1 /*
2 * VC-1 and WMV3 decoder
3 * Copyright (c) 2011 Mashiat Sarker Shakkhar
4 * Copyright (c) 2006-2007 Konstantin Shishkov
5 * Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 /**
25 * @file
26 * VC-1 and WMV3 block decoding routines
27 */
28
29 #include "avcodec.h"
30 #include "h264chroma.h"
31 #include "mathops.h"
32 #include "mpegvideo.h"
33 #include "vc1.h"
34
vc1_scale_luma(uint8_t * srcY,int k,int linesize)35 static av_always_inline void vc1_scale_luma(uint8_t *srcY,
36 int k, int linesize)
37 {
38 int i, j;
39 for (j = 0; j < k; j++) {
40 for (i = 0; i < k; i++)
41 srcY[i] = ((srcY[i] - 128) >> 1) + 128;
42 srcY += linesize;
43 }
44 }
45
vc1_scale_chroma(uint8_t * srcU,uint8_t * srcV,int k,int uvlinesize)46 static av_always_inline void vc1_scale_chroma(uint8_t *srcU, uint8_t *srcV,
47 int k, int uvlinesize)
48 {
49 int i, j;
50 for (j = 0; j < k; j++) {
51 for (i = 0; i < k; i++) {
52 srcU[i] = ((srcU[i] - 128) >> 1) + 128;
53 srcV[i] = ((srcV[i] - 128) >> 1) + 128;
54 }
55 srcU += uvlinesize;
56 srcV += uvlinesize;
57 }
58 }
59
vc1_lut_scale_luma(uint8_t * srcY,uint8_t * lut1,uint8_t * lut2,int k,int linesize)60 static av_always_inline void vc1_lut_scale_luma(uint8_t *srcY,
61 uint8_t *lut1, uint8_t *lut2,
62 int k, int linesize)
63 {
64 int i, j;
65
66 for (j = 0; j < k; j += 2) {
67 for (i = 0; i < k; i++)
68 srcY[i] = lut1[srcY[i]];
69 srcY += linesize;
70
71 if (j + 1 == k)
72 break;
73
74 for (i = 0; i < k; i++)
75 srcY[i] = lut2[srcY[i]];
76 srcY += linesize;
77 }
78 }
79
vc1_lut_scale_chroma(uint8_t * srcU,uint8_t * srcV,uint8_t * lut1,uint8_t * lut2,int k,int uvlinesize)80 static av_always_inline void vc1_lut_scale_chroma(uint8_t *srcU, uint8_t *srcV,
81 uint8_t *lut1, uint8_t *lut2,
82 int k, int uvlinesize)
83 {
84 int i, j;
85
86 for (j = 0; j < k; j += 2) {
87 for (i = 0; i < k; i++) {
88 srcU[i] = lut1[srcU[i]];
89 srcV[i] = lut1[srcV[i]];
90 }
91 srcU += uvlinesize;
92 srcV += uvlinesize;
93
94 if (j + 1 == k)
95 break;
96
97 for (i = 0; i < k; i++) {
98 srcU[i] = lut2[srcU[i]];
99 srcV[i] = lut2[srcV[i]];
100 }
101 srcU += uvlinesize;
102 srcV += uvlinesize;
103 }
104 }
105
106 static const uint8_t popcount4[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4 };
107
get_luma_mv(VC1Context * v,int dir,int16_t * tx,int16_t * ty)108 static av_always_inline int get_luma_mv(VC1Context *v, int dir, int16_t *tx, int16_t *ty)
109 {
110 MpegEncContext *s = &v->s;
111 int idx = v->mv_f[dir][s->block_index[0] + v->blocks_off] |
112 (v->mv_f[dir][s->block_index[1] + v->blocks_off] << 1) |
113 (v->mv_f[dir][s->block_index[2] + v->blocks_off] << 2) |
114 (v->mv_f[dir][s->block_index[3] + v->blocks_off] << 3);
115 static const uint8_t index2[16] = { 0, 0, 0, 0x23, 0, 0x13, 0x03, 0, 0, 0x12, 0x02, 0, 0x01, 0, 0, 0 };
116 int opp_count = popcount4[idx];
117
118 switch (opp_count) {
119 case 0:
120 case 4:
121 *tx = median4(s->mv[dir][0][0], s->mv[dir][1][0], s->mv[dir][2][0], s->mv[dir][3][0]);
122 *ty = median4(s->mv[dir][0][1], s->mv[dir][1][1], s->mv[dir][2][1], s->mv[dir][3][1]);
123 break;
124 case 1:
125 *tx = mid_pred(s->mv[dir][idx < 2][0], s->mv[dir][1 + (idx < 4)][0], s->mv[dir][2 + (idx < 8)][0]);
126 *ty = mid_pred(s->mv[dir][idx < 2][1], s->mv[dir][1 + (idx < 4)][1], s->mv[dir][2 + (idx < 8)][1]);
127 break;
128 case 3:
129 *tx = mid_pred(s->mv[dir][idx > 0xd][0], s->mv[dir][1 + (idx > 0xb)][0], s->mv[dir][2 + (idx > 0x7)][0]);
130 *ty = mid_pred(s->mv[dir][idx > 0xd][1], s->mv[dir][1 + (idx > 0xb)][1], s->mv[dir][2 + (idx > 0x7)][1]);
131 break;
132 case 2:
133 *tx = (s->mv[dir][index2[idx] >> 4][0] + s->mv[dir][index2[idx] & 0xf][0]) / 2;
134 *ty = (s->mv[dir][index2[idx] >> 4][1] + s->mv[dir][index2[idx] & 0xf][1]) / 2;
135 break;
136 }
137 return opp_count;
138 }
139
get_chroma_mv(VC1Context * v,int dir,int16_t * tx,int16_t * ty)140 static av_always_inline int get_chroma_mv(VC1Context *v, int dir, int16_t *tx, int16_t *ty)
141 {
142 MpegEncContext *s = &v->s;
143 int idx = !v->mb_type[0][s->block_index[0]] |
144 (!v->mb_type[0][s->block_index[1]] << 1) |
145 (!v->mb_type[0][s->block_index[2]] << 2) |
146 (!v->mb_type[0][s->block_index[3]] << 3);
147 static const uint8_t index2[16] = { 0, 0, 0, 0x01, 0, 0x02, 0x12, 0, 0, 0x03, 0x13, 0, 0x23, 0, 0, 0 };
148 int valid_count = popcount4[idx];
149
150 switch (valid_count) {
151 case 4:
152 *tx = median4(s->mv[dir][0][0], s->mv[dir][1][0], s->mv[dir][2][0], s->mv[dir][3][0]);
153 *ty = median4(s->mv[dir][0][1], s->mv[dir][1][1], s->mv[dir][2][1], s->mv[dir][3][1]);
154 break;
155 case 3:
156 *tx = mid_pred(s->mv[dir][idx > 0xd][0], s->mv[dir][1 + (idx > 0xb)][0], s->mv[dir][2 + (idx > 0x7)][0]);
157 *ty = mid_pred(s->mv[dir][idx > 0xd][1], s->mv[dir][1 + (idx > 0xb)][1], s->mv[dir][2 + (idx > 0x7)][1]);
