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 decoder
31 #include "error_resilience.h"
32 #include "mpegutils.h"
33 #include "mpegvideo.h"
35 #include "h264chroma.h"
38 #include "vc1acdata.h"
39 #include "msmpeg4data.h"
42 #include "vdpau_internal.h"
43 #include "libavutil/avassert.h"
48 #define MB_INTRA_VLC_BITS 9
52 // offset tables for interlaced picture MVDATA decoding
53 static const int offset_table1[9] = { 0, 1, 2, 4, 8, 16, 32, 64, 128 };
54 static const int offset_table2[9] = { 0, 1, 3, 7, 15, 31, 63, 127, 255 };
56 /***********************************************************************/
58 * @name VC-1 Bitplane decoding
64 static void init_block_index(VC1Context *v)
66 MpegEncContext *s = &v->s;
67 ff_init_block_index(s);
68 if (v->field_mode && !(v->second_field ^ v->tff)) {
69 s->dest[0] += s->current_picture_ptr->f->linesize[0];
70 s->dest[1] += s->current_picture_ptr->f->linesize[1];
71 s->dest[2] += s->current_picture_ptr->f->linesize[2];
75 /** @} */ //Bitplane group
77 static void vc1_put_signed_blocks_clamped(VC1Context *v)
79 MpegEncContext *s = &v->s;
80 int topleft_mb_pos, top_mb_pos;
81 int stride_y, fieldtx = 0;
84 /* The put pixels loop is always one MB row behind the decoding loop,
85 * because we can only put pixels when overlap filtering is done, and
86 * for filtering of the bottom edge of a MB, we need the next MB row
88 * Within the row, the put pixels loop is also one MB col behind the
89 * decoding loop. The reason for this is again, because for filtering
90 * of the right MB edge, we need the next MB present. */
91 if (!s->first_slice_line) {
93 topleft_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x - 1;
94 if (v->fcm == ILACE_FRAME)
95 fieldtx = v->fieldtx_plane[topleft_mb_pos];
96 stride_y = s->linesize << fieldtx;
97 v_dist = (16 - fieldtx) >> (fieldtx == 0);
98 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][0],
99 s->dest[0] - 16 * s->linesize - 16,
101 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][1],
102 s->dest[0] - 16 * s->linesize - 8,
104 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][2],
105 s->dest[0] - v_dist * s->linesize - 16,
107 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3],
108 s->dest[0] - v_dist * s->linesize - 8,
110 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4],
111 s->dest[1] - 8 * s->uvlinesize - 8,
113 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][5],
114 s->dest[2] - 8 * s->uvlinesize - 8,
117 if (s->mb_x == s->mb_width - 1) {
118 top_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x;
119 if (v->fcm == ILACE_FRAME)
120 fieldtx = v->fieldtx_plane[top_mb_pos];
121 stride_y = s->linesize << fieldtx;
122 v_dist = fieldtx ? 15 : 8;
123 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][0],
124 s->dest[0] - 16 * s->linesize,
126 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][1],
127 s->dest[0] - 16 * s->linesize + 8,
129 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][2],
130 s->dest[0] - v_dist * s->linesize,
132 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3],
133 s->dest[0] - v_dist * s->linesize + 8,
135 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4],
136 s->dest[1] - 8 * s->uvlinesize,
138 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][5],
139 s->dest[2] - 8 * s->uvlinesize,
144 #define inc_blk_idx(idx) do { \
146 if (idx >= v->n_allocated_blks) \
150 inc_blk_idx(v->topleft_blk_idx);
151 inc_blk_idx(v->top_blk_idx);
152 inc_blk_idx(v->left_blk_idx);
153 inc_blk_idx(v->cur_blk_idx);
156 static void vc1_loop_filter_iblk(VC1Context *v, int pq)
158 MpegEncContext *s = &v->s;
160 if (!s->first_slice_line) {
161 v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
163 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
164 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
165 for (j = 0; j < 2; j++) {
166 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1], s->uvlinesize, pq);
168 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
171 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8 * s->linesize, s->linesize, pq);
173 if (s->mb_y == s->end_mb_y - 1) {
175 v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
176 v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
177 v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
179 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
183 static void vc1_loop_filter_iblk_delayed(VC1Context *v, int pq)
185 MpegEncContext *s = &v->s;
188 /* The loopfilter runs 1 row and 1 column behind the overlap filter, which
189 * means it runs two rows/cols behind the decoding loop. */
190 if (!s->first_slice_line) {
192 if (s->mb_y >= s->start_mb_y + 2) {
193 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
196 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
197 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
198 for (j = 0; j < 2; j++) {
199 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
201 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq);
205 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq);
208 if (s->mb_x == s->mb_width - 1) {
209 if (s->mb_y >= s->start_mb_y + 2) {
210 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
213 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
214 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
215 for (j = 0; j < 2; j++) {
216 v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
218 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize, s->uvlinesize, pq);
222 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq);
225 if (s->mb_y == s->end_mb_y) {
228 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
229 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
231 for (j = 0; j < 2; j++) {
232 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
237 if (s->mb_x == s->mb_width - 1) {
239 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
240 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
242 for (j = 0; j < 2; j++) {
243 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
251 static void vc1_smooth_overlap_filter_iblk(VC1Context *v)
253 MpegEncContext *s = &v->s;
256 if (v->condover == CONDOVER_NONE)
259 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
261 /* Within a MB, the horizontal overlap always runs before the vertical.
262 * To accomplish that, we run the H on left and internal borders of the
263 * currently decoded MB. Then, we wait for the next overlap iteration
264 * to do H overlap on the right edge of this MB, before moving over and
265 * running the V overlap. Therefore, the V overlap makes us trail by one
266 * MB col and the H overlap filter makes us trail by one MB row. This
267 * is reflected in the time at which we run the put_pixels loop. */
268 if (v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) {
269 if (s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
270 v->over_flags_plane[mb_pos - 1])) {
271 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1],
272 v->block[v->cur_blk_idx][0]);
273 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
274 v->block[v->cur_blk_idx][2]);
275 if (!(s->flags & CODEC_FLAG_GRAY)) {
276 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
277 v->block[v->cur_blk_idx][4]);
278 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
279 v->block[v->cur_blk_idx][5]);
282 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0],
283 v->block[v->cur_blk_idx][1]);
284 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2],
285 v->block[v->cur_blk_idx][3]);
287 if (s->mb_x == s->mb_width - 1) {
288 if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
289 v->over_flags_plane[mb_pos - s->mb_stride])) {
290 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2],
291 v->block[v->cur_blk_idx][0]);
292 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
293 v->block[v->cur_blk_idx][1]);
294 if (!(s->flags & CODEC_FLAG_GRAY)) {
295 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
296 v->block[v->cur_blk_idx][4]);
297 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
298 v->block[v->cur_blk_idx][5]);
301 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0],
302 v->block[v->cur_blk_idx][2]);
303 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1],
304 v->block[v->cur_blk_idx][3]);
307 if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) {
308 if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
309 v->over_flags_plane[mb_pos - s->mb_stride - 1])) {
310 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2],
311 v->block[v->left_blk_idx][0]);
312 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
313 v->block[v->left_blk_idx][1]);
314 if (!(s->flags & CODEC_FLAG_GRAY)) {
315 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
316 v->block[v->left_blk_idx][4]);
317 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
318 v->block[v->left_blk_idx][5]);
321 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0],
322 v->block[v->left_blk_idx][2]);
323 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1],
324 v->block[v->left_blk_idx][3]);
328 /** Do motion compensation over 1 macroblock
329 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
331 static void vc1_mc_1mv(VC1Context *v, int dir)
333 MpegEncContext *s = &v->s;
334 H264ChromaContext *h264chroma = &v->h264chroma;
335 uint8_t *srcY, *srcU, *srcV;
336 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
337 int v_edge_pos = s->v_edge_pos >> v->field_mode;
339 uint8_t (*luty)[256], (*lutuv)[256];
342 if ((!v->field_mode ||
343 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
344 !v->s.last_picture.f->data[0])
347 mx = s->mv[dir][0][0];
348 my = s->mv[dir][0][1];
350 // store motion vectors for further use in B frames
351 if (s->pict_type == AV_PICTURE_TYPE_P) {
352 for (i = 0; i < 4; i++) {
353 s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][0] = mx;
354 s->current_picture.motion_val[1][s->block_index[i] + v->blocks_off][1] = my;
358 uvmx = (mx + ((mx & 3) == 3)) >> 1;
359 uvmy = (my + ((my & 3) == 3)) >> 1;
360 v->luma_mv[s->mb_x][0] = uvmx;
361 v->luma_mv[s->mb_x][1] = uvmy;
364 v->cur_field_type != v->ref_field_type[dir]) {
365 my = my - 2 + 4 * v->cur_field_type;
366 uvmy = uvmy - 2 + 4 * v->cur_field_type;
369 // fastuvmc shall be ignored for interlaced frame picture
370 if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
371 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
372 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
375 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
376 srcY = s->current_picture.f->data[0];
377 srcU = s->current_picture.f->data[1];
378 srcV = s->current_picture.f->data[2];
380 lutuv = v->curr_lutuv;
381 use_ic = *v->curr_use_ic;
383 srcY = s->last_picture.f->data[0];
384 srcU = s->last_picture.f->data[1];
385 srcV = s->last_picture.f->data[2];
387 lutuv = v->last_lutuv;
388 use_ic = v->last_use_ic;
391 srcY = s->next_picture.f->data[0];
392 srcU = s->next_picture.f->data[1];
393 srcV = s->next_picture.f->data[2];
395 lutuv = v->next_lutuv;
