e47bd5aff2c826911d69130dfd20afce7ea06a58
[libav.git] / libavcodec / vc1dsp.c
1 /*
2 * VC-1 and WMV3 decoder - DSP functions
3 * Copyright (c) 2006 Konstantin Shishkov
4 *
5 * This file is part of Libav.
6 *
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * Libav 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 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * VC-1 and WMV3 decoder
25 *
26 */
27
28 #include "libavutil/common.h"
29 #include "h264chroma.h"
30 #include "vc1dsp.h"
31
32 /* Apply overlap transform to horizontal edge */
33 static void vc1_v_overlap_c(uint8_t *src, int stride)
34 {
35 int i;
36 int a, b, c, d;
37 int d1, d2;
38 int rnd = 1;
39 for (i = 0; i < 8; i++) {
40 a = src[-2 * stride];
41 b = src[-stride];
42 c = src[0];
43 d = src[stride];
44 d1 = (a - d + 3 + rnd) >> 3;
45 d2 = (a - d + b - c + 4 - rnd) >> 3;
46
47 src[-2 * stride] = a - d1;
48 src[-stride] = av_clip_uint8(b - d2);
49 src[0] = av_clip_uint8(c + d2);
50 src[stride] = d + d1;
51 src++;
52 rnd = !rnd;
53 }
54 }
55
56 /* Apply overlap transform to vertical edge */
57 static void vc1_h_overlap_c(uint8_t *src, int stride)
58 {
59 int i;
60 int a, b, c, d;
61 int d1, d2;
62 int rnd = 1;
63 for (i = 0; i < 8; i++) {
64 a = src[-2];
65 b = src[-1];
66 c = src[0];
67 d = src[1];
68 d1 = (a - d + 3 + rnd) >> 3;
69 d2 = (a - d + b - c + 4 - rnd) >> 3;
70
71 src[-2] = a - d1;
72 src[-1] = av_clip_uint8(b - d2);
73 src[0] = av_clip_uint8(c + d2);
74 src[1] = d + d1;
75 src += stride;
76 rnd = !rnd;
77 }
78 }
79
80 static void vc1_v_s_overlap_c(int16_t *top, int16_t *bottom)
81 {
82 int i;
83 int a, b, c, d;
84 int d1, d2;
85 int rnd1 = 4, rnd2 = 3;
86 for (i = 0; i < 8; i++) {
87 a = top[48];
88 b = top[56];
89 c = bottom[0];
90 d = bottom[8];
91 d1 = a - d;
92 d2 = a - d + b - c;
93
94 top[48] = ((a << 3) - d1 + rnd1) >> 3;
95 top[56] = ((b << 3) - d2 + rnd2) >> 3;
96 bottom[0] = ((c << 3) + d2 + rnd1) >> 3;
97 bottom[8] = ((d << 3) + d1 + rnd2) >> 3;
98
99 bottom++;
100 top++;
101 rnd2 = 7 - rnd2;
102 rnd1 = 7 - rnd1;
103 }
104 }
105
106 static void vc1_h_s_overlap_c(int16_t *left, int16_t *right)
107 {
108 int i;
109 int a, b, c, d;
110 int d1, d2;
111 int rnd1 = 4, rnd2 = 3;
112 for (i = 0; i < 8; i++) {
113 a = left[6];
114 b = left[7];
115 c = right[0];
116 d = right[1];
117 d1 = a - d;
118 d2 = a - d + b - c;
119
120 left[6] = ((a << 3) - d1 + rnd1) >> 3;
121 left[7] = ((b << 3) - d2 + rnd2) >> 3;
122 right[0] = ((c << 3) + d2 + rnd1) >> 3;
123 right[1] = ((d << 3) + d1 + rnd2) >> 3;
124
125 right += 8;
126 left += 8;
127 rnd2 = 7 - rnd2;
128 rnd1 = 7 - rnd1;
129 }
130 }
131
132 /**
133 * VC-1 in-loop deblocking filter for one line
134 * @param src source block type
135 * @param stride block stride
136 * @param pq block quantizer
137 * @return whether other 3 pairs should be filtered or not
138 * @see 8.6
139 */
140 static av_always_inline int vc1_filter_line(uint8_t *src, int stride, int pq)
141 {
142 int a0 = (2 * (src[-2 * stride] - src[1 * stride]) -
143 5 * (src[-1 * stride] - src[0 * stride]) + 4) >> 3;
144 int a0_sign = a0 >> 31; /* Store sign */
145
146 a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */
147 if (a0 < pq) {
148 int a1 = FFABS((2 * (src[-4 * stride] - src[-1 * stride]) -
149 5 * (src[-3 * stride] - src[-2 * stride]) + 4) >> 3);
150 int a2 = FFABS((2 * (src[ 0 * stride] - src[ 3 * stride]) -
