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 static const uint8_t popcount4[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4 };
108 static av_always_inline int get_luma_mv(VC1Context *v, int dir, int16_t *tx, int16_t *ty)
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];
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]);
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]);
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]);
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;
140 static av_always_inline int get_chroma_mv(VC1Context *v, int dir, int16_t *tx, int16_t *ty)
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];
150 switch (valid_count) {
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]);
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]);
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;
169 /** Do motion compensation over 1 macroblock
170 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
172 void ff_vc1_mc_1mv(VC1Context *v, int dir)
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;
180 uint8_t (*luty)[256], (*lutuv)[256];
183 int linesize, uvlinesize;
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])
190 linesize = s->current_picture_ptr->f->linesize[0];
191 uvlinesize = s->current_picture_ptr->f->linesize[1];
193 mx = s->mv[dir][0][0];
194 my = s->mv[dir][0][1];
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;
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;
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;
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));
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];
226 lutuv = v->curr_lutuv;
227 use_ic = *v->curr_use_ic;
230 srcY = s->last_picture.f->data[0];
231 srcU = s->last_picture.f->data[1];
232 srcV = s->last_picture.f->data[2];
234 lutuv = v->last_lutuv;
235 use_ic = v->last_use_ic;
236 interlace = s->last_picture.f->interlaced_frame;
239 srcY = s->next_picture.f->data[0];
240 srcU = s->next_picture.f->data[1];
241 srcV = s->next_picture.f->data[2];
243 lutuv = v->next_lutuv;
244 use_ic = v->next_use_ic;
245 interlace = s->next_picture.f->interlaced_frame;
248 if (!srcY || !srcU) {
249 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
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);
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);
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));
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);
275 srcY += src_y * s->linesize + src_x;
276 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
277 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
279 if (v->field_mode && v->ref_field_type[dir]) {
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;
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;
299 srcY -= s->mspel * (1 + s->linesize);
301 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
306 v->field_mode ? k : k + 1 >> 1,
308 src_y - s->mspel >> !v->field_mode,
312 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + linesize,
319 src_y - s->mspel + 1 >> 1,
323 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
328 v->field_mode ? (k << 1) - 1 : k,
330 v->field_mode ? 2 * (src_y - s->mspel) + v->ref_field_type[dir] :
334 srcY = s->sc.edge_emu_buffer;
336 s->vdsp.emulated_edge_mc(ubuf,
341 v->field_mode ? 9 : 5,
343 uvsrc_y >> !v->field_mode,
346 s->vdsp.emulated_edge_mc(vbuf,
351 v->field_mode ? 9 : 5,
353 uvsrc_y >> !v->field_mode,
356 if (!v->field_mode) {
357 s->vdsp.emulated_edge_mc(ubuf + uvlinesize,
367 s->vdsp.emulated_edge_mc(vbuf + uvlinesize,
379 s->vdsp.emulated_edge_mc(ubuf,
384 v->field_mode ? 17 : 9,
386 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[dir] : uvsrc_y,
389 s->vdsp.emulated_edge_mc(vbuf,
394 v->field_mode ? 17 : 9,
396 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[dir] : uvsrc_y,
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);
407 /* if we deal with intensity compensation we need to scale source blocks */
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)],
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)],
418 srcY += s->mspel * (1 + s->linesize);
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);
427 s->hdsp.put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
429 s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
432 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
434 /* Chroma MC always uses qpel bilinear */
435 uvmx = (uvmx & 3) << 1;
436 uvmy = (uvmy & 3) << 1;
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);
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);
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];
450 /** Do motion compensation for 4-MV macroblock - luminance block
452 void ff_vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
454 MpegEncContext *s = &v->s;
456 int dxy, mx, my, src_x, src_y;
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];
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])
470 linesize = s->current_picture_ptr->f->linesize[0];
472 mx = s->mv[dir][n][0];
473 my = s->mv[dir][n][1];
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];
479 use_ic = *v->curr_use_ic;
482 srcY = s->last_picture.f->data[0];
484 use_ic = v->last_use_ic;
485 interlace = s->last_picture.f->interlaced_frame;
488 srcY = s->next_picture.f->data[0];