158 break;
159 case 2:
160 *tx = (s->mv[dir][index2[idx] >> 4][0] + s->mv[dir][index2[idx] & 0xf][0]) / 2;
161 *ty = (s->mv[dir][index2[idx] >> 4][1] + s->mv[dir][index2[idx] & 0xf][1]) / 2;
162 break;
163 default:
164 return 0;
165 }
166 return valid_count;
167 }
168
169 /** Do motion compensation over 1 macroblock
170 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
171 */
ff_vc1_mc_1mv(VC1Context * v,int dir)172 void ff_vc1_mc_1mv(VC1Context *v, int dir)
173 {
174 MpegEncContext *s = &v->s;
175 H264ChromaContext *h264chroma = &v->h264chroma;
176 uint8_t *srcY, *srcU, *srcV;
177 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
178 int v_edge_pos = s->v_edge_pos >> v->field_mode;
179 int i;
180 uint8_t (*luty)[256], (*lutuv)[256];
181 int use_ic;
182 int interlace;
183 int linesize, uvlinesize;
184
185 if ((!v->field_mode ||
186 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
187 !v->s.last_picture.f->data[0])
188 return;
189
190 linesize = s->current_picture_ptr->f->linesize[0];
191 uvlinesize = s->current_picture_ptr->f->linesize[1];
192
193 mx = s->mv[dir][0][0];
194 my = s->mv[dir][0][1];
195
196 // store motion vectors for further use in B-frames
197 if (s->pict_type == AV_PICTURE_TYPE_P) {
198 for (i = 0; i < 4; i++) {
199 s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][0] = mx;
200 s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][1] = my;
201 }
202 }
203
204 uvmx = (mx + ((mx & 3) == 3)) >> 1;
205 uvmy = (my + ((my & 3) == 3)) >> 1;
206 v->luma_mv[s->mb_x][0] = uvmx;
207 v->luma_mv[s->mb_x][1] = uvmy;
208
209 if (v->field_mode &&
210 v->cur_field_type != v->ref_field_type[dir]) {
211 my = my - 2 + 4 * v->cur_field_type;
212 uvmy = uvmy - 2 + 4 * v->cur_field_type;
213 }
214
215 // fastuvmc shall be ignored for interlaced frame picture
216 if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
217 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
218 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
219 }
220 if (!dir) {
221 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
222 srcY = s->current_picture.f->data[0];
223 srcU = s->current_picture.f->data[1];
224 srcV = s->current_picture.f->data[2];
225 luty = v->curr_luty;
226 lutuv = v->curr_lutuv;
227 use_ic = *v->curr_use_ic;
228 interlace = 1;
229 } else {
230 srcY = s->last_picture.f->data[0];
231 srcU = s->last_picture.f->data[1];
232 srcV = s->last_picture.f->data[2];
233 luty = v->last_luty;
234 lutuv = v->last_lutuv;
235 use_ic = v->last_use_ic;
236 interlace = s->last_picture.f->interlaced_frame;
237 }
238 } else {
239 srcY = s->next_picture.f->data[0];
240 srcU = s->next_picture.f->data[1];
241 srcV = s->next_picture.f->data[2];
242 luty = v->next_luty;
243 lutuv = v->next_lutuv;
244 use_ic = v->next_use_ic;
245 interlace = s->next_picture.f->interlaced_frame;
246 }
247
248 if (!srcY || !srcU) {
249 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
250 return;
251 }
252
253 src_x = s->mb_x * 16 + (mx >> 2);
254 src_y = s->mb_y * 16 + (my >> 2);
255 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
256 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
257
258 if (v->profile != PROFILE_ADVANCED) {
259 src_x = av_clip( src_x, -16, s->mb_width * 16);
260 src_y = av_clip( src_y, -16, s->mb_height * 16);
261 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
262 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
263 } else {
264 src_x = av_clip( src_x, -17, s->avctx->coded_width);
265 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
266 if (v->fcm == ILACE_FRAME) {
267 src_y = av_clip(src_y, -18 + (src_y & 1), s->avctx->coded_height + (src_y & 1));
268 uvsrc_y = av_clip(uvsrc_y, -8 + (uvsrc_y & 1), (s->avctx->coded_height >> 1) + (uvsrc_y & 1));
269 } else {
270 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
271 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
272 }
273 }
274
275 srcY += src_y * s->linesize + src_x;
276 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
277 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
278
279 if (v->field_mode && v->ref_field_type[dir]) {
280 srcY += linesize;
281 srcU += uvlinesize;
282 srcV += uvlinesize;
283 }
284
285 /* for grayscale we should not try to read from unknown area */
286 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY) {
287 srcU = s->sc.edge_emu_buffer + 18 * s->linesize;
288 srcV = s->sc.edge_emu_buffer + 18 * s->linesize;
289 }
290
291 if (v->rangeredfrm || use_ic
292 || s->h_edge_pos < 22 || v_edge_pos < 22
293 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
294 || (unsigned)(src_y - 1) > v_edge_pos - (my&3) - 16 - 3) {
295 uint8_t *ubuf = s->sc.edge_emu_buffer + 19 * s->linesize;
296 uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
