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
24 * @file libavcodec/vc1.c
25 * VC-1 and WMV3 decoder
31 #include "mpegvideo.h"
34 #include "vc1acdata.h"
35 #include "msmpeg4data.h"
37 #include "simple_idct.h"
39 #include "vdpau_internal.h"
44 #define MB_INTRA_VLC_BITS 9
47 static const uint16_t table_mb_intra[64][2];
51 * Init VC-1 specific tables and VC1Context members
52 * @param v The VC1Context to initialize
55 static int vc1_init_common(VC1Context *v)
60 v->hrd_rate = v->hrd_buffer = NULL;
66 init_vlc(&ff_vc1_bfraction_vlc, VC1_BFRACTION_VLC_BITS, 23,
67 ff_vc1_bfraction_bits, 1, 1,
68 ff_vc1_bfraction_codes, 1, 1, INIT_VLC_USE_STATIC);
69 init_vlc(&ff_vc1_norm2_vlc, VC1_NORM2_VLC_BITS, 4,
70 ff_vc1_norm2_bits, 1, 1,
71 ff_vc1_norm2_codes, 1, 1, INIT_VLC_USE_STATIC);
72 init_vlc(&ff_vc1_norm6_vlc, VC1_NORM6_VLC_BITS, 64,
73 ff_vc1_norm6_bits, 1, 1,
74 ff_vc1_norm6_codes, 2, 2, INIT_VLC_USE_STATIC);
75 init_vlc(&ff_vc1_imode_vlc, VC1_IMODE_VLC_BITS, 7,
76 ff_vc1_imode_bits, 1, 1,
77 ff_vc1_imode_codes, 1, 1, INIT_VLC_USE_STATIC);
80 init_vlc(&ff_vc1_ttmb_vlc[i], VC1_TTMB_VLC_BITS, 16,
81 ff_vc1_ttmb_bits[i], 1, 1,
82 ff_vc1_ttmb_codes[i], 2, 2, INIT_VLC_USE_STATIC);
83 init_vlc(&ff_vc1_ttblk_vlc[i], VC1_TTBLK_VLC_BITS, 8,
84 ff_vc1_ttblk_bits[i], 1, 1,
85 ff_vc1_ttblk_codes[i], 1, 1, INIT_VLC_USE_STATIC);
86 init_vlc(&ff_vc1_subblkpat_vlc[i], VC1_SUBBLKPAT_VLC_BITS, 15,
87 ff_vc1_subblkpat_bits[i], 1, 1,
88 ff_vc1_subblkpat_codes[i], 1, 1, INIT_VLC_USE_STATIC);
92 init_vlc(&ff_vc1_4mv_block_pattern_vlc[i], VC1_4MV_BLOCK_PATTERN_VLC_BITS, 16,
93 ff_vc1_4mv_block_pattern_bits[i], 1, 1,
94 ff_vc1_4mv_block_pattern_codes[i], 1, 1, INIT_VLC_USE_STATIC);
95 init_vlc(&ff_vc1_cbpcy_p_vlc[i], VC1_CBPCY_P_VLC_BITS, 64,
96 ff_vc1_cbpcy_p_bits[i], 1, 1,
97 ff_vc1_cbpcy_p_codes[i], 2, 2, INIT_VLC_USE_STATIC);
98 init_vlc(&ff_vc1_mv_diff_vlc[i], VC1_MV_DIFF_VLC_BITS, 73,
99 ff_vc1_mv_diff_bits[i], 1, 1,
100 ff_vc1_mv_diff_codes[i], 2, 2, INIT_VLC_USE_STATIC);
103 init_vlc(&ff_vc1_ac_coeff_table[i], AC_VLC_BITS, vc1_ac_sizes[i],
104 &vc1_ac_tables[i][0][1], 8, 4,
105 &vc1_ac_tables[i][0][0], 8, 4, INIT_VLC_USE_STATIC);
106 init_vlc(&ff_msmp4_mb_i_vlc, MB_INTRA_VLC_BITS, 64,
107 &ff_msmp4_mb_i_table[0][1], 4, 2,
108 &ff_msmp4_mb_i_table[0][0], 4, 2, INIT_VLC_USE_STATIC);
113 v->mvrange = 0; /* 7.1.1.18, p80 */
118 /***********************************************************************/
120 * @defgroup vc1bitplane VC-1 Bitplane decoding
138 /** @} */ //imode defines
140 /** Decode rows by checking if they are skipped
141 * @param plane Buffer to store decoded bits
142 * @param[in] width Width of this buffer
143 * @param[in] height Height of this buffer
144 * @param[in] stride of this buffer
146 static void decode_rowskip(uint8_t* plane, int width, int height, int stride, GetBitContext *gb){
149 for (y=0; y<height; y++){
150 if (!get_bits1(gb)) //rowskip
151 memset(plane, 0, width);
153 for (x=0; x<width; x++)
154 plane[x] = get_bits1(gb);
159 /** Decode columns by checking if they are skipped
160 * @param plane Buffer to store decoded bits
161 * @param[in] width Width of this buffer
162 * @param[in] height Height of this buffer
163 * @param[in] stride of this buffer
164 * @todo FIXME: Optimize
166 static void decode_colskip(uint8_t* plane, int width, int height, int stride, GetBitContext *gb){
169 for (x=0; x<width; x++){
170 if (!get_bits1(gb)) //colskip
171 for (y=0; y<height; y++)
174 for (y=0; y<height; y++)
175 plane[y*stride] = get_bits1(gb);
180 /** Decode a bitplane's bits
181 * @param data bitplane where to store the decode bits
182 * @param[out] raw_flag pointer to the flag indicating that this bitplane is not coded explicitly
183 * @param v VC-1 context for bit reading and logging
185 * @todo FIXME: Optimize
187 static int bitplane_decoding(uint8_t* data, int *raw_flag, VC1Context *v)
189 GetBitContext *gb = &v->s.gb;
191 int imode, x, y, code, offset;
192 uint8_t invert, *planep = data;
193 int width, height, stride;
195 width = v->s.mb_width;
196 height = v->s.mb_height;
197 stride = v->s.mb_stride;
198 invert = get_bits1(gb);
199 imode = get_vlc2(gb, ff_vc1_imode_vlc.table, VC1_IMODE_VLC_BITS, 1);
205 //Data is actually read in the MB layer (same for all tests == "raw")
206 *raw_flag = 1; //invert ignored
210 if ((height * width) & 1)
212 *planep++ = get_bits1(gb);
216 // decode bitplane as one long line
217 for (y = offset; y < height * width; y += 2) {
218 code = get_vlc2(gb, ff_vc1_norm2_vlc.table, VC1_NORM2_VLC_BITS, 1);
219 *planep++ = code & 1;
221 if(offset == width) {
223 planep += stride - width;
225 *planep++ = code >> 1;
227 if(offset == width) {
229 planep += stride - width;
235 if(!(height % 3) && (width % 3)) { // use 2x3 decoding
236 for(y = 0; y < height; y+= 3) {
237 for(x = width & 1; x < width; x += 2) {
238 code = get_vlc2(gb, ff_vc1_norm6_vlc.table, VC1_NORM6_VLC_BITS, 2);
240 av_log(v->s.avctx, AV_LOG_DEBUG, "invalid NORM-6 VLC\n");
243 planep[x + 0] = (code >> 0) & 1;
244 planep[x + 1] = (code >> 1) & 1;
245 planep[x + 0 + stride] = (code >> 2) & 1;
246 planep[x + 1 + stride] = (code >> 3) & 1;
247 planep[x + 0 + stride * 2] = (code >> 4) & 1;
248 planep[x + 1 + stride * 2] = (code >> 5) & 1;
250 planep += stride * 3;
252 if(width & 1) decode_colskip(data, 1, height, stride, &v->s.gb);
254 planep += (height & 1) * stride;
255 for(y = height & 1; y < height; y += 2) {
256 for(x = width % 3; x < width; x += 3) {
257 code = get_vlc2(gb, ff_vc1_norm6_vlc.table, VC1_NORM6_VLC_BITS, 2);
259 av_log(v->s.avctx, AV_LOG_DEBUG, "invalid NORM-6 VLC\n");
262 planep[x + 0] = (code >> 0) & 1;
263 planep[x + 1] = (code >> 1) & 1;
264 planep[x + 2] = (code >> 2) & 1;
265 planep[x + 0 + stride] = (code >> 3) & 1;
266 planep[x + 1 + stride] = (code >> 4) & 1;
267 planep[x + 2 + stride] = (code >> 5) & 1;
269 planep += stride * 2;
272 if(x) decode_colskip(data , x, height , stride, &v->s.gb);
273 if(height & 1) decode_rowskip(data+x, width - x, 1, stride, &v->s.gb);
277 decode_rowskip(data, width, height, stride, &v->s.gb);
280 decode_colskip(data, width, height, stride, &v->s.gb);
285 /* Applying diff operator */
286 if (imode == IMODE_DIFF2 || imode == IMODE_DIFF6)
290 for (x=1; x<width; x++)
291 planep[x] ^= planep[x-1];
292 for (y=1; y<height; y++)
295 planep[0] ^= planep[-stride];
296 for (x=1; x<width; x++)
298 if (planep[x-1] != planep[x-stride]) planep[x] ^= invert;
299 else planep[x] ^= planep[x-1];
306 for (x=0; x<stride*height; x++) planep[x] = !planep[x]; //FIXME stride
308 return (imode<<1) + invert;
311 /** @} */ //Bitplane group
314 * VC-1 in-loop deblocking filter for one line
315 * @param src source block type
316 * @param stride block stride
317 * @param pq block quantizer
318 * @return whether other 3 pairs should be filtered or not
321 static av_always_inline int vc1_filter_line(uint8_t* src, int stride, int pq){
322 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
324 int a0 = (2*(src[-2*stride] - src[ 1*stride]) - 5*(src[-1*stride] - src[ 0*stride]) + 4) >> 3;
325 int a0_sign = a0 >> 31; /* Store sign */
326 a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */
328 int a1 = FFABS((2*(src[-4*stride] - src[-1*stride]) - 5*(src[-3*stride] - src[-2*stride]) + 4) >> 3);
329 int a2 = FFABS((2*(src[ 0*stride] - src[ 3*stride]) - 5*(src[ 1*stride] - src[ 2*stride]) + 4) >> 3);
330 if(a1 < a0 || a2 < a0){
331 int clip = src[-1*stride] - src[ 0*stride];
332 int clip_sign = clip >> 31;
333 clip = ((clip ^ clip_sign) - clip_sign)>>1;
335 int a3 = FFMIN(a1, a2);
336 int d = 5 * (a3 - a0);
337 int d_sign = (d >> 31);
338 d = ((d ^ d_sign) - d_sign) >> 3;
341 if( d_sign ^ clip_sign )
345 d = (d ^ d_sign) - d_sign; /* Restore sign */
346 src[-1*stride] = cm[src[-1*stride] - d];
347 src[ 0*stride] = cm[src[ 0*stride] + d];
357 * VC-1 in-loop deblocking filter
358 * @param src source block type
359 * @param step distance between horizontally adjacent elements
360 * @param stride distance between vertically adjacent elements
361 * @param len edge length to filter (4 or 8 pixels)
362 * @param pq block quantizer
365 static void vc1_loop_filter(uint8_t* src, int step, int stride, int len, int pq)
370 for(i = 0; i < len; i += 4){
371 filt3 = vc1_filter_line(src + 2*step, stride, pq);
373 vc1_filter_line(src + 0*step, stride, pq);
374 vc1_filter_line(src + 1*step, stride, pq);
375 vc1_filter_line(src + 3*step, stride, pq);
381 static void vc1_loop_filter_iblk(MpegEncContext *s, int pq)
384 if(!s->first_slice_line)
385 vc1_loop_filter(s->dest[0], 1, s->linesize, 16, pq);
386 vc1_loop_filter(s->dest[0] + 8*s->linesize, 1, s->linesize, 16, pq);
387 for(i = !s->mb_x*8; i < 16; i += 8)
388 vc1_loop_filter(s->dest[0] + i, s->linesize, 1, 16, pq);
389 for(j = 0; j < 2; j++){
390 if(!s->first_slice_line)
391 vc1_loop_filter(s->dest[j+1], 1, s->uvlinesize, 8, pq);
393 vc1_loop_filter(s->dest[j+1], s->uvlinesize, 1, 8, pq);
397 /***********************************************************************/
398 /** VOP Dquant decoding
399 * @param v VC-1 Context
401 static int vop_dquant_decoding(VC1Context *v)
403 GetBitContext *gb = &v->s.gb;
409 pqdiff = get_bits(gb, 3);
410 if (pqdiff == 7) v->altpq = get_bits(gb, 5);
411 else v->altpq = v->pq + pqdiff + 1;
415 v->dquantfrm = get_bits1(gb);
418 v->dqprofile = get_bits(gb, 2);
419 switch (v->dqprofile)
421 case DQPROFILE_SINGLE_EDGE:
422 case DQPROFILE_DOUBLE_EDGES:
423 v->dqsbedge = get_bits(gb, 2);
425 case DQPROFILE_ALL_MBS:
426 v->dqbilevel = get_bits1(gb);
429 default: break; //Forbidden ?
