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
7 * This file is part of FFmpeg.
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.
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.
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
26 * VC-1 and WMV3 block decoding routines
30 #include "h264chroma.h"
32 #include "mpegvideo.h"
35 static av_always_inline void vc1_scale_luma(uint8_t *srcY,
39 for (j = 0; j < k; j++) {
40 for (i = 0; i < k; i++)
41 srcY[i] = ((srcY[i] - 128) >> 1) + 128;
46 static av_always_inline void vc1_scale_chroma(uint8_t *srcU, uint8_t *srcV,
47 int k, int uvlinesize)
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;
60 static av_always_inline void vc1_lut_scale_luma(uint8_t *srcY,
61 uint8_t *lut1, uint8_t *lut2,
66 for (j = 0; j < k; j += 2) {
67 for (i = 0; i < k; i++)
68 srcY[i] = lut1[srcY[i]];
74 for (i = 0; i < k; i++)
75 srcY[i] = lut2[srcY[i]];
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)
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]];
97 for (i = 0; i < k; i++) {
98 srcU[i] = lut2[srcU[i]];
99 srcV[i] = lut2[srcV[i]];
106 /** Do motion compensation over 1 macroblock
107 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
109 void ff_vc1_mc_1mv(VC1Context *v, int dir)
111 MpegEncContext *s = &v->s;
112 H264ChromaContext *h264chroma = &v->h264chroma;
113 uint8_t *srcY, *srcU, *srcV;
114 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
115 int v_edge_pos = s->v_edge_pos >> v->field_mode;
117 uint8_t (*luty)[256], (*lutuv)[256];
120 if ((!v->field_mode ||
121 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
122 !v->s.last_picture.f->data[0])
125 mx = s->mv[dir][0][0];
126 my = s->mv[dir][0][1];
128 // store motion vectors for further use in B frames
129 if (s->pict_type == AV_PICTURE_TYPE_P) {
130 for (i = 0; i < 4; i++) {
131 s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][0] = mx;
132 s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][1] = my;
136 uvmx = (mx + ((mx & 3) == 3)) >> 1;
137 uvmy = (my + ((my & 3) == 3)) >> 1;
138 v->luma_mv[s->mb_x][0] = uvmx;
139 v->luma_mv[s->mb_x][1] = uvmy;
142 v->cur_field_type != v->ref_field_type[dir]) {
143 my = my - 2 + 4 * v->cur_field_type;
144 uvmy = uvmy - 2 + 4 * v->cur_field_type;
147 // fastuvmc shall be ignored for interlaced frame picture
148 if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
149 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
150 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
153 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
154 srcY = s->current_picture.f->data[0];
155 srcU = s->current_picture.f->data[1];
156 srcV = s->current_picture.f->data[2];
158 lutuv = v->curr_lutuv;
159 use_ic = *v->curr_use_ic;
161 srcY = s->last_picture.f->data[0];
162 srcU = s->last_picture.f->data[1];
163 srcV = s->last_picture.f->data[2];
165 lutuv = v->last_lutuv;
166 use_ic = v->last_use_ic;
169 srcY = s->next_picture.f->data[0];
170 srcU = s->next_picture.f->data[1];
171 srcV = s->next_picture.f->data[2];
173 lutuv = v->next_lutuv;
174 use_ic = v->next_use_ic;
177 if (!srcY || !srcU) {
178 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
182 src_x = s->mb_x * 16 + (mx >> 2);
183 src_y = s->mb_y * 16 + (my >> 2);
184 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
185 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
187 if (v->profile != PROFILE_ADVANCED) {
188 src_x = av_clip( src_x, -16, s->mb_width * 16);
189 src_y = av_clip( src_y, -16, s->mb_height * 16);
190 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
191 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
193 src_x = av_clip( src_x, -17, s->avctx->coded_width);
194 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
195 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
196 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
199 srcY += src_y * s->linesize + src_x;
200 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
201 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
203 if (v->field_mode && v->ref_field_type[dir]) {
204 srcY += s->current_picture_ptr->f->linesize[0];