396 use_ic = v->next_use_ic;
399 if (!srcY || !srcU) {
400 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
404 src_x = s->mb_x * 16 + (mx >> 2);
405 src_y = s->mb_y * 16 + (my >> 2);
406 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
407 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
409 if (v->profile != PROFILE_ADVANCED) {
410 src_x = av_clip( src_x, -16, s->mb_width * 16);
411 src_y = av_clip( src_y, -16, s->mb_height * 16);
412 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
413 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
415 src_x = av_clip( src_x, -17, s->avctx->coded_width);
416 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
417 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
418 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
421 srcY += src_y * s->linesize + src_x;
422 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
423 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
425 if (v->field_mode && v->ref_field_type[dir]) {
426 srcY += s->current_picture_ptr->f->linesize[0];
427 srcU += s->current_picture_ptr->f->linesize[1];
428 srcV += s->current_picture_ptr->f->linesize[2];
431 /* for grayscale we should not try to read from unknown area */
432 if (s->flags & CODEC_FLAG_GRAY) {
433 srcU = s->edge_emu_buffer + 18 * s->linesize;
434 srcV = s->edge_emu_buffer + 18 * s->linesize;
437 if (v->rangeredfrm || use_ic
438 || s->h_edge_pos < 22 || v_edge_pos < 22
439 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
440 || (unsigned)(src_y - 1) > v_edge_pos - (my&3) - 16 - 3) {
441 uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
443 srcY -= s->mspel * (1 + s->linesize);
444 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
445 s->linesize, s->linesize,
446 17 + s->mspel * 2, 17 + s->mspel * 2,
447 src_x - s->mspel, src_y - s->mspel,
448 s->h_edge_pos, v_edge_pos);
449 srcY = s->edge_emu_buffer;
450 s->vdsp.emulated_edge_mc(uvbuf, srcU,
451 s->uvlinesize, s->uvlinesize,
454 s->h_edge_pos >> 1, v_edge_pos >> 1);
455 s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
456 s->uvlinesize, s->uvlinesize,
459 s->h_edge_pos >> 1, v_edge_pos >> 1);
462 /* if we deal with range reduction we need to scale source blocks */
463 if (v->rangeredfrm) {
468 for (j = 0; j < 17 + s->mspel * 2; j++) {
469 for (i = 0; i < 17 + s->mspel * 2; i++)
470 src[i] = ((src[i] - 128) >> 1) + 128;
475 for (j = 0; j < 9; j++) {
476 for (i = 0; i < 9; i++) {
477 src[i] = ((src[i] - 128) >> 1) + 128;
478 src2[i] = ((src2[i] - 128) >> 1) + 128;
480 src += s->uvlinesize;
481 src2 += s->uvlinesize;
484 /* if we deal with intensity compensation we need to scale source blocks */
490 for (j = 0; j < 17 + s->mspel * 2; j++) {
491 int f = v->field_mode ? v->ref_field_type[dir] : ((j + src_y - s->mspel) & 1) ;
492 for (i = 0; i < 17 + s->mspel * 2; i++)
493 src[i] = luty[f][src[i]];
498 for (j = 0; j < 9; j++) {
499 int f = v->field_mode ? v->ref_field_type[dir] : ((j + uvsrc_y) & 1);
500 for (i = 0; i < 9; i++) {
501 src[i] = lutuv[f][src[i]];
502 src2[i] = lutuv[f][src2[i]];
504 src += s->uvlinesize;
505 src2 += s->uvlinesize;
508 srcY += s->mspel * (1 + s->linesize);
512 dxy = ((my & 3) << 2) | (mx & 3);
513 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
514 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
515 srcY += s->linesize * 8;
516 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
517 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
518 } else { // hpel mc - always used for luma
519 dxy = (my & 2) | ((mx & 2) >> 1);
521 s->hdsp.put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
523 s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
526 if (s->flags & CODEC_FLAG_GRAY) return;
527 /* Chroma MC always uses qpel bilinear */
528 uvmx = (uvmx & 3) << 1;
529 uvmy = (uvmy & 3) << 1;
531 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
532 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
534 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
535 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
539 static inline int median4(int a, int b, int c, int d)
542 if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
543 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
545 if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
546 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
550 /** Do motion compensation for 4-MV macroblock - luminance block
552 static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
554 MpegEncContext *s = &v->s;
556 int dxy, mx, my, src_x, src_y;
558 int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
559 int v_edge_pos = s->v_edge_pos >> v->field_mode;
560 uint8_t (*luty)[256];
563 if ((!v->field_mode ||
564 (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
565 !v->s.last_picture.f->data[0])
568 mx = s->mv[dir][n][0];
569 my = s->mv[dir][n][1];
572 if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
573 srcY = s->current_picture.f->data[0];
575 use_ic = *v->curr_use_ic;
577 srcY = s->last_picture.f->data[0];
579 use_ic = v->last_use_ic;
582 srcY = s->next_picture.f->data[0];
584 use_ic = v->next_use_ic;
588 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
593 if (v->cur_field_type != v->ref_field_type[dir])
594 my = my - 2 + 4 * v->cur_field_type;
597 if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
598 int same_count = 0, opp_count = 0, k;
599 int chosen_mv[2][4][2], f;
601 for (k = 0; k < 4; k++) {
602 f = v->mv_f[0][s->block_index[k] + v->blocks_off];
603 chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
604 chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
608 f = opp_count > same_count;
609 switch (f ? opp_count : same_count) {
611 tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
612 chosen_mv[f][2][0], chosen_mv[f][3][0]);
613 ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
614 chosen_mv[f][2][1], chosen_mv[f][3][1]);
617 tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
618 ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
621 tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
622 ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
627 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
628 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
629 for (k = 0; k < 4; k++)
630 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
633 if (v->fcm == ILACE_FRAME) { // not sure if needed for other types of picture
635 int width = s->avctx->coded_width;
636 int height = s->avctx->coded_height >> 1;
637 if (s->pict_type == AV_PICTURE_TYPE_P) {
638 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][0] = mx;
639 s->current_picture.motion_val[1][s->block_index[n] + v->blocks_off][1] = my;
641 qx = (s->mb_x * 16) + (mx >> 2);
642 qy = (s->mb_y * 8) + (my >> 3);
647 mx -= 4 * (qx - width);
650 else if (qy > height + 1)
651 my -= 8 * (qy - height - 1);
654 if ((v->fcm == ILACE_FRAME) && fieldmv)
655 off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
657 off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
659 src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
661 src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
663 src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
665 if (v->profile != PROFILE_ADVANCED) {
666 src_x = av_clip(src_x, -16, s->mb_width * 16);
667 src_y = av_clip(src_y, -16, s->mb_height * 16);
669 src_x = av_clip(src_x, -17, s->avctx->coded_width);
670 if (v->fcm == ILACE_FRAME) {
672 src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
674 src_y = av_clip(src_y, -18, s->avctx->coded_height);
676 src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
680 srcY += src_y * s->linesize + src_x;
681 if (v->field_mode && v->ref_field_type[dir])
682 srcY += s->current_picture_ptr->f->linesize[0];
684 if (fieldmv && !(src_y & 1))
686 if (fieldmv && (src_y & 1) && src_y < 4)
688 if (v->rangeredfrm || use_ic
689 || s->h_edge_pos < 13 || v_edge_pos < 23
690 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
691 || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
692 srcY -= s->mspel * (1 + (s->linesize << fieldmv));
693 /* check emulate edge stride and offset */
694 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
695 s->linesize, s->linesize,
696 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
697 src_x - s->mspel, src_y - (s->mspel << fieldmv),
698 s->h_edge_pos, v_edge_pos);
699 srcY = s->edge_emu_buffer;
700 /* if we deal with range reduction we need to scale source blocks */
701 if (v->rangeredfrm) {
706 for (j = 0; j < 9 + s->mspel * 2; j++) {
707 for (i = 0; i < 9 + s->mspel * 2; i++)
708 src[i] = ((src[i] - 128) >> 1) + 128;
709 src += s->linesize << fieldmv;
712 /* if we deal with intensity compensation we need to scale source blocks */
718 for (j = 0; j < 9 + s->mspel * 2; j++) {
719 int f = v->field_mode ? v->ref_field_type[dir] : (((j<<fieldmv)+src_y - (s->mspel << fieldmv)) & 1);
720 for (i = 0; i < 9 + s->mspel * 2; i++)
721 src[i] = luty[f][src[i]];
722 src += s->linesize << fieldmv;
725 srcY += s->mspel * (1 + (s->linesize << fieldmv));
729 dxy = ((my & 3) << 2) | (mx & 3);
731 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
733 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
734 } else { // hpel mc - always used for luma
735 dxy = (my & 2) | ((mx & 2) >> 1);
737 s->hdsp.put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
739 s->hdsp.put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
743 static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
746 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
748 idx = ((a[3] != flag) << 3)
749 | ((a[2] != flag) << 2)
750 | ((a[1] != flag) << 1)
753 *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
754 *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
756 } else if (count[idx] == 1) {
759 *tx = mid_pred(mvx[1], mvx[2], mvx[3]);
760 *ty = mid_pred(mvy[1], mvy[2], mvy[3]);
763 *tx = mid_pred(mvx[0], mvx[2], mvx[3]);
764 *ty = mid_pred(mvy[0], mvy[2], mvy[3]);
767 *tx = mid_pred(mvx[0], mvx[1], mvx[3]);
768 *ty = mid_pred(mvy[0], mvy[1], mvy[3]);
771 *tx = mid_pred(mvx[0], mvx[1], mvx[2]);
772 *ty = mid_pred(mvy[0], mvy[1], mvy[2]);
775 } else if (count[idx] == 2) {
777 for (i = 0; i < 3; i++)
782 for (i = t1 + 1; i < 4; i++)
787 *tx = (mvx[t1] + mvx[t2]) / 2;
788 *ty = (mvy[t1] + mvy[t2]) / 2;
796 /** Do motion compensation for 4-MV macroblock - both chroma blocks
798 static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
800 MpegEncContext *s = &v->s;
801 H264ChromaContext *h264chroma = &v->h264chroma;
802 uint8_t *srcU, *srcV;
803 int uvmx, uvmy, uvsrc_x, uvsrc_y;
804 int k, tx = 0, ty = 0;
805 int mvx[4], mvy[4], intra[4], mv_f[4];
807 int chroma_ref_type = v->cur_field_type;
808 int v_edge_pos = s->v_edge_pos >> v->field_mode;
809 uint8_t (*lutuv)[256];
812 if (!v->field_mode && !v->s.last_picture.f->data[0])
814 if (s->flags & CODEC_FLAG_GRAY)
817 for (k = 0; k < 4; k++) {
818 mvx[k] = s->mv[dir][k][0];
819 mvy[k] = s->mv[dir][k][1];