431 if (v->dqbilevel || v->dqprofile != DQPROFILE_ALL_MBS)
433 pqdiff = get_bits(gb, 3);
434 if (pqdiff == 7) v->altpq = get_bits(gb, 5);
435 else v->altpq = v->pq + pqdiff + 1;
442 /** Put block onto picture
444 static void vc1_put_block(VC1Context *v, DCTELEM block[6][64])
448 DSPContext *dsp = &v->s.dsp;
452 for(k = 0; k < 6; k++)
453 for(j = 0; j < 8; j++)
454 for(i = 0; i < 8; i++)
455 block[k][i + j*8] = ((block[k][i + j*8] - 128) << 1) + 128;
458 ys = v->s.current_picture.linesize[0];
459 us = v->s.current_picture.linesize[1];
460 vs = v->s.current_picture.linesize[2];
463 dsp->put_pixels_clamped(block[0], Y, ys);
464 dsp->put_pixels_clamped(block[1], Y + 8, ys);
466 dsp->put_pixels_clamped(block[2], Y, ys);
467 dsp->put_pixels_clamped(block[3], Y + 8, ys);
469 if(!(v->s.flags & CODEC_FLAG_GRAY)) {
470 dsp->put_pixels_clamped(block[4], v->s.dest[1], us);
471 dsp->put_pixels_clamped(block[5], v->s.dest[2], vs);
475 /** Do motion compensation over 1 macroblock
476 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
478 static void vc1_mc_1mv(VC1Context *v, int dir)
480 MpegEncContext *s = &v->s;
481 DSPContext *dsp = &v->s.dsp;
482 uint8_t *srcY, *srcU, *srcV;
483 int dxy, uvdxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
485 if(!v->s.last_picture.data[0])return;
487 mx = s->mv[dir][0][0];
488 my = s->mv[dir][0][1];
490 // store motion vectors for further use in B frames
491 if(s->pict_type == FF_P_TYPE) {
492 s->current_picture.motion_val[1][s->block_index[0]][0] = mx;
493 s->current_picture.motion_val[1][s->block_index[0]][1] = my;
495 uvmx = (mx + ((mx & 3) == 3)) >> 1;
496 uvmy = (my + ((my & 3) == 3)) >> 1;
498 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
499 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
502 srcY = s->last_picture.data[0];
503 srcU = s->last_picture.data[1];
504 srcV = s->last_picture.data[2];
506 srcY = s->next_picture.data[0];
507 srcU = s->next_picture.data[1];
508 srcV = s->next_picture.data[2];
511 src_x = s->mb_x * 16 + (mx >> 2);
512 src_y = s->mb_y * 16 + (my >> 2);
513 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
514 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
516 if(v->profile != PROFILE_ADVANCED){
517 src_x = av_clip( src_x, -16, s->mb_width * 16);
518 src_y = av_clip( src_y, -16, s->mb_height * 16);
519 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
520 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
522 src_x = av_clip( src_x, -17, s->avctx->coded_width);
523 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
524 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
525 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
528 srcY += src_y * s->linesize + src_x;
529 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
530 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
532 /* for grayscale we should not try to read from unknown area */
533 if(s->flags & CODEC_FLAG_GRAY) {
534 srcU = s->edge_emu_buffer + 18 * s->linesize;
535 srcV = s->edge_emu_buffer + 18 * s->linesize;
538 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
539 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
540 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
541 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
543 srcY -= s->mspel * (1 + s->linesize);
544 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
545 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
546 srcY = s->edge_emu_buffer;
547 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
548 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
549 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
550 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
553 /* if we deal with range reduction we need to scale source blocks */
559 for(j = 0; j < 17 + s->mspel*2; j++) {
560 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
563 src = srcU; src2 = srcV;
564 for(j = 0; j < 9; j++) {
565 for(i = 0; i < 9; i++) {
566 src[i] = ((src[i] - 128) >> 1) + 128;
567 src2[i] = ((src2[i] - 128) >> 1) + 128;
569 src += s->uvlinesize;
570 src2 += s->uvlinesize;
573 /* if we deal with intensity compensation we need to scale source blocks */
574 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
579 for(j = 0; j < 17 + s->mspel*2; j++) {
580 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = v->luty[src[i]];
583 src = srcU; src2 = srcV;
584 for(j = 0; j < 9; j++) {
585 for(i = 0; i < 9; i++) {
586 src[i] = v->lutuv[src[i]];
587 src2[i] = v->lutuv[src2[i]];
589 src += s->uvlinesize;
590 src2 += s->uvlinesize;
593 srcY += s->mspel * (1 + s->linesize);
597 dxy = ((my & 3) << 2) | (mx & 3);
598 dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
599 dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
600 srcY += s->linesize * 8;
601 dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
602 dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
603 } else { // hpel mc - always used for luma
604 dxy = (my & 2) | ((mx & 2) >> 1);
607 dsp->put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
609 dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
612 if(s->flags & CODEC_FLAG_GRAY) return;
613 /* Chroma MC always uses qpel bilinear */
614 uvdxy = ((uvmy & 3) << 2) | (uvmx & 3);
618 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
619 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
621 dsp->put_no_rnd_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
622 dsp->put_no_rnd_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
626 /** Do motion compensation for 4-MV macroblock - luminance block
628 static void vc1_mc_4mv_luma(VC1Context *v, int n)
630 MpegEncContext *s = &v->s;
631 DSPContext *dsp = &v->s.dsp;
633 int dxy, mx, my, src_x, src_y;
636 if(!v->s.last_picture.data[0])return;
639 srcY = s->last_picture.data[0];
641 off = s->linesize * 4 * (n&2) + (n&1) * 8;
643 src_x = s->mb_x * 16 + (n&1) * 8 + (mx >> 2);
644 src_y = s->mb_y * 16 + (n&2) * 4 + (my >> 2);
646 if(v->profile != PROFILE_ADVANCED){
647 src_x = av_clip( src_x, -16, s->mb_width * 16);
648 src_y = av_clip( src_y, -16, s->mb_height * 16);
650 src_x = av_clip( src_x, -17, s->avctx->coded_width);
651 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
654 srcY += src_y * s->linesize + src_x;
656 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
657 || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 8 - s->mspel*2
658 || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 8 - s->mspel*2){
659 srcY -= s->mspel * (1 + s->linesize);
660 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 9+s->mspel*2, 9+s->mspel*2,
661 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
662 srcY = s->edge_emu_buffer;
663 /* if we deal with range reduction we need to scale source blocks */
669 for(j = 0; j < 9 + s->mspel*2; j++) {
670 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
674 /* if we deal with intensity compensation we need to scale source blocks */
675 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
680 for(j = 0; j < 9 + s->mspel*2; j++) {
681 for(i = 0; i < 9 + s->mspel*2; i++) src[i] = v->luty[src[i]];
685 srcY += s->mspel * (1 + s->linesize);
689 dxy = ((my & 3) << 2) | (mx & 3);
690 dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, v->rnd);
691 } else { // hpel mc - always used for luma
692 dxy = (my & 2) | ((mx & 2) >> 1);
694 dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
696 dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
700 static inline int median4(int a, int b, int c, int d)
703 if(c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
704 else return (FFMIN(b, c) + FFMAX(a, d)) / 2;
706 if(c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
707 else return (FFMIN(a, c) + FFMAX(b, d)) / 2;
712 /** Do motion compensation for 4-MV macroblock - both chroma blocks
714 static void vc1_mc_4mv_chroma(VC1Context *v)
716 MpegEncContext *s = &v->s;
717 DSPContext *dsp = &v->s.dsp;
718 uint8_t *srcU, *srcV;
719 int uvdxy, uvmx, uvmy, uvsrc_x, uvsrc_y;
720 int i, idx, tx = 0, ty = 0;
721 int mvx[4], mvy[4], intra[4];
722 static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
724 if(!v->s.last_picture.data[0])return;
725 if(s->flags & CODEC_FLAG_GRAY) return;
727 for(i = 0; i < 4; i++) {
728 mvx[i] = s->mv[0][i][0];
729 mvy[i] = s->mv[0][i][1];
730 intra[i] = v->mb_type[0][s->block_index[i]];
733 /* calculate chroma MV vector from four luma MVs */
734 idx = (intra[3] << 3) | (intra[2] << 2) | (intra[1] << 1) | intra[0];
735 if(!idx) { // all blocks are inter
736 tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
737 ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
738 } else if(count[idx] == 1) { // 3 inter blocks
741 tx = mid_pred(mvx[1], mvx[2], mvx[3]);
742 ty = mid_pred(mvy[1], mvy[2], mvy[3]);
745 tx = mid_pred(mvx[0], mvx[2], mvx[3]);
746 ty = mid_pred(mvy[0], mvy[2], mvy[3]);
749 tx = mid_pred(mvx[0], mvx[1], mvx[3]);
750 ty = mid_pred(mvy[0], mvy[1], mvy[3]);
753 tx = mid_pred(mvx[0], mvx[1], mvx[2]);
754 ty = mid_pred(mvy[0], mvy[1], mvy[2]);
757 } else if(count[idx] == 2) {
759 for(i=0; i<3;i++) if(!intra[i]) {t1 = i; break;}
760 for(i= t1+1; i<4; i++)if(!intra[i]) {t2 = i; break;}
761 tx = (mvx[t1] + mvx[t2]) / 2;
762 ty = (mvy[t1] + mvy[t2]) / 2;
764 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
765 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
766 return; //no need to do MC for inter blocks
769 s->current_picture.motion_val[1][s->block_index[0]][0] = tx;
770 s->current_picture.motion_val[1][s->block_index[0]][1] = ty;
771 uvmx = (tx + ((tx&3) == 3)) >> 1;
772 uvmy = (ty + ((ty&3) == 3)) >> 1;
774 uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
775 uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
778 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
779 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
781 if(v->profile != PROFILE_ADVANCED){
782 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
783 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
785 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
786 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
789 srcU = s->last_picture.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
790 srcV = s->last_picture.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
791 if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
792 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
793 || (unsigned)uvsrc_y > (s->v_edge_pos >> 1) - 9){
794 ff_emulated_edge_mc(s->edge_emu_buffer , srcU, s->uvlinesize, 8+1, 8+1,
795 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
796 ff_emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize, 8+1, 8+1,
797 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
798 srcU = s->edge_emu_buffer;
799 srcV = s->edge_emu_buffer + 16;
801 /* if we deal with range reduction we need to scale source blocks */
806 src = srcU; src2 = srcV;
807 for(j = 0; j < 9; j++) {
808 for(i = 0; i < 9; i++) {
809 src[i] = ((src[i] - 128) >> 1) + 128;
810 src2[i] = ((src2[i] - 128) >> 1) + 128;
812 src += s->uvlinesize;
813 src2 += s->uvlinesize;
816 /* if we deal with intensity compensation we need to scale source blocks */
817 if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
821 src = srcU; src2 = srcV;
822 for(j = 0; j < 9; j++) {
823 for(i = 0; i < 9; i++) {
824 src[i] = v->lutuv[src[i]];
825 src2[i] = v->lutuv[src2[i]];
827 src += s->uvlinesize;
828 src2 += s->uvlinesize;
833 /* Chroma MC always uses qpel bilinear */
834 uvdxy = ((uvmy & 3) << 2) | (uvmx & 3);
838 dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
839 dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
841 dsp->put_no_rnd_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
842 dsp->put_no_rnd_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
846 static int decode_sequence_header_adv(VC1Context *v, GetBitContext *gb);
849 * Decode Simple/Main Profiles sequence header
850 * @see Figure 7-8, p16-17
851 * @param avctx Codec context
852 * @param gb GetBit context initialized from Codec context extra_data
855 static int decode_sequence_header(AVCodecContext *avctx, GetBitContext *gb)
857 VC1Context *v = avctx->priv_data;
859 av_log(avctx, AV_LOG_DEBUG, "Header: %0X\n", show_bits(gb, 32));
860 v->profile = get_bits(gb, 2);
861 if (v->profile == PROFILE_COMPLEX)
863 av_log(avctx, AV_LOG_ERROR, "WMV3 Complex Profile is not fully supported\n");
866 if (v->profile == PROFILE_ADVANCED)
868 v->zz_8x4 = ff_vc1_adv_progressive_8x4_zz;
869 v->zz_4x8 = ff_vc1_adv_progressive_4x8_zz;
870 return decode_sequence_header_adv(v, gb);
874 v->zz_8x4 = wmv2_scantableA;
875 v->zz_4x8 = wmv2_scantableB;
876 v->res_sm = get_bits(gb, 2); //reserved
879 av_log(avctx, AV_LOG_ERROR,
880 "Reserved RES_SM=%i is forbidden\n", v->res_sm);
886 v->frmrtq_postproc = get_bits(gb, 3); //common
887 // (bitrate-32kbps)/64kbps
888 v->bitrtq_postproc = get_bits(gb, 5); //common
889 v->s.loop_filter = get_bits1(gb); //common
890 if(v->s.loop_filter == 1 && v->profile == PROFILE_SIMPLE)
892 av_log(avctx, AV_LOG_ERROR,
893 "LOOPFILTER shell not be enabled in simple profile\n");
895 if(v->s.avctx->skip_loop_filter >= AVDISCARD_ALL)
896 v->s.loop_filter = 0;
898 v->res_x8 = get_bits1(gb); //reserved
899 v->multires = get_bits1(gb);
900 v->res_fasttx = get_bits1(gb);
903 v->s.dsp.vc1_inv_trans_8x8 = ff_simple_idct;
904 v->s.dsp.vc1_inv_trans_8x4 = ff_simple_idct84_add;
905 v->s.dsp.vc1_inv_trans_4x8 = ff_simple_idct48_add;
906 v->s.dsp.vc1_inv_trans_4x4 = ff_simple_idct44_add;
909 v->fastuvmc = get_bits1(gb); //common
910 if (!v->profile && !v->fastuvmc)
912 av_log(avctx, AV_LOG_ERROR,
913 "FASTUVMC unavailable in Simple Profile\n");
916 v->extended_mv = get_bits1(gb); //common
917 if (!v->profile && v->extended_mv)
919 av_log(avctx, AV_LOG_ERROR,
920 "Extended MVs unavailable in Simple Profile\n");
923 v->dquant = get_bits(gb, 2); //common
924 v->vstransform = get_bits1(gb); //common
926 v->res_transtab = get_bits1(gb);
929 av_log(avctx, AV_LOG_ERROR,
930 "1 for reserved RES_TRANSTAB is forbidden\n");
934 v->overlap = get_bits1(gb); //common
936 v->s.resync_marker = get_bits1(gb);
937 v->rangered = get_bits1(gb);
938 if (v->rangered && v->profile == PROFILE_SIMPLE)
940 av_log(avctx, AV_LOG_INFO,
941 "RANGERED should be set to 0 in simple profile\n");
944 v->s.max_b_frames = avctx->max_b_frames = get_bits(gb, 3); //common
945 v->quantizer_mode = get_bits(gb, 2); //common
947 v->finterpflag = get_bits1(gb); //common
948 v->res_rtm_flag = get_bits1(gb); //reserved
949 if (!v->res_rtm_flag)
951 // av_log(avctx, AV_LOG_ERROR,
952 // "0 for reserved RES_RTM_FLAG is forbidden\n");
953 av_log(avctx, AV_LOG_ERROR,
954 "Old WMV3 version detected, only I-frames will be decoded\n");
957 //TODO: figure out what they mean (always 0x402F)
958 if(!v->res_fasttx) skip_bits(gb, 16);
959 av_log(avctx, AV_LOG_DEBUG,
960 "Profile %i:\nfrmrtq_postproc=%i, bitrtq_postproc=%i\n"
961 "LoopFilter=%i, MultiRes=%i, FastUVMC=%i, Extended MV=%i\n"
962 "Rangered=%i, VSTransform=%i, Overlap=%i, SyncMarker=%i\n"
963 "DQuant=%i, Quantizer mode=%i, Max B frames=%i\n",
964 v->profile, v->frmrtq_postproc, v->bitrtq_postproc,
965 v->s.loop_filter, v->multires, v->fastuvmc, v->extended_mv,
966 v->rangered, v->vstransform, v->overlap, v->s.resync_marker,
967 v->dquant, v->quantizer_mode, avctx->max_b_frames
972 static int decode_sequence_header_adv(VC1Context *v, GetBitContext *gb)
975 v->level = get_bits(gb, 3);
978 av_log(v->s.avctx, AV_LOG_ERROR, "Reserved LEVEL %i\n",v->level);
980 v->chromaformat = get_bits(gb, 2);
981 if (v->chromaformat != 1)
