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_loop_filter_iblk(VC1Context *v, int pq)
165 MpegEncContext *s = &v->s;
167 if (!s->first_slice_line) {
168 v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
170 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16*s->linesize, s->linesize, pq);
171 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16*s->linesize+8, s->linesize, pq);
172 for(j = 0; j < 2; j++){
173 v->vc1dsp.vc1_v_loop_filter8(s->dest[j+1], s->uvlinesize, pq);
175 v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1]-8*s->uvlinesize, s->uvlinesize, pq);
178 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8*s->linesize, s->linesize, pq);
180 if (s->mb_y == s->mb_height-1) {
182 v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
183 v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
184 v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
186 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
190 /** Do motion compensation over 1 macroblock
191 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
193 static void vc1_mc_1mv(VC1Context *v, int dir)
195 MpegEncContext *s = &v->s;
196 DSPContext *dsp = &v->s.dsp;
197 uint8_t *srcY, *srcU, *srcV;
198 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
200 if(!v->s.last_picture.data[0])return;
202 mx = s->mv[dir][0][0];
203 my = s->mv[dir][0][1];
205 // store motion vectors for further use in B frames
206 if(s->pict_type == FF_P_TYPE) {
207 s->current_picture.motion_val[1][s->block_index[0]][0] = mx;
208 s->current_picture.motion_val[1][s->block_index[0]][1] = my;
210 uvmx = (mx + ((mx & 3) == 3)) >> 1;
211 uvmy = (my + ((my & 3) == 3)) >> 1;
213 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
214 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
217 srcY = s->last_picture.data[0];
218 srcU = s->last_picture.data[1];
219 srcV = s->last_picture.data[2];
221 srcY = s->next_picture.data[0];
222 srcU = s->next_picture.data[1];
223 srcV = s->next_picture.data[2];
226 src_x = s->mb_x * 16 + (mx >> 2);
227 src_y = s->mb_y * 16 + (my >> 2);
228 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
229 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
231 if(v->profile != PROFILE_ADVANCED){
232 src_x = av_clip( src_x, -16, s->mb_width * 16);
233 src_y = av_clip( src_y, -16, s->mb_height * 16);
234 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
235 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
237 src_x = av_clip( src_x, -17, s->avctx->coded_width);
238 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
239 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
240 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
243 srcY += src_y * s->linesize + src_x;
244 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
245 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
247 /* for grayscale we should not try to read from unknown area */
248 if(s->flags & CODEC_FLAG_GRAY) {
249 srcU = s->edge_emu_buffer + 18 * s->linesize;
250 srcV = s->edge_emu_buffer + 18 * s->linesize;
253 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
254 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
255 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
256 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
258 srcY -= s->mspel * (1 + s->linesize);
259 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
260 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
261 srcY = s->edge_emu_buffer;
262 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
263 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
264 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
265 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
268 /* if we deal with range reduction we need to scale source blocks */
274 for(j = 0; j < 17 + s->mspel*2; j++) {
275 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
278 src = srcU; src2 = srcV;
279 for(j = 0; j < 9; j++) {
280 for(i = 0; i < 9; i++) {
281 src[i] = ((src[i] - 128) >> 1) + 128;
282 src2[i] = ((src2[i] - 128) >> 1) + 128;
284 src += s->uvlinesize;
285 src2 += s->uvlinesize;
288 /* if we deal with intensity compensation we need to scale source blocks */
289 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
294 for(j = 0; j < 17 + s->mspel*2; j++) {
295 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = v->luty[src[i]];
298 src = srcU; src2 = srcV;
299 for(j = 0; j < 9; j++) {
300 for(i = 0; i < 9; i++) {
301 src[i] = v->lutuv[src[i]];
302 src2[i] = v->lutuv[src2[i]];
304 src += s->uvlinesize;
305 src2 += s->uvlinesize;
308 srcY += s->mspel * (1 + s->linesize);
312 dxy = ((my & 3) << 2) | (mx & 3);
313 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
314 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
315 srcY += s->linesize * 8;
316 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
317 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
318 } else { // hpel mc - always used for luma
319 dxy = (my & 2) | ((mx & 2) >> 1);
322 dsp->put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
324 dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
327 if(s->flags & CODEC_FLAG_GRAY) return;
328 /* Chroma MC always uses qpel bilinear */
332 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
333 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
335 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
336 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
340 /** Do motion compensation for 4-MV macroblock - luminance block
342 static void vc1_mc_4mv_luma(VC1Context *v, int n)
344 MpegEncContext *s = &v->s;
345 DSPContext *dsp = &v->s.dsp;
347 int dxy, mx, my, src_x, src_y;
350 if(!v->s.last_picture.data[0])return;
353 srcY = s->last_picture.data[0];
355 off = s->linesize * 4 * (n&2) + (n&1) * 8;
357 src_x = s->mb_x * 16 + (n&1) * 8 + (mx >> 2);
358 src_y = s->mb_y * 16 + (n&2) * 4 + (my >> 2);
360 if(v->profile != PROFILE_ADVANCED){
361 src_x = av_clip( src_x, -16, s->mb_width * 16);
362 src_y = av_clip( src_y, -16, s->mb_height * 16);
364 src_x = av_clip( src_x, -17, s->avctx->coded_width);
365 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
368 srcY += src_y * s->linesize + src_x;
370 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
371 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 8 - s->mspel*2
372 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 8 - s->mspel*2){
373 srcY -= s->mspel * (1 + s->linesize);
374 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 9+s->mspel*2, 9+s->mspel*2,
375 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
376 srcY = s->edge_emu_buffer;
377 /* if we deal with range reduction we need to scale source blocks */
383 for(j = 0; j < 9 + s->mspel*2; j++) {
384 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
388 /* if we deal with intensity compensation we need to scale source blocks */
389 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
394 for(j = 0; j < 9 + s->mspel*2; j++) {
395 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = v->luty[src[i]];
399 srcY += s->mspel * (1 + s->linesize);
403 dxy = ((my & 3) << 2) | (mx & 3);
404 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, v->rnd);
405 } else { // hpel mc - always used for luma
406 dxy = (my & 2) | ((mx & 2) >> 1);
408 dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
410 dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
414 static inline int median4(int a, int b, int c, int d)
417 if(c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
418 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
420 if(c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
421 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
426 /** Do motion compensation for 4-MV macroblock - both chroma blocks
428 static void vc1_mc_4mv_chroma(VC1Context *v)
430 MpegEncContext *s = &v->s;
431 DSPContext *dsp = &v->s.dsp;
432 uint8_t *srcU, *srcV;
433 int uvmx, uvmy, uvsrc_x, uvsrc_y;
434 int i, idx, tx = 0, ty = 0;
435 int mvx[4], mvy[4], intra[4];
436 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
438 if(!v->s.last_picture.data[0])return;
439 if(s->flags & CODEC_FLAG_GRAY) return;
441 for(i = 0; i < 4; i++) {
442 mvx[i] = s->mv[0][i][0];
443 mvy[i] = s->mv[0][i][1];
444 intra[i] = v->mb_type[0][s->block_index[i]];
447 /* calculate chroma MV vector from four luma MVs */
448 idx = (intra[3] << 3) | (intra[2] << 2) | (intra[1] << 1) | intra[0];
449 if(!idx) { // all blocks are inter
450 tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
451 ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
452 } else if(count[idx] == 1) { // 3 inter blocks
455 tx = mid_pred(mvx[1], mvx[2], mvx[3]);
456 ty = mid_pred(mvy[1], mvy[2], mvy[3]);
459 tx = mid_pred(mvx[0], mvx[2], mvx[3]);
460 ty = mid_pred(mvy[0], mvy[2], mvy[3]);
463 tx = mid_pred(mvx[0], mvx[1], mvx[3]);
464 ty = mid_pred(mvy[0], mvy[1], mvy[3]);
467 tx = mid_pred(mvx[0], mvx[1], mvx[2]);
468 ty = mid_pred(mvy[0], mvy[1], mvy[2]);
471 } else if(count[idx] == 2) {
473 for(i=0; i<3;i++) if(!intra[i]) {t1 = i; break;}
474 for(i= t1+1; i<4; i++)if(!intra[i]) {t2 = i; break;}
475 tx = (mvx[t1] + mvx[t2]) / 2;
476 ty = (mvy[t1] + mvy[t2]) / 2;
478 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
479 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
480 return; //no need to do MC for inter blocks
483 s->current_picture.motion_val[1][s->block_index[0]][0] = tx;
484 s->current_picture.motion_val[1][s->block_index[0]][1] = ty;
485 uvmx = (tx + ((tx&3) == 3)) >> 1;
486 uvmy = (ty + ((ty&3) == 3)) >> 1;
488 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
489 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
492 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
493 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
495 if(v->profile != PROFILE_ADVANCED){
496 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
497 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
499 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
500 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
503 srcU = s->last_picture.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
504 srcV = s->last_picture.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
505 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
506 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
507 || (unsigned)uvsrc_y > (s->v_edge_pos >> 1) - 9){
508 s->dsp.emulated_edge_mc(s->edge_emu_buffer , srcU, s->uvlinesize, 8+1, 8+1,
509 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
510 s->dsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize, 8+1, 8+1,
511 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
512 srcU = s->edge_emu_buffer;
513 srcV = s->edge_emu_buffer + 16;
