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;
212 v->luma_mv[s->mb_x][0] = uvmx;
213 v->luma_mv[s->mb_x][1] = uvmy;
215 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
216 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
219 srcY = s->last_picture.data[0];
220 srcU = s->last_picture.data[1];
221 srcV = s->last_picture.data[2];
223 srcY = s->next_picture.data[0];
224 srcU = s->next_picture.data[1];
225 srcV = s->next_picture.data[2];
228 src_x = s->mb_x * 16 + (mx >> 2);
229 src_y = s->mb_y * 16 + (my >> 2);
230 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
231 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
233 if(v->profile != PROFILE_ADVANCED){
234 src_x = av_clip( src_x, -16, s->mb_width * 16);
235 src_y = av_clip( src_y, -16, s->mb_height * 16);
236 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
237 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
239 src_x = av_clip( src_x, -17, s->avctx->coded_width);
240 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
241 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
242 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
245 srcY += src_y * s->linesize + src_x;
246 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
247 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
249 /* for grayscale we should not try to read from unknown area */
250 if(s->flags & CODEC_FLAG_GRAY) {
251 srcU = s->edge_emu_buffer + 18 * s->linesize;
252 srcV = s->edge_emu_buffer + 18 * s->linesize;
255 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
256 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
257 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
258 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
260 srcY -= s->mspel * (1 + s->linesize);
261 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
262 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
263 srcY = s->edge_emu_buffer;
264 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
265 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
266 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
267 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
270 /* if we deal with range reduction we need to scale source blocks */
276 for(j = 0; j < 17 + s->mspel*2; j++) {
277 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
280 src = srcU; src2 = srcV;
281 for(j = 0; j < 9; j++) {
282 for(i = 0; i < 9; i++) {
283 src[i] = ((src[i] - 128) >> 1) + 128;
284 src2[i] = ((src2[i] - 128) >> 1) + 128;
286 src += s->uvlinesize;
287 src2 += s->uvlinesize;
290 /* if we deal with intensity compensation we need to scale source blocks */
291 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
296 for(j = 0; j < 17 + s->mspel*2; j++) {
297 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = v->luty[src[i]];
300 src = srcU; src2 = srcV;
301 for(j = 0; j < 9; j++) {
302 for(i = 0; i < 9; i++) {
303 src[i] = v->lutuv[src[i]];
304 src2[i] = v->lutuv[src2[i]];
306 src += s->uvlinesize;
307 src2 += s->uvlinesize;
310 srcY += s->mspel * (1 + s->linesize);
314 dxy = ((my & 3) << 2) | (mx & 3);
315 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
316 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
317 srcY += s->linesize * 8;
318 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
319 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
320 } else { // hpel mc - always used for luma
321 dxy = (my & 2) | ((mx & 2) >> 1);
324 dsp->put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
326 dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
329 if(s->flags & CODEC_FLAG_GRAY) return;
330 /* Chroma MC always uses qpel bilinear */
334 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
335 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
337 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
338 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
342 /** Do motion compensation for 4-MV macroblock - luminance block
344 static void vc1_mc_4mv_luma(VC1Context *v, int n)
346 MpegEncContext *s = &v->s;
347 DSPContext *dsp = &v->s.dsp;
349 int dxy, mx, my, src_x, src_y;
352 if(!v->s.last_picture.data[0])return;
355 srcY = s->last_picture.data[0];
357 off = s->linesize * 4 * (n&2) + (n&1) * 8;
359 src_x = s->mb_x * 16 + (n&1) * 8 + (mx >> 2);
360 src_y = s->mb_y * 16 + (n&2) * 4 + (my >> 2);
362 if(v->profile != PROFILE_ADVANCED){
363 src_x = av_clip( src_x, -16, s->mb_width * 16);
364 src_y = av_clip( src_y, -16, s->mb_height * 16);
366 src_x = av_clip( src_x, -17, s->avctx->coded_width);
367 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
370 srcY += src_y * s->linesize + src_x;
372 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
373 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 8 - s->mspel*2
374 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 8 - s->mspel*2){
375 srcY -= s->mspel * (1 + s->linesize);
376 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 9+s->mspel*2, 9+s->mspel*2,
377 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
378 srcY = s->edge_emu_buffer;
379 /* if we deal with range reduction we need to scale source blocks */
385 for(j = 0; j < 9 + s->mspel*2; j++) {
386 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
390 /* if we deal with intensity compensation we need to scale source blocks */
391 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
396 for(j = 0; j < 9 + s->mspel*2; j++) {
397 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = v->luty[src[i]];
401 srcY += s->mspel * (1 + s->linesize);
405 dxy = ((my & 3) << 2) | (mx & 3);
406 v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, v->rnd);
407 } else { // hpel mc - always used for luma
408 dxy = (my & 2) | ((mx & 2) >> 1);
410 dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
412 dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
416 static inline int median4(int a, int b, int c, int d)
419 if(c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
420 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
422 if(c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
423 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
428 /** Do motion compensation for 4-MV macroblock - both chroma blocks
430 static void vc1_mc_4mv_chroma(VC1Context *v)
432 MpegEncContext *s = &v->s;
433 DSPContext *dsp = &v->s.dsp;
434 uint8_t *srcU, *srcV;
435 int uvmx, uvmy, uvsrc_x, uvsrc_y;
436 int i, idx, tx = 0, ty = 0;
437 int mvx[4], mvy[4], intra[4];
438 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
440 if(!v->s.last_picture.data[0])return;
441 if(s->flags & CODEC_FLAG_GRAY) return;
443 for(i = 0; i < 4; i++) {
444 mvx[i] = s->mv[0][i][0];
445 mvy[i] = s->mv[0][i][1];
446 intra[i] = v->mb_type[0][s->block_index[i]];
449 /* calculate chroma MV vector from four luma MVs */
450 idx = (intra[3] << 3) | (intra[2] << 2) | (intra[1] << 1) | intra[0];
451 if(!idx) { // all blocks are inter
452 tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
453 ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
454 } else if(count[idx] == 1) { // 3 inter blocks
457 tx = mid_pred(mvx[1], mvx[2], mvx[3]);
458 ty = mid_pred(mvy[1], mvy[2], mvy[3]);
461 tx = mid_pred(mvx[0], mvx[2], mvx[3]);
462 ty = mid_pred(mvy[0], mvy[2], mvy[3]);
465 tx = mid_pred(mvx[0], mvx[1], mvx[3]);
466 ty = mid_pred(mvy[0], mvy[1], mvy[3]);
469 tx = mid_pred(mvx[0], mvx[1], mvx[2]);
470 ty = mid_pred(mvy[0], mvy[1], mvy[2]);
473 } else if(count[idx] == 2) {
475 for(i=0; i<3;i++) if(!intra[i]) {t1 = i; break;}
476 for(i= t1+1; i<4; i++)if(!intra[i]) {t2 = i; break;}
477 tx = (mvx[t1] + mvx[t2]) / 2;
478 ty = (mvy[t1] + mvy[t2]) / 2;
480 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
481 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
482 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
483 return; //no need to do MC for inter blocks
486 s->current_picture.motion_val[1][s->block_index[0]][0] = tx;
487 s->current_picture.motion_val[1][s->block_index[0]][1] = ty;
488 uvmx = (tx + ((tx&3) == 3)) >> 1;
489 uvmy = (ty + ((ty&3) == 3)) >> 1;
490 v->luma_mv[s->mb_x][0] = uvmx;
491 v->luma_mv[s->mb_x][1] = uvmy;
493 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
494 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
497 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
498 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
500 if(v->profile != PROFILE_ADVANCED){
501 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
502 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
504 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
505 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
508 srcU = s->last_picture.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
509 srcV = s->last_picture.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
510 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
511 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
512 || (unsigned)uvsrc_y > (s->v_edge_pos >> 1) - 9){
513 s->dsp.emulated_edge_mc(s->edge_emu_buffer , srcU, s->uvlinesize, 8+1, 8+1,
514 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
515 s->dsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize, 8+1, 8+1,
516 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
517 srcU = s->edge_emu_buffer;
518 srcV = s->edge_emu_buffer + 16;
520 /* if we deal with range reduction we need to scale source blocks */
525 src = srcU; src2 = srcV;
526 for(j = 0; j < 9; j++) {
527 for(i = 0; i < 9; i++) {
528 src[i] = ((src[i] - 128) >> 1) + 128;
529 src2[i] = ((src2[i] - 128) >> 1) + 128;
531 src += s->uvlinesize;
532 src2 += s->uvlinesize;
535 /* if we deal with intensity compensation we need to scale source blocks */
536 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
540 src = srcU; src2 = srcV;
541 for(j = 0; j < 9; j++) {
542 for(i = 0; i < 9; i++) {
543 src[i] = v->lutuv[src[i]];
544 src2[i] = v->lutuv[src2[i]];
546 src += s->uvlinesize;
547 src2 += s->uvlinesize;
552 /* Chroma MC always uses qpel bilinear */
556 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
557 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
559 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
560 v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
564 /***********************************************************************/
566 * @defgroup vc1block VC-1 Block-level functions
567 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
573 * @brief Get macroblock-level quantizer scale
575 #define GET_MQUANT() \
579 if (v->dqprofile == DQPROFILE_ALL_MBS) \
583 mquant = (get_bits1(gb)) ? v->altpq : v->pq; \
587 mqdiff = get_bits(gb, 3); \
588 if (mqdiff != 7) mquant = v->pq + mqdiff; \
589 else mquant = get_bits(gb, 5); \
592 if(v->dqprofile == DQPROFILE_SINGLE_EDGE) \
593 edges = 1 << v->dqsbedge; \
594 else if(v->dqprofile == DQPROFILE_DOUBLE_EDGES) \
595 edges = (3 << v->dqsbedge) % 15; \
596 else if(v->dqprofile == DQPROFILE_FOUR_EDGES) \
598 if((edges&1) && !s->mb_x) \
600 if((edges&2) && s->first_slice_line) \
602 if((edges&4) && s->mb_x == (s->mb_width - 1)) \
604 if((edges&8) && s->mb_y == (s->mb_height - 1)) \
609 * @def GET_MVDATA(_dmv_x, _dmv_y)
610 * @brief Get MV differentials
611 * @see MVDATA decoding from 8.3.5.2, p(1)20
612 * @param _dmv_x Horizontal differential for decoded MV
613 * @param _dmv_y Vertical differential for decoded MV
615 #define GET_MVDATA(_dmv_x, _dmv_y) \
616 index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table,\
617 VC1_MV_DIFF_VLC_BITS, 2); \