820 intra[k] = v->mb_type[0][s->block_index[k]];
822 mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
825 /* calculate chroma MV vector from four luma MVs */
826 if (!v->field_mode || (v->field_mode && !v->numref)) {
827 valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
828 chroma_ref_type = v->reffield;
830 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
831 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
832 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
833 return; //no need to do MC for intra blocks
837 if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
839 valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
841 chroma_ref_type = !v->cur_field_type;
843 if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f->data[0])
845 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
846 s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
847 uvmx = (tx + ((tx & 3) == 3)) >> 1;
848 uvmy = (ty + ((ty & 3) == 3)) >> 1;
850 v->luma_mv[s->mb_x][0] = uvmx;
851 v->luma_mv[s->mb_x][1] = uvmy;
854 uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
855 uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
857 // Field conversion bias
858 if (v->cur_field_type != chroma_ref_type)
859 uvmy += 2 - 4 * chroma_ref_type;
861 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
862 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
864 if (v->profile != PROFILE_ADVANCED) {
865 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
866 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
868 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
869 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
873 if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
874 srcU = s->current_picture.f->data[1];
875 srcV = s->current_picture.f->data[2];
876 lutuv = v->curr_lutuv;
877 use_ic = *v->curr_use_ic;
879 srcU = s->last_picture.f->data[1];
880 srcV = s->last_picture.f->data[2];
881 lutuv = v->last_lutuv;
882 use_ic = v->last_use_ic;
885 srcU = s->next_picture.f->data[1];
886 srcV = s->next_picture.f->data[2];
887 lutuv = v->next_lutuv;
888 use_ic = v->next_use_ic;
892 av_log(v->s.avctx, AV_LOG_ERROR, "Referenced frame missing.\n");
896 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
897 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
900 if (chroma_ref_type) {
901 srcU += s->current_picture_ptr->f->linesize[1];
902 srcV += s->current_picture_ptr->f->linesize[2];
906 if (v->rangeredfrm || use_ic
907 || s->h_edge_pos < 18 || v_edge_pos < 18
908 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
909 || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) {
910 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
911 s->uvlinesize, s->uvlinesize,
912 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
913 s->h_edge_pos >> 1, v_edge_pos >> 1);
914 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
915 s->uvlinesize, s->uvlinesize,
916 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
917 s->h_edge_pos >> 1, v_edge_pos >> 1);
918 srcU = s->edge_emu_buffer;
919 srcV = s->edge_emu_buffer + 16;
921 /* if we deal with range reduction we need to scale source blocks */
922 if (v->rangeredfrm) {
928 for (j = 0; j < 9; j++) {
929 for (i = 0; i < 9; i++) {
930 src[i] = ((src[i] - 128) >> 1) + 128;
931 src2[i] = ((src2[i] - 128) >> 1) + 128;
933 src += s->uvlinesize;
934 src2 += s->uvlinesize;
937 /* if we deal with intensity compensation we need to scale source blocks */
944 for (j = 0; j < 9; j++) {
945 int f = v->field_mode ? chroma_ref_type : ((j + uvsrc_y) & 1);
946 for (i = 0; i < 9; i++) {
947 src[i] = lutuv[f][src[i]];
948 src2[i] = lutuv[f][src2[i]];
950 src += s->uvlinesize;
951 src2 += s->uvlinesize;
956 /* Chroma MC always uses qpel bilinear */
957 uvmx = (uvmx & 3) << 1;
958 uvmy = (uvmy & 3) << 1;
960 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
961 h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
963 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
964 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
968 /** Do motion compensation for 4-MV interlaced frame chroma macroblock (both U and V)
970 static void vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg)
972 MpegEncContext *s = &v->s;
973 H264ChromaContext *h264chroma = &v->h264chroma;
974 uint8_t *srcU, *srcV;
975 int uvsrc_x, uvsrc_y;
976 int uvmx_field[4], uvmy_field[4];
978 int fieldmv = v->blk_mv_type[s->block_index[0]];
979 static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
980 int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
981 int v_edge_pos = s->v_edge_pos >> 1;
983 uint8_t (*lutuv)[256];
985 if (s->flags & CODEC_FLAG_GRAY)
988 for (i = 0; i < 4; i++) {
989 int d = i < 2 ? dir: dir2;
991 uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
994 uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
996 uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
999 for (i = 0; i < 4; i++) {
1000 off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
1001 uvsrc_x = s->mb_x * 8 + (i & 1) * 4 + (uvmx_field[i] >> 2);
1002 uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
1003 // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
1004 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
1005 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
1006 if (i < 2 ? dir : dir2) {
1007 srcU = s->next_picture.f->data[1];
1008 srcV = s->next_picture.f->data[2];
1009 lutuv = v->next_lutuv;
1010 use_ic = v->next_use_ic;
1012 srcU = s->last_picture.f->data[1];
1013 srcV = s->last_picture.f->data[2];
1014 lutuv = v->last_lutuv;
1015 use_ic = v->last_use_ic;
1019 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1020 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1021 uvmx_field[i] = (uvmx_field[i] & 3) << 1;
1022 uvmy_field[i] = (uvmy_field[i] & 3) << 1;
1024 if (fieldmv && !(uvsrc_y & 1))
1025 v_edge_pos = (s->v_edge_pos >> 1) - 1;
1027 if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
1030 || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
1031 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
1032 || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
1033 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU,
1034 s->uvlinesize, s->uvlinesize,
1035 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
1036 s->h_edge_pos >> 1, v_edge_pos);
1037 s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV,
1038 s->uvlinesize, s->uvlinesize,
1039 5, (5 << fieldmv), uvsrc_x, uvsrc_y,
1040 s->h_edge_pos >> 1, v_edge_pos);
1041 srcU = s->edge_emu_buffer;
1042 srcV = s->edge_emu_buffer + 16;
1044 /* if we deal with intensity compensation we need to scale source blocks */
1047 uint8_t *src, *src2;
1051 for (j = 0; j < 5; j++) {
1052 int f = (uvsrc_y + (j << fieldmv)) & 1;
1053 for (i = 0; i < 5; i++) {
1054 src[i] = lutuv[f][src[i]];
1055 src2[i] = lutuv[f][src2[i]];
1057 src += s->uvlinesize << fieldmv;
1058 src2 += s->uvlinesize << fieldmv;
1064 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1065 h264chroma->avg_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1067 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]);
1068 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]);
1072 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1073 h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1075 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]);
1076 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]);
1082 /***********************************************************************/
1084 * @name VC-1 Block-level functions
1085 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1091 * @brief Get macroblock-level quantizer scale
1093 #define GET_MQUANT() \
1094 if (v->dquantfrm) { \
1096 if (v->dqprofile == DQPROFILE_ALL_MBS) { \
1097 if (v->dqbilevel) { \
1098 mquant = (get_bits1(gb)) ? v->altpq : v->pq; \
1100 mqdiff = get_bits(gb, 3); \
1102 mquant = v->pq + mqdiff; \
1104 mquant = get_bits(gb, 5); \
1107 if (v->dqprofile == DQPROFILE_SINGLE_EDGE) \
1108 edges = 1 << v->dqsbedge; \
1109 else if (v->dqprofile == DQPROFILE_DOUBLE_EDGES) \
1110 edges = (3 << v->dqsbedge) % 15; \
1111 else if (v->dqprofile == DQPROFILE_FOUR_EDGES) \
1113 if ((edges&1) && !s->mb_x) \
1114 mquant = v->altpq; \
1115 if ((edges&2) && s->first_slice_line) \
1116 mquant = v->altpq; \
1117 if ((edges&4) && s->mb_x == (s->mb_width - 1)) \
1118 mquant = v->altpq; \
1119 if ((edges&8) && s->mb_y == (s->mb_height - 1)) \
1120 mquant = v->altpq; \
1121 if (!mquant || mquant > 31) { \
1122 av_log(v->s.avctx, AV_LOG_ERROR, \
1123 "Overriding invalid mquant %d\n", mquant); \
1129 * @def GET_MVDATA(_dmv_x, _dmv_y)
1130 * @brief Get MV differentials
1131 * @see MVDATA decoding from 8.3.5.2, p(1)20
1132 * @param _dmv_x Horizontal differential for decoded MV
1133 * @param _dmv_y Vertical differential for decoded MV
1135 #define GET_MVDATA(_dmv_x, _dmv_y) \
1136 index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table, \
1137 VC1_MV_DIFF_VLC_BITS, 2); \
1139 mb_has_coeffs = 1; \
1142 mb_has_coeffs = 0; \
1145 _dmv_x = _dmv_y = 0; \
1146 } else if (index == 35) { \
1147 _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample); \
1148 _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample); \
1149 } else if (index == 36) { \
1154 index1 = index % 6; \
1155 if (!s->quarter_sample && index1 == 5) val = 1; \
1157 if (size_table[index1] - val > 0) \
1158 val = get_bits(gb, size_table[index1] - val); \
1160 sign = 0 - (val&1); \
1161 _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
1163 index1 = index / 6; \
1164 if (!s->quarter_sample && index1 == 5) val = 1; \
1166 if (size_table[index1] - val > 0) \
1167 val = get_bits(gb, size_table[index1] - val); \
1169 sign = 0 - (val & 1); \
1170 _dmv_y = (sign ^ ((val >> 1) + offset_table[index1])) - sign; \
1173 static av_always_inline void get_mvdata_interlaced(VC1Context *v, int *dmv_x,
1174 int *dmv_y, int *pred_flag)
1177 int extend_x = 0, extend_y = 0;
1178 GetBitContext *gb = &v->s.gb;
1181 const int* offs_tab;
1184 bits = VC1_2REF_MVDATA_VLC_BITS;
1187 bits = VC1_1REF_MVDATA_VLC_BITS;
1190 switch (v->dmvrange) {
1198 extend_x = extend_y = 1;
1201 index = get_vlc2(gb, v->imv_vlc->table, bits, 3);
1203 *dmv_x = get_bits(gb, v->k_x);
1204 *dmv_y = get_bits(gb, v->k_y);
1207 *pred_flag = *dmv_y & 1;
1208 *dmv_y = (*dmv_y + *pred_flag) >> 1;
1210 *dmv_y = (*dmv_y + (*dmv_y & 1)) >> 1;
1215 av_assert0(index < esc);
1217 offs_tab = offset_table2;
1219 offs_tab = offset_table1;
1220 index1 = (index + 1) % 9;
1222 val = get_bits(gb, index1 + extend_x);
1223 sign = 0 -(val & 1);
1224 *dmv_x = (sign ^ ((val >> 1) + offs_tab[index1])) - sign;
1228 offs_tab = offset_table2;
1230 offs_tab = offset_table1;
1231 index1 = (index + 1) / 9;
1232 if (index1 > v->numref) {
1233 val = get_bits(gb, (index1 + (extend_y << v->numref)) >> v->numref);
1234 sign = 0 - (val & 1);
1235 *dmv_y = (sign ^ ((val >> 1) + offs_tab[index1 >> v->numref])) - sign;
1238 if (v->numref && pred_flag)
1239 *pred_flag = index1 & 1;