515 /* if we deal with range reduction we need to scale source blocks */
520 src = srcU; src2 = srcV;
521 for(j = 0; j < 9; j++) {
522 for(i = 0; i < 9; i++) {
523 src[i] = ((src[i] - 128) >> 1) + 128;
524 src2[i] = ((src2[i] - 128) >> 1) + 128;
526 src += s->uvlinesize;
527 src2 += s->uvlinesize;
530 /* if we deal with intensity compensation we need to scale source blocks */
531 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
535 src = srcU; src2 = srcV;
536 for(j = 0; j < 9; j++) {
537 for(i = 0; i < 9; i++) {
538 src[i] = v->lutuv[src[i]];
539 src2[i] = v->lutuv[src2[i]];
541 src += s->uvlinesize;
542 src2 += s->uvlinesize;
547 /* Chroma MC always uses qpel bilinear */
551 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
552 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
554 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
555 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
559 /***********************************************************************/
561 * @defgroup vc1block VC-1 Block-level functions
562 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
568 * @brief Get macroblock-level quantizer scale
570 #define GET_MQUANT() \
574 if (v->dqprofile == DQPROFILE_ALL_MBS) \
578 mquant = (get_bits1(gb)) ? v->altpq : v->pq; \
582 mqdiff = get_bits(gb, 3); \
583 if (mqdiff != 7) mquant = v->pq + mqdiff; \
584 else mquant = get_bits(gb, 5); \
587 if(v->dqprofile == DQPROFILE_SINGLE_EDGE) \
588 edges = 1 << v->dqsbedge; \
589 else if(v->dqprofile == DQPROFILE_DOUBLE_EDGES) \
590 edges = (3 << v->dqsbedge) % 15; \
591 else if(v->dqprofile == DQPROFILE_FOUR_EDGES) \
593 if((edges&1) && !s->mb_x) \
595 if((edges&2) && s->first_slice_line) \
597 if((edges&4) && s->mb_x == (s->mb_width - 1)) \
599 if((edges&8) && s->mb_y == (s->mb_height - 1)) \
604 * @def GET_MVDATA(_dmv_x, _dmv_y)
605 * @brief Get MV differentials
606 * @see MVDATA decoding from 8.3.5.2, p(1)20
607 * @param _dmv_x Horizontal differential for decoded MV
608 * @param _dmv_y Vertical differential for decoded MV
610 #define GET_MVDATA(_dmv_x, _dmv_y) \
611 index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table,\
612 VC1_MV_DIFF_VLC_BITS, 2); \
618 else mb_has_coeffs = 0; \
620 if (!index) { _dmv_x = _dmv_y = 0; } \
621 else if (index == 35) \
623 _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample); \
624 _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample); \
626 else if (index == 36) \
635 if (!s->quarter_sample && index1 == 5) val = 1; \
637 if(size_table[index1] - val > 0) \
638 val = get_bits(gb, size_table[index1] - val); \
640 sign = 0 - (val&1); \
641 _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
644 if (!s->quarter_sample && index1 == 5) val = 1; \
646 if(size_table[index1] - val > 0) \
647 val = get_bits(gb, size_table[index1] - val); \
649 sign = 0 - (val&1); \
650 _dmv_y = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
653 /** Predict and set motion vector
655 static inline void vc1_pred_mv(MpegEncContext *s, int n, int dmv_x, int dmv_y, int mv1, int r_x, int r_y, uint8_t* is_intra)
657 int xy, wrap, off = 0;
662 /* scale MV difference to be quad-pel */
663 dmv_x <<= 1 - s->quarter_sample;
664 dmv_y <<= 1 - s->quarter_sample;
667 xy = s->block_index[n];
670 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = 0;
671 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = 0;
672 s->current_picture.motion_val[1][xy][0] = 0;
673 s->current_picture.motion_val[1][xy][1] = 0;
674 if(mv1) { /* duplicate motion data for 1-MV block */
675 s->current_picture.motion_val[0][xy + 1][0] = 0;
676 s->current_picture.motion_val[0][xy + 1][1] = 0;
677 s->current_picture.motion_val[0][xy + wrap][0] = 0;
678 s->current_picture.motion_val[0][xy + wrap][1] = 0;
679 s->current_picture.motion_val[0][xy + wrap + 1][0] = 0;
680 s->current_picture.motion_val[0][xy + wrap + 1][1] = 0;
681 s->current_picture.motion_val[1][xy + 1][0] = 0;
682 s->current_picture.motion_val[1][xy + 1][1] = 0;
683 s->current_picture.motion_val[1][xy + wrap][0] = 0;
684 s->current_picture.motion_val[1][xy + wrap][1] = 0;
685 s->current_picture.motion_val[1][xy + wrap + 1][0] = 0;
686 s->current_picture.motion_val[1][xy + wrap + 1][1] = 0;
691 C = s->current_picture.motion_val[0][xy - 1];
692 A = s->current_picture.motion_val[0][xy - wrap];
694 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
696 //in 4-MV mode different blocks have different B predictor position
699 off = (s->mb_x > 0) ? -1 : 1;
702 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
711 B = s->current_picture.motion_val[0][xy - wrap + off];
713 if(!s->first_slice_line || (n==2 || n==3)) { // predictor A is not out of bounds
714 if(s->mb_width == 1) {
718 px = mid_pred(A[0], B[0], C[0]);
719 py = mid_pred(A[1], B[1], C[1]);
721 } else if(s->mb_x || (n==1 || n==3)) { // predictor C is not out of bounds
727 /* Pullback MV as specified in 8.3.5.3.4 */
730 qx = (s->mb_x << 6) + ((n==1 || n==3) ? 32 : 0);
731 qy = (s->mb_y << 6) + ((n==2 || n==3) ? 32 : 0);
732 X = (s->mb_width << 6) - 4;
733 Y = (s->mb_height << 6) - 4;
735 if(qx + px < -60) px = -60 - qx;
736 if(qy + py < -60) py = -60 - qy;
738 if(qx + px < -28) px = -28 - qx;
739 if(qy + py < -28) py = -28 - qy;
741 if(qx + px > X) px = X - qx;
742 if(qy + py > Y) py = Y - qy;
744 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
745 if((!s->first_slice_line || (n==2 || n==3)) && (s->mb_x || (n==1 || n==3))) {
746 if(is_intra[xy - wrap])
747 sum = FFABS(px) + FFABS(py);
749 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
751 if(get_bits1(&s->gb)) {
760 sum = FFABS(px) + FFABS(py);
762 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
764 if(get_bits1(&s->gb)) {
774 /* store MV using signed modulus of MV range defined in 4.11 */
775 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
776 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
777 if(mv1) { /* duplicate motion data for 1-MV block */
778 s->current_picture.motion_val[0][xy + 1][0] = s->current_picture.motion_val[0][xy][0];
779 s->current_picture.motion_val[0][xy + 1][1] = s->current_picture.motion_val[0][xy][1];
780 s->current_picture.motion_val[0][xy + wrap][0] = s->current_picture.motion_val[0][xy][0];
781 s->current_picture.motion_val[0][xy + wrap][1] = s->current_picture.motion_val[0][xy][1];
782 s->current_picture.motion_val[0][xy + wrap + 1][0] = s->current_picture.motion_val[0][xy][0];
783 s->current_picture.motion_val[0][xy + wrap + 1][1] = s->current_picture.motion_val[0][xy][1];
787 /** Motion compensation for direct or interpolated blocks in B-frames
789 static void vc1_interp_mc(VC1Context *v)
791 MpegEncContext *s = &v->s;
792 DSPContext *dsp = &v->s.dsp;
793 uint8_t *srcY, *srcU, *srcV;
794 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
796 if(!v->s.next_picture.data[0])return;
800 uvmx = (mx + ((mx & 3) == 3)) >> 1;
801 uvmy = (my + ((my & 3) == 3)) >> 1;
803 uvmx = uvmx + ((uvmx<0)?-(uvmx&1):(uvmx&1));
804 uvmy = uvmy + ((uvmy<0)?-(uvmy&1):(uvmy&1));
806 srcY = s->next_picture.data[0];
807 srcU = s->next_picture.data[1];
808 srcV = s->next_picture.data[2];
810 src_x = s->mb_x * 16 + (mx >> 2);
811 src_y = s->mb_y * 16 + (my >> 2);
812 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
813 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
815 if(v->profile != PROFILE_ADVANCED){
816 src_x = av_clip( src_x, -16, s->mb_width * 16);
817 src_y = av_clip( src_y, -16, s->mb_height * 16);
818 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
819 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
821 src_x = av_clip( src_x, -17, s->avctx->coded_width);
822 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
823 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
824 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
827 srcY += src_y * s->linesize + src_x;
828 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
829 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
831 /* for grayscale we should not try to read from unknown area */
832 if(s->flags & CODEC_FLAG_GRAY) {
833 srcU = s->edge_emu_buffer + 18 * s->linesize;
834 srcV = s->edge_emu_buffer + 18 * s->linesize;
838 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
839 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
840 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
842 srcY -= s->mspel * (1 + s->linesize);
843 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
844 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
845 srcY = s->edge_emu_buffer;
846 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
847 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
848 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
849 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
852 /* if we deal with range reduction we need to scale source blocks */
858 for(j = 0; j < 17 + s->mspel*2; j++) {
859 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
862 src = srcU; src2 = srcV;
863 for(j = 0; j < 9; j++) {
864 for(i = 0; i < 9; i++) {
865 src[i] = ((src[i] - 128) >> 1) + 128;
866 src2[i] = ((src2[i] - 128) >> 1) + 128;
868 src += s->uvlinesize;
869 src2 += s->uvlinesize;
872 srcY += s->mspel * (1 + s->linesize);
876 dxy = ((my & 3) << 2) | (mx & 3);
877 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
878 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
879 srcY += s->linesize * 8;
880 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
881 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
883 dxy = (my & 2) | ((mx & 2) >> 1);
886 dsp->avg_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
888 dsp->avg_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
891 if(s->flags & CODEC_FLAG_GRAY) return;
892 /* Chroma MC always uses qpel blilinear */
896 dsp->avg_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
897 dsp->avg_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
899 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
900 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
904 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
908 #if B_FRACTION_DEN==256
912 return 2 * ((value * n + 255) >> 9);
913 return (value * n + 128) >> 8;
918 return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
919 return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
923 /** Reconstruct motion vector for B-frame and do motion compensation
925 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mode)
928 v->mv_mode2 = v->mv_mode;
929 v->mv_mode = MV_PMODE_INTENSITY_COMP;
934 if(v->use_ic) v->mv_mode = v->mv_mode2;
937 if(mode == BMV_TYPE_INTERPOLATED) {
940 if(v->use_ic) v->mv_mode = v->mv_mode2;
944 if(v->use_ic && (mode == BMV_TYPE_BACKWARD)) v->mv_mode = v->mv_mode2;