151 5 * (src[ 1 * stride] - src[ 2 * stride]) + 4) >> 3);
152 if (a1 < a0 || a2 < a0) {
153 int clip = src[-1 * stride] - src[0 * stride];
154 int clip_sign = clip >> 31;
155
156 clip = ((clip ^ clip_sign) - clip_sign) >> 1;
157 if (clip) {
158 int a3 = FFMIN(a1, a2);
159 int d = 5 * (a3 - a0);
160 int d_sign = (d >> 31);
161
162 d = ((d ^ d_sign) - d_sign) >> 3;
163 d_sign ^= a0_sign;
164
165 if (d_sign ^ clip_sign)
166 d = 0;
167 else {
168 d = FFMIN(d, clip);
169 d = (d ^ d_sign) - d_sign; /* Restore sign */
170 src[-1 * stride] = av_clip_uint8(src[-1 * stride] - d);
171 src[ 0 * stride] = av_clip_uint8(src[ 0 * stride] + d);
172 }
173 return 1;
174 }
175 }
176 }
177 return 0;
178 }
179
180 /**
181 * VC-1 in-loop deblocking filter
182 * @param src source block type
183 * @param step distance between horizontally adjacent elements
184 * @param stride distance between vertically adjacent elements
185 * @param len edge length to filter (4 or 8 pixels)
186 * @param pq block quantizer
187 * @see 8.6
188 */
189 static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
190 int len, int pq)
191 {
192 int i;
193 int filt3;
194
195 for (i = 0; i < len; i += 4) {
196 filt3 = vc1_filter_line(src + 2 * step, stride, pq);
197 if (filt3) {
198 vc1_filter_line(src + 0 * step, stride, pq);
199 vc1_filter_line(src + 1 * step, stride, pq);
200 vc1_filter_line(src + 3 * step, stride, pq);
201 }
202 src += step * 4;
203 }
204 }
205
206 static void vc1_v_loop_filter4_c(uint8_t *src, int stride, int pq)
207 {
208 vc1_loop_filter(src, 1, stride, 4, pq);
209 }
210
211 static void vc1_h_loop_filter4_c(uint8_t *src, int stride, int pq)
212 {
213 vc1_loop_filter(src, stride, 1, 4, pq);
214 }
215
216 static void vc1_v_loop_filter8_c(uint8_t *src, int stride, int pq)
217 {
218 vc1_loop_filter(src, 1, stride, 8, pq);
219 }
220
221 static void vc1_h_loop_filter8_c(uint8_t *src, int stride, int pq)
222 {
223 vc1_loop_filter(src, stride, 1, 8, pq);
224 }
225
226 static void vc1_v_loop_filter16_c(uint8_t *src, int stride, int pq)
227 {
228 vc1_loop_filter(src, 1, stride, 16, pq);
229 }
230
231 static void vc1_h_loop_filter16_c(uint8_t *src, int stride, int pq)
232 {
233 vc1_loop_filter(src, stride, 1, 16, pq);
234 }
235
236 /* Do inverse transform on 8x8 block */
237 static void vc1_inv_trans_8x8_dc_c(uint8_t *dest, int linesize, int16_t *block)
238 {
239 int i;
240 int dc = block[0];
241
242 dc = (3 * dc + 1) >> 1;
243 dc = (3 * dc + 16) >> 5;
244
245 for (i = 0; i < 8; i++) {
246 dest[0] = av_clip_uint8(dest[0] + dc);
247 dest[1] = av_clip_uint8(dest[1] + dc);
248 dest[2] = av_clip_uint8(dest[2] + dc);
249 dest[3] = av_clip_uint8(dest[3] + dc);
250 dest[4] = av_clip_uint8(dest[4] + dc);
251 dest[5] = av_clip_uint8(dest[5] + dc);
252 dest[6] = av_clip_uint8(dest[6] + dc);
253 dest[7] = av_clip_uint8(dest[7] + dc);
254 dest += linesize;
255 }
256 }
257
258 static void vc1_inv_trans_8x8_c(int16_t block[64])
259 {
260 int i;
261 register int t1, t2, t3, t4, t5, t6, t7, t8;
262 int16_t *src, *dst, temp[64];
263
264 src = block;
265 dst = temp;
266 for (i = 0; i < 8; i++) {
267 t1 = 12 * (src[ 0] + src[32]) + 4;
268 t2 = 12 * (src[ 0] - src[32]) + 4;
269 t3 = 16 * src[16] + 6 * src[48];
270 t4 = 6 * src[16] - 16 * src[48];
271
272 t5 = t1 + t3;
273 t6 = t2 + t4;
274 t7 = t2 - t4;
275 t8 = t1 - t3;
276
277 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
