbw %%mm7, %%mm3 \n\t"
259 "paddusw %%mm2, %%mm0 \n\t"
260 "paddusw %%mm3, %%mm1 \n\t"
261 "paddusw %%mm6, %%mm4 \n\t"
262 "paddusw %%mm6, %%mm5 \n\t"
263 "paddusw %%mm0, %%mm4 \n\t"
264 "paddusw %%mm1, %%mm5 \n\t"
265 "psrlw $2, %%mm4 \n\t"
266 "psrlw $2, %%mm5 \n\t"
267 "packuswb %%mm5, %%mm4 \n\t"
268 "movq %%mm4, (%2, %%eax) \n\t"
269 "addl %3, %%eax \n\t"
270
271 "movq (%1, %%eax), %%mm2 \n\t" // 0 <-> 2 1 <-> 3
272 "movq 1(%1, %%eax), %%mm4 \n\t"
273 "movq %%mm2, %%mm3 \n\t"
274 "movq %%mm4, %%mm5 \n\t"
275 "punpcklbw %%mm7, %%mm2 \n\t"
276 "punpcklbw %%mm7, %%mm4 \n\t"
277 "punpckhbw %%mm7, %%mm3 \n\t"
278 "punpckhbw %%mm7, %%mm5 \n\t"
279 "paddusw %%mm2, %%mm4 \n\t"
280 "paddusw %%mm3, %%mm5 \n\t"
281 "paddusw %%mm6, %%mm0 \n\t"
282 "paddusw %%mm6, %%mm1 \n\t"
283 "paddusw %%mm4, %%mm0 \n\t"
284 "paddusw %%mm5, %%mm1 \n\t"
285 "psrlw $2, %%mm0 \n\t"
286 "psrlw $2, %%mm1 \n\t"
287 "packuswb %%mm1, %%mm0 \n\t"
288 "movq %%mm0, (%2, %%eax) \n\t"
289 "addl %3, %%eax \n\t"
290
291 "subl $2, %0 \n\t"
292 "jnz 1b \n\t"
293 :"+g"(h), "+S"(pixels)
294 :"D"(block), "r"(line_size)
295 :"eax", "memory");
296 }
297
298 // avg_pixels
299 static void DEF(avg, pixels4)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
300 {
301 MOVQ_BFE(mm6);
302 JUMPALIGN();
303 do {
304 __asm __volatile(
305 "movd %0, %%mm0 \n\t"
306 "movd %1, %%mm1 \n\t"
307 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
308 "movd %%mm2, %0 \n\t"
309 :"+m"(*block)
310 :"m"(*pixels)
311 :"memory");
312 pixels += line_size;
313 block += line_size;
314 }
315 while (--h);
316 }
317
318 // in case more speed is needed - unroling would certainly help
319 static void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
320 {
321 MOVQ_BFE(mm6);
322 JUMPALIGN();
323 do {
324 __asm __volatile(
325 "movq %0, %%mm0 \n\t"
326 "movq %1, %%mm1 \n\t"
327 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
328 "movq %%mm2, %0 \n\t"
329 :"+m"(*block)
330 :"m"(*pixels)
331 :"memory");
332 pixels += line_size;
333 block += line_size;
334 }
335 while (--h);
336 }
337
338 static void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
339 {
340 MOVQ_BFE(mm6);
341 JUMPALIGN();
342 do {
343 __asm __volatile(
344 "movq %0, %%mm0 \n\t"
345 "movq %1, %%mm1 \n\t"
346 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
347 "movq %%mm2, %0 \n\t"
348 "movq 8%0, %%mm0 \n\t"
349 "movq 8%1, %%mm1 \n\t"
350 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
351 "movq %%mm2, 8%0 \n\t"
352 :"+m"(*block)
353 :"m"(*pixels)
354 :"memory");
355 pixels += line_size;
356 block += line_size;
357 }
358 while (--h);
359 }
360
361 static void DEF(avg, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
362 {
363 MOVQ_BFE(mm6);
364 JUMPALIGN();
365 do {
366 __asm __volatile(
367 "movq %1, %%mm0 \n\t"
368 "movq 1%1, %%mm1 \n\t"
369 "movq %0, %%mm3 \n\t"
370 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
371 PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
372 "movq %%mm0, %0 \n\t"
373 :"+m"(*block)
374 :"m"(*pixels)
375 :"memory");
376 pixels += line_size;
377 block += line_size;
378 } while (--h);
379 }
380