945 vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
946 if(v->use_ic) v->mv_mode = v->mv_mode2;
949 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mvtype)
951 MpegEncContext *s = &v->s;
952 int xy, wrap, off = 0;
957 const uint8_t *is_intra = v->mb_type[0];
961 /* scale MV difference to be quad-pel */
962 dmv_x[0] <<= 1 - s->quarter_sample;
963 dmv_y[0] <<= 1 - s->quarter_sample;
964 dmv_x[1] <<= 1 - s->quarter_sample;
965 dmv_y[1] <<= 1 - s->quarter_sample;
968 xy = s->block_index[0];
971 s->current_picture.motion_val[0][xy][0] =
972 s->current_picture.motion_val[0][xy][1] =
973 s->current_picture.motion_val[1][xy][0] =
974 s->current_picture.motion_val[1][xy][1] = 0;
977 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
978 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
979 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
980 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
982 /* Pullback predicted motion vectors as specified in 8.4.5.4 */
983 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));
984 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));
985 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));
986 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));
988 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
989 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
990 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
991 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
995 if((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
996 C = s->current_picture.motion_val[0][xy - 2];
997 A = s->current_picture.motion_val[0][xy - wrap*2];
998 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
999 B = s->current_picture.motion_val[0][xy - wrap*2 + off];
1001 if(!s->mb_x) C[0] = C[1] = 0;
1002 if(!s->first_slice_line) { // predictor A is not out of bounds
1003 if(s->mb_width == 1) {
1007 px = mid_pred(A[0], B[0], C[0]);
1008 py = mid_pred(A[1], B[1], C[1]);
1010 } else if(s->mb_x) { // predictor C is not out of bounds
1016 /* Pullback MV as specified in 8.3.5.3.4 */
1019 if(v->profile < PROFILE_ADVANCED) {
1020 qx = (s->mb_x << 5);
1021 qy = (s->mb_y << 5);
1022 X = (s->mb_width << 5) - 4;
1023 Y = (s->mb_height << 5) - 4;
1024 if(qx + px < -28) px = -28 - qx;
1025 if(qy + py < -28) py = -28 - qy;
1026 if(qx + px > X) px = X - qx;
1027 if(qy + py > Y) py = Y - qy;
1029 qx = (s->mb_x << 6);
1030 qy = (s->mb_y << 6);
1031 X = (s->mb_width << 6) - 4;
1032 Y = (s->mb_height << 6) - 4;
1033 if(qx + px < -60) px = -60 - qx;
1034 if(qy + py < -60) py = -60 - qy;
1035 if(qx + px > X) px = X - qx;
1036 if(qy + py > Y) py = Y - qy;
1039 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
1040 if(0 && !s->first_slice_line && s->mb_x) {
1041 if(is_intra[xy - wrap])
1042 sum = FFABS(px) + FFABS(py);
1044 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
1046 if(get_bits1(&s->gb)) {
1054 if(is_intra[xy - 2])
1055 sum = FFABS(px) + FFABS(py);
1057 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
1059 if(get_bits1(&s->gb)) {
1069 /* store MV using signed modulus of MV range defined in 4.11 */
1070 s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
1071 s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
1073 if((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
1074 C = s->current_picture.motion_val[1][xy - 2];
1075 A = s->current_picture.motion_val[1][xy - wrap*2];
1076 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1077 B = s->current_picture.motion_val[1][xy - wrap*2 + off];
1079 if(!s->mb_x) C[0] = C[1] = 0;
1080 if(!s->first_slice_line) { // predictor A is not out of bounds
1081 if(s->mb_width == 1) {
1085 px = mid_pred(A[0], B[0], C[0]);
1086 py = mid_pred(A[1], B[1], C[1]);
1088 } else if(s->mb_x) { // predictor C is not out of bounds
1094 /* Pullback MV as specified in 8.3.5.3.4 */
1097 if(v->profile < PROFILE_ADVANCED) {
1098 qx = (s->mb_x << 5);
1099 qy = (s->mb_y << 5);
1100 X = (s->mb_width << 5) - 4;
1101 Y = (s->mb_height << 5) - 4;
1102 if(qx + px < -28) px = -28 - qx;
1103 if(qy + py < -28) py = -28 - qy;
1104 if(qx + px > X) px = X - qx;
1105 if(qy + py > Y) py = Y - qy;
1107 qx = (s->mb_x << 6);
1108 qy = (s->mb_y << 6);
1109 X = (s->mb_width << 6) - 4;
1110 Y = (s->mb_height << 6) - 4;
1111 if(qx + px < -60) px = -60 - qx;
1112 if(qy + py < -60) py = -60 - qy;
1113 if(qx + px > X) px = X - qx;
1114 if(qy + py > Y) py = Y - qy;
1117 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
1118 if(0 && !s->first_slice_line && s->mb_x) {
1119 if(is_intra[xy - wrap])
1120 sum = FFABS(px) + FFABS(py);
1122 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
1124 if(get_bits1(&s->gb)) {
1132 if(is_intra[xy - 2])
1133 sum = FFABS(px) + FFABS(py);
1135 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
1137 if(get_bits1(&s->gb)) {
1147 /* store MV using signed modulus of MV range defined in 4.11 */
1149 s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
1150 s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
1152 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
1153 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
1154 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
1155 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
1158 /** Get predicted DC value for I-frames only
1159 * prediction dir: left=0, top=1
1160 * @param s MpegEncContext
1161 * @param overlap flag indicating that overlap filtering is used
1162 * @param pq integer part of picture quantizer
1163 * @param[in] n block index in the current MB
1164 * @param dc_val_ptr Pointer to DC predictor
1165 * @param dir_ptr Prediction direction for use in AC prediction
1167 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
1168 int16_t **dc_val_ptr, int *dir_ptr)
1170 int a, b, c, wrap, pred, scale;
1172 static const uint16_t dcpred[32] = {
1173 -1, 1024, 512, 341, 256, 205, 171, 146, 128,
1174 114, 102, 93, 85, 79, 73, 68, 64,
1175 60, 57, 54, 51, 49, 47, 45, 43,
1176 41, 39, 38, 37, 35, 34, 33
1179 /* find prediction - wmv3_dc_scale always used here in fact */
1180 if (n < 4) scale = s->y_dc_scale;
1181 else scale = s->c_dc_scale;
1183 wrap = s->block_wrap[n];
1184 dc_val= s->dc_val[0] + s->block_index[n];
1190 b = dc_val[ - 1 - wrap];
1191 a = dc_val[ - wrap];
1193 if (pq < 9 || !overlap)
1195 /* Set outer values */
1196 if (s->first_slice_line && (n!=2 && n!=3)) b=a=dcpred[scale];
1197 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=dcpred[scale];
1201 /* Set outer values */
1202 if (s->first_slice_line && (n!=2 && n!=3)) b=a=0;
1203 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=0;
1206 if (abs(a - b) <= abs(b - c)) {
1214 /* update predictor */
1215 *dc_val_ptr = &dc_val[0];
1220 /** Get predicted DC value
1221 * prediction dir: left=0, top=1
1222 * @param s MpegEncContext
1223 * @param overlap flag indicating that overlap filtering is used
1224 * @param pq integer part of picture quantizer
1225 * @param[in] n block index in the current MB
1226 * @param a_avail flag indicating top block availability
1227 * @param c_avail flag indicating left block availability
1228 * @param dc_val_ptr Pointer to DC predictor
1229 * @param dir_ptr Prediction direction for use in AC prediction
1231 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
1232 int a_avail, int c_avail,
1233 int16_t **dc_val_ptr, int *dir_ptr)
1235 int a, b, c, wrap, pred;
1237 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1240 wrap = s->block_wrap[n];
1241 dc_val= s->dc_val[0] + s->block_index[n];
1247 b = dc_val[ - 1 - wrap];
1248 a = dc_val[ - wrap];
1249 /* scale predictors if needed */
1250 q1 = s->current_picture.qscale_table[mb_pos];
1251 if(c_avail && (n!= 1 && n!=3)) {
1252 q2 = s->current_picture.qscale_table[mb_pos - 1];
1254 c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1256 if(a_avail && (n!= 2 && n!=3)) {
1257 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1259 a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1261 if(a_avail && c_avail && (n!=3)) {
1264 if(n != 2) off -= s->mb_stride;
1265 q2 = s->current_picture.qscale_table[off];
1267 b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1270 if(a_avail && c_avail) {
1271 if(abs(a - b) <= abs(b - c)) {
1278 } else if(a_avail) {
1281 } else if(c_avail) {
1289 /* update predictor */
1290 *dc_val_ptr = &dc_val[0];
1294 /** @} */ // Block group
1297 * @defgroup vc1_std_mb VC1 Macroblock-level functions in Simple/Main Profiles
1298 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1302 static inline int vc1_coded_block_pred(MpegEncContext * s, int n, uint8_t **coded_block_ptr)
1304 int xy, wrap, pred, a, b, c;
1306 xy = s->block_index[n];
1307 wrap = s->b8_stride;
1312 a = s->coded_block[xy - 1 ];
1313 b = s->coded_block[xy - 1 - wrap];
1314 c = s->coded_block[xy - wrap];
1323 *coded_block_ptr = &s->coded_block[xy];
1329 * Decode one AC coefficient
1330 * @param v The VC1 context
1331 * @param last Last coefficient
1332 * @param skip How much zero coefficients to skip
1333 * @param value Decoded AC coefficient value
1334 * @param codingset set of VLC to decode data
1337 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip, int *value, int codingset)
1339 GetBitContext *gb = &v->s.gb;
1340 int index, escape, run = 0, level = 0, lst = 0;
1342 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
1343 if (index != vc1_ac_sizes[codingset] - 1) {
1344 run = vc1_index_decode_table[codingset][index][0];
1345 level = vc1_index_decode_table[codingset][index][1];
1346 lst = index >= vc1_last_decode_table[codingset] || get_bits_left(gb) < 0;
1350 escape = decode210(gb);
1352 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
1353 run = vc1_index_decode_table[codingset][index][0];
1354 level = vc1_index_decode_table[codingset][index][1];
1355 lst = index >= vc1_last_decode_table[codingset];
1358 level += vc1_last_delta_level_table[codingset][run];
1360 level += vc1_delta_level_table[codingset][run];
1363 run += vc1_last_delta_run_table[codingset][level] + 1;
1365 run += vc1_delta_run_table[codingset][level] + 1;
1371 lst = get_bits1(gb);
1372 if(v->s.esc3_level_length == 0) {
1373 if(v->pq < 8 || v->dquantfrm) { // table 59
1374 v->s.esc3_level_length = get_bits(gb, 3);
1375 if(!v->s.esc3_level_length)
1376 v->s.esc3_level_length = get_bits(gb, 2) + 8;
1378 v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
1380 v->s.esc3_run_length = 3 + get_bits(gb, 2);
1382 run = get_bits(gb, v->s.esc3_run_length);
1383 sign = get_bits1(gb);
1384 level = get_bits(gb, v->s.esc3_level_length);
1395 /** Decode intra block in intra frames - should be faster than decode_intra_block
1396 * @param v VC1Context
1397 * @param block block to decode
1398 * @param[in] n subblock index