623 else mb_has_coeffs = 0; \
625 if (!index) { _dmv_x = _dmv_y = 0; } \
626 else if (index == 35) \
628 _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample); \
629 _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample); \
631 else if (index == 36) \
640 if (!s->quarter_sample && index1 == 5) val = 1; \
642 if(size_table[index1] - val > 0) \
643 val = get_bits(gb, size_table[index1] - val); \
645 sign = 0 - (val&1); \
646 _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
649 if (!s->quarter_sample && index1 == 5) val = 1; \
651 if(size_table[index1] - val > 0) \
652 val = get_bits(gb, size_table[index1] - val); \
654 sign = 0 - (val&1); \
655 _dmv_y = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
658 /** Predict and set motion vector
660 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)
662 MpegEncContext *s = &v->s;
663 int xy, wrap, off = 0;
668 /* scale MV difference to be quad-pel */
669 dmv_x <<= 1 - s->quarter_sample;
670 dmv_y <<= 1 - s->quarter_sample;
673 xy = s->block_index[n];
676 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = 0;
677 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = 0;
678 s->current_picture.motion_val[1][xy][0] = 0;
679 s->current_picture.motion_val[1][xy][1] = 0;
680 if(mv1) { /* duplicate motion data for 1-MV block */
681 s->current_picture.motion_val[0][xy + 1][0] = 0;
682 s->current_picture.motion_val[0][xy + 1][1] = 0;
683 s->current_picture.motion_val[0][xy + wrap][0] = 0;
684 s->current_picture.motion_val[0][xy + wrap][1] = 0;
685 s->current_picture.motion_val[0][xy + wrap + 1][0] = 0;
686 s->current_picture.motion_val[0][xy + wrap + 1][1] = 0;
687 v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
688 s->current_picture.motion_val[1][xy + 1][0] = 0;
689 s->current_picture.motion_val[1][xy + 1][1] = 0;
690 s->current_picture.motion_val[1][xy + wrap][0] = 0;
691 s->current_picture.motion_val[1][xy + wrap][1] = 0;
692 s->current_picture.motion_val[1][xy + wrap + 1][0] = 0;
693 s->current_picture.motion_val[1][xy + wrap + 1][1] = 0;
698 C = s->current_picture.motion_val[0][xy - 1];
699 A = s->current_picture.motion_val[0][xy - wrap];
701 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
703 //in 4-MV mode different blocks have different B predictor position
706 off = (s->mb_x > 0) ? -1 : 1;
709 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
718 B = s->current_picture.motion_val[0][xy - wrap + off];
720 if(!s->first_slice_line || (n==2 || n==3)) { // predictor A is not out of bounds
721 if(s->mb_width == 1) {
725 px = mid_pred(A[0], B[0], C[0]);
726 py = mid_pred(A[1], B[1], C[1]);
728 } else if(s->mb_x || (n==1 || n==3)) { // predictor C is not out of bounds
734 /* Pullback MV as specified in 8.3.5.3.4 */
737 qx = (s->mb_x << 6) + ((n==1 || n==3) ? 32 : 0);
738 qy = (s->mb_y << 6) + ((n==2 || n==3) ? 32 : 0);
739 X = (s->mb_width << 6) - 4;
740 Y = (s->mb_height << 6) - 4;
742 if(qx + px < -60) px = -60 - qx;
743 if(qy + py < -60) py = -60 - qy;
745 if(qx + px < -28) px = -28 - qx;
746 if(qy + py < -28) py = -28 - qy;
748 if(qx + px > X) px = X - qx;
749 if(qy + py > Y) py = Y - qy;
751 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
752 if((!s->first_slice_line || (n==2 || n==3)) && (s->mb_x || (n==1 || n==3))) {
753 if(is_intra[xy - wrap])
754 sum = FFABS(px) + FFABS(py);
756 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
758 if(get_bits1(&s->gb)) {
767 sum = FFABS(px) + FFABS(py);
769 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
771 if(get_bits1(&s->gb)) {
781 /* store MV using signed modulus of MV range defined in 4.11 */
782 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
783 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
784 if(mv1) { /* duplicate motion data for 1-MV block */
785 s->current_picture.motion_val[0][xy + 1][0] = s->current_picture.motion_val[0][xy][0];
786 s->current_picture.motion_val[0][xy + 1][1] = s->current_picture.motion_val[0][xy][1];
787 s->current_picture.motion_val[0][xy + wrap][0] = s->current_picture.motion_val[0][xy][0];
788 s->current_picture.motion_val[0][xy + wrap][1] = s->current_picture.motion_val[0][xy][1];
789 s->current_picture.motion_val[0][xy + wrap + 1][0] = s->current_picture.motion_val[0][xy][0];
790 s->current_picture.motion_val[0][xy + wrap + 1][1] = s->current_picture.motion_val[0][xy][1];
794 /** Motion compensation for direct or interpolated blocks in B-frames
796 static void vc1_interp_mc(VC1Context *v)
798 MpegEncContext *s = &v->s;
799 DSPContext *dsp = &v->s.dsp;
800 uint8_t *srcY, *srcU, *srcV;
801 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
803 if(!v->s.next_picture.data[0])return;
807 uvmx = (mx + ((mx & 3) == 3)) >> 1;
808 uvmy = (my + ((my & 3) == 3)) >> 1;
810 uvmx = uvmx + ((uvmx<0)?-(uvmx&1):(uvmx&1));
811 uvmy = uvmy + ((uvmy<0)?-(uvmy&1):(uvmy&1));
813 srcY = s->next_picture.data[0];
814 srcU = s->next_picture.data[1];
815 srcV = s->next_picture.data[2];
817 src_x = s->mb_x * 16 + (mx >> 2);
818 src_y = s->mb_y * 16 + (my >> 2);
819 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
820 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
822 if(v->profile != PROFILE_ADVANCED){
823 src_x = av_clip( src_x, -16, s->mb_width * 16);
824 src_y = av_clip( src_y, -16, s->mb_height * 16);
825 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
826 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
828 src_x = av_clip( src_x, -17, s->avctx->coded_width);
829 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
830 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
831 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
834 srcY += src_y * s->linesize + src_x;
835 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
836 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
838 /* for grayscale we should not try to read from unknown area */
839 if(s->flags & CODEC_FLAG_GRAY) {
840 srcU = s->edge_emu_buffer + 18 * s->linesize;
841 srcV = s->edge_emu_buffer + 18 * s->linesize;
845 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
846 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
847 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
849 srcY -= s->mspel * (1 + s->linesize);
850 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
851 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
852 srcY = s->edge_emu_buffer;
853 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
854 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
855 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
856 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
859 /* if we deal with range reduction we need to scale source blocks */
865 for(j = 0; j < 17 + s->mspel*2; j++) {
866 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
869 src = srcU; src2 = srcV;
870 for(j = 0; j < 9; j++) {
871 for(i = 0; i < 9; i++) {
872 src[i] = ((src[i] - 128) >> 1) + 128;
873 src2[i] = ((src2[i] - 128) >> 1) + 128;
875 src += s->uvlinesize;
876 src2 += s->uvlinesize;
879 srcY += s->mspel * (1 + s->linesize);
883 dxy = ((my & 3) << 2) | (mx & 3);
884 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
885 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
886 srcY += s->linesize * 8;
887 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
888 v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
890 dxy = (my & 2) | ((mx & 2) >> 1);
893 dsp->avg_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
895 dsp->avg_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
898 if(s->flags & CODEC_FLAG_GRAY) return;
899 /* Chroma MC always uses qpel blilinear */
903 dsp->avg_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
904 dsp->avg_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
906 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
907 v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
911 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
915 #if B_FRACTION_DEN==256
919 return 2 * ((value * n + 255) >> 9);
920 return (value * n + 128) >> 8;
925 return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
926 return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
930 /** Reconstruct motion vector for B-frame and do motion compensation
932 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mode)
935 v->mv_mode2 = v->mv_mode;
936 v->mv_mode = MV_PMODE_INTENSITY_COMP;
941 if(v->use_ic) v->mv_mode = v->mv_mode2;
944 if(mode == BMV_TYPE_INTERPOLATED) {
947 if(v->use_ic) v->mv_mode = v->mv_mode2;
951 if(v->use_ic && (mode == BMV_TYPE_BACKWARD)) v->mv_mode = v->mv_mode2;
952 vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
953 if(v->use_ic) v->mv_mode = v->mv_mode2;
956 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mvtype)
958 MpegEncContext *s = &v->s;
959 int xy, wrap, off = 0;
964 const uint8_t *is_intra = v->mb_type[0];
968 /* scale MV difference to be quad-pel */
969 dmv_x[0] <<= 1 - s->quarter_sample;
970 dmv_y[0] <<= 1 - s->quarter_sample;
971 dmv_x[1] <<= 1 - s->quarter_sample;
972 dmv_y[1] <<= 1 - s->quarter_sample;
975 xy = s->block_index[0];
978 s->current_picture.motion_val[0][xy][0] =
979 s->current_picture.motion_val[0][xy][1] =
980 s->current_picture.motion_val[1][xy][0] =
981 s->current_picture.motion_val[1][xy][1] = 0;
984 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
985 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
986 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
987 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
989 /* Pullback predicted motion vectors as specified in 8.4.5.4 */
990 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));
991 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));
992 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));
993 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));
995 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
996 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
997 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
998 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
1002 if((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
1003 C = s->current_picture.motion_val[0][xy - 2];
1004 A = s->current_picture.motion_val[0][xy - wrap*2];
1005 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1006 B = s->current_picture.motion_val[0][xy - wrap*2 + off];
1008 if(!s->mb_x) C[0] = C[1] = 0;
1009 if(!s->first_slice_line) { // predictor A is not out of bounds
1010 if(s->mb_width == 1) {
1014 px = mid_pred(A[0], B[0], C[0]);
1015 py = mid_pred(A[1], B[1], C[1]);
1017 } else if(s->mb_x) { // predictor C is not out of bounds
1023 /* Pullback MV as specified in 8.3.5.3.4 */
1026 if(v->profile < PROFILE_ADVANCED) {
1027 qx = (s->mb_x << 5);
1028 qy = (s->mb_y << 5);
1029 X = (s->mb_width << 5) - 4;
1030 Y = (s->mb_height << 5) - 4;
1031 if(qx + px < -28) px = -28 - qx;
1032 if(qy + py < -28) py = -28 - qy;
1033 if(qx + px > X) px = X - qx;
1034 if(qy + py > Y) py = Y - qy;
1036 qx = (s->mb_x << 6);
1037 qy = (s->mb_y << 6);
1038 X = (s->mb_width << 6) - 4;
1039 Y = (s->mb_height << 6) - 4;
1040 if(qx + px < -60) px = -60 - qx;
1041 if(qy + py < -60) py = -60 - qy;
1042 if(qx + px > X) px = X - qx;
1043 if(qy + py > Y) py = Y - qy;
1046 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
1047 if(0 && !s->first_slice_line && s->mb_x) {
1048 if(is_intra[xy - wrap])
1049 sum = FFABS(px) + FFABS(py);
1051 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
1053 if(get_bits1(&s->gb)) {
1061 if(is_intra[xy - 2])
1062 sum = FFABS(px) + FFABS(py);