983 av_log(v->s.avctx, AV_LOG_ERROR,
984 "Only 4:2:0 chroma format supported\n");
989 v->frmrtq_postproc = get_bits(gb, 3); //common
990 // (bitrate-32kbps)/64kbps
991 v->bitrtq_postproc = get_bits(gb, 5); //common
992 v->postprocflag = get_bits1(gb); //common
994 v->s.avctx->coded_width = (get_bits(gb, 12) + 1) << 1;
995 v->s.avctx->coded_height = (get_bits(gb, 12) + 1) << 1;
996 v->s.avctx->width = v->s.avctx->coded_width;
997 v->s.avctx->height = v->s.avctx->coded_height;
998 v->broadcast = get_bits1(gb);
999 v->interlace = get_bits1(gb);
1000 v->tfcntrflag = get_bits1(gb);
1001 v->finterpflag = get_bits1(gb);
1002 skip_bits1(gb); // reserved
1004 v->s.h_edge_pos = v->s.avctx->coded_width;
1005 v->s.v_edge_pos = v->s.avctx->coded_height;
1007 av_log(v->s.avctx, AV_LOG_DEBUG,
1008 "Advanced Profile level %i:\nfrmrtq_postproc=%i, bitrtq_postproc=%i\n"
1009 "LoopFilter=%i, ChromaFormat=%i, Pulldown=%i, Interlace: %i\n"
1010 "TFCTRflag=%i, FINTERPflag=%i\n",
1011 v->level, v->frmrtq_postproc, v->bitrtq_postproc,
1012 v->s.loop_filter, v->chromaformat, v->broadcast, v->interlace,
1013 v->tfcntrflag, v->finterpflag
1016 v->psf = get_bits1(gb);
1017 if(v->psf) { //PsF, 6.1.13
1018 av_log(v->s.avctx, AV_LOG_ERROR, "Progressive Segmented Frame mode: not supported (yet)\n");
1021 v->s.max_b_frames = v->s.avctx->max_b_frames = 7;
1022 if(get_bits1(gb)) { //Display Info - decoding is not affected by it
1024 av_log(v->s.avctx, AV_LOG_DEBUG, "Display extended info:\n");
1025 v->s.avctx->coded_width = w = get_bits(gb, 14) + 1;
1026 v->s.avctx->coded_height = h = get_bits(gb, 14) + 1;
1027 av_log(v->s.avctx, AV_LOG_DEBUG, "Display dimensions: %ix%i\n", w, h);
1029 ar = get_bits(gb, 4);
1031 v->s.avctx->sample_aspect_ratio = ff_vc1_pixel_aspect[ar];
1033 w = get_bits(gb, 8);
1034 h = get_bits(gb, 8);
1035 v->s.avctx->sample_aspect_ratio = (AVRational){w, h};
1037 av_log(v->s.avctx, AV_LOG_DEBUG, "Aspect: %i:%i\n", v->s.avctx->sample_aspect_ratio.num, v->s.avctx->sample_aspect_ratio.den);
1039 if(get_bits1(gb)){ //framerate stuff
1041 v->s.avctx->time_base.num = 32;
1042 v->s.avctx->time_base.den = get_bits(gb, 16) + 1;
1045 nr = get_bits(gb, 8);
1046 dr = get_bits(gb, 4);
1047 if(nr && nr < 8 && dr && dr < 3){
1048 v->s.avctx->time_base.num = ff_vc1_fps_dr[dr - 1];
1049 v->s.avctx->time_base.den = ff_vc1_fps_nr[nr - 1] * 1000;
1055 v->color_prim = get_bits(gb, 8);
1056 v->transfer_char = get_bits(gb, 8);
1057 v->matrix_coef = get_bits(gb, 8);
1061 v->hrd_param_flag = get_bits1(gb);
1062 if(v->hrd_param_flag) {
1064 v->hrd_num_leaky_buckets = get_bits(gb, 5);
1065 skip_bits(gb, 4); //bitrate exponent
1066 skip_bits(gb, 4); //buffer size exponent
1067 for(i = 0; i < v->hrd_num_leaky_buckets; i++) {
1068 skip_bits(gb, 16); //hrd_rate[n]
1069 skip_bits(gb, 16); //hrd_buffer[n]
1075 static int decode_entry_point(AVCodecContext *avctx, GetBitContext *gb)
1077 VC1Context *v = avctx->priv_data;
1080 av_log(avctx, AV_LOG_DEBUG, "Entry point: %08X\n", show_bits_long(gb, 32));
1081 v->broken_link = get_bits1(gb);
1082 v->closed_entry = get_bits1(gb);
1083 v->panscanflag = get_bits1(gb);
1084 v->refdist_flag = get_bits1(gb);
1085 v->s.loop_filter = get_bits1(gb);
1086 v->fastuvmc = get_bits1(gb);
1087 v->extended_mv = get_bits1(gb);
1088 v->dquant = get_bits(gb, 2);
1089 v->vstransform = get_bits1(gb);
1090 v->overlap = get_bits1(gb);
1091 v->quantizer_mode = get_bits(gb, 2);
1093 if(v->hrd_param_flag){
1094 for(i = 0; i < v->hrd_num_leaky_buckets; i++) {
1095 skip_bits(gb, 8); //hrd_full[n]
1100 avctx->coded_width = (get_bits(gb, 12)+1)<<1;
1101 avctx->coded_height = (get_bits(gb, 12)+1)<<1;
1104 v->extended_dmv = get_bits1(gb);
1105 if((v->range_mapy_flag = get_bits1(gb))) {
1106 av_log(avctx, AV_LOG_ERROR, "Luma scaling is not supported, expect wrong picture\n");
1107 v->range_mapy = get_bits(gb, 3);
1109 if((v->range_mapuv_flag = get_bits1(gb))) {
1110 av_log(avctx, AV_LOG_ERROR, "Chroma scaling is not supported, expect wrong picture\n");
1111 v->range_mapuv = get_bits(gb, 3);
1114 av_log(avctx, AV_LOG_DEBUG, "Entry point info:\n"
1115 "BrokenLink=%i, ClosedEntry=%i, PanscanFlag=%i\n"
1116 "RefDist=%i, Postproc=%i, FastUVMC=%i, ExtMV=%i\n"
1117 "DQuant=%i, VSTransform=%i, Overlap=%i, Qmode=%i\n",
1118 v->broken_link, v->closed_entry, v->panscanflag, v->refdist_flag, v->s.loop_filter,
1119 v->fastuvmc, v->extended_mv, v->dquant, v->vstransform, v->overlap, v->quantizer_mode);
1124 static int vc1_parse_frame_header(VC1Context *v, GetBitContext* gb)
1126 int pqindex, lowquant, status;
1128 if(v->finterpflag) v->interpfrm = get_bits1(gb);
1129 skip_bits(gb, 2); //framecnt unused
1131 if (v->rangered) v->rangeredfrm = get_bits1(gb);
1132 v->s.pict_type = get_bits1(gb);
1133 if (v->s.avctx->max_b_frames) {
1134 if (!v->s.pict_type) {
1135 if (get_bits1(gb)) v->s.pict_type = FF_I_TYPE;
1136 else v->s.pict_type = FF_B_TYPE;
1137 } else v->s.pict_type = FF_P_TYPE;
1138 } else v->s.pict_type = v->s.pict_type ? FF_P_TYPE : FF_I_TYPE;
1141 if(v->s.pict_type == FF_B_TYPE) {
1142 v->bfraction_lut_index = get_vlc2(gb, ff_vc1_bfraction_vlc.table, VC1_BFRACTION_VLC_BITS, 1);
1143 v->bfraction = ff_vc1_bfraction_lut[v->bfraction_lut_index];
1144 if(v->bfraction == 0) {
1145 v->s.pict_type = FF_BI_TYPE;
1148 if(v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_BI_TYPE)
1149 skip_bits(gb, 7); // skip buffer fullness
1152 if(v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_BI_TYPE)
1154 if(v->s.pict_type == FF_P_TYPE)
1157 /* Quantizer stuff */
1158 pqindex = get_bits(gb, 5);
1159 if(!pqindex) return -1;
1160 if (v->quantizer_mode == QUANT_FRAME_IMPLICIT)
1161 v->pq = ff_vc1_pquant_table[0][pqindex];
1163 v->pq = ff_vc1_pquant_table[1][pqindex];
1166 if (v->quantizer_mode == QUANT_FRAME_IMPLICIT)
1167 v->pquantizer = pqindex < 9;
1168 if (v->quantizer_mode == QUANT_NON_UNIFORM)
1170 v->pqindex = pqindex;
1171 if (pqindex < 9) v->halfpq = get_bits1(gb);
1173 if (v->quantizer_mode == QUANT_FRAME_EXPLICIT)
1174 v->pquantizer = get_bits1(gb);
1176 if (v->extended_mv == 1) v->mvrange = get_unary(gb, 0, 3);
1177 v->k_x = v->mvrange + 9 + (v->mvrange >> 1); //k_x can be 9 10 12 13
1178 v->k_y = v->mvrange + 8; //k_y can be 8 9 10 11
1179 v->range_x = 1 << (v->k_x - 1);
1180 v->range_y = 1 << (v->k_y - 1);
1181 if (v->multires && v->s.pict_type != FF_B_TYPE) v->respic = get_bits(gb, 2);
1183 if(v->res_x8 && (v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_BI_TYPE)){
1184 v->x8_type = get_bits1(gb);
1185 }else v->x8_type = 0;
1186 //av_log(v->s.avctx, AV_LOG_INFO, "%c Frame: QP=[%i]%i (+%i/2) %i\n",
1187 // (v->s.pict_type == FF_P_TYPE) ? 'P' : ((v->s.pict_type == FF_I_TYPE) ? 'I' : 'B'), pqindex, v->pq, v->halfpq, v->rangeredfrm);
1189 if(v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_P_TYPE) v->use_ic = 0;
1191 switch(v->s.pict_type) {
1193 if (v->pq < 5) v->tt_index = 0;
1194 else if(v->pq < 13) v->tt_index = 1;
1195 else v->tt_index = 2;
1197 lowquant = (v->pq > 12) ? 0 : 1;
1198 v->mv_mode = ff_vc1_mv_pmode_table[lowquant][get_unary(gb, 1, 4)];
1199 if (v->mv_mode == MV_PMODE_INTENSITY_COMP)
1201 int scale, shift, i;
1202 v->mv_mode2 = ff_vc1_mv_pmode_table2[lowquant][get_unary(gb, 1, 3)];
1203 v->lumscale = get_bits(gb, 6);
1204 v->lumshift = get_bits(gb, 6);
1206 /* fill lookup tables for intensity compensation */
1209 shift = (255 - v->lumshift * 2) << 6;
1210 if(v->lumshift > 31)
1213 scale = v->lumscale + 32;
1214 if(v->lumshift > 31)
1215 shift = (v->lumshift - 64) << 6;
1217 shift = v->lumshift << 6;
1219 for(i = 0; i < 256; i++) {
1220 v->luty[i] = av_clip_uint8((scale * i + shift + 32) >> 6);
1221 v->lutuv[i] = av_clip_uint8((scale * (i - 128) + 128*64 + 32) >> 6);
1224 if(v->mv_mode == MV_PMODE_1MV_HPEL || v->mv_mode == MV_PMODE_1MV_HPEL_BILIN)
1225 v->s.quarter_sample = 0;
1226 else if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
1227 if(v->mv_mode2 == MV_PMODE_1MV_HPEL || v->mv_mode2 == MV_PMODE_1MV_HPEL_BILIN)
1228 v->s.quarter_sample = 0;
1230 v->s.quarter_sample = 1;
1232 v->s.quarter_sample = 1;
1233 v->s.mspel = !(v->mv_mode == MV_PMODE_1MV_HPEL_BILIN || (v->mv_mode == MV_PMODE_INTENSITY_COMP && v->mv_mode2 == MV_PMODE_1MV_HPEL_BILIN));
1235 if ((v->mv_mode == MV_PMODE_INTENSITY_COMP &&
1236 v->mv_mode2 == MV_PMODE_MIXED_MV)
1237 || v->mv_mode == MV_PMODE_MIXED_MV)
1239 status = bitplane_decoding(v->mv_type_mb_plane, &v->mv_type_is_raw, v);
1240 if (status < 0) return -1;
1241 av_log(v->s.avctx, AV_LOG_DEBUG, "MB MV Type plane encoding: "
1242 "Imode: %i, Invert: %i\n", status>>1, status&1);
1244 v->mv_type_is_raw = 0;
1245 memset(v->mv_type_mb_plane, 0, v->s.mb_stride * v->s.mb_height);
1247 status = bitplane_decoding(v->s.mbskip_table, &v->skip_is_raw, v);
1248 if (status < 0) return -1;
1249 av_log(v->s.avctx, AV_LOG_DEBUG, "MB Skip plane encoding: "
1250 "Imode: %i, Invert: %i\n", status>>1, status&1);
1252 /* Hopefully this is correct for P frames */
1253 v->s.mv_table_index = get_bits(gb, 2); //but using ff_vc1_ tables
1254 v->cbpcy_vlc = &ff_vc1_cbpcy_p_vlc[get_bits(gb, 2)];
1258 av_log(v->s.avctx, AV_LOG_DEBUG, "VOP DQuant info\n");
1259 vop_dquant_decoding(v);
1262 v->ttfrm = 0; //FIXME Is that so ?
1265 v->ttmbf = get_bits1(gb);
1268 v->ttfrm = ff_vc1_ttfrm_to_tt[get_bits(gb, 2)];
1276 if (v->pq < 5) v->tt_index = 0;
1277 else if(v->pq < 13) v->tt_index = 1;
1278 else v->tt_index = 2;
1280 lowquant = (v->pq > 12) ? 0 : 1;
1281 v->mv_mode = get_bits1(gb) ? MV_PMODE_1MV : MV_PMODE_1MV_HPEL_BILIN;
1282 v->s.quarter_sample = (v->mv_mode == MV_PMODE_1MV);
1283 v->s.mspel = v->s.quarter_sample;
1285 status = bitplane_decoding(v->direct_mb_plane, &v->dmb_is_raw, v);
1286 if (status < 0) return -1;
1287 av_log(v->s.avctx, AV_LOG_DEBUG, "MB Direct Type plane encoding: "
1288 "Imode: %i, Invert: %i\n", status>>1, status&1);
1289 status = bitplane_decoding(v->s.mbskip_table, &v->skip_is_raw, v);
1290 if (status < 0) return -1;
1291 av_log(v->s.avctx, AV_LOG_DEBUG, "MB Skip plane encoding: "
1292 "Imode: %i, Invert: %i\n", status>>1, status&1);
1294 v->s.mv_table_index = get_bits(gb, 2);
1295 v->cbpcy_vlc = &ff_vc1_cbpcy_p_vlc[get_bits(gb, 2)];
1299 av_log(v->s.avctx, AV_LOG_DEBUG, "VOP DQuant info\n");
1300 vop_dquant_decoding(v);
1306 v->ttmbf = get_bits1(gb);
1309 v->ttfrm = ff_vc1_ttfrm_to_tt[get_bits(gb, 2)];
1321 v->c_ac_table_index = decode012(gb);
1322 if (v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_BI_TYPE)
1324 v->y_ac_table_index = decode012(gb);
1327 v->s.dc_table_index = get_bits1(gb);
1330 if(v->s.pict_type == FF_BI_TYPE) {
1331 v->s.pict_type = FF_B_TYPE;
1337 static int vc1_parse_frame_header_adv(VC1Context *v, GetBitContext* gb)
1339 int pqindex, lowquant;
1342 v->p_frame_skipped = 0;
1345 v->fcm = decode012(gb);
1346 if(v->fcm) return -1; // interlaced frames/fields are not implemented
1348 switch(get_unary(gb, 0, 4)) {
1350 v->s.pict_type = FF_P_TYPE;
1353 v->s.pict_type = FF_B_TYPE;
1356 v->s.pict_type = FF_I_TYPE;
1359 v->s.pict_type = FF_BI_TYPE;
1362 v->s.pict_type = FF_P_TYPE; // skipped pic
1363 v->p_frame_skipped = 1;
1369 if(!v->interlace || v->psf) {
1370 v->rptfrm = get_bits(gb, 2);
1372 v->tff = get_bits1(gb);
1373 v->rptfrm = get_bits1(gb);
1376 if(v->panscanflag) {
1379 v->rnd = get_bits1(gb);
1381 v->uvsamp = get_bits1(gb);
1382 if(v->finterpflag) v->interpfrm = get_bits1(gb);
1383 if(v->s.pict_type == FF_B_TYPE) {
1384 v->bfraction_lut_index = get_vlc2(gb, ff_vc1_bfraction_vlc.table, VC1_BFRACTION_VLC_BITS, 1);
1385 v->bfraction = ff_vc1_bfraction_lut[v->bfraction_lut_index];
1386 if(v->bfraction == 0) {
1387 v->s.pict_type = FF_BI_TYPE; /* XXX: should not happen here */
1390 pqindex = get_bits(gb, 5);
1391 if(!pqindex) return -1;
1392 v->pqindex = pqindex;
1393 if (v->quantizer_mode == QUANT_FRAME_IMPLICIT)
1394 v->pq = ff_vc1_pquant_table[0][pqindex];
1396 v->pq = ff_vc1_pquant_table[1][pqindex];
1399 if (v->quantizer_mode == QUANT_FRAME_IMPLICIT)
1400 v->pquantizer = pqindex < 9;
1401 if (v->quantizer_mode == QUANT_NON_UNIFORM)
1403 v->pqindex = pqindex;
1404 if (pqindex < 9) v->halfpq = get_bits1(gb);
1406 if (v->quantizer_mode == QUANT_FRAME_EXPLICIT)
1407 v->pquantizer = get_bits1(gb);
1409 v->postproc = get_bits(gb, 2);
1411 if(v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_P_TYPE) v->use_ic = 0;
1413 switch(v->s.pict_type) {
1416 status = bitplane_decoding(v->acpred_plane, &v->acpred_is_raw, v);
1417 if (status < 0) return -1;
1418 av_log(v->s.avctx, AV_LOG_DEBUG, "ACPRED plane encoding: "
1419 "Imode: %i, Invert: %i\n", status>>1, status&1);
1420 v->condover = CONDOVER_NONE;
1421 if(v->overlap && v->pq <= 8) {
1422 v->condover = decode012(gb);
1423 if(v->condover == CONDOVER_SELECT) {
1424 status = bitplane_decoding(v->over_flags_plane, &v->overflg_is_raw, v);
1425 if (status < 0) return -1;
1426 av_log(v->s.avctx, AV_LOG_DEBUG, "CONDOVER plane encoding: "
1427 "Imode: %i, Invert: %i\n", status>>1, status&1);
1432 if (v->extended_mv) v->mvrange = get_unary(gb, 0, 3);
1433 else v->mvrange = 0;
1434 v->k_x = v->mvrange + 9 + (v->mvrange >> 1); //k_x can be 9 10 12 13
1435 v->k_y = v->mvrange + 8; //k_y can be 8 9 10 11
1436 v->range_x = 1 << (v->k_x - 1);
1437 v->range_y = 1 << (v->k_y - 1);
1439 if (v->pq < 5) v->tt_index = 0;
1440 else if(v->pq < 13) v->tt_index = 1;
1441 else v->tt_index = 2;
1443 lowquant = (v->pq > 12) ? 0 : 1;
1444 v->mv_mode = ff_vc1_mv_pmode_table[lowquant][get_unary(gb, 1, 4)];
1445 if (v->mv_mode == MV_PMODE_INTENSITY_COMP)
1447 int scale, shift, i;
1448 v->mv_mode2 = ff_vc1_mv_pmode_table2[lowquant][get_unary(gb, 1, 3)];
1449 v->lumscale = get_bits(gb, 6);
1450 v->lumshift = get_bits(gb, 6);
1451 /* fill lookup tables for intensity compensation */
1454 shift = (255 - v->lumshift * 2) << 6;
1455 if(v->lumshift > 31)
1458 scale = v->lumscale + 32;
1459 if(v->lumshift > 31)
1460 shift = (v->lumshift - 64) << 6;
1462 shift = v->lumshift << 6;
1464 for(i = 0; i < 256; i++) {
1465 v->luty[i] = av_clip_uint8((scale * i + shift + 32) >> 6);
1466 v->lutuv[i] = av_clip_uint8((scale * (i - 128) + 128*64 + 32) >> 6);
1470 if(v->mv_mode == MV_PMODE_1MV_HPEL || v->mv_mode == MV_PMODE_1MV_HPEL_BILIN)
1471 v->s.quarter_sample = 0;
1472 else if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
1473 if(v->mv_mode2 == MV_PMODE_1MV_HPEL || v->mv_mode2 == MV_PMODE_1MV_HPEL_BILIN)
1474 v->s.quarter_sample = 0;
1476 v->s.quarter_sample = 1;
1478 v->s.quarter_sample = 1;
1479 v->s.mspel = !(v->mv_mode == MV_PMODE_1MV_HPEL_BILIN || (v->mv_mode == MV_PMODE_INTENSITY_COMP && v->mv_mode2 == MV_PMODE_1MV_HPEL_BILIN));
1481 if ((v->mv_mode == MV_PMODE_INTENSITY_COMP &&
1482 v->mv_mode2 == MV_PMODE_MIXED_MV)
1483 || v->mv_mode == MV_PMODE_MIXED_MV)
1485 status = bitplane_decoding(v->mv_type_mb_plane, &v->mv_type_is_raw, v);
1486 if (status < 0) return -1;
1487 av_log(v->s.avctx, AV_LOG_DEBUG, "MB MV Type plane encoding: "
1488 "Imode: %i, Invert: %i\n", status>>1, status&1);
1490 v->mv_type_is_raw = 0;
1491 memset(v->mv_type_mb_plane, 0, v->s.mb_stride * v->s.mb_height);
1493 status = bitplane_decoding(v->s.mbskip_table, &v->skip_is_raw, v);
1494 if (status < 0) return -1;
1495 av_log(v->s.avctx, AV_LOG_DEBUG, "MB Skip plane encoding: "
1496 "Imode: %i, Invert: %i\n", status>>1, status&1);
1498 /* Hopefully this is correct for P frames */
1499 v->s.mv_table_index = get_bits(gb, 2); //but using ff_vc1_ tables
1500 v->cbpcy_vlc = &ff_vc1_cbpcy_p_vlc[get_bits(gb, 2)];
1503 av_log(v->s.avctx, AV_LOG_DEBUG, "VOP DQuant info\n");
1504 vop_dquant_decoding(v);
1507 v->ttfrm = 0; //FIXME Is that so ?