1399 * @param coded are AC coeffs present or not
1400 * @param codingset set of VLC to decode data
1402 static int vc1_decode_i_block(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset)
1404 GetBitContext *gb = &v->s.gb;
1405 MpegEncContext *s = &v->s;
1406 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1409 int16_t *ac_val, *ac_val2;
1412 /* Get DC differential */
1414 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1416 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1419 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1424 if (dcdiff == 119 /* ESC index value */)
1426 /* TODO: Optimize */
1427 if (v->pq == 1) dcdiff = get_bits(gb, 10);
1428 else if (v->pq == 2) dcdiff = get_bits(gb, 9);
1429 else dcdiff = get_bits(gb, 8);
1434 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1435 else if (v->pq == 2)
1436 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1443 dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
1446 /* Store the quantized DC coeff, used for prediction */
1448 block[0] = dcdiff * s->y_dc_scale;
1450 block[0] = dcdiff * s->c_dc_scale;
1461 int last = 0, skip, value;
1462 const uint8_t *zz_table;
1466 scale = v->pq * 2 + v->halfpq;
1470 zz_table = v->zz_8x8[2];
1472 zz_table = v->zz_8x8[3];
1474 zz_table = v->zz_8x8[1];
1476 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1478 if(dc_pred_dir) //left
1481 ac_val -= 16 * s->block_wrap[n];
1484 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1488 block[zz_table[i++]] = value;
1491 /* apply AC prediction if needed */
1493 if(dc_pred_dir) { //left
1494 for(k = 1; k < 8; k++)
1495 block[k] += ac_val[k];
1497 for(k = 1; k < 8; k++)
1498 block[k << 3] += ac_val[k + 8];
1501 /* save AC coeffs for further prediction */
1502 for(k = 1; k < 8; k++) {
1503 ac_val2[k] = block[k];
1504 ac_val2[k + 8] = block[k << 3];
1507 /* scale AC coeffs */
1508 for(k = 1; k < 64; k++)
1512 block[k] += (block[k] < 0) ? -v->pq : v->pq;
1515 if(s->ac_pred) i = 63;
1521 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1525 scale = v->pq * 2 + v->halfpq;
1526 memset(ac_val2, 0, 16 * 2);
1527 if(dc_pred_dir) {//left
1530 memcpy(ac_val2, ac_val, 8 * 2);
1532 ac_val -= 16 * s->block_wrap[n];
1534 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1537 /* apply AC prediction if needed */
1539 if(dc_pred_dir) { //left
1540 for(k = 1; k < 8; k++) {
1541 block[k] = ac_val[k] * scale;
1542 if(!v->pquantizer && block[k])
1543 block[k] += (block[k] < 0) ? -v->pq : v->pq;
1546 for(k = 1; k < 8; k++) {
1547 block[k << 3] = ac_val[k + 8] * scale;
1548 if(!v->pquantizer && block[k << 3])
1549 block[k << 3] += (block[k << 3] < 0) ? -v->pq : v->pq;
1555 s->block_last_index[n] = i;
1560 /** Decode intra block in intra frames - should be faster than decode_intra_block
1561 * @param v VC1Context
1562 * @param block block to decode
1563 * @param[in] n subblock number
1564 * @param coded are AC coeffs present or not
1565 * @param codingset set of VLC to decode data
1566 * @param mquant quantizer value for this macroblock
1568 static int vc1_decode_i_block_adv(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset, int mquant)
1570 GetBitContext *gb = &v->s.gb;
1571 MpegEncContext *s = &v->s;
1572 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1575 int16_t *ac_val, *ac_val2;
1577 int a_avail = v->a_avail, c_avail = v->c_avail;
1578 int use_pred = s->ac_pred;
1581 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1583 /* Get DC differential */
1585 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1587 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1590 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1595 if (dcdiff == 119 /* ESC index value */)
1597 /* TODO: Optimize */
1598 if (mquant == 1) dcdiff = get_bits(gb, 10);
1599 else if (mquant == 2) dcdiff = get_bits(gb, 9);
1600 else dcdiff = get_bits(gb, 8);
1605 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1606 else if (mquant == 2)
1607 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1614 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
1617 /* Store the quantized DC coeff, used for prediction */
1619 block[0] = dcdiff * s->y_dc_scale;
1621 block[0] = dcdiff * s->c_dc_scale;
1627 /* check if AC is needed at all */
1628 if(!a_avail && !c_avail) use_pred = 0;
1629 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1632 scale = mquant * 2 + ((mquant == v->pq) ? v->halfpq : 0);
1634 if(dc_pred_dir) //left
1637 ac_val -= 16 * s->block_wrap[n];
1639 q1 = s->current_picture.qscale_table[mb_pos];
1640 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
1641 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1642 if(dc_pred_dir && n==1) q2 = q1;
1643 if(!dc_pred_dir && n==2) q2 = q1;
1647 int last = 0, skip, value;
1648 const uint8_t *zz_table;
1653 zz_table = v->zz_8x8[2];
1655 zz_table = v->zz_8x8[3];
1657 zz_table = v->zz_8x8[1];
1660 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1664 block[zz_table[i++]] = value;
1667 /* apply AC prediction if needed */
1669 /* scale predictors if needed*/
1671 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1672 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1674 if(dc_pred_dir) { //left
1675 for(k = 1; k < 8; k++)
1676 block[k] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1678 for(k = 1; k < 8; k++)
1679 block[k << 3] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1682 if(dc_pred_dir) { //left
1683 for(k = 1; k < 8; k++)
1684 block[k] += ac_val[k];
1686 for(k = 1; k < 8; k++)
1687 block[k << 3] += ac_val[k + 8];
1691 /* save AC coeffs for further prediction */
1692 for(k = 1; k < 8; k++) {
1693 ac_val2[k] = block[k];
1694 ac_val2[k + 8] = block[k << 3];
1697 /* scale AC coeffs */
1698 for(k = 1; k < 64; k++)
1702 block[k] += (block[k] < 0) ? -mquant : mquant;
1705 if(use_pred) i = 63;
1706 } else { // no AC coeffs
1709 memset(ac_val2, 0, 16 * 2);
1710 if(dc_pred_dir) {//left
1712 memcpy(ac_val2, ac_val, 8 * 2);
1714 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1715 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1716 for(k = 1; k < 8; k++)
1717 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1722 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1724 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1725 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1726 for(k = 1; k < 8; k++)
1727 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1732 /* apply AC prediction if needed */
1734 if(dc_pred_dir) { //left
1735 for(k = 1; k < 8; k++) {
1736 block[k] = ac_val2[k] * scale;
1737 if(!v->pquantizer && block[k])
1738 block[k] += (block[k] < 0) ? -mquant : mquant;
1741 for(k = 1; k < 8; k++) {
1742 block[k << 3] = ac_val2[k + 8] * scale;
1743 if(!v->pquantizer && block[k << 3])
1744 block[k << 3] += (block[k << 3] < 0) ? -mquant : mquant;
1750 s->block_last_index[n] = i;
1755 /** Decode intra block in inter frames - more generic version than vc1_decode_i_block
1756 * @param v VC1Context
1757 * @param block block to decode
1758 * @param[in] n subblock index
1759 * @param coded are AC coeffs present or not
1760 * @param mquant block quantizer
1761 * @param codingset set of VLC to decode data
1763 static int vc1_decode_intra_block(VC1Context *v, DCTELEM block[64], int n, int coded, int mquant, int codingset)
1765 GetBitContext *gb = &v->s.gb;
1766 MpegEncContext *s = &v->s;
1767 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1770 int16_t *ac_val, *ac_val2;
1772 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1773 int a_avail = v->a_avail, c_avail = v->c_avail;
1774 int use_pred = s->ac_pred;
1778 s->dsp.clear_block(block);
1780 /* XXX: Guard against dumb values of mquant */
1781 mquant = (mquant < 1) ? 0 : ( (mquant>31) ? 31 : mquant );
1783 /* Set DC scale - y and c use the same */
1784 s->y_dc_scale = s->y_dc_scale_table[mquant];
1785 s->c_dc_scale = s->c_dc_scale_table[mquant];
1787 /* Get DC differential */
1789 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1791 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1794 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1799 if (dcdiff == 119 /* ESC index value */)
1801 /* TODO: Optimize */
1802 if (mquant == 1) dcdiff = get_bits(gb, 10);
1803 else if (mquant == 2) dcdiff = get_bits(gb, 9);
1804 else dcdiff = get_bits(gb, 8);
1809 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1810 else if (mquant == 2)
1811 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1818 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
1821 /* Store the quantized DC coeff, used for prediction */
1824 block[0] = dcdiff * s->y_dc_scale;
1826 block[0] = dcdiff * s->c_dc_scale;
1832 /* check if AC is needed at all and adjust direction if needed */
1833 if(!a_avail) dc_pred_dir = 1;
1834 if(!c_avail) dc_pred_dir = 0;
1835 if(!a_avail && !c_avail) use_pred = 0;
1836 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1839 scale = mquant * 2 + v->halfpq;
1841 if(dc_pred_dir) //left
1844 ac_val -= 16 * s->block_wrap[n];
1846 q1 = s->current_picture.qscale_table[mb_pos];
1847 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
1848 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1849 if(dc_pred_dir && n==1) q2 = q1;
1850 if(!dc_pred_dir && n==2) q2 = q1;
1854 int last = 0, skip, value;
1858 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1862 block[v->zz_8x8[0][i++]] = value;
1865 /* apply AC prediction if needed */
1867 /* scale predictors if needed*/
1869 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1870 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1872 if(dc_pred_dir) { //left
1873 for(k = 1; k < 8; k++)
1874 block[k] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1876 for(k = 1; k < 8; k++)
1877 block[k << 3] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1880 if(dc_pred_dir) { //left
1881 for(k = 1; k < 8; k++)
1882 block[k] += ac_val[k];
1884 for(k = 1; k < 8; k++)
1885 block[k << 3] += ac_val[k + 8];
1889 /* save AC coeffs for further prediction */
1890 for(k = 1; k < 8; k++) {
1891 ac_val2[k] = block[k];
1892 ac_val2[k + 8] = block[k << 3];
1895 /* scale AC coeffs */
1896 for(k = 1; k < 64; k++)
1900 block[k] += (block[k] < 0) ? -mquant : mquant;
1903 if(use_pred) i = 63;
1904 } else { // no AC coeffs
1907 memset(ac_val2, 0, 16 * 2);
1908 if(dc_pred_dir) {//left
1910 memcpy(ac_val2, ac_val, 8 * 2);
1912 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1913 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1914 for(k = 1; k < 8; k++)
1915 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1920 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1922 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1923 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1924 for(k = 1; k < 8; k++)
