spelling
[libav.git] / libavcodec / i386 / cavsdsp_mmx.c
1 /*
2 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
3 * Copyright (c) 2006 Stefan Gehrer <stefan.gehrer@gmx.de>
4 *
5 * MMX optimized DSP functions, based on H.264 optimizations by
6 * Michael Niedermayer and Loren Merritt
7 *
8 * This file is part of FFmpeg.
9 *
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
14 *
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25 #include "dsputil.h"
26 #include "dsputil_mmx.h"
27 #include "common.h"
28
29 /*****************************************************************************
30 *
31 * inverse transform
32 *
33 ****************************************************************************/
34
35 #define SUMSUB_BA( a, b ) \
36 "paddw "#b", "#a" \n\t"\
37 "paddw "#b", "#b" \n\t"\
38 "psubw "#a", "#b" \n\t"
39
40 #define SBUTTERFLY(a,b,t,n)\
41 "movq " #a ", " #t " \n\t" /* abcd */\
42 "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\
43 "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */
44
45 #define TRANSPOSE4(a,b,c,d,t)\
46 SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
47 SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
48 SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
49 SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
50
51 static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)
52 {
53 asm volatile(
54 "movq 112(%0), %%mm4 \n\t" /* mm4 = src7 */
55 "movq 16(%0), %%mm5 \n\t" /* mm5 = src1 */
56 "movq 80(%0), %%mm2 \n\t" /* mm2 = src5 */
57 "movq 48(%0), %%mm7 \n\t" /* mm7 = src3 */
58 "movq %%mm4, %%mm0 \n\t"
59 "movq %%mm5, %%mm3 \n\t"
60 "movq %%mm2, %%mm6 \n\t"
61 "movq %%mm7, %%mm1 \n\t"
62
63 "paddw %%mm4, %%mm4 \n\t" /* mm4 = 2*src7 */
64 "paddw %%mm3, %%mm3 \n\t" /* mm3 = 2*src1 */
65 "paddw %%mm6, %%mm6 \n\t" /* mm6 = 2*src5 */
66 "paddw %%mm1, %%mm1 \n\t" /* mm1 = 2*src3 */
67 "paddw %%mm4, %%mm0 \n\t" /* mm0 = 3*src7 */
68 "paddw %%mm3, %%mm5 \n\t" /* mm5 = 3*src1 */
69 "paddw %%mm6, %%mm2 \n\t" /* mm2 = 3*src5 */
70 "paddw %%mm1, %%mm7 \n\t" /* mm7 = 3*src3 */
71 "psubw %%mm4, %%mm5 \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */
72 "paddw %%mm6, %%mm7 \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */
73 "psubw %%mm2, %%mm1 \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */
74 "paddw %%mm0, %%mm3 \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */
75
76 "movq %%mm5, %%mm4 \n\t"
77 "movq %%mm7, %%mm6 \n\t"
78 "movq %%mm3, %%mm0 \n\t"
79 "movq %%mm1, %%mm2 \n\t"
80 SUMSUB_BA( %%mm7, %%mm5 ) /* mm7 = a0 + a1 mm5 = a0 - a1 */
81 "paddw %%mm3, %%mm7 \n\t" /* mm7 = a0 + a1 + a3 */
82 "paddw %%mm1, %%mm5 \n\t" /* mm5 = a0 - a1 + a2 */
83 "paddw %%mm7, %%mm7 \n\t"
84 "paddw %%mm5, %%mm5 \n\t"
85 "paddw %%mm6, %%mm7 \n\t" /* mm7 = b4 */
86 "paddw %%mm4, %%mm5 \n\t" /* mm5 = b5 */
87
88 SUMSUB_BA( %%mm1, %%mm3 ) /* mm1 = a3 + a2 mm3 = a3 - a2 */
89 "psubw %%mm1, %%mm4 \n\t" /* mm4 = a0 - a2 - a3 */
90 "movq %%mm4, %%mm1 \n\t" /* mm1 = a0 - a2 - a3 */
91 "psubw %%mm6, %%mm3 \n\t" /* mm3 = a3 - a2 - a1 */
92 "paddw %%mm1, %%mm1 \n\t"
93 "paddw %%mm3, %%mm3 \n\t"
94 "psubw %%mm2, %%mm1 \n\t" /* mm1 = b7 */
95 "paddw %%mm0, %%mm3 \n\t" /* mm3 = b6 */
96
97 "movq 32(%0), %%mm2 \n\t" /* mm2 = src2 */
98 "movq 96(%0), %%mm6 \n\t" /* mm6 = src6 */
99 "movq %%mm2, %%mm4 \n\t"
100 "movq %%mm6, %%mm0 \n\t"