297 const int k = 17 + s->mspel * 2;
298
299 srcY -= s->mspel * (1 + s->linesize);
300 if (interlace) {
301 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
302 srcY,
303 linesize << 1,
304 linesize << 1,
305 k,
306 v->field_mode ? k : k + 1 >> 1,
307 src_x - s->mspel,
308 src_y - s->mspel >> !v->field_mode,
309 s->h_edge_pos,
310 s->v_edge_pos >> 1);
311 if (!v->field_mode)
312 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + linesize,
313 srcY + linesize,
314 linesize << 1,
315 linesize << 1,
316 k,
317 k >> 1,
318 src_x - s->mspel,
319 src_y - s->mspel + 1 >> 1,
320 s->h_edge_pos,
321 s->v_edge_pos >> 1);
322 } else
323 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
324 srcY,
325 linesize,
326 linesize,
327 k,
328 v->field_mode ? (k << 1) - 1 : k,
329 src_x - s->mspel,
330 v->field_mode ? 2 * (src_y - s->mspel) + v->ref_field_type[dir] :
331 src_y - s->mspel,
332 s->h_edge_pos,
333 s->v_edge_pos);
334 srcY = s->sc.edge_emu_buffer;
335 if (interlace) {
336 s->vdsp.emulated_edge_mc(ubuf,
337 srcU,
338 uvlinesize << 1,
339 uvlinesize << 1,
340 9,
341 v->field_mode ? 9 : 5,
342 uvsrc_x,
343 uvsrc_y >> !v->field_mode,
344 s->h_edge_pos >> 1,
345 s->v_edge_pos >> 2);
346 s->vdsp.emulated_edge_mc(vbuf,
347 srcV,
348 uvlinesize << 1,
349 uvlinesize << 1,
350 9,
351 v->field_mode ? 9 : 5,
352 uvsrc_x,
353 uvsrc_y >> !v->field_mode,
354 s->h_edge_pos >> 1,
355 s->v_edge_pos >> 2);
356 if (!v->field_mode) {
357 s->vdsp.emulated_edge_mc(ubuf + uvlinesize,
358 srcU + uvlinesize,
359 uvlinesize << 1,
360 uvlinesize << 1,
361 9,
362 4,
363 uvsrc_x,
364 uvsrc_y + 1 >> 1,
365 s->h_edge_pos >> 1,
366 s->v_edge_pos >> 2);
367 s->vdsp.emulated_edge_mc(vbuf + uvlinesize,
368 srcV + uvlinesize,
369 uvlinesize << 1,
370 uvlinesize << 1,
371 9,
372 4,
373 uvsrc_x,
374 uvsrc_y + 1 >> 1,
375 s->h_edge_pos >> 1,
376 s->v_edge_pos >> 2);
377 }
378 } else {
379 s->vdsp.emulated_edge_mc(ubuf,
380 srcU,
381 uvlinesize,
382 uvlinesize,
383 9,
384 v->field_mode ? 17 : 9,
385 uvsrc_x,
386 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[dir] : uvsrc_y,
387 s->h_edge_pos >> 1,
388 s->v_edge_pos >> 1);
389 s->vdsp.emulated_edge_mc(vbuf,
390 srcV,
391 uvlinesize,
392 uvlinesize,
393 9,
394 v->field_mode ? 17 : 9,
395 uvsrc_x,
396 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[dir] : uvsrc_y,
397 s->h_edge_pos >> 1,
398 s->v_edge_pos >> 1);
399 }
400 srcU = ubuf;
401 srcV = vbuf;
402 /* if we deal with range reduction we need to scale source blocks */
403 if (v->rangeredfrm) {
404 vc1_scale_luma(srcY, k, s->linesize);
405 vc1_scale_chroma(srcU, srcV, 9, s->uvlinesize);
406 }
407 /* if we deal with intensity compensation we need to scale source blocks */
408 if (use_ic) {
409 vc1_lut_scale_luma(srcY,
410 luty[v->field_mode ? v->ref_field_type[dir] : ((0 + src_y - s->mspel) & 1)],
411 luty[v->field_mode ? v->ref_field_type[dir] : ((1 + src_y - s->mspel) & 1)],
412 k, s->linesize);
413 vc1_lut_scale_chroma(srcU, srcV,
414 lutuv[v->field_mode ? v->ref_field_type[dir] : ((0 + uvsrc_y) & 1)],
415 lutuv[v->field_mode ? v->ref_field_type[dir] : ((1 + uvsrc_y) & 1)],
416 9, s->uvlinesize);
417 }
418 srcY += s->mspel * (1 + s->linesize);
419 }
420
421 if (s->mspel) {
422 dxy = ((my & 3) << 2) | (mx & 3);
423 v->vc1dsp.put_vc1_mspel_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, v->rnd);
424 } else { // hpel mc - always used for luma
425 dxy = (my & 2) | ((mx & 2) >> 1);
426 if (!v->rnd)
427 s->hdsp.put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
428 else
429 s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
430 }
431
432 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
433 return;
434 /* Chroma MC always uses qpel bilinear */
435 uvmx = (uvmx & 3) << 1;
436 uvmy = (uvmy & 3) << 1;
437 if (!v->rnd) {
438 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
439 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
440 } else {
441 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
442 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
443 }
444 if (v->field_mode) {
445 v->mv_f[dir][s->block_index[4] + v->mb_off] = v->cur_field_type != v->ref_field_type[dir];
446 v->mv_f[dir][s->block_index[5] + v->mb_off] = v->cur_field_type != v->ref_field_type[dir];
447 }
448 }
449
450 /** Do motion compensation for 4-MV macroblock - luminance block
451 */
ff_vc1_mc_4mv_luma(VC1Context * v,int n,int dir,int avg)452 void ff_vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
453 {
454 MpegEncContext *s = &v->s;
455 uint8_t *srcY;
456 int dxy, mx, my, src_x, src_y;
457 int off;
458 int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
459 int v_edge_pos = s->v_edge_pos >> v->field_mode;
460 uint8_t (*luty)[256];
461 int use_ic;
462 int interlace;
