2 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
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
21 #include "libavutil/intmath.h"
22 #include "libavutil/log.h"
23 #include "libavutil/opt.h"
30 #include "rangecoder.h"
33 #include "mpegvideo.h"
39 static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, int plane_index, int add, int mb_y){
40 Plane *p= &s->plane[plane_index];
41 const int mb_w= s->b_width << s->block_max_depth;
42 const int mb_h= s->b_height << s->block_max_depth;
44 int block_size = MB_SIZE >> s->block_max_depth;
45 int block_w = plane_index ? block_size/2 : block_size;
46 const uint8_t *obmc = plane_index ? ff_obmc_tab[s->block_max_depth+1] : ff_obmc_tab[s->block_max_depth];
47 int obmc_stride= plane_index ? block_size : 2*block_size;
48 int ref_stride= s->current_picture.linesize[plane_index];
49 uint8_t *dst8= s->current_picture.data[plane_index];
53 if(s->keyframe || (s->avctx->debug&512)){
58 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
59 // DWTELEM * line = slice_buffer_get_line(sb, y);
60 IDWTELEM * line = sb->line[y];
62 // int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
63 int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
65 if(v&(~255)) v= ~(v>>31);
66 dst8[x + y*ref_stride]= v;
70 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
71 // DWTELEM * line = slice_buffer_get_line(sb, y);
72 IDWTELEM * line = sb->line[y];
74 line[x] -= 128 << FRAC_BITS;
75 // buf[x + y*w]-= 128<<FRAC_BITS;
83 for(mb_x=0; mb_x<=mb_w; mb_x++){
84 add_yblock(s, 1, sb, old_buffer, dst8, obmc,
85 block_w*mb_x - block_w/2,
86 block_w*mb_y - block_w/2,
89 w, ref_stride, obmc_stride,
95 static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){
96 const int w= b->width;
98 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
99 int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
100 int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
103 if(b->ibuf == s->spatial_idwt_buffer || s->qlog == LOSSLESS_QLOG){
108 /* If we are on the second or later slice, restore our index. */
110 new_index = save_state[0];
113 for(y=start_y; y<h; y++){
116 IDWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
117 memset(line, 0, b->width*sizeof(IDWTELEM));
118 v = b->x_coeff[new_index].coeff;
119 x = b->x_coeff[new_index++].x;
121 register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT;
122 register int u= -(v&1);
125 v = b->x_coeff[new_index].coeff;
126 x = b->x_coeff[new_index++].x;
130 /* Save our variables for the next slice. */
131 save_state[0] = new_index;
136 static int decode_q_branch(SnowContext *s, int level, int x, int y){
137 const int w= s->b_width << s->block_max_depth;
138 const int rem_depth= s->block_max_depth - level;
139 const int index= (x + y*w) << rem_depth;
140 int trx= (x+1)<<rem_depth;
141 const BlockNode *left = x ? &s->block[index-1] : &null_block;
142 const BlockNode *top = y ? &s->block[index-w] : &null_block;
143 const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
144 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
145 int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
149 set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA);
153 if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
155 int l = left->color[0];
156 int cb= left->color[1];
157 int cr= left->color[2];
159 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
160 int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx));
161 int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my));
163 type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;
166 pred_mv(s, &mx, &my, 0, left, top, tr);
167 l += get_symbol(&s->c, &s->block_state[32], 1);
168 cb+= get_symbol(&s->c, &s->block_state[64], 1);
169 cr+= get_symbol(&s->c, &s->block_state[96], 1);
171 if(s->ref_frames > 1)
172 ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0);
173 if (ref >= s->ref_frames) {
174 av_log(s->avctx, AV_LOG_ERROR, "Invalid ref\n");
175 return AVERROR_INVALIDDATA;
177 pred_mv(s, &mx, &my, ref, left, top, tr);
178 mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1);
179 my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1);
181 set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type);
183 if ((res = decode_q_branch(s, level+1, 2*x+0, 2*y+0)) < 0 ||
184 (res = decode_q_branch(s, level+1, 2*x+1, 2*y+0)) < 0 ||
185 (res = decode_q_branch(s, level+1, 2*x+0, 2*y+1)) < 0 ||
186 (res = decode_q_branch(s, level+1, 2*x+1, 2*y+1)) < 0)
192 static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){
193 const int w= b->width;
194 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