101 "psllw $2, %%mm4 \n\t" /* mm4 = 4*src2 */
102 "psllw $2, %%mm6 \n\t" /* mm6 = 4*src6 */
103 "paddw %%mm4, %%mm2 \n\t" /* mm2 = 5*src2 */
104 "paddw %%mm6, %%mm0 \n\t" /* mm0 = 5*src6 */
105 "paddw %%mm2, %%mm2 \n\t"
106 "paddw %%mm0, %%mm0 \n\t"
107 "psubw %%mm0, %%mm4 \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */
108 "paddw %%mm2, %%mm6 \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */
109
110 "movq (%0), %%mm2 \n\t" /* mm2 = src0 */
111 "movq 64(%0), %%mm0 \n\t" /* mm0 = src4 */
112 SUMSUB_BA( %%mm0, %%mm2 ) /* mm0 = src0+src4 mm2 = src0-src4 */
113 "psllw $3, %%mm0 \n\t"
114 "psllw $3, %%mm2 \n\t"
115 "paddw %1, %%mm0 \n\t" /* add rounding bias */
116 "paddw %1, %%mm2 \n\t" /* add rounding bias */
117
118 SUMSUB_BA( %%mm6, %%mm0 ) /* mm6 = a4 + a6 mm0 = a4 - a6 */
119 SUMSUB_BA( %%mm4, %%mm2 ) /* mm4 = a5 + a7 mm2 = a5 - a7 */
120 SUMSUB_BA( %%mm7, %%mm6 ) /* mm7 = dst0 mm6 = dst7 */
121 SUMSUB_BA( %%mm5, %%mm4 ) /* mm5 = dst1 mm4 = dst6 */
122 SUMSUB_BA( %%mm3, %%mm2 ) /* mm3 = dst2 mm2 = dst5 */
123 SUMSUB_BA( %%mm1, %%mm0 ) /* mm1 = dst3 mm0 = dst4 */
124 :: "r"(block), "m"(bias)
125 );
126 }
127
128 static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
129 {
130 int i;
131 DECLARE_ALIGNED_8(int16_t, b2[64]);
132
133 for(i=0; i<2; i++){
134 DECLARE_ALIGNED_8(uint64_t, tmp);
135
136 cavs_idct8_1d(block+4*i, ff_pw_4);
137
138 asm volatile(
139 "psraw $3, %%mm7 \n\t"
140 "psraw $3, %%mm6 \n\t"
141 "psraw $3, %%mm5 \n\t"
142 "psraw $3, %%mm4 \n\t"
143 "psraw $3, %%mm3 \n\t"
144 "psraw $3, %%mm2 \n\t"
145 "psraw $3, %%mm1 \n\t"
146 "psraw $3, %%mm0 \n\t"
147 "movq %%mm7, %0 \n\t"
148 TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
149 "movq %%mm0, 8(%1) \n\t"
150 "movq %%mm6, 24(%1) \n\t"
151 "movq %%mm7, 40(%1) \n\t"
152 "movq %%mm4, 56(%1) \n\t"
153 "movq %0, %%mm7 \n\t"
154 TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
155 "movq %%mm7, (%1) \n\t"
156 "movq %%mm1, 16(%1) \n\t"
157 "movq %%mm0, 32(%1) \n\t"
158 "movq %%mm3, 48(%1) \n\t"
159 : "=m"(tmp)
160 : "r"(b2+32*i)
161 : "memory"
162 );
163 }
164
165 for(i=0; i<2; i++){
166 cavs_idct8_1d(b2+4*i, ff_pw_64);
167
168 asm volatile(
169 "psraw $7, %%mm7 \n\t"
170 "psraw $7, %%mm6 \n\t"
171 "psraw $7, %%mm5 \n\t"
172 "psraw $7, %%mm4 \n\t"
173 "psraw $7, %%mm3 \n\t"
174 "psraw $7, %%mm2 \n\t"
175 "psraw $7, %%mm1 \n\t"
176 "psraw $7, %%mm0 \n\t"
177 "movq %%mm7, (%0) \n\t"
178 "movq %%mm5, 16(%0) \n\t"
179 "movq %%mm3, 32(%0) \n\t"
180 "movq %%mm1, 48(%0) \n\t"
181 "movq %%mm0, 64(%0) \n\t"
182 "movq %%mm2, 80(%0) \n\t"
183 "movq %%mm4, 96(%0) \n\t"
184 "movq %%mm6, 112(%0) \n\t"
185 :: "r"(b2+4*i)
186 : "memory"
187 );
188 }
189
190 add_pixels_clamped_mmx(b2, dst, stride);
191
192 /* clear block */
193 asm volatile(
194 "pxor %%mm7, %%mm7 \n\t"
195 "movq %%mm7, (%0) \n\t"
196 "movq %%mm7, 8(%0) \n\t"
197 "movq %%mm7, 16(%0) \n\t"
198 "movq %%mm7, 24(%0) \n\t"
199 "movq %%mm7, 32(%0) \n\t"
200 "movq %%mm7, 40(%0) \n\t"
201 "movq %%mm7, 48(%0) \n\t"
202 "movq %%mm7, 56(%0) \n\t"
203 "movq %%mm7, 64(%0) \n\t"
204 "movq %%mm7, 72(%0) \n\t"
205 "movq %%mm7, 80(%0) \n\t"
206 "movq %%mm7, 88(%0) \n\t"
207 "movq %%mm7, 96(%0) \n\t"
208 "movq %%mm7, 104(%0) \n\t"
209 "movq %%mm7, 112(%0) \n\t"
210 "movq %%mm7, 120(%0) \n\t"
211 :: "r" (block)
212 );
213 }
214