490 use_ic = v->next_use_ic;
491 interlace = s->next_picture.f->interlaced_frame;
495 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
500 if (v->cur_field_type != v->ref_field_type[dir])
501 my = my - 2 + 4 * v->cur_field_type;
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;
513 if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
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;
521 qx = (s->mb_x * 16) + (mx >> 2);
522 qy = (s->mb_y * 8) + (my >> 3);
527 mx -= 4 * (qx - width);
530 else if (qy > height + 1)
531 my -= 8 * (qy - height - 1);
534 if ((v->fcm == ILACE_FRAME) && fieldmv)
535 off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
537 off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
539 src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
541 src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
543 src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
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);
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));
553 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
556 srcY += src_y * s->linesize + src_x;
557 if (v->field_mode && v->ref_field_type[dir])
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;
566 srcY -= s->mspel * (1 + (s->linesize << fieldmv));
567 /* check emulate edge stride and offset */
569 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
574 v->field_mode ? k : (k << fieldmv) + 1 >> 1,
576 src_y - (s->mspel << fieldmv) >> !v->field_mode,
579 if (!v->field_mode && !fieldmv)
580 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + linesize,
587 src_y - s->mspel + 1 >> 1,
591 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
596 v->field_mode ? (k << 1) - 1 : k << fieldmv,
598 v->field_mode ? 2 * (src_y - s->mspel) + v->ref_field_type[dir] :
599 src_y - (s->mspel << fieldmv),
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);
607 /* if we deal with intensity compensation we need to scale source blocks */
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);
614 srcY += s->mspel * (1 + (s->linesize << fieldmv));
618 dxy = ((my & 3) << 2) | (mx & 3);
620 v->vc1dsp.avg_vc1_mspel_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
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);
626 s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
628 s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
632 /** Do motion compensation for 4-MV macroblock - both chroma blocks
634 void ff_vc1_mc_4mv_chroma(VC1Context *v, int dir)
636 MpegEncContext *s = &v->s;
637 H264ChromaContext *h264chroma = &v->h264chroma;
638 uint8_t *srcU, *srcV;
639 int uvmx, uvmy, uvsrc_x, uvsrc_y;
642 int v_edge_pos = s->v_edge_pos >> v->field_mode;
643 uint8_t (*lutuv)[256];
648 if (!v->field_mode && !v->s.last_picture.f->data[0])
650 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
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);
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
662 chroma_ref_type = v->ref_field_type[dir];
664 int opp_count = get_luma_mv(v, dir, &tx, &ty);
665 chroma_ref_type = v->cur_field_type ^ (opp_count > 2);
667 if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f->data[0])
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;
672 uvlinesize = s->current_picture_ptr->f->linesize[1];
674 uvmx = (tx + ((tx & 3) == 3)) >> 1;
675 uvmy = (ty + ((ty & 3) == 3)) >> 1;
677 v->luma_mv[s->mb_x][0] = uvmx;
678 v->luma_mv[s->mb_x][1] = uvmy;
681 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
682 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
684 // Field conversion bias
685 if (v->cur_field_type != chroma_ref_type)
686 uvmy += 2 - 4 * chroma_ref_type;
688 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
689 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
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);
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);
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;
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;
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;
722 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
726 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
727 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
730 if (chroma_ref_type) {
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) {
741 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
746 v->field_mode ? 9 : 5,
748 uvsrc_y >> !v->field_mode,
751 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
756 v->field_mode ? 9 : 5,
758 uvsrc_y >> !v->field_mode,
761 if (!v->field_mode) {
762 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + uvlinesize,
772 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16 + uvlinesize,
784 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
789 v->field_mode ? 17 : 9,
791 v->field_mode ? 2 * uvsrc_y + chroma_ref_type : uvsrc_y,
794 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
799 v->field_mode ? 17 : 9,
801 v->field_mode ? 2 * uvsrc_y + chroma_ref_type : uvsrc_y,
805 srcU = s->sc.edge_emu_buffer;
806 srcV = s->sc.edge_emu_buffer + 16;
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);
812 /* if we deal with intensity compensation we need to scale source blocks */
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)],
821 /* Chroma MC always uses qpel bilinear */