1925 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1930 /* apply AC prediction if needed */
1932 if(dc_pred_dir) { //left
1933 for(k = 1; k < 8; k++) {
1934 block[k] = ac_val2[k] * scale;
1935 if(!v->pquantizer && block[k])
1936 block[k] += (block[k] < 0) ? -mquant : mquant;
1939 for(k = 1; k < 8; k++) {
1940 block[k << 3] = ac_val2[k + 8] * scale;
1941 if(!v->pquantizer && block[k << 3])
1942 block[k << 3] += (block[k << 3] < 0) ? -mquant : mquant;
1948 s->block_last_index[n] = i;
1955 static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n, int mquant, int ttmb, int first_block,
1956 uint8_t *dst, int linesize, int skip_block, int apply_filter, int cbp_top, int cbp_left)
1958 MpegEncContext *s = &v->s;
1959 GetBitContext *gb = &s->gb;
1962 int scale, off, idx, last, skip, value;
1963 int ttblk = ttmb & 7;
1966 s->dsp.clear_block(block);
1969 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)];
1971 if(ttblk == TT_4X4) {
1972 subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
1974 if((ttblk != TT_8X8 && ttblk != TT_4X4)
1975 && ((v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))
1976 || (!v->res_rtm_flag && !first_block))) {
1977 subblkpat = decode012(gb);
1978 if(subblkpat) subblkpat ^= 3; //swap decoded pattern bits
1979 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) ttblk = TT_8X4;
1980 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) ttblk = TT_4X8;
1982 scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);
1984 // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
1985 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
1986 subblkpat = 2 - (ttblk == TT_8X4_TOP);
1989 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
1990 subblkpat = 2 - (ttblk == TT_4X8_LEFT);
1999 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2003 idx = v->zz_8x8[0][i++];
2004 block[idx] = value * scale;
2006 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2010 v->vc1dsp.vc1_inv_trans_8x8_dc(dst, linesize, block);
2012 v->vc1dsp.vc1_inv_trans_8x8_add(dst, linesize, block);
2014 if(apply_filter && cbp_top & 0xC)
2015 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
2016 if(apply_filter && cbp_left & 0xA)
2017 v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
2021 pat = ~subblkpat & 0xF;
2022 for(j = 0; j < 4; j++) {
2023 last = subblkpat & (1 << (3 - j));
2025 off = (j & 1) * 4 + (j & 2) * 16;
2027 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2031 idx = ff_vc1_simple_progressive_4x4_zz[i++];
2032 block[idx + off] = value * scale;
2034 block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
2036 if(!(subblkpat & (1 << (3 - j))) && !skip_block){
2038 v->vc1dsp.vc1_inv_trans_4x4_dc(dst + (j&1)*4 + (j&2)*2*linesize, linesize, block + off);
2040 v->vc1dsp.vc1_inv_trans_4x4(dst + (j&1)*4 + (j&2)*2*linesize, linesize, block + off);
2041 if(apply_filter && (j&2 ? pat & (1<<(j-2)) : (cbp_top & (1 << (j + 2)))))
2042 v->vc1dsp.vc1_v_loop_filter4(dst + (j&1)*4 + (j&2)*2*linesize, linesize, v->pq);
2043 if(apply_filter && (j&1 ? pat & (1<<(j-1)) : (cbp_left & (1 << (j + 1)))))
2044 v->vc1dsp.vc1_h_loop_filter4(dst + (j&1)*4 + (j&2)*2*linesize, linesize, v->pq);
2049 pat = ~((subblkpat & 2)*6 + (subblkpat & 1)*3) & 0xF;
2050 for(j = 0; j < 2; j++) {
2051 last = subblkpat & (1 << (1 - j));
2055 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2059 idx = v->zz_8x4[i++]+off;
2060 block[idx] = value * scale;
2062 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2064 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
2066 v->vc1dsp.vc1_inv_trans_8x4_dc(dst + j*4*linesize, linesize, block + off);
2068 v->vc1dsp.vc1_inv_trans_8x4(dst + j*4*linesize, linesize, block + off);
2069 if(apply_filter && j ? pat & 0x3 : (cbp_top & 0xC))
2070 v->vc1dsp.vc1_v_loop_filter8(dst + j*4*linesize, linesize, v->pq);
2071 if(apply_filter && cbp_left & (2 << j))
2072 v->vc1dsp.vc1_h_loop_filter4(dst + j*4*linesize, linesize, v->pq);
2077 pat = ~(subblkpat*5) & 0xF;
2078 for(j = 0; j < 2; j++) {
2079 last = subblkpat & (1 << (1 - j));
2083 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2087 idx = v->zz_4x8[i++]+off;
2088 block[idx] = value * scale;
2090 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2092 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
2094 v->vc1dsp.vc1_inv_trans_4x8_dc(dst + j*4, linesize, block + off);
2096 v->vc1dsp.vc1_inv_trans_4x8(dst + j*4, linesize, block + off);
2097 if(apply_filter && cbp_top & (2 << j))
2098 v->vc1dsp.vc1_v_loop_filter4(dst + j*4, linesize, v->pq);
2099 if(apply_filter && j ? pat & 0x5 : (cbp_left & 0xA))
2100 v->vc1dsp.vc1_h_loop_filter8(dst + j*4, linesize, v->pq);
2108 /** @} */ // Macroblock group
2110 static const int size_table [6] = { 0, 2, 3, 4, 5, 8 };
2111 static const int offset_table[6] = { 0, 1, 3, 7, 15, 31 };
2113 /** Decode one P-frame MB (in Simple/Main profile)
2115 static int vc1_decode_p_mb(VC1Context *v)
2117 MpegEncContext *s = &v->s;
2118 GetBitContext *gb = &s->gb;
2120 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2121 int cbp; /* cbp decoding stuff */
2122 int mqdiff, mquant; /* MB quantization */
2123 int ttmb = v->ttfrm; /* MB Transform type */
2125 int mb_has_coeffs = 1; /* last_flag */
2126 int dmv_x, dmv_y; /* Differential MV components */
2127 int index, index1; /* LUT indexes */
2128 int val, sign; /* temp values */
2129 int first_block = 1;
2131 int skipped, fourmv;
2132 int block_cbp = 0, pat;
2133 int apply_loop_filter;
2135 mquant = v->pq; /* Loosy initialization */
2137 if (v->mv_type_is_raw)
2138 fourmv = get_bits1(gb);
2140 fourmv = v->mv_type_mb_plane[mb_pos];
2142 skipped = get_bits1(gb);
2144 skipped = v->s.mbskip_table[mb_pos];
2146 apply_loop_filter = s->loop_filter && !(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY);
2147 if (!fourmv) /* 1MV mode */
2151 vc1_idct_func idct8x8_fn;
2153 GET_MVDATA(dmv_x, dmv_y);
2156 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2157 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2159 s->current_picture.mb_type[mb_pos] = s->mb_intra ? MB_TYPE_INTRA : MB_TYPE_16x16;
2160 vc1_pred_mv(s, 0, dmv_x, dmv_y, 1, v->range_x, v->range_y, v->mb_type[0]);
2162 /* FIXME Set DC val for inter block ? */
2163 if (s->mb_intra && !mb_has_coeffs)
2166 s->ac_pred = get_bits1(gb);
2169 else if (mb_has_coeffs)
2171 if (s->mb_intra) s->ac_pred = get_bits1(gb);
2172 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2180 s->current_picture.qscale_table[mb_pos] = mquant;
2182 if (!v->ttmbf && !s->mb_intra && mb_has_coeffs)
2183 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table,
2184 VC1_TTMB_VLC_BITS, 2);
2185 if(!s->mb_intra) vc1_mc_1mv(v, 0);
2187 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2190 s->dc_val[0][s->block_index[i]] = 0;
2192 val = ((cbp >> (5 - i)) & 1);
2193 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2194 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2196 /* check if prediction blocks A and C are available */
2197 v->a_avail = v->c_avail = 0;
2198 if(i == 2 || i == 3 || !s->first_slice_line)
2199 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2200 if(i == 1 || i == 3 || s->mb_x)
2201 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2203 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
2204 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2205 idct8x8_fn(s->dest[dst_idx] + off,
2206 i & 4 ? s->uvlinesize : s->linesize,
2208 if(v->pq >= 9 && v->overlap) {
2210 v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2212 v->vc1dsp.vc1_v_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2214 if(apply_loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
2215 int left_cbp, top_cbp;
2217 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
2218 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
2220 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
2221 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
2224 v->vc1dsp.vc1_v_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2226 v->vc1dsp.vc1_h_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2228 block_cbp |= 0xF << (i << 2);
2230 int left_cbp = 0, top_cbp = 0, filter = 0;
2231 if(apply_loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
2234 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
2235 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
2237 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
2238 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
2241 v->vc1dsp.vc1_v_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2243 v->vc1dsp.vc1_h_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2245 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), filter, left_cbp, top_cbp);
2246 block_cbp |= pat << (i << 2);
2247 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2255 for(i = 0; i < 6; i++) {
2256 v->mb_type[0][s->block_index[i]] = 0;
2257 s->dc_val[0][s->block_index[i]] = 0;
2259 s->current_picture.mb_type[mb_pos] = MB_TYPE_SKIP;
2260 s->current_picture.qscale_table[mb_pos] = 0;
2261 vc1_pred_mv(s, 0, 0, 0, 1, v->range_x, v->range_y, v->mb_type[0]);
2268 if (!skipped /* unskipped MB */)
2270 int intra_count = 0, coded_inter = 0;
2271 int is_intra[6], is_coded[6];
2272 vc1_idct_func idct8x8_fn;
2274 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2277 val = ((cbp >> (5 - i)) & 1);
2278 s->dc_val[0][s->block_index[i]] = 0;
2285 GET_MVDATA(dmv_x, dmv_y);
2287 vc1_pred_mv(s, i, dmv_x, dmv_y, 0, v->range_x, v->range_y, v->mb_type[0]);
2288 if(!s->mb_intra) vc1_mc_4mv_luma(v, i);
2289 intra_count += s->mb_intra;
2290 is_intra[i] = s->mb_intra;
2291 is_coded[i] = mb_has_coeffs;
2294 is_intra[i] = (intra_count >= 3);
2297 if(i == 4) vc1_mc_4mv_chroma(v);
2298 v->mb_type[0][s->block_index[i]] = is_intra[i];
2299 if(!coded_inter) coded_inter = !is_intra[i] & is_coded[i];
2301 // if there are no coded blocks then don't do anything more
2302 if(!intra_count && !coded_inter) return 0;
2305 s->current_picture.qscale_table[mb_pos] = mquant;
2306 /* test if block is intra and has pred */
2311 if(((!s->first_slice_line || (i==2 || i==3)) && v->mb_type[0][s->block_index[i] - s->block_wrap[i]])
2312 || ((s->mb_x || (i==1 || i==3)) && v->mb_type[0][s->block_index[i] - 1])) {
2317 if(intrapred)s->ac_pred = get_bits1(gb);
2318 else s->ac_pred = 0;
2320 if (!v->ttmbf && coded_inter)
2321 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2322 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2326 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2327 s->mb_intra = is_intra[i];
2329 /* check if prediction blocks A and C are available */
2330 v->a_avail = v->c_avail = 0;
2331 if(i == 2 || i == 3 || !s->first_slice_line)
