2 * VC-1 and WMV3 decoder
3 * Copyright (c) 2006-2007 Konstantin Shishkov
4 * Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * VC-1 and WMV3 decoder
31 #include "mpegvideo.h"
35 #include "vc1acdata.h"
36 #include "msmpeg4data.h"
38 #include "simple_idct.h"
40 #include "vdpau_internal.h"
45 #define MB_INTRA_VLC_BITS 9
48 static const uint16_t table_mb_intra[64][2];
51 static const uint16_t vlc_offs[] = {
52 0, 520, 552, 616, 1128, 1160, 1224, 1740, 1772, 1836, 1900, 2436,
53 2986, 3050, 3610, 4154, 4218, 4746, 5326, 5390, 5902, 6554, 7658, 8620,
54 9262, 10202, 10756, 11310, 12228, 15078
58 * Init VC-1 specific tables and VC1Context members
59 * @param v The VC1Context to initialize
62 static int vc1_init_common(VC1Context *v)
66 static VLC_TYPE vlc_table[15078][2];
68 v->hrd_rate = v->hrd_buffer = NULL;
73 INIT_VLC_STATIC(&ff_vc1_bfraction_vlc, VC1_BFRACTION_VLC_BITS, 23,
74 ff_vc1_bfraction_bits, 1, 1,
75 ff_vc1_bfraction_codes, 1, 1, 1 << VC1_BFRACTION_VLC_BITS);
76 INIT_VLC_STATIC(&ff_vc1_norm2_vlc, VC1_NORM2_VLC_BITS, 4,
77 ff_vc1_norm2_bits, 1, 1,
78 ff_vc1_norm2_codes, 1, 1, 1 << VC1_NORM2_VLC_BITS);
79 INIT_VLC_STATIC(&ff_vc1_norm6_vlc, VC1_NORM6_VLC_BITS, 64,
80 ff_vc1_norm6_bits, 1, 1,
81 ff_vc1_norm6_codes, 2, 2, 556);
82 INIT_VLC_STATIC(&ff_vc1_imode_vlc, VC1_IMODE_VLC_BITS, 7,
83 ff_vc1_imode_bits, 1, 1,
84 ff_vc1_imode_codes, 1, 1, 1 << VC1_IMODE_VLC_BITS);
87 ff_vc1_ttmb_vlc[i].table = &vlc_table[vlc_offs[i*3+0]];
88 ff_vc1_ttmb_vlc[i].table_allocated = vlc_offs[i*3+1] - vlc_offs[i*3+0];
89 init_vlc(&ff_vc1_ttmb_vlc[i], VC1_TTMB_VLC_BITS, 16,
90 ff_vc1_ttmb_bits[i], 1, 1,
91 ff_vc1_ttmb_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
92 ff_vc1_ttblk_vlc[i].table = &vlc_table[vlc_offs[i*3+1]];
93 ff_vc1_ttblk_vlc[i].table_allocated = vlc_offs[i*3+2] - vlc_offs[i*3+1];
94 init_vlc(&ff_vc1_ttblk_vlc[i], VC1_TTBLK_VLC_BITS, 8,
95 ff_vc1_ttblk_bits[i], 1, 1,
96 ff_vc1_ttblk_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
97 ff_vc1_subblkpat_vlc[i].table = &vlc_table[vlc_offs[i*3+2]];
98 ff_vc1_subblkpat_vlc[i].table_allocated = vlc_offs[i*3+3] - vlc_offs[i*3+2];
99 init_vlc(&ff_vc1_subblkpat_vlc[i], VC1_SUBBLKPAT_VLC_BITS, 15,
100 ff_vc1_subblkpat_bits[i], 1, 1,
101 ff_vc1_subblkpat_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
105 ff_vc1_4mv_block_pattern_vlc[i].table = &vlc_table[vlc_offs[i*3+9]];
106 ff_vc1_4mv_block_pattern_vlc[i].table_allocated = vlc_offs[i*3+10] - vlc_offs[i*3+9];
107 init_vlc(&ff_vc1_4mv_block_pattern_vlc[i], VC1_4MV_BLOCK_PATTERN_VLC_BITS, 16,
108 ff_vc1_4mv_block_pattern_bits[i], 1, 1,
109 ff_vc1_4mv_block_pattern_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
110 ff_vc1_cbpcy_p_vlc[i].table = &vlc_table[vlc_offs[i*3+10]];
111 ff_vc1_cbpcy_p_vlc[i].table_allocated = vlc_offs[i*3+11] - vlc_offs[i*3+10];
112 init_vlc(&ff_vc1_cbpcy_p_vlc[i], VC1_CBPCY_P_VLC_BITS, 64,
113 ff_vc1_cbpcy_p_bits[i], 1, 1,
114 ff_vc1_cbpcy_p_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
115 ff_vc1_mv_diff_vlc[i].table = &vlc_table[vlc_offs[i*3+11]];
116 ff_vc1_mv_diff_vlc[i].table_allocated = vlc_offs[i*3+12] - vlc_offs[i*3+11];
117 init_vlc(&ff_vc1_mv_diff_vlc[i], VC1_MV_DIFF_VLC_BITS, 73,
118 ff_vc1_mv_diff_bits[i], 1, 1,
119 ff_vc1_mv_diff_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
122 ff_vc1_ac_coeff_table[i].table = &vlc_table[vlc_offs[i+21]];
123 ff_vc1_ac_coeff_table[i].table_allocated = vlc_offs[i+22] - vlc_offs[i+21];
124 init_vlc(&ff_vc1_ac_coeff_table[i], AC_VLC_BITS, vc1_ac_sizes[i],
125 &vc1_ac_tables[i][0][1], 8, 4,
126 &vc1_ac_tables[i][0][0], 8, 4, INIT_VLC_USE_NEW_STATIC);
133 v->mvrange = 0; /* 7.1.1.18, p80 */
138 /***********************************************************************/
140 * @defgroup vc1bitplane VC-1 Bitplane decoding
158 /** @} */ //imode defines
161 /** @} */ //Bitplane group
163 static void vc1_put_signed_blocks_clamped(VC1Context *v)
165 MpegEncContext *s = &v->s;
167 /* The put pixels loop is always one MB row behind the decoding loop,
168 * because we can only put pixels when overlap filtering is done, and
169 * for filtering of the bottom edge of a MB, we need the next MB row
171 * Within the row, the put pixels loop is also one MB col behind the
172 * decoding loop. The reason for this is again, because for filtering
173 * of the right MB edge, we need the next MB present. */
174 if (!s->first_slice_line) {
176 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][0],
177 s->dest[0] - 16 * s->linesize - 16,
179 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][1],
180 s->dest[0] - 16 * s->linesize - 8,
182 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][2],
183 s->dest[0] - 8 * s->linesize - 16,
185 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3],
186 s->dest[0] - 8 * s->linesize - 8,
188 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4],
189 s->dest[1] - 8 * s->uvlinesize - 8,
191 s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][5],
192 s->dest[2] - 8 * s->uvlinesize - 8,
195 if (s->mb_x == s->mb_width - 1) {
196 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][0],
197 s->dest[0] - 16 * s->linesize,
199 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][1],
200 s->dest[0] - 16 * s->linesize + 8,
202 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][2],
203 s->dest[0] - 8 * s->linesize,
205 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3],
206 s->dest[0] - 8 * s->linesize + 8,
208 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4],
209 s->dest[1] - 8 * s->uvlinesize,
211 s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][5],
212 s->dest[2] - 8 * s->uvlinesize,
217 #define inc_blk_idx(idx) do { \
219 if (idx >= v->n_allocated_blks) \
223 inc_blk_idx(v->topleft_blk_idx);
224 inc_blk_idx(v->top_blk_idx);
225 inc_blk_idx(v->left_blk_idx);
226 inc_blk_idx(v->cur_blk_idx);
229 static void vc1_loop_filter_iblk(VC1Context *v, int pq)
231 MpegEncContext *s = &v->s;
233 if (!s->first_slice_line) {
234 v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
236 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16*s->linesize, s->linesize, pq);
237 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16*s->linesize+8, s->linesize, pq);
238 for(j = 0; j < 2; j++){
239 v->vc1dsp.vc1_v_loop_filter8(s->dest[j+1], s->uvlinesize, pq);
241 v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1]-8*s->uvlinesize, s->uvlinesize, pq);
244 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8*s->linesize, s->linesize, pq);
246 if (s->mb_y == s->mb_height-1) {
248 v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
249 v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
250 v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
252 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
256 static void vc1_loop_filter_iblk_delayed(VC1Context *v, int pq)
258 MpegEncContext *s = &v->s;
261 /* The loopfilter runs 1 row and 1 column behind the overlap filter, which
262 * means it runs two rows/cols behind the decoding loop. */
263 if (!s->first_slice_line) {
265 if (s->mb_y >= s->start_mb_y + 2) {
266 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
269 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
270 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
271 for(j = 0; j < 2; j++) {
272 v->vc1dsp.vc1_v_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
274 v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq);
278 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq);
281 if (s->mb_x == s->mb_width - 1) {
282 if (s->mb_y >= s->start_mb_y + 2) {
283 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
286 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
287 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
288 for(j = 0; j < 2; j++) {
289 v->vc1dsp.vc1_v_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize, s->uvlinesize, pq);
291 v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 16 * s->uvlinesize, s->uvlinesize, pq);
295 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq);
298 if (s->mb_y == s->mb_height) {
301 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
302 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
304 for(j = 0; j < 2; j++) {
305 v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
310 if (s->mb_x == s->mb_width - 1) {
312 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
313 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
315 for(j = 0; j < 2; j++) {
316 v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize, s->uvlinesize, pq);
324 static void vc1_smooth_overlap_filter_iblk(VC1Context *v)
326 MpegEncContext *s = &v->s;
329 if (v->condover == CONDOVER_NONE)
332 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
334 /* Within a MB, the horizontal overlap always runs before the vertical.
