2 * Copyright (c) 2003 The Libav Project
4 * This file is part of Libav.
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 * How to use this decoder:
23 * SVQ3 data is transported within Apple Quicktime files. Quicktime files
24 * have stsd atoms to describe media trak properties. A stsd atom for a
25 * video trak contains 1 or more ImageDescription atoms. These atoms begin
26 * with the 4-byte length of the atom followed by the codec fourcc. Some
27 * decoders need information in this atom to operate correctly. Such
28 * is the case with SVQ3. In order to get the best use out of this decoder,
29 * the calling app must make the SVQ3 ImageDescription atom available
30 * via the AVCodecContext's extradata[_size] field:
32 * AVCodecContext.extradata = pointer to ImageDescription, first characters
33 * are expected to be 'S', 'V', 'Q', and '3', NOT the 4-byte atom length
34 * AVCodecContext.extradata_size = size of ImageDescription atom memory
35 * buffer (which will be the same as the ImageDescription atom size field
36 * from the QT file, minus 4 bytes since the length is missing)
38 * You will know you have these parameters passed correctly when the decoder
39 * correctly decodes this file:
40 * http://samples.libav.org/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov
45 #include "mpegvideo.h"
48 #include "h264data.h" //FIXME FIXME FIXME
50 #include "h264_mvpred.h"
52 #include "rectangle.h"
53 #include "vdpau_internal.h"
72 uint32_t watermark_key;
75 #define FULLPEL_MODE 1
76 #define HALFPEL_MODE 2
77 #define THIRDPEL_MODE 3
78 #define PREDICT_MODE 4
80 /* dual scan (from some older h264 draft)
89 static const uint8_t svq3_scan[16] = {
90 0+0*4, 1+0*4, 2+0*4, 2+1*4,
91 2+2*4, 3+0*4, 3+1*4, 3+2*4,
92 0+1*4, 0+2*4, 1+1*4, 1+2*4,
93 0+3*4, 1+3*4, 2+3*4, 3+3*4,
96 static const uint8_t svq3_pred_0[25][2] = {
99 { 0, 2 }, { 1, 1 }, { 2, 0 },
100 { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },
101 { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },
102 { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },
103 { 2, 4 }, { 3, 3 }, { 4, 2 },
108 static const int8_t svq3_pred_1[6][6][5] = {
109 { { 2,-1,-1,-1,-1 }, { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 },
110 { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 }, { 1, 2,-1,-1,-1 } },
111 { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 4, 3 }, { 0, 1, 2, 4, 3 },
112 { 0, 2, 1, 4, 3 }, { 2, 0, 1, 3, 4 }, { 0, 4, 2, 1, 3 } },
113 { { 2, 0,-1,-1,-1 }, { 2, 1, 0, 4, 3 }, { 1, 2, 4, 0, 3 },
114 { 2, 1, 0, 4, 3 }, { 2, 1, 4, 3, 0 }, { 1, 2, 4, 0, 3 } },
115 { { 2, 0,-1,-1,-1 }, { 2, 0, 1, 4, 3 }, { 1, 2, 0, 4, 3 },
116 { 2, 1, 0, 4, 3 }, { 2, 1, 3, 4, 0 }, { 2, 4, 1, 0, 3 } },
117 { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 3, 4 }, { 1, 2, 3, 0, 4 },
118 { 2, 0, 1, 3, 4 }, { 2, 1, 3, 0, 4 }, { 2, 0, 4, 3, 1 } },
119 { { 0, 2,-1,-1,-1 }, { 0, 2, 4, 1, 3 }, { 1, 4, 2, 0, 3 },
120 { 4, 2, 0, 1, 3 }, { 2, 0, 1, 4, 3 }, { 4, 2, 1, 0, 3 } },
123 static const struct { uint8_t run; uint8_t level; } svq3_dct_tables[2][16] = {
124 { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },
125 { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },
126 { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },
127 { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }
130 static const uint32_t svq3_dequant_coeff[32] = {
131 3881, 4351, 4890, 5481, 6154, 6914, 7761, 8718,
132 9781, 10987, 12339, 13828, 15523, 17435, 19561, 21873,
133 24552, 27656, 30847, 34870, 38807, 43747, 49103, 54683,
134 61694, 68745, 77615, 89113,100253,109366,126635,141533
137 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp){
138 const int qmul = svq3_dequant_coeff[qp];