381 static __attribute__((unused)) void DEF(avg, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
382 {
383 MOVQ_BFE(mm6);
384 JUMPALIGN();
385 do {
386 __asm __volatile(
387 "movq %1, %%mm0 \n\t"
388 "movq %2, %%mm1 \n\t"
389 "movq %0, %%mm3 \n\t"
390 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
391 PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
392 "movq %%mm0, %0 \n\t"
393 :"+m"(*dst)
394 :"m"(*src1), "m"(*src2)
395 :"memory");
396 dst += dstStride;
397 src1 += src1Stride;
398 src2 += 8;
399 } while (--h);
400 }
401
402 static void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
403 {
404 MOVQ_BFE(mm6);
405 JUMPALIGN();
406 do {
407 __asm __volatile(
408 "movq %1, %%mm0 \n\t"
409 "movq 1%1, %%mm1 \n\t"
410 "movq %0, %%mm3 \n\t"
411 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
412 PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
413 "movq %%mm0, %0 \n\t"
414 "movq 8%1, %%mm0 \n\t"
415 "movq 9%1, %%mm1 \n\t"
416 "movq 8%0, %%mm3 \n\t"
417 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
418 PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
419 "movq %%mm0, 8%0 \n\t"
420 :"+m"(*block)
421 :"m"(*pixels)
422 :"memory");
423 pixels += line_size;
424 block += line_size;
425 } while (--h);
426 }
427
428 static __attribute__((unused)) void DEF(avg, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
429 {
430 MOVQ_BFE(mm6);
431 JUMPALIGN();
432 do {
433 __asm __volatile(
434 "movq %1, %%mm0 \n\t"
435 "movq %2, %%mm1 \n\t"
436 "movq %0, %%mm3 \n\t"
437 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
438 PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
439 "movq %%mm0, %0 \n\t"
440 "movq 8%1, %%mm0 \n\t"
441 "movq 8%2, %%mm1 \n\t"
442 "movq 8%0, %%mm3 \n\t"
443 PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
444 PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
445 "movq %%mm0, 8%0 \n\t"
446 :"+m"(*dst)
447 :"m"(*src1), "m"(*src2)
448 :"memory");
449 dst += dstStride;
450 src1 += src1Stride;
451 src2 += 16;
452 } while (--h);
453 }
454
455 static void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
456 {
457 MOVQ_BFE(mm6);
458 __asm __volatile(
459 "lea (%3, %3), %%eax \n\t"
460 "movq (%1), %%mm0 \n\t"
461 ".balign 8 \n\t"
462 "1: \n\t"
463 "movq (%1, %3), %%mm1 \n\t"
464 "movq (%1, %%eax), %%mm2 \n\t"
465 PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
466 "movq (%2), %%mm3 \n\t"
467 PAVGB(%%mm3, %%mm4, %%mm0, %%mm6)
468 "movq (%2, %3), %%mm3 \n\t"
469 PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)
470 "movq %%mm0, (%2) \n\t"
471 "movq %%mm1, (%2, %3) \n\t"
472 "addl %%eax, %1 \n\t"
473 "addl %%eax, %2 \n\t"
474
475 "movq (%1, %3), %%mm1 \n\t"
476 "movq (%1, %%eax), %%mm0 \n\t"
477 PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
478 "movq (%2), %%mm3 \n\t"
479 PAVGB(%%mm3, %%mm4, %%mm2, %%mm6)
480 "movq (%2, %3), %%mm3 \n\t"
481 PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)
482 "movq %%mm2, (%2) \n\t"
483 "movq %%mm1, (%2, %3) \n\t"
484 "addl %%eax, %1 \n\t"
485 "addl %%eax, %2 \n\t"
486
487 "subl $4, %0 \n\t"
488 "jnz 1b \n\t"
489 :"+g"(h), "+S"(pixels), "+D"(block)
490 :"r"(line_size)
491 :"eax", "memory");