335 * To accomplish that, we run the H on left and internal borders of the
336 * currently decoded MB. Then, we wait for the next overlap iteration
337 * to do H overlap on the right edge of this MB, before moving over and
338 * running the V overlap. Therefore, the V overlap makes us trail by one
339 * MB col and the H overlap filter makes us trail by one MB row. This
340 * is reflected in the time at which we run the put_pixels loop. */
341 if(v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) {
342 if(s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
343 v->over_flags_plane[mb_pos - 1])) {
344 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1],
345 v->block[v->cur_blk_idx][0]);
346 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
347 v->block[v->cur_blk_idx][2]);
348 if(!(s->flags & CODEC_FLAG_GRAY)) {
349 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
350 v->block[v->cur_blk_idx][4]);
351 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
352 v->block[v->cur_blk_idx][5]);
355 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0],
356 v->block[v->cur_blk_idx][1]);
357 v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2],
358 v->block[v->cur_blk_idx][3]);
360 if (s->mb_x == s->mb_width - 1) {
361 if(!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
362 v->over_flags_plane[mb_pos - s->mb_stride])) {
363 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2],
364 v->block[v->cur_blk_idx][0]);
365 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
366 v->block[v->cur_blk_idx][1]);
367 if(!(s->flags & CODEC_FLAG_GRAY)) {
368 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
369 v->block[v->cur_blk_idx][4]);
370 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
371 v->block[v->cur_blk_idx][5]);
374 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0],
375 v->block[v->cur_blk_idx][2]);
376 v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1],
377 v->block[v->cur_blk_idx][3]);
380 if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) {
381 if(!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
382 v->over_flags_plane[mb_pos - s->mb_stride - 1])) {
383 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2],
384 v->block[v->left_blk_idx][0]);
385 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
386 v->block[v->left_blk_idx][1]);
387 if(!(s->flags & CODEC_FLAG_GRAY)) {
388 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
389 v->block[v->left_blk_idx][4]);
390 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
391 v->block[v->left_blk_idx][5]);
394 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0],
395 v->block[v->left_blk_idx][2]);
396 v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1],
397 v->block[v->left_blk_idx][3]);
401 /** Do motion compensation over 1 macroblock
402 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
404 static void vc1_mc_1mv(VC1Context *v, int dir)
406 MpegEncContext *s = &v->s;
407 DSPContext *dsp = &v->s.dsp;
408 uint8_t *srcY, *srcU, *srcV;
409 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
411 if(!v->s.last_picture.data[0])return;
413 mx = s->mv[dir][0][0];
414 my = s->mv[dir][0][1];
416 // store motion vectors for further use in B frames
417 if(s->pict_type == AV_PICTURE_TYPE_P) {
418 s->current_picture.motion_val[1][s->block_index[0]][0] = mx;
419 s->current_picture.motion_val[1][s->block_index[0]][1] = my;
421 uvmx = (mx + ((mx & 3) == 3)) >> 1;
422 uvmy = (my + ((my & 3) == 3)) >> 1;
423 v->luma_mv[s->mb_x][0] = uvmx;
424 v->luma_mv[s->mb_x][1] = uvmy;
426 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
427 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
430 srcY = s->last_picture.data[0];
431 srcU = s->last_picture.data[1];
432 srcV = s->last_picture.data[2];
434 srcY = s->next_picture.data[0];
435 srcU = s->next_picture.data[1];
436 srcV = s->next_picture.data[2];
439 src_x = s->mb_x * 16 + (mx >> 2);
440 src_y = s->mb_y * 16 + (my >> 2);
441 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
442 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
444 if(v->profile != PROFILE_ADVANCED){
445 src_x = av_clip( src_x, -16, s->mb_width * 16);
446 src_y = av_clip( src_y, -16, s->mb_height * 16);
447 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
448 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
450 src_x = av_clip( src_x, -17, s->avctx->coded_width);
451 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
452 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
453 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
456 srcY += src_y * s->linesize + src_x;
457 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
458 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
460 /* for grayscale we should not try to read from unknown area */
461 if(s->flags & CODEC_FLAG_GRAY) {
462 srcU = s->edge_emu_buffer + 18 * s->linesize;
463 srcV = s->edge_emu_buffer + 18 * s->linesize;
466 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
467 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
468 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
469 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
471 srcY -= s->mspel * (1 + s->linesize);
472 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
473 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
474 srcY = s->edge_emu_buffer;
475 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
476 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
477 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
478 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
481 /* if we deal with range reduction we need to scale source blocks */
487 for(j = 0; j < 17 + s->mspel*2; j++) {
488 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
491 src = srcU; src2 = srcV;
492 for(j = 0; j < 9; j++) {
493 for(i = 0; i < 9; i++) {
494 src[i] = ((src[i] - 128) >> 1) + 128;
495 src2[i] = ((src2[i] - 128) >> 1) + 128;
497 src += s->uvlinesize;
498 src2 += s->uvlinesize;
501 /* if we deal with intensity compensation we need to scale source blocks */
502 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
507 for(j = 0; j < 17 + s->mspel*2; j++) {
508 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = v->luty[src[i]];
511 src = srcU; src2 = srcV;
512 for(j = 0; j < 9; j++) {
513 for(i = 0; i < 9; i++) {
514 src[i] = v->lutuv[src[i]];
515 src2[i] = v->lutuv[src2[i]];
517 src += s->uvlinesize;
518 src2 += s->uvlinesize;
521 srcY += s->mspel * (1 + s->linesize);
525 dxy = ((my & 3) << 2) | (mx & 3);
526 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
527 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
528 srcY += s->linesize * 8;
529 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
530 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
531 } else { // hpel mc - always used for luma
532 dxy = (my & 2) | ((mx & 2) >> 1);
535 dsp->put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
537 dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
540 if(s->flags & CODEC_FLAG_GRAY) return;
541 /* Chroma MC always uses qpel bilinear */
545 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
546 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
548 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
549 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
553 /** Do motion compensation for 4-MV macroblock - luminance block
555 static void vc1_mc_4mv_luma(VC1Context *v, int n)
557 MpegEncContext *s = &v->s;
558 DSPContext *dsp = &v->s.dsp;
560 int dxy, mx, my, src_x, src_y;
563 if(!v->s.last_picture.data[0])return;
566 srcY = s->last_picture.data[0];
568 off = s->linesize * 4 * (n&2) + (n&1) * 8;
570 src_x = s->mb_x * 16 + (n&1) * 8 + (mx >> 2);
571 src_y = s->mb_y * 16 + (n&2) * 4 + (my >> 2);
573 if(v->profile != PROFILE_ADVANCED){
574 src_x = av_clip( src_x, -16, s->mb_width * 16);
575 src_y = av_clip( src_y, -16, s->mb_height * 16);
577 src_x = av_clip( src_x, -17, s->avctx->coded_width);
578 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
581 srcY += src_y * s->linesize + src_x;
583 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
584 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 8 - s->mspel*2
585 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 8 - s->mspel*2){
586 srcY -= s->mspel * (1 + s->linesize);
587 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 9+s->mspel*2, 9+s->mspel*2,
588 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
589 srcY = s->edge_emu_buffer;
590 /* if we deal with range reduction we need to scale source blocks */
596 for(j = 0; j < 9 + s->mspel*2; j++) {
597 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
601 /* if we deal with intensity compensation we need to scale source blocks */
602 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
607 for(j = 0; j < 9 + s->mspel*2; j++) {
608 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = v->luty[src[i]];
612 srcY += s->mspel * (1 + s->linesize);
616 dxy = ((my & 3) << 2) | (mx & 3);
617 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, v->rnd);
618 } else { // hpel mc - always used for luma
619 dxy = (my & 2) | ((mx & 2) >> 1);
621 dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
623 dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
627 static inline int median4(int a, int b, int c, int d)
630 if(c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
631 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
633 if(c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
634 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
639 /** Do motion compensation for 4-MV macroblock - both chroma blocks
641 static void vc1_mc_4mv_chroma(VC1Context *v)
643 MpegEncContext *s = &v->s;
644 DSPContext *dsp = &v->s.dsp;
645 uint8_t *srcU, *srcV;
646 int uvmx, uvmy, uvsrc_x, uvsrc_y;
647 int i, idx, tx = 0, ty = 0;
648 int mvx[4], mvy[4], intra[4];
649 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
651 if(!v->s.last_picture.data[0])return;
652 if(s->flags & CODEC_FLAG_GRAY) return;
654 for(i = 0; i < 4; i++) {
655 mvx[i] = s->mv[0][i][0];
656 mvy[i] = s->mv[0][i][1];
657 intra[i] = v->mb_type[0][s->block_index[i]];
660 /* calculate chroma MV vector from four luma MVs */
661 idx = (intra[3] << 3) | (intra[2] << 2) | (intra[1] << 1) | intra[0];
662 if(!idx) { // all blocks are inter
663 tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
664 ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
665 } else if(count[idx] == 1) { // 3 inter blocks
668 tx = mid_pred(mvx[1], mvx[2], mvx[3]);
669 ty = mid_pred(mvy[1], mvy[2], mvy[3]);
672 tx = mid_pred(mvx[0], mvx[2], mvx[3]);
673 ty = mid_pred(mvy[0], mvy[2], mvy[3]);
676 tx = mid_pred(mvx[0], mvx[1], mvx[3]);
677 ty = mid_pred(mvy[0], mvy[1], mvy[3]);
680 tx = mid_pred(mvx[0], mvx[1], mvx[2]);
681 ty = mid_pred(mvy[0], mvy[1], mvy[2]);
684 } else if(count[idx] == 2) {
686 for(i=0; i<3;i++) if(!intra[i]) {t1 = i; break;}
687 for(i= t1+1; i<4; i++)if(!intra[i]) {t2 = i; break;}
688 tx = (mvx[t1] + mvx[t2]) / 2;
689 ty = (mvy[t1] + mvy[t2]) / 2;
691 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
692 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
693 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
694 return; //no need to do MC for inter blocks
697 s->current_picture.motion_val[1][s->block_index[0]][0] = tx;
698 s->current_picture.motion_val[1][s->block_index[0]][1] = ty;
699 uvmx = (tx + ((tx&3) == 3)) >> 1;
700 uvmy = (ty + ((ty&3) == 3)) >> 1;
701 v->luma_mv[s->mb_x][0] = uvmx;
702 v->luma_mv[s->mb_x][1] = uvmy;
704 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
705 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
708 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
709 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
711 if(v->profile != PROFILE_ADVANCED){
712 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
713 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
715 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
716 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
719 srcU = s->last_picture.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
720 srcV = s->last_picture.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
721 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
722 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
723 || (unsigned)uvsrc_y > (s->v_edge_pos >> 1) - 9){
724 s->dsp.emulated_edge_mc(s->edge_emu_buffer , srcU, s->uvlinesize, 8+1, 8+1,
725 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
726 s->dsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize, 8+1, 8+1,
727 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
728 srcU = s->edge_emu_buffer;
729 srcV = s->edge_emu_buffer + 16;
731 /* if we deal with range reduction we need to scale source blocks */
736 src = srcU; src2 = srcV;
737 for(j = 0; j < 9; j++) {
738 for(i = 0; i < 9; i++) {
739 src[i] = ((src[i] - 128) >> 1) + 128;
740 src2[i] = ((src2[i] - 128) >> 1) + 128;
742 src += s->uvlinesize;
743 src2 += s->uvlinesize;
746 /* if we deal with intensity compensation we need to scale source blocks */
747 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
751 src = srcU; src2 = srcV;
752 for(j = 0; j < 9; j++) {
753 for(i = 0; i < 9; i++) {
754 src[i] = v->lutuv[src[i]];
755 src2[i] = v->lutuv[src2[i]];
757 src += s->uvlinesize;
758 src2 += s->uvlinesize;
763 /* Chroma MC always uses qpel bilinear */
767 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
768 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
770 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
771 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
775 /***********************************************************************/
777 * @defgroup vc1block VC-1 Block-level functions
778 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
784 * @brief Get macroblock-level quantizer scale
786 #define GET_MQUANT() \
790 if (v->dqprofile == DQPROFILE_ALL_MBS) \
794 mquant = (get_bits1(gb)) ? v->altpq : v->pq; \
798 mqdiff = get_bits(gb, 3); \
799 if (mqdiff != 7) mquant = v->pq + mqdiff; \
800 else mquant = get_bits(gb, 5); \