2332 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2333 if(i == 1 || i == 3 || s->mb_x)
2334 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2336 vc1_decode_intra_block(v, s->block[i], i, is_coded[i], mquant, (i&4)?v->codingset2:v->codingset);
2337 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2338 idct8x8_fn(s->dest[dst_idx] + off,
2339 (i&4)?s->uvlinesize:s->linesize,
2341 if(v->pq >= 9 && v->overlap) {
2343 v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2345 v->vc1dsp.vc1_v_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2347 if(v->s.loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
2348 int left_cbp, top_cbp;
2350 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
2351 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
2353 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
2354 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
2357 v->vc1dsp.vc1_v_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2359 v->vc1dsp.vc1_h_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2361 block_cbp |= 0xF << (i << 2);
2362 } else if(is_coded[i]) {
2363 int left_cbp = 0, top_cbp = 0, filter = 0;
2364 if(v->s.loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
2367 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
2368 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
2370 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
2371 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
2374 v->vc1dsp.vc1_v_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2376 v->vc1dsp.vc1_h_loop_filter8(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, v->pq);
2378 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), filter, left_cbp, top_cbp);
2379 block_cbp |= pat << (i << 2);
2380 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2389 s->current_picture.qscale_table[mb_pos] = 0;
2390 for (i=0; i<6; i++) {
2391 v->mb_type[0][s->block_index[i]] = 0;
2392 s->dc_val[0][s->block_index[i]] = 0;
2396 vc1_pred_mv(s, i, 0, 0, 0, v->range_x, v->range_y, v->mb_type[0]);
2397 vc1_mc_4mv_luma(v, i);
2399 vc1_mc_4mv_chroma(v);
2400 s->current_picture.qscale_table[mb_pos] = 0;
2404 v->cbp[s->mb_x] = block_cbp;
2406 /* Should never happen */
2410 /** Decode one B-frame MB (in Main profile)
2412 static void vc1_decode_b_mb(VC1Context *v)
2414 MpegEncContext *s = &v->s;
2415 GetBitContext *gb = &s->gb;
2417 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2418 int cbp = 0; /* cbp decoding stuff */
2419 int mqdiff, mquant; /* MB quantization */
2420 int ttmb = v->ttfrm; /* MB Transform type */
2421 int mb_has_coeffs = 0; /* last_flag */
2422 int index, index1; /* LUT indexes */
2423 int val, sign; /* temp values */
2424 int first_block = 1;
2426 int skipped, direct;
2427 int dmv_x[2], dmv_y[2];
2428 int bmvtype = BMV_TYPE_BACKWARD;
2429 vc1_idct_func idct8x8_fn;
2431 mquant = v->pq; /* Loosy initialization */
2435 direct = get_bits1(gb);
2437 direct = v->direct_mb_plane[mb_pos];
2439 skipped = get_bits1(gb);
2441 skipped = v->s.mbskip_table[mb_pos];
2443 dmv_x[0] = dmv_x[1] = dmv_y[0] = dmv_y[1] = 0;
2444 for(i = 0; i < 6; i++) {
2445 v->mb_type[0][s->block_index[i]] = 0;
2446 s->dc_val[0][s->block_index[i]] = 0;
2448 s->current_picture.qscale_table[mb_pos] = 0;
2452 GET_MVDATA(dmv_x[0], dmv_y[0]);
2453 dmv_x[1] = dmv_x[0];
2454 dmv_y[1] = dmv_y[0];
2456 if(skipped || !s->mb_intra) {
2457 bmvtype = decode012(gb);
2460 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_BACKWARD : BMV_TYPE_FORWARD;
2463 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_FORWARD : BMV_TYPE_BACKWARD;
2466 bmvtype = BMV_TYPE_INTERPOLATED;
2467 dmv_x[0] = dmv_y[0] = 0;
2471 for(i = 0; i < 6; i++)
2472 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2475 if(direct) bmvtype = BMV_TYPE_INTERPOLATED;
2476 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2477 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2481 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2484 s->current_picture.qscale_table[mb_pos] = mquant;
2486 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2487 dmv_x[0] = dmv_y[0] = dmv_x[1] = dmv_y[1] = 0;
2488 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2489 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2491 if(!mb_has_coeffs && !s->mb_intra) {
2492 /* no coded blocks - effectively skipped */
2493 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2494 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2497 if(s->mb_intra && !mb_has_coeffs) {
2499 s->current_picture.qscale_table[mb_pos] = mquant;
2500 s->ac_pred = get_bits1(gb);
2502 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2504 if(bmvtype == BMV_TYPE_INTERPOLATED) {
2505 GET_MVDATA(dmv_x[0], dmv_y[0]);
2506 if(!mb_has_coeffs) {
2507 /* interpolated skipped block */
2508 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2509 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2513 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2515 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2518 s->ac_pred = get_bits1(gb);
2519 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2521 s->current_picture.qscale_table[mb_pos] = mquant;
2522 if(!v->ttmbf && !s->mb_intra && mb_has_coeffs)
2523 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2527 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2530 s->dc_val[0][s->block_index[i]] = 0;
2532 val = ((cbp >> (5 - i)) & 1);
2533 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2534 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2536 /* check if prediction blocks A and C are available */
2537 v->a_avail = v->c_avail = 0;
2538 if(i == 2 || i == 3 || !s->first_slice_line)
2539 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2540 if(i == 1 || i == 3 || s->mb_x)
2541 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2543 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
2544 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2545 idct8x8_fn(s->dest[dst_idx] + off,
2546 i & 4 ? s->uvlinesize : s->linesize,
2549 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), 0, 0, 0);
2550 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2556 /** Decode blocks of I-frame
2558 static void vc1_decode_i_blocks(VC1Context *v)
2561 MpegEncContext *s = &v->s;
2565 vc1_idct_func idct8x8_fn;
2567 /* select codingmode used for VLC tables selection */
2568 switch(v->y_ac_table_index){
2570 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2573 v->codingset = CS_HIGH_MOT_INTRA;
2576 v->codingset = CS_MID_RATE_INTRA;
2580 switch(v->c_ac_table_index){
2582 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2585 v->codingset2 = CS_HIGH_MOT_INTER;
2588 v->codingset2 = CS_MID_RATE_INTER;
2592 /* Set DC scale - y and c use the same */
2593 s->y_dc_scale = s->y_dc_scale_table[v->pq];
2594 s->c_dc_scale = s->c_dc_scale_table[v->pq];
2597 s->mb_x = s->mb_y = 0;
2599 s->first_slice_line = 1;
2600 if(v->pq >= 9 && v->overlap) {
2601 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2603 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put[!!v->rangeredfrm];
2604 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
2606 ff_init_block_index(s);
2607 for(; s->mb_x < s->mb_width; s->mb_x++) {
2609 ff_update_block_index(s);
2610 dst[0] = s->dest[0];
2611 dst[1] = dst[0] + 8;
2612 dst[2] = s->dest[0] + s->linesize * 8;
2613 dst[3] = dst[2] + 8;
2614 dst[4] = s->dest[1];
2615 dst[5] = s->dest[2];
2616 s->dsp.clear_blocks(s->block[0]);
2617 mb_pos = s->mb_x + s->mb_y * s->mb_width;
2618 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
2619 s->current_picture.qscale_table[mb_pos] = v->pq;
2620 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2621 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2623 // do actual MB decoding and displaying
2624 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
2625 v->s.ac_pred = get_bits1(&v->s.gb);
2627 for(k = 0; k < 6; k++) {
2628 val = ((cbp >> (5 - k)) & 1);
2631 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
2635 cbp |= val << (5 - k);
2637 vc1_decode_i_block(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2);
2639 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
2641 k & 4 ? s->uvlinesize : s->linesize,
2645 if(v->pq >= 9 && v->overlap) {
2647 v->vc1dsp.vc1_h_overlap(s->dest[0], s->linesize);
2648 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2649 if(!(s->flags & CODEC_FLAG_GRAY)) {
2650 v->vc1dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
2651 v->vc1dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
2654 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
2655 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2656 if(!s->first_slice_line) {
2657 v->vc1dsp.vc1_v_overlap(s->dest[0], s->linesize);
2658 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
2659 if(!(s->flags & CODEC_FLAG_GRAY)) {
2660 v->vc1dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
2661 v->vc1dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
2664 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2665 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2667 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
2669 if(get_bits_count(&s->gb) > v->bits) {
2670 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2671 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
2675 if (!v->s.loop_filter)
2676 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2678 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
2680 s->first_slice_line = 0;
2682 if (v->s.loop_filter)
2683 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
2684 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
2687 /** Decode blocks of I-frame for advanced profile
2689 static void vc1_decode_i_blocks_adv(VC1Context *v)
2692 MpegEncContext *s = &v->s;
2699 GetBitContext *gb = &s->gb;
2700 vc1_idct_func idct8x8_fn;
2702 /* select codingmode used for VLC tables selection */
2703 switch(v->y_ac_table_index){
2705 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2708 v->codingset = CS_HIGH_MOT_INTRA;
2711 v->codingset = CS_MID_RATE_INTRA;
2715 switch(v->c_ac_table_index){
2717 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2720 v->codingset2 = CS_HIGH_MOT_INTER;
2723 v->codingset2 = CS_MID_RATE_INTER;
2728 s->mb_x = s->mb_y = 0;
2730 s->first_slice_line = 1;
2731 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[0];
2732 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
2734 ff_init_block_index(s);
2735 for(;s->mb_x < s->mb_width; s->mb_x++) {
2737 ff_update_block_index(s);
2738 dst[0] = s->dest[0];
2739 dst[1] = dst[0] + 8;
2740 dst[2] = s->dest[0] + s->linesize * 8;
2741 dst[3] = dst[2] + 8;
2742 dst[4] = s->dest[1];
2743 dst[5] = s->dest[2];
2744 s->dsp.clear_blocks(s->block[0]);