142 static const uint8_t x_offset[4]={0, 1*stride, 4*stride, 5*stride};
145 const int z0 = 13*(input[4*i+0] + input[4*i+2]);
146 const int z1 = 13*(input[4*i+0] - input[4*i+2]);
147 const int z2 = 7* input[4*i+1] - 17*input[4*i+3];
148 const int z3 = 17* input[4*i+1] + 7*input[4*i+3];
157 const int offset= x_offset[i];
158 const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
159 const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
160 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
161 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
163 output[stride* 0+offset] = ((z0 + z3)*qmul + 0x80000) >> 20;
164 output[stride* 2+offset] = ((z1 + z2)*qmul + 0x80000) >> 20;
165 output[stride* 8+offset] = ((z1 - z2)*qmul + 0x80000) >> 20;
166 output[stride*10+offset] = ((z0 - z3)*qmul + 0x80000) >> 20;
171 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp,
174 const int qmul = svq3_dequant_coeff[qp];
178 dc = 13*13*((dc == 1) ? 1538*block[0] : ((qmul*(block[0] >> 3)) / 2));
182 for (i = 0; i < 4; i++) {
183 const int z0 = 13*(block[0 + 4*i] + block[2 + 4*i]);
184 const int z1 = 13*(block[0 + 4*i] - block[2 + 4*i]);
185 const int z2 = 7* block[1 + 4*i] - 17*block[3 + 4*i];
186 const int z3 = 17* block[1 + 4*i] + 7*block[3 + 4*i];
188 block[0 + 4*i] = z0 + z3;
189 block[1 + 4*i] = z1 + z2;
190 block[2 + 4*i] = z1 - z2;
191 block[3 + 4*i] = z0 - z3;
194 for (i = 0; i < 4; i++) {
195 const int z0 = 13*(block[i + 4*0] + block[i + 4*2]);
196 const int z1 = 13*(block[i + 4*0] - block[i + 4*2]);
197 const int z2 = 7* block[i + 4*1] - 17*block[i + 4*3];
198 const int z3 = 17* block[i + 4*1] + 7*block[i + 4*3];
199 const int rr = (dc + 0x80000);
201 dst[i + stride*0] = av_clip_uint8( dst[i + stride*0] + (((z0 + z3)*qmul + rr) >> 20) );
202 dst[i + stride*1] = av_clip_uint8( dst[i + stride*1] + (((z1 + z2)*qmul + rr) >> 20) );
203 dst[i + stride*2] = av_clip_uint8( dst[i + stride*2] + (((z1 - z2)*qmul + rr) >> 20) );
204 dst[i + stride*3] = av_clip_uint8( dst[i + stride*3] + (((z0 - z3)*qmul + rr) >> 20) );
208 static inline int svq3_decode_block(GetBitContext *gb, DCTELEM *block,
209 int index, const int type)
211 static const uint8_t *const scan_patterns[4] =
212 { luma_dc_zigzag_scan, zigzag_scan, svq3_scan, chroma_dc_scan };
214 int run, level, sign, vlc, limit;
215 const int intra = (3 * type) >> 2;
216 const uint8_t *const scan = scan_patterns[type];
218 for (limit = (16 >> intra); index < 16; index = limit, limit += 8) {
219 for (; (vlc = svq3_get_ue_golomb(gb)) != 0; index++) {
221 if (vlc == INVALID_VLC)
224 sign = (vlc & 0x1) - 1;
225 vlc = (vlc + 1) >> 1;
231 } else if (vlc < 4) {
236 level = ((vlc + 9) >> 2) - run;
240 run = svq3_dct_tables[intra][vlc].run;
241 level = svq3_dct_tables[intra][vlc].level;
244 level = (vlc >> 3) + ((run == 0) ? 8 : ((run < 2) ? 2 : ((run < 5) ? 0 : -1)));
247 level = (vlc >> 4) + ((run == 0) ? 4 : ((run < 3) ? 2 : ((run < 10) ? 1 : 0)));
251 if ((index += run) >= limit)
254 block[scan[index]] = (level ^ sign) - sign;
265 static inline void svq3_mc_dir_part(MpegEncContext *s,
266 int x, int y, int width, int height,
267 int mx, int my, int dxy,
268 int thirdpel, int dir, int avg)
270 const Picture *pic = (dir == 0) ? &s->last_picture : &s->next_picture;
273 int blocksize = 2 - (width>>3); //16->0, 8->1, 4->2
278 if (mx < 0 || mx >= (s->h_edge_pos - width - 1) ||
279 my < 0 || my >= (s->v_edge_pos - height - 1)) {
281 if ((s->flags & CODEC_FLAG_EMU_EDGE)) {
285 mx = av_clip (mx, -16, (s->h_edge_pos - width + 15));
286 my = av_clip (my, -16, (s->v_edge_pos - height + 15));
289 /* form component predictions */
290 dest = s->current_picture.f.data[0] + x + y*s->linesize;
291 src = pic->f.data[0] + mx + my*s->linesize;