803 if(v->dqprofile == DQPROFILE_SINGLE_EDGE) \
804 edges = 1 << v->dqsbedge; \
805 else if(v->dqprofile == DQPROFILE_DOUBLE_EDGES) \
806 edges = (3 << v->dqsbedge) % 15; \
807 else if(v->dqprofile == DQPROFILE_FOUR_EDGES) \
809 if((edges&1) && !s->mb_x) \
811 if((edges&2) && s->first_slice_line) \
813 if((edges&4) && s->mb_x == (s->mb_width - 1)) \
815 if((edges&8) && s->mb_y == (s->mb_height - 1)) \
820 * @def GET_MVDATA(_dmv_x, _dmv_y)
821 * @brief Get MV differentials
822 * @see MVDATA decoding from 8.3.5.2, p(1)20
823 * @param _dmv_x Horizontal differential for decoded MV
824 * @param _dmv_y Vertical differential for decoded MV
826 #define GET_MVDATA(_dmv_x, _dmv_y) \
827 index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table,\
828 VC1_MV_DIFF_VLC_BITS, 2); \
834 else mb_has_coeffs = 0; \
836 if (!index) { _dmv_x = _dmv_y = 0; } \
837 else if (index == 35) \
839 _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample); \
840 _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample); \
842 else if (index == 36) \
851 if (!s->quarter_sample && index1 == 5) val = 1; \
853 if(size_table[index1] - val > 0) \
854 val = get_bits(gb, size_table[index1] - val); \
856 sign = 0 - (val&1); \
857 _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
860 if (!s->quarter_sample && index1 == 5) val = 1; \
862 if(size_table[index1] - val > 0) \
863 val = get_bits(gb, size_table[index1] - val); \
865 sign = 0 - (val&1); \
866 _dmv_y = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
869 /** Predict and set motion vector
871 static inline void vc1_pred_mv(VC1Context *v, int n, int dmv_x, int dmv_y, int mv1, int r_x, int r_y, uint8_t* is_intra)
873 MpegEncContext *s = &v->s;
874 int xy, wrap, off = 0;
879 /* scale MV difference to be quad-pel */
880 dmv_x <<= 1 - s->quarter_sample;
881 dmv_y <<= 1 - s->quarter_sample;
884 xy = s->block_index[n];
887 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = 0;
888 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = 0;
889 s->current_picture.motion_val[1][xy][0] = 0;
890 s->current_picture.motion_val[1][xy][1] = 0;
891 if(mv1) { /* duplicate motion data for 1-MV block */
892 s->current_picture.motion_val[0][xy + 1][0] = 0;
893 s->current_picture.motion_val[0][xy + 1][1] = 0;
894 s->current_picture.motion_val[0][xy + wrap][0] = 0;
895 s->current_picture.motion_val[0][xy + wrap][1] = 0;
896 s->current_picture.motion_val[0][xy + wrap + 1][0] = 0;
897 s->current_picture.motion_val[0][xy + wrap + 1][1] = 0;
898 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
899 s->current_picture.motion_val[1][xy + 1][0] = 0;
900 s->current_picture.motion_val[1][xy + 1][1] = 0;
901 s->current_picture.motion_val[1][xy + wrap][0] = 0;
902 s->current_picture.motion_val[1][xy + wrap][1] = 0;
903 s->current_picture.motion_val[1][xy + wrap + 1][0] = 0;
904 s->current_picture.motion_val[1][xy + wrap + 1][1] = 0;
909 C = s->current_picture.motion_val[0][xy - 1];
910 A = s->current_picture.motion_val[0][xy - wrap];
912 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
914 //in 4-MV mode different blocks have different B predictor position
917 off = (s->mb_x > 0) ? -1 : 1;
920 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
929 B = s->current_picture.motion_val[0][xy - wrap + off];
931 if(!s->first_slice_line || (n==2 || n==3)) { // predictor A is not out of bounds
932 if(s->mb_width == 1) {
936 px = mid_pred(A[0], B[0], C[0]);
937 py = mid_pred(A[1], B[1], C[1]);
939 } else if(s->mb_x || (n==1 || n==3)) { // predictor C is not out of bounds
945 /* Pullback MV as specified in 8.3.5.3.4 */
948 qx = (s->mb_x << 6) + ((n==1 || n==3) ? 32 : 0);
949 qy = (s->mb_y << 6) + ((n==2 || n==3) ? 32 : 0);
950 X = (s->mb_width << 6) - 4;
951 Y = (s->mb_height << 6) - 4;
953 if(qx + px < -60) px = -60 - qx;
954 if(qy + py < -60) py = -60 - qy;
956 if(qx + px < -28) px = -28 - qx;
957 if(qy + py < -28) py = -28 - qy;
959 if(qx + px > X) px = X - qx;
960 if(qy + py > Y) py = Y - qy;
962 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
963 if((!s->first_slice_line || (n==2 || n==3)) && (s->mb_x || (n==1 || n==3))) {
964 if(is_intra[xy - wrap])
965 sum = FFABS(px) + FFABS(py);
967 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
969 if(get_bits1(&s->gb)) {
978 sum = FFABS(px) + FFABS(py);
980 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
982 if(get_bits1(&s->gb)) {
992 /* store MV using signed modulus of MV range defined in 4.11 */
993 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
994 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
995 if(mv1) { /* duplicate motion data for 1-MV block */
996 s->current_picture.motion_val[0][xy + 1][0] = s->current_picture.motion_val[0][xy][0];
997 s->current_picture.motion_val[0][xy + 1][1] = s->current_picture.motion_val[0][xy][1];
998 s->current_picture.motion_val[0][xy + wrap][0] = s->current_picture.motion_val[0][xy][0];
999 s->current_picture.motion_val[0][xy + wrap][1] = s->current_picture.motion_val[0][xy][1];
1000 s->current_picture.motion_val[0][xy + wrap + 1][0] = s->current_picture.motion_val[0][xy][0];
1001 s->current_picture.motion_val[0][xy + wrap + 1][1] = s->current_picture.motion_val[0][xy][1];
1005 /** Motion compensation for direct or interpolated blocks in B-frames
1007 static void vc1_interp_mc(VC1Context *v)
1009 MpegEncContext *s = &v->s;
1010 DSPContext *dsp = &v->s.dsp;
1011 uint8_t *srcY, *srcU, *srcV;
1012 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
1014 if(!v->s.next_picture.data[0])return;
1016 mx = s->mv[1][0][0];
1017 my = s->mv[1][0][1];
1018 uvmx = (mx + ((mx & 3) == 3)) >> 1;
1019 uvmy = (my + ((my & 3) == 3)) >> 1;
1021 uvmx = uvmx + ((uvmx<0)?-(uvmx&1):(uvmx&1));
1022 uvmy = uvmy + ((uvmy<0)?-(uvmy&1):(uvmy&1));
1024 srcY = s->next_picture.data[0];
1025 srcU = s->next_picture.data[1];
1026 srcV = s->next_picture.data[2];
1028 src_x = s->mb_x * 16 + (mx >> 2);
1029 src_y = s->mb_y * 16 + (my >> 2);
1030 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
1031 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
1033 if(v->profile != PROFILE_ADVANCED){
1034 src_x = av_clip( src_x, -16, s->mb_width * 16);
1035 src_y = av_clip( src_y, -16, s->mb_height * 16);
1036 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
1037 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
1039 src_x = av_clip( src_x, -17, s->avctx->coded_width);
1040 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
1041 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
1042 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
1045 srcY += src_y * s->linesize + src_x;
1046 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1047 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1049 /* for grayscale we should not try to read from unknown area */
1050 if(s->flags & CODEC_FLAG_GRAY) {
1051 srcU = s->edge_emu_buffer + 18 * s->linesize;
1052 srcV = s->edge_emu_buffer + 18 * s->linesize;
1056 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
1057 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
1058 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
1060 srcY -= s->mspel * (1 + s->linesize);
1061 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
1062 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
1063 srcY = s->edge_emu_buffer;
1064 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
1065 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
1066 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
1067 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
1070 /* if we deal with range reduction we need to scale source blocks */
1071 if(v->rangeredfrm) {
1073 uint8_t *src, *src2;
1076 for(j = 0; j < 17 + s->mspel*2; j++) {
1077 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
1080 src = srcU; src2 = srcV;
1081 for(j = 0; j < 9; j++) {
1082 for(i = 0; i < 9; i++) {
1083 src[i] = ((src[i] - 128) >> 1) + 128;
1084 src2[i] = ((src2[i] - 128) >> 1) + 128;
1086 src += s->uvlinesize;
1087 src2 += s->uvlinesize;
1090 srcY += s->mspel * (1 + s->linesize);
1094 dxy = ((my & 3) << 2) | (mx & 3);
1095 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
1096 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
1097 srcY += s->linesize * 8;
1098 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
1099 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
1101 dxy = (my & 2) | ((mx & 2) >> 1);
1104 dsp->avg_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
1106 dsp->avg_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
1109 if(s->flags & CODEC_FLAG_GRAY) return;
1110 /* Chroma MC always uses qpel blilinear */
1114 dsp->avg_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
1115 dsp->avg_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
1117 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
1118 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
1122 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
1126 #if B_FRACTION_DEN==256
1130 return 2 * ((value * n + 255) >> 9);
1131 return (value * n + 128) >> 8;
1134 n -= B_FRACTION_DEN;
1136 return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
1137 return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
1141 /** Reconstruct motion vector for B-frame and do motion compensation
1143 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mode)
1146 v->mv_mode2 = v->mv_mode;
1147 v->mv_mode = MV_PMODE_INTENSITY_COMP;
1152 if(v->use_ic) v->mv_mode = v->mv_mode2;
1155 if(mode == BMV_TYPE_INTERPOLATED) {
1158 if(v->use_ic) v->mv_mode = v->mv_mode2;
1162 if(v->use_ic && (mode == BMV_TYPE_BACKWARD)) v->mv_mode = v->mv_mode2;
1163 vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
1164 if(v->use_ic) v->mv_mode = v->mv_mode2;
1167 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mvtype)
1169 MpegEncContext *s = &v->s;
1170 int xy, wrap, off = 0;
1175 const uint8_t *is_intra = v->mb_type[0];
1179 /* scale MV difference to be quad-pel */
1180 dmv_x[0] <<= 1 - s->quarter_sample;
1181 dmv_y[0] <<= 1 - s->quarter_sample;
1182 dmv_x[1] <<= 1 - s->quarter_sample;
1183 dmv_y[1] <<= 1 - s->quarter_sample;
1185 wrap = s->b8_stride;
1186 xy = s->block_index[0];
1189 s->current_picture.motion_val[0][xy][0] =
1190 s->current_picture.motion_val[0][xy][1] =
1191 s->current_picture.motion_val[1][xy][0] =
1192 s->current_picture.motion_val[1][xy][1] = 0;
1195 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
1196 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
1197 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
1198 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
1200 /* Pullback predicted motion vectors as specified in 8.4.5.4 */
1201 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));
1202 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));
1203 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));
1204 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));
1206 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
1207 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
1208 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
1209 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
1213 if((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
1214 C = s->current_picture.motion_val[0][xy - 2];
1215 A = s->current_picture.motion_val[0][xy - wrap*2];
1216 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1217 B = s->current_picture.motion_val[0][xy - wrap*2 + off];
1219 if(!s->mb_x) C[0] = C[1] = 0;
1220 if(!s->first_slice_line) { // predictor A is not out of bounds
1221 if(s->mb_width == 1) {
1225 px = mid_pred(A[0], B[0], C[0]);
1226 py = mid_pred(A[1], B[1], C[1]);
1228 } else if(s->mb_x) { // predictor C is not out of bounds
1234 /* Pullback MV as specified in 8.3.5.3.4 */
1237 if(v->profile < PROFILE_ADVANCED) {
1238 qx = (s->mb_x << 5);
1239 qy = (s->mb_y << 5);
1240 X = (s->mb_width << 5) - 4;
1241 Y = (s->mb_height << 5) - 4;
1242 if(qx + px < -28) px = -28 - qx;
1243 if(qy + py < -28) py = -28 - qy;
1244 if(qx + px > X) px = X - qx;
1245 if(qy + py > Y) py = Y - qy;
1247 qx = (s->mb_x << 6);
1248 qy = (s->mb_y << 6);
1249 X = (s->mb_width << 6) - 4;
1250 Y = (s->mb_height << 6) - 4;
1251 if(qx + px < -60) px = -60 - qx;
1252 if(qy + py < -60) py = -60 - qy;
1253 if(qx + px > X) px = X - qx;
1254 if(qy + py > Y) py = Y - qy;
1257 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
1258 if(0 && !s->first_slice_line && s->mb_x) {
1259 if(is_intra[xy - wrap])
1260 sum = FFABS(px) + FFABS(py);
1262 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
1264 if(get_bits1(&s->gb)) {
1272 if(is_intra[xy - 2])
1273 sum = FFABS(px) + FFABS(py);
1275 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
1277 if(get_bits1(&s->gb)) {
1287 /* store MV using signed modulus of MV range defined in 4.11 */
1288 s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
1289 s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
1291 if((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
1292 C = s->current_picture.motion_val[1][xy - 2];
1293 A = s->current_picture.motion_val[1][xy - wrap*2];
1294 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1295 B = s->current_picture.motion_val[1][xy - wrap*2 + off];
1297 if(!s->mb_x) C[0] = C[1] = 0;
1298 if(!s->first_slice_line) { // predictor A is not out of bounds
1299 if(s->mb_width == 1) {
1303 px = mid_pred(A[0], B[0], C[0]);
1304 py = mid_pred(A[1], B[1], C[1]);
1306 } else if(s->mb_x) { // predictor C is not out of bounds
1312 /* Pullback MV as specified in 8.3.5.3.4 */
1315 if(v->profile < PROFILE_ADVANCED) {
1316 qx = (s->mb_x << 5);
1317 qy = (s->mb_y << 5);
1318 X = (s->mb_width << 5) - 4;
1319 Y = (s->mb_height << 5) - 4;
1320 if(qx + px < -28) px = -28 - qx;
1321 if(qy + py < -28) py = -28 - qy;
1322 if(qx + px > X) px = X - qx;
1323 if(qy + py > Y) py = Y - qy;
1325 qx = (s->mb_x << 6);
1326 qy = (s->mb_y << 6);
1327 X = (s->mb_width << 6) - 4;
1328 Y = (s->mb_height << 6) - 4;
1329 if(qx + px < -60) px = -60 - qx;
1330 if(qy + py < -60) py = -60 - qy;
1331 if(qx + px > X) px = X - qx;
1332 if(qy + py > Y) py = Y - qy;