205 srcU += s->current_picture_ptr->f->linesize[1];
206 srcV += s->current_picture_ptr->f->linesize[2];
209 /* for grayscale we should not try to read from unknown area */
210 if (s->flags & CODEC_FLAG_GRAY) {
211 srcU = s->edge_emu_buffer + 18 * s->linesize;
212 srcV = s->edge_emu_buffer + 18 * s->linesize;
215 if (v->rangeredfrm || use_ic
216 || s->h_edge_pos < 22 || v_edge_pos < 22
217 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
218 || (unsigned)(src_y - 1) > v_edge_pos - (my&3) - 16 - 3) {
219 uint8_t *ubuf = s->edge_emu_buffer + 19 * s->linesize;
220 uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
221 const int k = 17 + s->mspel * 2;
223 srcY -= s->mspel * (1 + s->linesize);
224 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
225 s->linesize, s->linesize,
227 src_x - s->mspel, src_y - s->mspel,
228 s->h_edge_pos, v_edge_pos);
229 srcY = s->edge_emu_buffer;
230 s->vdsp.emulated_edge_mc(ubuf, srcU,
231 s->uvlinesize, s->uvlinesize,
234 s->h_edge_pos >> 1, v_edge_pos >> 1);
235 s->vdsp.emulated_edge_mc(vbuf, srcV,
236 s->uvlinesize, s->uvlinesize,
239 s->h_edge_pos >> 1, v_edge_pos >> 1);
242 /* if we deal with range reduction we need to scale source blocks */
243 if (v->rangeredfrm) {
244 vc1_scale_luma(srcY, k, s->linesize);
245 vc1_scale_chroma(srcU, srcV, 9, s->uvlinesize);
247 /* if we deal with intensity compensation we need to scale source blocks */
249 vc1_lut_scale_luma(srcY,
250 luty[v->field_mode ? v->ref_field_type[dir] : ((0 + src_y - s->mspel) & 1)],
251 luty[v->field_mode ? v->ref_field_type[dir] : ((1 + src_y - s->mspel) & 1)],
253 vc1_lut_scale_chroma(srcU, srcV,
254 lutuv[v->field_mode ? v->ref_field_type[dir] : ((0 + uvsrc_y) & 1)],
255 lutuv[v->field_mode ? v->ref_field_type[dir] : ((1 + uvsrc_y) & 1)],
258 srcY += s->mspel * (1 + s->linesize);
262 dxy = ((my & 3) << 2) | (mx & 3);
263 v->vc1dsp.put_vc1_mspel_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, v->rnd);
264 } else { // hpel mc - always used for luma
265 dxy = (my & 2) | ((mx & 2) >> 1);
267 s->hdsp.put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
269 s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
272 if (s->flags & CODEC_FLAG_GRAY) return;
273 /* Chroma MC always uses qpel bilinear */
274 uvmx = (uvmx & 3) << 1;
275 uvmy = (uvmy & 3) << 1;
277 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
278 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
280 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
281 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
285 /** Do motion compensation for 4-MV macroblock - luminance block
287 void ff_vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
289 MpegEncContext *s = &v->s;
291 int dxy, mx, my, src_x, src_y;
293 int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
294 int v_edge_pos = s->v_edge_pos >> v->field_mode;
295 uint8_t (*luty)[256];
298 if ((!v->field_mode ||
299 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
300 !v->s.last_picture.f->data[0])
303 mx = s->mv[dir][n][0];
304 my = s->mv[dir][n][1];
307 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
308 srcY = s->current_picture.f->data[0];
310 use_ic = *v->curr_use_ic;
312 srcY = s->last_picture.f->data[0];
314 use_ic = v->last_use_ic;
317 srcY = s->next_picture.f->data[0];
319 use_ic = v->next_use_ic;
323 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
328 if (v->cur_field_type != v->ref_field_type[dir])
329 my = my - 2 + 4 * v->cur_field_type;
332 if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
333 int same_count = 0, opp_count = 0, k;
334 int chosen_mv[2][4][2], f;
336 for (k = 0; k < 4; k++) {
337 f = v->mv_f[0][s->block_index[k] + v->blocks_off];
338 chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
339 chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
343 f = opp_count > same_count;
344 switch (f ? opp_count : same_count) {
346 tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
347 chosen_mv[f][2][0], chosen_mv[f][3][0]);
348 ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