294 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, (width + 1), (height + 1),
295 mx, my, s->h_edge_pos, s->v_edge_pos);
296 src = s->edge_emu_buffer;
299 (avg ? s->dsp.avg_tpel_pixels_tab : s->dsp.put_tpel_pixels_tab)[dxy](dest, src, s->linesize, width, height);
301 (avg ? s->dsp.avg_pixels_tab : s->dsp.put_pixels_tab)[blocksize][dxy](dest, src, s->linesize, height);
303 if (!(s->flags & CODEC_FLAG_GRAY)) {
304 mx = (mx + (mx < (int) x)) >> 1;
305 my = (my + (my < (int) y)) >> 1;
306 width = (width >> 1);
307 height = (height >> 1);
310 for (i = 1; i < 3; i++) {
311 dest = s->current_picture.f.data[i] + (x >> 1) + (y >> 1) * s->uvlinesize;
312 src = pic->f.data[i] + mx + my * s->uvlinesize;
315 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->uvlinesize, (width + 1), (height + 1),
316 mx, my, (s->h_edge_pos >> 1), (s->v_edge_pos >> 1));
317 src = s->edge_emu_buffer;
320 (avg ? s->dsp.avg_tpel_pixels_tab : s->dsp.put_tpel_pixels_tab)[dxy](dest, src, s->uvlinesize, width, height);
322 (avg ? s->dsp.avg_pixels_tab : s->dsp.put_pixels_tab)[blocksize][dxy](dest, src, s->uvlinesize, height);
327 static inline int svq3_mc_dir(H264Context *h, int size, int mode, int dir,
330 int i, j, k, mx, my, dx, dy, x, y;
331 MpegEncContext *const s = (MpegEncContext *) h;
332 const int part_width = ((size & 5) == 4) ? 4 : 16 >> (size & 1);
333 const int part_height = 16 >> ((unsigned) (size + 1) / 3);
334 const int extra_width = (mode == PREDICT_MODE) ? -16*6 : 0;
335 const int h_edge_pos = 6*(s->h_edge_pos - part_width ) - extra_width;
336 const int v_edge_pos = 6*(s->v_edge_pos - part_height) - extra_width;
338 for (i = 0; i < 16; i += part_height) {
339 for (j = 0; j < 16; j += part_width) {
340 const int b_xy = (4*s->mb_x + (j >> 2)) + (4*s->mb_y + (i >> 2))*h->b_stride;
344 k = ((j >> 2) & 1) + ((i >> 1) & 2) + ((j >> 1) & 4) + (i & 8);
346 if (mode != PREDICT_MODE) {
347 pred_motion(h, k, (part_width >> 2), dir, 1, &mx, &my);
349 mx = s->next_picture.f.motion_val[0][b_xy][0] << 1;
350 my = s->next_picture.f.motion_val[0][b_xy][1] << 1;
353 mx = ((mx * h->frame_num_offset) / h->prev_frame_num_offset + 1) >> 1;
354 my = ((my * h->frame_num_offset) / h->prev_frame_num_offset + 1) >> 1;
356 mx = ((mx * (h->frame_num_offset - h->prev_frame_num_offset)) / h->prev_frame_num_offset + 1) >> 1;
357 my = ((my * (h->frame_num_offset - h->prev_frame_num_offset)) / h->prev_frame_num_offset + 1) >> 1;
361 /* clip motion vector prediction to frame border */
362 mx = av_clip(mx, extra_width - 6*x, h_edge_pos - 6*x);
363 my = av_clip(my, extra_width - 6*y, v_edge_pos - 6*y);
365 /* get (optional) motion vector differential */
366 if (mode == PREDICT_MODE) {
369 dy = svq3_get_se_golomb(&s->gb);
370 dx = svq3_get_se_golomb(&s->gb);
372 if (dx == INVALID_VLC || dy == INVALID_VLC) {
373 av_log(h->s.avctx, AV_LOG_ERROR, "invalid MV vlc\n");
378 /* compute motion vector */
379 if (mode == THIRDPEL_MODE) {
381 mx = ((mx + 1)>>1) + dx;
382 my = ((my + 1)>>1) + dy;
383 fx = ((unsigned)(mx + 0x3000))/3 - 0x1000;
384 fy = ((unsigned)(my + 0x3000))/3 - 0x1000;
385 dxy = (mx - 3*fx) + 4*(my - 3*fy);
387 svq3_mc_dir_part(s, x, y, part_width, part_height, fx, fy, dxy, 1, dir, avg);
390 } else if (mode == HALFPEL_MODE || mode == PREDICT_MODE) {
391 mx = ((unsigned)(mx + 1 + 0x3000))/3 + dx - 0x1000;
392 my = ((unsigned)(my + 1 + 0x3000))/3 + dy - 0x1000;
393 dxy = (mx&1) + 2*(my&1);
395 svq3_mc_dir_part(s, x, y, part_width, part_height, mx>>1, my>>1, dxy, 0, dir, avg);
399 mx = ((unsigned)(mx + 3 + 0x6000))/6 + dx - 0x1000;
400 my = ((unsigned)(my + 3 + 0x6000))/6 + dy - 0x1000;
402 svq3_mc_dir_part(s, x, y, part_width, part_height, mx, my, 0, 0, dir, avg);
407 /* update mv_cache */
408 if (mode != PREDICT_MODE) {