1335 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
1336 if(0 && !s->first_slice_line && s->mb_x) {
1337 if(is_intra[xy - wrap])
1338 sum = FFABS(px) + FFABS(py);
1340 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
1342 if(get_bits1(&s->gb)) {
1350 if(is_intra[xy - 2])
1351 sum = FFABS(px) + FFABS(py);
1353 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
1355 if(get_bits1(&s->gb)) {
1365 /* store MV using signed modulus of MV range defined in 4.11 */
1367 s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
1368 s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
1370 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
1371 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
1372 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
1373 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
1376 /** Get predicted DC value for I-frames only
1377 * prediction dir: left=0, top=1
1378 * @param s MpegEncContext
1379 * @param overlap flag indicating that overlap filtering is used
1380 * @param pq integer part of picture quantizer
1381 * @param[in] n block index in the current MB
1382 * @param dc_val_ptr Pointer to DC predictor
1383 * @param dir_ptr Prediction direction for use in AC prediction
1385 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
1386 int16_t **dc_val_ptr, int *dir_ptr)
1388 int a, b, c, wrap, pred, scale;
1390 static const uint16_t dcpred[32] = {
1391 -1, 1024, 512, 341, 256, 205, 171, 146, 128,
1392 114, 102, 93, 85, 79, 73, 68, 64,
1393 60, 57, 54, 51, 49, 47, 45, 43,
1394 41, 39, 38, 37, 35, 34, 33
1397 /* find prediction - wmv3_dc_scale always used here in fact */
1398 if (n < 4) scale = s->y_dc_scale;
1399 else scale = s->c_dc_scale;
1401 wrap = s->block_wrap[n];
1402 dc_val= s->dc_val[0] + s->block_index[n];
1408 b = dc_val[ - 1 - wrap];
1409 a = dc_val[ - wrap];
1411 if (pq < 9 || !overlap)
1413 /* Set outer values */
1414 if (s->first_slice_line && (n!=2 && n!=3)) b=a=dcpred[scale];
1415 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=dcpred[scale];
1419 /* Set outer values */
1420 if (s->first_slice_line && (n!=2 && n!=3)) b=a=0;
1421 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=0;
1424 if (abs(a - b) <= abs(b - c)) {
1432 /* update predictor */
1433 *dc_val_ptr = &dc_val[0];
1438 /** Get predicted DC value
1439 * prediction dir: left=0, top=1
1440 * @param s MpegEncContext
1441 * @param overlap flag indicating that overlap filtering is used
1442 * @param pq integer part of picture quantizer
1443 * @param[in] n block index in the current MB
1444 * @param a_avail flag indicating top block availability
1445 * @param c_avail flag indicating left block availability
1446 * @param dc_val_ptr Pointer to DC predictor
1447 * @param dir_ptr Prediction direction for use in AC prediction
1449 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
1450 int a_avail, int c_avail,
1451 int16_t **dc_val_ptr, int *dir_ptr)
1453 int a, b, c, wrap, pred;
1455 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1458 wrap = s->block_wrap[n];
1459 dc_val= s->dc_val[0] + s->block_index[n];
1465 b = dc_val[ - 1 - wrap];
1466 a = dc_val[ - wrap];
1467 /* scale predictors if needed */
1468 q1 = s->current_picture.qscale_table[mb_pos];
1469 if(c_avail && (n!= 1 && n!=3)) {
1470 q2 = s->current_picture.qscale_table[mb_pos - 1];
1472 c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1474 if(a_avail && (n!= 2 && n!=3)) {
1475 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1477 a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1479 if(a_avail && c_avail && (n!=3)) {
1482 if(n != 2) off -= s->mb_stride;
1483 q2 = s->current_picture.qscale_table[off];
1485 b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1488 if(a_avail && c_avail) {
1489 if(abs(a - b) <= abs(b - c)) {
1496 } else if(a_avail) {
1499 } else if(c_avail) {
1507 /* update predictor */
1508 *dc_val_ptr = &dc_val[0];
1512 /** @} */ // Block group
1515 * @defgroup vc1_std_mb VC1 Macroblock-level functions in Simple/Main Profiles
1516 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1520 static inline int vc1_coded_block_pred(MpegEncContext * s, int n, uint8_t **coded_block_ptr)
1522 int xy, wrap, pred, a, b, c;
1524 xy = s->block_index[n];
1525 wrap = s->b8_stride;
1530 a = s->coded_block[xy - 1 ];
1531 b = s->coded_block[xy - 1 - wrap];
1532 c = s->coded_block[xy - wrap];
1541 *coded_block_ptr = &s->coded_block[xy];
1547 * Decode one AC coefficient
1548 * @param v The VC1 context
1549 * @param last Last coefficient
1550 * @param skip How much zero coefficients to skip
1551 * @param value Decoded AC coefficient value
1552 * @param codingset set of VLC to decode data
1555 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip, int *value, int codingset)
1557 GetBitContext *gb = &v->s.gb;
1558 int index, escape, run = 0, level = 0, lst = 0;
1560 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
1561 if (index != vc1_ac_sizes[codingset] - 1) {
1562 run = vc1_index_decode_table[codingset][index][0];
1563 level = vc1_index_decode_table[codingset][index][1];
1564 lst = index >= vc1_last_decode_table[codingset] || get_bits_left(gb) < 0;
1568 escape = decode210(gb);
1570 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
1571 run = vc1_index_decode_table[codingset][index][0];
1572 level = vc1_index_decode_table[codingset][index][1];
1573 lst = index >= vc1_last_decode_table[codingset];
1576 level += vc1_last_delta_level_table[codingset][run];
1578 level += vc1_delta_level_table[codingset][run];
1581 run += vc1_last_delta_run_table[codingset][level] + 1;
1583 run += vc1_delta_run_table[codingset][level] + 1;
1589 lst = get_bits1(gb);
1590 if(v->s.esc3_level_length == 0) {
1591 if(v->pq < 8 || v->dquantfrm) { // table 59
1592 v->s.esc3_level_length = get_bits(gb, 3);
1593 if(!v->s.esc3_level_length)
1594 v->s.esc3_level_length = get_bits(gb, 2) + 8;
1596 v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
1598 v->s.esc3_run_length = 3 + get_bits(gb, 2);
1600 run = get_bits(gb, v->s.esc3_run_length);
1601 sign = get_bits1(gb);
1602 level = get_bits(gb, v->s.esc3_level_length);
1613 /** Decode intra block in intra frames - should be faster than decode_intra_block
1614 * @param v VC1Context
1615 * @param block block to decode
1616 * @param[in] n subblock index
1617 * @param coded are AC coeffs present or not
1618 * @param codingset set of VLC to decode data
1620 static int vc1_decode_i_block(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset)
1622 GetBitContext *gb = &v->s.gb;
1623 MpegEncContext *s = &v->s;
1624 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1627 int16_t *ac_val, *ac_val2;
1630 /* Get DC differential */
1632 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1634 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1637 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1642 if (dcdiff == 119 /* ESC index value */)
1644 /* TODO: Optimize */
1645 if (v->pq == 1) dcdiff = get_bits(gb, 10);
1646 else if (v->pq == 2) dcdiff = get_bits(gb, 9);
1647 else dcdiff = get_bits(gb, 8);
1652 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1653 else if (v->pq == 2)
1654 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1661 dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
1664 /* Store the quantized DC coeff, used for prediction */
1666 block[0] = dcdiff * s->y_dc_scale;
1668 block[0] = dcdiff * s->c_dc_scale;
1679 int last = 0, skip, value;
1680 const uint8_t *zz_table;
1684 scale = v->pq * 2 + v->halfpq;
1688 zz_table = v->zz_8x8[2];
1690 zz_table = v->zz_8x8[3];
1692 zz_table = v->zz_8x8[1];
1694 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1696 if(dc_pred_dir) //left
1699 ac_val -= 16 * s->block_wrap[n];
1702 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1706 block[zz_table[i++]] = value;
1709 /* apply AC prediction if needed */
1711 if(dc_pred_dir) { //left
1712 for(k = 1; k < 8; k++)
1713 block[k << v->left_blk_sh] += ac_val[k];
1715 for(k = 1; k < 8; k++)
1716 block[k << v->top_blk_sh] += ac_val[k + 8];
1719 /* save AC coeffs for further prediction */
1720 for(k = 1; k < 8; k++) {
1721 ac_val2[k] = block[k << v->left_blk_sh];
1722 ac_val2[k + 8] = block[k << v->top_blk_sh];
1725 /* scale AC coeffs */
1726 for(k = 1; k < 64; k++)
1730 block[k] += (block[k] < 0) ? -v->pq : v->pq;
1733 if(s->ac_pred) i = 63;
1739 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1743 scale = v->pq * 2 + v->halfpq;
1744 memset(ac_val2, 0, 16 * 2);
1745 if(dc_pred_dir) {//left
1748 memcpy(ac_val2, ac_val, 8 * 2);
1750 ac_val -= 16 * s->block_wrap[n];
1752 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1755 /* apply AC prediction if needed */
1757 if(dc_pred_dir) { //left
1758 for(k = 1; k < 8; k++) {
1759 block[k << v->left_blk_sh] = ac_val[k] * scale;
1760 if(!v->pquantizer && block[k << v->left_blk_sh])
1761 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -v->pq : v->pq;
1764 for(k = 1; k < 8; k++) {
1765 block[k << v->top_blk_sh] = ac_val[k + 8] * scale;
1766 if(!v->pquantizer && block[k << v->top_blk_sh])
1767 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -v->pq : v->pq;
1773 s->block_last_index[n] = i;
1778 /** Decode intra block in intra frames - should be faster than decode_intra_block
1779 * @param v VC1Context
1780 * @param block block to decode
1781 * @param[in] n subblock number
1782 * @param coded are AC coeffs present or not
1783 * @param codingset set of VLC to decode data
1784 * @param mquant quantizer value for this macroblock
1786 static int vc1_decode_i_block_adv(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset, int mquant)
1788 GetBitContext *gb = &v->s.gb;
1789 MpegEncContext *s = &v->s;
1790 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1793 int16_t *ac_val, *ac_val2;
1795 int a_avail = v->a_avail, c_avail = v->c_avail;
1796 int use_pred = s->ac_pred;
1799 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1801 /* Get DC differential */
1803 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1805 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1808 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1813 if (dcdiff == 119 /* ESC index value */)
1815 /* TODO: Optimize */
1816 if (mquant == 1) dcdiff = get_bits(gb, 10);
1817 else if (mquant == 2) dcdiff = get_bits(gb, 9);
1818 else dcdiff = get_bits(gb, 8);
1823 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1824 else if (mquant == 2)
1825 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1832 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
1835 /* Store the quantized DC coeff, used for prediction */
1837 block[0] = dcdiff * s->y_dc_scale;
1839 block[0] = dcdiff * s->c_dc_scale;
1845 /* check if AC is needed at all */
1846 if(!a_avail && !c_avail) use_pred = 0;
1847 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1850 scale = mquant * 2 + ((mquant == v->pq) ? v->halfpq : 0);
1852 if(dc_pred_dir) //left
1855 ac_val -= 16 * s->block_wrap[n];
1857 q1 = s->current_picture.qscale_table[mb_pos];
1858 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
1859 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1860 if(dc_pred_dir && n==1) q2 = q1;
1861 if(!dc_pred_dir && n==2) q2 = q1;
1865 int last = 0, skip, value;
1866 const uint8_t *zz_table;
1871 zz_table = v->zz_8x8[2];
1873 zz_table = v->zz_8x8[3];
1875 zz_table = v->zz_8x8[1];
1878 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1882 block[zz_table[i++]] = value;
1885 /* apply AC prediction if needed */
1887 /* scale predictors if needed*/
1889 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1890 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1892 if(dc_pred_dir) { //left
1893 for(k = 1; k < 8; k++)
1894 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1896 for(k = 1; k < 8; k++)
1897 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1900 if(dc_pred_dir) { //left
1901 for(k = 1; k < 8; k++)
1902 block[k << v->left_blk_sh] += ac_val[k];
1904 for(k = 1; k < 8; k++)
1905 block[k << v->top_blk_sh] += ac_val[k + 8];
1909 /* save AC coeffs for further prediction */
1910 for(k = 1; k < 8; k++) {
1911 ac_val2[k ] = block[k << v->left_blk_sh];
1912 ac_val2[k + 8] = block[k << v->top_blk_sh];
1915 /* scale AC coeffs */
1916 for(k = 1; k < 64; k++)
1920 block[k] += (block[k] < 0) ? -mquant : mquant;
1923 if(use_pred) i = 63;
1924 } else { // no AC coeffs
1927 memset(ac_val2, 0, 16 * 2);
1928 if(dc_pred_dir) {//left
1930 memcpy(ac_val2, ac_val, 8 * 2);
1932 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1933 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1934 for(k = 1; k < 8; k++)
1935 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1940 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1942 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1943 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1944 for(k = 1; k < 8; k++)
1945 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1950 /* apply AC prediction if needed */
1952 if(dc_pred_dir) { //left
1953 for(k = 1; k < 8; k++) {
1954 block[k << v->left_blk_sh] = ac_val2[k] * scale;
1955 if(!v->pquantizer && block[k << v->left_blk_sh])
1956 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
1959 for(k = 1; k < 8; k++) {
1960 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
1961 if(!v->pquantizer && block[k << v->top_blk_sh])
1962 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
1968 s->block_last_index[n] = i;
1973 /** Decode intra block in inter frames - more generic version than vc1_decode_i_block
1974 * @param v VC1Context
1975 * @param block block to decode
1976 * @param[in] n subblock index
1977 * @param coded are AC coeffs present or not
1978 * @param mquant block quantizer
1979 * @param codingset set of VLC to decode data
1981 static int vc1_decode_intra_block(VC1Context *v, DCTELEM block[64], int n, int coded, int mquant, int codingset)
1983 GetBitContext *gb = &v->s.gb;
1984 MpegEncContext *s = &v->s;
1985 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1988 int16_t *ac_val, *ac_val2;
1990 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1991 int a_avail = v->a_avail, c_avail = v->c_avail;
1992 int use_pred = s->ac_pred;