1243 static av_always_inline int scaleforsame_x(VC1Context *v, int n /* MV */, int dir)
1245 int scaledvalue, refdist;
1246 int scalesame1, scalesame2;
1247 int scalezone1_x, zone1offset_x;
1248 int table_index = dir ^ v->second_field;
1250 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1251 refdist = v->refdist;
1253 refdist = dir ? v->brfd : v->frfd;
1256 scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1257 scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1258 scalezone1_x = ff_vc1_field_mvpred_scales[table_index][3][refdist];
1259 zone1offset_x = ff_vc1_field_mvpred_scales[table_index][5][refdist];
1264 if (FFABS(n) < scalezone1_x)
1265 scaledvalue = (n * scalesame1) >> 8;
1268 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_x;
1270 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_x;
1273 return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1276 static av_always_inline int scaleforsame_y(VC1Context *v, int i, int n /* MV */, int dir)
1278 int scaledvalue, refdist;
1279 int scalesame1, scalesame2;
1280 int scalezone1_y, zone1offset_y;
1281 int table_index = dir ^ v->second_field;
1283 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1284 refdist = v->refdist;
1286 refdist = dir ? v->brfd : v->frfd;
1289 scalesame1 = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1290 scalesame2 = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1291 scalezone1_y = ff_vc1_field_mvpred_scales[table_index][4][refdist];
1292 zone1offset_y = ff_vc1_field_mvpred_scales[table_index][6][refdist];
1297 if (FFABS(n) < scalezone1_y)
1298 scaledvalue = (n * scalesame1) >> 8;
1301 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_y;
1303 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_y;
1307 if (v->cur_field_type && !v->ref_field_type[dir])
1308 return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1310 return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1313 static av_always_inline int scaleforopp_x(VC1Context *v, int n /* MV */)
1315 int scalezone1_x, zone1offset_x;
1316 int scaleopp1, scaleopp2, brfd;
1319 brfd = FFMIN(v->brfd, 3);
1320 scalezone1_x = ff_vc1_b_field_mvpred_scales[3][brfd];
1321 zone1offset_x = ff_vc1_b_field_mvpred_scales[5][brfd];
1322 scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
1323 scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
1328 if (FFABS(n) < scalezone1_x)
1329 scaledvalue = (n * scaleopp1) >> 8;
1332 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_x;
1334 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_x;
1337 return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1340 static av_always_inline int scaleforopp_y(VC1Context *v, int n /* MV */, int dir)
1342 int scalezone1_y, zone1offset_y;
1343 int scaleopp1, scaleopp2, brfd;
1346 brfd = FFMIN(v->brfd, 3);
1347 scalezone1_y = ff_vc1_b_field_mvpred_scales[4][brfd];
1348 zone1offset_y = ff_vc1_b_field_mvpred_scales[6][brfd];
1349 scaleopp1 = ff_vc1_b_field_mvpred_scales[1][brfd];
1350 scaleopp2 = ff_vc1_b_field_mvpred_scales[2][brfd];
1355 if (FFABS(n) < scalezone1_y)
1356 scaledvalue = (n * scaleopp1) >> 8;
1359 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_y;
1361 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_y;
1364 if (v->cur_field_type && !v->ref_field_type[dir]) {
1365 return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1367 return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1371 static av_always_inline int scaleforsame(VC1Context *v, int i, int n /* MV */,
1374 int brfd, scalesame;
1375 int hpel = 1 - v->s.quarter_sample;
1378 if (v->s.pict_type != AV_PICTURE_TYPE_B || v->second_field || !dir) {
1380 n = scaleforsame_y(v, i, n, dir) << hpel;
1382 n = scaleforsame_x(v, n, dir) << hpel;
1385 brfd = FFMIN(v->brfd, 3);
1386 scalesame = ff_vc1_b_field_mvpred_scales[0][brfd];
1388 n = (n * scalesame >> 8) << hpel;
1392 static av_always_inline int scaleforopp(VC1Context *v, int n /* MV */,
1395 int refdist, scaleopp;
1396 int hpel = 1 - v->s.quarter_sample;
1399 if (v->s.pict_type == AV_PICTURE_TYPE_B && !v->second_field && dir == 1) {
1401 n = scaleforopp_y(v, n, dir) << hpel;
1403 n = scaleforopp_x(v, n) << hpel;
1406 if (v->s.pict_type != AV_PICTURE_TYPE_B)
1407 refdist = FFMIN(v->refdist, 3);
1409 refdist = dir ? v->brfd : v->frfd;
1410 scaleopp = ff_vc1_field_mvpred_scales[dir ^ v->second_field][0][refdist];
1412 n = (n * scaleopp >> 8) << hpel;
1416 /** Predict and set motion vector
1418 static inline void vc1_pred_mv(VC1Context *v, int n, int dmv_x, int dmv_y,
1419 int mv1, int r_x, int r_y, uint8_t* is_intra,
1420 int pred_flag, int dir)
1422 MpegEncContext *s = &v->s;
1423 int xy, wrap, off = 0;
1427 int mixedmv_pic, num_samefield = 0, num_oppfield = 0;
1428 int opposite, a_f, b_f, c_f;
1429 int16_t field_predA[2];
1430 int16_t field_predB[2];
1431 int16_t field_predC[2];
1432 int a_valid, b_valid, c_valid;
1433 int hybridmv_thresh, y_bias = 0;
1435 if (v->mv_mode == MV_PMODE_MIXED_MV ||
1436 ((v->mv_mode == MV_PMODE_INTENSITY_COMP) && (v->mv_mode2 == MV_PMODE_MIXED_MV)))
1440 /* scale MV difference to be quad-pel */
1441 dmv_x <<= 1 - s->quarter_sample;
1442 dmv_y <<= 1 - s->quarter_sample;
1444 wrap = s->b8_stride;
1445 xy = s->block_index[n];
1448 s->mv[0][n][0] = s->current_picture.motion_val[0][xy + v->blocks_off][0] = 0;
1449 s->mv[0][n][1] = s->current_picture.motion_val[0][xy + v->blocks_off][1] = 0;
1450 s->current_picture.motion_val[1][xy + v->blocks_off][0] = 0;
1451 s->current_picture.motion_val[1][xy + v->blocks_off][1] = 0;
1452 if (mv1) { /* duplicate motion data for 1-MV block */
1453 s->current_picture.motion_val[0][xy + 1 + v->blocks_off][0] = 0;
1454 s->current_picture.motion_val[0][xy + 1 + v->blocks_off][1] = 0;
1455 s->current_picture.motion_val[0][xy + wrap + v->blocks_off][0] = 0;
1456 s->current_picture.motion_val[0][xy + wrap + v->blocks_off][1] = 0;
1457 s->current_picture.motion_val[0][xy + wrap + 1 + v->blocks_off][0] = 0;
1458 s->current_picture.motion_val[0][xy + wrap + 1 + v->blocks_off][1] = 0;
1459 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1460 s->current_picture.motion_val[1][xy + 1 + v->blocks_off][0] = 0;
1461 s->current_picture.motion_val[1][xy + 1 + v->blocks_off][1] = 0;
1462 s->current_picture.motion_val[1][xy + wrap][0] = 0;
1463 s->current_picture.motion_val[1][xy + wrap + v->blocks_off][1] = 0;
1464 s->current_picture.motion_val[1][xy + wrap + 1 + v->blocks_off][0] = 0;
1465 s->current_picture.motion_val[1][xy + wrap + 1 + v->blocks_off][1] = 0;
1470 C = s->current_picture.motion_val[dir][xy - 1 + v->blocks_off];
1471 A = s->current_picture.motion_val[dir][xy - wrap + v->blocks_off];
1473 if (v->field_mode && mixedmv_pic)
1474 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1476 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
1478 //in 4-MV mode different blocks have different B predictor position
1481 off = (s->mb_x > 0) ? -1 : 1;
1484 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
1493 B = s->current_picture.motion_val[dir][xy - wrap + off + v->blocks_off];
1495 a_valid = !s->first_slice_line || (n == 2 || n == 3);
1496 b_valid = a_valid && (s->mb_width > 1);
1497 c_valid = s->mb_x || (n == 1 || n == 3);
1498 if (v->field_mode) {
1499 a_valid = a_valid && !is_intra[xy - wrap];
1500 b_valid = b_valid && !is_intra[xy - wrap + off];
1501 c_valid = c_valid && !is_intra[xy - 1];
1505 a_f = v->mv_f[dir][xy - wrap + v->blocks_off];
1506 num_oppfield += a_f;
1507 num_samefield += 1 - a_f;
1508 field_predA[0] = A[0];
1509 field_predA[1] = A[1];
1511 field_predA[0] = field_predA[1] = 0;
1515 b_f = v->mv_f[dir][xy - wrap + off + v->blocks_off];
1516 num_oppfield += b_f;
1517 num_samefield += 1 - b_f;
1518 field_predB[0] = B[0];
1519 field_predB[1] = B[1];
1521 field_predB[0] = field_predB[1] = 0;
1525 c_f = v->mv_f[dir][xy - 1 + v->blocks_off];
1526 num_oppfield += c_f;
1527 num_samefield += 1 - c_f;
1528 field_predC[0] = C[0];
1529 field_predC[1] = C[1];
1531 field_predC[0] = field_predC[1] = 0;
1535 if (v->field_mode) {
1537 // REFFIELD determines if the last field or the second-last field is
1538 // to be used as reference
1539 opposite = 1 - v->reffield;
1541 if (num_samefield <= num_oppfield)
1542 opposite = 1 - pred_flag;
1544 opposite = pred_flag;
1549 if (a_valid && !a_f) {
1550 field_predA[0] = scaleforopp(v, field_predA[0], 0, dir);
1551 field_predA[1] = scaleforopp(v, field_predA[1], 1, dir);
1553 if (b_valid && !b_f) {
1554 field_predB[0] = scaleforopp(v, field_predB[0], 0, dir);
1555 field_predB[1] = scaleforopp(v, field_predB[1], 1, dir);
1557 if (c_valid && !c_f) {
1558 field_predC[0] = scaleforopp(v, field_predC[0], 0, dir);
1559 field_predC[1] = scaleforopp(v, field_predC[1], 1, dir);
1561 v->mv_f[dir][xy + v->blocks_off] = 1;
1562 v->ref_field_type[dir] = !v->cur_field_type;
1564 if (a_valid && a_f) {
1565 field_predA[0] = scaleforsame(v, n, field_predA[0], 0, dir);
1566 field_predA[1] = scaleforsame(v, n, field_predA[1], 1, dir);
1568 if (b_valid && b_f) {
1569 field_predB[0] = scaleforsame(v, n, field_predB[0], 0, dir);
1570 field_predB[1] = scaleforsame(v, n, field_predB[1], 1, dir);
1572 if (c_valid && c_f) {
1573 field_predC[0] = scaleforsame(v, n, field_predC[0], 0, dir);
1574 field_predC[1] = scaleforsame(v, n, field_predC[1], 1, dir);
1576 v->mv_f[dir][xy + v->blocks_off] = 0;
1577 v->ref_field_type[dir] = v->cur_field_type;
1581 px = field_predA[0];
1582 py = field_predA[1];
1583 } else if (c_valid) {
1584 px = field_predC[0];
1585 py = field_predC[1];
1586 } else if (b_valid) {
1587 px = field_predB[0];
1588 py = field_predB[1];
1594 if (num_samefield + num_oppfield > 1) {
1595 px = mid_pred(field_predA[0], field_predB[0], field_predC[0]);
1596 py = mid_pred(field_predA[1], field_predB[1], field_predC[1]);
1599 /* Pullback MV as specified in 8.3.5.3.4 */
1600 if (!v->field_mode) {
1602 qx = (s->mb_x << 6) + ((n == 1 || n == 3) ? 32 : 0);
1603 qy = (s->mb_y << 6) + ((n == 2 || n == 3) ? 32 : 0);
1604 X = (s->mb_width << 6) - 4;
1605 Y = (s->mb_height << 6) - 4;
1607 if (qx + px < -60) px = -60 - qx;
1608 if (qy + py < -60) py = -60 - qy;
1610 if (qx + px < -28) px = -28 - qx;
1611 if (qy + py < -28) py = -28 - qy;
1613 if (qx + px > X) px = X - qx;
1614 if (qy + py > Y) py = Y - qy;
1617 if (!v->field_mode || s->pict_type != AV_PICTURE_TYPE_B) {
1618 /* Calculate hybrid prediction as specified in 8.3.5.3.5 (also 10.3.5.4.3.5) */
1619 hybridmv_thresh = 32;
1620 if (a_valid && c_valid) {
1621 if (is_intra[xy - wrap])
1622 sum = FFABS(px) + FFABS(py);
1624 sum = FFABS(px - field_predA[0]) + FFABS(py - field_predA[1]);
1625 if (sum > hybridmv_thresh) {
1626 if (get_bits1(&s->gb)) { // read HYBRIDPRED bit
1627 px = field_predA[0];
1628 py = field_predA[1];
1630 px = field_predC[0];
1631 py = field_predC[1];
1634 if (is_intra[xy - 1])
1635 sum = FFABS(px) + FFABS(py);
1637 sum = FFABS(px - field_predC[0]) + FFABS(py - field_predC[1]);
1638 if (sum > hybridmv_thresh) {
1639 if (get_bits1(&s->gb)) {
1640 px = field_predA[0];