2745 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2746 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
2747 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2748 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2750 // do actual MB decoding and displaying
2751 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
2752 if(v->acpred_is_raw)
2753 v->s.ac_pred = get_bits1(&v->s.gb);
2755 v->s.ac_pred = v->acpred_plane[mb_pos];
2757 if(v->condover == CONDOVER_SELECT) {
2758 if(v->overflg_is_raw)
2759 overlap = get_bits1(&v->s.gb);
2761 overlap = v->over_flags_plane[mb_pos];
2763 overlap = (v->condover == CONDOVER_ALL);
2767 s->current_picture.qscale_table[mb_pos] = mquant;
2768 /* Set DC scale - y and c use the same */
2769 s->y_dc_scale = s->y_dc_scale_table[mquant];
2770 s->c_dc_scale = s->c_dc_scale_table[mquant];
2772 for(k = 0; k < 6; k++) {
2773 val = ((cbp >> (5 - k)) & 1);
2776 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
2780 cbp |= val << (5 - k);
2782 v->a_avail = !s->first_slice_line || (k==2 || k==3);
2783 v->c_avail = !!s->mb_x || (k==1 || k==3);
2785 vc1_decode_i_block_adv(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant);
2787 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
2789 k & 4 ? s->uvlinesize : s->linesize,
2795 v->vc1dsp.vc1_h_overlap(s->dest[0], s->linesize);
2796 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2797 if(!(s->flags & CODEC_FLAG_GRAY)) {
2798 v->vc1dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
2799 v->vc1dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
2802 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
2803 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2804 if(!s->first_slice_line) {
2805 v->vc1dsp.vc1_v_overlap(s->dest[0], s->linesize);
2806 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
2807 if(!(s->flags & CODEC_FLAG_GRAY)) {
2808 v->vc1dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
2809 v->vc1dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
2812 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2813 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2815 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
2817 if(get_bits_count(&s->gb) > v->bits) {
2818 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2819 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
2823 if (!v->s.loop_filter)
2824 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2826 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
2827 s->first_slice_line = 0;
2829 if (v->s.loop_filter)
2830 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
2831 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
2834 static void vc1_decode_p_blocks(VC1Context *v)
2836 MpegEncContext *s = &v->s;
2838 /* select codingmode used for VLC tables selection */
2839 switch(v->c_ac_table_index){
2841 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2844 v->codingset = CS_HIGH_MOT_INTRA;
2847 v->codingset = CS_MID_RATE_INTRA;
2851 switch(v->c_ac_table_index){
2853 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2856 v->codingset2 = CS_HIGH_MOT_INTER;
2859 v->codingset2 = CS_MID_RATE_INTER;
2863 s->first_slice_line = 1;
2864 memset(v->cbp_base, 0, sizeof(v->cbp_base[0])*2*s->mb_stride);
2865 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
2867 ff_init_block_index(s);
2868 for(; s->mb_x < s->mb_width; s->mb_x++) {
2869 ff_update_block_index(s);
2872 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
2873 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2874 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);
2878 memmove(v->cbp_base, v->cbp, sizeof(v->cbp_base[0])*s->mb_stride);
2879 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2880 s->first_slice_line = 0;
2882 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
2885 static void vc1_decode_b_blocks(VC1Context *v)
2887 MpegEncContext *s = &v->s;
2889 /* select codingmode used for VLC tables selection */
2890 switch(v->c_ac_table_index){
2892 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2895 v->codingset = CS_HIGH_MOT_INTRA;
2898 v->codingset = CS_MID_RATE_INTRA;
2902 switch(v->c_ac_table_index){
2904 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2907 v->codingset2 = CS_HIGH_MOT_INTER;
2910 v->codingset2 = CS_MID_RATE_INTER;
2914 s->first_slice_line = 1;
2915 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
2917 ff_init_block_index(s);
2918 for(; s->mb_x < s->mb_width; s->mb_x++) {
2919 ff_update_block_index(s);
2922 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
2923 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2924 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);
2927 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
2929 if (!v->s.loop_filter)
2930 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2932 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
2933 s->first_slice_line = 0;
2935 if (v->s.loop_filter)
2936 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
2937 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
2940 static void vc1_decode_skip_blocks(VC1Context *v)
2942 MpegEncContext *s = &v->s;
2944 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
2945 s->first_slice_line = 1;
2946 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
2948 ff_init_block_index(s);
2949 ff_update_block_index(s);
2950 memcpy(s->dest[0], s->last_picture.data[0] + s->mb_y * 16 * s->linesize, s->linesize * 16);
2951 memcpy(s->dest[1], s->last_picture.data[1] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
2952 memcpy(s->dest[2], s->last_picture.data[2] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
2953 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2954 s->first_slice_line = 0;
2956 s->pict_type = FF_P_TYPE;
2959 static void vc1_decode_blocks(VC1Context *v)
2962 v->s.esc3_level_length = 0;
2964 ff_intrax8_decode_picture(&v->x8, 2*v->pq+v->halfpq, v->pq*(!v->pquantizer) );
2967 switch(v->s.pict_type) {
2969 if(v->profile == PROFILE_ADVANCED)
2970 vc1_decode_i_blocks_adv(v);
2972 vc1_decode_i_blocks(v);
2975 if(v->p_frame_skipped)
2976 vc1_decode_skip_blocks(v);
2978 vc1_decode_p_blocks(v);
2982 if(v->profile == PROFILE_ADVANCED)
2983 vc1_decode_i_blocks_adv(v);
2985 vc1_decode_i_blocks(v);
2987 vc1_decode_b_blocks(v);
2993 /** Initialize a VC1/WMV3 decoder
2994 * @todo TODO: Handle VC-1 IDUs (Transport level?)
2995 * @todo TODO: Decypher remaining bits in extra_data
2997 static av_cold int vc1_decode_init(AVCodecContext *avctx)
2999 VC1Context *v = avctx->priv_data;
3000 MpegEncContext *s = &v->s;
3004 if (!avctx->extradata_size || !avctx->extradata) return -1;
3005 if (!(avctx->flags & CODEC_FLAG_GRAY))
3006 avctx->pix_fmt = avctx->get_format(avctx, avctx->codec->pix_fmts);
3008 avctx->pix_fmt = PIX_FMT_GRAY8;
3009 avctx->hwaccel = ff_find_hwaccel(avctx->codec->id, avctx->pix_fmt);
3011 avctx->flags |= CODEC_FLAG_EMU_EDGE;
3012 v->s.flags |= CODEC_FLAG_EMU_EDGE;
3014 if(avctx->idct_algo==FF_IDCT_AUTO){
3015 avctx->idct_algo=FF_IDCT_WMV2;
3018 if(ff_msmpeg4_decode_init(avctx) < 0)
3020 if (vc1_init_common(v) < 0) return -1;
3021 ff_vc1dsp_init(&v->vc1dsp);
3022 for (i = 0; i < 64; i++) {
3023 #define transpose(x) ((x>>3) | ((x&7)<<3))
3024 v->zz_8x8[0][i] = transpose(wmv1_scantable[0][i]);
3025 v->zz_8x8[1][i] = transpose(wmv1_scantable[1][i]);
3026 v->zz_8x8[2][i] = transpose(wmv1_scantable[2][i]);
3027 v->zz_8x8[3][i] = transpose(wmv1_scantable[3][i]);
3030 avctx->coded_width = avctx->width;
3031 avctx->coded_height = avctx->height;
3032 if (avctx->codec_id == CODEC_ID_WMV3)
3036 // looks like WMV3 has a sequence header stored in the extradata
3037 // advanced sequence header may be before the first frame
3038 // the last byte of the extradata is a version number, 1 for the
3039 // samples we can decode
3041 init_get_bits(&gb, avctx->extradata, avctx->extradata_size*8);
3043 if (vc1_decode_sequence_header(avctx, v, &gb) < 0)
3046 count = avctx->extradata_size*8 - get_bits_count(&gb);
3049 av_log(avctx, AV_LOG_INFO, "Extra data: %i bits left, value: %X\n",
3050 count, get_bits(&gb, count));
3054 av_log(avctx, AV_LOG_INFO, "Read %i bits in overflow\n", -count);
3056 } else { // VC1/WVC1
3057 const uint8_t *start = avctx->extradata;
3058 uint8_t *end = avctx->extradata + avctx->extradata_size;
3059 const uint8_t *next;
3060 int size, buf2_size;
3061 uint8_t *buf2 = NULL;
3062 int seq_initialized = 0, ep_initialized = 0;
3064 if(avctx->extradata_size < 16) {
3065 av_log(avctx, AV_LOG_ERROR, "Extradata size too small: %i\n", avctx->extradata_size);
3069 buf2 = av_mallocz(avctx->extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
3070 start = find_next_marker(start, end); // in WVC1 extradata first byte is its size, but can be 0 in mkv
3072 for(; next < end; start = next){
3073 next = find_next_marker(start + 4, end);
3074 size = next - start - 4;
3075 if(size <= 0) continue;
3076 buf2_size = vc1_unescape_buffer(start + 4, size, buf2);
3077 init_get_bits(&gb, buf2, buf2_size * 8);
3078 switch(AV_RB32(start)){
3079 case VC1_CODE_SEQHDR:
3080 if(vc1_decode_sequence_header(avctx, v, &gb) < 0){
3084 seq_initialized = 1;
3086 case VC1_CODE_ENTRYPOINT:
3087 if(vc1_decode_entry_point(avctx, v, &gb) < 0){
3096 if(!seq_initialized || !ep_initialized){
3097 av_log(avctx, AV_LOG_ERROR, "Incomplete extradata\n");
3101 avctx->has_b_frames= !!(avctx->max_b_frames);
3102 s->low_delay = !avctx->has_b_frames;
3104 s->mb_width = (avctx->coded_width+15)>>4;
3105 s->mb_height = (avctx->coded_height+15)>>4;
3107 /* Allocate mb bitplanes */
3108 v->mv_type_mb_plane = av_malloc(s->mb_stride * s->mb_height);
3109 v->direct_mb_plane = av_malloc(s->mb_stride * s->mb_height);
3110 v->acpred_plane = av_malloc(s->mb_stride * s->mb_height);
3111 v->over_flags_plane = av_malloc(s->mb_stride * s->mb_height);
3113 v->cbp_base = av_malloc(sizeof(v->cbp_base[0]) * 2 * s->mb_stride);
3114 v->cbp = v->cbp_base + s->mb_stride;
3116 /* allocate block type info in that way so it could be used with s->block_index[] */
3117 v->mb_type_base = av_malloc(s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride * (s->mb_height + 1) * 2);
3118 v->mb_type[0] = v->mb_type_base + s->b8_stride + 1;
3119 v->mb_type[1] = v->mb_type_base + s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride + 1;
3120 v->mb_type[2] = v->mb_type[1] + s->mb_stride * (s->mb_height + 1);
3122 /* Init coded blocks info */
3123 if (v->profile == PROFILE_ADVANCED)
3125 // if (alloc_bitplane(&v->over_flags_plane, s->mb_width, s->mb_height) < 0)
3127 // if (alloc_bitplane(&v->ac_pred_plane, s->mb_width, s->mb_height) < 0)
3131 ff_intrax8_common_init(&v->x8,s);
3136 /** Decode a VC1/WMV3 frame
3137 * @todo TODO: Handle VC-1 IDUs (Transport level?)