1510 v->ttmbf = get_bits1(gb);
1513 v->ttfrm = ff_vc1_ttfrm_to_tt[get_bits(gb, 2)];
1521 if (v->extended_mv) v->mvrange = get_unary(gb, 0, 3);
1522 else v->mvrange = 0;
1523 v->k_x = v->mvrange + 9 + (v->mvrange >> 1); //k_x can be 9 10 12 13
1524 v->k_y = v->mvrange + 8; //k_y can be 8 9 10 11
1525 v->range_x = 1 << (v->k_x - 1);
1526 v->range_y = 1 << (v->k_y - 1);
1528 if (v->pq < 5) v->tt_index = 0;
1529 else if(v->pq < 13) v->tt_index = 1;
1530 else v->tt_index = 2;
1532 lowquant = (v->pq > 12) ? 0 : 1;
1533 v->mv_mode = get_bits1(gb) ? MV_PMODE_1MV : MV_PMODE_1MV_HPEL_BILIN;
1534 v->s.quarter_sample = (v->mv_mode == MV_PMODE_1MV);
1535 v->s.mspel = v->s.quarter_sample;
1537 status = bitplane_decoding(v->direct_mb_plane, &v->dmb_is_raw, v);
1538 if (status < 0) return -1;
1539 av_log(v->s.avctx, AV_LOG_DEBUG, "MB Direct Type plane encoding: "
1540 "Imode: %i, Invert: %i\n", status>>1, status&1);
1541 status = bitplane_decoding(v->s.mbskip_table, &v->skip_is_raw, v);
1542 if (status < 0) return -1;
1543 av_log(v->s.avctx, AV_LOG_DEBUG, "MB Skip plane encoding: "
1544 "Imode: %i, Invert: %i\n", status>>1, status&1);
1546 v->s.mv_table_index = get_bits(gb, 2);
1547 v->cbpcy_vlc = &ff_vc1_cbpcy_p_vlc[get_bits(gb, 2)];
1551 av_log(v->s.avctx, AV_LOG_DEBUG, "VOP DQuant info\n");
1552 vop_dquant_decoding(v);
1558 v->ttmbf = get_bits1(gb);
1561 v->ttfrm = ff_vc1_ttfrm_to_tt[get_bits(gb, 2)];
1571 v->c_ac_table_index = decode012(gb);
1572 if (v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_BI_TYPE)
1574 v->y_ac_table_index = decode012(gb);
1577 v->s.dc_table_index = get_bits1(gb);
1578 if ((v->s.pict_type == FF_I_TYPE || v->s.pict_type == FF_BI_TYPE) && v->dquant) {
1579 av_log(v->s.avctx, AV_LOG_DEBUG, "VOP DQuant info\n");
1580 vop_dquant_decoding(v);
1584 if(v->s.pict_type == FF_BI_TYPE) {
1585 v->s.pict_type = FF_B_TYPE;
1591 /***********************************************************************/
1593 * @defgroup vc1block VC-1 Block-level functions
1594 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1600 * @brief Get macroblock-level quantizer scale
1602 #define GET_MQUANT() \
1606 if (v->dqprofile == DQPROFILE_ALL_MBS) \
1610 mquant = (get_bits1(gb)) ? v->altpq : v->pq; \
1614 mqdiff = get_bits(gb, 3); \
1615 if (mqdiff != 7) mquant = v->pq + mqdiff; \
1616 else mquant = get_bits(gb, 5); \
1619 if(v->dqprofile == DQPROFILE_SINGLE_EDGE) \
1620 edges = 1 << v->dqsbedge; \
1621 else if(v->dqprofile == DQPROFILE_DOUBLE_EDGES) \
1622 edges = (3 << v->dqsbedge) % 15; \
1623 else if(v->dqprofile == DQPROFILE_FOUR_EDGES) \
1625 if((edges&1) && !s->mb_x) \
1626 mquant = v->altpq; \
1627 if((edges&2) && s->first_slice_line) \
1628 mquant = v->altpq; \
1629 if((edges&4) && s->mb_x == (s->mb_width - 1)) \
1630 mquant = v->altpq; \
1631 if((edges&8) && s->mb_y == (s->mb_height - 1)) \
1632 mquant = v->altpq; \
1636 * @def GET_MVDATA(_dmv_x, _dmv_y)
1637 * @brief Get MV differentials
1638 * @see MVDATA decoding from 8.3.5.2, p(1)20
1639 * @param _dmv_x Horizontal differential for decoded MV
1640 * @param _dmv_y Vertical differential for decoded MV
1642 #define GET_MVDATA(_dmv_x, _dmv_y) \
1643 index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table,\
1644 VC1_MV_DIFF_VLC_BITS, 2); \
1647 mb_has_coeffs = 1; \
1650 else mb_has_coeffs = 0; \
1652 if (!index) { _dmv_x = _dmv_y = 0; } \
1653 else if (index == 35) \
1655 _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample); \
1656 _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample); \
1658 else if (index == 36) \
1667 if (!s->quarter_sample && index1 == 5) val = 1; \
1669 if(size_table[index1] - val > 0) \
1670 val = get_bits(gb, size_table[index1] - val); \
1672 sign = 0 - (val&1); \
1673 _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
1676 if (!s->quarter_sample && index1 == 5) val = 1; \
1678 if(size_table[index1] - val > 0) \
1679 val = get_bits(gb, size_table[index1] - val); \
1681 sign = 0 - (val&1); \
1682 _dmv_y = (sign ^ ((val>>1) + offset_table[index1])) - sign; \
1685 /** Predict and set motion vector
1687 static inline void vc1_pred_mv(MpegEncContext *s, int n, int dmv_x, int dmv_y, int mv1, int r_x, int r_y, uint8_t* is_intra)
1689 int xy, wrap, off = 0;
1694 /* scale MV difference to be quad-pel */
1695 dmv_x <<= 1 - s->quarter_sample;
1696 dmv_y <<= 1 - s->quarter_sample;
1698 wrap = s->b8_stride;
1699 xy = s->block_index[n];
1702 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = 0;
1703 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = 0;
1704 s->current_picture.motion_val[1][xy][0] = 0;
1705 s->current_picture.motion_val[1][xy][1] = 0;
1706 if(mv1) { /* duplicate motion data for 1-MV block */
1707 s->current_picture.motion_val[0][xy + 1][0] = 0;
1708 s->current_picture.motion_val[0][xy + 1][1] = 0;
1709 s->current_picture.motion_val[0][xy + wrap][0] = 0;
1710 s->current_picture.motion_val[0][xy + wrap][1] = 0;
1711 s->current_picture.motion_val[0][xy + wrap + 1][0] = 0;
1712 s->current_picture.motion_val[0][xy + wrap + 1][1] = 0;
1713 s->current_picture.motion_val[1][xy + 1][0] = 0;
1714 s->current_picture.motion_val[1][xy + 1][1] = 0;
1715 s->current_picture.motion_val[1][xy + wrap][0] = 0;
1716 s->current_picture.motion_val[1][xy + wrap][1] = 0;
1717 s->current_picture.motion_val[1][xy + wrap + 1][0] = 0;
1718 s->current_picture.motion_val[1][xy + wrap + 1][1] = 0;
1723 C = s->current_picture.motion_val[0][xy - 1];
1724 A = s->current_picture.motion_val[0][xy - wrap];
1726 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
1728 //in 4-MV mode different blocks have different B predictor position
1731 off = (s->mb_x > 0) ? -1 : 1;
1734 off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
1743 B = s->current_picture.motion_val[0][xy - wrap + off];
1745 if(!s->first_slice_line || (n==2 || n==3)) { // predictor A is not out of bounds
1746 if(s->mb_width == 1) {
1750 px = mid_pred(A[0], B[0], C[0]);
1751 py = mid_pred(A[1], B[1], C[1]);
1753 } else if(s->mb_x || (n==1 || n==3)) { // predictor C is not out of bounds
1759 /* Pullback MV as specified in 8.3.5.3.4 */
1762 qx = (s->mb_x << 6) + ((n==1 || n==3) ? 32 : 0);
1763 qy = (s->mb_y << 6) + ((n==2 || n==3) ? 32 : 0);
1764 X = (s->mb_width << 6) - 4;
1765 Y = (s->mb_height << 6) - 4;
1767 if(qx + px < -60) px = -60 - qx;
1768 if(qy + py < -60) py = -60 - qy;
1770 if(qx + px < -28) px = -28 - qx;
1771 if(qy + py < -28) py = -28 - qy;
1773 if(qx + px > X) px = X - qx;
1774 if(qy + py > Y) py = Y - qy;
1776 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
1777 if((!s->first_slice_line || (n==2 || n==3)) && (s->mb_x || (n==1 || n==3))) {
1778 if(is_intra[xy - wrap])
1779 sum = FFABS(px) + FFABS(py);
1781 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
1783 if(get_bits1(&s->gb)) {
1791 if(is_intra[xy - 1])
1792 sum = FFABS(px) + FFABS(py);
1794 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
1796 if(get_bits1(&s->gb)) {
1806 /* store MV using signed modulus of MV range defined in 4.11 */
1807 s->mv[0][n][0] = s->current_picture.motion_val[0][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1808 s->mv[0][n][1] = s->current_picture.motion_val[0][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
1809 if(mv1) { /* duplicate motion data for 1-MV block */
1810 s->current_picture.motion_val[0][xy + 1][0] = s->current_picture.motion_val[0][xy][0];
1811 s->current_picture.motion_val[0][xy + 1][1] = s->current_picture.motion_val[0][xy][1];
1812 s->current_picture.motion_val[0][xy + wrap][0] = s->current_picture.motion_val[0][xy][0];
1813 s->current_picture.motion_val[0][xy + wrap][1] = s->current_picture.motion_val[0][xy][1];
1814 s->current_picture.motion_val[0][xy + wrap + 1][0] = s->current_picture.motion_val[0][xy][0];
1815 s->current_picture.motion_val[0][xy + wrap + 1][1] = s->current_picture.motion_val[0][xy][1];
1819 /** Motion compensation for direct or interpolated blocks in B-frames
1821 static void vc1_interp_mc(VC1Context *v)
1823 MpegEncContext *s = &v->s;
1824 DSPContext *dsp = &v->s.dsp;
1825 uint8_t *srcY, *srcU, *srcV;
1826 int dxy, uvdxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
1828 if(!v->s.next_picture.data[0])return;
1830 mx = s->mv[1][0][0];
1831 my = s->mv[1][0][1];
1832 uvmx = (mx + ((mx & 3) == 3)) >> 1;
1833 uvmy = (my + ((my & 3) == 3)) >> 1;
1835 uvmx = uvmx + ((uvmx<0)?-(uvmx&1):(uvmx&1));
1836 uvmy = uvmy + ((uvmy<0)?-(uvmy&1):(uvmy&1));
1838 srcY = s->next_picture.data[0];
1839 srcU = s->next_picture.data[1];
1840 srcV = s->next_picture.data[2];
1842 src_x = s->mb_x * 16 + (mx >> 2);
1843 src_y = s->mb_y * 16 + (my >> 2);
1844 uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
1845 uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
1847 if(v->profile != PROFILE_ADVANCED){
1848 src_x = av_clip( src_x, -16, s->mb_width * 16);
1849 src_y = av_clip( src_y, -16, s->mb_height * 16);
1850 uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width * 8);
1851 uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
1853 src_x = av_clip( src_x, -17, s->avctx->coded_width);
1854 src_y = av_clip( src_y, -18, s->avctx->coded_height + 1);
1855 uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width >> 1);
1856 uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
1859 srcY += src_y * s->linesize + src_x;
1860 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1861 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1863 /* for grayscale we should not try to read from unknown area */
1864 if(s->flags & CODEC_FLAG_GRAY) {
1865 srcU = s->edge_emu_buffer + 18 * s->linesize;
1866 srcV = s->edge_emu_buffer + 18 * s->linesize;
1870 || (unsigned)src_x > s->h_edge_pos - (mx&3) - 16
1871 || (unsigned)src_y > s->v_edge_pos - (my&3) - 16){
1872 uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;
1874 srcY -= s->mspel * (1 + s->linesize);
1875 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
1876 src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
1877 srcY = s->edge_emu_buffer;
1878 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8+1, 8+1,
1879 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
1880 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
1881 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
1884 /* if we deal with range reduction we need to scale source blocks */
1885 if(v->rangeredfrm) {
1887 uint8_t *src, *src2;
1890 for(j = 0; j < 17 + s->mspel*2; j++) {
1891 for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
1894 src = srcU; src2 = srcV;
1895 for(j = 0; j < 9; j++) {
1896 for(i = 0; i < 9; i++) {
1897 src[i] = ((src[i] - 128) >> 1) + 128;
1898 src2[i] = ((src2[i] - 128) >> 1) + 128;
1900 src += s->uvlinesize;
1901 src2 += s->uvlinesize;
1904 srcY += s->mspel * (1 + s->linesize);
1908 dxy = ((my & 3) << 2) | (mx & 3);
1909 dsp->avg_vc1_mspel_pixels_tab[dxy](s->dest[0] , srcY , s->linesize, v->rnd);
1910 dsp->avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
1911 srcY += s->linesize * 8;
1912 dsp->avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize , srcY , s->linesize, v->rnd);
1913 dsp->avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
1915 dxy = (my & 2) | ((mx & 2) >> 1);
1918 dsp->avg_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
1920 dsp->avg_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
1923 if(s->flags & CODEC_FLAG_GRAY) return;
1924 /* Chroma MC always uses qpel blilinear */
1925 uvdxy = ((uvmy & 3) << 2) | (uvmx & 3);
1928 dsp->avg_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
1929 dsp->avg_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
1932 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
1936 #if B_FRACTION_DEN==256
1940 return 2 * ((value * n + 255) >> 9);
1941 return (value * n + 128) >> 8;
1944 n -= B_FRACTION_DEN;
1946 return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
1947 return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
1951 /** Reconstruct motion vector for B-frame and do motion compensation
1953 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mode)
1956 v->mv_mode2 = v->mv_mode;
1957 v->mv_mode = MV_PMODE_INTENSITY_COMP;
1962 if(v->use_ic) v->mv_mode = v->mv_mode2;
1965 if(mode == BMV_TYPE_INTERPOLATED) {
1968 if(v->use_ic) v->mv_mode = v->mv_mode2;
1972 if(v->use_ic && (mode == BMV_TYPE_BACKWARD)) v->mv_mode = v->mv_mode2;
1973 vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
1974 if(v->use_ic) v->mv_mode = v->mv_mode2;
1977 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2], int direct, int mvtype)
1979 MpegEncContext *s = &v->s;
1980 int xy, wrap, off = 0;
1985 const uint8_t *is_intra = v->mb_type[0];
1989 /* scale MV difference to be quad-pel */
1990 dmv_x[0] <<= 1 - s->quarter_sample;
1991 dmv_y[0] <<= 1 - s->quarter_sample;
1992 dmv_x[1] <<= 1 - s->quarter_sample;
1993 dmv_y[1] <<= 1 - s->quarter_sample;
1995 wrap = s->b8_stride;
1996 xy = s->block_index[0];
1999 s->current_picture.motion_val[0][xy][0] =
2000 s->current_picture.motion_val[0][xy][1] =
2001 s->current_picture.motion_val[1][xy][0] =
2002 s->current_picture.motion_val[1][xy][1] = 0;
2005 s->mv[0][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
2006 s->mv[0][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
2007 s->mv[1][0][0] = scale_mv(s->next_picture.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
2008 s->mv[1][0][1] = scale_mv(s->next_picture.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
2010 /* Pullback predicted motion vectors as specified in 8.4.5.4 */
2011 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));
2012 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));
2013 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));
2014 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));
2016 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
2017 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
2018 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
2019 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
2023 if((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2024 C = s->current_picture.motion_val[0][xy - 2];
2025 A = s->current_picture.motion_val[0][xy - wrap*2];
2026 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2027 B = s->current_picture.motion_val[0][xy - wrap*2 + off];
2029 if(!s->mb_x) C[0] = C[1] = 0;
2030 if(!s->first_slice_line) { // predictor A is not out of bounds
2031 if(s->mb_width == 1) {
2035 px = mid_pred(A[0], B[0], C[0]);
2036 py = mid_pred(A[1], B[1], C[1]);
2038 } else if(s->mb_x) { // predictor C is not out of bounds
2044 /* Pullback MV as specified in 8.3.5.3.4 */
2047 if(v->profile < PROFILE_ADVANCED) {
2048 qx = (s->mb_x << 5);
2049 qy = (s->mb_y << 5);
2050 X = (s->mb_width << 5) - 4;
2051 Y = (s->mb_height << 5) - 4;
2052 if(qx + px < -28) px = -28 - qx;
2053 if(qy + py < -28) py = -28 - qy;
2054 if(qx + px > X) px = X - qx;
2055 if(qy + py > Y) py = Y - qy;
2057 qx = (s->mb_x << 6);
2058 qy = (s->mb_y << 6);
2059 X = (s->mb_width << 6) - 4;
2060 Y = (s->mb_height << 6) - 4;
2061 if(qx + px < -60) px = -60 - qx;
2062 if(qy + py < -60) py = -60 - qy;
2063 if(qx + px > X) px = X - qx;
2064 if(qy + py > Y) py = Y - qy;
2067 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2068 if(0 && !s->first_slice_line && s->mb_x) {
2069 if(is_intra[xy - wrap])
2070 sum = FFABS(px) + FFABS(py);
2072 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2074 if(get_bits1(&s->gb)) {
2082 if(is_intra[xy - 2])
2083 sum = FFABS(px) + FFABS(py);
2085 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2087 if(get_bits1(&s->gb)) {