1641 py = field_predA[1];
1643 px = field_predC[0];
1644 py = field_predC[1];
1651 if (v->field_mode && v->numref)
1653 if (v->field_mode && v->cur_field_type && v->ref_field_type[dir] == 0)
1655 /* store MV using signed modulus of MV range defined in 4.11 */
1656 s->mv[dir][n][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1657 s->mv[dir][n][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1] = ((py + dmv_y + r_y - y_bias) & ((r_y << 1) - 1)) - r_y + y_bias;
1658 if (mv1) { /* duplicate motion data for 1-MV block */
1659 s->current_picture.motion_val[dir][xy + 1 + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
1660 s->current_picture.motion_val[dir][xy + 1 + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
1661 s->current_picture.motion_val[dir][xy + wrap + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
1662 s->current_picture.motion_val[dir][xy + wrap + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
1663 s->current_picture.motion_val[dir][xy + wrap + 1 + v->blocks_off][0] = s->current_picture.motion_val[dir][xy + v->blocks_off][0];
1664 s->current_picture.motion_val[dir][xy + wrap + 1 + v->blocks_off][1] = s->current_picture.motion_val[dir][xy + v->blocks_off][1];
1665 v->mv_f[dir][xy + 1 + v->blocks_off] = v->mv_f[dir][xy + v->blocks_off];
1666 v->mv_f[dir][xy + wrap + v->blocks_off] = v->mv_f[dir][xy + wrap + 1 + v->blocks_off] = v->mv_f[dir][xy + v->blocks_off];
1670 /** Predict and set motion vector for interlaced frame picture MBs
1672 static inline void vc1_pred_mv_intfr(VC1Context *v, int n, int dmv_x, int dmv_y,
1673 int mvn, int r_x, int r_y, uint8_t* is_intra, int dir)
1675 MpegEncContext *s = &v->s;
1676 int xy, wrap, off = 0;
1677 int A[2], B[2], C[2];
1679 int a_valid = 0, b_valid = 0, c_valid = 0;
1680 int field_a, field_b, field_c; // 0: same, 1: opposit
1681 int total_valid, num_samefield, num_oppfield;
1682 int pos_c, pos_b, n_adj;
1684 wrap = s->b8_stride;
1685 xy = s->block_index[n];
1688 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = 0;
1689 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = 0;
1690 s->current_picture.motion_val[1][xy][0] = 0;
1691 s->current_picture.motion_val[1][xy][1] = 0;
1692 if (mvn == 1) { /* duplicate motion data for 1-MV block */
1693 s->current_picture.motion_val[0][xy + 1][0] = 0;
1694 s->current_picture.motion_val[0][xy + 1][1] = 0;
1695 s->current_picture.motion_val[0][xy + wrap][0] = 0;
1696 s->current_picture.motion_val[0][xy + wrap][1] = 0;
1697 s->current_picture.motion_val[0][xy + wrap + 1][0] = 0;
1698 s->current_picture.motion_val[0][xy + wrap + 1][1] = 0;
1699 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1700 s->current_picture.motion_val[1][xy + 1][0] = 0;
1701 s->current_picture.motion_val[1][xy + 1][1] = 0;
1702 s->current_picture.motion_val[1][xy + wrap][0] = 0;
1703 s->current_picture.motion_val[1][xy + wrap][1] = 0;
1704 s->current_picture.motion_val[1][xy + wrap + 1][0] = 0;
1705 s->current_picture.motion_val[1][xy + wrap + 1][1] = 0;
1710 off = ((n == 0) || (n == 1)) ? 1 : -1;
1712 if (s->mb_x || (n == 1) || (n == 3)) {
1713 if ((v->blk_mv_type[xy]) // current block (MB) has a field MV
1714 || (!v->blk_mv_type[xy] && !v->blk_mv_type[xy - 1])) { // or both have frame MV
1715 A[0] = s->current_picture.motion_val[dir][xy - 1][0];
1716 A[1] = s->current_picture.motion_val[dir][xy - 1][1];
1718 } else { // current block has frame mv and cand. has field MV (so average)
1719 A[0] = (s->current_picture.motion_val[dir][xy - 1][0]
1720 + s->current_picture.motion_val[dir][xy - 1 + off * wrap][0] + 1) >> 1;
1721 A[1] = (s->current_picture.motion_val[dir][xy - 1][1]
1722 + s->current_picture.motion_val[dir][xy - 1 + off * wrap][1] + 1) >> 1;
1725 if (!(n & 1) && v->is_intra[s->mb_x - 1]) {
1731 /* Predict B and C */
1732 B[0] = B[1] = C[0] = C[1] = 0;
1733 if (n == 0 || n == 1 || v->blk_mv_type[xy]) {
1734 if (!s->first_slice_line) {
1735 if (!v->is_intra[s->mb_x - s->mb_stride]) {
1738 pos_b = s->block_index[n_adj] - 2 * wrap;
1739 if (v->blk_mv_type[pos_b] && v->blk_mv_type[xy]) {
1740 n_adj = (n & 2) | (n & 1);
1742 B[0] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap][0];
1743 B[1] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap][1];
1744 if (v->blk_mv_type[pos_b] && !v->blk_mv_type[xy]) {
1745 B[0] = (B[0] + s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap][0] + 1) >> 1;
1746 B[1] = (B[1] + s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap][1] + 1) >> 1;
1749 if (s->mb_width > 1) {
1750 if (!v->is_intra[s->mb_x - s->mb_stride + 1]) {
1753 pos_c = s->block_index[2] - 2 * wrap + 2;
1754 if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1757 C[0] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap + 2][0];
1758 C[1] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap + 2][1];
1759 if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1760 C[0] = (1 + C[0] + (s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap + 2][0])) >> 1;
1761 C[1] = (1 + C[1] + (s->current_picture.motion_val[dir][s->block_index[n_adj ^ 2] - 2 * wrap + 2][1])) >> 1;
1763 if (s->mb_x == s->mb_width - 1) {
1764 if (!v->is_intra[s->mb_x - s->mb_stride - 1]) {
1767 pos_c = s->block_index[3] - 2 * wrap - 2;
1768 if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1771 C[0] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap - 2][0];
1772 C[1] = s->current_picture.motion_val[dir][s->block_index[n_adj] - 2 * wrap - 2][1];
1773 if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1774 C[0] = (1 + C[0] + s->current_picture.motion_val[dir][s->block_index[1] - 2 * wrap - 2][0]) >> 1;
1775 C[1] = (1 + C[1] + s->current_picture.motion_val[dir][s->block_index[1] - 2 * wrap - 2][1]) >> 1;
1784 pos_b = s->block_index[1];
1786 B[0] = s->current_picture.motion_val[dir][pos_b][0];
1787 B[1] = s->current_picture.motion_val[dir][pos_b][1];
1788 pos_c = s->block_index[0];
1790 C[0] = s->current_picture.motion_val[dir][pos_c][0];
1791 C[1] = s->current_picture.motion_val[dir][pos_c][1];
1794 total_valid = a_valid + b_valid + c_valid;
1795 // check if predictor A is out of bounds
1796 if (!s->mb_x && !(n == 1 || n == 3)) {
1799 // check if predictor B is out of bounds
1800 if ((s->first_slice_line && v->blk_mv_type[xy]) || (s->first_slice_line && !(n & 2))) {
1801 B[0] = B[1] = C[0] = C[1] = 0;
1803 if (!v->blk_mv_type[xy]) {
1804 if (s->mb_width == 1) {
1808 if (total_valid >= 2) {
1809 px = mid_pred(A[0], B[0], C[0]);
1810 py = mid_pred(A[1], B[1], C[1]);
1811 } else if (total_valid) {
1812 if (a_valid) { px = A[0]; py = A[1]; }
1813 else if (b_valid) { px = B[0]; py = B[1]; }
1814 else { px = C[0]; py = C[1]; }
1819 field_a = (A[1] & 4) ? 1 : 0;
1823 field_b = (B[1] & 4) ? 1 : 0;
1827 field_c = (C[1] & 4) ? 1 : 0;
1831 num_oppfield = field_a + field_b + field_c;
1832 num_samefield = total_valid - num_oppfield;
1833 if (total_valid == 3) {
1834 if ((num_samefield == 3) || (num_oppfield == 3)) {
1835 px = mid_pred(A[0], B[0], C[0]);
1836 py = mid_pred(A[1], B[1], C[1]);
1837 } else if (num_samefield >= num_oppfield) {
1838 /* take one MV from same field set depending on priority
1839 the check for B may not be necessary */
1840 px = !field_a ? A[0] : B[0];
1841 py = !field_a ? A[1] : B[1];
1843 px = field_a ? A[0] : B[0];
1844 py = field_a ? A[1] : B[1];
1846 } else if (total_valid == 2) {
1847 if (num_samefield >= num_oppfield) {
1848 if (!field_a && a_valid) {
1851 } else if (!field_b && b_valid) {
1854 } else /*if (c_valid)*/ {
1855 av_assert1(c_valid);
1858 } /*else px = py = 0;*/
1860 if (field_a && a_valid) {
1863 } else /*if (field_b && b_valid)*/ {
1864 av_assert1(field_b && b_valid);
1867 } /*else if (c_valid) {
1872 } else if (total_valid == 1) {
1873 px = (a_valid) ? A[0] : ((b_valid) ? B[0] : C[0]);
1874 py = (a_valid) ? A[1] : ((b_valid) ? B[1] : C[1]);
1878 /* store MV using signed modulus of MV range defined in 4.11 */
1879 s->mv[dir][n][0] = s->current_picture.motion_val[dir][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1880 s->mv[dir][n][1] = s->current_picture.motion_val[dir][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
1881 if (mvn == 1) { /* duplicate motion data for 1-MV block */
1882 s->current_picture.motion_val[dir][xy + 1 ][0] = s->current_picture.motion_val[dir][xy][0];
1883 s->current_picture.motion_val[dir][xy + 1 ][1] = s->current_picture.motion_val[dir][xy][1];
1884 s->current_picture.motion_val[dir][xy + wrap ][0] = s->current_picture.motion_val[dir][xy][0];
1885 s->current_picture.motion_val[dir][xy + wrap ][1] = s->current_picture.motion_val[dir][xy][1];
1886 s->current_picture.motion_val[dir][xy + wrap + 1][0] = s->current_picture.motion_val[dir][xy][0];
1887 s->current_picture.motion_val[dir][xy + wrap + 1][1] = s->current_picture.motion_val[dir][xy][1];
1888 } else if (mvn == 2) { /* duplicate motion data for 2-Field MV block */
1889 s->current_picture.motion_val[dir][xy + 1][0] = s->current_picture.motion_val[dir][xy][0];
1890 s->current_picture.motion_val[dir][xy + 1][1] = s->current_picture.motion_val[dir][xy][1];
1891 s->mv[dir][n + 1][0] = s->mv[dir][n][0];
1892 s->mv[dir][n + 1][1] = s->mv[dir][n][1];
1896 /** Motion compensation for direct or interpolated blocks in B-frames
1898 static void vc1_interp_mc(VC1Context *v)
1900 MpegEncContext *s = &v->s;
1901 H264ChromaContext *h264chroma = &v->h264chroma;
1902 uint8_t *srcY, *srcU, *srcV;
1903 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
1905 int v_edge_pos = s->v_edge_pos >> v->field_mode;
1906 int use_ic = v->next_use_ic;
1908 if (!v->field_mode && !v->s.next_picture.f->data[0])
1911 mx = s->mv[1][0][0];
1912 my = s->mv[1][0][1];
1913 uvmx = (mx + ((mx & 3) == 3)) >> 1;
1914 uvmy = (my + ((my & 3) == 3)) >> 1;
1915 if (v->field_mode) {
1916 if (v->cur_field_type != v->ref_field_type[1]) {
1917 my = my - 2 + 4 * v->cur_field_type;
1918 uvmy = uvmy - 2 + 4 * v->cur_field_type;
1922 uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
1923 uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
1925 srcY = s->next_picture.f->data[0];
1926 srcU = s->next_picture.f->data[1];
1927 srcV = s->next_picture.f->data[2];
1929 src_x = s->mb_x * 16 + (mx >> 2);
1930 src_y = s->mb_y * 16 + (my >> 2);
1931 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
1932 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
1934 if (v->profile != PROFILE_ADVANCED) {
1935 src_x = av_clip( src_x, -16, s->mb_width * 16);
1936 src_y = av_clip( src_y, -16, s->mb_height * 16);
1937 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
1938 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
1940 src_x = av_clip( src_x, -17, s->avctx->coded_width);
1941 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
1942 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
1943 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
1946 srcY += src_y * s->linesize + src_x;
1947 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1948 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1950 if (v->field_mode && v->ref_field_type[1]) {