215 /*****************************************************************************
216 *
217 * motion compensation
218 *
219 ****************************************************************************/
220
221 /* vertical filter [-1 -2 96 42 -7 0] */
222 #define QPEL_CAVSV1(A,B,C,D,E,F,OP) \
223 "movd (%0), "#F" \n\t"\
224 "movq "#C", %%mm6 \n\t"\
225 "pmullw %5, %%mm6 \n\t"\
226 "movq "#D", %%mm7 \n\t"\
227 "pmullw %6, %%mm7 \n\t"\
228 "psllw $3, "#E" \n\t"\
229 "psubw "#E", %%mm6 \n\t"\
230 "psraw $3, "#E" \n\t"\
231 "paddw %%mm7, %%mm6 \n\t"\
232 "paddw "#E", %%mm6 \n\t"\
233 "paddw "#B", "#B" \n\t"\
234 "pxor %%mm7, %%mm7 \n\t"\
235 "add %2, %0 \n\t"\
236 "punpcklbw %%mm7, "#F" \n\t"\
237 "psubw "#B", %%mm6 \n\t"\
238 "psraw $1, "#B" \n\t"\
239 "psubw "#A", %%mm6 \n\t"\
240 "paddw %4, %%mm6 \n\t"\
241 "psraw $7, %%mm6 \n\t"\
242 "packuswb %%mm6, %%mm6 \n\t"\
243 OP(%%mm6, (%1), A, d) \
244 "add %3, %1 \n\t"
245
246 /* vertical filter [ 0 -1 5 5 -1 0] */
247 #define QPEL_CAVSV2(A,B,C,D,E,F,OP) \
248 "movd (%0), "#F" \n\t"\
249 "movq "#C", %%mm6 \n\t"\
250 "paddw "#D", %%mm6 \n\t"\
251 "pmullw %5, %%mm6 \n\t"\
252 "add %2, %0 \n\t"\
253 "punpcklbw %%mm7, "#F" \n\t"\
254 "psubw "#B", %%mm6 \n\t"\
255 "psubw "#E", %%mm6 \n\t"\
256 "paddw %4, %%mm6 \n\t"\
257 "psraw $3, %%mm6 \n\t"\
258 "packuswb %%mm6, %%mm6 \n\t"\
259 OP(%%mm6, (%1), A, d) \
260 "add %3, %1 \n\t"
261
262 /* vertical filter [ 0 -7 42 96 -2 -1] */
263 #define QPEL_CAVSV3(A,B,C,D,E,F,OP) \
264 "movd (%0), "#F" \n\t"\
265 "movq "#C", %%mm6 \n\t"\
266 "pmullw %6, %%mm6 \n\t"\
267 "movq "#D", %%mm7 \n\t"\
268 "pmullw %5, %%mm7 \n\t"\
269 "psllw $3, "#B" \n\t"\
270 "psubw "#B", %%mm6 \n\t"\
271 "psraw $3, "#B" \n\t"\
272 "paddw %%mm7, %%mm6 \n\t"\
273 "paddw "#B", %%mm6 \n\t"\
274 "paddw "#E", "#E" \n\t"\
275 "pxor %%mm7, %%mm7 \n\t"\
276 "add %2, %0 \n\t"\
277 "punpcklbw %%mm7, "#F" \n\t"\
278 "psubw "#E", %%mm6 \n\t"\
279 "psraw $1, "#E" \n\t"\
280 "psubw "#F", %%mm6 \n\t"\
281 "paddw %4, %%mm6 \n\t"\
282 "psraw $7, %%mm6 \n\t"\
283 "packuswb %%mm6, %%mm6 \n\t"\
284 OP(%%mm6, (%1), A, d) \
285 "add %3, %1 \n\t"
286
287
288 #define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
289 int w= 2;\
290 src -= 2*srcStride;\
291 \
292 while(w--){\
293 asm volatile(\
294 "pxor %%mm7, %%mm7 \n\t"\
295 "movd (%0), %%mm0 \n\t"\
296 "add %2, %0 \n\t"\
297 "movd (%0), %%mm1 \n\t"\
298 "add %2, %0 \n\t"\
299 "movd (%0), %%mm2 \n\t"\
300 "add %2, %0 \n\t"\
301 "movd (%0), %%mm3 \n\t"\
302 "add %2, %0 \n\t"\
303 "movd (%0), %%mm4 \n\t"\
304 "add %2, %0 \n\t"\
305 "punpcklbw %%mm7, %%mm0 \n\t"\
306 "punpcklbw %%mm7, %%mm1 \n\t"\
307 "punpcklbw %%mm7, %%mm2 \n\t"\
308 "punpcklbw %%mm7, %%mm3 \n\t"\
309 "punpcklbw %%mm7, %%mm4 \n\t"\
310 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
311 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
312 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
313 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
314 VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
315 VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
316 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
317 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
318 \
319 : "+a"(src), "+c"(dst)\
320 : "S"((long)srcStride), "D"((long)dstStride), "m"(ADD), "m"(MUL1), "m"(MUL2)\
321 : "memory"\
322 );\
323 if(h==16){\
324 asm volatile(\