463 int linesize;
464
465 if ((!v->field_mode ||
466 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
467 !v->s.last_picture.f->data[0])
468 return;
469
470 linesize = s->current_picture_ptr->f->linesize[0];
471
472 mx = s->mv[dir][n][0];
473 my = s->mv[dir][n][1];
474
475 if (!dir) {
476 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
477 srcY = s->current_picture.f->data[0];
478 luty = v->curr_luty;
479 use_ic = *v->curr_use_ic;
480 interlace = 1;
481 } else {
482 srcY = s->last_picture.f->data[0];
483 luty = v->last_luty;
484 use_ic = v->last_use_ic;
485 interlace = s->last_picture.f->interlaced_frame;
486 }
487 } else {
488 srcY = s->next_picture.f->data[0];
489 luty = v->next_luty;
490 use_ic = v->next_use_ic;
491 interlace = s->next_picture.f->interlaced_frame;
492 }
493
494 if (!srcY) {
495 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
496 return;
497 }
498
499 if (v->field_mode) {
500 if (v->cur_field_type != v->ref_field_type[dir])
501 my = my - 2 + 4 * v->cur_field_type;
502 }
503
504 if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
505 int opp_count = get_luma_mv(v, 0,
506 &s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0],
507 &s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1]);
508 int k, f = opp_count > 2;
509 for (k = 0; k < 4; k++)
510 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
511 }
512
513 if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
514 int qx, qy;
515 int width = s->avctx->coded_width;
516 int height = s->avctx->coded_height >> 1;
517 if (s->pict_type == AV_PICTURE_TYPE_P) {
518 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][0] = mx;
519 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][1] = my;
520 }
521 qx = (s->mb_x * 16) + (mx >> 2);
522 qy = (s->mb_y * 8) + (my >> 3);
523
524 if (qx < -17)
525 mx -= 4 * (qx + 17);
526 else if (qx > width)
527 mx -= 4 * (qx - width);
528 if (qy < -18)
529 my -= 8 * (qy + 18);
530 else if (qy > height + 1)
531 my -= 8 * (qy - height - 1);
532 }
533
534 if ((v->fcm == ILACE_FRAME) && fieldmv)
535 off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
536 else
537 off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
538
539 src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
540 if (!fieldmv)
541 src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
542 else
543 src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
544
545 if (v->profile != PROFILE_ADVANCED) {
546 src_x = av_clip(src_x, -16, s->mb_width * 16);
547 src_y = av_clip(src_y, -16, s->mb_height * 16);
548 } else {
549 src_x = av_clip(src_x, -17, s->avctx->coded_width);
550 if (v->fcm == ILACE_FRAME)
551 src_y = av_clip(src_y, -18 + (src_y & 1), s->avctx->coded_height + (src_y & 1));
552 else
553 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
554 }
555
556 srcY += src_y * s->linesize + src_x;
557 if (v->field_mode && v->ref_field_type[dir])
558 srcY += linesize;
559
560 if (v->rangeredfrm || use_ic
561 || s->h_edge_pos < 13 || v_edge_pos < 23
562 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
563 || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
564 const int k = 9 + s->mspel * 2;
565
566 srcY -= s->mspel * (1 + (s->linesize << fieldmv));
567 /* check emulate edge stride and offset */
568 if (interlace) {
569 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
570 srcY,
571 linesize << 1,
572 linesize << 1,
573 k,
574 v->field_mode ? k : (k << fieldmv) + 1 >> 1,
575 src_x - s->mspel,
576 src_y - (s->mspel << fieldmv) >> !v->field_mode,
577 s->h_edge_pos,
578 s->v_edge_pos >> 1);
579 if (!v->field_mode && !fieldmv)
580 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + linesize,
581 srcY + linesize,
582 linesize << 1,
583 linesize << 1,
584 k,
585 k >> 1,
586 src_x - s->mspel,
587 src_y - s->mspel + 1 >> 1,
588 s->h_edge_pos,
589 s->v_edge_pos >> 1);
590 } else
591 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
592 srcY,
593 linesize,
594 linesize,
595 k,
596 v->field_mode ? (k << 1) - 1 : k << fieldmv,
597 src_x - s->mspel,
598 v->field_mode ? 2 * (src_y - s->mspel) + v->ref_field_type[dir] :
599 src_y - (s->mspel << fieldmv),
600 s->h_edge_pos,
601 s->v_edge_pos);
602 srcY = s->sc.edge_emu_buffer;
603 /* if we deal with range reduction we need to scale source blocks */
604 if (v->rangeredfrm) {
605 vc1_scale_luma(srcY, k, s->linesize << fieldmv);
606 }
607 /* if we deal with intensity compensation we need to scale source blocks */
608 if (use_ic) {
609 vc1_lut_scale_luma(srcY,
610 luty[v->field_mode ? v->ref_field_type[dir] : (((0<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1)],
611 luty[v->field_mode ? v->ref_field_type[dir] : (((1<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1)],
612 k, s->linesize << fieldmv);