278 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
279 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
280 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
281
282 dst[0] = (t5 + t1) >> 3;
283 dst[1] = (t6 + t2) >> 3;
284 dst[2] = (t7 + t3) >> 3;
285 dst[3] = (t8 + t4) >> 3;
286 dst[4] = (t8 - t4) >> 3;
287 dst[5] = (t7 - t3) >> 3;
288 dst[6] = (t6 - t2) >> 3;
289 dst[7] = (t5 - t1) >> 3;
290
291 src += 1;
292 dst += 8;
293 }
294
295 src = temp;
296 dst = block;
297 for (i = 0; i < 8; i++) {
298 t1 = 12 * (src[ 0] + src[32]) + 64;
299 t2 = 12 * (src[ 0] - src[32]) + 64;
300 t3 = 16 * src[16] + 6 * src[48];
301 t4 = 6 * src[16] - 16 * src[48];
302
303 t5 = t1 + t3;
304 t6 = t2 + t4;
305 t7 = t2 - t4;
306 t8 = t1 - t3;
307
308 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
309 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
310 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
311 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
312
313 dst[ 0] = (t5 + t1) >> 7;
314 dst[ 8] = (t6 + t2) >> 7;
315 dst[16] = (t7 + t3) >> 7;
316 dst[24] = (t8 + t4) >> 7;
317 dst[32] = (t8 - t4 + 1) >> 7;
318 dst[40] = (t7 - t3 + 1) >> 7;
319 dst[48] = (t6 - t2 + 1) >> 7;
320 dst[56] = (t5 - t1 + 1) >> 7;
321
322 src++;
323 dst++;
324 }
325 }
326
327 /* Do inverse transform on 8x4 part of block */
328 static void vc1_inv_trans_8x4_dc_c(uint8_t *dest, int linesize, int16_t *block)
329 {
330 int i;
331 int dc = block[0];
332
333 dc = (3 * dc + 1) >> 1;
334 dc = (17 * dc + 64) >> 7;
335
336 for (i = 0; i < 4; i++) {
337 dest[0] = av_clip_uint8(dest[0] + dc);
338 dest[1] = av_clip_uint8(dest[1] + dc);
339 dest[2] = av_clip_uint8(dest[2] + dc);
340 dest[3] = av_clip_uint8(dest[3] + dc);
341 dest[4] = av_clip_uint8(dest[4] + dc);
342 dest[5] = av_clip_uint8(dest[5] + dc);
343 dest[6] = av_clip_uint8(dest[6] + dc);
344 dest[7] = av_clip_uint8(dest[7] + dc);
345 dest += linesize;
346 }
347 }
348
349 static void vc1_inv_trans_8x4_c(uint8_t *dest, int linesize, int16_t *block)
350 {
351 int i;
352 register int t1, t2, t3, t4, t5, t6, t7, t8;
353 int16_t *src, *dst;
354
355 src = block;
356 dst = block;
357
358 for (i = 0; i < 4; i++) {
359 t1 = 12 * (src[0] + src[4]) + 4;
360 t2 = 12 * (src[0] - src[4]) + 4;
361 t3 = 16 * src[2] + 6 * src[6];
362 t4 = 6 * src[2] - 16 * src[6];
363
364 t5 = t1 + t3;
365 t6 = t2 + t4;
366 t7 = t2 - t4;
367 t8 = t1 - t3;
368
369 t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7];
370 t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7];
371 t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7];
372 t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7];
373
374 dst[0] = (t5 + t1) >> 3;
375 dst[1] = (t6 + t2) >> 3;
376 dst[2] = (t7 + t3) >> 3;
377 dst[3] = (t8 + t4) >> 3;
378 dst[4] = (t8 - t4) >> 3;
379 dst[5] = (t7 - t3) >> 3;
380 dst[6] = (t6 - t2) >> 3;
381 dst[7] = (t5 - t1) >> 3;
382
383 src += 8;
384 dst += 8;
385 }
386
387 src = block;
388 for (i = 0; i < 8; i++) {
389 t1 = 17 * (src[ 0] + src[16]) + 64;
390 t2 = 17 * (src[ 0] - src[16]) + 64;
391 t3 = 22 * src[ 8] + 10 * src[24];
392 t4 = 22 * src[24] - 10 * src[ 8];
393
394 dest[0 * linesize] = av_clip_uint8(dest[0 * linesize] + ((t1 + t3) >> 7));
395 dest[1 * linesize] = av_clip_uint8(dest[1 * linesize] + ((t2 - t4) >> 7));
396 dest[2 * linesize] = av_clip_uint8(dest[2 * linesize] + ((t2 + t4) >> 7));