409 int32_t mv = pack16to32(mx,my);
411 if (part_height == 8 && i < 8) {
412 *(int32_t *) h->mv_cache[dir][scan8[k] + 1*8] = mv;
414 if (part_width == 8 && j < 8) {
415 *(int32_t *) h->mv_cache[dir][scan8[k] + 1 + 1*8] = mv;
418 if (part_width == 8 && j < 8) {
419 *(int32_t *) h->mv_cache[dir][scan8[k] + 1] = mv;
421 if (part_width == 4 || part_height == 4) {
422 *(int32_t *) h->mv_cache[dir][scan8[k]] = mv;
426 /* write back motion vectors */
427 fill_rectangle(s->current_picture.f.motion_val[dir][b_xy],
428 part_width >> 2, part_height >> 2, h->b_stride,
429 pack16to32(mx, my), 4);
436 static int svq3_decode_mb(SVQ3Context *svq3, unsigned int mb_type)
438 H264Context *h = &svq3->h;
439 int i, j, k, m, dir, mode;
443 MpegEncContext *const s = (MpegEncContext *) h;
444 const int mb_xy = h->mb_xy;
445 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
447 h->top_samples_available = (s->mb_y == 0) ? 0x33FF : 0xFFFF;
448 h->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF;
449 h->topright_samples_available = 0xFFFF;
451 if (mb_type == 0) { /* SKIP */
452 if (s->pict_type == AV_PICTURE_TYPE_P || s->next_picture.f.mb_type[mb_xy] == -1) {
453 svq3_mc_dir_part(s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0, 0, 0, 0);
455 if (s->pict_type == AV_PICTURE_TYPE_B) {
456 svq3_mc_dir_part(s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0, 0, 1, 1);
459 mb_type = MB_TYPE_SKIP;
461 mb_type = FFMIN(s->next_picture.f.mb_type[mb_xy], 6);
462 if (svq3_mc_dir(h, mb_type, PREDICT_MODE, 0, 0) < 0)
464 if (svq3_mc_dir(h, mb_type, PREDICT_MODE, 1, 1) < 0)
467 mb_type = MB_TYPE_16x16;
469 } else if (mb_type < 8) { /* INTER */
470 if (svq3->thirdpel_flag && svq3->halfpel_flag == !get_bits1 (&s->gb)) {
471 mode = THIRDPEL_MODE;
472 } else if (svq3->halfpel_flag && svq3->thirdpel_flag == !get_bits1 (&s->gb)) {
479 /* note ref_cache should contain here:
487 for (m = 0; m < 2; m++) {
488 if (s->mb_x > 0 && h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1]+6] != -1) {
489 for (i = 0; i < 4; i++) {
490 *(uint32_t *) h->mv_cache[m][scan8[0] - 1 + i*8] = *(uint32_t *) s->current_picture.f.motion_val[m][b_xy - 1 + i*h->b_stride];
493 for (i = 0; i < 4; i++) {
494 *(uint32_t *) h->mv_cache[m][scan8[0] - 1 + i*8] = 0;
498 memcpy(h->mv_cache[m][scan8[0] - 1*8], s->current_picture.f.motion_val[m][b_xy - h->b_stride], 4*2*sizeof(int16_t));
499 memset(&h->ref_cache[m][scan8[0] - 1*8], (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1, 4);
501 if (s->mb_x < (s->mb_width - 1)) {
502 *(uint32_t *) h->mv_cache[m][scan8[0] + 4 - 1*8] = *(uint32_t *) s->current_picture.f.motion_val[m][b_xy - h->b_stride + 4];
503 h->ref_cache[m][scan8[0] + 4 - 1*8] =
504 (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride + 1]+6] == -1 ||
505 h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride ] ] == -1) ? PART_NOT_AVAILABLE : 1;
507 h->ref_cache[m][scan8[0] + 4 - 1*8] = PART_NOT_AVAILABLE;
509 *(uint32_t *) h->mv_cache[m][scan8[0] - 1 - 1*8] = *(uint32_t *) s->current_picture.f.motion_val[m][b_xy - h->b_stride - 1];
510 h->ref_cache[m][scan8[0] - 1 - 1*8] = (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride - 1]+3] == -1) ? PART_NOT_AVAILABLE : 1;
512 h->ref_cache[m][scan8[0] - 1 - 1*8] = PART_NOT_AVAILABLE;
514 memset(&h->ref_cache[m][scan8[0] - 1*8 - 1], PART_NOT_AVAILABLE, 8);
516 if (s->pict_type != AV_PICTURE_TYPE_B)
520 /* decode motion vector(s) and form prediction(s) */
521 if (s->pict_type == AV_PICTURE_TYPE_P) {
522 if (svq3_mc_dir(h, (mb_type - 1), mode, 0, 0) < 0)
524 } else { /* AV_PICTURE_TYPE_B */
526 if (svq3_mc_dir(h, 0, mode, 0, 0) < 0)
529 for (i = 0; i < 4; i++) {
530 memset(s->current_picture.f.motion_val[0][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