1951 srcY += s->current_picture_ptr->f->linesize[0];
1952 srcU += s->current_picture_ptr->f->linesize[1];
1953 srcV += s->current_picture_ptr->f->linesize[2];
1956 /* for grayscale we should not try to read from unknown area */
1957 if (s->flags & CODEC_FLAG_GRAY) {
1958 srcU = s->edge_emu_buffer + 18 * s->linesize;
1959 srcV = s->edge_emu_buffer + 18 * s->linesize;
1962 if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22 || use_ic
1963 || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
1964 || (unsigned)(src_y - 1) > v_edge_pos - (my & 3) - 16 - 3) {
1965 uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
1967 srcY -= s->mspel * (1 + s->linesize);
1968 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,
1969 s->linesize, s->linesize,
1970 17 + s->mspel * 2, 17 + s->mspel * 2,
1971 src_x - s->mspel, src_y - s->mspel,
1972 s->h_edge_pos, v_edge_pos);
1973 srcY = s->edge_emu_buffer;
1974 s->vdsp.emulated_edge_mc(uvbuf, srcU,
1975 s->uvlinesize, s->uvlinesize,
1978 s->h_edge_pos >> 1, v_edge_pos >> 1);
1979 s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,
1980 s->uvlinesize, s->uvlinesize,
1983 s->h_edge_pos >> 1, v_edge_pos >> 1);
1986 /* if we deal with range reduction we need to scale source blocks */
1987 if (v->rangeredfrm) {
1989 uint8_t *src, *src2;
1992 for (j = 0; j < 17 + s->mspel * 2; j++) {
1993 for (i = 0; i < 17 + s->mspel * 2; i++)
1994 src[i] = ((src[i] - 128) >> 1) + 128;
1999 for (j = 0; j < 9; j++) {
2000 for (i = 0; i < 9; i++) {
2001 src[i] = ((src[i] - 128) >> 1) + 128;
2002 src2[i] = ((src2[i] - 128) >> 1) + 128;
2004 src += s->uvlinesize;
2005 src2 += s->uvlinesize;
2010 uint8_t (*luty )[256] = v->next_luty;
2011 uint8_t (*lutuv)[256] = v->next_lutuv;
2013 uint8_t *src, *src2;
2016 for (j = 0; j < 17 + s->mspel * 2; j++) {
2017 int f = v->field_mode ? v->ref_field_type[1] : ((j+src_y - s->mspel) & 1);
2018 for (i = 0; i < 17 + s->mspel * 2; i++)
2019 src[i] = luty[f][src[i]];
2024 for (j = 0; j < 9; j++) {
2025 int f = v->field_mode ? v->ref_field_type[1] : ((j+uvsrc_y) & 1);
2026 for (i = 0; i < 9; i++) {
2027 src[i] = lutuv[f][src[i]];
2028 src2[i] = lutuv[f][src2[i]];
2030 src += s->uvlinesize;
2031 src2 += s->uvlinesize;
2034 srcY += s->mspel * (1 + s->linesize);
2041 dxy = ((my & 3) << 2) | (mx & 3);
2042 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off , srcY , s->linesize, v->rnd);
2043 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
2044 srcY += s->linesize * 8;
2045 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize , srcY , s->linesize, v->rnd);
2046 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
2048 dxy = (my & 2) | ((mx & 2) >> 1);
2051 s->hdsp.avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
2053 s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, 16);
2056 if (s->flags & CODEC_FLAG_GRAY) return;
2057 /* Chroma MC always uses qpel blilinear */
2058 uvmx = (uvmx & 3) << 1;
2059 uvmy = (uvmy & 3) << 1;
2061 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
2062 h264chroma->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
2064 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
2065 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
2069 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
2073 #if B_FRACTION_DEN==256
2077 return 2 * ((value * n + 255) >> 9);
2078 return (value * n + 128) >> 8;
2081 n -= B_FRACTION_DEN;
2083 return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
2084 return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
2088 /** Reconstruct motion vector for B-frame and do motion compensation
2090 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2],
2091 int direct, int mode)
2098 if (mode == BMV_TYPE_INTERPOLATED) {
2104 vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
2107 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2],
2108 int direct, int mvtype)
2110 MpegEncContext *s = &v->s;
2111 int xy, wrap, off = 0;
2116 const uint8_t *is_intra = v->mb_type[0];
2118 av_assert0(!v->field_mode);
2122 /* scale MV difference to be quad-pel */
2123 dmv_x[0] <<= 1 - s->quarter_sample;
2124 dmv_y[0] <<= 1 - s->quarter_sample;
2125 dmv_x[1] <<= 1 - s->quarter_sample;
2126 dmv_y[1] <<= 1 - s->quarter_sample;
2128 wrap = s->b8_stride;
2129 xy = s->block_index[0];
2132 s->current_picture.motion_val[0][xy][0] =
2133 s->current_picture.motion_val[0][xy][1] =
2134 s->current_picture.motion_val[1][xy][0] =
2135 s->current_picture.motion_val[1][xy][1] = 0;
2138 if (direct && s->next_picture_ptr->field_picture)
2139 av_log(s->avctx, AV_LOG_WARNING, "Mixed frame/field direct mode not supported\n");
2141 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
2142 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
2143 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
2144 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
2146 /* Pullback predicted motion vectors as specified in 8.4.5.4 */
2147 s->mv[0][0][0] = av_clip(s->mv[0][0][0], -60 - (s->mb_x << 6), (s->mb_width << 6) - 4 - (s->mb_x << 6));
2148 s->mv[0][0][1] = av_clip(s->mv[0][0][1], -60 - (s->mb_y << 6), (s->mb_height << 6) - 4 - (s->mb_y << 6));
2149 s->mv[1][0][0] = av_clip(s->mv[1][0][0], -60 - (s->mb_x << 6), (s->mb_width << 6) - 4 - (s->mb_x << 6));
2150 s->mv[1][0][1] = av_clip(s->mv[1][0][1], -60 - (s->mb_y << 6), (s->mb_height << 6) - 4 - (s->mb_y << 6));
2152 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
2153 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
2154 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
2155 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
2159 if ((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2160 C = s->current_picture.motion_val[0][xy - 2];
2161 A = s->current_picture.motion_val[0][xy - wrap * 2];
2162 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2163 B = s->current_picture.motion_val[0][xy - wrap * 2 + off];
2165 if (!s->mb_x) C[0] = C[1] = 0;
2166 if (!s->first_slice_line) { // predictor A is not out of bounds
2167 if (s->mb_width == 1) {
2171 px = mid_pred(A[0], B[0], C[0]);
2172 py = mid_pred(A[1], B[1], C[1]);
2174 } else if (s->mb_x) { // predictor C is not out of bounds
2180 /* Pullback MV as specified in 8.3.5.3.4 */
2183 if (v->profile < PROFILE_ADVANCED) {
2184 qx = (s->mb_x << 5);
2185 qy = (s->mb_y << 5);
2186 X = (s->mb_width << 5) - 4;
2187 Y = (s->mb_height << 5) - 4;
2188 if (qx + px < -28) px = -28 - qx;
2189 if (qy + py < -28) py = -28 - qy;
2190 if (qx + px > X) px = X - qx;
2191 if (qy + py > Y) py = Y - qy;
2193 qx = (s->mb_x << 6);
2194 qy = (s->mb_y << 6);
2195 X = (s->mb_width << 6) - 4;
2196 Y = (s->mb_height << 6) - 4;
2197 if (qx + px < -60) px = -60 - qx;
2198 if (qy + py < -60) py = -60 - qy;
2199 if (qx + px > X) px = X - qx;
2200 if (qy + py > Y) py = Y - qy;
2203 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2204 if (0 && !s->first_slice_line && s->mb_x) {
2205 if (is_intra[xy - wrap])
2206 sum = FFABS(px) + FFABS(py);
2208 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2210 if (get_bits1(&s->gb)) {
2218 if (is_intra[xy - 2])
2219 sum = FFABS(px) + FFABS(py);
2221 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2223 if (get_bits1(&s->gb)) {
2233 /* store MV using signed modulus of MV range defined in 4.11 */
2234 s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
2235 s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
2237 if ((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2238 C = s->current_picture.motion_val[1][xy - 2];
2239 A = s->current_picture.motion_val[1][xy - wrap * 2];
2240 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2241 B = s->current_picture.motion_val[1][xy - wrap * 2 + off];
2245 if (!s->first_slice_line) { // predictor A is not out of bounds
2246 if (s->mb_width == 1) {
2250 px = mid_pred(A[0], B[0], C[0]);
2251 py = mid_pred(A[1], B[1], C[1]);
2253 } else if (s->mb_x) { // predictor C is not out of bounds
2259 /* Pullback MV as specified in 8.3.5.3.4 */
2262 if (v->profile < PROFILE_ADVANCED) {
2263 qx = (s->mb_x << 5);
2264 qy = (s->mb_y << 5);
2265 X = (s->mb_width << 5) - 4;
2266 Y = (s->mb_height << 5) - 4;
2267 if (qx + px < -28) px = -28 - qx;
2268 if (qy + py < -28) py = -28 - qy;
2269 if (qx + px > X) px = X - qx;
2270 if (qy + py > Y) py = Y - qy;
2272 qx = (s->mb_x << 6);
2273 qy = (s->mb_y << 6);
2274 X = (s->mb_width << 6) - 4;
2275 Y = (s->mb_height << 6) - 4;
2276 if (qx + px < -60) px = -60 - qx;
2277 if (qy + py < -60) py = -60 - qy;
2278 if (qx + px > X) px = X - qx;
2279 if (qy + py > Y) py = Y - qy;
2282 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2283 if (0 && !s->first_slice_line && s->mb_x) {
2284 if (is_intra[xy - wrap])
2285 sum = FFABS(px) + FFABS(py);
2287 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2289 if (get_bits1(&s->gb)) {
2297 if (is_intra[xy - 2])
2298 sum = FFABS(px) + FFABS(py);
2300 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2302 if (get_bits1(&s->gb)) {
2312 /* store MV using signed modulus of MV range defined in 4.11 */
2314 s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
2315 s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
2317 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
2318 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
2319 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
2320 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
2323 static inline void vc1_pred_b_mv_intfi(VC1Context *v, int n, int *dmv_x, int *dmv_y, int mv1, int *pred_flag)
2325 int dir = (v->bmvtype == BMV_TYPE_BACKWARD) ? 1 : 0;
2326 MpegEncContext *s = &v->s;
2327 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2329 if (v->bmvtype == BMV_TYPE_DIRECT) {
2330 int total_opp, k, f;
2331 if (s->next_picture.mb_type[mb_pos + v->mb_off] != MB_TYPE_INTRA) {
2332 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][0],
2333 v->bfraction, 0, s->quarter_sample);
2334 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][1],
2335 v->bfraction, 0, s->quarter_sample);
2336 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][0],
2337 v->bfraction, 1, s->quarter_sample);
2338 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][s->block_index[0] + v->blocks_off][1],
2339 v->bfraction, 1, s->quarter_sample);
2341 total_opp = v->mv_f_next[0][s->block_index[0] + v->blocks_off]
2342 + v->mv_f_next[0][s->block_index[1] + v->blocks_off]
2343 + v->mv_f_next[0][s->block_index[2] + v->blocks_off]
2344 + v->mv_f_next[0][s->block_index[3] + v->blocks_off];
2345 f = (total_opp > 2) ? 1 : 0;
2347 s->mv[0][0][0] = s->mv[0][0][1] = 0;
2348 s->mv[1][0][0] = s->mv[1][0][1] = 0;
2351 v->ref_field_type[0] = v->ref_field_type[1] = v->cur_field_type ^ f;
2352 for (k = 0; k < 4; k++) {
2353 s->current_picture.motion_val[0][s->block_index[k] + v->blocks_off][0] = s->mv[0][0][0];
2354 s->current_picture.motion_val[0][s->block_index[k] + v->blocks_off][1] = s->mv[0][0][1];
2355 s->current_picture.motion_val[1][s->block_index[k] + v->blocks_off][0] = s->mv[1][0][0];