1996 s->dsp.clear_block(block);
1998 /* XXX: Guard against dumb values of mquant */
1999 mquant = (mquant < 1) ? 0 : ( (mquant>31) ? 31 : mquant );
2001 /* Set DC scale - y and c use the same */
2002 s->y_dc_scale = s->y_dc_scale_table[mquant];
2003 s->c_dc_scale = s->c_dc_scale_table[mquant];
2005 /* Get DC differential */
2007 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2009 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2012 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2017 if (dcdiff == 119 /* ESC index value */)
2019 /* TODO: Optimize */
2020 if (mquant == 1) dcdiff = get_bits(gb, 10);
2021 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2022 else dcdiff = get_bits(gb, 8);
2027 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
2028 else if (mquant == 2)
2029 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
2036 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
2039 /* Store the quantized DC coeff, used for prediction */
2042 block[0] = dcdiff * s->y_dc_scale;
2044 block[0] = dcdiff * s->c_dc_scale;
2050 /* check if AC is needed at all and adjust direction if needed */
2051 if(!a_avail) dc_pred_dir = 1;
2052 if(!c_avail) dc_pred_dir = 0;
2053 if(!a_avail && !c_avail) use_pred = 0;
2054 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2057 scale = mquant * 2 + v->halfpq;
2059 if(dc_pred_dir) //left
2062 ac_val -= 16 * s->block_wrap[n];
2064 q1 = s->current_picture.qscale_table[mb_pos];
2065 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
2066 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2067 if(dc_pred_dir && n==1) q2 = q1;
2068 if(!dc_pred_dir && n==2) q2 = q1;
2072 int last = 0, skip, value;
2076 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2080 block[v->zz_8x8[0][i++]] = value;
2083 /* apply AC prediction if needed */
2085 /* scale predictors if needed*/
2087 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2088 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2090 if(dc_pred_dir) { //left
2091 for(k = 1; k < 8; k++)
2092 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2094 for(k = 1; k < 8; k++)
2095 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2098 if(dc_pred_dir) { //left
2099 for(k = 1; k < 8; k++)
2100 block[k << v->left_blk_sh] += ac_val[k];
2102 for(k = 1; k < 8; k++)
2103 block[k << v->top_blk_sh] += ac_val[k + 8];
2107 /* save AC coeffs for further prediction */
2108 for(k = 1; k < 8; k++) {
2109 ac_val2[k ] = block[k << v->left_blk_sh];
2110 ac_val2[k + 8] = block[k << v->top_blk_sh];
2113 /* scale AC coeffs */
2114 for(k = 1; k < 64; k++)
2118 block[k] += (block[k] < 0) ? -mquant : mquant;
2121 if(use_pred) i = 63;
2122 } else { // no AC coeffs
2125 memset(ac_val2, 0, 16 * 2);
2126 if(dc_pred_dir) {//left
2128 memcpy(ac_val2, ac_val, 8 * 2);
2130 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2131 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2132 for(k = 1; k < 8; k++)
2133 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2138 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2140 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2141 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2142 for(k = 1; k < 8; k++)
2143 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2148 /* apply AC prediction if needed */
2150 if(dc_pred_dir) { //left
2151 for(k = 1; k < 8; k++) {
2152 block[k << v->left_blk_sh] = ac_val2[k] * scale;
2153 if(!v->pquantizer && block[k << v->left_blk_sh])
2154 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
2157 for(k = 1; k < 8; k++) {
2158 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
2159 if(!v->pquantizer && block[k << v->top_blk_sh])
2160 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
2166 s->block_last_index[n] = i;
2173 static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n, int mquant, int ttmb, int first_block,
2174 uint8_t *dst, int linesize, int skip_block, int *ttmb_out)
2176 MpegEncContext *s = &v->s;
2177 GetBitContext *gb = &s->gb;
2180 int scale, off, idx, last, skip, value;
2181 int ttblk = ttmb & 7;
2184 s->dsp.clear_block(block);
2187 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)];
2189 if(ttblk == TT_4X4) {
2190 subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
2192 if((ttblk != TT_8X8 && ttblk != TT_4X4)
2193 && ((v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))
2194 || (!v->res_rtm_flag && !first_block))) {
2195 subblkpat = decode012(gb);
2196 if(subblkpat) subblkpat ^= 3; //swap decoded pattern bits
2197 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) ttblk = TT_8X4;
2198 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) ttblk = TT_4X8;
2200 scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);
2202 // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
2203 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
2204 subblkpat = 2 - (ttblk == TT_8X4_TOP);
2207 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
2208 subblkpat = 2 - (ttblk == TT_4X8_LEFT);
2217 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2221 idx = v->zz_8x8[0][i++];
2222 block[idx] = value * scale;
2224 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2228 v->vc1dsp.vc1_inv_trans_8x8_dc(dst, linesize, block);
2230 v->vc1dsp.vc1_inv_trans_8x8(block);
2231 s->dsp.add_pixels_clamped(block, dst, linesize);
2236 pat = ~subblkpat & 0xF;
2237 for(j = 0; j < 4; j++) {
2238 last = subblkpat & (1 << (3 - j));
2240 off = (j & 1) * 4 + (j & 2) * 16;
2242 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2246 idx = ff_vc1_simple_progressive_4x4_zz[i++];
2247 block[idx + off] = value * scale;
2249 block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
2251 if(!(subblkpat & (1 << (3 - j))) && !skip_block){
2253 v->vc1dsp.vc1_inv_trans_4x4_dc(dst + (j&1)*4 + (j&2)*2*linesize, linesize, block + off);
2255 v->vc1dsp.vc1_inv_trans_4x4(dst + (j&1)*4 + (j&2)*2*linesize, linesize, block + off);
2260 pat = ~((subblkpat & 2)*6 + (subblkpat & 1)*3) & 0xF;
2261 for(j = 0; j < 2; j++) {
2262 last = subblkpat & (1 << (1 - j));
2266 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2270 idx = v->zz_8x4[i++]+off;
2271 block[idx] = value * scale;
2273 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2275 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
2277 v->vc1dsp.vc1_inv_trans_8x4_dc(dst + j*4*linesize, linesize, block + off);
2279 v->vc1dsp.vc1_inv_trans_8x4(dst + j*4*linesize, linesize, block + off);
2284 pat = ~(subblkpat*5) & 0xF;
2285 for(j = 0; j < 2; j++) {
2286 last = subblkpat & (1 << (1 - j));
2290 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2294 idx = v->zz_4x8[i++]+off;
2295 block[idx] = value * scale;
2297 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2299 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
2301 v->vc1dsp.vc1_inv_trans_4x8_dc(dst + j*4, linesize, block + off);
2303 v->vc1dsp.vc1_inv_trans_4x8(dst + j*4, linesize, block + off);
2309 *ttmb_out |= ttblk << (n * 4);
2313 /** @} */ // Macroblock group
2315 static const int size_table [6] = { 0, 2, 3, 4, 5, 8 };
2316 static const int offset_table[6] = { 0, 1, 3, 7, 15, 31 };
2318 static av_always_inline void vc1_apply_p_v_loop_filter(VC1Context *v, int block_num)
2320 MpegEncContext *s = &v->s;
2321 int mb_cbp = v->cbp[s->mb_x - s->mb_stride],
2322 block_cbp = mb_cbp >> (block_num * 4), bottom_cbp,
2323 mb_is_intra = v->is_intra[s->mb_x - s->mb_stride],
2324 block_is_intra = mb_is_intra >> (block_num * 4), bottom_is_intra;
2325 int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
2329 dst = s->dest[block_num - 3];
2331 dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 8) * linesize;
2333 if (s->mb_y != s->mb_height || block_num < 2) {
2338 bottom_cbp = v->cbp[s->mb_x] >> (block_num * 4);
2339 bottom_is_intra = v->is_intra[s->mb_x] >> (block_num * 4);
2340 mv = &v->luma_mv[s->mb_x - s->mb_stride];
2341 mv_stride = s->mb_stride;
2343 bottom_cbp = (block_num < 2) ? (mb_cbp >> ((block_num + 2) * 4)) :
2344 (v->cbp[s->mb_x] >> ((block_num - 2) * 4));
2345 bottom_is_intra = (block_num < 2) ? (mb_is_intra >> ((block_num + 2) * 4)) :
2346 (v->is_intra[s->mb_x] >> ((block_num - 2) * 4));
2347 mv_stride = s->b8_stride;
2348 mv = &s->current_picture.motion_val[0][s->block_index[block_num] - 2 * mv_stride];
2351 if (bottom_is_intra & 1 || block_is_intra & 1 ||
2352 mv[0][0] != mv[mv_stride][0] || mv[0][1] != mv[mv_stride][1]) {
2353 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
2355 idx = ((bottom_cbp >> 2) | block_cbp) & 3;
2357 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
2360 v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);
2362 v->vc1dsp.vc1_v_loop_filter4(dst, linesize, v->pq);
2367 dst -= 4 * linesize;
2368 ttblk = (v->ttblk[s->mb_x - s->mb_stride] >> (block_num * 4)) & 0xf;
2369 if (ttblk == TT_4X4 || ttblk == TT_8X4) {
2370 idx = (block_cbp | (block_cbp >> 2)) & 3;
2372 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
2375 v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);
2377 v->vc1dsp.vc1_v_loop_filter4(dst, linesize, v->pq);
2382 static av_always_inline void vc1_apply_p_h_loop_filter(VC1Context *v, int block_num)
2384 MpegEncContext *s = &v->s;
2385 int mb_cbp = v->cbp[s->mb_x - 1 - s->mb_stride],
2386 block_cbp = mb_cbp >> (block_num * 4), right_cbp,
2387 mb_is_intra = v->is_intra[s->mb_x - 1 - s->mb_stride],
2388 block_is_intra = mb_is_intra >> (block_num * 4), right_is_intra;
2389 int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
2392 if (block_num > 3) {
2393 dst = s->dest[block_num - 3] - 8 * linesize;
2395 dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 16) * linesize - 8;
2398 if (s->mb_x != s->mb_width || !(block_num & 5)) {
2402 right_cbp = v->cbp[s->mb_x - s->mb_stride] >> (block_num * 4);
2403 right_is_intra = v->is_intra[s->mb_x - s->mb_stride] >> (block_num * 4);
2404 mv = &v->luma_mv[s->mb_x - s->mb_stride - 1];
2406 right_cbp = (block_num & 1) ? (v->cbp[s->mb_x - s->mb_stride] >> ((block_num - 1) * 4)) :
2407 (mb_cbp >> ((block_num + 1) * 4));
2408 right_is_intra = (block_num & 1) ? (v->is_intra[s->mb_x - s->mb_stride] >> ((block_num - 1) * 4)) :
2409 (mb_is_intra >> ((block_num + 1) * 4));
2410 mv = &s->current_picture.motion_val[0][s->block_index[block_num] - s->b8_stride * 2 - 2];
2412 if (block_is_intra & 1 || right_is_intra & 1 || mv[0][0] != mv[1][0] || mv[0][1] != mv[1][1]) {
2413 v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
2415 idx = ((right_cbp >> 1) | block_cbp) & 5; // FIXME check
2417 v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
2420 v->vc1dsp.vc1_h_loop_filter4(dst+4*linesize, linesize, v->pq);
2422 v->vc1dsp.vc1_h_loop_filter4(dst, linesize, v->pq);
2428 ttblk = (v->ttblk[s->mb_x - s->mb_stride - 1] >> (block_num * 4)) & 0xf;
2429 if (ttblk == TT_4X4 || ttblk == TT_4X8) {
2430 idx = (block_cbp | (block_cbp >> 1)) & 5;
2432 v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
2435 v->vc1dsp.vc1_h_loop_filter4(dst + linesize*4, linesize, v->pq);
2437 v->vc1dsp.vc1_h_loop_filter4(dst, linesize, v->pq);
2442 static void vc1_apply_p_loop_filter(VC1Context *v)
2444 MpegEncContext *s = &v->s;
2447 for (i = 0; i < 6; i++) {
2448 vc1_apply_p_v_loop_filter(v, i);
2451 /* V always preceedes H, therefore we run H one MB before V;
2452 * at the end of a row, we catch up to complete the row */
2454 for (i = 0; i < 6; i++) {
2455 vc1_apply_p_h_loop_filter(v, i);
2457 if (s->mb_x == s->mb_width - 1) {
2459 ff_update_block_index(s);
2460 for (i = 0; i < 6; i++) {
2461 vc1_apply_p_h_loop_filter(v, i);
2467 /** Decode one P-frame MB (in Simple/Main profile)
2469 static int vc1_decode_p_mb(VC1Context *v)
2471 MpegEncContext *s = &v->s;
2472 GetBitContext *gb = &s->gb;
2474 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2475 int cbp; /* cbp decoding stuff */
2476 int mqdiff, mquant; /* MB quantization */
2477 int ttmb = v->ttfrm; /* MB Transform type */
2479 int mb_has_coeffs = 1; /* last_flag */
2480 int dmv_x, dmv_y; /* Differential MV components */
2481 int index, index1; /* LUT indexes */
2482 int val, sign; /* temp values */
2483 int first_block = 1;
2485 int skipped, fourmv;
2486 int block_cbp = 0, pat, block_tt = 0, block_intra = 0;
2488 mquant = v->pq; /* Loosy initialization */
2490 if (v->mv_type_is_raw)
2491 fourmv = get_bits1(gb);
2493 fourmv = v->mv_type_mb_plane[mb_pos];
2495 skipped = get_bits1(gb);
2497 skipped = v->s.mbskip_table[mb_pos];
2499 if (!fourmv) /* 1MV mode */
2503 GET_MVDATA(dmv_x, dmv_y);
2506 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2507 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2509 s->current_picture.mb_type[mb_pos] = s->mb_intra ? MB_TYPE_INTRA : MB_TYPE_16x16;
2510 vc1_pred_mv(v, 0, dmv_x, dmv_y, 1, v->range_x, v->range_y, v->mb_type[0]);
2512 /* FIXME Set DC val for inter block ? */
2513 if (s->mb_intra && !mb_has_coeffs)
2516 s->ac_pred = get_bits1(gb);
2519 else if (mb_has_coeffs)
2521 if (s->mb_intra) s->ac_pred = get_bits1(gb);
2522 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2530 s->current_picture.qscale_table[mb_pos] = mquant;
2532 if (!v->ttmbf && !s->mb_intra && mb_has_coeffs)
2533 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table,
2534 VC1_TTMB_VLC_BITS, 2);
2535 if(!s->mb_intra) vc1_mc_1mv(v, 0);
2539 s->dc_val[0][s->block_index[i]] = 0;
2541 val = ((cbp >> (5 - i)) & 1);
2542 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2543 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2545 /* check if prediction blocks A and C are available */
2546 v->a_avail = v->c_avail = 0;
2547 if(i == 2 || i == 3 || !s->first_slice_line)
2548 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2549 if(i == 1 || i == 3 || s->mb_x)
2550 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2552 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
2553 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2554 v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
2555 if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
2556 s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2557 if(v->pq >= 9 && v->overlap) {