3139 static int vc1_decode_frame(AVCodecContext *avctx,
3140 void *data, int *data_size,
3143 const uint8_t *buf = avpkt->data;
3144 int buf_size = avpkt->size;
3145 VC1Context *v = avctx->priv_data;
3146 MpegEncContext *s = &v->s;
3147 AVFrame *pict = data;
3148 uint8_t *buf2 = NULL;
3149 const uint8_t *buf_start = buf;
3151 /* no supplementary picture */
3152 if (buf_size == 0) {
3153 /* special case for last picture */
3154 if (s->low_delay==0 && s->next_picture_ptr) {
3155 *pict= *(AVFrame*)s->next_picture_ptr;
3156 s->next_picture_ptr= NULL;
3158 *data_size = sizeof(AVFrame);
3164 /* We need to set current_picture_ptr before reading the header,
3165 * otherwise we cannot store anything in there. */
3166 if(s->current_picture_ptr==NULL || s->current_picture_ptr->data[0]){
3167 int i= ff_find_unused_picture(s, 0);
3168 s->current_picture_ptr= &s->picture[i];
3171 if (s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3172 if (v->profile < PROFILE_ADVANCED)
3173 avctx->pix_fmt = PIX_FMT_VDPAU_WMV3;
3175 avctx->pix_fmt = PIX_FMT_VDPAU_VC1;
3178 //for advanced profile we may need to parse and unescape data
3179 if (avctx->codec_id == CODEC_ID_VC1) {
3181 buf2 = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
3183 if(IS_MARKER(AV_RB32(buf))){ /* frame starts with marker and needs to be parsed */
3184 const uint8_t *start, *end, *next;
3188 for(start = buf, end = buf + buf_size; next < end; start = next){
3189 next = find_next_marker(start + 4, end);
3190 size = next - start - 4;
3191 if(size <= 0) continue;
3192 switch(AV_RB32(start)){
3193 case VC1_CODE_FRAME:
3194 if (avctx->hwaccel ||
3195 s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3197 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
3199 case VC1_CODE_ENTRYPOINT: /* it should be before frame data */
3200 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
3201 init_get_bits(&s->gb, buf2, buf_size2*8);
3202 vc1_decode_entry_point(avctx, v, &s->gb);
3204 case VC1_CODE_SLICE:
3205 av_log(avctx, AV_LOG_ERROR, "Sliced decoding is not implemented (yet)\n");
3210 }else if(v->interlace && ((buf[0] & 0xC0) == 0xC0)){ /* WVC1 interlaced stores both fields divided by marker */
3211 const uint8_t *divider;
3213 divider = find_next_marker(buf, buf + buf_size);
3214 if((divider == (buf + buf_size)) || AV_RB32(divider) != VC1_CODE_FIELD){
3215 av_log(avctx, AV_LOG_ERROR, "Error in WVC1 interlaced frame\n");
3220 buf_size2 = vc1_unescape_buffer(buf, divider - buf, buf2);
3222 if(!v->warn_interlaced++)
3223 av_log(v->s.avctx, AV_LOG_ERROR, "Interlaced WVC1 support is not implemented\n");
3224 av_free(buf2);return -1;
3226 buf_size2 = vc1_unescape_buffer(buf, buf_size, buf2);
3228 init_get_bits(&s->gb, buf2, buf_size2*8);
3230 init_get_bits(&s->gb, buf, buf_size*8);
3231 // do parse frame header
3232 if(v->profile < PROFILE_ADVANCED) {
3233 if(vc1_parse_frame_header(v, &s->gb) == -1) {
3238 if(vc1_parse_frame_header_adv(v, &s->gb) == -1) {
3244 if(v->res_sprite && (s->pict_type!=FF_I_TYPE)){
3250 s->current_picture.pict_type= s->pict_type;
3251 s->current_picture.key_frame= s->pict_type == FF_I_TYPE;
3253 /* skip B-frames if we don't have reference frames */
3254 if(s->last_picture_ptr==NULL && (s->pict_type==FF_B_TYPE || s->dropable)){
3256 return -1;//buf_size;
3258 /* skip b frames if we are in a hurry */
3259 if(avctx->hurry_up && s->pict_type==FF_B_TYPE) return -1;//buf_size;
3260 if( (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type==FF_B_TYPE)
3261 || (avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type!=FF_I_TYPE)
3262 || avctx->skip_frame >= AVDISCARD_ALL) {
3266 /* skip everything if we are in a hurry>=5 */
3267 if(avctx->hurry_up>=5) {
3269 return -1;//buf_size;
3272 if(s->next_p_frame_damaged){
3273 if(s->pict_type==FF_B_TYPE)
3276 s->next_p_frame_damaged=0;
3279 if(MPV_frame_start(s, avctx) < 0) {
3284 s->me.qpel_put= s->dsp.put_qpel_pixels_tab;
3285 s->me.qpel_avg= s->dsp.avg_qpel_pixels_tab;
3287 if ((CONFIG_VC1_VDPAU_DECODER)
3288 &&s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3289 ff_vdpau_vc1_decode_picture(s, buf_start, (buf + buf_size) - buf_start);
3290 else if (avctx->hwaccel) {
3291 if (avctx->hwaccel->start_frame(avctx, buf, buf_size) < 0)
3293 if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
3295 if (avctx->hwaccel->end_frame(avctx) < 0)
3298 ff_er_frame_start(s);
3300 v->bits = buf_size * 8;
3301 vc1_decode_blocks(v);
3302 //av_log(s->avctx, AV_LOG_INFO, "Consumed %i/%i bits\n", get_bits_count(&s->gb), buf_size*8);
3303 // if(get_bits_count(&s->gb) > buf_size * 8)
3310 assert(s->current_picture.pict_type == s->current_picture_ptr->pict_type);
3311 assert(s->current_picture.pict_type == s->pict_type);
3312 if (s->pict_type == FF_B_TYPE || s->low_delay) {
3313 *pict= *(AVFrame*)s->current_picture_ptr;
3314 } else if (s->last_picture_ptr != NULL) {
3315 *pict= *(AVFrame*)s->last_picture_ptr;
3318 if(s->last_picture_ptr || s->low_delay){
3319 *data_size = sizeof(AVFrame);
3320 ff_print_debug_info(s, pict);
3328 /** Close a VC1/WMV3 decoder
3329 * @warning Initial try at using MpegEncContext stuff
3331 static av_cold int vc1_decode_end(AVCodecContext *avctx)
3333 VC1Context *v = avctx->priv_data;
3335 av_freep(&v->hrd_rate);
3336 av_freep(&v->hrd_buffer);
3337 MPV_common_end(&v->s);
3338 av_freep(&v->mv_type_mb_plane);
3339 av_freep(&v->direct_mb_plane);
3340 av_freep(&v->acpred_plane);
3341 av_freep(&v->over_flags_plane);
3342 av_freep(&v->mb_type_base);
3343 av_freep(&v->cbp_base);
3344 ff_intrax8_common_end(&v->x8);
3349 AVCodec ff_vc1_decoder = {
3358 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3360 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1"),
3361 .pix_fmts = ff_hwaccel_pixfmt_list_420
3364 #if CONFIG_WMV3_DECODER
3365 AVCodec ff_wmv3_decoder = {
3374 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3376 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9"),
3377 .pix_fmts = ff_hwaccel_pixfmt_list_420
3381 #if CONFIG_WMV3_VDPAU_DECODER
3382 AVCodec ff_wmv3_vdpau_decoder = {
3391 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3393 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 VDPAU"),
3394 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_WMV3, PIX_FMT_NONE}
3398 #if CONFIG_VC1_VDPAU_DECODER
3399 AVCodec ff_vc1_vdpau_decoder = {
3408 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3410 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1 VDPAU"),
3411 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_VC1, PIX_FMT_NONE}