2356 s->current_picture.motion_val[1][s->block_index[k] + v->blocks_off][1] = s->mv[1][0][1];
2357 v->mv_f[0][s->block_index[k] + v->blocks_off] = f;
2358 v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
2362 if (v->bmvtype == BMV_TYPE_INTERPOLATED) {
2363 vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0], 1, v->range_x, v->range_y, v->mb_type[0], pred_flag[0], 0);
2364 vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1], 1, v->range_x, v->range_y, v->mb_type[0], pred_flag[1], 1);
2367 if (dir) { // backward
2368 vc1_pred_mv(v, n, dmv_x[1], dmv_y[1], mv1, v->range_x, v->range_y, v->mb_type[0], pred_flag[1], 1);
2369 if (n == 3 || mv1) {
2370 vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0], 1, v->range_x, v->range_y, v->mb_type[0], 0, 0);
2373 vc1_pred_mv(v, n, dmv_x[0], dmv_y[0], mv1, v->range_x, v->range_y, v->mb_type[0], pred_flag[0], 0);
2374 if (n == 3 || mv1) {
2375 vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1], 1, v->range_x, v->range_y, v->mb_type[0], 0, 1);
2380 /** Get predicted DC value for I-frames only
2381 * prediction dir: left=0, top=1
2382 * @param s MpegEncContext
2383 * @param overlap flag indicating that overlap filtering is used
2384 * @param pq integer part of picture quantizer
2385 * @param[in] n block index in the current MB
2386 * @param dc_val_ptr Pointer to DC predictor
2387 * @param dir_ptr Prediction direction for use in AC prediction
2389 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2390 int16_t **dc_val_ptr, int *dir_ptr)
2392 int a, b, c, wrap, pred, scale;
2394 static const uint16_t dcpred[32] = {
2395 -1, 1024, 512, 341, 256, 205, 171, 146, 128,
2396 114, 102, 93, 85, 79, 73, 68, 64,
2397 60, 57, 54, 51, 49, 47, 45, 43,
2398 41, 39, 38, 37, 35, 34, 33
2401 /* find prediction - wmv3_dc_scale always used here in fact */
2402 if (n < 4) scale = s->y_dc_scale;
2403 else scale = s->c_dc_scale;
2405 wrap = s->block_wrap[n];
2406 dc_val = s->dc_val[0] + s->block_index[n];
2412 b = dc_val[ - 1 - wrap];
2413 a = dc_val[ - wrap];
2415 if (pq < 9 || !overlap) {
2416 /* Set outer values */
2417 if (s->first_slice_line && (n != 2 && n != 3))
2418 b = a = dcpred[scale];
2419 if (s->mb_x == 0 && (n != 1 && n != 3))
2420 b = c = dcpred[scale];
2422 /* Set outer values */
2423 if (s->first_slice_line && (n != 2 && n != 3))
2425 if (s->mb_x == 0 && (n != 1 && n != 3))
2429 if (abs(a - b) <= abs(b - c)) {
2431 *dir_ptr = 1; // left
2434 *dir_ptr = 0; // top
2437 /* update predictor */
2438 *dc_val_ptr = &dc_val[0];
2443 /** Get predicted DC value
2444 * prediction dir: left=0, top=1
2445 * @param s MpegEncContext
2446 * @param overlap flag indicating that overlap filtering is used
2447 * @param pq integer part of picture quantizer
2448 * @param[in] n block index in the current MB
2449 * @param a_avail flag indicating top block availability
2450 * @param c_avail flag indicating left block availability
2451 * @param dc_val_ptr Pointer to DC predictor
2452 * @param dir_ptr Prediction direction for use in AC prediction
2454 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2455 int a_avail, int c_avail,
2456 int16_t **dc_val_ptr, int *dir_ptr)
2458 int a, b, c, wrap, pred;
2460 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2464 wrap = s->block_wrap[n];
2465 dc_val = s->dc_val[0] + s->block_index[n];
2471 b = dc_val[ - 1 - wrap];
2472 a = dc_val[ - wrap];
2473 /* scale predictors if needed */
2474 q1 = s->current_picture.qscale_table[mb_pos];
2475 dqscale_index = s->y_dc_scale_table[q1] - 1;
2476 if (dqscale_index < 0)
2478 if (c_avail && (n != 1 && n != 3)) {
2479 q2 = s->current_picture.qscale_table[mb_pos - 1];
2481 c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2483 if (a_avail && (n != 2 && n != 3)) {
2484 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2486 a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2488 if (a_avail && c_avail && (n != 3)) {
2493 off -= s->mb_stride;
2494 q2 = s->current_picture.qscale_table[off];
2496 b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2499 if (a_avail && c_avail) {
2500 if (abs(a - b) <= abs(b - c)) {
2502 *dir_ptr = 1; // left
2505 *dir_ptr = 0; // top
2507 } else if (a_avail) {
2509 *dir_ptr = 0; // top
2510 } else if (c_avail) {
2512 *dir_ptr = 1; // left
2515 *dir_ptr = 1; // left
2518 /* update predictor */
2519 *dc_val_ptr = &dc_val[0];
2523 /** @} */ // Block group
2526 * @name VC1 Macroblock-level functions in Simple/Main Profiles
2527 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
2531 static inline int vc1_coded_block_pred(MpegEncContext * s, int n,
2532 uint8_t **coded_block_ptr)
2534 int xy, wrap, pred, a, b, c;
2536 xy = s->block_index[n];
2537 wrap = s->b8_stride;
2542 a = s->coded_block[xy - 1 ];
2543 b = s->coded_block[xy - 1 - wrap];
2544 c = s->coded_block[xy - wrap];
2553 *coded_block_ptr = &s->coded_block[xy];
2559 * Decode one AC coefficient
2560 * @param v The VC1 context
2561 * @param last Last coefficient
2562 * @param skip How much zero coefficients to skip
2563 * @param value Decoded AC coefficient value
2564 * @param codingset set of VLC to decode data
2567 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip,
2568 int *value, int codingset)
2570 GetBitContext *gb = &v->s.gb;
2571 int index, escape, run = 0, level = 0, lst = 0;
2573 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2574 if (index != ff_vc1_ac_sizes[codingset] - 1) {
2575 run = vc1_index_decode_table[codingset][index][0];
2576 level = vc1_index_decode_table[codingset][index][1];
2577 lst = index >= vc1_last_decode_table[codingset] || get_bits_left(gb) < 0;
2581 escape = decode210(gb);
2583 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2584 run = vc1_index_decode_table[codingset][index][0];
2585 level = vc1_index_decode_table[codingset][index][1];
2586 lst = index >= vc1_last_decode_table[codingset];
2589 level += vc1_last_delta_level_table[codingset][run];
2591 level += vc1_delta_level_table[codingset][run];
2594 run += vc1_last_delta_run_table[codingset][level] + 1;
2596 run += vc1_delta_run_table[codingset][level] + 1;
2602 lst = get_bits1(gb);
2603 if (v->s.esc3_level_length == 0) {
2604 if (v->pq < 8 || v->dquantfrm) { // table 59
2605 v->s.esc3_level_length = get_bits(gb, 3);
2606 if (!v->s.esc3_level_length)
2607 v->s.esc3_level_length = get_bits(gb, 2) + 8;
2608 } else { // table 60
2609 v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
2611 v->s.esc3_run_length = 3 + get_bits(gb, 2);
2613 run = get_bits(gb, v->s.esc3_run_length);
2614 sign = get_bits1(gb);
2615 level = get_bits(gb, v->s.esc3_level_length);
2626 /** Decode intra block in intra frames - should be faster than decode_intra_block
2627 * @param v VC1Context
2628 * @param block block to decode
2629 * @param[in] n subblock index
2630 * @param coded are AC coeffs present or not
2631 * @param codingset set of VLC to decode data
2633 static int vc1_decode_i_block(VC1Context *v, int16_t block[64], int n,
2634 int coded, int codingset)
2636 GetBitContext *gb = &v->s.gb;
2637 MpegEncContext *s = &v->s;
2638 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2641 int16_t *ac_val, *ac_val2;
2644 /* Get DC differential */
2646 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2648 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2651 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2655 if (dcdiff == 119 /* ESC index value */) {
2656 /* TODO: Optimize */
2657 if (v->pq == 1) dcdiff = get_bits(gb, 10);
2658 else if (v->pq == 2) dcdiff = get_bits(gb, 9);
2659 else dcdiff = get_bits(gb, 8);
2662 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2663 else if (v->pq == 2)
2664 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2671 dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
2674 /* Store the quantized DC coeff, used for prediction */
2676 block[0] = dcdiff * s->y_dc_scale;
2678 block[0] = dcdiff * s->c_dc_scale;
2689 int last = 0, skip, value;
2690 const uint8_t *zz_table;
2694 scale = v->pq * 2 + v->halfpq;
2698 zz_table = v->zz_8x8[2];
2700 zz_table = v->zz_8x8[3];
2702 zz_table = v->zz_8x8[1];
2704 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2706 if (dc_pred_dir) // left
2709 ac_val -= 16 * s->block_wrap[n];
2712 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2716 block[zz_table[i++]] = value;
2719 /* apply AC prediction if needed */
2721 if (dc_pred_dir) { // left
2722 for (k = 1; k < 8; k++)
2723 block[k << v->left_blk_sh] += ac_val[k];
2725 for (k = 1; k < 8; k++)
2726 block[k << v->top_blk_sh] += ac_val[k + 8];
2729 /* save AC coeffs for further prediction */
2730 for (k = 1; k < 8; k++) {
2731 ac_val2[k] = block[k << v->left_blk_sh];
2732 ac_val2[k + 8] = block[k << v->top_blk_sh];
2735 /* scale AC coeffs */
2736 for (k = 1; k < 64; k++)
2740 block[k] += (block[k] < 0) ? -v->pq : v->pq;
2743 if (s->ac_pred) i = 63;
2749 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2753 scale = v->pq * 2 + v->halfpq;
2754 memset(ac_val2, 0, 16 * 2);
2755 if (dc_pred_dir) { // left
2758 memcpy(ac_val2, ac_val, 8 * 2);
2760 ac_val -= 16 * s->block_wrap[n];
2762 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2765 /* apply AC prediction if needed */
2767 if (dc_pred_dir) { //left
2768 for (k = 1; k < 8; k++) {
2769 block[k << v->left_blk_sh] = ac_val[k] * scale;
2770 if (!v->pquantizer && block[k << v->left_blk_sh])
2771 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -v->pq : v->pq;
2774 for (k = 1; k < 8; k++) {
2775 block[k << v->top_blk_sh] = ac_val[k + 8] * scale;
2776 if (!v->pquantizer && block[k << v->top_blk_sh])
2777 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -v->pq : v->pq;
2783 s->block_last_index[n] = i;
2788 /** Decode intra block in intra frames - should be faster than decode_intra_block
2789 * @param v VC1Context
2790 * @param block block to decode
2791 * @param[in] n subblock number
2792 * @param coded are AC coeffs present or not
2793 * @param codingset set of VLC to decode data
2794 * @param mquant quantizer value for this macroblock
2796 static int vc1_decode_i_block_adv(VC1Context *v, int16_t block[64], int n,
2797 int coded, int codingset, int mquant)
2799 GetBitContext *gb = &v->s.gb;
2800 MpegEncContext *s = &v->s;
2801 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2803 int16_t *dc_val = NULL;
2804 int16_t *ac_val, *ac_val2;
2806 int a_avail = v->a_avail, c_avail = v->c_avail;
2807 int use_pred = s->ac_pred;
2810 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2812 /* Get DC differential */
2814 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2816 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2819 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2823 if (dcdiff == 119 /* ESC index value */) {
2824 /* TODO: Optimize */
2825 if (mquant == 1) dcdiff = get_bits(gb, 10);
2826 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2827 else dcdiff = get_bits(gb, 8);
2830 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2831 else if (mquant == 2)