2097 /* store MV using signed modulus of MV range defined in 4.11 */
2098 s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
2099 s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
2101 if((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2102 C = s->current_picture.motion_val[1][xy - 2];
2103 A = s->current_picture.motion_val[1][xy - wrap*2];
2104 off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2105 B = s->current_picture.motion_val[1][xy - wrap*2 + off];
2107 if(!s->mb_x) C[0] = C[1] = 0;
2108 if(!s->first_slice_line) { // predictor A is not out of bounds
2109 if(s->mb_width == 1) {
2113 px = mid_pred(A[0], B[0], C[0]);
2114 py = mid_pred(A[1], B[1], C[1]);
2116 } else if(s->mb_x) { // predictor C is not out of bounds
2122 /* Pullback MV as specified in 8.3.5.3.4 */
2125 if(v->profile < PROFILE_ADVANCED) {
2126 qx = (s->mb_x << 5);
2127 qy = (s->mb_y << 5);
2128 X = (s->mb_width << 5) - 4;
2129 Y = (s->mb_height << 5) - 4;
2130 if(qx + px < -28) px = -28 - qx;
2131 if(qy + py < -28) py = -28 - qy;
2132 if(qx + px > X) px = X - qx;
2133 if(qy + py > Y) py = Y - qy;
2135 qx = (s->mb_x << 6);
2136 qy = (s->mb_y << 6);
2137 X = (s->mb_width << 6) - 4;
2138 Y = (s->mb_height << 6) - 4;
2139 if(qx + px < -60) px = -60 - qx;
2140 if(qy + py < -60) py = -60 - qy;
2141 if(qx + px > X) px = X - qx;
2142 if(qy + py > Y) py = Y - qy;
2145 /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2146 if(0 && !s->first_slice_line && s->mb_x) {
2147 if(is_intra[xy - wrap])
2148 sum = FFABS(px) + FFABS(py);
2150 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2152 if(get_bits1(&s->gb)) {
2160 if(is_intra[xy - 2])
2161 sum = FFABS(px) + FFABS(py);
2163 sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2165 if(get_bits1(&s->gb)) {
2175 /* store MV using signed modulus of MV range defined in 4.11 */
2177 s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
2178 s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
2180 s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
2181 s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
2182 s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
2183 s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
2186 /** Get predicted DC value for I-frames only
2187 * prediction dir: left=0, top=1
2188 * @param s MpegEncContext
2189 * @param overlap flag indicating that overlap filtering is used
2190 * @param pq integer part of picture quantizer
2191 * @param[in] n block index in the current MB
2192 * @param dc_val_ptr Pointer to DC predictor
2193 * @param dir_ptr Prediction direction for use in AC prediction
2195 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2196 int16_t **dc_val_ptr, int *dir_ptr)
2198 int a, b, c, wrap, pred, scale;
2200 static const uint16_t dcpred[32] = {
2201 -1, 1024, 512, 341, 256, 205, 171, 146, 128,
2202 114, 102, 93, 85, 79, 73, 68, 64,
2203 60, 57, 54, 51, 49, 47, 45, 43,
2204 41, 39, 38, 37, 35, 34, 33
2207 /* find prediction - wmv3_dc_scale always used here in fact */
2208 if (n < 4) scale = s->y_dc_scale;
2209 else scale = s->c_dc_scale;
2211 wrap = s->block_wrap[n];
2212 dc_val= s->dc_val[0] + s->block_index[n];
2218 b = dc_val[ - 1 - wrap];
2219 a = dc_val[ - wrap];
2221 if (pq < 9 || !overlap)
2223 /* Set outer values */
2224 if (s->first_slice_line && (n!=2 && n!=3)) b=a=dcpred[scale];
2225 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=dcpred[scale];
2229 /* Set outer values */
2230 if (s->first_slice_line && (n!=2 && n!=3)) b=a=0;
2231 if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=0;
2234 if (abs(a - b) <= abs(b - c)) {
2242 /* update predictor */
2243 *dc_val_ptr = &dc_val[0];
2248 /** Get predicted DC value
2249 * prediction dir: left=0, top=1
2250 * @param s MpegEncContext
2251 * @param overlap flag indicating that overlap filtering is used
2252 * @param pq integer part of picture quantizer
2253 * @param[in] n block index in the current MB
2254 * @param a_avail flag indicating top block availability
2255 * @param c_avail flag indicating left block availability
2256 * @param dc_val_ptr Pointer to DC predictor
2257 * @param dir_ptr Prediction direction for use in AC prediction
2259 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2260 int a_avail, int c_avail,
2261 int16_t **dc_val_ptr, int *dir_ptr)
2263 int a, b, c, wrap, pred, scale;
2265 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2268 /* find prediction - wmv3_dc_scale always used here in fact */
2269 if (n < 4) scale = s->y_dc_scale;
2270 else scale = s->c_dc_scale;
2272 wrap = s->block_wrap[n];
2273 dc_val= s->dc_val[0] + s->block_index[n];
2279 b = dc_val[ - 1 - wrap];
2280 a = dc_val[ - wrap];
2281 /* scale predictors if needed */
2282 q1 = s->current_picture.qscale_table[mb_pos];
2283 if(c_avail && (n!= 1 && n!=3)) {
2284 q2 = s->current_picture.qscale_table[mb_pos - 1];
2286 c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
2288 if(a_avail && (n!= 2 && n!=3)) {
2289 q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2291 a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
2293 if(a_avail && c_avail && (n!=3)) {
2296 if(n != 2) off -= s->mb_stride;
2297 q2 = s->current_picture.qscale_table[off];
2299 b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
2302 if(a_avail && c_avail) {
2303 if(abs(a - b) <= abs(b - c)) {
2310 } else if(a_avail) {
2313 } else if(c_avail) {
2321 /* update predictor */
2322 *dc_val_ptr = &dc_val[0];
2326 /** @} */ // Block group
2329 * @defgroup vc1_std_mb VC1 Macroblock-level functions in Simple/Main Profiles
2330 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
2334 static inline int vc1_coded_block_pred(MpegEncContext * s, int n, uint8_t **coded_block_ptr)
2336 int xy, wrap, pred, a, b, c;
2338 xy = s->block_index[n];
2339 wrap = s->b8_stride;
2344 a = s->coded_block[xy - 1 ];
2345 b = s->coded_block[xy - 1 - wrap];
2346 c = s->coded_block[xy - wrap];
2355 *coded_block_ptr = &s->coded_block[xy];
2361 * Decode one AC coefficient
2362 * @param v The VC1 context
2363 * @param last Last coefficient
2364 * @param skip How much zero coefficients to skip
2365 * @param value Decoded AC coefficient value
2366 * @param codingset set of VLC to decode data
2369 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip, int *value, int codingset)
2371 GetBitContext *gb = &v->s.gb;
2372 int index, escape, run = 0, level = 0, lst = 0;
2374 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2375 if (index != vc1_ac_sizes[codingset] - 1) {
2376 run = vc1_index_decode_table[codingset][index][0];
2377 level = vc1_index_decode_table[codingset][index][1];
2378 lst = index >= vc1_last_decode_table[codingset];
2382 escape = decode210(gb);
2384 index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2385 run = vc1_index_decode_table[codingset][index][0];
2386 level = vc1_index_decode_table[codingset][index][1];
2387 lst = index >= vc1_last_decode_table[codingset];
2390 level += vc1_last_delta_level_table[codingset][run];
2392 level += vc1_delta_level_table[codingset][run];
2395 run += vc1_last_delta_run_table[codingset][level] + 1;
2397 run += vc1_delta_run_table[codingset][level] + 1;
2403 lst = get_bits1(gb);
2404 if(v->s.esc3_level_length == 0) {
2405 if(v->pq < 8 || v->dquantfrm) { // table 59
2406 v->s.esc3_level_length = get_bits(gb, 3);
2407 if(!v->s.esc3_level_length)
2408 v->s.esc3_level_length = get_bits(gb, 2) + 8;
2410 v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
2412 v->s.esc3_run_length = 3 + get_bits(gb, 2);
2414 run = get_bits(gb, v->s.esc3_run_length);
2415 sign = get_bits1(gb);
2416 level = get_bits(gb, v->s.esc3_level_length);
2427 /** Decode intra block in intra frames - should be faster than decode_intra_block
2428 * @param v VC1Context
2429 * @param block block to decode
2430 * @param[in] n subblock index
2431 * @param coded are AC coeffs present or not
2432 * @param codingset set of VLC to decode data
2434 static int vc1_decode_i_block(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset)
2436 GetBitContext *gb = &v->s.gb;
2437 MpegEncContext *s = &v->s;
2438 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2441 int16_t *ac_val, *ac_val2;
2444 /* Get DC differential */
2446 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2448 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2451 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2456 if (dcdiff == 119 /* ESC index value */)
2458 /* TODO: Optimize */
2459 if (v->pq == 1) dcdiff = get_bits(gb, 10);
2460 else if (v->pq == 2) dcdiff = get_bits(gb, 9);
2461 else dcdiff = get_bits(gb, 8);
2466 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
2467 else if (v->pq == 2)
2468 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
2475 dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
2478 /* Store the quantized DC coeff, used for prediction */
2480 block[0] = dcdiff * s->y_dc_scale;
2482 block[0] = dcdiff * s->c_dc_scale;
2495 int last = 0, skip, value;
2496 const int8_t *zz_table;
2500 scale = v->pq * 2 + v->halfpq;
2504 zz_table = wmv1_scantable[2];
2506 zz_table = wmv1_scantable[3];
2508 zz_table = wmv1_scantable[1];
2510 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2512 if(dc_pred_dir) //left
2515 ac_val -= 16 * s->block_wrap[n];
2518 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2522 block[zz_table[i++]] = value;
2525 /* apply AC prediction if needed */
2527 if(dc_pred_dir) { //left
2528 for(k = 1; k < 8; k++)
2529 block[k << 3] += ac_val[k];
2531 for(k = 1; k < 8; k++)
2532 block[k] += ac_val[k + 8];
2535 /* save AC coeffs for further prediction */
2536 for(k = 1; k < 8; k++) {
2537 ac_val2[k] = block[k << 3];
2538 ac_val2[k + 8] = block[k];
2541 /* scale AC coeffs */
2542 for(k = 1; k < 64; k++)
2546 block[k] += (block[k] < 0) ? -v->pq : v->pq;
2549 if(s->ac_pred) i = 63;
2555 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2558 scale = v->pq * 2 + v->halfpq;
2559 memset(ac_val2, 0, 16 * 2);
2560 if(dc_pred_dir) {//left
2563 memcpy(ac_val2, ac_val, 8 * 2);
2565 ac_val -= 16 * s->block_wrap[n];
2567 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2570 /* apply AC prediction if needed */
2572 if(dc_pred_dir) { //left
2573 for(k = 1; k < 8; k++) {
2574 block[k << 3] = ac_val[k] * scale;
2575 if(!v->pquantizer && block[k << 3])
2576 block[k << 3] += (block[k << 3] < 0) ? -v->pq : v->pq;
2579 for(k = 1; k < 8; k++) {
2580 block[k] = ac_val[k + 8] * scale;
2581 if(!v->pquantizer && block[k])
2582 block[k] += (block[k] < 0) ? -v->pq : v->pq;
2588 s->block_last_index[n] = i;
2593 /** Decode intra block in intra frames - should be faster than decode_intra_block
2594 * @param v VC1Context
2595 * @param block block to decode
2596 * @param[in] n subblock number
2597 * @param coded are AC coeffs present or not
2598 * @param codingset set of VLC to decode data
2599 * @param mquant quantizer value for this macroblock
2601 static int vc1_decode_i_block_adv(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset, int mquant)
2603 GetBitContext *gb = &v->s.gb;
2604 MpegEncContext *s = &v->s;
2605 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2608 int16_t *ac_val, *ac_val2;
2610 int a_avail = v->a_avail, c_avail = v->c_avail;
2611 int use_pred = s->ac_pred;
2614 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2616 /* Get DC differential */
2618 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2620 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2623 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2628 if (dcdiff == 119 /* ESC index value */)
2630 /* TODO: Optimize */
2631 if (mquant == 1) dcdiff = get_bits(gb, 10);
2632 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2633 else dcdiff = get_bits(gb, 8);
2638 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
2639 else if (mquant == 2)
2640 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
2647 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
2650 /* Store the quantized DC coeff, used for prediction */
2652 block[0] = dcdiff * s->y_dc_scale;
2654 block[0] = dcdiff * s->c_dc_scale;
2663 /* check if AC is needed at all */
2664 if(!a_avail && !c_avail) use_pred = 0;
2665 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2668 scale = mquant * 2 + ((mquant == v->pq) ? v->halfpq : 0);
2670 if(dc_pred_dir) //left
2673 ac_val -= 16 * s->block_wrap[n];
2675 q1 = s->current_picture.qscale_table[mb_pos];
2676 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
2677 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2678 if(dc_pred_dir && n==1) q2 = q1;
2679 if(!dc_pred_dir && n==2) q2 = q1;
2683 int last = 0, skip, value;
2684 const int8_t *zz_table;
2689 zz_table = wmv1_scantable[2];
2691 zz_table = wmv1_scantable[3];
2693 zz_table = wmv1_scantable[1];
2696 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2700 block[zz_table[i++]] = value;
2703 /* apply AC prediction if needed */
2705 /* scale predictors if needed*/
2707 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2708 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2710 if(dc_pred_dir) { //left
2711 for(k = 1; k < 8; k++)
2712 block[k << 3] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2714 for(k = 1; k < 8; k++)
2715 block[k] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2718 if(dc_pred_dir) { //left
2719 for(k = 1; k < 8; k++)
2720 block[k << 3] += ac_val[k];
2722 for(k = 1; k < 8; k++)
2723 block[k] += ac_val[k + 8];
2727 /* save AC coeffs for further prediction */
2728 for(k = 1; k < 8; k++) {
2729 ac_val2[k] = block[k << 3];
2730 ac_val2[k + 8] = block[k];
2733 /* scale AC coeffs */
2734 for(k = 1; k < 64; k++)
2738 block[k] += (block[k] < 0) ? -mquant : mquant;
2741 if(use_pred) i = 63;
2742 } else { // no AC coeffs
2745 memset(ac_val2, 0, 16 * 2);
2746 if(dc_pred_dir) {//left
2748 memcpy(ac_val2, ac_val, 8 * 2);
2750 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2751 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2752 for(k = 1; k < 8; k++)
2753 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2758 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2760 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2761 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2762 for(k = 1; k < 8; k++)
2763 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2768 /* apply AC prediction if needed */
2770 if(dc_pred_dir) { //left
2771 for(k = 1; k < 8; k++) {
2772 block[k << 3] = ac_val2[k] * scale;
2773 if(!v->pquantizer && block[k << 3])
2774 block[k << 3] += (block[k << 3] < 0) ? -mquant : mquant;
2777 for(k = 1; k < 8; k++) {
2778 block[k] = ac_val2[k + 8] * scale;
2779 if(!v->pquantizer && block[k])
2780 block[k] += (block[k] < 0) ? -mquant : mquant;
2786 s->block_last_index[n] = i;
2791 /** Decode intra block in inter frames - more generic version than vc1_decode_i_block
2792 * @param v VC1Context
2793 * @param block block to decode
2794 * @param[in] n subblock index
2795 * @param coded are AC coeffs present or not
2796 * @param mquant block quantizer
2797 * @param codingset set of VLC to decode data
2799 static int vc1_decode_intra_block(VC1Context *v, DCTELEM block[64], int n, int coded, int mquant, int codingset)
2801 GetBitContext *gb = &v->s.gb;
2802 MpegEncContext *s = &v->s;
2803 int dc_pred_dir = 0; /* Direction of the DC prediction used */
2806 int16_t *ac_val, *ac_val2;
2808 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2809 int a_avail = v->a_avail, c_avail = v->c_avail;
2810 int use_pred = s->ac_pred;
2814 /* XXX: Guard against dumb values of mquant */