325 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
326 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
327 VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
328 VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
329 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
330 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
331 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
332 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
333 \
334 : "+a"(src), "+c"(dst)\
335 : "S"((long)srcStride), "D"((long)dstStride), "m"(ADD), "m"(MUL1), "m"(MUL2)\
336 : "memory"\
337 );\
338 }\
339 src += 4-(h+5)*srcStride;\
340 dst += 4-h*dstStride;\
341 }
342
343 #define QPEL_CAVS(OPNAME, OP, MMX)\
344 static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
345 int h=8;\
346 asm volatile(\
347 "pxor %%mm7, %%mm7 \n\t"\
348 "movq %5, %%mm6 \n\t"\
349 "1: \n\t"\
350 "movq (%0), %%mm0 \n\t"\
351 "movq 1(%0), %%mm2 \n\t"\
352 "movq %%mm0, %%mm1 \n\t"\
353 "movq %%mm2, %%mm3 \n\t"\
354 "punpcklbw %%mm7, %%mm0 \n\t"\
355 "punpckhbw %%mm7, %%mm1 \n\t"\
356 "punpcklbw %%mm7, %%mm2 \n\t"\
357 "punpckhbw %%mm7, %%mm3 \n\t"\
358 "paddw %%mm2, %%mm0 \n\t"\
359 "paddw %%mm3, %%mm1 \n\t"\
360 "pmullw %%mm6, %%mm0 \n\t"\
361 "pmullw %%mm6, %%mm1 \n\t"\
362 "movq -1(%0), %%mm2 \n\t"\
363 "movq 2(%0), %%mm4 \n\t"\
364 "movq %%mm2, %%mm3 \n\t"\
365 "movq %%mm4, %%mm5 \n\t"\
366 "punpcklbw %%mm7, %%mm2 \n\t"\
367 "punpckhbw %%mm7, %%mm3 \n\t"\
368 "punpcklbw %%mm7, %%mm4 \n\t"\
369 "punpckhbw %%mm7, %%mm5 \n\t"\
370 "paddw %%mm4, %%mm2 \n\t"\
371 "paddw %%mm3, %%mm5 \n\t"\
372 "psubw %%mm2, %%mm0 \n\t"\
373 "psubw %%mm5, %%mm1 \n\t"\
374 "movq %6, %%mm5 \n\t"\
375 "paddw %%mm5, %%mm0 \n\t"\
376 "paddw %%mm5, %%mm1 \n\t"\
377 "psraw $3, %%mm0 \n\t"\
378 "psraw $3, %%mm1 \n\t"\
379 "packuswb %%mm1, %%mm0 \n\t"\
380 OP(%%mm0, (%1),%%mm5, q) \
381 "add %3, %0 \n\t"\
382 "add %4, %1 \n\t"\
383 "decl %2 \n\t"\
384 " jnz 1b \n\t"\
385 : "+a"(src), "+c"(dst), "+m"(h)\
386 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\
387 : "memory"\
388 );\
389 }\
390 \
391 static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
392 QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42) \
393 }\
394 \
395 static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
396 QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5) \
397 }\
398 \
399 static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
400 QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42) \
401 }\
402 \
403 static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
404 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 8);\
405 }\
406 static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
407 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 16);\
408 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
409 }\
410 \
411 static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
412 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 8);\
413 }\
414 static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
415 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 16);\
416 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
417 }\
418 \
419 static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
420 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 8);\
421 }\