349 chosen_mv[f][2][1], chosen_mv[f][3][1]);
352 tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
353 ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
356 tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
357 ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
362 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
363 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
364 for (k = 0; k < 4; k++)
365 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
368 if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
370 int width = s->avctx->coded_width;
371 int height = s->avctx->coded_height >> 1;
372 if (s->pict_type == AV_PICTURE_TYPE_P) {
373 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][0] = mx;
374 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][1] = my;
376 qx = (s->mb_x * 16) + (mx >> 2);
377 qy = (s->mb_y * 8) + (my >> 3);
382 mx -= 4 * (qx - width);
385 else if (qy > height + 1)
386 my -= 8 * (qy - height - 1);
389 if ((v->fcm == ILACE_FRAME) && fieldmv)
390 off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
392 off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
394 src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
396 src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
398 src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
400 if (v->profile != PROFILE_ADVANCED) {
401 src_x = av_clip(src_x, -16, s->mb_width * 16);
402 src_y = av_clip(src_y, -16, s->mb_height * 16);
404 src_x = av_clip(src_x, -17, s->avctx->coded_width);
405 if (v->fcm == ILACE_FRAME) {
407 src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
409 src_y = av_clip(src_y, -18, s->avctx->coded_height);
411 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
415 srcY += src_y * s->linesize + src_x;
416 if (v->field_mode && v->ref_field_type[dir])
417 srcY += s->current_picture_ptr->f->linesize[0];
419 if (fieldmv && !(src_y & 1))
421 if (fieldmv && (src_y & 1) && src_y < 4)
423 if (v->rangeredfrm || use_ic
424 || s->h_edge_pos < 13 || v_edge_pos < 23
425 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
426 || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
427 const int k = 9 + s->mspel * 2;
429 srcY -= s->mspel * (1 + (s->linesize << fieldmv));
430 /* check emulate edge stride and offset */
431 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
432 s->linesize, s->linesize,
434 src_x - s->mspel, src_y - (s->mspel << fieldmv),
435 s->h_edge_pos, v_edge_pos);
436 srcY = s->edge_emu_buffer;
437 /* if we deal with range reduction we need to scale source blocks */
438 if (v->rangeredfrm) {
439 vc1_scale_luma(srcY, k, s->linesize << fieldmv);
441 /* if we deal with intensity compensation we need to scale source blocks */
443 vc1_lut_scale_luma(srcY,
444 luty[v->field_mode ? v->ref_field_type[dir] : (((0<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1)],
445 luty[v->field_mode ? v->ref_field_type[dir] : (((1<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1)],
446 k, s->linesize << fieldmv);
448 srcY += s->mspel * (1 + (s->linesize << fieldmv));
452 dxy = ((my & 3) << 2) | (mx & 3);
454 v->vc1dsp.avg_vc1_mspel_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
456 v->vc1dsp.put_vc1_mspel_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
457 } else { // hpel mc - always used for luma
458 dxy = (my & 2) | ((mx & 2) >> 1);
460 s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
462 s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
466 static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
469 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
471 idx = ((a[3] != flag) << 3)
472 | ((a[2] != flag) << 2)
473 | ((a[1] != flag) << 1)
476 *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
477 *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
479 } else if (count[idx] == 1) {
482 *tx = mid_pred(mvx[1], mvx[2], mvx[3]);
483 *ty = mid_pred(mvy[1], mvy[2], mvy[3]);
486 *tx = mid_pred(mvx[0], mvx[2], mvx[3]);
487 *ty = mid_pred(mvy[0], mvy[2], mvy[3]);
490 *tx = mid_pred(mvx[0], mvx[1], mvx[3]);
491 *ty = mid_pred(mvy[0], mvy[1], mvy[3]);