1064 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
1066 if(get_bits1(&s->gb)) {
1076 /* store MV using signed modulus of MV range defined in 4.11 */
1077 s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
1078 s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
1080 if((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
1081 C = s->current_picture.motion_val[1][xy - 2];
1082 A = s->current_picture.motion_val[1][xy - wrap*2];
1083 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1084 B = s->current_picture.motion_val[1][xy - wrap*2 + off];
1086 if(!s->mb_x) C[0] = C[1] = 0;
1087 if(!s->first_slice_line) { // predictor A is not out of bounds
1088 if(s->mb_width == 1) {
1092 px = mid_pred(A[0], B[0], C[0]);
1093 py = mid_pred(A[1], B[1], C[1]);
1095 } else if(s->mb_x) { // predictor C is not out of bounds
1101 /* Pullback MV as specified in 8.3.5.3.4 */
1104 if(v->profile < PROFILE_ADVANCED) {
1105 qx = (s->mb_x << 5);
1106 qy = (s->mb_y << 5);
1107 X = (s->mb_width << 5) - 4;
1108 Y = (s->mb_height << 5) - 4;
1109 if(qx + px < -28) px = -28 - qx;
1110 if(qy + py < -28) py = -28 - qy;
1111 if(qx + px > X) px = X - qx;
1112 if(qy + py > Y) py = Y - qy;
1114 qx = (s->mb_x << 6);
1115 qy = (s->mb_y << 6);
1116 X = (s->mb_width << 6) - 4;
1117 Y = (s->mb_height << 6) - 4;
1118 if(qx + px < -60) px = -60 - qx;
1119 if(qy + py < -60) py = -60 - qy;
1120 if(qx + px > X) px = X - qx;
1121 if(qy + py > Y) py = Y - qy;
1124 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
1125 if(0 && !s->first_slice_line && s->mb_x) {
1126 if(is_intra[xy - wrap])
1127 sum = FFABS(px) + FFABS(py);
1129 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
1131 if(get_bits1(&s->gb)) {
1139 if(is_intra[xy - 2])
1140 sum = FFABS(px) + FFABS(py);
1142 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
1144 if(get_bits1(&s->gb)) {
1154 /* store MV using signed modulus of MV range defined in 4.11 */
1156 s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
1157 s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
1159 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
1160 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
1161 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
1162 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
1165 /** Get predicted DC value for I-frames only
1166 * prediction dir: left=0, top=1
1167 * @param s MpegEncContext
1168 * @param overlap flag indicating that overlap filtering is used
1169 * @param pq integer part of picture quantizer
1170 * @param[in] n block index in the current MB
1171 * @param dc_val_ptr Pointer to DC predictor
1172 * @param dir_ptr Prediction direction for use in AC prediction
1174 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
1175 int16_t **dc_val_ptr, int *dir_ptr)
1177 int a, b, c, wrap, pred, scale;
1179 static const uint16_t dcpred[32] = {
1180 -1, 1024, 512, 341, 256, 205, 171, 146, 128,
1181 114, 102, 93, 85, 79, 73, 68, 64,
1182 60, 57, 54, 51, 49, 47, 45, 43,
1183 41, 39, 38, 37, 35, 34, 33
1186 /* find prediction - wmv3_dc_scale always used here in fact */
1187 if (n < 4) scale = s->y_dc_scale;
1188 else scale = s->c_dc_scale;
1190 wrap = s->block_wrap[n];
1191 dc_val= s->dc_val[0] + s->block_index[n];
1197 b = dc_val[ - 1 - wrap];
1198 a = dc_val[ - wrap];
1200 if (pq < 9 || !overlap)
1202 /* Set outer values */
1203 if (s->first_slice_line && (n!=2 && n!=3)) b=a=dcpred[scale];
1204 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=dcpred[scale];
1208 /* Set outer values */
1209 if (s->first_slice_line && (n!=2 && n!=3)) b=a=0;
1210 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=0;
1213 if (abs(a - b) <= abs(b - c)) {
1221 /* update predictor */
1222 *dc_val_ptr = &dc_val[0];
1227 /** Get predicted DC value
1228 * prediction dir: left=0, top=1
1229 * @param s MpegEncContext
1230 * @param overlap flag indicating that overlap filtering is used
1231 * @param pq integer part of picture quantizer
1232 * @param[in] n block index in the current MB
1233 * @param a_avail flag indicating top block availability
1234 * @param c_avail flag indicating left block availability
1235 * @param dc_val_ptr Pointer to DC predictor
1236 * @param dir_ptr Prediction direction for use in AC prediction
1238 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
1239 int a_avail, int c_avail,
1240 int16_t **dc_val_ptr, int *dir_ptr)
1242 int a, b, c, wrap, pred;
1244 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1247 wrap = s->block_wrap[n];
1248 dc_val= s->dc_val[0] + s->block_index[n];
1254 b = dc_val[ - 1 - wrap];
1255 a = dc_val[ - wrap];
1256 /* scale predictors if needed */
1257 q1 = s->current_picture.qscale_table[mb_pos];
1258 if(c_avail && (n!= 1 && n!=3)) {
1259 q2 = s->current_picture.qscale_table[mb_pos - 1];
1261 c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1263 if(a_avail && (n!= 2 && n!=3)) {
1264 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1266 a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1268 if(a_avail && c_avail && (n!=3)) {
1271 if(n != 2) off -= s->mb_stride;
1272 q2 = s->current_picture.qscale_table[off];
1274 b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
1277 if(a_avail && c_avail) {
1278 if(abs(a - b) <= abs(b - c)) {
1285 } else if(a_avail) {
1288 } else if(c_avail) {
1296 /* update predictor */
1297 *dc_val_ptr = &dc_val[0];
1301 /** @} */ // Block group
1304 * @defgroup vc1_std_mb VC1 Macroblock-level functions in Simple/Main Profiles
1305 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1309 static inline int vc1_coded_block_pred(MpegEncContext * s, int n, uint8_t **coded_block_ptr)
1311 int xy, wrap, pred, a, b, c;
1313 xy = s->block_index[n];
1314 wrap = s->b8_stride;
1319 a = s->coded_block[xy - 1 ];
1320 b = s->coded_block[xy - 1 - wrap];
1321 c = s->coded_block[xy - wrap];
1330 *coded_block_ptr = &s->coded_block[xy];
1336 * Decode one AC coefficient
1337 * @param v The VC1 context
1338 * @param last Last coefficient
1339 * @param skip How much zero coefficients to skip
1340 * @param value Decoded AC coefficient value
1341 * @param codingset set of VLC to decode data
1344 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip, int *value, int codingset)
1346 GetBitContext *gb = &v->s.gb;
1347 int index, escape, run = 0, level = 0, lst = 0;
1349 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
1350 if (index != vc1_ac_sizes[codingset] - 1) {
1351 run = vc1_index_decode_table[codingset][index][0];
1352 level = vc1_index_decode_table[codingset][index][1];
1353 lst = index >= vc1_last_decode_table[codingset] || get_bits_left(gb) < 0;
1357 escape = decode210(gb);
1359 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
1360 run = vc1_index_decode_table[codingset][index][0];
1361 level = vc1_index_decode_table[codingset][index][1];
1362 lst = index >= vc1_last_decode_table[codingset];
1365 level += vc1_last_delta_level_table[codingset][run];
1367 level += vc1_delta_level_table[codingset][run];
1370 run += vc1_last_delta_run_table[codingset][level] + 1;
1372 run += vc1_delta_run_table[codingset][level] + 1;
1378 lst = get_bits1(gb);
1379 if(v->s.esc3_level_length == 0) {
1380 if(v->pq < 8 || v->dquantfrm) { // table 59
1381 v->s.esc3_level_length = get_bits(gb, 3);
1382 if(!v->s.esc3_level_length)
1383 v->s.esc3_level_length = get_bits(gb, 2) + 8;
1385 v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
1387 v->s.esc3_run_length = 3 + get_bits(gb, 2);
1389 run = get_bits(gb, v->s.esc3_run_length);
1390 sign = get_bits1(gb);
1391 level = get_bits(gb, v->s.esc3_level_length);
1402 /** Decode intra block in intra frames - should be faster than decode_intra_block
1403 * @param v VC1Context
1404 * @param block block to decode
1405 * @param[in] n subblock index
1406 * @param coded are AC coeffs present or not
1407 * @param codingset set of VLC to decode data
1409 static int vc1_decode_i_block(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset)
1411 GetBitContext *gb = &v->s.gb;
1412 MpegEncContext *s = &v->s;
1413 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1416 int16_t *ac_val, *ac_val2;
1419 /* Get DC differential */
1421 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1423 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1426 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1431 if (dcdiff == 119 /* ESC index value */)
1433 /* TODO: Optimize */
1434 if (v->pq == 1) dcdiff = get_bits(gb, 10);
1435 else if (v->pq == 2) dcdiff = get_bits(gb, 9);
1436 else dcdiff = get_bits(gb, 8);
1441 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1442 else if (v->pq == 2)
1443 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1450 dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
1453 /* Store the quantized DC coeff, used for prediction */
1455 block[0] = dcdiff * s->y_dc_scale;
1457 block[0] = dcdiff * s->c_dc_scale;
1468 int last = 0, skip, value;
1469 const uint8_t *zz_table;
1473 scale = v->pq * 2 + v->halfpq;
1477 zz_table = v->zz_8x8[2];
1479 zz_table = v->zz_8x8[3];
1481 zz_table = v->zz_8x8[1];
1483 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1485 if(dc_pred_dir) //left
1488 ac_val -= 16 * s->block_wrap[n];
1491 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1495 block[zz_table[i++]] = value;
1498 /* apply AC prediction if needed */
1500 if(dc_pred_dir) { //left
1501 for(k = 1; k < 8; k++)
1502 block[k << v->left_blk_sh] += ac_val[k];
1504 for(k = 1; k < 8; k++)
1505 block[k << v->top_blk_sh] += ac_val[k + 8];
1508 /* save AC coeffs for further prediction */
1509 for(k = 1; k < 8; k++) {
1510 ac_val2[k] = block[k << v->left_blk_sh];
1511 ac_val2[k + 8] = block[k << v->top_blk_sh];
1514 /* scale AC coeffs */
1515 for(k = 1; k < 64; k++)
1519 block[k] += (block[k] < 0) ? -v->pq : v->pq;
1522 if(s->ac_pred) i = 63;
1528 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1532 scale = v->pq * 2 + v->halfpq;
1533 memset(ac_val2, 0, 16 * 2);
1534 if(dc_pred_dir) {//left
1537 memcpy(ac_val2, ac_val, 8 * 2);
1539 ac_val -= 16 * s->block_wrap[n];
1541 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1544 /* apply AC prediction if needed */
1546 if(dc_pred_dir) { //left
1547 for(k = 1; k < 8; k++) {
1548 block[k << v->left_blk_sh] = ac_val[k] * scale;
1549 if(!v->pquantizer && block[k << v->left_blk_sh])
1550 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -v->pq : v->pq;
1553 for(k = 1; k < 8; k++) {
1554 block[k << v->top_blk_sh] = ac_val[k + 8] * scale;
1555 if(!v->pquantizer && block[k << v->top_blk_sh])
1556 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -v->pq : v->pq;
1562 s->block_last_index[n] = i;
1567 /** Decode intra block in intra frames - should be faster than decode_intra_block
1568 * @param v VC1Context
1569 * @param block block to decode
1570 * @param[in] n subblock number
1571 * @param coded are AC coeffs present or not
1572 * @param codingset set of VLC to decode data
1573 * @param mquant quantizer value for this macroblock
1575 static int vc1_decode_i_block_adv(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset, int mquant)
1577 GetBitContext *gb = &v->s.gb;
1578 MpegEncContext *s = &v->s;
1579 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1582 int16_t *ac_val, *ac_val2;
1584 int a_avail = v->a_avail, c_avail = v->c_avail;
1585 int use_pred = s->ac_pred;
1588 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1590 /* Get DC differential */