397 dest[3 * linesize] = av_clip_uint8(dest[3 * linesize] + ((t1 - t3) >> 7));
398
399 src++;
400 dest++;
401 }
402 }
403
404 /* Do inverse transform on 4x8 parts of block */
405 static void vc1_inv_trans_4x8_dc_c(uint8_t *dest, int linesize, int16_t *block)
406 {
407 int i;
408 int dc = block[0];
409
410 dc = (17 * dc + 4) >> 3;
411 dc = (12 * dc + 64) >> 7;
412
413 for (i = 0; i < 8; i++) {
414 dest[0] = av_clip_uint8(dest[0] + dc);
415 dest[1] = av_clip_uint8(dest[1] + dc);
416 dest[2] = av_clip_uint8(dest[2] + dc);
417 dest[3] = av_clip_uint8(dest[3] + dc);
418 dest += linesize;
419 }
420 }
421
422 static void vc1_inv_trans_4x8_c(uint8_t *dest, int linesize, int16_t *block)
423 {
424 int i;
425 register int t1, t2, t3, t4, t5, t6, t7, t8;
426 int16_t *src, *dst;
427
428 src = block;
429 dst = block;
430
431 for (i = 0; i < 8; i++) {
432 t1 = 17 * (src[0] + src[2]) + 4;
433 t2 = 17 * (src[0] - src[2]) + 4;
434 t3 = 22 * src[1] + 10 * src[3];
435 t4 = 22 * src[3] - 10 * src[1];
436
437 dst[0] = (t1 + t3) >> 3;
438 dst[1] = (t2 - t4) >> 3;
439 dst[2] = (t2 + t4) >> 3;
440 dst[3] = (t1 - t3) >> 3;
441
442 src += 8;
443 dst += 8;
444 }
445
446 src = block;
447 for (i = 0; i < 4; i++) {
448 t1 = 12 * (src[ 0] + src[32]) + 64;
449 t2 = 12 * (src[ 0] - src[32]) + 64;
450 t3 = 16 * src[16] + 6 * src[48];
451 t4 = 6 * src[16] - 16 * src[48];
452
453 t5 = t1 + t3;
454 t6 = t2 + t4;
455 t7 = t2 - t4;
456 t8 = t1 - t3;
457
458 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
459 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
460 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
461 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
462
463 dest[0 * linesize] = av_clip_uint8(dest[0 * linesize] + ((t5 + t1) >> 7));
464 dest[1 * linesize] = av_clip_uint8(dest[1 * linesize] + ((t6 + t2) >> 7));
465 dest[2 * linesize] = av_clip_uint8(dest[2 * linesize] + ((t7 + t3) >> 7));
466 dest[3 * linesize] = av_clip_uint8(dest[3 * linesize] + ((t8 + t4) >> 7));
467 dest[4 * linesize] = av_clip_uint8(dest[4 * linesize] + ((t8 - t4 + 1) >> 7));
468 dest[5 * linesize] = av_clip_uint8(dest[5 * linesize] + ((t7 - t3 + 1) >> 7));
469 dest[6 * linesize] = av_clip_uint8(dest[6 * linesize] + ((t6 - t2 + 1) >> 7));
470 dest[7 * linesize] = av_clip_uint8(dest[7 * linesize] + ((t5 - t1 + 1) >> 7));
471
472 src++;
473 dest++;
474 }
475 }
476
477 /* Do inverse transform on 4x4 part of block */
478 static void vc1_inv_trans_4x4_dc_c(uint8_t *dest, int linesize, int16_t *block)
479 {
480 int i;
481 int dc = block[0];
482
483 dc = (17 * dc + 4) >> 3;
484 dc = (17 * dc + 64) >> 7;
485
486 for (i = 0; i < 4; i++) {
487 dest[0] = av_clip_uint8(dest[0] + dc);
488 dest[1] = av_clip_uint8(dest[1] + dc);
489 dest[2] = av_clip_uint8(dest[2] + dc);
490 dest[3] = av_clip_uint8(dest[3] + dc);
491 dest += linesize;
492 }
493 }
494
495 static void vc1_inv_trans_4x4_c(uint8_t *dest, int linesize, int16_t *block)
496 {
497 int i;
498 register int t1, t2, t3, t4;
499 int16_t *src, *dst;
500
501 src = block;
502 dst = block;
503 for (i = 0; i < 4; i++) {
504 t1 = 17 * (src[0] + src[2]) + 4;
505 t2 = 17 * (src[0] - src[2]) + 4;
506 t3 = 22 * src[1] + 10 * src[3];
507 t4 = 22 * src[3] - 10 * src[1];
508
509 dst[0] = (t1 + t3) >> 3;
510 dst[1] = (t2 - t4) >> 3;
511 dst[2] = (t2 + t4) >> 3;
512 dst[3] = (t1 - t3) >> 3;
513
514 src += 8;
515 dst += 8;
516 }
517