2559 v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2561 v->vc1dsp.vc1_v_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2563 block_cbp |= 0xF << (i << 2);
2564 block_intra |= 1 << i;
2566 pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block, s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize, (i&4) && (s->flags & CODEC_FLAG_GRAY), &block_tt);
2567 block_cbp |= pat << (i << 2);
2568 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2576 for(i = 0; i < 6; i++) {
2577 v->mb_type[0][s->block_index[i]] = 0;
2578 s->dc_val[0][s->block_index[i]] = 0;
2580 s->current_picture.mb_type[mb_pos] = MB_TYPE_SKIP;
2581 s->current_picture.qscale_table[mb_pos] = 0;
2582 vc1_pred_mv(v, 0, 0, 0, 1, v->range_x, v->range_y, v->mb_type[0]);
2588 if (!skipped /* unskipped MB */)
2590 int intra_count = 0, coded_inter = 0;
2591 int is_intra[6], is_coded[6];
2593 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2596 val = ((cbp >> (5 - i)) & 1);
2597 s->dc_val[0][s->block_index[i]] = 0;
2604 GET_MVDATA(dmv_x, dmv_y);
2606 vc1_pred_mv(v, i, dmv_x, dmv_y, 0, v->range_x, v->range_y, v->mb_type[0]);
2607 if(!s->mb_intra) vc1_mc_4mv_luma(v, i);
2608 intra_count += s->mb_intra;
2609 is_intra[i] = s->mb_intra;
2610 is_coded[i] = mb_has_coeffs;
2613 is_intra[i] = (intra_count >= 3);
2616 if(i == 4) vc1_mc_4mv_chroma(v);
2617 v->mb_type[0][s->block_index[i]] = is_intra[i];
2618 if(!coded_inter) coded_inter = !is_intra[i] & is_coded[i];
2620 // if there are no coded blocks then don't do anything more
2622 if(!intra_count && !coded_inter)
2625 s->current_picture.qscale_table[mb_pos] = mquant;
2626 /* test if block is intra and has pred */
2631 if(((!s->first_slice_line || (i==2 || i==3)) && v->mb_type[0][s->block_index[i] - s->block_wrap[i]])
2632 || ((s->mb_x || (i==1 || i==3)) && v->mb_type[0][s->block_index[i] - 1])) {
2637 if(intrapred)s->ac_pred = get_bits1(gb);
2638 else s->ac_pred = 0;
2640 if (!v->ttmbf && coded_inter)
2641 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2645 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2646 s->mb_intra = is_intra[i];
2648 /* check if prediction blocks A and C are available */
2649 v->a_avail = v->c_avail = 0;
2650 if(i == 2 || i == 3 || !s->first_slice_line)
2651 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2652 if(i == 1 || i == 3 || s->mb_x)
2653 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2655 vc1_decode_intra_block(v, s->block[i], i, is_coded[i], mquant, (i&4)?v->codingset2:v->codingset);
2656 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2657 v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
2658 if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
2659 s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize);
2660 if(v->pq >= 9 && v->overlap) {
2662 v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2664 v->vc1dsp.vc1_v_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2666 block_cbp |= 0xF << (i << 2);
2667 block_intra |= 1 << i;
2668 } else if(is_coded[i]) {
2669 pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block, s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize, (i&4) && (s->flags & CODEC_FLAG_GRAY), &block_tt);
2670 block_cbp |= pat << (i << 2);
2671 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2679 s->current_picture.qscale_table[mb_pos] = 0;
2680 for (i=0; i<6; i++) {
2681 v->mb_type[0][s->block_index[i]] = 0;
2682 s->dc_val[0][s->block_index[i]] = 0;
2686 vc1_pred_mv(v, i, 0, 0, 0, v->range_x, v->range_y, v->mb_type[0]);
2687 vc1_mc_4mv_luma(v, i);
2689 vc1_mc_4mv_chroma(v);
2690 s->current_picture.qscale_table[mb_pos] = 0;
2694 v->cbp[s->mb_x] = block_cbp;
2695 v->ttblk[s->mb_x] = block_tt;
2696 v->is_intra[s->mb_x] = block_intra;
2701 /** Decode one B-frame MB (in Main profile)
2703 static void vc1_decode_b_mb(VC1Context *v)
2705 MpegEncContext *s = &v->s;
2706 GetBitContext *gb = &s->gb;
2708 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2709 int cbp = 0; /* cbp decoding stuff */
2710 int mqdiff, mquant; /* MB quantization */
2711 int ttmb = v->ttfrm; /* MB Transform type */
2712 int mb_has_coeffs = 0; /* last_flag */
2713 int index, index1; /* LUT indexes */
2714 int val, sign; /* temp values */
2715 int first_block = 1;
2717 int skipped, direct;
2718 int dmv_x[2], dmv_y[2];
2719 int bmvtype = BMV_TYPE_BACKWARD;
2721 mquant = v->pq; /* Loosy initialization */
2725 direct = get_bits1(gb);
2727 direct = v->direct_mb_plane[mb_pos];
2729 skipped = get_bits1(gb);
2731 skipped = v->s.mbskip_table[mb_pos];
2733 dmv_x[0] = dmv_x[1] = dmv_y[0] = dmv_y[1] = 0;
2734 for(i = 0; i < 6; i++) {
2735 v->mb_type[0][s->block_index[i]] = 0;
2736 s->dc_val[0][s->block_index[i]] = 0;
2738 s->current_picture.qscale_table[mb_pos] = 0;
2742 GET_MVDATA(dmv_x[0], dmv_y[0]);
2743 dmv_x[1] = dmv_x[0];
2744 dmv_y[1] = dmv_y[0];
2746 if(skipped || !s->mb_intra) {
2747 bmvtype = decode012(gb);
2750 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_BACKWARD : BMV_TYPE_FORWARD;
2753 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_FORWARD : BMV_TYPE_BACKWARD;
2756 bmvtype = BMV_TYPE_INTERPOLATED;
2757 dmv_x[0] = dmv_y[0] = 0;
2761 for(i = 0; i < 6; i++)
2762 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2765 if(direct) bmvtype = BMV_TYPE_INTERPOLATED;
2766 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2767 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2771 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2774 s->current_picture.qscale_table[mb_pos] = mquant;
2776 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2777 dmv_x[0] = dmv_y[0] = dmv_x[1] = dmv_y[1] = 0;
2778 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2779 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2781 if(!mb_has_coeffs && !s->mb_intra) {
2782 /* no coded blocks - effectively skipped */
2783 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2784 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2787 if(s->mb_intra && !mb_has_coeffs) {
2789 s->current_picture.qscale_table[mb_pos] = mquant;
2790 s->ac_pred = get_bits1(gb);
2792 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2794 if(bmvtype == BMV_TYPE_INTERPOLATED) {
2795 GET_MVDATA(dmv_x[0], dmv_y[0]);
2796 if(!mb_has_coeffs) {
2797 /* interpolated skipped block */
2798 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2799 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2803 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2805 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2808 s->ac_pred = get_bits1(gb);
2809 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2811 s->current_picture.qscale_table[mb_pos] = mquant;
2812 if(!v->ttmbf && !s->mb_intra && mb_has_coeffs)
2813 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2819 s->dc_val[0][s->block_index[i]] = 0;
2821 val = ((cbp >> (5 - i)) & 1);
2822 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2823 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2825 /* check if prediction blocks A and C are available */
2826 v->a_avail = v->c_avail = 0;
2827 if(i == 2 || i == 3 || !s->first_slice_line)
2828 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2829 if(i == 1 || i == 3 || s->mb_x)
2830 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2832 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
2833 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2834 v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
2835 if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
2836 s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2838 vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block, s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize, (i&4) && (s->flags & CODEC_FLAG_GRAY), NULL);
2839 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2845 /** Decode blocks of I-frame
2847 static void vc1_decode_i_blocks(VC1Context *v)
2850 MpegEncContext *s = &v->s;
2855 /* select codingmode used for VLC tables selection */
2856 switch(v->y_ac_table_index){
2858 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2861 v->codingset = CS_HIGH_MOT_INTRA;
2864 v->codingset = CS_MID_RATE_INTRA;
2868 switch(v->c_ac_table_index){
2870 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2873 v->codingset2 = CS_HIGH_MOT_INTER;
2876 v->codingset2 = CS_MID_RATE_INTER;
2880 /* Set DC scale - y and c use the same */
2881 s->y_dc_scale = s->y_dc_scale_table[v->pq];
2882 s->c_dc_scale = s->c_dc_scale_table[v->pq];
2885 s->mb_x = s->mb_y = 0;
2887 s->first_slice_line = 1;
2888 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
2890 ff_init_block_index(s);
2891 for(; s->mb_x < s->mb_width; s->mb_x++) {
2893 ff_update_block_index(s);
2894 dst[0] = s->dest[0];
2895 dst[1] = dst[0] + 8;
2896 dst[2] = s->dest[0] + s->linesize * 8;
2897 dst[3] = dst[2] + 8;
2898 dst[4] = s->dest[1];
2899 dst[5] = s->dest[2];
2900 s->dsp.clear_blocks(s->block[0]);
2901 mb_pos = s->mb_x + s->mb_y * s->mb_width;
2902 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
2903 s->current_picture.qscale_table[mb_pos] = v->pq;
2904 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2905 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2907 // do actual MB decoding and displaying
2908 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
2909 v->s.ac_pred = get_bits1(&v->s.gb);
2911 for(k = 0; k < 6; k++) {
2912 val = ((cbp >> (5 - k)) & 1);
2915 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
2919 cbp |= val << (5 - k);
2921 vc1_decode_i_block(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2);
2923 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
2924 v->vc1dsp.vc1_inv_trans_8x8(s->block[k]);
2925 if(v->pq >= 9 && v->overlap) {
2926 if (v->rangeredfrm) for(j = 0; j < 64; j++) s->block[k][j] <<= 1;
2927 s->dsp.put_signed_pixels_clamped(s->block[k], dst[k], k & 4 ? s->uvlinesize : s->linesize);
2929 if (v->rangeredfrm) for(j = 0; j < 64; j++) s->block[k][j] = (s->block[k][j] - 64) << 1;
2930 s->dsp.put_pixels_clamped(s->block[k], dst[k], k & 4 ? s->uvlinesize : s->linesize);
2934 if(v->pq >= 9 && v->overlap) {
2936 v->vc1dsp.vc1_h_overlap(s->dest[0], s->linesize);
2937 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2938 if(!(s->flags & CODEC_FLAG_GRAY)) {
2939 v->vc1dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
2940 v->vc1dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
2943 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
2944 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2945 if(!s->first_slice_line) {
2946 v->vc1dsp.vc1_v_overlap(s->dest[0], s->linesize);
2947 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
2948 if(!(s->flags & CODEC_FLAG_GRAY)) {
2949 v->vc1dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
2950 v->vc1dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
2953 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2954 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2956 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
2958 if(get_bits_count(&s->gb) > v->bits) {
2959 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2960 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
2964 if (!v->s.loop_filter)
2965 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2967 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
2969 s->first_slice_line = 0;
2971 if (v->s.loop_filter)
2972 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
2973 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
2976 /** Decode blocks of I-frame for advanced profile
2978 static void vc1_decode_i_blocks_adv(VC1Context *v)
2981 MpegEncContext *s = &v->s;
2987 GetBitContext *gb = &s->gb;
2989 /* select codingmode used for VLC tables selection */
2990 switch(v->y_ac_table_index){
2992 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2995 v->codingset = CS_HIGH_MOT_INTRA;
2998 v->codingset = CS_MID_RATE_INTRA;
3002 switch(v->c_ac_table_index){
3004 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3007 v->codingset2 = CS_HIGH_MOT_INTER;
3010 v->codingset2 = CS_MID_RATE_INTER;
3015 s->mb_x = s->mb_y = 0;
3017 s->first_slice_line = 1;
3018 s->mb_y = s->start_mb_y;
3019 if (s->start_mb_y) {
3021 ff_init_block_index(s);
3022 memset(&s->coded_block[s->block_index[0]-s->b8_stride], 0,
3023 s->b8_stride * sizeof(*s->coded_block));
3025 for(; s->mb_y < s->end_mb_y; s->mb_y++) {
3027 ff_init_block_index(s);
3028 for(;s->mb_x < s->mb_width; s->mb_x++) {
3029 DCTELEM (*block)[64] = v->block[v->cur_blk_idx];
3030 ff_update_block_index(s);
3031 s->dsp.clear_blocks(block[0]);
3032 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
3033 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
3034 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
3035 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
3037 // do actual MB decoding and displaying
3038 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
3039 if(v->acpred_is_raw)
3040 v->s.ac_pred = get_bits1(&v->s.gb);
3042 v->s.ac_pred = v->acpred_plane[mb_pos];
3044 if (v->condover == CONDOVER_SELECT && v->overflg_is_raw)
3045 v->over_flags_plane[mb_pos] = get_bits1(&v->s.gb);
3049 s->current_picture.qscale_table[mb_pos] = mquant;
3050 /* Set DC scale - y and c use the same */
3051 s->y_dc_scale = s->y_dc_scale_table[mquant];
3052 s->c_dc_scale = s->c_dc_scale_table[mquant];
3054 for(k = 0; k < 6; k++) {
3055 val = ((cbp >> (5 - k)) & 1);
3058 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
3062 cbp |= val << (5 - k);
3064 v->a_avail = !s->first_slice_line || (k==2 || k==3);
3065 v->c_avail = !!s->mb_x || (k==1 || k==3);
3067 vc1_decode_i_block_adv(v, block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant);
3069 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
3070 v->vc1dsp.vc1_inv_trans_8x8(block[k]);
3073 vc1_smooth_overlap_filter_iblk(v);
3074 vc1_put_signed_blocks_clamped(v);
3075 if(v->s.loop_filter) vc1_loop_filter_iblk_delayed(v, v->pq);
3077 if(get_bits_count(&s->gb) > v->bits) {
3078 ff_er_add_slice(s, 0, s->start_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3079 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
3083 if (!v->s.loop_filter)