2815 mquant = (mquant < 1) ? 0 : ( (mquant>31) ? 31 : mquant );
2817 /* Set DC scale - y and c use the same */
2818 s->y_dc_scale = s->y_dc_scale_table[mquant];
2819 s->c_dc_scale = s->c_dc_scale_table[mquant];
2821 /* Get DC differential */
2823 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2825 dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2828 av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2833 if (dcdiff == 119 /* ESC index value */)
2835 /* TODO: Optimize */
2836 if (mquant == 1) dcdiff = get_bits(gb, 10);
2837 else if (mquant == 2) dcdiff = get_bits(gb, 9);
2838 else dcdiff = get_bits(gb, 8);
2843 dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
2844 else if (mquant == 2)
2845 dcdiff = (dcdiff<<1) + get_bits1(gb) - 1;
2852 dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
2855 /* Store the quantized DC coeff, used for prediction */
2858 block[0] = dcdiff * s->y_dc_scale;
2860 block[0] = dcdiff * s->c_dc_scale;
2869 /* check if AC is needed at all and adjust direction if needed */
2870 if(!a_avail) dc_pred_dir = 1;
2871 if(!c_avail) dc_pred_dir = 0;
2872 if(!a_avail && !c_avail) use_pred = 0;
2873 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
2876 scale = mquant * 2 + v->halfpq;
2878 if(dc_pred_dir) //left
2881 ac_val -= 16 * s->block_wrap[n];
2883 q1 = s->current_picture.qscale_table[mb_pos];
2884 if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
2885 if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
2886 if(dc_pred_dir && n==1) q2 = q1;
2887 if(!dc_pred_dir && n==2) q2 = q1;
2891 int last = 0, skip, value;
2892 const int8_t *zz_table;
2895 zz_table = wmv1_scantable[0];
2898 vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2902 block[zz_table[i++]] = value;
2905 /* apply AC prediction if needed */
2907 /* scale predictors if needed*/
2909 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2910 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2912 if(dc_pred_dir) { //left
2913 for(k = 1; k < 8; k++)
2914 block[k << 3] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2916 for(k = 1; k < 8; k++)
2917 block[k] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2920 if(dc_pred_dir) { //left
2921 for(k = 1; k < 8; k++)
2922 block[k << 3] += ac_val[k];
2924 for(k = 1; k < 8; k++)
2925 block[k] += ac_val[k + 8];
2929 /* save AC coeffs for further prediction */
2930 for(k = 1; k < 8; k++) {
2931 ac_val2[k] = block[k << 3];
2932 ac_val2[k + 8] = block[k];
2935 /* scale AC coeffs */
2936 for(k = 1; k < 64; k++)
2940 block[k] += (block[k] < 0) ? -mquant : mquant;
2943 if(use_pred) i = 63;
2944 } else { // no AC coeffs
2947 memset(ac_val2, 0, 16 * 2);
2948 if(dc_pred_dir) {//left
2950 memcpy(ac_val2, ac_val, 8 * 2);
2952 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2953 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2954 for(k = 1; k < 8; k++)
2955 ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2960 memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
2962 q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
2963 q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
2964 for(k = 1; k < 8; k++)
2965 ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
2970 /* apply AC prediction if needed */
2972 if(dc_pred_dir) { //left
2973 for(k = 1; k < 8; k++) {
2974 block[k << 3] = ac_val2[k] * scale;
2975 if(!v->pquantizer && block[k << 3])
2976 block[k << 3] += (block[k << 3] < 0) ? -mquant : mquant;
2979 for(k = 1; k < 8; k++) {
2980 block[k] = ac_val2[k + 8] * scale;
2981 if(!v->pquantizer && block[k])
2982 block[k] += (block[k] < 0) ? -mquant : mquant;
2988 s->block_last_index[n] = i;
2995 static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n, int mquant, int ttmb, int first_block,
2996 uint8_t *dst, int linesize, int skip_block, int apply_filter, int cbp_top, int cbp_left)
2998 MpegEncContext *s = &v->s;
2999 GetBitContext *gb = &s->gb;
3002 int scale, off, idx, last, skip, value;
3003 int ttblk = ttmb & 7;
3007 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)];
3009 if(ttblk == TT_4X4) {
3010 subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
3012 if((ttblk != TT_8X8 && ttblk != TT_4X4) && (v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))) {
3013 subblkpat = decode012(gb);
3014 if(subblkpat) subblkpat ^= 3; //swap decoded pattern bits
3015 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) ttblk = TT_8X4;
3016 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) ttblk = TT_4X8;
3018 scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);
3020 // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
3021 if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
3022 subblkpat = 2 - (ttblk == TT_8X4_TOP);
3025 if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
3026 subblkpat = 2 - (ttblk == TT_4X8_LEFT);
3035 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3039 idx = wmv1_scantable[0][i++];
3040 block[idx] = value * scale;
3042 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3045 s->dsp.vc1_inv_trans_8x8(block);
3046 s->dsp.add_pixels_clamped(block, dst, linesize);
3047 if(apply_filter && cbp_top & 0xC)
3048 vc1_loop_filter(dst, 1, linesize, 8, mquant);
3049 if(apply_filter && cbp_left & 0xA)
3050 vc1_loop_filter(dst, linesize, 1, 8, mquant);
3054 pat = ~subblkpat & 0xF;
3055 for(j = 0; j < 4; j++) {
3056 last = subblkpat & (1 << (3 - j));
3058 off = (j & 1) * 4 + (j & 2) * 16;
3060 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3064 idx = ff_vc1_simple_progressive_4x4_zz[i++];
3065 block[idx + off] = value * scale;
3067 block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
3069 if(!(subblkpat & (1 << (3 - j))) && !skip_block){
3070 s->dsp.vc1_inv_trans_4x4(dst + (j&1)*4 + (j&2)*2*linesize, linesize, block + off);
3071 if(apply_filter && (j&2 ? pat & (1<<(j-2)) : (cbp_top & (1 << (j + 2)))))
3072 vc1_loop_filter(dst + (j&1)*4 + (j&2)*2*linesize, 1, linesize, 4, mquant);
3073 if(apply_filter && (j&1 ? pat & (1<<(j-1)) : (cbp_left & (1 << (j + 1)))))
3074 vc1_loop_filter(dst + (j&1)*4 + (j&2)*2*linesize, linesize, 1, 4, mquant);
3079 pat = ~((subblkpat & 2)*6 + (subblkpat & 1)*3) & 0xF;
3080 for(j = 0; j < 2; j++) {
3081 last = subblkpat & (1 << (1 - j));
3085 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3089 idx = v->zz_8x4[i++]+off;
3090 block[idx] = value * scale;
3092 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3094 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
3095 s->dsp.vc1_inv_trans_8x4(dst + j*4*linesize, linesize, block + off);
3096 if(apply_filter && j ? pat & 0x3 : (cbp_top & 0xC))
3097 vc1_loop_filter(dst + j*4*linesize, 1, linesize, 8, mquant);
3098 if(apply_filter && cbp_left & (2 << j))
3099 vc1_loop_filter(dst + j*4*linesize, linesize, 1, 4, mquant);
3104 pat = ~(subblkpat*5) & 0xF;
3105 for(j = 0; j < 2; j++) {
3106 last = subblkpat & (1 << (1 - j));
3110 vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
3114 idx = v->zz_4x8[i++]+off;
3115 block[idx] = value * scale;
3117 block[idx] += (block[idx] < 0) ? -mquant : mquant;
3119 if(!(subblkpat & (1 << (1 - j))) && !skip_block){
3120 s->dsp.vc1_inv_trans_4x8(dst + j*4, linesize, block + off);
3121 if(apply_filter && cbp_top & (2 << j))
3122 vc1_loop_filter(dst + j*4, 1, linesize, 4, mquant);
3123 if(apply_filter && j ? pat & 0x5 : (cbp_left & 0xA))
3124 vc1_loop_filter(dst + j*4, linesize, 1, 8, mquant);
3132 /** @} */ // Macroblock group
3134 static const int size_table [6] = { 0, 2, 3, 4, 5, 8 };
3135 static const int offset_table[6] = { 0, 1, 3, 7, 15, 31 };
3137 /** Decode one P-frame MB (in Simple/Main profile)
3139 static int vc1_decode_p_mb(VC1Context *v)
3141 MpegEncContext *s = &v->s;
3142 GetBitContext *gb = &s->gb;
3144 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
3145 int cbp; /* cbp decoding stuff */
3146 int mqdiff, mquant; /* MB quantization */
3147 int ttmb = v->ttfrm; /* MB Transform type */
3149 int mb_has_coeffs = 1; /* last_flag */
3150 int dmv_x, dmv_y; /* Differential MV components */
3151 int index, index1; /* LUT indexes */
3152 int val, sign; /* temp values */
3153 int first_block = 1;
3155 int skipped, fourmv;
3156 int block_cbp = 0, pat;
3157 int apply_loop_filter;
3159 mquant = v->pq; /* Loosy initialization */
3161 if (v->mv_type_is_raw)
3162 fourmv = get_bits1(gb);
3164 fourmv = v->mv_type_mb_plane[mb_pos];
3166 skipped = get_bits1(gb);
3168 skipped = v->s.mbskip_table[mb_pos];
3170 s->dsp.clear_blocks(s->block[0]);
3172 apply_loop_filter = s->loop_filter && !(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY);
3173 if (!fourmv) /* 1MV mode */
3177 GET_MVDATA(dmv_x, dmv_y);
3180 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
3181 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
3183 s->current_picture.mb_type[mb_pos] = s->mb_intra ? MB_TYPE_INTRA : MB_TYPE_16x16;
3184 vc1_pred_mv(s, 0, dmv_x, dmv_y, 1, v->range_x, v->range_y, v->mb_type[0]);
3186 /* FIXME Set DC val for inter block ? */
3187 if (s->mb_intra && !mb_has_coeffs)
3190 s->ac_pred = get_bits1(gb);
3193 else if (mb_has_coeffs)
3195 if (s->mb_intra) s->ac_pred = get_bits1(gb);
3196 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
3204 s->current_picture.qscale_table[mb_pos] = mquant;
3206 if (!v->ttmbf && !s->mb_intra && mb_has_coeffs)
3207 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table,
3208 VC1_TTMB_VLC_BITS, 2);
3209 if(!s->mb_intra) vc1_mc_1mv(v, 0);
3213 s->dc_val[0][s->block_index[i]] = 0;
3215 val = ((cbp >> (5 - i)) & 1);
3216 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
3217 v->mb_type[0][s->block_index[i]] = s->mb_intra;
3219 /* check if prediction blocks A and C are available */
3220 v->a_avail = v->c_avail = 0;
3221 if(i == 2 || i == 3 || !s->first_slice_line)
3222 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
3223 if(i == 1 || i == 3 || s->mb_x)
3224 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
3226 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
3227 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
3228 s->dsp.vc1_inv_trans_8x8(s->block[i]);
3229 if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
3230 s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
3231 if(v->pq >= 9 && v->overlap) {
3233 s->dsp.vc1_h_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
3235 s->dsp.vc1_v_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
3237 if(apply_loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
3238 int left_cbp, top_cbp;
3240 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
3241 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
3243 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
3244 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
3247 vc1_loop_filter(s->dest[dst_idx] + off, 1, i & 4 ? s->uvlinesize : s->linesize, 8, mquant);
3249 vc1_loop_filter(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, 1, 8, mquant);
3251 block_cbp |= 0xF << (i << 2);
3253 int left_cbp = 0, top_cbp = 0, filter = 0;
3254 if(apply_loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
3257 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
3258 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
3260 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
3261 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
3264 vc1_loop_filter(s->dest[dst_idx] + off, 1, i & 4 ? s->uvlinesize : s->linesize, 8, mquant);
3266 vc1_loop_filter(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, 1, 8, mquant);
3268 pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block, s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize, (i&4) && (s->flags & CODEC_FLAG_GRAY), filter, left_cbp, top_cbp);
3269 block_cbp |= pat << (i << 2);
3270 if(!v->ttmbf && ttmb < 8) ttmb = -1;
3278 for(i = 0; i < 6; i++) {
3279 v->mb_type[0][s->block_index[i]] = 0;
3280 s->dc_val[0][s->block_index[i]] = 0;
3282 s->current_picture.mb_type[mb_pos] = MB_TYPE_SKIP;
3283 s->current_picture.qscale_table[mb_pos] = 0;
3284 vc1_pred_mv(s, 0, 0, 0, 1, v->range_x, v->range_y, v->mb_type[0]);
3291 if (!skipped /* unskipped MB */)
3293 int intra_count = 0, coded_inter = 0;
3294 int is_intra[6], is_coded[6];
3296 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
3299 val = ((cbp >> (5 - i)) & 1);
3300 s->dc_val[0][s->block_index[i]] = 0;
3307 GET_MVDATA(dmv_x, dmv_y);
3309 vc1_pred_mv(s, i, dmv_x, dmv_y, 0, v->range_x, v->range_y, v->mb_type[0]);
3310 if(!s->mb_intra) vc1_mc_4mv_luma(v, i);
3311 intra_count += s->mb_intra;
3312 is_intra[i] = s->mb_intra;
3313 is_coded[i] = mb_has_coeffs;
3316 is_intra[i] = (intra_count >= 3);
3319 if(i == 4) vc1_mc_4mv_chroma(v);
3320 v->mb_type[0][s->block_index[i]] = is_intra[i];
3321 if(!coded_inter) coded_inter = !is_intra[i] & is_coded[i];
3323 // if there are no coded blocks then don't do anything more
3324 if(!intra_count && !coded_inter) return 0;
3327 s->current_picture.qscale_table[mb_pos] = mquant;
3328 /* test if block is intra and has pred */
3333 if(((!s->first_slice_line || (i==2 || i==3)) && v->mb_type[0][s->block_index[i] - s->block_wrap[i]])
3334 || ((s->mb_x || (i==1 || i==3)) && v->mb_type[0][s->block_index[i] - 1])) {
3339 if(intrapred)s->ac_pred = get_bits1(gb);
3340 else s->ac_pred = 0;
3342 if (!v->ttmbf && coded_inter)
3343 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
3347 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
3348 s->mb_intra = is_intra[i];
3350 /* check if prediction blocks A and C are available */
3351 v->a_avail = v->c_avail = 0;
3352 if(i == 2 || i == 3 || !s->first_slice_line)
3353 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
3354 if(i == 1 || i == 3 || s->mb_x)
3355 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
3357 vc1_decode_intra_block(v, s->block[i], i, is_coded[i], mquant, (i&4)?v->codingset2:v->codingset);
3358 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
3359 s->dsp.vc1_inv_trans_8x8(s->block[i]);
3360 if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
3361 s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize);
3362 if(v->pq >= 9 && v->overlap) {
3364 s->dsp.vc1_h_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
3366 s->dsp.vc1_v_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
3368 if(v->s.loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
3369 int left_cbp, top_cbp;
3371 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
3372 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
3374 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
3375 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
3378 vc1_loop_filter(s->dest[dst_idx] + off, 1, i & 4 ? s->uvlinesize : s->linesize, 8, mquant);
3380 vc1_loop_filter(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, 1, 8, mquant);
3382 block_cbp |= 0xF << (i << 2);
3383 } else if(is_coded[i]) {
3384 int left_cbp = 0, top_cbp = 0, filter = 0;
3385 if(v->s.loop_filter && s->mb_x && s->mb_x != (s->mb_width - 1) && s->mb_y && s->mb_y != (s->mb_height - 1)){
3388 left_cbp = v->cbp[s->mb_x - 1] >> (i * 4);
3389 top_cbp = v->cbp[s->mb_x - s->mb_stride] >> (i * 4);
3391 left_cbp = (i & 1) ? (cbp >> ((i-1)*4)) : (v->cbp[s->mb_x - 1] >> ((i+1)*4));
3392 top_cbp = (i & 2) ? (cbp >> ((i-2)*4)) : (v->cbp[s->mb_x - s->mb_stride] >> ((i+2)*4));
3395 vc1_loop_filter(s->dest[dst_idx] + off, 1, i & 4 ? s->uvlinesize : s->linesize, 8, mquant);
3397 vc1_loop_filter(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize, 1, 8, mquant);