534 if (svq3_mc_dir(h, 0, mode, 1, (mb_type == 3)) < 0)
537 for (i = 0; i < 4; i++) {
538 memset(s->current_picture.f.motion_val[1][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
543 mb_type = MB_TYPE_16x16;
544 } else if (mb_type == 8 || mb_type == 33) { /* INTRA4x4 */
545 memset(h->intra4x4_pred_mode_cache, -1, 8*5*sizeof(int8_t));
549 for (i = 0; i < 4; i++) {
550 h->intra4x4_pred_mode_cache[scan8[0] - 1 + i*8] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1]+6-i];
552 if (h->intra4x4_pred_mode_cache[scan8[0] - 1] == -1) {
553 h->left_samples_available = 0x5F5F;
557 h->intra4x4_pred_mode_cache[4+8*0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride]+0];
558 h->intra4x4_pred_mode_cache[5+8*0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride]+1];
559 h->intra4x4_pred_mode_cache[6+8*0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride]+2];
560 h->intra4x4_pred_mode_cache[7+8*0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride]+3];
562 if (h->intra4x4_pred_mode_cache[4+8*0] == -1) {
563 h->top_samples_available = 0x33FF;
567 /* decode prediction codes for luma blocks */
568 for (i = 0; i < 16; i+=2) {
569 vlc = svq3_get_ue_golomb(&s->gb);
572 av_log(h->s.avctx, AV_LOG_ERROR, "luma prediction:%d\n", vlc);
576 left = &h->intra4x4_pred_mode_cache[scan8[i] - 1];
577 top = &h->intra4x4_pred_mode_cache[scan8[i] - 8];
579 left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]];
580 left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]];
582 if (left[1] == -1 || left[2] == -1){
583 av_log(h->s.avctx, AV_LOG_ERROR, "weird prediction\n");
587 } else { /* mb_type == 33, DC_128_PRED block type */
588 for (i = 0; i < 4; i++) {
589 memset(&h->intra4x4_pred_mode_cache[scan8[0] + 8*i], DC_PRED, 4);
593 write_back_intra_pred_mode(h);
596 ff_h264_check_intra4x4_pred_mode(h);
598 h->top_samples_available = (s->mb_y == 0) ? 0x33FF : 0xFFFF;
599 h->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF;
601 for (i = 0; i < 4; i++) {
602 memset(&h->intra4x4_pred_mode_cache[scan8[0] + 8*i], DC_128_PRED, 4);
605 h->top_samples_available = 0x33FF;
606 h->left_samples_available = 0x5F5F;
609 mb_type = MB_TYPE_INTRA4x4;
610 } else { /* INTRA16x16 */
611 dir = i_mb_type_info[mb_type - 8].pred_mode;
612 dir = (dir >> 1) ^ 3*(dir & 1) ^ 1;
614 if ((h->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, dir, 0)) == -1){
615 av_log(h->s.avctx, AV_LOG_ERROR, "check_intra_pred_mode = -1\n");
619 cbp = i_mb_type_info[mb_type - 8].cbp;
620 mb_type = MB_TYPE_INTRA16x16;
623 if (!IS_INTER(mb_type) && s->pict_type != AV_PICTURE_TYPE_I) {
624 for (i = 0; i < 4; i++) {
625 memset(s->current_picture.f.motion_val[0][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
627 if (s->pict_type == AV_PICTURE_TYPE_B) {
628 for (i = 0; i < 4; i++) {
629 memset(s->current_picture.f.motion_val[1][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
633 if (!IS_INTRA4x4(mb_type)) {
634 memset(h->intra4x4_pred_mode+h->mb2br_xy[mb_xy], DC_PRED, 8);
636 if (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B) {
637 memset(h->non_zero_count_cache + 8, 0, 14*8*sizeof(uint8_t));
638 s->dsp.clear_blocks(h->mb+ 0);
639 s->dsp.clear_blocks(h->mb+384);
642 if (!IS_INTRA16x16(mb_type) && (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B)) {
643 if ((vlc = svq3_get_ue_golomb(&s->gb)) >= 48){
644 av_log(h->s.avctx, AV_LOG_ERROR, "cbp_vlc=%d\n", vlc);
648 cbp = IS_INTRA(mb_type) ? golomb_to_intra4x4_cbp[vlc] : golomb_to_inter_cbp[vlc];
650 if (IS_INTRA16x16(mb_type) || (s->pict_type != AV_PICTURE_TYPE_I && s->adaptive_quant && cbp)) {
651 s->qscale += svq3_get_se_golomb(&s->gb);
653 if (s->qscale > 31u){
654 av_log(h->s.avctx, AV_LOG_ERROR, "qscale:%d\n", s->qscale);