195 const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
196 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
199 if(s->qlog == LOSSLESS_QLOG) return;
201 for(y=start_y; y<end_y; y++){
202 // DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride));
203 IDWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
207 line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
209 line[x]= (( i*qmul + qadd)>>(QEXPSHIFT));
215 static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){
216 const int w= b->width;
219 IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning
223 line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
225 for(y=start_y; y<end_y; y++){
227 // line = slice_buffer_get_line_from_address(sb, src + (y * stride));
228 line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
232 if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]);
233 else line[x] += line[x - 1];
235 if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]);
236 else line[x] += line[x - 1];
239 if(y) line[x] += prev[x];
245 static void decode_qlogs(SnowContext *s){
246 int plane_index, level, orientation;
248 for(plane_index=0; plane_index<3; plane_index++){
249 for(level=0; level<s->spatial_decomposition_count; level++){
250 for(orientation=level ? 1:0; orientation<4; orientation++){
252 if (plane_index==2) q= s->plane[1].band[level][orientation].qlog;
253 else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog;
254 else q= get_symbol(&s->c, s->header_state, 1);
255 s->plane[plane_index].band[level][orientation].qlog= q;
261 #define GET_S(dst, check) \
262 tmp= get_symbol(&s->c, s->header_state, 0);\
264 av_log(s->avctx, AV_LOG_ERROR, "Error " #dst " is %d\n", tmp);\
269 static int decode_header(SnowContext *s){
270 int plane_index, tmp;
273 memset(kstate, MID_STATE, sizeof(kstate));
275 s->keyframe= get_rac(&s->c, kstate);
276 if(s->keyframe || s->always_reset){
277 ff_snow_reset_contexts(s);
278 s->spatial_decomposition_type=
282 s->block_max_depth= 0;
285 GET_S(s->version, tmp <= 0U)
286 s->always_reset= get_rac(&s->c, s->header_state);
287 s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0);
288 s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0);
289 GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
290 s->colorspace_type= get_symbol(&s->c, s->header_state, 0);
291 s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0);
292 s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0);
293 s->spatial_scalability= get_rac(&s->c, s->header_state);
294 // s->rate_scalability= get_rac(&s->c, s->header_state);
295 GET_S(s->max_ref_frames, tmp < (unsigned)MAX_REF_FRAMES)
302 if(get_rac(&s->c, s->header_state)){
303 for(plane_index=0; plane_index<2; plane_index++){
305 Plane *p= &s->plane[plane_index];
306 p->diag_mc= get_rac(&s->c, s->header_state);
307 htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
308 if((unsigned)htaps > HTAPS_MAX || htaps==0)
311 for(i= htaps/2; i; i--){
312 p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
315 p->hcoeff[0]= 32-sum;
317 s->plane[2].diag_mc= s->plane[1].diag_mc;
318 s->plane[2].htaps = s->plane[1].htaps;
319 memcpy(s->plane[2].hcoeff, s->plane[1].hcoeff, sizeof(s->plane[1].hcoeff));
321 if(get_rac(&s->c, s->header_state)){
322 GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
327 s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1);
328 if(s->spatial_decomposition_type > 1U){
329 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
332 if(FFMIN(s->avctx-> width>>s->chroma_h_shift,
333 s->avctx->height>>s->chroma_v_shift) >> (s->spatial_decomposition_count-1) <= 0){
334 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_count %d too large for size", s->spatial_decomposition_count);
338 if (s->chroma_h_shift != 1 || s->chroma_v_shift != 1) {
339 av_log(s->avctx, AV_LOG_ERROR, "Invalid chroma shift\n");
340 return AVERROR_PATCHWELCOME;
343 s->qlog += get_symbol(&s->c, s->header_state, 1);
344 s->mv_scale += get_symbol(&s->c, s->header_state, 1);
345 s->qbias += get_symbol(&s->c, s->header_state, 1);
346 s->block_max_depth+= get_symbol(&s->c, s->header_state, 1);
347 if(s->block_max_depth > 1 || s->block_max_depth < 0){
348 av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
349 s->block_max_depth= 0;
356 static av_cold int decode_init(AVCodecContext *avctx)
360 avctx->pix_fmt= AV_PIX_FMT_YUV420P;
362 if ((ret = ff_snow_common_init(avctx)) < 0) {
363 ff_snow_common_end(avctx->priv_data);
370 static int decode_blocks(SnowContext *s){
378 if ((res = decode_q_branch(s, 0, x, y)) < 0)
385 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
388 const uint8_t *buf = avpkt->data;