3399 pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block, s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize, (i&4) && (s->flags & CODEC_FLAG_GRAY), filter, left_cbp, top_cbp);
3400 block_cbp |= pat << (i << 2);
3401 if(!v->ttmbf && ttmb < 8) ttmb = -1;
3410 s->current_picture.qscale_table[mb_pos] = 0;
3411 for (i=0; i<6; i++) {
3412 v->mb_type[0][s->block_index[i]] = 0;
3413 s->dc_val[0][s->block_index[i]] = 0;
3417 vc1_pred_mv(s, i, 0, 0, 0, v->range_x, v->range_y, v->mb_type[0]);
3418 vc1_mc_4mv_luma(v, i);
3420 vc1_mc_4mv_chroma(v);
3421 s->current_picture.qscale_table[mb_pos] = 0;
3425 v->cbp[s->mb_x] = block_cbp;
3427 /* Should never happen */
3431 /** Decode one B-frame MB (in Main profile)
3433 static void vc1_decode_b_mb(VC1Context *v)
3435 MpegEncContext *s = &v->s;
3436 GetBitContext *gb = &s->gb;
3438 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
3439 int cbp = 0; /* cbp decoding stuff */
3440 int mqdiff, mquant; /* MB quantization */
3441 int ttmb = v->ttfrm; /* MB Transform type */
3442 int mb_has_coeffs = 0; /* last_flag */
3443 int index, index1; /* LUT indexes */
3444 int val, sign; /* temp values */
3445 int first_block = 1;
3447 int skipped, direct;
3448 int dmv_x[2], dmv_y[2];
3449 int bmvtype = BMV_TYPE_BACKWARD;
3451 mquant = v->pq; /* Loosy initialization */
3455 direct = get_bits1(gb);
3457 direct = v->direct_mb_plane[mb_pos];
3459 skipped = get_bits1(gb);
3461 skipped = v->s.mbskip_table[mb_pos];
3463 s->dsp.clear_blocks(s->block[0]);
3464 dmv_x[0] = dmv_x[1] = dmv_y[0] = dmv_y[1] = 0;
3465 for(i = 0; i < 6; i++) {
3466 v->mb_type[0][s->block_index[i]] = 0;
3467 s->dc_val[0][s->block_index[i]] = 0;
3469 s->current_picture.qscale_table[mb_pos] = 0;
3473 GET_MVDATA(dmv_x[0], dmv_y[0]);
3474 dmv_x[1] = dmv_x[0];
3475 dmv_y[1] = dmv_y[0];
3477 if(skipped || !s->mb_intra) {
3478 bmvtype = decode012(gb);
3481 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_BACKWARD : BMV_TYPE_FORWARD;
3484 bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_FORWARD : BMV_TYPE_BACKWARD;
3487 bmvtype = BMV_TYPE_INTERPOLATED;
3488 dmv_x[0] = dmv_y[0] = 0;
3492 for(i = 0; i < 6; i++)
3493 v->mb_type[0][s->block_index[i]] = s->mb_intra;
3496 if(direct) bmvtype = BMV_TYPE_INTERPOLATED;
3497 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
3498 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
3502 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
3506 s->current_picture.qscale_table[mb_pos] = mquant;
3508 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
3509 dmv_x[0] = dmv_y[0] = dmv_x[1] = dmv_y[1] = 0;
3510 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
3511 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
3513 if(!mb_has_coeffs && !s->mb_intra) {
3514 /* no coded blocks - effectively skipped */
3515 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
3516 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
3519 if(s->mb_intra && !mb_has_coeffs) {
3521 s->current_picture.qscale_table[mb_pos] = mquant;
3522 s->ac_pred = get_bits1(gb);
3524 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
3526 if(bmvtype == BMV_TYPE_INTERPOLATED) {
3527 GET_MVDATA(dmv_x[0], dmv_y[0]);
3528 if(!mb_has_coeffs) {
3529 /* interpolated skipped block */
3530 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
3531 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
3535 vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
3537 vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
3540 s->ac_pred = get_bits1(gb);
3541 cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
3543 s->current_picture.qscale_table[mb_pos] = mquant;
3544 if(!v->ttmbf && !s->mb_intra && mb_has_coeffs)
3545 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
3551 s->dc_val[0][s->block_index[i]] = 0;
3553 val = ((cbp >> (5 - i)) & 1);
3554 off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
3555 v->mb_type[0][s->block_index[i]] = s->mb_intra;
3557 /* check if prediction blocks A and C are available */
3558 v->a_avail = v->c_avail = 0;
3559 if(i == 2 || i == 3 || !s->first_slice_line)
3560 v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
3561 if(i == 1 || i == 3 || s->mb_x)
3562 v->c_avail = v->mb_type[0][s->block_index[i] - 1];
3564 vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
3565 if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
3566 s->dsp.vc1_inv_trans_8x8(s->block[i]);
3567 if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
3568 s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
3570 vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block, s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize, (i&4) && (s->flags & CODEC_FLAG_GRAY), 0, 0, 0);
3571 if(!v->ttmbf && ttmb < 8) ttmb = -1;
3577 /** Decode blocks of I-frame
3579 static void vc1_decode_i_blocks(VC1Context *v)
3582 MpegEncContext *s = &v->s;
3587 /* select codingmode used for VLC tables selection */
3588 switch(v->y_ac_table_index){
3590 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
3593 v->codingset = CS_HIGH_MOT_INTRA;
3596 v->codingset = CS_MID_RATE_INTRA;
3600 switch(v->c_ac_table_index){
3602 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3605 v->codingset2 = CS_HIGH_MOT_INTER;
3608 v->codingset2 = CS_MID_RATE_INTER;
3612 /* Set DC scale - y and c use the same */
3613 s->y_dc_scale = s->y_dc_scale_table[v->pq];
3614 s->c_dc_scale = s->c_dc_scale_table[v->pq];
3617 s->mb_x = s->mb_y = 0;
3619 s->first_slice_line = 1;
3620 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
3621 for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
3622 ff_init_block_index(s);
3623 ff_update_block_index(s);
3624 s->dsp.clear_blocks(s->block[0]);
3625 mb_pos = s->mb_x + s->mb_y * s->mb_width;
3626 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
3627 s->current_picture.qscale_table[mb_pos] = v->pq;
3628 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
3629 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
3631 // do actual MB decoding and displaying
3632 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
3633 v->s.ac_pred = get_bits1(&v->s.gb);
3635 for(k = 0; k < 6; k++) {
3636 val = ((cbp >> (5 - k)) & 1);
3639 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
3643 cbp |= val << (5 - k);
3645 vc1_decode_i_block(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2);
3647 s->dsp.vc1_inv_trans_8x8(s->block[k]);
3648 if(v->pq >= 9 && v->overlap) {
3649 for(j = 0; j < 64; j++) s->block[k][j] += 128;
3653 vc1_put_block(v, s->block);
3654 if(v->pq >= 9 && v->overlap) {
3656 s->dsp.vc1_h_overlap(s->dest[0], s->linesize);
3657 s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
3658 if(!(s->flags & CODEC_FLAG_GRAY)) {
3659 s->dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
3660 s->dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
3663 s->dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
3664 s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
3665 if(!s->first_slice_line) {
3666 s->dsp.vc1_v_overlap(s->dest[0], s->linesize);
3667 s->dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
3668 if(!(s->flags & CODEC_FLAG_GRAY)) {
3669 s->dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
3670 s->dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
3673 s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
3674 s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
3676 if(v->s.loop_filter) vc1_loop_filter_iblk(s, s->current_picture.qscale_table[mb_pos]);
3678 if(get_bits_count(&s->gb) > v->bits) {
3679 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3680 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
3684 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3685 s->first_slice_line = 0;
3687 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
3690 /** Decode blocks of I-frame for advanced profile
3692 static void vc1_decode_i_blocks_adv(VC1Context *v)
3695 MpegEncContext *s = &v->s;
3702 GetBitContext *gb = &s->gb;
3704 /* select codingmode used for VLC tables selection */
3705 switch(v->y_ac_table_index){
3707 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
3710 v->codingset = CS_HIGH_MOT_INTRA;
3713 v->codingset = CS_MID_RATE_INTRA;
3717 switch(v->c_ac_table_index){
3719 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3722 v->codingset2 = CS_HIGH_MOT_INTER;
3725 v->codingset2 = CS_MID_RATE_INTER;
3730 s->mb_x = s->mb_y = 0;
3732 s->first_slice_line = 1;
3733 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
3734 for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
3735 ff_init_block_index(s);
3736 ff_update_block_index(s);
3737 s->dsp.clear_blocks(s->block[0]);
3738 mb_pos = s->mb_x + s->mb_y * s->mb_stride;
3739 s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
3740 s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
3741 s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
3743 // do actual MB decoding and displaying
3744 cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
3745 if(v->acpred_is_raw)
3746 v->s.ac_pred = get_bits1(&v->s.gb);
3748 v->s.ac_pred = v->acpred_plane[mb_pos];
3750 if(v->condover == CONDOVER_SELECT) {
3751 if(v->overflg_is_raw)
3752 overlap = get_bits1(&v->s.gb);
3754 overlap = v->over_flags_plane[mb_pos];
3756 overlap = (v->condover == CONDOVER_ALL);
3760 s->current_picture.qscale_table[mb_pos] = mquant;
3761 /* Set DC scale - y and c use the same */
3762 s->y_dc_scale = s->y_dc_scale_table[mquant];
3763 s->c_dc_scale = s->c_dc_scale_table[mquant];
3765 for(k = 0; k < 6; k++) {
3766 val = ((cbp >> (5 - k)) & 1);
3769 int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
3773 cbp |= val << (5 - k);
3775 v->a_avail = !s->first_slice_line || (k==2 || k==3);
3776 v->c_avail = !!s->mb_x || (k==1 || k==3);
3778 vc1_decode_i_block_adv(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant);
3780 s->dsp.vc1_inv_trans_8x8(s->block[k]);
3781 for(j = 0; j < 64; j++) s->block[k][j] += 128;
3784 vc1_put_block(v, s->block);
3787 s->dsp.vc1_h_overlap(s->dest[0], s->linesize);
3788 s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
3789 if(!(s->flags & CODEC_FLAG_GRAY)) {
3790 s->dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
3791 s->dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
3794 s->dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
3795 s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
3796 if(!s->first_slice_line) {
3797 s->dsp.vc1_v_overlap(s->dest[0], s->linesize);
3798 s->dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
3799 if(!(s->flags & CODEC_FLAG_GRAY)) {
3800 s->dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
3801 s->dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
3804 s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
3805 s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
3807 if(v->s.loop_filter) vc1_loop_filter_iblk(s, s->current_picture.qscale_table[mb_pos]);
3809 if(get_bits_count(&s->gb) > v->bits) {
3810 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3811 av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
3815 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3816 s->first_slice_line = 0;
3818 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
3821 static void vc1_decode_p_blocks(VC1Context *v)
3823 MpegEncContext *s = &v->s;
3825 /* select codingmode used for VLC tables selection */
3826 switch(v->c_ac_table_index){
3828 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
3831 v->codingset = CS_HIGH_MOT_INTRA;
3834 v->codingset = CS_MID_RATE_INTRA;
3838 switch(v->c_ac_table_index){
3840 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3843 v->codingset2 = CS_HIGH_MOT_INTER;
3846 v->codingset2 = CS_MID_RATE_INTER;
3850 s->first_slice_line = 1;
3851 memset(v->cbp_base, 0, sizeof(v->cbp_base[0])*2*s->mb_stride);
3852 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
3853 for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
3854 ff_init_block_index(s);
3855 ff_update_block_index(s);
3856 s->dsp.clear_blocks(s->block[0]);
3859 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
3860 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3861 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);
3865 memmove(v->cbp_base, v->cbp, sizeof(v->cbp_base[0])*s->mb_stride);
3866 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3867 s->first_slice_line = 0;
3869 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
3872 static void vc1_decode_b_blocks(VC1Context *v)
3874 MpegEncContext *s = &v->s;
3876 /* select codingmode used for VLC tables selection */
3877 switch(v->c_ac_table_index){
3879 v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
3882 v->codingset = CS_HIGH_MOT_INTRA;
3885 v->codingset = CS_MID_RATE_INTRA;
3889 switch(v->c_ac_table_index){
3891 v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
3894 v->codingset2 = CS_HIGH_MOT_INTER;
3897 v->codingset2 = CS_MID_RATE_INTER;
3901 s->first_slice_line = 1;
3902 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
3903 for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
3904 ff_init_block_index(s);
3905 ff_update_block_index(s);
3906 s->dsp.clear_blocks(s->block[0]);
3909 if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
3910 ff_er_add_slice(s, 0, 0, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END));
3911 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);
3914 if(v->s.loop_filter) vc1_loop_filter_iblk(s, s->current_picture.qscale_table[s->mb_x + s->mb_y *s->mb_stride]);
3916 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3917 s->first_slice_line = 0;
3919 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
3922 static void vc1_decode_skip_blocks(VC1Context *v)
3924 MpegEncContext *s = &v->s;
3926 ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
3927 s->first_slice_line = 1;
3928 for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
3930 ff_init_block_index(s);
3931 ff_update_block_index(s);
3932 memcpy(s->dest[0], s->last_picture.data[0] + s->mb_y * 16 * s->linesize, s->linesize * 16);
3933 memcpy(s->dest[1], s->last_picture.data[1] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
3934 memcpy(s->dest[2], s->last_picture.data[2] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
3935 ff_draw_horiz_band(s, s->mb_y * 16, 16);
3936 s->first_slice_line = 0;
3938 s->pict_type = FF_P_TYPE;
3941 static void vc1_decode_blocks(VC1Context *v)
3944 v->s.esc3_level_length = 0;
3946 ff_intrax8_decode_picture(&v->x8, 2*v->pq+v->halfpq, v->pq*(!v->pquantizer) );
3949 switch(v->s.pict_type) {
3951 if(v->profile == PROFILE_ADVANCED)
3952 vc1_decode_i_blocks_adv(v);
3954 vc1_decode_i_blocks(v);
3957 if(v->p_frame_skipped)
3958 vc1_decode_skip_blocks(v);
3960 vc1_decode_p_blocks(v);
3964 if(v->profile == PROFILE_ADVANCED)
3965 vc1_decode_i_blocks_adv(v);
3967 vc1_decode_i_blocks(v);
3969 vc1_decode_b_blocks(v);
3975 /** Find VC-1 marker in buffer
3976 * @return position where next marker starts or end of buffer if no marker found
3978 static av_always_inline const uint8_t* find_next_marker(const uint8_t *src, const uint8_t *end)
3980 uint32_t mrk = 0xFFFFFFFF;
3982 if(end-src < 4) return end;
3984 mrk = (mrk << 8) | *src++;
3991 static av_always_inline int vc1_unescape_buffer(const uint8_t *src, int size, uint8_t *dst)
3996 for(dsize = 0; dsize < size; dsize++) *dst++ = *src++;
3999 for(i = 0; i < size; i++, src++) {
4000 if(src[0] == 3 && i >= 2 && !src[-1] && !src[-2] && i < size-1 && src[1] < 4) {
4001 dst[dsize++] = src[1];
4005 dst[dsize++] = *src;
4010 /** Initialize a VC1/WMV3 decoder
4011 * @todo TODO: Handle VC-1 IDUs (Transport level?)