1592 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1594 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1597 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1602 if (dcdiff == 119 /* ESC index value */)
1604 /* TODO: Optimize */
1605 if (mquant == 1) dcdiff = get_bits(gb, 10);
1606 else if (mquant == 2) dcdiff = get_bits(gb, 9);
1607 else dcdiff = get_bits(gb, 8);
1612 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1613 else if (mquant == 2)
1614 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1621 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
1624 /* Store the quantized DC coeff, used for prediction */
1626 block[0] = dcdiff * s->y_dc_scale;
1628 block[0] = dcdiff * s->c_dc_scale;
1634 /* check if AC is needed at all */
1635 if(!a_avail && !c_avail) use_pred = 0;
1636 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1639 scale = mquant * 2 + ((mquant == v->pq) ? v->halfpq : 0);
1641 if(dc_pred_dir) //left
1644 ac_val -= 16 * s->block_wrap[n];
1646 q1 = s->current_picture.qscale_table[mb_pos];
1647 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
1648 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1649 if(dc_pred_dir && n==1) q2 = q1;
1650 if(!dc_pred_dir && n==2) q2 = q1;
1654 int last = 0, skip, value;
1655 const uint8_t *zz_table;
1660 zz_table = v->zz_8x8[2];
1662 zz_table = v->zz_8x8[3];
1664 zz_table = v->zz_8x8[1];
1667 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1671 block[zz_table[i++]] = value;
1674 /* apply AC prediction if needed */
1676 /* scale predictors if needed*/
1678 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1679 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1681 if(dc_pred_dir) { //left
1682 for(k = 1; k < 8; k++)
1683 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1685 for(k = 1; k < 8; k++)
1686 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1689 if(dc_pred_dir) { //left
1690 for(k = 1; k < 8; k++)
1691 block[k << v->left_blk_sh] += ac_val[k];
1693 for(k = 1; k < 8; k++)
1694 block[k << v->top_blk_sh] += ac_val[k + 8];
1698 /* save AC coeffs for further prediction */
1699 for(k = 1; k < 8; k++) {
1700 ac_val2[k ] = block[k << v->left_blk_sh];
1701 ac_val2[k + 8] = block[k << v->top_blk_sh];
1704 /* scale AC coeffs */
1705 for(k = 1; k < 64; k++)
1709 block[k] += (block[k] < 0) ? -mquant : mquant;
1712 if(use_pred) i = 63;
1713 } else { // no AC coeffs
1716 memset(ac_val2, 0, 16 * 2);
1717 if(dc_pred_dir) {//left
1719 memcpy(ac_val2, ac_val, 8 * 2);
1721 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1722 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1723 for(k = 1; k < 8; k++)
1724 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1729 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1731 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1732 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1733 for(k = 1; k < 8; k++)
1734 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1739 /* apply AC prediction if needed */
1741 if(dc_pred_dir) { //left
1742 for(k = 1; k < 8; k++) {
1743 block[k << v->left_blk_sh] = ac_val2[k] * scale;
1744 if(!v->pquantizer && block[k << v->left_blk_sh])
1745 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
1748 for(k = 1; k < 8; k++) {
1749 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
1750 if(!v->pquantizer && block[k << v->top_blk_sh])
1751 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
1757 s->block_last_index[n] = i;
1762 /** Decode intra block in inter frames - more generic version than vc1_decode_i_block
1763 * @param v VC1Context
1764 * @param block block to decode
1765 * @param[in] n subblock index
1766 * @param coded are AC coeffs present or not
1767 * @param mquant block quantizer
1768 * @param codingset set of VLC to decode data
1770 static int vc1_decode_intra_block(VC1Context *v, DCTELEM block[64], int n, int coded, int mquant, int codingset)
1772 GetBitContext *gb = &v->s.gb;
1773 MpegEncContext *s = &v->s;
1774 int dc_pred_dir = 0; /* Direction of the DC prediction used */
1777 int16_t *ac_val, *ac_val2;
1779 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1780 int a_avail = v->a_avail, c_avail = v->c_avail;
1781 int use_pred = s->ac_pred;
1785 s->dsp.clear_block(block);
1787 /* XXX: Guard against dumb values of mquant */
1788 mquant = (mquant < 1) ? 0 : ( (mquant>31) ? 31 : mquant );
1790 /* Set DC scale - y and c use the same */
1791 s->y_dc_scale = s->y_dc_scale_table[mquant];
1792 s->c_dc_scale = s->c_dc_scale_table[mquant];
1794 /* Get DC differential */
1796 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1798 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
1801 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
1806 if (dcdiff == 119 /* ESC index value */)
1808 /* TODO: Optimize */
1809 if (mquant == 1) dcdiff = get_bits(gb, 10);
1810 else if (mquant == 2) dcdiff = get_bits(gb, 9);
1811 else dcdiff = get_bits(gb, 8);
1816 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
1817 else if (mquant == 2)
1818 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
1825 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
1828 /* Store the quantized DC coeff, used for prediction */
1831 block[0] = dcdiff * s->y_dc_scale;
1833 block[0] = dcdiff * s->c_dc_scale;
1839 /* check if AC is needed at all and adjust direction if needed */
1840 if(!a_avail) dc_pred_dir = 1;
1841 if(!c_avail) dc_pred_dir = 0;
1842 if(!a_avail && !c_avail) use_pred = 0;
1843 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
1846 scale = mquant * 2 + v->halfpq;
1848 if(dc_pred_dir) //left
1851 ac_val -= 16 * s->block_wrap[n];
1853 q1 = s->current_picture.qscale_table[mb_pos];
1854 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
1855 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
1856 if(dc_pred_dir && n==1) q2 = q1;
1857 if(!dc_pred_dir && n==2) q2 = q1;
1861 int last = 0, skip, value;
1865 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
1869 block[v->zz_8x8[0][i++]] = value;
1872 /* apply AC prediction if needed */
1874 /* scale predictors if needed*/
1876 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1877 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1879 if(dc_pred_dir) { //left
1880 for(k = 1; k < 8; k++)
1881 block[k << v->left_blk_sh] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1883 for(k = 1; k < 8; k++)
1884 block[k << v->top_blk_sh] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1887 if(dc_pred_dir) { //left
1888 for(k = 1; k < 8; k++)
1889 block[k << v->left_blk_sh] += ac_val[k];
1891 for(k = 1; k < 8; k++)
1892 block[k << v->top_blk_sh] += ac_val[k + 8];
1896 /* save AC coeffs for further prediction */
1897 for(k = 1; k < 8; k++) {
1898 ac_val2[k ] = block[k << v->left_blk_sh];
1899 ac_val2[k + 8] = block[k << v->top_blk_sh];
1902 /* scale AC coeffs */
1903 for(k = 1; k < 64; k++)
1907 block[k] += (block[k] < 0) ? -mquant : mquant;
1910 if(use_pred) i = 63;
1911 } else { // no AC coeffs
1914 memset(ac_val2, 0, 16 * 2);
1915 if(dc_pred_dir) {//left
1917 memcpy(ac_val2, ac_val, 8 * 2);
1919 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1920 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1921 for(k = 1; k < 8; k++)
1922 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1927 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
1929 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
1930 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
1931 for(k = 1; k < 8; k++)
1932 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
1937 /* apply AC prediction if needed */
1939 if(dc_pred_dir) { //left
1940 for(k = 1; k < 8; k++) {
1941 block[k << v->left_blk_sh] = ac_val2[k] * scale;
1942 if(!v->pquantizer && block[k << v->left_blk_sh])
1943 block[k << v->left_blk_sh] += (block[k << v->left_blk_sh] < 0) ? -mquant : mquant;
1946 for(k = 1; k < 8; k++) {
1947 block[k << v->top_blk_sh] = ac_val2[k + 8] * scale;
1948 if(!v->pquantizer && block[k << v->top_blk_sh])
1949 block[k << v->top_blk_sh] += (block[k << v->top_blk_sh] < 0) ? -mquant : mquant;
1955 s->block_last_index[n] = i;
1962 static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n, int mquant, int ttmb, int first_block,
1963 uint8_t *dst, int linesize, int skip_block, int *ttmb_out)
1965 MpegEncContext *s = &v->s;
1966 GetBitContext *gb = &s->gb;
1969 int scale, off, idx, last, skip, value;
1970 int ttblk = ttmb & 7;
1973 s->dsp.clear_block(block);
1976 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)];
1978 if(ttblk == TT_4X4) {
1979 subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
1981 if((ttblk != TT_8X8 && ttblk != TT_4X4)
1982 && ((v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))
1983 || (!v->res_rtm_flag && !first_block))) {
1984 subblkpat = decode012(gb);
1985 if(subblkpat) subblkpat ^= 3; //swap decoded pattern bits
1986 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) ttblk = TT_8X4;
1987 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) ttblk = TT_4X8;
1989 scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);
1991 // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
1992 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
1993 subblkpat = 2 - (ttblk == TT_8X4_TOP);
1996 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
1997 subblkpat = 2 - (ttblk == TT_4X8_LEFT);
2006 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2010 idx = v->zz_8x8[0][i++];
2011 block[idx] = value * scale;
2013 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2017 v->vc1dsp.vc1_inv_trans_8x8_dc(dst, linesize, block);
2019 v->vc1dsp.vc1_inv_trans_8x8_add(dst, linesize, block);
2024 pat = ~subblkpat & 0xF;
2025 for(j = 0; j < 4; j++) {
2026 last = subblkpat & (1 << (3 - j));
2028 off = (j & 1) * 4 + (j & 2) * 16;
2030 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2034 idx = ff_vc1_simple_progressive_4x4_zz[i++];
2035 block[idx + off] = value * scale;
2037 block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
2039 if(!(subblkpat & (1 << (3 - j))) && !skip_block){
2041 v->vc1dsp.vc1_inv_trans_4x4_dc(dst + (j&1)*4 + (j&2)*2*linesize, linesize, block + off);
2043 v->vc1dsp.vc1_inv_trans_4x4(dst + (j&1)*4 + (j&2)*2*linesize, linesize, block + off);
2048 pat = ~((subblkpat & 2)*6 + (subblkpat & 1)*3) & 0xF;
2049 for(j = 0; j < 2; j++) {
2050 last = subblkpat & (1 << (1 - j));
2054 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2058 idx = v->zz_8x4[i++]+off;
2059 block[idx] = value * scale;
2061 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2063 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
2065 v->vc1dsp.vc1_inv_trans_8x4_dc(dst + j*4*linesize, linesize, block + off);
2067 v->vc1dsp.vc1_inv_trans_8x4(dst + j*4*linesize, linesize, block + off);
2072 pat = ~(subblkpat*5) & 0xF;
2073 for(j = 0; j < 2; j++) {
2074 last = subblkpat & (1 << (1 - j));
2078 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
2082 idx = v->zz_4x8[i++]+off;
2083 block[idx] = value * scale;
2085 block[idx] += (block[idx] < 0) ? -mquant : mquant;
2087 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
2089 v->vc1dsp.vc1_inv_trans_4x8_dc(dst + j*4, linesize, block + off);
2091 v->vc1dsp.vc1_inv_trans_4x8(dst + j*4, linesize, block + off);
2097 *ttmb_out |= ttblk << (n * 4);
2101 /** @} */ // Macroblock group
2103 static const int size_table [6] = { 0, 2, 3, 4, 5, 8 };
2104 static const int offset_table[6] = { 0, 1, 3, 7, 15, 31 };
2106 static av_always_inline void vc1_apply_p_v_loop_filter(VC1Context *v, int block_num)
2108 MpegEncContext *s = &v->s;
2109 int mb_cbp = v->cbp[s->mb_x - s->mb_stride],