613 }
614 srcY += s->mspel * (1 + (s->linesize << fieldmv));
615 }
616
617 if (s->mspel) {
618 dxy = ((my & 3) << 2) | (mx & 3);
619 if (avg)
620 v->vc1dsp.avg_vc1_mspel_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
621 else
622 v->vc1dsp.put_vc1_mspel_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
623 } else { // hpel mc - always used for luma
624 dxy = (my & 2) | ((mx & 2) >> 1);
625 if (!v->rnd)
626 s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
627 else
628 s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
629 }
630 }
631
632 /** Do motion compensation for 4-MV macroblock - both chroma blocks
633 */
ff_vc1_mc_4mv_chroma(VC1Context * v,int dir)634 void ff_vc1_mc_4mv_chroma(VC1Context *v, int dir)
635 {
636 MpegEncContext *s = &v->s;
637 H264ChromaContext *h264chroma = &v->h264chroma;
638 uint8_t *srcU, *srcV;
639 int uvmx, uvmy, uvsrc_x, uvsrc_y;
640 int16_t tx, ty;
641 int chroma_ref_type;
642 int v_edge_pos = s->v_edge_pos >> v->field_mode;
643 uint8_t (*lutuv)[256];
644 int use_ic;
645 int interlace;
646 int uvlinesize;
647
648 if (!v->field_mode && !v->s.last_picture.f->data[0])
649 return;
650 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
651 return;
652
653 /* calculate chroma MV vector from four luma MVs */
654 if (!v->field_mode || !v->numref) {
655 int valid_count = get_chroma_mv(v, dir, &tx, &ty);
656 if (!valid_count) {
657 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
658 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
659 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
660 return; //no need to do MC for intra blocks
661 }
662 chroma_ref_type = v->ref_field_type[dir];
663 } else {
664 int opp_count = get_luma_mv(v, dir, &tx, &ty);
665 chroma_ref_type = v->cur_field_type ^ (opp_count > 2);
666 }
667 if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f->data[0])
668 return;
669 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
670 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
671
672 uvlinesize = s->current_picture_ptr->f->linesize[1];
673
674 uvmx = (tx + ((tx & 3) == 3)) >> 1;
675 uvmy = (ty + ((ty & 3) == 3)) >> 1;
676
677 v->luma_mv[s->mb_x][0] = uvmx;
678 v->luma_mv[s->mb_x][1] = uvmy;
679
680 if (v->fastuvmc) {
681 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
682 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
683 }
684 // Field conversion bias
685 if (v->cur_field_type != chroma_ref_type)
686 uvmy += 2 - 4 * chroma_ref_type;
687
688 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
689 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
690
691 if (v->profile != PROFILE_ADVANCED) {
692 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
693 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
694 } else {
695 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
696 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
697 }
698
699 if (!dir) {
700 if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
701 srcU = s->current_picture.f->data[1];
702 srcV = s->current_picture.f->data[2];
703 lutuv = v->curr_lutuv;
704 use_ic = *v->curr_use_ic;
705 interlace = 1;
706 } else {
707 srcU = s->last_picture.f->data[1];
708 srcV = s->last_picture.f->data[2];
709 lutuv = v->last_lutuv;
710 use_ic = v->last_use_ic;
711 interlace = s->last_picture.f->interlaced_frame;
712 }
713 } else {
714 srcU = s->next_picture.f->data[1];
715 srcV = s->next_picture.f->data[2];
716 lutuv = v->next_lutuv;
717 use_ic = v->next_use_ic;
718 interlace = s->next_picture.f->interlaced_frame;
719 }
720
721 if (!srcU) {
722 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
723 return;
724 }
725
726 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
727 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
728
729 if (v->field_mode) {
730 if (chroma_ref_type) {
731 srcU += uvlinesize;
732 srcV += uvlinesize;
733 }
734 }
735
736 if (v->rangeredfrm || use_ic
737 || s->h_edge_pos < 18 || v_edge_pos < 18
738 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
739 || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) {
740 if (interlace) {
741 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
742 srcU,
743 uvlinesize << 1,
744 uvlinesize << 1,
745 9,
746 v->field_mode ? 9 : 5,
747 uvsrc_x,
748 uvsrc_y >> !v->field_mode,
749 s->h_edge_pos >> 1,
750 s->v_edge_pos >> 2);
751 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
752 srcV,
753 uvlinesize << 1,
754 uvlinesize << 1,
755 9,
756 v->field_mode ? 9 : 5,
757 uvsrc_x,
758 uvsrc_y >> !v->field_mode,
759 s->h_edge_pos >> 1,
760 s->v_edge_pos >> 2);
761 if (!v->field_mode) {
762 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + uvlinesize,
763 srcU + uvlinesize,
764 uvlinesize << 1,
765 uvlinesize << 1,
766 9,
767 4,