494 *tx = mid_pred(mvx[0], mvx[1], mvx[2]);
495 *ty = mid_pred(mvy[0], mvy[1], mvy[2]);
498 } else if (count[idx] == 2) {
500 for (i = 0; i < 3; i++)
505 for (i = t1 + 1; i < 4; i++)
510 *tx = (mvx[t1] + mvx[t2]) / 2;
511 *ty = (mvy[t1] + mvy[t2]) / 2;
519 /** Do motion compensation for 4-MV macroblock - both chroma blocks
521 void ff_vc1_mc_4mv_chroma(VC1Context *v, int dir)
523 MpegEncContext *s = &v->s;
524 H264ChromaContext *h264chroma = &v->h264chroma;
525 uint8_t *srcU, *srcV;
526 int uvmx, uvmy, uvsrc_x, uvsrc_y;
527 int k, tx = 0, ty = 0;
528 int mvx[4], mvy[4], intra[4], mv_f[4];
530 int chroma_ref_type = v->cur_field_type;
531 int v_edge_pos = s->v_edge_pos >> v->field_mode;
532 uint8_t (*lutuv)[256];
535 if (!v->field_mode && !v->s.last_picture.f->data[0])
537 if (s->flags & CODEC_FLAG_GRAY)
540 for (k = 0; k < 4; k++) {
541 mvx[k] = s->mv[dir][k][0];
542 mvy[k] = s->mv[dir][k][1];
543 intra[k] = v->mb_type[0][s->block_index[k]];
545 mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
548 /* calculate chroma MV vector from four luma MVs */
549 if (!v->field_mode || (v->field_mode && !v->numref)) {
550 valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
551 chroma_ref_type = v->ref_field_type[dir];
553 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
554 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
555 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
556 return; //no need to do MC for intra blocks
560 if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
562 valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
564 chroma_ref_type = !v->cur_field_type;
566 if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f->data[0])
568 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
569 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
570 uvmx = (tx + ((tx & 3) == 3)) >> 1;
571 uvmy = (ty + ((ty & 3) == 3)) >> 1;
573 v->luma_mv[s->mb_x][0] = uvmx;
574 v->luma_mv[s->mb_x][1] = uvmy;
577 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
578 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
580 // Field conversion bias
581 if (v->cur_field_type != chroma_ref_type)
582 uvmy += 2 - 4 * chroma_ref_type;
584 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
585 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
587 if (v->profile != PROFILE_ADVANCED) {
588 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
589 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
591 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
592 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
596 if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
597 srcU = s->current_picture.f->data[1];
598 srcV = s->current_picture.f->data[2];
599 lutuv = v->curr_lutuv;
600 use_ic = *v->curr_use_ic;
602 srcU = s->last_picture.f->data[1];
603 srcV = s->last_picture.f->data[2];
604 lutuv = v->last_lutuv;
605 use_ic = v->last_use_ic;
608 srcU = s->next_picture.f->data[1];
609 srcV = s->next_picture.f->data[2];
610 lutuv = v->next_lutuv;
611 use_ic = v->next_use_ic;
615 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
619 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
620 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
623 if (chroma_ref_type) {
624 srcU += s->current_picture_ptr->f->linesize[1];
625 srcV += s->current_picture_ptr->f->linesize[2];
629 if (v->rangeredfrm || use_ic
630 || s->h_edge_pos < 18 || v_edge_pos < 18
631 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
632 || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) {
633 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
634 s->uvlinesize, s->uvlinesize,
635 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
636 s->h_edge_pos >> 1, v_edge_pos >> 1);
637 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
638 s->uvlinesize, s->uvlinesize,
639 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
640 s->h_edge_pos >> 1, v_edge_pos >> 1);
641 srcU = s->edge_emu_buffer;
642 srcV = s->edge_emu_buffer + 16;
644 /* if we deal with range reduction we need to scale source blocks */
645 if (v->rangeredfrm) {