2110 block_cbp = mb_cbp >> (block_num * 4), bottom_cbp,
2111 mb_is_intra = v->is_intra[s->mb_x - s->mb_stride],
2112 block_is_intra = mb_is_intra >> (block_num * 4), bottom_is_intra;
2113 int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
2117 dst = s->dest[block_num - 3];
2119 dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 8) * linesize;
2121 if (s->mb_y != s->mb_height || block_num < 2) {
2126 bottom_cbp = v->cbp[s->mb_x] >> (block_num * 4);
2127 bottom_is_intra = v->is_intra[s->mb_x] >> (block_num * 4);
2128 mv = &v->luma_mv[s->mb_x - s->mb_stride];
2129 mv_stride = s->mb_stride;
2131 bottom_cbp = (block_num < 2) ? (mb_cbp >> ((block_num + 2) * 4)) :
2132 (v->cbp[s->mb_x] >> ((block_num - 2) * 4));
2133 bottom_is_intra = (block_num < 2) ? (mb_is_intra >> ((block_num + 2) * 4)) :
2134 (v->is_intra[s->mb_x] >> ((block_num - 2) * 4));
2135 mv_stride = s->b8_stride;
2136 mv = &s->current_picture.motion_val[0][s->block_index[block_num] - 2 * mv_stride];
2139 if (bottom_is_intra & 1 || block_is_intra & 1 ||
2140 mv[0][0] != mv[mv_stride][0] || mv[0][1] != mv[mv_stride][1]) {
2141 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
2143 idx = ((bottom_cbp >> 2) | block_cbp) & 3;
2145 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
2148 v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);
2150 v->vc1dsp.vc1_v_loop_filter4(dst, linesize, v->pq);
2155 dst -= 4 * linesize;
2156 ttblk = (v->ttblk[s->mb_x - s->mb_stride] >> (block_num * 4)) & 0xf;
2157 if (ttblk == TT_4X4 || ttblk == TT_8X4) {
2158 idx = (block_cbp | (block_cbp >> 2)) & 3;
2160 v->vc1dsp.vc1_v_loop_filter8(dst, linesize, v->pq);
2163 v->vc1dsp.vc1_v_loop_filter4(dst + 4, linesize, v->pq);
2165 v->vc1dsp.vc1_v_loop_filter4(dst, linesize, v->pq);
2170 static av_always_inline void vc1_apply_p_h_loop_filter(VC1Context *v, int block_num)
2172 MpegEncContext *s = &v->s;
2173 int mb_cbp = v->cbp[s->mb_x - 1 - s->mb_stride],
2174 block_cbp = mb_cbp >> (block_num * 4), right_cbp,
2175 mb_is_intra = v->is_intra[s->mb_x - 1 - s->mb_stride],
2176 block_is_intra = mb_is_intra >> (block_num * 4), right_is_intra;
2177 int idx, linesize = block_num > 3 ? s->uvlinesize : s->linesize, ttblk;
2180 if (block_num > 3) {
2181 dst = s->dest[block_num - 3] - 8 * linesize;
2183 dst = s->dest[0] + (block_num & 1) * 8 + ((block_num & 2) * 4 - 16) * linesize - 8;
2186 if (s->mb_x != s->mb_width || !(block_num & 5)) {
2190 right_cbp = v->cbp[s->mb_x - s->mb_stride] >> (block_num * 4);
2191 right_is_intra = v->is_intra[s->mb_x - s->mb_stride] >> (block_num * 4);
2192 mv = &v->luma_mv[s->mb_x - s->mb_stride - 1];
2194 right_cbp = (block_num & 1) ? (v->cbp[s->mb_x - s->mb_stride] >> ((block_num - 1) * 4)) :
2195 (mb_cbp >> ((block_num + 1) * 4));
2196 right_is_intra = (block_num & 1) ? (v->is_intra[s->mb_x - s->mb_stride] >> ((block_num - 1) * 4)) :
2197 (mb_is_intra >> ((block_num + 1) * 4));
2198 mv = &s->current_picture.motion_val[0][s->block_index[block_num] - s->b8_stride * 2 - 2];
2200 if (block_is_intra & 1 || right_is_intra & 1 || mv[0][0] != mv[1][0] || mv[0][1] != mv[1][1]) {
2201 v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
2203 idx = ((right_cbp >> 1) | block_cbp) & 5; // FIXME check
2205 v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
2208 v->vc1dsp.vc1_h_loop_filter4(dst+4*linesize, linesize, v->pq);
2210 v->vc1dsp.vc1_h_loop_filter4(dst, linesize, v->pq);
2216 ttblk = (v->ttblk[s->mb_x - s->mb_stride - 1] >> (block_num * 4)) & 0xf;
2217 if (ttblk == TT_4X4 || ttblk == TT_4X8) {
2218 idx = (block_cbp | (block_cbp >> 1)) & 5;
2220 v->vc1dsp.vc1_h_loop_filter8(dst, linesize, v->pq);
2223 v->vc1dsp.vc1_h_loop_filter4(dst + linesize*4, linesize, v->pq);
2225 v->vc1dsp.vc1_h_loop_filter4(dst, linesize, v->pq);
2230 static void vc1_apply_p_loop_filter(VC1Context *v)
2232 MpegEncContext *s = &v->s;
2235 for (i = 0; i < 6; i++) {
2236 vc1_apply_p_v_loop_filter(v, i);
2239 /* V always preceedes H, therefore we run H one MB before V;
2240 * at the end of a row, we catch up to complete the row */
2242 for (i = 0; i < 6; i++) {
2243 vc1_apply_p_h_loop_filter(v, i);
2245 if (s->mb_x == s->mb_width - 1) {
2247 ff_update_block_index(s);
2248 for (i = 0; i < 6; i++) {
2249 vc1_apply_p_h_loop_filter(v, i);
2255 /** Decode one P-frame MB (in Simple/Main profile)
2257 static int vc1_decode_p_mb(VC1Context *v)
2259 MpegEncContext *s = &v->s;
2260 GetBitContext *gb = &s->gb;
2262 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2263 int cbp; /* cbp decoding stuff */
2264 int mqdiff, mquant; /* MB quantization */
2265 int ttmb = v->ttfrm; /* MB Transform type */
2267 int mb_has_coeffs = 1; /* last_flag */
2268 int dmv_x, dmv_y; /* Differential MV components */
2269 int index, index1; /* LUT indexes */
2270 int val, sign; /* temp values */
2271 int first_block = 1;
2273 int skipped, fourmv;
2274 int block_cbp = 0, pat, block_tt = 0, block_intra = 0;
2276 mquant = v->pq; /* Loosy initialization */
2278 if (v->mv_type_is_raw)
2279 fourmv = get_bits1(gb);
2281 fourmv = v->mv_type_mb_plane[mb_pos];
2283 skipped = get_bits1(gb);
2285 skipped = v->s.mbskip_table[mb_pos];
2287 if (!fourmv) /* 1MV mode */
2291 vc1_idct_func idct8x8_fn;
2293 GET_MVDATA(dmv_x, dmv_y);
2296 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2297 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2299 s->current_picture.mb_type[mb_pos] = s->mb_intra ? MB_TYPE_INTRA : MB_TYPE_16x16;
2300 vc1_pred_mv(v, 0, dmv_x, dmv_y, 1, v->range_x, v->range_y, v->mb_type[0]);
2302 /* FIXME Set DC val for inter block ? */
2303 if (s->mb_intra && !mb_has_coeffs)
2306 s->ac_pred = get_bits1(gb);
2309 else if (mb_has_coeffs)
2311 if (s->mb_intra) s->ac_pred = get_bits1(gb);
2312 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2320 s->current_picture.qscale_table[mb_pos] = mquant;
2322 if (!v->ttmbf && !s->mb_intra && mb_has_coeffs)
2323 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table,
2324 VC1_TTMB_VLC_BITS, 2);
2325 if(!s->mb_intra) vc1_mc_1mv(v, 0);
2327 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2330 s->dc_val[0][s->block_index[i]] = 0;
2332 val = ((cbp >> (5 - i)) & 1);
2333 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2334 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2336 /* check if prediction blocks A and C are available */
2337 v->a_avail = v->c_avail = 0;
2338 if(i == 2 || i == 3 || !s->first_slice_line)
2339 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2340 if(i == 1 || i == 3 || s->mb_x)
2341 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2343 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
2344 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2345 idct8x8_fn(s->dest[dst_idx] + off,
2346 i & 4 ? s->uvlinesize : s->linesize,
2348 if(v->pq >= 9 && v->overlap) {
2350 v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2352 v->vc1dsp.vc1_v_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2354 block_cbp |= 0xF << (i << 2);
2355 block_intra |= 1 << i;
2357 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);
2358 block_cbp |= pat << (i << 2);
2359 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2367 for(i = 0; i < 6; i++) {
2368 v->mb_type[0][s->block_index[i]] = 0;
2369 s->dc_val[0][s->block_index[i]] = 0;
2371 s->current_picture.mb_type[mb_pos] = MB_TYPE_SKIP;
2372 s->current_picture.qscale_table[mb_pos] = 0;
2373 vc1_pred_mv(v, 0, 0, 0, 1, v->range_x, v->range_y, v->mb_type[0]);
2379 if (!skipped /* unskipped MB */)
2381 int intra_count = 0, coded_inter = 0;
2382 int is_intra[6], is_coded[6];
2383 vc1_idct_func idct8x8_fn;
2385 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2388 val = ((cbp >> (5 - i)) & 1);
2389 s->dc_val[0][s->block_index[i]] = 0;
2396 GET_MVDATA(dmv_x, dmv_y);
2398 vc1_pred_mv(v, i, dmv_x, dmv_y, 0, v->range_x, v->range_y, v->mb_type[0]);
2399 if(!s->mb_intra) vc1_mc_4mv_luma(v, i);
2400 intra_count += s->mb_intra;
2401 is_intra[i] = s->mb_intra;
2402 is_coded[i] = mb_has_coeffs;
2405 is_intra[i] = (intra_count >= 3);
2408 if(i == 4) vc1_mc_4mv_chroma(v);
2409 v->mb_type[0][s->block_index[i]] = is_intra[i];
2410 if(!coded_inter) coded_inter = !is_intra[i] & is_coded[i];
2412 // if there are no coded blocks then don't do anything more
2414 if(!intra_count && !coded_inter)
2417 s->current_picture.qscale_table[mb_pos] = mquant;
2418 /* test if block is intra and has pred */
2423 if(((!s->first_slice_line || (i==2 || i==3)) && v->mb_type[0][s->block_index[i] - s->block_wrap[i]])
2424 || ((s->mb_x || (i==1 || i==3)) && v->mb_type[0][s->block_index[i] - 1])) {
2429 if(intrapred)s->ac_pred = get_bits1(gb);
2430 else s->ac_pred = 0;
2432 if (!v->ttmbf && coded_inter)
2433 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2434 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2438 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2439 s->mb_intra = is_intra[i];
2441 /* check if prediction blocks A and C are available */
2442 v->a_avail = v->c_avail = 0;
2443 if(i == 2 || i == 3 || !s->first_slice_line)
2444 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2445 if(i == 1 || i == 3 || s->mb_x)
2446 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2448 vc1_decode_intra_block(v, s->block[i], i, is_coded[i], mquant, (i&4)?v->codingset2:v->codingset);
2449 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2450 idct8x8_fn(s->dest[dst_idx] + off,
2451 (i&4)?s->uvlinesize:s->linesize,
2453 if(v->pq >= 9 && v->overlap) {
2455 v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2457 v->vc1dsp.vc1_v_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
2459 block_cbp |= 0xF << (i << 2);
2460 block_intra |= 1 << i;
2461 } else if(is_coded[i]) {
2462 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);
2463 block_cbp |= pat << (i << 2);
2464 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2472 s->current_picture.qscale_table[mb_pos] = 0;
2473 for (i=0; i<6; i++) {
2474 v->mb_type[0][s->block_index[i]] = 0;
2475 s->dc_val[0][s->block_index[i]] = 0;
2479 vc1_pred_mv(v, i, 0, 0, 0, v->range_x, v->range_y, v->mb_type[0]);
2480 vc1_mc_4mv_luma(v, i);
2482 vc1_mc_4mv_chroma(v);
2483 s->current_picture.qscale_table[mb_pos] = 0;
2487 v->cbp[s->mb_x] = block_cbp;
2488 v->ttblk[s->mb_x] = block_tt;
2489 v->is_intra[s->mb_x] = block_intra;
2494 /** Decode one B-frame MB (in Main profile)
2496 static void vc1_decode_b_mb(VC1Context *v)
2498 MpegEncContext *s = &v->s;
2499 GetBitContext *gb = &s->gb;
2501 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2502 int cbp = 0; /* cbp decoding stuff */
2503 int mqdiff, mquant; /* MB quantization */
2504 int ttmb = v->ttfrm; /* MB Transform type */
2505 int mb_has_coeffs = 0; /* last_flag */
2506 int index, index1; /* LUT indexes */
2507 int val, sign; /* temp values */
2508 int first_block = 1;
2510 int skipped, direct;
2511 int dmv_x[2], dmv_y[2];
2512 int bmvtype = BMV_TYPE_BACKWARD;
2513 vc1_idct_func idct8x8_fn;
2515 mquant = v->pq; /* Loosy initialization */
2519 direct = get_bits1(gb);
2521 direct = v->direct_mb_plane[mb_pos];
2523 skipped = get_bits1(gb);
2525 skipped = v->s.mbskip_table[mb_pos];
2527 dmv_x[0] = dmv_x[1] = dmv_y[0] = dmv_y[1] = 0;
2528 for(i = 0; i < 6; i++) {
2529 v->mb_type[0][s->block_index[i]] = 0;
2530 s->dc_val[0][s->block_index[i]] = 0;
2532 s->current_picture.qscale_table[mb_pos] = 0;
2536 GET_MVDATA(dmv_x[0], dmv_y[0]);
2537 dmv_x[1] = dmv_x[0];
2538 dmv_y[1] = dmv_y[0];
2540 if(skipped || !s->mb_intra) {
2541 bmvtype = decode012(gb);
2544 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_BACKWARD : BMV_TYPE_FORWARD;