518 src = block;
519 for (i = 0; i < 4; i++) {
520 t1 = 17 * (src[0] + src[16]) + 64;
521 t2 = 17 * (src[0] - src[16]) + 64;
522 t3 = 22 * src[8] + 10 * src[24];
523 t4 = 22 * src[24] - 10 * src[8];
524
525 dest[0 * linesize] = av_clip_uint8(dest[0 * linesize] + ((t1 + t3) >> 7));
526 dest[1 * linesize] = av_clip_uint8(dest[1 * linesize] + ((t2 - t4) >> 7));
527 dest[2 * linesize] = av_clip_uint8(dest[2 * linesize] + ((t2 + t4) >> 7));
528 dest[3 * linesize] = av_clip_uint8(dest[3 * linesize] + ((t1 - t3) >> 7));
529
530 src++;
531 dest++;
532 }
533 }
534
535 /* motion compensation functions */
536
537 /* Filter in case of 2 filters */
538 #define VC1_MSPEL_FILTER_16B(DIR, TYPE) \
539 static av_always_inline int vc1_mspel_ ## DIR ## _filter_16bits(const TYPE *src, \
540 int stride, \
541 int mode) \
542 { \
543 switch(mode) { \
544 case 0: /* no shift - should not occur */ \
545 return 0; \
546 case 1: /* 1/4 shift */ \
547 return -4 * src[-stride] + 53 * src[0] + \
548 18 * src[stride] - 3 * src[stride * 2]; \
549 case 2: /* 1/2 shift */ \
550 return -1 * src[-stride] + 9 * src[0] + \
551 9 * src[stride] - 1 * src[stride * 2]; \
552 case 3: /* 3/4 shift */ \
553 return -3 * src[-stride] + 18 * src[0] + \
554 53 * src[stride] - 4 * src[stride * 2]; \
555 } \
556 return 0; /* should not occur */ \
557 }
558
559 VC1_MSPEL_FILTER_16B(ver, uint8_t)
560 VC1_MSPEL_FILTER_16B(hor, int16_t)
561
562 /* Filter used to interpolate fractional pel values */
563 static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride,
564 int mode, int r)
565 {
566 switch (mode) {
567 case 0: // no shift
568 return src[0];
569 case 1: // 1/4 shift
570 return (-4 * src[-stride] + 53 * src[0] +
571 18 * src[stride] - 3 * src[stride * 2] + 32 - r) >> 6;
572 case 2: // 1/2 shift
573 return (-1 * src[-stride] + 9 * src[0] +
574 9 * src[stride] - 1 * src[stride * 2] + 8 - r) >> 4;
575 case 3: // 3/4 shift
576 return (-3 * src[-stride] + 18 * src[0] +
577 53 * src[stride] - 4 * src[stride * 2] + 32 - r) >> 6;
578 }
579 return 0; // should not occur
580 }
581
582 /* Function used to do motion compensation with bicubic interpolation */
583 #define VC1_MSPEL_MC(OP, OPNAME) \
584 static av_always_inline void OPNAME ## vc1_mspel_mc(uint8_t *dst, \
585 const uint8_t *src, \
586 int stride, \
587 int hmode, \
588 int vmode, \
589 int rnd) \
590 { \
591 int i, j; \
592 \
593 if (vmode) { /* Horizontal filter to apply */ \
594 int r; \
595 \
596 if (hmode) { /* Vertical filter to apply, output to tmp */ \
597 static const int shift_value[] = { 0, 5, 1, 5 }; \
598 int shift = (shift_value[hmode] + shift_value[vmode]) >> 1; \
599 int16_t tmp[11 * 8], *tptr = tmp; \
600 \
601 r = (1 << (shift - 1)) + rnd - 1; \
602 \
603 src -= 1; \
604 for (j = 0; j < 8; j++) { \
605 for (i = 0; i < 11; i++) \
606 tptr[i] = (vc1_mspel_ver_filter_16bits(src + i, stride, vmode) + r) >> shift; \
607 src += stride; \
608 tptr += 11; \
609 } \
610 \
611 r = 64 - rnd; \
612 tptr = tmp + 1; \
613 for (j = 0; j < 8; j++) { \
614 for (i = 0; i < 8; i++) \
615 OP(dst[i], (vc1_mspel_hor_filter_16bits(tptr + i, 1, hmode) + r) >> 7); \
616 dst += stride; \
617 tptr += 11; \
618 } \
619 \
620 return; \
621 } else { /* No horizontal filter, output 8 lines to dst */ \
622 r = 1 - rnd; \
623 \
624 for (j = 0; j < 8; j++) { \
625 for (i = 0; i < 8; i++) \
626 OP(dst[i], vc1_mspel_filter(src + i, stride, vmode, r)); \