3084 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3086 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
3087 s->first_slice_line = 0;
3090 /* raw bottom MB row */
3092 ff_init_block_index(s);
3093 for(;s->mb_x < s->mb_width; s->mb_x++) {
3094 ff_update_block_index(s);
3095 vc1_put_signed_blocks_clamped(v);
3096 if(v->s.loop_filter) vc1_loop_filter_iblk_delayed(v, v->pq);
3098 if (v->s.loop_filter)
3099 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
3100 ff_er_add_slice(s, 0, s->start_mb_y, s->mb_width - 1, s->end_mb_y - 1, (AC_END|DC_END|MV_END));
3103 static void vc1_decode_p_blocks(VC1Context *v)
3105 MpegEncContext *s = &v->s;
3106 int apply_loop_filter;
3108 /* select codingmode used for VLC tables selection */
3109 switch(v->c_ac_table_index){
3111 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
3114 v->codingset = CS_HIGH_MOT_INTRA;
3117 v->codingset = CS_MID_RATE_INTRA;
3121 switch(v->c_ac_table_index){
3123 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3126 v->codingset2 = CS_HIGH_MOT_INTER;
3129 v->codingset2 = CS_MID_RATE_INTER;
3133 apply_loop_filter = s->loop_filter && !(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY);
3134 s->first_slice_line = 1;
3135 memset(v->cbp_base, 0, sizeof(v->cbp_base[0])*2*s->mb_stride);
3136 for(s->mb_y = s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
3138 ff_init_block_index(s);
3139 for(; s->mb_x < s->mb_width; s->mb_x++) {
3140 ff_update_block_index(s);
3143 if (s->mb_y != s->start_mb_y && apply_loop_filter)
3144 vc1_apply_p_loop_filter(v);
3145 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
3146 ff_er_add_slice(s, 0, s->start_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3147 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i at %ix%i\n", get_bits_count(&s->gb), v->bits,s->mb_x,s->mb_y);
3151 memmove(v->cbp_base, v->cbp, sizeof(v->cbp_base[0])*s->mb_stride);
3152 memmove(v->ttblk_base, v->ttblk, sizeof(v->ttblk_base[0])*s->mb_stride);
3153 memmove(v->is_intra_base, v->is_intra, sizeof(v->is_intra_base[0])*s->mb_stride);
3154 memmove(v->luma_mv_base, v->luma_mv, sizeof(v->luma_mv_base[0])*s->mb_stride);
3155 if (s->mb_y != s->start_mb_y) ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
3156 s->first_slice_line = 0;
3158 if (apply_loop_filter) {
3160 ff_init_block_index(s);
3161 for (; s->mb_x < s->mb_width; s->mb_x++) {
3162 ff_update_block_index(s);
3163 vc1_apply_p_loop_filter(v);
3166 if (s->end_mb_y >= s->start_mb_y)
3167 ff_draw_horiz_band(s, (s->end_mb_y-1) * 16, 16);
3168 ff_er_add_slice(s, 0, s->start_mb_y, s->mb_width - 1, s->end_mb_y - 1, (AC_END|DC_END|MV_END));
3171 static void vc1_decode_b_blocks(VC1Context *v)
3173 MpegEncContext *s = &v->s;
3175 /* select codingmode used for VLC tables selection */
3176 switch(v->c_ac_table_index){
3178 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
3181 v->codingset = CS_HIGH_MOT_INTRA;
3184 v->codingset = CS_MID_RATE_INTRA;
3188 switch(v->c_ac_table_index){
3190 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3193 v->codingset2 = CS_HIGH_MOT_INTER;
3196 v->codingset2 = CS_MID_RATE_INTER;
3200 s->first_slice_line = 1;
3201 for(s->mb_y = s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
3203 ff_init_block_index(s);
3204 for(; s->mb_x < s->mb_width; s->mb_x++) {
3205 ff_update_block_index(s);
3208 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
3209 ff_er_add_slice(s, 0, s->start_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3210 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i at %ix%i\n", get_bits_count(&s->gb), v->bits,s->mb_x,s->mb_y);
3213 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
3215 if (!v->s.loop_filter)
3216 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3218 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
3219 s->first_slice_line = 0;
3221 if (v->s.loop_filter)
3222 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
3223 ff_er_add_slice(s, 0, s->start_mb_y, s->mb_width - 1, s->end_mb_y - 1, (AC_END|DC_END|MV_END));
3226 static void vc1_decode_skip_blocks(VC1Context *v)
3228 MpegEncContext *s = &v->s;
3230 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
3231 s->first_slice_line = 1;
3232 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
3234 ff_init_block_index(s);
3235 ff_update_block_index(s);
3236 memcpy(s->dest[0], s->last_picture.data[0] + s->mb_y * 16 * s->linesize, s->linesize * 16);
3237 memcpy(s->dest[1], s->last_picture.data[1] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
3238 memcpy(s->dest[2], s->last_picture.data[2] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
3239 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3240 s->first_slice_line = 0;
3242 s->pict_type = AV_PICTURE_TYPE_P;
3245 static void vc1_decode_blocks(VC1Context *v)
3248 v->s.esc3_level_length = 0;
3250 ff_intrax8_decode_picture(&v->x8, 2*v->pq+v->halfpq, v->pq*(!v->pquantizer) );
3253 v->left_blk_idx = -1;
3254 v->topleft_blk_idx = 1;
3256 switch(v->s.pict_type) {
3257 case AV_PICTURE_TYPE_I:
3258 if(v->profile == PROFILE_ADVANCED)
3259 vc1_decode_i_blocks_adv(v);
3261 vc1_decode_i_blocks(v);
3263 case AV_PICTURE_TYPE_P:
3264 if(v->p_frame_skipped)
3265 vc1_decode_skip_blocks(v);
3267 vc1_decode_p_blocks(v);
3269 case AV_PICTURE_TYPE_B:
3271 if(v->profile == PROFILE_ADVANCED)
3272 vc1_decode_i_blocks_adv(v);
3274 vc1_decode_i_blocks(v);
3276 vc1_decode_b_blocks(v);
3282 static inline float get_float_val(GetBitContext* gb)
3284 return (float)get_bits_long(gb, 30) / (1<<15) - (1<<14);
3287 static void vc1_sprite_parse_transform(VC1Context *v, GetBitContext* gb, float c[7])
3291 switch (get_bits(gb, 2)) {
3294 c[2] = get_float_val(gb);
3298 c[0] = c[4] = get_float_val(gb);
3299 c[2] = get_float_val(gb);
3302 c[0] = get_float_val(gb);
3303 c[2] = get_float_val(gb);
3304 c[4] = get_float_val(gb);
3307 av_log_ask_for_sample(v->s.avctx, NULL);
3308 c[0] = get_float_val(gb);
3309 c[1] = get_float_val(gb);
3310 c[2] = get_float_val(gb);
3311 c[3] = get_float_val(gb);
3312 c[4] = get_float_val(gb);
3315 c[5] = get_float_val(gb);
3317 c[6] = get_float_val(gb);
3322 static void vc1_parse_sprites(VC1Context *v, GetBitContext* gb)
3324 int effect_type, effect_flag, effect_pcount1, effect_pcount2, i;
3325 float effect_params1[14], effect_params2[10];
3328 vc1_sprite_parse_transform(v, gb, coefs[0]);
3329 av_log(v->s.avctx, AV_LOG_DEBUG, "S1:");
3330 for (i = 0; i < 7; i++)
3331 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", coefs[0][i]);
3332 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3334 if (v->two_sprites) {
3335 vc1_sprite_parse_transform(v, gb, coefs[1]);
3336 av_log(v->s.avctx, AV_LOG_DEBUG, "S2:");
3337 for (i = 0; i < 7; i++)
3338 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", coefs[1][i]);
3339 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3342 if (effect_type = get_bits_long(gb, 30)){
3343 switch (effect_pcount1 = get_bits(gb, 4)) {
3345 effect_params1[0] = get_float_val(gb);
3346 effect_params1[1] = get_float_val(gb);
3349 vc1_sprite_parse_transform(v, gb, effect_params1);
3352 vc1_sprite_parse_transform(v, gb, effect_params1);
3353 vc1_sprite_parse_transform(v, gb, &effect_params1[7]);
3356 av_log_ask_for_sample(v->s.avctx, NULL);
3359 if (effect_type != 13 || effect_params1[0] != coefs[0][6]) {
3360 // effect 13 is simple alpha blending and matches the opacity above
3361 av_log(v->s.avctx, AV_LOG_DEBUG, "Effect: %d; params: ", effect_type);
3362 for (i = 0; i < effect_pcount1; i++)
3363 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", effect_params1[i]);
3364 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3367 effect_pcount2 = get_bits(gb, 16);
3368 if (effect_pcount2 > 10) {
3369 av_log(v->s.avctx, AV_LOG_ERROR, "Too many effect parameters\n");
3371 } else if (effect_pcount2) {
3373 av_log(v->s.avctx, AV_LOG_DEBUG, "Effect params 2: ");
3374 while (i < effect_pcount2){
3375 effect_params2[i] = get_float_val(gb);
3376 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", effect_params2[i]);
3379 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3382 if (effect_flag = get_bits1(gb))
3383 av_log(v->s.avctx, AV_LOG_DEBUG, "Effect flag set\n");
3385 if (get_bits_count(gb) >= gb->size_in_bits +
3386 (v->s.avctx->codec_id == CODEC_ID_WMV3 ? 64 : 0))
3387 av_log(v->s.avctx, AV_LOG_ERROR, "Buffer overrun\n");
3388 if (get_bits_count(gb) < gb->size_in_bits - 8)
3389 av_log(v->s.avctx, AV_LOG_WARNING, "Buffer not fully read\n");
3392 /** Initialize a VC1/WMV3 decoder
3393 * @todo TODO: Handle VC-1 IDUs (Transport level?)
3394 * @todo TODO: Decypher remaining bits in extra_data
3396 static av_cold int vc1_decode_init(AVCodecContext *avctx)
3398 VC1Context *v = avctx->priv_data;
3399 MpegEncContext *s = &v->s;
3401 int i, cur_width, cur_height;
3403 if (!avctx->extradata_size || !avctx->extradata) return -1;
3404 if (!(avctx->flags & CODEC_FLAG_GRAY))
3405 avctx->pix_fmt = avctx->get_format(avctx, avctx->codec->pix_fmts);
3407 avctx->pix_fmt = PIX_FMT_GRAY8;
3408 avctx->hwaccel = ff_find_hwaccel(avctx->codec->id, avctx->pix_fmt);
3410 avctx->flags |= CODEC_FLAG_EMU_EDGE;
3411 v->s.flags |= CODEC_FLAG_EMU_EDGE;
3413 if(avctx->idct_algo==FF_IDCT_AUTO){
3414 avctx->idct_algo=FF_IDCT_WMV2;
3417 if(ff_msmpeg4_decode_init(avctx) < 0)
3419 if (vc1_init_common(v) < 0) return -1;
3420 ff_vc1dsp_init(&v->vc1dsp);
3422 cur_width = avctx->coded_width = avctx->width;
3423 cur_height = avctx->coded_height = avctx->height;
3424 if (avctx->codec_id == CODEC_ID_WMV3)
3428 // looks like WMV3 has a sequence header stored in the extradata
3429 // advanced sequence header may be before the first frame
3430 // the last byte of the extradata is a version number, 1 for the
3431 // samples we can decode
3433 init_get_bits(&gb, avctx->extradata, avctx->extradata_size*8);
3435 if (vc1_decode_sequence_header(avctx, v, &gb) < 0)
3438 count = avctx->extradata_size*8 - get_bits_count(&gb);
3441 av_log(avctx, AV_LOG_INFO, "Extra data: %i bits left, value: %X\n",
3442 count, get_bits(&gb, count));
3446 av_log(avctx, AV_LOG_INFO, "Read %i bits in overflow\n", -count);
3448 } else { // VC1/WVC1/WVP2
3449 const uint8_t *start = avctx->extradata;
3450 uint8_t *end = avctx->extradata + avctx->extradata_size;
3451 const uint8_t *next;
3452 int size, buf2_size;
3453 uint8_t *buf2 = NULL;
3454 int seq_initialized = 0, ep_initialized = 0;
3456 if(avctx->extradata_size < 16) {
3457 av_log(avctx, AV_LOG_ERROR, "Extradata size too small: %i\n", avctx->extradata_size);
3461 buf2 = av_mallocz(avctx->extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
3462 start = find_next_marker(start, end); // in WVC1 extradata first byte is its size, but can be 0 in mkv
3464 for(; next < end; start = next){
3465 next = find_next_marker(start + 4, end);
3466 size = next - start - 4;
3467 if(size <= 0) continue;
3468 buf2_size = vc1_unescape_buffer(start + 4, size, buf2);
3469 init_get_bits(&gb, buf2, buf2_size * 8);
3470 switch(AV_RB32(start)){
3471 case VC1_CODE_SEQHDR:
3472 if(vc1_decode_sequence_header(avctx, v, &gb) < 0){
3476 seq_initialized = 1;
3478 case VC1_CODE_ENTRYPOINT:
3479 if(vc1_decode_entry_point(avctx, v, &gb) < 0){
3488 if(!seq_initialized || !ep_initialized){
3489 av_log(avctx, AV_LOG_ERROR, "Incomplete extradata\n");
3492 v->res_sprite = (avctx->codec_tag == MKTAG('W','V','P','2'));
3494 // Sequence header information may not have been parsed
3495 // yet when ff_msmpeg4_decode_init was called the fist time
3496 // above. If sequence information changes, we need to call
3498 if (cur_width != avctx->width ||
3499 cur_height != avctx->height) {
3501 if(ff_msmpeg4_decode_init(avctx) < 0)
3503 avctx->coded_width = avctx->width;
3504 avctx->coded_height = avctx->height;
3507 avctx->profile = v->profile;
3508 if (v->profile == PROFILE_ADVANCED)
3509 avctx->level = v->level;
3511 avctx->has_b_frames= !!(avctx->max_b_frames);
3512 s->low_delay = !avctx->has_b_frames;
3514 s->mb_width = (avctx->coded_width+15)>>4;
3515 s->mb_height = (avctx->coded_height+15)>>4;
3517 if (v->profile == PROFILE_ADVANCED || v->res_fasttx) {
3518 for (i = 0; i < 64; i++) {
3519 #define transpose(x) ((x>>3) | ((x&7)<<3))
3520 v->zz_8x8[0][i] = transpose(wmv1_scantable[0][i]);
3521 v->zz_8x8[1][i] = transpose(wmv1_scantable[1][i]);
3522 v->zz_8x8[2][i] = transpose(wmv1_scantable[2][i]);
3523 v->zz_8x8[3][i] = transpose(wmv1_scantable[3][i]);
3528 memcpy(v->zz_8x8, wmv1_scantable, 4*64);
3533 /* Allocate mb bitplanes */
3534 v->mv_type_mb_plane = av_malloc(s->mb_stride * s->mb_height);
3535 v->direct_mb_plane = av_malloc(s->mb_stride * s->mb_height);
3536 v->acpred_plane = av_malloc(s->mb_stride * s->mb_height);
3537 v->over_flags_plane = av_malloc(s->mb_stride * s->mb_height);
3539 v->n_allocated_blks = s->mb_width + 2;
3540 v->block = av_malloc(sizeof(*v->block) * v->n_allocated_blks);
3541 v->cbp_base = av_malloc(sizeof(v->cbp_base[0]) * 2 * s->mb_stride);
3542 v->cbp = v->cbp_base + s->mb_stride;
3543 v->ttblk_base = av_malloc(sizeof(v->ttblk_base[0]) * 2 * s->mb_stride);
3544 v->ttblk = v->ttblk_base + s->mb_stride;
3545 v->is_intra_base = av_malloc(sizeof(v->is_intra_base[0]) * 2 * s->mb_stride);
3546 v->is_intra = v->is_intra_base + s->mb_stride;
3547 v->luma_mv_base = av_malloc(sizeof(v->luma_mv_base[0]) * 2 * s->mb_stride);
3548 v->luma_mv = v->luma_mv_base + s->mb_stride;
3550 /* allocate block type info in that way so it could be used with s->block_index[] */
3551 v->mb_type_base = av_malloc(s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride * (s->mb_height + 1) * 2);
3552 v->mb_type[0] = v->mb_type_base + s->b8_stride + 1;
3553 v->mb_type[1] = v->mb_type_base + s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride + 1;
3554 v->mb_type[2] = v->mb_type[1] + s->mb_stride * (s->mb_height + 1);
3556 /* Init coded blocks info */
3557 if (v->profile == PROFILE_ADVANCED)
3559 // if (alloc_bitplane(&v->over_flags_plane, s->mb_width, s->mb_height) < 0)
3561 // if (alloc_bitplane(&v->ac_pred_plane, s->mb_width, s->mb_height) < 0)
3565 ff_intrax8_common_init(&v->x8,s);
3570 /** Decode a VC1/WMV3 frame
3571 * @todo TODO: Handle VC-1 IDUs (Transport level?)