768 uvsrc_x,
769 uvsrc_y + 1 >> 1,
770 s->h_edge_pos >> 1,
771 s->v_edge_pos >> 2);
772 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16 + uvlinesize,
773 srcV + uvlinesize,
774 uvlinesize << 1,
775 uvlinesize << 1,
776 9,
777 4,
778 uvsrc_x,
779 uvsrc_y + 1 >> 1,
780 s->h_edge_pos >> 1,
781 s->v_edge_pos >> 2);
782 }
783 } else {
784 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
785 srcU,
786 uvlinesize,
787 uvlinesize,
788 9,
789 v->field_mode ? 17 : 9,
790 uvsrc_x,
791 v->field_mode ? 2 * uvsrc_y + chroma_ref_type : uvsrc_y,
792 s->h_edge_pos >> 1,
793 s->v_edge_pos >> 1);
794 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
795 srcV,
796 uvlinesize,
797 uvlinesize,
798 9,
799 v->field_mode ? 17 : 9,
800 uvsrc_x,
801 v->field_mode ? 2 * uvsrc_y + chroma_ref_type : uvsrc_y,
802 s->h_edge_pos >> 1,
803 s->v_edge_pos >> 1);
804 }
805 srcU = s->sc.edge_emu_buffer;
806 srcV = s->sc.edge_emu_buffer + 16;
807
808 /* if we deal with range reduction we need to scale source blocks */
809 if (v->rangeredfrm) {
810 vc1_scale_chroma(srcU, srcV, 9, s->uvlinesize);
811 }
812 /* if we deal with intensity compensation we need to scale source blocks */
813 if (use_ic) {
814 vc1_lut_scale_chroma(srcU, srcV,
815 lutuv[v->field_mode ? chroma_ref_type : ((0 + uvsrc_y) & 1)],
816 lutuv[v->field_mode ? chroma_ref_type : ((1 + uvsrc_y) & 1)],
817 9, s->uvlinesize);
818 }
819 }
820
821 /* Chroma MC always uses qpel bilinear */
822 uvmx = (uvmx & 3) << 1;
823 uvmy = (uvmy & 3) << 1;
824 if (!v->rnd) {
825 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
826 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
827 } else {
828 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
829 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
830 }
831 if (v->field_mode) {
832 v->mv_f[dir][s->block_index[4] + v->mb_off] = v->cur_field_type != chroma_ref_type;
833 v->mv_f[dir][s->block_index[5] + v->mb_off] = v->cur_field_type != chroma_ref_type;
834 }
835 }
836
837 /** Do motion compensation for 4-MV interlaced frame chroma macroblock (both U and V)
838 */
ff_vc1_mc_4mv_chroma4(VC1Context * v,int dir,int dir2,int avg)839 void ff_vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg)
840 {
841 MpegEncContext *s = &v->s;
842 H264ChromaContext *h264chroma = &v->h264chroma;
843 uint8_t *srcU, *srcV;
844 int uvsrc_x, uvsrc_y;
845 int uvmx_field[4], uvmy_field[4];
846 int i, off, tx, ty;
847 int fieldmv = v->blk_mv_type[s->block_index[0]];
848 static const uint8_t s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
849 int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
850 int v_edge_pos = s->v_edge_pos >> 1;
851 int use_ic;
852 int interlace;
853 int uvlinesize;
854 uint8_t (*lutuv)[256];
855
856 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
857 return;
858
859 uvlinesize = s->current_picture_ptr->f->linesize[1];
860
861 for (i = 0; i < 4; i++) {
862 int d = i < 2 ? dir: dir2;
863 tx = s->mv[d][i][0];
864 uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
865 ty = s->mv[d][i][1];
866 if (fieldmv)
867 uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
868 else
869 uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
870 }
871
872 for (i = 0; i < 4; i++) {
873 off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
874 uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2);
875 uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
876 // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
877 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
878 if (v->fcm == ILACE_FRAME)
879 uvsrc_y = av_clip(uvsrc_y, -8 + (uvsrc_y & 1), (s->avctx->coded_height >> 1) + (uvsrc_y & 1));
880 else
881 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
882 if (i < 2 ? dir : dir2) {
883 srcU = s->next_picture.f->data[1];
884 srcV = s->next_picture.f->data[2];
885 lutuv = v->next_lutuv;
886 use_ic = v->next_use_ic;
887 interlace = s->next_picture.f->interlaced_frame;
888 } else {
889 srcU = s->last_picture.f->data[1];
890 srcV = s->last_picture.f->data[2];
891 lutuv = v->last_lutuv;
892 use_ic = v->last_use_ic;
893 interlace = s->last_picture.f->interlaced_frame;
894 }
895 if (!srcU)
896 return;
897 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
898 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
899 uvmx_field[i] = (uvmx_field[i] & 3) << 1;
900 uvmy_field[i] = (uvmy_field[i] & 3) << 1;
901
902 if (use_ic
903 || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
904 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
905 || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
906 if (interlace) {
907 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
908 srcU,
909 uvlinesize << 1,
910 uvlinesize << 1,
911 5,
912 (5 << fieldmv) + 1 >> 1,