646 vc1_scale_chroma(srcU, srcV, 9, s->uvlinesize);
648 /* if we deal with intensity compensation we need to scale source blocks */
650 vc1_lut_scale_chroma(srcU, srcV,
651 lutuv[v->field_mode ? chroma_ref_type : ((0 + uvsrc_y) & 1)],
652 lutuv[v->field_mode ? chroma_ref_type : ((1 + uvsrc_y) & 1)],
657 /* Chroma MC always uses qpel bilinear */
658 uvmx = (uvmx & 3) << 1;
659 uvmy = (uvmy & 3) << 1;
661 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
662 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
664 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
665 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
669 /** Do motion compensation for 4-MV interlaced frame chroma macroblock (both U and V)
671 void ff_vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg)
673 MpegEncContext *s = &v->s;
674 H264ChromaContext *h264chroma = &v->h264chroma;
675 uint8_t *srcU, *srcV;
676 int uvsrc_x, uvsrc_y;
677 int uvmx_field[4], uvmy_field[4];
679 int fieldmv = v->blk_mv_type[s->block_index[0]];
680 static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
681 int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
682 int v_edge_pos = s->v_edge_pos >> 1;
684 uint8_t (*lutuv)[256];
686 if (s->flags & CODEC_FLAG_GRAY)
689 for (i = 0; i < 4; i++) {
690 int d = i < 2 ? dir: dir2;
692 uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
695 uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
697 uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
700 for (i = 0; i < 4; i++) {
701 off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
702 uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2);
703 uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
704 // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
705 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
706 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
707 if (i < 2 ? dir : dir2) {
708 srcU = s->next_picture.f->data[1];
709 srcV = s->next_picture.f->data[2];
710 lutuv = v->next_lutuv;
711 use_ic = v->next_use_ic;
713 srcU = s->last_picture.f->data[1];
714 srcV = s->last_picture.f->data[2];
715 lutuv = v->last_lutuv;
716 use_ic = v->last_use_ic;
720 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
721 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
722 uvmx_field[i] = (uvmx_field[i] & 3) << 1;
723 uvmy_field[i] = (uvmy_field[i] & 3) << 1;
725 if (fieldmv && !(uvsrc_y & 1))
726 v_edge_pos = (s->v_edge_pos >> 1) - 1;
728 if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
731 || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
732 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
733 || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
734 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
735 s->uvlinesize, s->uvlinesize,
736 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
737 s->h_edge_pos >> 1, v_edge_pos);
738 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
739 s->uvlinesize, s->uvlinesize,
740 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
741 s->h_edge_pos >> 1, v_edge_pos);
742 srcU = s->edge_emu_buffer;
743 srcV = s->edge_emu_buffer + 16;
745 /* if we deal with intensity compensation we need to scale source blocks */
747 vc1_lut_scale_chroma(srcU, srcV,
748 lutuv[(uvsrc_y + (0 << fieldmv)) & 1],
749 lutuv[(uvsrc_y + (1 << fieldmv)) & 1],
750 5, s->uvlinesize << fieldmv);
755 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
756 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
758 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]);
759 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]);
763 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
764 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
766 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]);
767 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]);
773 /** Motion compensation for direct or interpolated blocks in B-frames
775 void ff_vc1_interp_mc(VC1Context *v)
777 MpegEncContext *s = &v->s;
778 H264ChromaContext *h264chroma = &v->h264chroma;
779 uint8_t *srcY, *srcU, *srcV;