4012 * @todo TODO: Decypher remaining bits in extra_data
4014 static av_cold int vc1_decode_init(AVCodecContext *avctx)
4016 VC1Context *v = avctx->priv_data;
4017 MpegEncContext *s = &v->s;
4020 if (!avctx->extradata_size || !avctx->extradata) return -1;
4021 if (!(avctx->flags & CODEC_FLAG_GRAY))
4022 avctx->pix_fmt = avctx->get_format(avctx, avctx->codec->pix_fmts);
4024 avctx->pix_fmt = PIX_FMT_GRAY8;
4025 avctx->hwaccel = ff_find_hwaccel(avctx->codec->id, avctx->pix_fmt);
4027 avctx->flags |= CODEC_FLAG_EMU_EDGE;
4028 v->s.flags |= CODEC_FLAG_EMU_EDGE;
4030 if(avctx->idct_algo==FF_IDCT_AUTO){
4031 avctx->idct_algo=FF_IDCT_WMV2;
4034 if(ff_h263_decode_init(avctx) < 0)
4036 if (vc1_init_common(v) < 0) return -1;
4038 avctx->coded_width = avctx->width;
4039 avctx->coded_height = avctx->height;
4040 if (avctx->codec_id == CODEC_ID_WMV3)
4044 // looks like WMV3 has a sequence header stored in the extradata
4045 // advanced sequence header may be before the first frame
4046 // the last byte of the extradata is a version number, 1 for the
4047 // samples we can decode
4049 init_get_bits(&gb, avctx->extradata, avctx->extradata_size*8);
4051 if (decode_sequence_header(avctx, &gb) < 0)
4054 count = avctx->extradata_size*8 - get_bits_count(&gb);
4057 av_log(avctx, AV_LOG_INFO, "Extra data: %i bits left, value: %X\n",
4058 count, get_bits(&gb, count));
4062 av_log(avctx, AV_LOG_INFO, "Read %i bits in overflow\n", -count);
4064 } else { // VC1/WVC1
4065 const uint8_t *start = avctx->extradata;
4066 uint8_t *end = avctx->extradata + avctx->extradata_size;
4067 const uint8_t *next;
4068 int size, buf2_size;
4069 uint8_t *buf2 = NULL;
4070 int seq_initialized = 0, ep_initialized = 0;
4072 if(avctx->extradata_size < 16) {
4073 av_log(avctx, AV_LOG_ERROR, "Extradata size too small: %i\n", avctx->extradata_size);
4077 buf2 = av_mallocz(avctx->extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
4078 if(start[0]) start++; // in WVC1 extradata first byte is its size
4080 for(; next < end; start = next){
4081 next = find_next_marker(start + 4, end);
4082 size = next - start - 4;
4083 if(size <= 0) continue;
4084 buf2_size = vc1_unescape_buffer(start + 4, size, buf2);
4085 init_get_bits(&gb, buf2, buf2_size * 8);
4086 switch(AV_RB32(start)){
4087 case VC1_CODE_SEQHDR:
4088 if(decode_sequence_header(avctx, &gb) < 0){
4092 seq_initialized = 1;
4094 case VC1_CODE_ENTRYPOINT:
4095 if(decode_entry_point(avctx, &gb) < 0){
4104 if(!seq_initialized || !ep_initialized){
4105 av_log(avctx, AV_LOG_ERROR, "Incomplete extradata\n");
4109 avctx->has_b_frames= !!(avctx->max_b_frames);
4110 s->low_delay = !avctx->has_b_frames;
4112 s->mb_width = (avctx->coded_width+15)>>4;
4113 s->mb_height = (avctx->coded_height+15)>>4;
4115 /* Allocate mb bitplanes */
4116 v->mv_type_mb_plane = av_malloc(s->mb_stride * s->mb_height);
4117 v->direct_mb_plane = av_malloc(s->mb_stride * s->mb_height);
4118 v->acpred_plane = av_malloc(s->mb_stride * s->mb_height);
4119 v->over_flags_plane = av_malloc(s->mb_stride * s->mb_height);
4121 v->cbp_base = av_malloc(sizeof(v->cbp_base[0]) * 2 * s->mb_stride);
4122 v->cbp = v->cbp_base + s->mb_stride;
4124 /* allocate block type info in that way so it could be used with s->block_index[] */
4125 v->mb_type_base = av_malloc(s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride * (s->mb_height + 1) * 2);
4126 v->mb_type[0] = v->mb_type_base + s->b8_stride + 1;
4127 v->mb_type[1] = v->mb_type_base + s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride + 1;
4128 v->mb_type[2] = v->mb_type[1] + s->mb_stride * (s->mb_height + 1);
4130 /* Init coded blocks info */
4131 if (v->profile == PROFILE_ADVANCED)
4133 // if (alloc_bitplane(&v->over_flags_plane, s->mb_width, s->mb_height) < 0)
4135 // if (alloc_bitplane(&v->ac_pred_plane, s->mb_width, s->mb_height) < 0)
4139 ff_intrax8_common_init(&v->x8,s);
4144 /** Decode a VC1/WMV3 frame
4145 * @todo TODO: Handle VC-1 IDUs (Transport level?)
4147 static int vc1_decode_frame(AVCodecContext *avctx,
4148 void *data, int *data_size,
4151 const uint8_t *buf = avpkt->data;
4152 int buf_size = avpkt->size;
4153 VC1Context *v = avctx->priv_data;
4154 MpegEncContext *s = &v->s;
4155 AVFrame *pict = data;
4156 uint8_t *buf2 = NULL;
4157 const uint8_t *buf_start = buf;
4159 /* no supplementary picture */
4160 if (buf_size == 0) {
4161 /* special case for last picture */
4162 if (s->low_delay==0 && s->next_picture_ptr) {
4163 *pict= *(AVFrame*)s->next_picture_ptr;
4164 s->next_picture_ptr= NULL;
4166 *data_size = sizeof(AVFrame);
4172 /* We need to set current_picture_ptr before reading the header,
4173 * otherwise we cannot store anything in there. */
4174 if(s->current_picture_ptr==NULL || s->current_picture_ptr->data[0]){
4175 int i= ff_find_unused_picture(s, 0);
4176 s->current_picture_ptr= &s->picture[i];
4179 if (s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
4180 if (v->profile < PROFILE_ADVANCED)
4181 avctx->pix_fmt = PIX_FMT_VDPAU_WMV3;
4183 avctx->pix_fmt = PIX_FMT_VDPAU_VC1;
4186 //for advanced profile we may need to parse and unescape data
4187 if (avctx->codec_id == CODEC_ID_VC1) {
4189 buf2 = av_mallocz(buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
4191 if(IS_MARKER(AV_RB32(buf))){ /* frame starts with marker and needs to be parsed */
4192 const uint8_t *start, *end, *next;
4196 for(start = buf, end = buf + buf_size; next < end; start = next){
4197 next = find_next_marker(start + 4, end);
4198 size = next - start - 4;
4199 if(size <= 0) continue;
4200 switch(AV_RB32(start)){
4201 case VC1_CODE_FRAME:
4202 if (avctx->hwaccel ||
4203 s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
4205 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
4207 case VC1_CODE_ENTRYPOINT: /* it should be before frame data */
4208 buf_size2 = vc1_unescape_buffer(start + 4, size, buf2);
4209 init_get_bits(&s->gb, buf2, buf_size2*8);
4210 decode_entry_point(avctx, &s->gb);
4212 case VC1_CODE_SLICE:
4213 av_log(avctx, AV_LOG_ERROR, "Sliced decoding is not implemented (yet)\n");
4218 }else if(v->interlace && ((buf[0] & 0xC0) == 0xC0)){ /* WVC1 interlaced stores both fields divided by marker */
4219 const uint8_t *divider;
4221 divider = find_next_marker(buf, buf + buf_size);
4222 if((divider == (buf + buf_size)) || AV_RB32(divider) != VC1_CODE_FIELD){
4223 av_log(avctx, AV_LOG_ERROR, "Error in WVC1 interlaced frame\n");
4228 buf_size2 = vc1_unescape_buffer(buf, divider - buf, buf2);
4230 av_free(buf2);return -1;
4232 buf_size2 = vc1_unescape_buffer(buf, buf_size, buf2);
4234 init_get_bits(&s->gb, buf2, buf_size2*8);
4236 init_get_bits(&s->gb, buf, buf_size*8);
4237 // do parse frame header
4238 if(v->profile < PROFILE_ADVANCED) {
4239 if(vc1_parse_frame_header(v, &s->gb) == -1) {
4244 if(vc1_parse_frame_header_adv(v, &s->gb) == -1) {
4250 if(s->pict_type != FF_I_TYPE && !v->res_rtm_flag){
4256 s->current_picture.pict_type= s->pict_type;
4257 s->current_picture.key_frame= s->pict_type == FF_I_TYPE;
4259 /* skip B-frames if we don't have reference frames */
4260 if(s->last_picture_ptr==NULL && (s->pict_type==FF_B_TYPE || s->dropable)){
4262 return -1;//buf_size;
4264 /* skip b frames if we are in a hurry */
4265 if(avctx->hurry_up && s->pict_type==FF_B_TYPE) return -1;//buf_size;
4266 if( (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type==FF_B_TYPE)
4267 || (avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type!=FF_I_TYPE)
4268 || avctx->skip_frame >= AVDISCARD_ALL) {
4272 /* skip everything if we are in a hurry>=5 */
4273 if(avctx->hurry_up>=5) {
4275 return -1;//buf_size;
4278 if(s->next_p_frame_damaged){
4279 if(s->pict_type==FF_B_TYPE)
4282 s->next_p_frame_damaged=0;
4285 if(MPV_frame_start(s, avctx) < 0) {
4290 s->me.qpel_put= s->dsp.put_qpel_pixels_tab;
4291 s->me.qpel_avg= s->dsp.avg_qpel_pixels_tab;
4293 if ((CONFIG_VC1_VDPAU_DECODER || CONFIG_WMV3_VDPAU_DECODER)
4294 &&s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
4295 ff_vdpau_vc1_decode_picture(s, buf_start, (buf + buf_size) - buf_start);
4296 else if (avctx->hwaccel) {
4297 if (avctx->hwaccel->start_frame(avctx, buf, buf_size) < 0)
4299 if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
4301 if (avctx->hwaccel->end_frame(avctx) < 0)
4304 ff_er_frame_start(s);
4306 v->bits = buf_size * 8;
4307 vc1_decode_blocks(v);
4308 //av_log(s->avctx, AV_LOG_INFO, "Consumed %i/%i bits\n", get_bits_count(&s->gb), buf_size*8);
4309 // if(get_bits_count(&s->gb) > buf_size * 8)
4316 assert(s->current_picture.pict_type == s->current_picture_ptr->pict_type);
4317 assert(s->current_picture.pict_type == s->pict_type);
4318 if (s->pict_type == FF_B_TYPE || s->low_delay) {
4319 *pict= *(AVFrame*)s->current_picture_ptr;
4320 } else if (s->last_picture_ptr != NULL) {
4321 *pict= *(AVFrame*)s->last_picture_ptr;
4324 if(s->last_picture_ptr || s->low_delay){
4325 *data_size = sizeof(AVFrame);
4326 ff_print_debug_info(s, pict);
4329 /* Return the Picture timestamp as the frame number */
4330 /* we subtract 1 because it is added on utils.c */
4331 avctx->frame_number = s->picture_number - 1;
4338 /** Close a VC1/WMV3 decoder
4339 * @warning Initial try at using MpegEncContext stuff
4341 static av_cold int vc1_decode_end(AVCodecContext *avctx)
4343 VC1Context *v = avctx->priv_data;
4345 av_freep(&v->hrd_rate);
4346 av_freep(&v->hrd_buffer);
4347 MPV_common_end(&v->s);
4348 av_freep(&v->mv_type_mb_plane);
4349 av_freep(&v->direct_mb_plane);
4350 av_freep(&v->acpred_plane);
4351 av_freep(&v->over_flags_plane);
4352 av_freep(&v->mb_type_base);
4353 av_freep(&v->cbp_base);
4354 ff_intrax8_common_end(&v->x8);
4359 AVCodec vc1_decoder = {
4370 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1"),
4371 .pix_fmts = ff_hwaccel_pixfmt_list_420
4374 AVCodec wmv3_decoder = {
4385 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9"),
4386 .pix_fmts = ff_hwaccel_pixfmt_list_420
4389 #if CONFIG_WMV3_VDPAU_DECODER
4390 AVCodec wmv3_vdpau_decoder = {
4399 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4401 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 VDPAU"),
4402 .pix_fmts = (enum PixelFormat[]){PIX_FMT_VDPAU_WMV3, PIX_FMT_NONE}
4406 #if CONFIG_VC1_VDPAU_DECODER
4407 AVCodec vc1_vdpau_decoder = {
4416 CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4418 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1 VDPAU"),
4419 .pix_fmts = (enum PixelFormat[]){PIX_FMT_VDPAU_VC1, PIX_FMT_NONE}