2547 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_FORWARD : BMV_TYPE_BACKWARD;
2550 bmvtype = BMV_TYPE_INTERPOLATED;
2551 dmv_x[0] = dmv_y[0] = 0;
2555 for(i = 0; i < 6; i++)
2556 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2559 if(direct) bmvtype = BMV_TYPE_INTERPOLATED;
2560 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2561 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2565 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2568 s->current_picture.qscale_table[mb_pos] = mquant;
2570 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2571 dmv_x[0] = dmv_y[0] = dmv_x[1] = dmv_y[1] = 0;
2572 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2573 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2575 if(!mb_has_coeffs && !s->mb_intra) {
2576 /* no coded blocks - effectively skipped */
2577 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2578 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2581 if(s->mb_intra && !mb_has_coeffs) {
2583 s->current_picture.qscale_table[mb_pos] = mquant;
2584 s->ac_pred = get_bits1(gb);
2586 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2588 if(bmvtype == BMV_TYPE_INTERPOLATED) {
2589 GET_MVDATA(dmv_x[0], dmv_y[0]);
2590 if(!mb_has_coeffs) {
2591 /* interpolated skipped block */
2592 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2593 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2597 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
2599 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
2602 s->ac_pred = get_bits1(gb);
2603 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
2605 s->current_picture.qscale_table[mb_pos] = mquant;
2606 if(!v->ttmbf && !s->mb_intra && mb_has_coeffs)
2607 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
2611 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2614 s->dc_val[0][s->block_index[i]] = 0;
2616 val = ((cbp >> (5 - i)) & 1);
2617 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
2618 v->mb_type[0][s->block_index[i]] = s->mb_intra;
2620 /* check if prediction blocks A and C are available */
2621 v->a_avail = v->c_avail = 0;
2622 if(i == 2 || i == 3 || !s->first_slice_line)
2623 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
2624 if(i == 1 || i == 3 || s->mb_x)
2625 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
2627 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
2628 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
2629 idct8x8_fn(s->dest[dst_idx] + off,
2630 i & 4 ? s->uvlinesize : s->linesize,
2633 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);
2634 if(!v->ttmbf && ttmb < 8) ttmb = -1;
2640 /** Decode blocks of I-frame
2642 static void vc1_decode_i_blocks(VC1Context *v)
2645 MpegEncContext *s = &v->s;
2649 vc1_idct_func idct8x8_fn;
2651 /* select codingmode used for VLC tables selection */
2652 switch(v->y_ac_table_index){
2654 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2657 v->codingset = CS_HIGH_MOT_INTRA;
2660 v->codingset = CS_MID_RATE_INTRA;
2664 switch(v->c_ac_table_index){
2666 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2669 v->codingset2 = CS_HIGH_MOT_INTER;
2672 v->codingset2 = CS_MID_RATE_INTER;
2676 /* Set DC scale - y and c use the same */
2677 s->y_dc_scale = s->y_dc_scale_table[v->pq];
2678 s->c_dc_scale = s->c_dc_scale_table[v->pq];
2681 s->mb_x = s->mb_y = 0;
2683 s->first_slice_line = 1;
2684 if(v->pq >= 9 && v->overlap) {
2685 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
2687 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put[!!v->rangeredfrm];
2688 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
2690 ff_init_block_index(s);
2691 for(; s->mb_x < s->mb_width; s->mb_x++) {
2693 ff_update_block_index(s);
2694 dst[0] = s->dest[0];
2695 dst[1] = dst[0] + 8;
2696 dst[2] = s->dest[0] + s->linesize * 8;
2697 dst[3] = dst[2] + 8;
2698 dst[4] = s->dest[1];
2699 dst[5] = s->dest[2];
2700 s->dsp.clear_blocks(s->block[0]);
2701 mb_pos = s->mb_x + s->mb_y * s->mb_width;
2702 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
2703 s->current_picture.qscale_table[mb_pos] = v->pq;
2704 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2705 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2707 // do actual MB decoding and displaying
2708 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
2709 v->s.ac_pred = get_bits1(&v->s.gb);
2711 for(k = 0; k < 6; k++) {
2712 val = ((cbp >> (5 - k)) & 1);
2715 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
2719 cbp |= val << (5 - k);
2721 vc1_decode_i_block(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2);
2723 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
2725 k & 4 ? s->uvlinesize : s->linesize,
2729 if(v->pq >= 9 && v->overlap) {
2731 v->vc1dsp.vc1_h_overlap(s->dest[0], s->linesize);
2732 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2733 if(!(s->flags & CODEC_FLAG_GRAY)) {
2734 v->vc1dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
2735 v->vc1dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
2738 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
2739 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2740 if(!s->first_slice_line) {
2741 v->vc1dsp.vc1_v_overlap(s->dest[0], s->linesize);
2742 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
2743 if(!(s->flags & CODEC_FLAG_GRAY)) {
2744 v->vc1dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
2745 v->vc1dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
2748 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2749 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2751 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
2753 if(get_bits_count(&s->gb) > v->bits) {
2754 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2755 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
2759 if (!v->s.loop_filter)
2760 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2762 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
2764 s->first_slice_line = 0;
2766 if (v->s.loop_filter)
2767 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
2768 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
2771 /** Decode blocks of I-frame for advanced profile
2773 static void vc1_decode_i_blocks_adv(VC1Context *v, int mby_start, int mby_end)
2776 MpegEncContext *s = &v->s;
2783 GetBitContext *gb = &s->gb;
2784 vc1_idct_func idct8x8_fn;
2786 /* select codingmode used for VLC tables selection */
2787 switch(v->y_ac_table_index){
2789 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2792 v->codingset = CS_HIGH_MOT_INTRA;
2795 v->codingset = CS_MID_RATE_INTRA;
2799 switch(v->c_ac_table_index){
2801 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2804 v->codingset2 = CS_HIGH_MOT_INTER;
2807 v->codingset2 = CS_MID_RATE_INTER;
2812 s->mb_x = s->mb_y = 0;
2814 s->first_slice_line = 1;
2815 s->mb_y = mby_start;
2818 ff_init_block_index(s);
2819 memset(&s->coded_block[s->block_index[0]-s->b8_stride], 0,
2820 s->b8_stride * sizeof(*s->coded_block));
2822 idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[0];
2823 for(; s->mb_y < mby_end; s->mb_y++) {
2825 ff_init_block_index(s);
2826 for(;s->mb_x < s->mb_width; s->mb_x++) {
2828 ff_update_block_index(s);
2829 dst[0] = s->dest[0];
2830 dst[1] = dst[0] + 8;
2831 dst[2] = s->dest[0] + s->linesize * 8;
2832 dst[3] = dst[2] + 8;
2833 dst[4] = s->dest[1];
2834 dst[5] = s->dest[2];
2835 s->dsp.clear_blocks(s->block[0]);
2836 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2837 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
2838 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
2839 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
2841 // do actual MB decoding and displaying
2842 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
2843 if(v->acpred_is_raw)
2844 v->s.ac_pred = get_bits1(&v->s.gb);
2846 v->s.ac_pred = v->acpred_plane[mb_pos];
2848 if(v->condover == CONDOVER_SELECT) {
2849 if(v->overflg_is_raw)
2850 overlap = get_bits1(&v->s.gb);
2852 overlap = v->over_flags_plane[mb_pos];
2854 overlap = (v->condover == CONDOVER_ALL);
2858 s->current_picture.qscale_table[mb_pos] = mquant;
2859 /* Set DC scale - y and c use the same */
2860 s->y_dc_scale = s->y_dc_scale_table[mquant];
2861 s->c_dc_scale = s->c_dc_scale_table[mquant];
2863 for(k = 0; k < 6; k++) {
2864 val = ((cbp >> (5 - k)) & 1);
2867 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
2871 cbp |= val << (5 - k);
2873 v->a_avail = !s->first_slice_line || (k==2 || k==3);
2874 v->c_avail = !!s->mb_x || (k==1 || k==3);
2876 vc1_decode_i_block_adv(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant);
2878 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
2880 k & 4 ? s->uvlinesize : s->linesize,
2886 v->vc1dsp.vc1_h_overlap(s->dest[0], s->linesize);
2887 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2888 if(!(s->flags & CODEC_FLAG_GRAY)) {
2889 v->vc1dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
2890 v->vc1dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
2893 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
2894 v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2895 if(!s->first_slice_line) {
2896 v->vc1dsp.vc1_v_overlap(s->dest[0], s->linesize);
2897 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
2898 if(!(s->flags & CODEC_FLAG_GRAY)) {
2899 v->vc1dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
2900 v->vc1dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
2903 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
2904 v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
2906 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
2908 if(get_bits_count(&s->gb) > v->bits) {
2909 ff_er_add_slice(s, 0, mby_start, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2910 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
2914 if (!v->s.loop_filter)
2915 ff_draw_horiz_band(s, s->mb_y * 16, 16);
2917 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
2918 s->first_slice_line = 0;
2920 if (v->s.loop_filter)
2921 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
2922 ff_er_add_slice(s, 0, mby_start, s->mb_width - 1, mby_end - 1, (AC_END|DC_END|MV_END));
2925 static void vc1_decode_p_blocks(VC1Context *v, int mby_start, int mby_end)
2927 MpegEncContext *s = &v->s;
2928 int apply_loop_filter;
2930 /* select codingmode used for VLC tables selection */
2931 switch(v->c_ac_table_index){
2933 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
2936 v->codingset = CS_HIGH_MOT_INTRA;
2939 v->codingset = CS_MID_RATE_INTRA;
2943 switch(v->c_ac_table_index){
2945 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
2948 v->codingset2 = CS_HIGH_MOT_INTER;
2951 v->codingset2 = CS_MID_RATE_INTER;
2955 apply_loop_filter = s->loop_filter && !(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY);
2956 s->first_slice_line = 1;
2957 memset(v->cbp_base, 0, sizeof(v->cbp_base[0])*2*s->mb_stride);
2958 for(s->mb_y = mby_start; s->mb_y < mby_end; s->mb_y++) {
2960 ff_init_block_index(s);
2961 for(; s->mb_x < s->mb_width; s->mb_x++) {
2962 ff_update_block_index(s);
2965 if (s->mb_y != mby_start && apply_loop_filter)
2966 vc1_apply_p_loop_filter(v);
2967 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
2968 ff_er_add_slice(s, 0, mby_start, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
2969 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);
2973 memmove(v->cbp_base, v->cbp, sizeof(v->cbp_base[0])*s->mb_stride);
2974 memmove(v->ttblk_base, v->ttblk, sizeof(v->ttblk_base[0])*s->mb_stride);