422 static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
423 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 16);\
424 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
425 }\
426 \
427 static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
428 OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\
429 OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
430 src += 8*srcStride;\
431 dst += 8*dstStride;\
432 OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\
433 OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
434 }\
435
436 #define CAVS_MC(OPNAME, SIZE, MMX) \
437 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
438 OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
439 }\
440 \
441 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
442 OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
443 }\
444 \
445 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
446 OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
447 }\
448 \
449 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
450 OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
451 }\
452
453 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
454 #define AVG_3DNOW_OP(a,b,temp, size) \
455 "mov" #size " " #b ", " #temp " \n\t"\
456 "pavgusb " #temp ", " #a " \n\t"\
457 "mov" #size " " #a ", " #b " \n\t"
458 #define AVG_MMX2_OP(a,b,temp, size) \
459 "mov" #size " " #b ", " #temp " \n\t"\
460 "pavgb " #temp ", " #a " \n\t"\
461 "mov" #size " " #a ", " #b " \n\t"
462
463 QPEL_CAVS(put_, PUT_OP, 3dnow)
464 QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
465 QPEL_CAVS(put_, PUT_OP, mmx2)
466 QPEL_CAVS(avg_, AVG_MMX2_OP, mmx2)
467
468 CAVS_MC(put_, 8, 3dnow)
469 CAVS_MC(put_, 16,3dnow)
470 CAVS_MC(avg_, 8, 3dnow)
471 CAVS_MC(avg_, 16,3dnow)
472 CAVS_MC(put_, 8, mmx2)
473 CAVS_MC(put_, 16,mmx2)
474 CAVS_MC(avg_, 8, mmx2)
475 CAVS_MC(avg_, 16,mmx2)
476
477 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
478 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
479 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
480 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
481
482 void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx) {
483 #define dspfunc(PFX, IDX, NUM) \
484 c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
485 c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \
486 c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \
487 c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \
488 c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \
489
490 dspfunc(put_cavs_qpel, 0, 16);
491 dspfunc(put_cavs_qpel, 1, 8);
492 dspfunc(avg_cavs_qpel, 0, 16);
493 dspfunc(avg_cavs_qpel, 1, 8);
494 #undef dspfunc
495 c->cavs_idct8_add = cavs_idct8_add_mmx;
496 }
497
498 void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx) {
499 #define dspfunc(PFX, IDX, NUM) \
500 c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
501 c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
502 c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
503 c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
504 c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \
505
506 dspfunc(put_cavs_qpel, 0, 16);
507 dspfunc(put_cavs_qpel, 1, 8);
508 dspfunc(avg_cavs_qpel, 0, 16);
509 dspfunc(avg_cavs_qpel, 1, 8);
510 #undef dspfunc
511 c->cavs_idct8_add = cavs_idct8_add_mmx;
512 }