658 if (IS_INTRA16x16(mb_type)) {
659 AV_ZERO128(h->mb_luma_dc[0]+0);
660 AV_ZERO128(h->mb_luma_dc[0]+8);
661 if (svq3_decode_block(&s->gb, h->mb_luma_dc, 0, 1)){
662 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding intra luma dc\n");
668 const int index = IS_INTRA16x16(mb_type) ? 1 : 0;
669 const int type = ((s->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1);
671 for (i = 0; i < 4; i++) {
672 if ((cbp & (1 << i))) {
673 for (j = 0; j < 4; j++) {
674 k = index ? ((j&1) + 2*(i&1) + 2*(j&2) + 4*(i&2)) : (4*i + j);
675 h->non_zero_count_cache[ scan8[k] ] = 1;
677 if (svq3_decode_block(&s->gb, &h->mb[16*k], index, type)){
678 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding block\n");
686 for (i = 1; i < 3; ++i) {
687 if (svq3_decode_block(&s->gb, &h->mb[16*16*i], 0, 3)){
688 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding chroma dc block\n");
694 for (i = 1; i < 3; i++) {
695 for (j = 0; j < 4; j++) {
697 h->non_zero_count_cache[ scan8[k] ] = 1;
699 if (svq3_decode_block(&s->gb, &h->mb[16*k], 1, 1)){
700 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding chroma ac block\n");
710 s->current_picture.f.mb_type[mb_xy] = mb_type;
712 if (IS_INTRA(mb_type)) {
713 h->chroma_pred_mode = ff_h264_check_intra_pred_mode(h, DC_PRED8x8, 1);
719 static int svq3_decode_slice_header(AVCodecContext *avctx)
721 SVQ3Context *svq3 = avctx->priv_data;
722 H264Context *h = &svq3->h;
723 MpegEncContext *s = &h->s;
724 const int mb_xy = h->mb_xy;
727 header = get_bits(&s->gb, 8);
729 if (((header & 0x9F) != 1 && (header & 0x9F) != 2) || (header & 0x60) == 0) {
731 av_log(avctx, AV_LOG_ERROR, "unsupported slice header (%02X)\n", header);
734 int length = (header >> 5) & 3;
736 svq3->next_slice_index = get_bits_count(&s->gb) + 8*show_bits(&s->gb, 8*length) + 8*length;
738 if (svq3->next_slice_index > s->gb.size_in_bits) {
739 av_log(avctx, AV_LOG_ERROR, "slice after bitstream end\n");
743 s->gb.size_in_bits = svq3->next_slice_index - 8*(length - 1);
744 skip_bits(&s->gb, 8);
746 if (svq3->watermark_key) {
747 uint32_t header = AV_RL32(&s->gb.buffer[(get_bits_count(&s->gb)>>3)+1]);
748 AV_WL32(&s->gb.buffer[(get_bits_count(&s->gb)>>3)+1], header ^ svq3->watermark_key);
751 memcpy((uint8_t *) &s->gb.buffer[get_bits_count(&s->gb) >> 3],
752 &s->gb.buffer[s->gb.size_in_bits >> 3], (length - 1));
754 skip_bits_long(&s->gb, 0);
757 if ((i = svq3_get_ue_golomb(&s->gb)) == INVALID_VLC || i >= 3){
758 av_log(h->s.avctx, AV_LOG_ERROR, "illegal slice type %d \n", i);
762 h->slice_type = golomb_to_pict_type[i];
764 if ((header & 0x9F) == 2) {
765 i = (s->mb_num < 64) ? 6 : (1 + av_log2 (s->mb_num - 1));
766 s->mb_skip_run = get_bits(&s->gb, i) - (s->mb_x + (s->mb_y * s->mb_width));
772 h->slice_num = get_bits(&s->gb, 8);
773 s->qscale = get_bits(&s->gb, 5);
774 s->adaptive_quant = get_bits1(&s->gb);
779 if (svq3->unknown_flag) {
784 skip_bits(&s->gb, 2);
786 while (get_bits1(&s->gb)) {
787 skip_bits(&s->gb, 8);
790 /* reset intra predictors and invalidate motion vector references */
792 memset(h->intra4x4_pred_mode+h->mb2br_xy[mb_xy - 1 ]+3, -1, 4*sizeof(int8_t));
793 memset(h->intra4x4_pred_mode+h->mb2br_xy[mb_xy - s->mb_x] , -1, 8*sizeof(int8_t)*s->mb_x);
796 memset(h->intra4x4_pred_mode+h->mb2br_xy[mb_xy - s->mb_stride], -1, 8*sizeof(int8_t)*(s->mb_width - s->mb_x));
799 h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - s->mb_stride - 1]+3] = -1;
806 static av_cold int svq3_decode_init(AVCodecContext *avctx)
808 SVQ3Context *svq3 = avctx->priv_data;
809 H264Context *h = &svq3->h;
810 MpegEncContext *s = &h->s;
812 unsigned char *extradata;
813 unsigned char *extradata_end;
815 int marker_found = 0;
817 if (ff_h264_decode_init(avctx) < 0)