2975 memmove(v->is_intra_base, v->is_intra, sizeof(v->is_intra_base[0])*s->mb_stride);
2976 memmove(v->luma_mv_base, v->luma_mv, sizeof(v->luma_mv_base[0])*s->mb_stride);
2977 if (s->mb_y != mby_start) ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
2978 s->first_slice_line = 0;
2980 if (apply_loop_filter) {
2982 ff_init_block_index(s);
2983 for (; s->mb_x < s->mb_width; s->mb_x++) {
2984 ff_update_block_index(s);
2985 vc1_apply_p_loop_filter(v);
2988 if (mby_end >= mby_start)
2989 ff_draw_horiz_band(s, (mby_end-1) * 16, 16);
2990 ff_er_add_slice(s, 0, mby_start, s->mb_width - 1, mby_end - 1, (AC_END|DC_END|MV_END));
2993 static void vc1_decode_b_blocks(VC1Context *v, int mby_start, int mby_end)
2995 MpegEncContext *s = &v->s;
2997 /* select codingmode used for VLC tables selection */
2998 switch(v->c_ac_table_index){
3000 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
3003 v->codingset = CS_HIGH_MOT_INTRA;
3006 v->codingset = CS_MID_RATE_INTRA;
3010 switch(v->c_ac_table_index){
3012 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3015 v->codingset2 = CS_HIGH_MOT_INTER;
3018 v->codingset2 = CS_MID_RATE_INTER;
3022 s->first_slice_line = 1;
3023 for(s->mb_y = mby_start; s->mb_y < mby_end; s->mb_y++) {
3025 ff_init_block_index(s);
3026 for(; s->mb_x < s->mb_width; s->mb_x++) {
3027 ff_update_block_index(s);
3030 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
3031 ff_er_add_slice(s, 0, mby_start, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3032 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);
3035 if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq);
3037 if (!v->s.loop_filter)
3038 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3040 ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16);
3041 s->first_slice_line = 0;
3043 if (v->s.loop_filter)
3044 ff_draw_horiz_band(s, (s->mb_height-1)*16, 16);
3045 ff_er_add_slice(s, 0, mby_start, s->mb_width - 1, mby_end - 1, (AC_END|DC_END|MV_END));
3048 static void vc1_decode_skip_blocks(VC1Context *v)
3050 MpegEncContext *s = &v->s;
3052 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
3053 s->first_slice_line = 1;
3054 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
3056 ff_init_block_index(s);
3057 ff_update_block_index(s);
3058 memcpy(s->dest[0], s->last_picture.data[0] + s->mb_y * 16 * s->linesize, s->linesize * 16);
3059 memcpy(s->dest[1], s->last_picture.data[1] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
3060 memcpy(s->dest[2], s->last_picture.data[2] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
3061 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3062 s->first_slice_line = 0;
3064 s->pict_type = FF_P_TYPE;
3067 static void vc1_decode_blocks(VC1Context *v, int mby_start, int mby_end)
3070 v->s.esc3_level_length = 0;
3072 ff_intrax8_decode_picture(&v->x8, 2*v->pq+v->halfpq, v->pq*(!v->pquantizer) );
3074 switch(v->s.pict_type) {
3076 if(v->profile == PROFILE_ADVANCED)
3077 vc1_decode_i_blocks_adv(v, mby_start, mby_end);
3079 vc1_decode_i_blocks(v);
3082 if(v->p_frame_skipped)
3083 vc1_decode_skip_blocks(v);
3085 vc1_decode_p_blocks(v, mby_start, mby_end);
3089 if(v->profile == PROFILE_ADVANCED)
3090 vc1_decode_i_blocks_adv(v, mby_start, mby_end);
3092 vc1_decode_i_blocks(v);
3094 vc1_decode_b_blocks(v, mby_start, mby_end);
3100 static inline float get_float_val(GetBitContext* gb)
3102 return (float)get_bits_long(gb, 30) / (1<<15) - (1<<14);
3105 static void vc1_sprite_parse_transform(VC1Context *v, GetBitContext* gb, float c[7])
3109 switch (get_bits(gb, 2)) {
3112 c[2] = get_float_val(gb);
3116 c[0] = c[4] = get_float_val(gb);
3117 c[2] = get_float_val(gb);
3120 c[0] = get_float_val(gb);
3121 c[2] = get_float_val(gb);
3122 c[4] = get_float_val(gb);
3125 av_log_ask_for_sample(v->s.avctx, NULL);
3126 c[0] = get_float_val(gb);
3127 c[1] = get_float_val(gb);
3128 c[2] = get_float_val(gb);
3129 c[3] = get_float_val(gb);
3130 c[4] = get_float_val(gb);
3133 c[5] = get_float_val(gb);
3135 c[6] = get_float_val(gb);
3140 static void vc1_parse_sprites(VC1Context *v, GetBitContext* gb)
3142 int effect_type, effect_flag, effect_pcount1, effect_pcount2, i;
3143 float effect_params1[14], effect_params2[10];
3146 vc1_sprite_parse_transform(v, gb, coefs[0]);
3147 av_log(v->s.avctx, AV_LOG_DEBUG, "S1:");
3148 for (i = 0; i < 7; i++)
3149 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", coefs[0][i]);
3150 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3152 if (v->two_sprites) {
3153 vc1_sprite_parse_transform(v, gb, coefs[1]);
3154 av_log(v->s.avctx, AV_LOG_DEBUG, "S2:");
3155 for (i = 0; i < 7; i++)
3156 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", coefs[1][i]);
3157 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3160 if (effect_type = get_bits_long(gb, 30)){
3161 switch (effect_pcount1 = get_bits(gb, 4)) {
3163 effect_params1[0] = get_float_val(gb);
3164 effect_params1[1] = get_float_val(gb);
3167 vc1_sprite_parse_transform(v, gb, effect_params1);
3170 vc1_sprite_parse_transform(v, gb, effect_params1);
3171 vc1_sprite_parse_transform(v, gb, &effect_params1[7]);
3174 av_log_ask_for_sample(v->s.avctx, NULL);
3177 if (effect_type != 13 || effect_params1[0] != coefs[0][6]) {
3178 // effect 13 is simple alpha blending and matches the opacity above
3179 av_log(v->s.avctx, AV_LOG_DEBUG, "Effect: %d; params: ", effect_type);
3180 for (i = 0; i < effect_pcount1; i++)
3181 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", effect_params1[i]);
3182 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3185 effect_pcount2 = get_bits(gb, 16);
3186 if (effect_pcount2 > 10) {
3187 av_log(v->s.avctx, AV_LOG_ERROR, "Too many effect parameters\n");
3189 } else if (effect_pcount2) {
3191 av_log(v->s.avctx, AV_LOG_DEBUG, "Effect params 2: ");
3192 while (i < effect_pcount2){
3193 effect_params2[i] = get_float_val(gb);
3194 av_log(v->s.avctx, AV_LOG_DEBUG, " %.3f", effect_params2[i]);
3197 av_log(v->s.avctx, AV_LOG_DEBUG, "\n");
3200 if (effect_flag = get_bits1(gb))
3201 av_log(v->s.avctx, AV_LOG_DEBUG, "Effect flag set\n");
3203 if (get_bits_count(gb) >= gb->size_in_bits +
3204 (v->s.avctx->codec_id == CODEC_ID_WMV3 ? 64 : 0))
3205 av_log(v->s.avctx, AV_LOG_ERROR, "Buffer overrun\n");
3206 if (get_bits_count(gb) < gb->size_in_bits - 8)
3207 av_log(v->s.avctx, AV_LOG_WARNING, "Buffer not fully read\n");
3210 /** Initialize a VC1/WMV3 decoder
3211 * @todo TODO: Handle VC-1 IDUs (Transport level?)
3212 * @todo TODO: Decypher remaining bits in extra_data
3214 static av_cold int vc1_decode_init(AVCodecContext *avctx)
3216 VC1Context *v = avctx->priv_data;
3217 MpegEncContext *s = &v->s;
3221 if (!avctx->extradata_size || !avctx->extradata) return -1;
3222 if (!(avctx->flags & CODEC_FLAG_GRAY))
3223 avctx->pix_fmt = avctx->get_format(avctx, avctx->codec->pix_fmts);
3225 avctx->pix_fmt = PIX_FMT_GRAY8;
3226 avctx->hwaccel = ff_find_hwaccel(avctx->codec->id, avctx->pix_fmt);
3228 avctx->flags |= CODEC_FLAG_EMU_EDGE;
3229 v->s.flags |= CODEC_FLAG_EMU_EDGE;
3231 if(avctx->idct_algo==FF_IDCT_AUTO){
3232 avctx->idct_algo=FF_IDCT_WMV2;
3235 if(ff_msmpeg4_decode_init(avctx) < 0)
3237 if (vc1_init_common(v) < 0) return -1;
3238 ff_vc1dsp_init(&v->vc1dsp);
3240 avctx->coded_width = avctx->width;
3241 avctx->coded_height = avctx->height;
3242 if (avctx->codec_id == CODEC_ID_WMV3)
3246 // looks like WMV3 has a sequence header stored in the extradata
3247 // advanced sequence header may be before the first frame
3248 // the last byte of the extradata is a version number, 1 for the
3249 // samples we can decode
3251 init_get_bits(&gb, avctx->extradata, avctx->extradata_size*8);
3253 if (vc1_decode_sequence_header(avctx, v, &gb) < 0)
3256 count = avctx->extradata_size*8 - get_bits_count(&gb);
3259 av_log(avctx, AV_LOG_INFO, "Extra data: %i bits left, value: %X\n",
3260 count, get_bits(&gb, count));
3264 av_log(avctx, AV_LOG_INFO, "Read %i bits in overflow\n", -count);
3266 } else { // VC1/WVC1/WVP2
3267 const uint8_t *start = avctx->extradata;
3268 uint8_t *end = avctx->extradata + avctx->extradata_size;
3269 const uint8_t *next;
3270 int size, buf2_size;
3271 uint8_t *buf2 = NULL;
3272 int seq_initialized = 0, ep_initialized = 0;
3274 if(avctx->extradata_size < 16) {
3275 av_log(avctx, AV_LOG_ERROR, "Extradata size too small: %i\n", avctx->extradata_size);
3279 buf2 = av_mallocz(avctx->extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
3280 start = find_next_marker(start, end); // in WVC1 extradata first byte is its size, but can be 0 in mkv
3282 for(; next < end; start = next){
3283 next = find_next_marker(start + 4, end);
3284 size = next - start - 4;
3285 if(size <= 0) continue;
3286 buf2_size = vc1_unescape_buffer(start + 4, size, buf2);
3287 init_get_bits(&gb, buf2, buf2_size * 8);
3288 switch(AV_RB32(start)){
3289 case VC1_CODE_SEQHDR:
3290 if(vc1_decode_sequence_header(avctx, v, &gb) < 0){
3294 seq_initialized = 1;
3296 case VC1_CODE_ENTRYPOINT:
3297 if(vc1_decode_entry_point(avctx, v, &gb) < 0){
3306 if(!seq_initialized || !ep_initialized){
3307 av_log(avctx, AV_LOG_ERROR, "Incomplete extradata\n");
3310 v->res_sprite = (avctx->codec_tag == MKTAG('W','V','P','2'));
3312 avctx->profile = v->profile;
3313 if (v->profile == PROFILE_ADVANCED)
3314 avctx->level = v->level;
3316 avctx->has_b_frames= !!(avctx->max_b_frames);
3317 s->low_delay = !avctx->has_b_frames;
3319 s->mb_width = (avctx->coded_width+15)>>4;
3320 s->mb_height = (avctx->coded_height+15)>>4;
3322 if (v->profile == PROFILE_ADVANCED || v->res_fasttx) {
3323 for (i = 0; i < 64; i++) {
3324 #define transpose(x) ((x>>3) | ((x&7)<<3))
3325 v->zz_8x8[0][i] = transpose(wmv1_scantable[0][i]);
3326 v->zz_8x8[1][i] = transpose(wmv1_scantable[1][i]);
3327 v->zz_8x8[2][i] = transpose(wmv1_scantable[2][i]);
3328 v->zz_8x8[3][i] = transpose(wmv1_scantable[3][i]);
3333 memcpy(v->zz_8x8, wmv1_scantable, 4*64);
3338 /* Allocate mb bitplanes */
3339 v->mv_type_mb_plane = av_malloc(s->mb_stride * s->mb_height);
3340 v->direct_mb_plane = av_malloc(s->mb_stride * s->mb_height);
3341 v->acpred_plane = av_malloc(s->mb_stride * s->mb_height);
3342 v->over_flags_plane = av_malloc(s->mb_stride * s->mb_height);
3344 v->cbp_base = av_malloc(sizeof(v->cbp_base[0]) * 2 * s->mb_stride);
3345 v->cbp = v->cbp_base + s->mb_stride;
3346 v->ttblk_base = av_malloc(sizeof(v->ttblk_base[0]) * 2 * s->mb_stride);
3347 v->ttblk = v->ttblk_base + s->mb_stride;
3348 v->is_intra_base = av_malloc(sizeof(v->is_intra_base[0]) * 2 * s->mb_stride);
3349 v->is_intra = v->is_intra_base + s->mb_stride;
3350 v->luma_mv_base = av_malloc(sizeof(v->luma_mv_base[0]) * 2 * s->mb_stride);
3351 v->luma_mv = v->luma_mv_base + s->mb_stride;
3353 /* allocate block type info in that way so it could be used with s->block_index[] */
3354 v->mb_type_base = av_malloc(s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride * (s->mb_height + 1) * 2);
3355 v->mb_type[0] = v->mb_type_base + s->b8_stride + 1;
3356 v->mb_type[1] = v->mb_type_base + s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride + 1;
3357 v->mb_type[2] = v->mb_type[1] + s->mb_stride * (s->mb_height + 1);
3359 /* Init coded blocks info */
3360 if (v->profile == PROFILE_ADVANCED)
3362 // if (alloc_bitplane(&v->over_flags_plane, s->mb_width, s->mb_height) < 0)
3364 // if (alloc_bitplane(&v->ac_pred_plane, s->mb_width, s->mb_height) < 0)
3368 ff_intrax8_common_init(&v->x8,s);
3373 /** Decode a VC1/WMV3 frame
3374 * @todo TODO: Handle VC-1 IDUs (Transport level?)