3573 static int vc1_decode_frame(AVCodecContext *avctx,
3574 void *data, int *data_size,
3577 const uint8_t *buf = avpkt->data;
3578 int buf_size = avpkt->size, n_slices = 0, i;
3579 VC1Context *v = avctx->priv_data;
3580 MpegEncContext *s = &v->s;
3581 AVFrame *pict = data;
3582 uint8_t *buf2 = NULL;
3583 const uint8_t *buf_start = buf;
3590 /* no supplementary picture */
3591 if (buf_size == 0 || (buf_size == 4 && AV_RB32(buf) == VC1_CODE_ENDOFSEQ)) {
3592 /* special case for last picture */
3593 if (s->low_delay==0 && s->next_picture_ptr) {
3594 *pict= *(AVFrame*)s->next_picture_ptr;
3595 s->next_picture_ptr= NULL;
3597 *data_size = sizeof(AVFrame);
3603 /* We need to set current_picture_ptr before reading the header,
3604 * otherwise we cannot store anything in there. */
3605 if(s->current_picture_ptr==NULL || s->current_picture_ptr->data[0]){
3606 int i= ff_find_unused_picture(s, 0);
3607 s->current_picture_ptr= &s->picture[i];
3610 if (s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3611 if (v->profile < PROFILE_ADVANCED)
3612 avctx->pix_fmt = PIX_FMT_VDPAU_WMV3;
3614 avctx->pix_fmt = PIX_FMT_VDPAU_VC1;
3617 //for advanced profile we may need to parse and unescape data
3618 if (avctx->codec_id == CODEC_ID_VC1) {
3620 buf2 = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
3622 if(IS_MARKER(AV_RB32(buf))){ /* frame starts with marker and needs to be parsed */
3623 const uint8_t *start, *end, *next;
3627 for(start = buf, end = buf + buf_size; next < end; start = next){
3628 next = find_next_marker(start + 4, end);
3629 size = next - start - 4;
3630 if(size <= 0) continue;
3631 switch(AV_RB32(start)){
3632 case VC1_CODE_FRAME:
3633 if (avctx->hwaccel ||
3634 s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3636 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
3638 case VC1_CODE_ENTRYPOINT: /* it should be before frame data */
3639 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
3640 init_get_bits(&s->gb, buf2, buf_size2*8);
3641 vc1_decode_entry_point(avctx, v, &s->gb);
3643 case VC1_CODE_SLICE: {
3645 slices = av_realloc(slices, sizeof(*slices) * (n_slices+1));
3646 if (!slices) goto err;
3647 slices[n_slices].buf = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
3648 if (!slices[n_slices].buf) goto err;
3649 buf_size3 = vc1_unescape_buffer(start + 4, size,
3650 slices[n_slices].buf);
3651 init_get_bits(&slices[n_slices].gb, slices[n_slices].buf,
3653 slices[n_slices].mby_start = get_bits(&slices[n_slices].gb, 9);
3659 }else if(v->interlace && ((buf[0] & 0xC0) == 0xC0)){ /* WVC1 interlaced stores both fields divided by marker */
3660 const uint8_t *divider;
3662 divider = find_next_marker(buf, buf + buf_size);
3663 if((divider == (buf + buf_size)) || AV_RB32(divider) != VC1_CODE_FIELD){
3664 av_log(avctx, AV_LOG_ERROR, "Error in WVC1 interlaced frame\n");
3668 buf_size2 = vc1_unescape_buffer(buf, divider - buf, buf2);
3670 if(!v->warn_interlaced++)
3671 av_log(v->s.avctx, AV_LOG_ERROR, "Interlaced WVC1 support is not implemented\n");
3674 buf_size2 = vc1_unescape_buffer(buf, buf_size, buf2);
3676 init_get_bits(&s->gb, buf2, buf_size2*8);
3678 init_get_bits(&s->gb, buf, buf_size*8);
3680 if (v->res_sprite) {
3681 v->new_sprite = !get_bits1(&s->gb);
3682 v->two_sprites = get_bits1(&s->gb);
3687 // do parse frame header
3688 if(v->profile < PROFILE_ADVANCED) {
3689 if(vc1_parse_frame_header(v, &s->gb) == -1) {
3693 if(vc1_parse_frame_header_adv(v, &s->gb) == -1) {
3698 if (v->res_sprite && s->pict_type!=AV_PICTURE_TYPE_I) {
3699 av_log(v->s.avctx, AV_LOG_WARNING, "Sprite decoder: expected I-frame\n");
3702 s->current_picture_ptr->repeat_pict = 0;
3704 s->current_picture_ptr->repeat_pict = 1;
3705 }else if (v->rptfrm){
3706 s->current_picture_ptr->repeat_pict = v->rptfrm * 2;
3709 s->current_picture_ptr->top_field_first = v->tff;
3711 // for skipping the frame
3712 s->current_picture.pict_type= s->pict_type;
3713 s->current_picture.key_frame= s->pict_type == AV_PICTURE_TYPE_I;
3715 /* skip B-frames if we don't have reference frames */
3716 if(s->last_picture_ptr==NULL && (s->pict_type==AV_PICTURE_TYPE_B || s->dropable)){
3719 if( (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type==AV_PICTURE_TYPE_B)
3720 || (avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type!=AV_PICTURE_TYPE_I)
3721 || avctx->skip_frame >= AVDISCARD_ALL) {
3725 if(s->next_p_frame_damaged){
3726 if(s->pict_type==AV_PICTURE_TYPE_B)
3729 s->next_p_frame_damaged=0;
3732 if(MPV_frame_start(s, avctx) < 0) {
3736 s->me.qpel_put= s->dsp.put_qpel_pixels_tab;
3737 s->me.qpel_avg= s->dsp.avg_qpel_pixels_tab;
3739 if ((CONFIG_VC1_VDPAU_DECODER)
3740 &&s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3741 ff_vdpau_vc1_decode_picture(s, buf_start, (buf + buf_size) - buf_start);
3742 else if (avctx->hwaccel) {
3743 if (avctx->hwaccel->start_frame(avctx, buf, buf_size) < 0)
3745 if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
3747 if (avctx->hwaccel->end_frame(avctx) < 0)
3750 ff_er_frame_start(s);
3752 v->bits = buf_size * 8;
3753 for (i = 0; i <= n_slices; i++) {
3754 if (i && get_bits1(&s->gb))
3755 vc1_parse_frame_header_adv(v, &s->gb);
3756 s->start_mb_y = (i == 0) ? 0 : FFMAX(0, slices[i-1].mby_start);
3757 s->end_mb_y = (i == n_slices) ? s->mb_height : FFMIN(s->mb_height, slices[i].mby_start);
3758 vc1_decode_blocks(v);
3759 if (i != n_slices) s->gb = slices[i].gb;
3761 //av_log(s->avctx, AV_LOG_INFO, "Consumed %i/%i bits\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3762 // if(get_bits_count(&s->gb) > buf_size * 8)
3769 assert(s->current_picture.pict_type == s->current_picture_ptr->pict_type);
3770 assert(s->current_picture.pict_type == s->pict_type);
3771 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
3772 *pict= *(AVFrame*)s->current_picture_ptr;
3773 } else if (s->last_picture_ptr != NULL) {
3774 *pict= *(AVFrame*)s->last_picture_ptr;
3777 if(s->last_picture_ptr || s->low_delay){
3778 *data_size = sizeof(AVFrame);
3779 ff_print_debug_info(s, pict);
3784 vc1_parse_sprites(v, &s->gb);
3786 for (i = 0; i < n_slices; i++)
3787 av_free(slices[i].buf);
3793 for (i = 0; i < n_slices; i++)
3794 av_free(slices[i].buf);
3800 /** Close a VC1/WMV3 decoder
3801 * @warning Initial try at using MpegEncContext stuff
3803 static av_cold int vc1_decode_end(AVCodecContext *avctx)
3805 VC1Context *v = avctx->priv_data;
3807 av_freep(&v->hrd_rate);
3808 av_freep(&v->hrd_buffer);
3809 MPV_common_end(&v->s);
3810 av_freep(&v->mv_type_mb_plane);
3811 av_freep(&v->direct_mb_plane);
3812 av_freep(&v->acpred_plane);
3813 av_freep(&v->over_flags_plane);
3814 av_freep(&v->mb_type_base);
3815 av_freep(&v->block);
3816 av_freep(&v->cbp_base);
3817 av_freep(&v->ttblk_base);
3818 av_freep(&v->is_intra_base); // FIXME use v->mb_type[]
3819 av_freep(&v->luma_mv_base);
3820 ff_intrax8_common_end(&v->x8);
3824 static const AVProfile profiles[] = {
3825 { FF_PROFILE_VC1_SIMPLE, "Simple" },
3826 { FF_PROFILE_VC1_MAIN, "Main" },
3827 { FF_PROFILE_VC1_COMPLEX, "Complex" },
3828 { FF_PROFILE_VC1_ADVANCED, "Advanced" },
3829 { FF_PROFILE_UNKNOWN },
3832 AVCodec ff_vc1_decoder = {
3841 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3843 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1"),
3844 .pix_fmts = ff_hwaccel_pixfmt_list_420,
3845 .profiles = NULL_IF_CONFIG_SMALL(profiles)
3848 #if CONFIG_WMV3_DECODER
3849 AVCodec ff_wmv3_decoder = {
3858 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3860 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9"),
3861 .pix_fmts = ff_hwaccel_pixfmt_list_420,
3862 .profiles = NULL_IF_CONFIG_SMALL(profiles)
3866 #if CONFIG_WMV3_VDPAU_DECODER
3867 AVCodec ff_wmv3_vdpau_decoder = {
3876 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3878 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 VDPAU"),
3879 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_WMV3, PIX_FMT_NONE},
3880 .profiles = NULL_IF_CONFIG_SMALL(profiles)
3884 #if CONFIG_VC1_VDPAU_DECODER
3885 AVCodec ff_vc1_vdpau_decoder = {
3894 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3896 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1 VDPAU"),
3897 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_VC1, PIX_FMT_NONE},
3898 .profiles = NULL_IF_CONFIG_SMALL(profiles)