913 uvsrc_x,
914 uvsrc_y >> 1,
915 s->h_edge_pos >> 1,
916 s->v_edge_pos >> 2);
917 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
918 srcV,
919 uvlinesize << 1,
920 uvlinesize << 1,
921 5,
922 (5 << fieldmv) + 1 >> 1,
923 uvsrc_x,
924 uvsrc_y >> 1,
925 s->h_edge_pos >> 1,
926 s->v_edge_pos >> 2);
927 if (!fieldmv) {
928 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + uvlinesize,
929 srcU + uvlinesize,
930 uvlinesize << 1,
931 uvlinesize << 1,
932 5,
933 2,
934 uvsrc_x,
935 uvsrc_y + 1 >> 1,
936 s->h_edge_pos >> 1,
937 s->v_edge_pos >> 2);
938 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16 + uvlinesize,
939 srcV + uvlinesize,
940 uvlinesize << 1,
941 uvlinesize << 1,
942 5,
943 2,
944 uvsrc_x,
945 uvsrc_y + 1 >> 1,
946 s->h_edge_pos >> 1,
947 s->v_edge_pos >> 2);
948 }
949 } else {
950 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
951 srcU,
952 uvlinesize,
953 uvlinesize,
954 5,
955 5 << fieldmv,
956 uvsrc_x,
957 uvsrc_y,
958 s->h_edge_pos >> 1,
959 s->v_edge_pos >> 1);
960 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
961 srcV,
962 uvlinesize,
963 uvlinesize,
964 5,
965 5 << fieldmv,
966 uvsrc_x,
967 uvsrc_y,
968 s->h_edge_pos >> 1,
969 s->v_edge_pos >> 1);
970 }
971 srcU = s->sc.edge_emu_buffer;
972 srcV = s->sc.edge_emu_buffer + 16;
973
974 /* if we deal with intensity compensation we need to scale source blocks */
975 if (use_ic) {
976 vc1_lut_scale_chroma(srcU, srcV,
977 lutuv[(uvsrc_y + (0 << fieldmv)) & 1],
978 lutuv[(uvsrc_y + (1 << fieldmv)) & 1],
979 5, s->uvlinesize << fieldmv);
980 }
981 }
982 if (avg) {
983 if (!v->rnd) {
984 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
985 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
986 } else {
987 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
988 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
989 }
990 } else {
991 if (!v->rnd) {
992 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
993 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
994 } else {
995 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
996 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
997 }
998 }
999 }
1000 }
1001
1002 /** Motion compensation for direct or interpolated blocks in B-frames
1003 */
ff_vc1_interp_mc(VC1Context * v)1004 void ff_vc1_interp_mc(VC1Context *v)
1005 {
1006 MpegEncContext *s = &v->s;
1007 H264ChromaContext *h264chroma = &v->h264chroma;
1008 uint8_t *srcY, *srcU, *srcV;
1009 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
1010 int v_edge_pos = s->v_edge_pos >> v->field_mode;
1011 int use_ic = v->next_use_ic;
1012 int interlace;
1013 int linesize, uvlinesize;
1014
1015 if (!v->field_mode && !v->s.next_picture.f->data[0])
1016 return;
1017
1018 linesize = s->current_picture_ptr->f->linesize[0];
1019 uvlinesize = s->current_picture_ptr->f->linesize[1];
1020
1021 mx = s->mv[1][0][0];
1022 my = s->mv[1][0][1];
1023 uvmx = (mx + ((mx & 3) == 3)) >> 1;
1024 uvmy = (my + ((my & 3) == 3)) >> 1;
1025 if (v->field_mode && v->cur_field_type != v->ref_field_type[1]) {
1026 my = my - 2 + 4 * v->cur_field_type;
1027 uvmy = uvmy - 2 + 4 * v->cur_field_type;
1028 }
1029 if (v->fastuvmc) {
1030 uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
1031 uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
1032 }
1033 srcY = s->next_picture.f->data[0];
1034 srcU = s->next_picture.f->data[1];
1035 srcV = s->next_picture.f->data[2];
1036
1037 interlace = s->next_picture.f->interlaced_frame;
1038
1039 src_x = s->mb_x * 16 + (mx >> 2);
1040 src_y = s->mb_y * 16 + (my >> 2);
1041 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
1042 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
1043
1044 if (v->profile != PROFILE_ADVANCED) {
1045 src_x = av_clip( src_x, -16, s->mb_width * 16);
1046 src_y = av_clip( src_y, -16, s->mb_height * 16);
1047 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
1048 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
1049 } else {
1050 src_x = av_clip( src_x, -17, s->avctx->coded_width);
1051 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
1052 if (v->fcm == ILACE_FRAME) {
1053 src_y = av_clip(src_y, -18 + (src_y & 1), s->avctx->coded_height + (src_y & 1));
1054 uvsrc_y = av_clip(uvsrc_y, -8 + (uvsrc_y & 1), (s->avctx->coded_height >> 1) + (uvsrc_y & 1));
1055 } else {
1056 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
1057 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
1058 }
1059 }
1060
1061 srcY += src_y * s->linesize + src_x;
1062 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1063 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1064
1065 if (v->field_mode && v->ref_field_type[1]) {
1066 srcY += linesize;
1067 srcU += uvlinesize;
1068 srcV += uvlinesize;
1069 }
1070