492 }
493
494 // this routine is 'slightly' suboptimal but mostly unused
495 static void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
496 {
497 MOVQ_ZERO(mm7);
498 SET_RND(mm6); // =2 for rnd and =1 for no_rnd version
499 __asm __volatile(
500 "movq (%1), %%mm0 \n\t"
501 "movq 1(%1), %%mm4 \n\t"
502 "movq %%mm0, %%mm1 \n\t"
503 "movq %%mm4, %%mm5 \n\t"
504 "punpcklbw %%mm7, %%mm0 \n\t"
505 "punpcklbw %%mm7, %%mm4 \n\t"
506 "punpckhbw %%mm7, %%mm1 \n\t"
507 "punpckhbw %%mm7, %%mm5 \n\t"
508 "paddusw %%mm0, %%mm4 \n\t"
509 "paddusw %%mm1, %%mm5 \n\t"
510 "xorl %%eax, %%eax \n\t"
511 "addl %3, %1 \n\t"
512 ".balign 8 \n\t"
513 "1: \n\t"
514 "movq (%1, %%eax), %%mm0 \n\t"
515 "movq 1(%1, %%eax), %%mm2 \n\t"
516 "movq %%mm0, %%mm1 \n\t"
517 "movq %%mm2, %%mm3 \n\t"
518 "punpcklbw %%mm7, %%mm0 \n\t"
519 "punpcklbw %%mm7, %%mm2 \n\t"
520 "punpckhbw %%mm7, %%mm1 \n\t"
521 "punpckhbw %%mm7, %%mm3 \n\t"
522 "paddusw %%mm2, %%mm0 \n\t"
523 "paddusw %%mm3, %%mm1 \n\t"
524 "paddusw %%mm6, %%mm4 \n\t"
525 "paddusw %%mm6, %%mm5 \n\t"
526 "paddusw %%mm0, %%mm4 \n\t"
527 "paddusw %%mm1, %%mm5 \n\t"
528 "psrlw $2, %%mm4 \n\t"
529 "psrlw $2, %%mm5 \n\t"
530 "movq (%2, %%eax), %%mm3 \n\t"
531 "packuswb %%mm5, %%mm4 \n\t"
532 "pcmpeqd %%mm2, %%mm2 \n\t"
533 "paddb %%mm2, %%mm2 \n\t"
534 PAVGB(%%mm3, %%mm4, %%mm5, %%mm2)
535 "movq %%mm5, (%2, %%eax) \n\t"
536 "addl %3, %%eax \n\t"
537
538 "movq (%1, %%eax), %%mm2 \n\t" // 0 <-> 2 1 <-> 3
539 "movq 1(%1, %%eax), %%mm4 \n\t"
540 "movq %%mm2, %%mm3 \n\t"
541 "movq %%mm4, %%mm5 \n\t"
542 "punpcklbw %%mm7, %%mm2 \n\t"
543 "punpcklbw %%mm7, %%mm4 \n\t"
544 "punpckhbw %%mm7, %%mm3 \n\t"
545 "punpckhbw %%mm7, %%mm5 \n\t"
546 "paddusw %%mm2, %%mm4 \n\t"
547 "paddusw %%mm3, %%mm5 \n\t"
548 "paddusw %%mm6, %%mm0 \n\t"
549 "paddusw %%mm6, %%mm1 \n\t"
550 "paddusw %%mm4, %%mm0 \n\t"
551 "paddusw %%mm5, %%mm1 \n\t"
552 "psrlw $2, %%mm0 \n\t"
553 "psrlw $2, %%mm1 \n\t"
554 "movq (%2, %%eax), %%mm3 \n\t"
555 "packuswb %%mm1, %%mm0 \n\t"
556 "pcmpeqd %%mm2, %%mm2 \n\t"
557 "paddb %%mm2, %%mm2 \n\t"
558 PAVGB(%%mm3, %%mm0, %%mm1, %%mm2)
559 "movq %%mm1, (%2, %%eax) \n\t"
560 "addl %3, %%eax \n\t"
561
562 "subl $2, %0 \n\t"
563 "jnz 1b \n\t"
564 :"+g"(h), "+S"(pixels)
565 :"D"(block), "r"(line_size)
566 :"eax", "memory");
567 }
568
569 //FIXME optimize
570 static void DEF(put, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
571 DEF(put, pixels8_y2)(block , pixels , line_size, h);
572 DEF(put, pixels8_y2)(block+8, pixels+8, line_size, h);
573 }
574
575 static void DEF(put, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
576 DEF(put, pixels8_xy2)(block , pixels , line_size, h);
577 DEF(put, pixels8_xy2)(block+8, pixels+8, line_size, h);
578 }
579
580 static void DEF(avg, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
581 DEF(avg, pixels8_y2)(block , pixels , line_size, h);
582 DEF(avg, pixels8_y2)(block+8, pixels+8, line_size, h);
583 }
584
585 static void DEF(avg, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
586 DEF(avg, pixels8_xy2)(block , pixels , line_size, h);
587 DEF(avg, pixels8_xy2)(block+8, pixels+8, line_size, h);
588 }
589
590