820 s->flags = avctx->flags;
821 s->flags2 = avctx->flags2;
822 s->unrestricted_mv = 1;
824 avctx->pix_fmt = avctx->codec->pix_fmts[0];
826 if (!s->context_initialized) {
827 h->chroma_qp[0] = h->chroma_qp[1] = 4;
829 svq3->halfpel_flag = 1;
830 svq3->thirdpel_flag = 1;
831 svq3->unknown_flag = 0;
833 /* prowl for the "SEQH" marker in the extradata */
834 extradata = (unsigned char *)avctx->extradata;
835 extradata_end = avctx->extradata + avctx->extradata_size;
837 for (m = 0; m + 8 < avctx->extradata_size; m++) {
838 if (!memcmp(extradata, "SEQH", 4)) {
846 /* if a match was found, parse the extra data */
852 size = AV_RB32(&extradata[4]);
853 if (size > extradata_end - extradata - 8)
854 return AVERROR_INVALIDDATA;
855 init_get_bits(&gb, extradata + 8, size*8);
857 /* 'frame size code' and optional 'width, height' */
858 frame_size_code = get_bits(&gb, 3);
859 switch (frame_size_code) {
860 case 0: avctx->width = 160; avctx->height = 120; break;
861 case 1: avctx->width = 128; avctx->height = 96; break;
862 case 2: avctx->width = 176; avctx->height = 144; break;
863 case 3: avctx->width = 352; avctx->height = 288; break;
864 case 4: avctx->width = 704; avctx->height = 576; break;
865 case 5: avctx->width = 240; avctx->height = 180; break;
866 case 6: avctx->width = 320; avctx->height = 240; break;
868 avctx->width = get_bits(&gb, 12);
869 avctx->height = get_bits(&gb, 12);
873 svq3->halfpel_flag = get_bits1(&gb);
874 svq3->thirdpel_flag = get_bits1(&gb);
882 s->low_delay = get_bits1(&gb);
887 while (get_bits1(&gb)) {
891 svq3->unknown_flag = get_bits1(&gb);
892 avctx->has_b_frames = !s->low_delay;
893 if (svq3->unknown_flag) {
895 unsigned watermark_width = svq3_get_ue_golomb(&gb);
896 unsigned watermark_height = svq3_get_ue_golomb(&gb);
897 int u1 = svq3_get_ue_golomb(&gb);
898 int u2 = get_bits(&gb, 8);
899 int u3 = get_bits(&gb, 2);
900 int u4 = svq3_get_ue_golomb(&gb);
901 unsigned long buf_len = watermark_width*watermark_height*4;
902 int offset = (get_bits_count(&gb)+7)>>3;
905 if ((uint64_t)watermark_width*4 > UINT_MAX/watermark_height)
908 buf = av_malloc(buf_len);
909 av_log(avctx, AV_LOG_DEBUG, "watermark size: %dx%d\n", watermark_width, watermark_height);
910 av_log(avctx, AV_LOG_DEBUG, "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n", u1, u2, u3, u4, offset);
911 if (uncompress(buf, &buf_len, extradata + 8 + offset, size - offset) != Z_OK) {
912 av_log(avctx, AV_LOG_ERROR, "could not uncompress watermark logo\n");
916 svq3->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);
917 svq3->watermark_key = svq3->watermark_key << 16 | svq3->watermark_key;
918 av_log(avctx, AV_LOG_DEBUG, "watermark key %#x\n", svq3->watermark_key);
921 av_log(avctx, AV_LOG_ERROR, "this svq3 file contains watermark which need zlib support compiled in\n");
927 s->width = avctx->width;
928 s->height = avctx->height;
930 if (ff_MPV_common_init(s) < 0)
933 h->b_stride = 4*s->mb_width;
935 if (ff_h264_alloc_tables(h) < 0) {
936 av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");
937 return AVERROR(ENOMEM);
944 static int svq3_decode_frame(AVCodecContext *avctx,
945 void *data, int *data_size,
948 const uint8_t *buf = avpkt->data;
949 SVQ3Context *svq3 = avctx->priv_data;
950 H264Context *h = &svq3->h;
951 MpegEncContext *s = &h->s;
952 int buf_size = avpkt->size;
955 /* special case for last picture */
957 if (s->next_picture_ptr && !s->low_delay) {
958 *(AVFrame *) data = s->next_picture.f;
959 s->next_picture_ptr = NULL;
960 *data_size = sizeof(AVFrame);
965 init_get_bits (&s->gb, buf, 8*buf_size);
967 s->mb_x = s->mb_y = h->mb_xy = 0;
969 if (svq3_decode_slice_header(avctx))
972 s->pict_type = h->slice_type;
973 s->picture_number = h->slice_num;