389 int buf_size = avpkt->size;
390 SnowContext *s = avctx->priv_data;
391 RangeCoder * const c= &s->c;
393 AVFrame *picture = data;
394 int level, orientation, plane_index;
397 ff_init_range_decoder(c, buf, buf_size);
398 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
400 s->current_picture.pict_type= AV_PICTURE_TYPE_I; //FIXME I vs. P
401 if(decode_header(s)<0)
403 ff_snow_common_init_after_header(avctx);
405 // realloc slice buffer for the case that spatial_decomposition_count changed
406 ff_slice_buffer_destroy(&s->sb);
407 if ((res = ff_slice_buffer_init(&s->sb, s->plane[0].height,
408 (MB_SIZE >> s->block_max_depth) +
409 s->spatial_decomposition_count * 8 + 1,
411 s->spatial_idwt_buffer)) < 0)
414 for(plane_index=0; plane_index<3; plane_index++){
415 Plane *p= &s->plane[plane_index];
416 p->fast_mc= p->diag_mc && p->htaps==6 && p->hcoeff[0]==40
421 ff_snow_alloc_blocks(s);
423 if(ff_snow_frame_start(s) < 0)
425 //keyframe flag duplication mess FIXME
426 if(avctx->debug&FF_DEBUG_PICT_INFO)
427 av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog);
429 if ((res = decode_blocks(s)) < 0)
432 for(plane_index=0; plane_index<3; plane_index++){
433 Plane *p= &s->plane[plane_index];
437 int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */
439 if(s->avctx->debug&2048){
440 memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
441 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
445 int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x];
446 s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v;
452 for(level=0; level<s->spatial_decomposition_count; level++){
453 for(orientation=level ? 1 : 0; orientation<4; orientation++){
454 SubBand *b= &p->band[level][orientation];
455 unpack_coeffs(s, b, b->parent, orientation);
461 const int mb_h= s->b_height << s->block_max_depth;
462 const int block_size = MB_SIZE >> s->block_max_depth;
463 const int block_w = plane_index ? block_size/2 : block_size;
465 DWTCompose cs[MAX_DECOMPOSITIONS];
470 ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count);
471 for(mb_y=0; mb_y<=mb_h; mb_y++){
473 int slice_starty = block_w*mb_y;
474 int slice_h = block_w*(mb_y+1);
475 if (!(s->keyframe || s->avctx->debug&512)){
476 slice_starty = FFMAX(0, slice_starty - (block_w >> 1));
477 slice_h -= (block_w >> 1);
480 for(level=0; level<s->spatial_decomposition_count; level++){
481 for(orientation=level ? 1 : 0; orientation<4; orientation++){
482 SubBand *b= &p->band[level][orientation];
485 int our_mb_start = mb_y;
486 int our_mb_end = (mb_y + 1);
488 start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0);
489 end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra);
490 if (!(s->keyframe || s->avctx->debug&512)){
491 start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level)));
492 end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level)));
494 start_y = FFMIN(b->height, start_y);
495 end_y = FFMIN(b->height, end_y);
497 if (start_y != end_y){
498 if (orientation == 0){
499 SubBand * correlate_band = &p->band[0][0];
500 int correlate_end_y = FFMIN(b->height, end_y + 1);
501 int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0));
502 decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]);
503 correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
504 dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y);
507 decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
512 for(; yd<slice_h; yd+=4){
513 ff_spatial_idwt_buffered_slice(&s->dwt, cs, &s->sb, s->temp_idwt_buffer, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
516 if(s->qlog == LOSSLESS_QLOG){
517 for(; yq<slice_h && yq<h; yq++){
518 IDWTELEM * line = slice_buffer_get_line(&s->sb, yq);
520 line[x] <<= FRAC_BITS;
525 predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y);
527 y = FFMIN(p->height, slice_starty);
528 end_y = FFMIN(p->height, slice_h);
530 ff_slice_buffer_release(&s->sb, y++);
533 ff_slice_buffer_flush(&s->sb);
540 ff_snow_release_buffer(avctx);
542 if(!(s->avctx->debug&2048))
543 *picture= s->current_picture;
545 *picture= s->mconly_picture;
549 bytes_read= c->bytestream - c->bytestream_start;
550 if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME
555 static av_cold int decode_end(AVCodecContext *avctx)
557 SnowContext *s = avctx->priv_data;
559 ff_slice_buffer_destroy(&s->sb);
561 ff_snow_common_end(s);
566 AVCodec ff_snow_decoder = {
568 .type = AVMEDIA_TYPE_VIDEO,
569 .id = AV_CODEC_ID_SNOW,
570 .priv_data_size = sizeof(SnowContext),
573 .decode = decode_frame,
574 .capabilities = CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
575 .long_name = NULL_IF_CONFIG_SMALL("Snow"),