627 src += stride; \
628 dst += stride; \
629 } \
630 return; \
631 } \
632 } \
633 \
634 /* Horizontal mode with no vertical mode */ \
635 for (j = 0; j < 8; j++) { \
636 for (i = 0; i < 8; i++) \
637 OP(dst[i], vc1_mspel_filter(src + i, 1, hmode, rnd)); \
638 dst += stride; \
639 src += stride; \
640 } \
641 }
642
643 #define op_put(a, b) a = av_clip_uint8(b)
644 #define op_avg(a, b) a = (a + av_clip_uint8(b) + 1) >> 1
645
646 VC1_MSPEL_MC(op_put, put_)
647 VC1_MSPEL_MC(op_avg, avg_)
648
649 /* pixel functions - really are entry points to vc1_mspel_mc */
650
651 #define PUT_VC1_MSPEL(a, b) \
652 static void put_vc1_mspel_mc ## a ## b ## _c(uint8_t *dst, \
653 const uint8_t *src, \
654 ptrdiff_t stride, int rnd) \
655 { \
656 put_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
657 } \
658 static void avg_vc1_mspel_mc ## a ## b ## _c(uint8_t *dst, \
659 const uint8_t *src, \
660 ptrdiff_t stride, int rnd) \
661 { \
662 avg_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
663 }
664
665 PUT_VC1_MSPEL(1, 0)
666 PUT_VC1_MSPEL(2, 0)
667 PUT_VC1_MSPEL(3, 0)
668
669 PUT_VC1_MSPEL(0, 1)
670 PUT_VC1_MSPEL(1, 1)
671 PUT_VC1_MSPEL(2, 1)
672 PUT_VC1_MSPEL(3, 1)
673
674 PUT_VC1_MSPEL(0, 2)
675 PUT_VC1_MSPEL(1, 2)
676 PUT_VC1_MSPEL(2, 2)
677 PUT_VC1_MSPEL(3, 2)
678
679 PUT_VC1_MSPEL(0, 3)
680 PUT_VC1_MSPEL(1, 3)
681 PUT_VC1_MSPEL(2, 3)
682 PUT_VC1_MSPEL(3, 3)
683
684 #define chroma_mc(a) \
685 ((A * src[a] + B * src[a + 1] + \
686 C * src[stride + a] + D * src[stride + a + 1] + 32 - 4) >> 6)
687 static void put_no_rnd_vc1_chroma_mc8_c(uint8_t *dst /* align 8 */,
688 uint8_t *src /* align 1 */,
689 int stride, int h, int x, int y)
690 {
691 const int A = (8 - x) * (8 - y);
692 const int B = (x) * (8 - y);
693 const int C = (8 - x) * (y);
694 const int D = (x) * (y);
695 int i;
696
697 assert(x < 8 && y < 8 && x >= 0 && y >= 0);
698
699 for (i = 0; i < h; i++) {
700 dst[0] = chroma_mc(0);
701 dst[1] = chroma_mc(1);
702 dst[2] = chroma_mc(2);
703 dst[3] = chroma_mc(3);
704 dst[4] = chroma_mc(4);
705 dst[5] = chroma_mc(5);
706 dst[6] = chroma_mc(6);
707 dst[7] = chroma_mc(7);
708 dst += stride;
709 src += stride;
710 }
711 }
712
713 static void put_no_rnd_vc1_chroma_mc4_c(uint8_t *dst, uint8_t *src,
714 int stride, int h, int x, int y)
715 {
716 const int A = (8 - x) * (8 - y);
717 const int B = (x) * (8 - y);
718 const int C = (8 - x) * (y);
719 const int D = (x) * (y);
720 int i;
721
722 assert(x < 8 && y < 8 && x >= 0 && y >= 0);
723
724 for (i = 0; i < h; i++) {
725 dst[0] = chroma_mc(0);
726 dst[1] = chroma_mc(1);
727 dst[2] = chroma_mc(2);
728 dst[3] = chroma_mc(3);
729 dst += stride;
730 src += stride;
731 }
732 }
733
734 #define avg2(a, b) (((a) + (b) + 1) >> 1)
735 static void avg_no_rnd_vc1_chroma_mc8_c(uint8_t *dst /* align 8 */,
736 uint8_t *src /* align 1 */,
737 int stride, int h, int x, int y)
738 {
739 const int A = (8 - x) * (8 - y);
740 const int B = (x) * (8 - y);
741 const int C = (8 - x) * (y);
742 const int D = (x) * (y);
743 int i;
744
745 assert(x < 8 && y < 8 && x >= 0 && y >= 0);
746
747 for (i = 0; i < h; i++) {
748 dst[0] = avg2(dst[0], chroma_mc(0));
749 dst[1] = avg2(dst[1], chroma_mc(1));
750 dst[2] = avg2(dst[2], chroma_mc(2));
751 dst[3] = avg2(dst[3], chroma_mc(3));
752 dst[4] = avg2(dst[4], chroma_mc(4));
753 dst[5] = avg2(dst[5], chroma_mc(5));
754 dst[6] = avg2(dst[6], chroma_mc(6));