3376 static int vc1_decode_frame(AVCodecContext *avctx,
3377 void *data, int *data_size,
3380 const uint8_t *buf = avpkt->data;
3381 int buf_size = avpkt->size, n_slices = 0, i;
3382 VC1Context *v = avctx->priv_data;
3383 MpegEncContext *s = &v->s;
3384 AVFrame *pict = data;
3385 uint8_t *buf2 = NULL;
3386 const uint8_t *buf_start = buf;
3393 /* no supplementary picture */
3394 if (buf_size == 0 || (buf_size == 4 && AV_RB32(buf) == VC1_CODE_ENDOFSEQ)) {
3395 /* special case for last picture */
3396 if (s->low_delay==0 && s->next_picture_ptr) {
3397 *pict= *(AVFrame*)s->next_picture_ptr;
3398 s->next_picture_ptr= NULL;
3400 *data_size = sizeof(AVFrame);
3406 /* We need to set current_picture_ptr before reading the header,
3407 * otherwise we cannot store anything in there. */
3408 if(s->current_picture_ptr==NULL || s->current_picture_ptr->data[0]){
3409 int i= ff_find_unused_picture(s, 0);
3410 s->current_picture_ptr= &s->picture[i];
3413 if (s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
3414 if (v->profile < PROFILE_ADVANCED)
3415 avctx->pix_fmt = PIX_FMT_VDPAU_WMV3;
3417 avctx->pix_fmt = PIX_FMT_VDPAU_VC1;
3420 //for advanced profile we may need to parse and unescape data
3421 if (avctx->codec_id == CODEC_ID_VC1) {
3423 buf2 = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
3425 if(IS_MARKER(AV_RB32(buf))){ /* frame starts with marker and needs to be parsed */
3426 const uint8_t *start, *end, *next;
3430 for(start = buf, end = buf + buf_size; next < end; start = next){
3431 next = find_next_marker(start + 4, end);
3432 size = next - start - 4;
3433 if(size <= 0) continue;
3434 switch(AV_RB32(start)){
3435 case VC1_CODE_FRAME:
3436 if (avctx->hwaccel ||
3437 s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3439 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
3441 case VC1_CODE_ENTRYPOINT: /* it should be before frame data */
3442 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
3443 init_get_bits(&s->gb, buf2, buf_size2*8);
3444 vc1_decode_entry_point(avctx, v, &s->gb);
3446 case VC1_CODE_SLICE: {
3448 slices = av_realloc(slices, sizeof(*slices) * (n_slices+1));
3449 if (!slices) goto err;
3450 slices[n_slices].buf = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
3451 if (!slices[n_slices].buf) goto err;
3452 buf_size3 = vc1_unescape_buffer(start + 4, size,
3453 slices[n_slices].buf);
3454 init_get_bits(&slices[n_slices].gb, slices[n_slices].buf,
3456 slices[n_slices].mby_start = get_bits(&slices[n_slices].gb, 9);
3462 }else if(v->interlace && ((buf[0] & 0xC0) == 0xC0)){ /* WVC1 interlaced stores both fields divided by marker */
3463 const uint8_t *divider;
3465 divider = find_next_marker(buf, buf + buf_size);
3466 if((divider == (buf + buf_size)) || AV_RB32(divider) != VC1_CODE_FIELD){
3467 av_log(avctx, AV_LOG_ERROR, "Error in WVC1 interlaced frame\n");
3471 buf_size2 = vc1_unescape_buffer(buf, divider - buf, buf2);
3473 if(!v->warn_interlaced++)
3474 av_log(v->s.avctx, AV_LOG_ERROR, "Interlaced WVC1 support is not implemented\n");
3477 buf_size2 = vc1_unescape_buffer(buf, buf_size, buf2);
3479 init_get_bits(&s->gb, buf2, buf_size2*8);
3481 init_get_bits(&s->gb, buf, buf_size*8);
3483 if (v->res_sprite) {
3484 v->new_sprite = !get_bits1(&s->gb);
3485 v->two_sprites = get_bits1(&s->gb);
3490 // do parse frame header
3491 if(v->profile < PROFILE_ADVANCED) {
3492 if(vc1_parse_frame_header(v, &s->gb) == -1) {
3496 if(vc1_parse_frame_header_adv(v, &s->gb) == -1) {
3501 if (v->res_sprite && s->pict_type!=FF_I_TYPE) {
3502 av_log(v->s.avctx, AV_LOG_WARNING, "Sprite decoder: expected I-frame\n");
3505 s->current_picture_ptr->repeat_pict = 0;
3507 s->current_picture_ptr->repeat_pict = 1;
3508 }else if (v->rptfrm){
3509 s->current_picture_ptr->repeat_pict = v->rptfrm * 2;
3512 s->current_picture_ptr->top_field_first = v->tff;
3514 // for skipping the frame
3515 s->current_picture.pict_type= s->pict_type;
3516 s->current_picture.key_frame= s->pict_type == FF_I_TYPE;
3518 /* skip B-frames if we don't have reference frames */
3519 if(s->last_picture_ptr==NULL && (s->pict_type==FF_B_TYPE || s->dropable)){
3523 /* skip b frames if we are in a hurry */
3524 if(avctx->hurry_up && s->pict_type==FF_B_TYPE) return -1;//buf_size;
3526 if( (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type==FF_B_TYPE)
3527 || (avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type!=FF_I_TYPE)
3528 || avctx->skip_frame >= AVDISCARD_ALL) {
3532 /* skip everything if we are in a hurry>=5 */
3533 if(avctx->hurry_up>=5) {
3538 if(s->next_p_frame_damaged){
3539 if(s->pict_type==FF_B_TYPE)
3542 s->next_p_frame_damaged=0;
3545 if(MPV_frame_start(s, avctx) < 0) {
3549 s->me.qpel_put= s->dsp.put_qpel_pixels_tab;
3550 s->me.qpel_avg= s->dsp.avg_qpel_pixels_tab;
3552 if ((CONFIG_VC1_VDPAU_DECODER)
3553 &&s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
3554 ff_vdpau_vc1_decode_picture(s, buf_start, (buf + buf_size) - buf_start);
3555 else if (avctx->hwaccel) {
3556 if (avctx->hwaccel->start_frame(avctx, buf, buf_size) < 0)
3558 if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
3560 if (avctx->hwaccel->end_frame(avctx) < 0)
3563 ff_er_frame_start(s);
3565 v->bits = buf_size * 8;
3566 for (i = 0; i <= n_slices; i++) {
3567 if (i && get_bits1(&s->gb))
3568 vc1_parse_frame_header_adv(v, &s->gb);
3569 vc1_decode_blocks(v, i == 0 ? 0 : FFMAX(0, slices[i-1].mby_start),
3570 i == n_slices ? s->mb_height : FFMIN(s->mb_height, slices[i].mby_start));
3571 if (i != n_slices) s->gb = slices[i].gb;
3573 //av_log(s->avctx, AV_LOG_INFO, "Consumed %i/%i bits\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3574 // if(get_bits_count(&s->gb) > buf_size * 8)
3581 assert(s->current_picture.pict_type == s->current_picture_ptr->pict_type);
3582 assert(s->current_picture.pict_type == s->pict_type);
3583 if (s->pict_type == FF_B_TYPE || s->low_delay) {
3584 *pict= *(AVFrame*)s->current_picture_ptr;
3585 } else if (s->last_picture_ptr != NULL) {
3586 *pict= *(AVFrame*)s->last_picture_ptr;
3589 if(s->last_picture_ptr || s->low_delay){
3590 *data_size = sizeof(AVFrame);
3591 ff_print_debug_info(s, pict);
3596 vc1_parse_sprites(v, &s->gb);
3598 for (i = 0; i < n_slices; i++)
3599 av_free(slices[i].buf);
3605 for (i = 0; i < n_slices; i++)
3606 av_free(slices[i].buf);
3612 /** Close a VC1/WMV3 decoder
3613 * @warning Initial try at using MpegEncContext stuff
3615 static av_cold int vc1_decode_end(AVCodecContext *avctx)
3617 VC1Context *v = avctx->priv_data;
3619 av_freep(&v->hrd_rate);
3620 av_freep(&v->hrd_buffer);
3621 MPV_common_end(&v->s);
3622 av_freep(&v->mv_type_mb_plane);
3623 av_freep(&v->direct_mb_plane);
3624 av_freep(&v->acpred_plane);
3625 av_freep(&v->over_flags_plane);
3626 av_freep(&v->mb_type_base);
3627 av_freep(&v->cbp_base);
3628 av_freep(&v->ttblk_base);
3629 av_freep(&v->is_intra_base); // FIXME use v->mb_type[]
3630 av_freep(&v->luma_mv_base);
3631 ff_intrax8_common_end(&v->x8);
3635 static const AVProfile profiles[] = {
3636 { FF_PROFILE_VC1_SIMPLE, "Simple" },
3637 { FF_PROFILE_VC1_MAIN, "Main" },
3638 { FF_PROFILE_VC1_COMPLEX, "Complex" },
3639 { FF_PROFILE_VC1_ADVANCED, "Advanced" },
3640 { FF_PROFILE_UNKNOWN },
3643 AVCodec ff_vc1_decoder = {
3652 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3654 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1"),
3655 .pix_fmts = ff_hwaccel_pixfmt_list_420,
3656 .profiles = NULL_IF_CONFIG_SMALL(profiles)
3659 #if CONFIG_WMV3_DECODER
3660 AVCodec ff_wmv3_decoder = {
3669 CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3671 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9"),
3672 .pix_fmts = ff_hwaccel_pixfmt_list_420,
3673 .profiles = NULL_IF_CONFIG_SMALL(profiles)
3677 #if CONFIG_WMV3_VDPAU_DECODER
3678 AVCodec ff_wmv3_vdpau_decoder = {
3687 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3689 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 VDPAU"),
3690 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_WMV3, PIX_FMT_NONE},
3691 .profiles = NULL_IF_CONFIG_SMALL(profiles)
3695 #if CONFIG_VC1_VDPAU_DECODER
3696 AVCodec ff_vc1_vdpau_decoder = {
3705 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3707 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1 VDPAU"),
3708 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_VC1, PIX_FMT_NONE},
3709 .profiles = NULL_IF_CONFIG_SMALL(profiles)