780 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
782 int v_edge_pos = s->v_edge_pos >> v->field_mode;
783 int use_ic = v->next_use_ic;
785 if (!v->field_mode && !v->s.next_picture.f->data[0])
790 uvmx = (mx + ((mx & 3) == 3)) >> 1;
791 uvmy = (my + ((my & 3) == 3)) >> 1;
792 if (v->field_mode && v->cur_field_type != v->ref_field_type[1]) {
793 my = my - 2 + 4 * v->cur_field_type;
794 uvmy = uvmy - 2 + 4 * v->cur_field_type;
797 uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
798 uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
800 srcY = s->next_picture.f->data[0];
801 srcU = s->next_picture.f->data[1];
802 srcV = s->next_picture.f->data[2];
804 src_x = s->mb_x * 16 + (mx >> 2);
805 src_y = s->mb_y * 16 + (my >> 2);
806 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
807 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
809 if (v->profile != PROFILE_ADVANCED) {
810 src_x = av_clip( src_x, -16, s->mb_width * 16);
811 src_y = av_clip( src_y, -16, s->mb_height * 16);
812 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
813 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
815 src_x = av_clip( src_x, -17, s->avctx->coded_width);
816 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
817 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
818 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
821 srcY += src_y * s->linesize + src_x;
822 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
823 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
825 if (v->field_mode && v->ref_field_type[1]) {
826 srcY += s->current_picture_ptr->f->linesize[0];
827 srcU += s->current_picture_ptr->f->linesize[1];
828 srcV += s->current_picture_ptr->f->linesize[2];
831 /* for grayscale we should not try to read from unknown area */
832 if (s->flags & CODEC_FLAG_GRAY) {
833 srcU = s->edge_emu_buffer + 18 * s->linesize;
834 srcV = s->edge_emu_buffer + 18 * s->linesize;
837 if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22 || use_ic
838 || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
839 || (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) {
840 uint8_t *ubuf = s->edge_emu_buffer + 19 * s->linesize;
841 uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
842 const int k = 17 + s->mspel * 2;
844 srcY -= s->mspel * (1 + s->linesize);
845 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
846 s->linesize, s->linesize,
848 src_x - s->mspel, src_y - s->mspel,
849 s->h_edge_pos, v_edge_pos);
850 srcY = s->edge_emu_buffer;
851 s->vdsp.emulated_edge_mc(ubuf, srcU,
852 s->uvlinesize, s->uvlinesize,
855 s->h_edge_pos >> 1, v_edge_pos >> 1);
856 s->vdsp.emulated_edge_mc(vbuf, srcV,
857 s->uvlinesize, s->uvlinesize,
860 s->h_edge_pos >> 1, v_edge_pos >> 1);
863 /* if we deal with range reduction we need to scale source blocks */
864 if (v->rangeredfrm) {
865 vc1_scale_luma(srcY, k, s->linesize);
866 vc1_scale_chroma(srcU, srcV, 9, s->uvlinesize);
870 uint8_t (*luty )[256] = v->next_luty;
871 uint8_t (*lutuv)[256] = v->next_lutuv;
872 vc1_lut_scale_luma(srcY,
873 luty[v->field_mode ? v->ref_field_type[1] : ((0+src_y - s->mspel) & 1)],
874 luty[v->field_mode ? v->ref_field_type[1] : ((1+src_y - s->mspel) & 1)],
876 vc1_lut_scale_chroma(srcU, srcV,
877 lutuv[v->field_mode ? v->ref_field_type[1] : ((0+uvsrc_y) & 1)],
878 lutuv[v->field_mode ? v->ref_field_type[1] : ((1+uvsrc_y) & 1)],
881 srcY += s->mspel * (1 + s->linesize);
888 dxy = ((my & 3) << 2) | (mx & 3);
889 v->vc1dsp.avg_vc1_mspel_pixels_tab[0][dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
891 dxy = (my & 2) | ((mx & 2) >> 1);
894 s->hdsp.avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
896 s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, 16);
899 if (s->flags & CODEC_FLAG_GRAY) return;
900 /* Chroma MC always uses qpel blilinear */
901 uvmx = (uvmx & 3) << 1;
902 uvmy = (uvmy & 3) << 1;
904 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
905 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
907 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
908 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);