1071 /* for grayscale we should not try to read from unknown area */
1072 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY) {
1073 srcU = s->sc.edge_emu_buffer + 18 * s->linesize;
1074 srcV = s->sc.edge_emu_buffer + 18 * s->linesize;
1075 }
1076
1077 if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22 || use_ic
1078 || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
1079 || (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) {
1080 uint8_t *ubuf = s->sc.edge_emu_buffer + 19 * s->linesize;
1081 uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
1082 const int k = 17 + s->mspel * 2;
1083
1084 srcY -= s->mspel * (1 + s->linesize);
1085 if (interlace) {
1086 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
1087 srcY,
1088 linesize << 1,
1089 linesize << 1,
1090 k,
1091 v->field_mode ? k : (k + 1 >> 1),
1092 src_x - s->mspel,
1093 src_y - s->mspel >> !v->field_mode,
1094 s->h_edge_pos,
1095 s->v_edge_pos >> 1);
1096 if (!v->field_mode)
1097 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + linesize,
1098 srcY + linesize,
1099 linesize << 1,
1100 linesize << 1,
1101 k,
1102 k >> 1,
1103 src_x - s->mspel,
1104 src_y - s->mspel + 1 >> 1,
1105 s->h_edge_pos,
1106 s->v_edge_pos >> 1);
1107 } else
1108 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
1109 srcY,
1110 linesize,
1111 linesize,
1112 k,
1113 v->field_mode ? (k << 1) - 1 : k,
1114 src_x - s->mspel,
1115 v->field_mode ? 2 * (src_y - s->mspel) + v->ref_field_type[1] :
1116 src_y - s->mspel,
1117 s->h_edge_pos,
1118 s->v_edge_pos);
1119 srcY = s->sc.edge_emu_buffer;
1120 if (interlace) {
1121 s->vdsp.emulated_edge_mc(ubuf,
1122 srcU,
1123 uvlinesize << 1,
1124 uvlinesize << 1,
1125 9,
1126 v->field_mode ? 9 : 5,
1127 uvsrc_x,
1128 uvsrc_y >> !v->field_mode,
1129 s->h_edge_pos >> 1,
1130 s->v_edge_pos >> 2);
1131 s->vdsp.emulated_edge_mc(vbuf,
1132 srcV,
1133 uvlinesize << 1,
1134 uvlinesize << 1,
1135 9,
1136 v->field_mode ? 9 : 5,
1137 uvsrc_x,
1138 uvsrc_y >> !v->field_mode,
1139 s->h_edge_pos >> 1,
1140 s->v_edge_pos >> 2);
1141 if (!v->field_mode) {
1142 s->vdsp.emulated_edge_mc(ubuf + uvlinesize,
1143 srcU + uvlinesize,
1144 uvlinesize << 1,
1145 uvlinesize << 1,
1146 9,
1147 4,
1148 uvsrc_x,
1149 uvsrc_y + 1 >> 1,
1150 s->h_edge_pos >> 1,
1151 s->v_edge_pos >> 2);
1152 s->vdsp.emulated_edge_mc(vbuf + uvlinesize,
1153 srcV + uvlinesize,
1154 uvlinesize << 1,
1155 uvlinesize << 1,
1156 9,
1157 4,
1158 uvsrc_x,
1159 uvsrc_y + 1 >> 1,
1160 s->h_edge_pos >> 1,
1161 s->v_edge_pos >> 2);
1162 }
1163 } else {
1164 s->vdsp.emulated_edge_mc(ubuf,
1165 srcU,
1166 uvlinesize,
1167 uvlinesize,
1168 9,
1169 v->field_mode ? 17 : 9,
1170 uvsrc_x,
1171 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[1] : uvsrc_y,
1172 s->h_edge_pos >> 1,
1173 s->v_edge_pos >> 1);
1174 s->vdsp.emulated_edge_mc(vbuf,
1175 srcV,
1176 uvlinesize,
1177 uvlinesize,
1178 9,
1179 v->field_mode ? 17 : 9,
1180 uvsrc_x,
1181 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[1] : uvsrc_y,
1182 s->h_edge_pos >> 1,
1183 s->v_edge_pos >> 1);
1184 }
1185 srcU = ubuf;
1186 srcV = vbuf;
1187 /* if we deal with range reduction we need to scale source blocks */
1188 if (v->rangeredfrm) {
1189 vc1_scale_luma(srcY, k, s->linesize);
1190 vc1_scale_chroma(srcU, srcV, 9, s->uvlinesize);
1191 }
1192
1193 if (use_ic) {
1194 uint8_t (*luty )[256] = v->next_luty;
1195 uint8_t (*lutuv)[256] = v->next_lutuv;
1196 vc1_lut_scale_luma(srcY,
1197 luty[v->field_mode ? v->ref_field_type[1] : ((0+src_y - s->mspel) & 1)],
1198 luty[v->field_mode ? v->ref_field_type[1] : ((1+src_y - s->mspel) & 1)],
1199 k, s->linesize);
1200 vc1_lut_scale_chroma(srcU, srcV,
1201 lutuv[v->field_mode ? v->ref_field_type[1] : ((0+uvsrc_y) & 1)],
1202 lutuv[v->field_mode ? v->ref_field_type[1] : ((1+uvsrc_y) & 1)],
1203 9, s->uvlinesize);
1204 }
1205 srcY += s->mspel * (1 + s->linesize);
1206 }
1207
1208 if (s->mspel) {
1209 dxy = ((my & 3) << 2) | (mx & 3);
1210 v->vc1dsp.avg_vc1_mspel_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, v->rnd);
1211 } else { // hpel mc
1212 dxy = (my & 2) | ((mx & 2) >> 1);
1213
1214 if (!v->rnd)
1215 s->hdsp.avg_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
1216 else
1217 s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0], srcY, s->linesize, 16);
1218 }
1219
1220 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
1221 return;
1222 /* Chroma MC always uses qpel bilinear */
1223 uvmx = (uvmx & 3) << 1;
1224 uvmy = (uvmy & 3) << 1;
1225 if (!v->rnd) {
1226 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
1227 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
1228 } else {
1229 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
1230 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
1231 }
1232 }
1233