2832 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
2839 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
2842 /* Store the quantized DC coeff, used for prediction */
2844 block[0] = dcdiff * s->y_dc_scale;
2846 block[0] = dcdiff * s->c_dc_scale;
2852 /* check if AC is needed at all */
2853 if (!a_avail && !c_avail)
2855 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2858 scale = mquant * 2 + ((mquant == v->pq) ? v->halfpq : 0);
2860 if (dc_pred_dir) // left
2863 ac_val -= 16 * s->block_wrap[n];
2865 q1 = s->current_picture.qscale_table[mb_pos];
2866 if ( dc_pred_dir && c_avail && mb_pos)
2867 q2 = s->current_picture.qscale_table[mb_pos - 1];
2868 if (!dc_pred_dir && a_avail && mb_pos >= s->mb_stride)
2869 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2870 if ( dc_pred_dir && n == 1)
2872 if (!dc_pred_dir && n == 2)
2878 int last = 0, skip, value;
2879 const uint8_t *zz_table;
2883 if (!use_pred && v->fcm == ILACE_FRAME) {
2884 zz_table = v->zzi_8x8;
2886 if (!dc_pred_dir) // top
2887 zz_table = v->zz_8x8[2];
2889 zz_table = v->zz_8x8[3];
2892 if (v->fcm != ILACE_FRAME)
2893 zz_table = v->zz_8x8[1];
2895 zz_table = v->zzi_8x8;
2899 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2903 block[zz_table[i++]] = value;
2906 /* apply AC prediction if needed */
2908 /* scale predictors if needed*/
2909 if (q2 && q1 != q2) {
2910 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2911 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2914 return AVERROR_INVALIDDATA;
2915 if (dc_pred_dir) { // left
2916 for (k = 1; k < 8; k++)
2917 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2919 for (k = 1; k < 8; k++)
2920 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2923 if (dc_pred_dir) { //left
2924 for (k = 1; k < 8; k++)
2925 block[k << v->left_blk_sh] += ac_val[k];
2927 for (k = 1; k < 8; k++)
2928 block[k << v->top_blk_sh] += ac_val[k + 8];
2932 /* save AC coeffs for further prediction */
2933 for (k = 1; k < 8; k++) {
2934 ac_val2[k ] = block[k << v->left_blk_sh];
2935 ac_val2[k + 8] = block[k << v->top_blk_sh];
2938 /* scale AC coeffs */
2939 for (k = 1; k < 64; k++)
2943 block[k] += (block[k] < 0) ? -mquant : mquant;
2946 if (use_pred) i = 63;
2947 } else { // no AC coeffs
2950 memset(ac_val2, 0, 16 * 2);
2951 if (dc_pred_dir) { // left
2953 memcpy(ac_val2, ac_val, 8 * 2);
2954 if (q2 && q1 != q2) {
2955 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2956 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2958 return AVERROR_INVALIDDATA;
2959 for (k = 1; k < 8; k++)
2960 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2965 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2966 if (q2 && q1 != q2) {
2967 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2968 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2970 return AVERROR_INVALIDDATA;
2971 for (k = 1; k < 8; k++)
2972 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2977 /* apply AC prediction if needed */
2979 if (dc_pred_dir) { // left
2980 for (k = 1; k < 8; k++) {
2981 block[k << v->left_blk_sh] = ac_val2[k] * scale;
2982 if (!v->pquantizer && block[k << v->left_blk_sh])
2983 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
2986 for (k = 1; k < 8; k++) {
2987 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
2988 if (!v->pquantizer && block[k << v->top_blk_sh])
2989 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
2995 s->block_last_index[n] = i;
3000 /** Decode intra block in inter frames - more generic version than vc1_decode_i_block
3001 * @param v VC1Context
3002 * @param block block to decode
3003 * @param[in] n subblock index
3004 * @param coded are AC coeffs present or not
3005 * @param mquant block quantizer
3006 * @param codingset set of VLC to decode data
3008 static int vc1_decode_intra_block(VC1Context *v, int16_t block[64], int n,
3009 int coded, int mquant, int codingset)
3011 GetBitContext *gb = &v->s.gb;
3012 MpegEncContext *s = &v->s;
3013 int dc_pred_dir = 0; /* Direction of the DC prediction used */
3015 int16_t *dc_val = NULL;
3016 int16_t *ac_val, *ac_val2;
3018 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
3019 int a_avail = v->a_avail, c_avail = v->c_avail;
3020 int use_pred = s->ac_pred;
3024 s->dsp.clear_block(block);
3026 /* XXX: Guard against dumb values of mquant */
3027 mquant = (mquant < 1) ? 0 : ((mquant > 31) ? 31 : mquant);
3029 /* Set DC scale - y and c use the same */
3030 s->y_dc_scale = s->y_dc_scale_table[mquant];
3031 s->c_dc_scale = s->c_dc_scale_table[mquant];
3033 /* Get DC differential */
3035 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
3037 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
3040 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
3044 if (dcdiff == 119 /* ESC index value */) {
3045 /* TODO: Optimize */
3046 if (mquant == 1) dcdiff = get_bits(gb, 10);
3047 else if (mquant == 2) dcdiff = get_bits(gb, 9);
3048 else dcdiff = get_bits(gb, 8);
3051 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
3052 else if (mquant == 2)
3053 dcdiff = (dcdiff << 1) + get_bits1(gb) - 1;
3060 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
3063 /* Store the quantized DC coeff, used for prediction */
3066 block[0] = dcdiff * s->y_dc_scale;
3068 block[0] = dcdiff * s->c_dc_scale;
3074 /* check if AC is needed at all and adjust direction if needed */
3075 if (!a_avail) dc_pred_dir = 1;
3076 if (!c_avail) dc_pred_dir = 0;
3077 if (!a_avail && !c_avail) use_pred = 0;
3078 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
3081 scale = mquant * 2 + v->halfpq;
3083 if (dc_pred_dir) //left
3086 ac_val -= 16 * s->block_wrap[n];
3088 q1 = s->current_picture.qscale_table[mb_pos];
3089 if (dc_pred_dir && c_avail && mb_pos)
3090 q2 = s->current_picture.qscale_table[mb_pos - 1];
3091 if (!dc_pred_dir && a_avail && mb_pos >= s->mb_stride)
3092 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
3093 if ( dc_pred_dir && n == 1)
3095 if (!dc_pred_dir && n == 2)
3097 if (n == 3) q2 = q1;
3100 int last = 0, skip, value;
3104 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
3108 if (v->fcm == PROGRESSIVE)
3109 block[v->zz_8x8[0][i++]] = value;
3111 if (use_pred && (v->fcm == ILACE_FRAME)) {
3112 if (!dc_pred_dir) // top
3113 block[v->zz_8x8[2][i++]] = value;
3115 block[v->zz_8x8[3][i++]] = value;
3117 block[v->zzi_8x8[i++]] = value;
3122 /* apply AC prediction if needed */
3124 /* scale predictors if needed*/
3125 if (q2 && q1 != q2) {
3126 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3127 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3130 return AVERROR_INVALIDDATA;
3131 if (dc_pred_dir) { // left
3132 for (k = 1; k < 8; k++)
3133 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3135 for (k = 1; k < 8; k++)
3136 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3139 if (dc_pred_dir) { // left
3140 for (k = 1; k < 8; k++)
3141 block[k << v->left_blk_sh] += ac_val[k];
3143 for (k = 1; k < 8; k++)
3144 block[k << v->top_blk_sh] += ac_val[k + 8];
3148 /* save AC coeffs for further prediction */
3149 for (k = 1; k < 8; k++) {
3150 ac_val2[k ] = block[k << v->left_blk_sh];
3151 ac_val2[k + 8] = block[k << v->top_blk_sh];
3154 /* scale AC coeffs */
3155 for (k = 1; k < 64; k++)
3159 block[k] += (block[k] < 0) ? -mquant : mquant;
3162 if (use_pred) i = 63;
3163 } else { // no AC coeffs
3166 memset(ac_val2, 0, 16 * 2);
3167 if (dc_pred_dir) { // left
3169 memcpy(ac_val2, ac_val, 8 * 2);
3170 if (q2 && q1 != q2) {
3171 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3172 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3174 return AVERROR_INVALIDDATA;
3175 for (k = 1; k < 8; k++)
3176 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3181 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
3182 if (q2 && q1 != q2) {
3183 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
3184 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
3186 return AVERROR_INVALIDDATA;
3187 for (k = 1; k < 8; k++)
3188 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
3193 /* apply AC prediction if needed */
3195 if (dc_pred_dir) { // left
3196 for (k = 1; k < 8; k++) {
3197 block[k << v->left_blk_sh] = ac_val2[k] * scale;
3198 if (!v->pquantizer && block[k << v->left_blk_sh])
3199 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
3202 for (k = 1; k < 8; k++) {
3203 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
3204 if (!v->pquantizer && block[k << v->top_blk_sh])
3205 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
3211 s->block_last_index[n] = i;
3218 static int vc1_decode_p_block(VC1Context *v, int16_t block[64], int n,
3219 int mquant, int ttmb, int first_block,
3220 uint8_t *dst, int linesize, int skip_block,
3223 MpegEncContext *s = &v->s;
3224 GetBitContext *gb = &s->gb;
3227 int scale, off, idx, last, skip, value;
3228 int ttblk = ttmb & 7;
3231 s->dsp.clear_block(block);
3234 ttblk = ff_vc1_ttblk_to_tt[v->tt_index][get_vlc2(gb, ff_vc1_ttblk_vlc[v->tt_index].table, VC1_TTBLK_VLC_BITS, 1)];
3236 if (ttblk == TT_4X4) {
3237 subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
3239 if ((ttblk != TT_8X8 && ttblk != TT_4X4)
3240 && ((v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))
3241 || (!v->res_rtm_flag && !first_block))) {
3242 subblkpat = decode012(gb);
3244 subblkpat ^= 3; // swap decoded pattern bits
3245 if (ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM)
3247 if (ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT)
3250 scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);
3252 // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
3253 if (ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
3254 subblkpat = 2 - (ttblk == TT_8X4_TOP);
3257 if (ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
3258 subblkpat = 2 - (ttblk == TT_4X8_LEFT);
3267 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3272 idx = v->zz_8x8[0][i++];
3274 idx = v->zzi_8x8[i++];
3275 block[idx] = value * scale;
3277 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3281 v->vc1dsp.vc1_inv_trans_8x8_dc(dst, linesize, block);
3283 v->vc1dsp.vc1_inv_trans_8x8(block);
3284 s->dsp.add_pixels_clamped(block, dst, linesize);
3289 pat = ~subblkpat & 0xF;
3290 for (j = 0; j < 4; j++) {
3291 last = subblkpat & (1 << (3 - j));
3293 off = (j & 1) * 4 + (j & 2) * 16;
3295 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3300 idx = ff_vc1_simple_progressive_4x4_zz[i++];
3302 idx = ff_vc1_adv_interlaced_4x4_zz[i++];
3303 block[idx + off] = value * scale;
3305 block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
3307 if (!(subblkpat & (1 << (3 - j))) && !skip_block) {
3309 v->vc1dsp.vc1_inv_trans_4x4_dc(dst + (j & 1) * 4 + (j & 2) * 2 * linesize, linesize, block + off);
3311 v->vc1dsp.vc1_inv_trans_4x4(dst + (j & 1) * 4 + (j & 2) * 2 * linesize, linesize, block + off);
3316 pat = ~((subblkpat & 2) * 6 + (subblkpat & 1) * 3) & 0xF;
3317 for (j = 0; j < 2; j++) {
3318 last = subblkpat & (1 << (1 - j));
3322 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);