822 uvmx = (uvmx & 3) << 1;
823 uvmy = (uvmy & 3) << 1;
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);
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);
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;
837 /** Do motion compensation for 4-MV interlaced frame chroma macroblock (both U and V)
839 void ff_vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg)
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];
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;
854 uint8_t (*lutuv)[256];
856 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
859 uvlinesize = s->current_picture_ptr->f->linesize[1];
861 for (i = 0; i < 4; i++) {
862 int d = i < 2 ? dir: dir2;
864 uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
867 uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
869 uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
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));
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;
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;
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;
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)) {
907 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
912 (5 << fieldmv) + 1 >> 1,
917 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
922 (5 << fieldmv) + 1 >> 1,
928 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + uvlinesize,
938 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16 + uvlinesize,
950 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
960 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + 16,
971 srcU = s->sc.edge_emu_buffer;
972 srcV = s->sc.edge_emu_buffer + 16;
974 /* if we deal with intensity compensation we need to scale source blocks */
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);
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]);
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]);
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]);
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]);
1002 /** Motion compensation for direct or interpolated blocks in B-frames
1004 void ff_vc1_interp_mc(VC1Context *v)
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;
1013 int linesize, uvlinesize;
1015 if (!v->field_mode && !v->s.next_picture.f->data[0])
1018 linesize = s->current_picture_ptr->f->linesize[0];
1019 uvlinesize = s->current_picture_ptr->f->linesize[1];
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;
1030 uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
1031 uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
1033 srcY = s->next_picture.f->data[0];
1034 srcU = s->next_picture.f->data[1];
1035 srcV = s->next_picture.f->data[2];
1037 interlace = s->next_picture.f->interlaced_frame;
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);
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);
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));
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);
1061 srcY += src_y * s->linesize + src_x;
1062 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1063 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1065 if (v->field_mode && v->ref_field_type[1]) {
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;
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;
1084 srcY -= s->mspel * (1 + s->linesize);
1086 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
1091 v->field_mode ? k : (k + 1 >> 1),
1093 src_y - s->mspel >> !v->field_mode,
1095 s->v_edge_pos >> 1);
1097 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer + linesize,
1104 src_y - s->mspel + 1 >> 1,
1106 s->v_edge_pos >> 1);
1108 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer,
1113 v->field_mode ? (k << 1) - 1 : k,
1115 v->field_mode ? 2 * (src_y - s->mspel) + v->ref_field_type[1] :
1119 srcY = s->sc.edge_emu_buffer;
1121 s->vdsp.emulated_edge_mc(ubuf,
1126 v->field_mode ? 9 : 5,
1128 uvsrc_y >> !v->field_mode,
1130 s->v_edge_pos >> 2);
1131 s->vdsp.emulated_edge_mc(vbuf,
1136 v->field_mode ? 9 : 5,
1138 uvsrc_y >> !v->field_mode,
1140 s->v_edge_pos >> 2);
1141 if (!v->field_mode) {
1142 s->vdsp.emulated_edge_mc(ubuf + uvlinesize,
1151 s->v_edge_pos >> 2);
1152 s->vdsp.emulated_edge_mc(vbuf + uvlinesize,
1161 s->v_edge_pos >> 2);
1164 s->vdsp.emulated_edge_mc(ubuf,
1169 v->field_mode ? 17 : 9,
1171 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[1] : uvsrc_y,
1173 s->v_edge_pos >> 1);
1174 s->vdsp.emulated_edge_mc(vbuf,
1179 v->field_mode ? 17 : 9,
1181 v->field_mode ? 2 * uvsrc_y + v->ref_field_type[1] : uvsrc_y,
1183 s->v_edge_pos >> 1);
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);
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)],
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)],
1205 srcY += s->mspel * (1 + s->linesize);
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);
1212 dxy = (my & 2) | ((mx & 2) >> 1);
1215 s->hdsp.avg_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
1217 s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0], srcY, s->linesize, 16);
1220 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
1222 /* Chroma MC always uses qpel bilinear */
1223 uvmx = (uvmx & 3) << 1;
1224 uvmy = (uvmy & 3) << 1;
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);
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);