755 dst[7] = avg2(dst[7], chroma_mc(7));
756 dst += stride;
757 src += stride;
758 }
759 }
760
761 static void avg_no_rnd_vc1_chroma_mc4_c(uint8_t *dst /* align 8 */,
762 uint8_t *src /* align 1 */,
763 int stride, int h, int x, int y)
764 {
765 const int A = (8 - x) * (8 - y);
766 const int B = ( x) * (8 - y);
767 const int C = (8 - x) * ( y);
768 const int D = ( x) * ( y);
769 int i;
770
771 assert(x < 8 && y < 8 && x >= 0 && y >= 0);
772
773 for (i = 0; i < h; i++) {
774 dst[0] = avg2(dst[0], chroma_mc(0));
775 dst[1] = avg2(dst[1], chroma_mc(1));
776 dst[2] = avg2(dst[2], chroma_mc(2));
777 dst[3] = avg2(dst[3], chroma_mc(3));
778 dst += stride;
779 src += stride;
780 }
781 }
782
783 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
784
785 static void sprite_h_c(uint8_t *dst, const uint8_t *src, int offset,
786 int advance, int count)
787 {
788 while (count--) {
789 int a = src[(offset >> 16)];
790 int b = src[(offset >> 16) + 1];
791 *dst++ = a + ((b - a) * (offset & 0xFFFF) >> 16);
792 offset += advance;
793 }
794 }
795
796 static av_always_inline void sprite_v_template(uint8_t *dst,
797 const uint8_t *src1a,
798 const uint8_t *src1b,
799 int offset1,
800 int two_sprites,
801 const uint8_t *src2a,
802 const uint8_t *src2b,
803 int offset2,
804 int alpha, int scaled,
805 int width)
806 {
807 int a1, b1, a2, b2;
808 while (width--) {
809 a1 = *src1a++;
810 if (scaled) {
811 b1 = *src1b++;
812 a1 = a1 + ((b1 - a1) * offset1 >> 16);
813 }
814 if (two_sprites) {
815 a2 = *src2a++;
816 if (scaled > 1) {
817 b2 = *src2b++;
818 a2 = a2 + ((b2 - a2) * offset2 >> 16);
819 }
820 a1 = a1 + ((a2 - a1) * alpha >> 16);
821 }
822 *dst++ = a1;
823 }
824 }
825
826 static void sprite_v_single_c(uint8_t *dst, const uint8_t *src1a,
827 const uint8_t *src1b,
828 int offset, int width)
829 {
830 sprite_v_template(dst, src1a, src1b, offset, 0, NULL, NULL, 0, 0, 1, width);
831 }
832
833 static void sprite_v_double_noscale_c(uint8_t *dst, const uint8_t *src1a,
834 const uint8_t *src2a,
835 int alpha, int width)
836 {
837 sprite_v_template(dst, src1a, NULL, 0, 1, src2a, NULL, 0, alpha, 0, width);
838 }
839
840 static void sprite_v_double_onescale_c(uint8_t *dst,
841 const uint8_t *src1a,
842 const uint8_t *src1b,
843 int offset1,
844 const uint8_t *src2a,
845 int alpha, int width)
846 {
847 sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, NULL, 0, alpha, 1,
848 width);
849 }
850
851 static void sprite_v_double_twoscale_c(uint8_t *dst,
852 const uint8_t *src1a,
853 const uint8_t *src1b,
854 int offset1,
855 const uint8_t *src2a,
856 const uint8_t *src2b,
857 int offset2,
858 int alpha,
859 int width)
860 {
861 sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, src2b, offset2,
862 alpha, 2, width);
863 }
864
865 #endif /* CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER */
866
867 av_cold void ff_vc1dsp_init(VC1DSPContext *dsp)
868 {
869 dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;
870 dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;
871 dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;
872 dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;
873 dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;
874 dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;
875 dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;
876 dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;