975 if (avctx->debug&FF_DEBUG_PICT_INFO){
976 av_log(h->s.avctx, AV_LOG_DEBUG, "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",
977 av_get_picture_type_char(s->pict_type), svq3->halfpel_flag, svq3->thirdpel_flag,
978 s->adaptive_quant, s->qscale, h->slice_num);
981 /* for skipping the frame */
982 s->current_picture.f.pict_type = s->pict_type;
983 s->current_picture.f.key_frame = (s->pict_type == AV_PICTURE_TYPE_I);
985 /* Skip B-frames if we do not have reference frames. */
986 if (s->last_picture_ptr == NULL && s->pict_type == AV_PICTURE_TYPE_B)
988 if ( (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
989 ||(avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
990 || avctx->skip_frame >= AVDISCARD_ALL)
993 if (s->next_p_frame_damaged) {
994 if (s->pict_type == AV_PICTURE_TYPE_B)
997 s->next_p_frame_damaged = 0;
1000 if (ff_h264_frame_start(h) < 0)
1003 if (s->pict_type == AV_PICTURE_TYPE_B) {
1004 h->frame_num_offset = (h->slice_num - h->prev_frame_num);
1006 if (h->frame_num_offset < 0) {
1007 h->frame_num_offset += 256;
1009 if (h->frame_num_offset == 0 || h->frame_num_offset >= h->prev_frame_num_offset) {
1010 av_log(h->s.avctx, AV_LOG_ERROR, "error in B-frame picture id\n");
1014 h->prev_frame_num = h->frame_num;
1015 h->frame_num = h->slice_num;
1016 h->prev_frame_num_offset = (h->frame_num - h->prev_frame_num);
1018 if (h->prev_frame_num_offset < 0) {
1019 h->prev_frame_num_offset += 256;
1023 for (m = 0; m < 2; m++){
1025 for (i = 0; i < 4; i++){
1027 for (j = -1; j < 4; j++)
1028 h->ref_cache[m][scan8[0] + 8*i + j]= 1;
1030 h->ref_cache[m][scan8[0] + 8*i + j]= PART_NOT_AVAILABLE;
1034 for (s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
1035 for (s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
1036 h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1038 if ( (get_bits_count(&s->gb) + 7) >= s->gb.size_in_bits &&
1039 ((get_bits_count(&s->gb) & 7) == 0 || show_bits(&s->gb, (-get_bits_count(&s->gb) & 7)) == 0)) {
1041 skip_bits(&s->gb, svq3->next_slice_index - get_bits_count(&s->gb));
1042 s->gb.size_in_bits = 8*buf_size;
1044 if (svq3_decode_slice_header(avctx))
1047 /* TODO: support s->mb_skip_run */
1050 mb_type = svq3_get_ue_golomb(&s->gb);
1052 if (s->pict_type == AV_PICTURE_TYPE_I) {
1054 } else if (s->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4) {
1057 if ((unsigned)mb_type > 33 || svq3_decode_mb(svq3, mb_type)) {
1058 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
1063 ff_h264_hl_decode_mb (h);
1066 if (s->pict_type != AV_PICTURE_TYPE_B && !s->low_delay) {
1067 s->current_picture.f.mb_type[s->mb_x + s->mb_y * s->mb_stride] =
1068 (s->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;
1072 ff_draw_horiz_band(s, 16*s->mb_y, 16);
1075 ff_MPV_frame_end(s);
1077 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1078 *(AVFrame *) data = s->current_picture.f;
1080 *(AVFrame *) data = s->last_picture.f;
1083 /* Do not output the last pic after seeking. */
1084 if (s->last_picture_ptr || s->low_delay) {
1085 *data_size = sizeof(AVFrame);
1091 static int svq3_decode_end(AVCodecContext *avctx)
1093 SVQ3Context *svq3 = avctx->priv_data;
1094 H264Context *h = &svq3->h;
1095 MpegEncContext *s = &h->s;
1097 ff_h264_free_context(h);
1099 ff_MPV_common_end(s);
1104 AVCodec ff_svq3_decoder = {
1106 .type = AVMEDIA_TYPE_VIDEO,
1107 .id = CODEC_ID_SVQ3,
1108 .priv_data_size = sizeof(SVQ3Context),
1109 .init = svq3_decode_init,
1110 .close = svq3_decode_end,
1111 .decode = svq3_decode_frame,
1112 .capabilities = CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 |
1114 .long_name = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),
1115 .pix_fmts = (const enum PixelFormat[]){ PIX_FMT_YUVJ420P, PIX_FMT_NONE },