877
878 dsp->vc1_h_overlap = vc1_h_overlap_c;
879 dsp->vc1_v_overlap = vc1_v_overlap_c;
880 dsp->vc1_h_s_overlap = vc1_h_s_overlap_c;
881 dsp->vc1_v_s_overlap = vc1_v_s_overlap_c;
882
883 dsp->vc1_v_loop_filter4 = vc1_v_loop_filter4_c;
884 dsp->vc1_h_loop_filter4 = vc1_h_loop_filter4_c;
885 dsp->vc1_v_loop_filter8 = vc1_v_loop_filter8_c;
886 dsp->vc1_h_loop_filter8 = vc1_h_loop_filter8_c;
887 dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_c;
888 dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_c;
889
890 dsp->put_vc1_mspel_pixels_tab[0] = ff_put_pixels8x8_c;
891 dsp->put_vc1_mspel_pixels_tab[1] = put_vc1_mspel_mc10_c;
892 dsp->put_vc1_mspel_pixels_tab[2] = put_vc1_mspel_mc20_c;
893 dsp->put_vc1_mspel_pixels_tab[3] = put_vc1_mspel_mc30_c;
894 dsp->put_vc1_mspel_pixels_tab[4] = put_vc1_mspel_mc01_c;
895 dsp->put_vc1_mspel_pixels_tab[5] = put_vc1_mspel_mc11_c;
896 dsp->put_vc1_mspel_pixels_tab[6] = put_vc1_mspel_mc21_c;
897 dsp->put_vc1_mspel_pixels_tab[7] = put_vc1_mspel_mc31_c;
898 dsp->put_vc1_mspel_pixels_tab[8] = put_vc1_mspel_mc02_c;
899 dsp->put_vc1_mspel_pixels_tab[9] = put_vc1_mspel_mc12_c;
900 dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_c;
901 dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_c;
902 dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_c;
903 dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_c;
904 dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_c;
905 dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_c;
906
907 dsp->avg_vc1_mspel_pixels_tab[0] = ff_avg_pixels8x8_c;
908 dsp->avg_vc1_mspel_pixels_tab[1] = avg_vc1_mspel_mc10_c;
909 dsp->avg_vc1_mspel_pixels_tab[2] = avg_vc1_mspel_mc20_c;
910 dsp->avg_vc1_mspel_pixels_tab[3] = avg_vc1_mspel_mc30_c;
911 dsp->avg_vc1_mspel_pixels_tab[4] = avg_vc1_mspel_mc01_c;
912 dsp->avg_vc1_mspel_pixels_tab[5] = avg_vc1_mspel_mc11_c;
913 dsp->avg_vc1_mspel_pixels_tab[6] = avg_vc1_mspel_mc21_c;
914 dsp->avg_vc1_mspel_pixels_tab[7] = avg_vc1_mspel_mc31_c;
915 dsp->avg_vc1_mspel_pixels_tab[8] = avg_vc1_mspel_mc02_c;
916 dsp->avg_vc1_mspel_pixels_tab[9] = avg_vc1_mspel_mc12_c;
917 dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_c;
918 dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_c;
919 dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_c;
920 dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_c;
921 dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_c;
922 dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_c;
923
924 dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_c;
925 dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_c;
926 dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = put_no_rnd_vc1_chroma_mc4_c;
927 dsp->avg_no_rnd_vc1_chroma_pixels_tab[1] = avg_no_rnd_vc1_chroma_mc4_c;
928
929 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
930 dsp->sprite_h = sprite_h_c;
931 dsp->sprite_v_single = sprite_v_single_c;
932 dsp->sprite_v_double_noscale = sprite_v_double_noscale_c;
933 dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;
934 dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;
935 #endif /* CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER */
936
937 if (ARCH_AARCH64)
938 ff_vc1dsp_init_aarch64(dsp);
939 if (ARCH_ARM)
940 ff_vc1dsp_init_arm(dsp);
941 if (ARCH_PPC)
942 ff_vc1dsp_init_ppc(dsp);
943 if (ARCH_X86)
944 ff_vc1dsp_init_x86(dsp);
945 }