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"
29 #include "rangecoder.h"
32 #include "mpegvideo.h"
38 static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, int plane_index, int add, int mb_y){
39 Plane *p= &s->plane[plane_index];
40 const int mb_w= s->b_width << s->block_max_depth;
41 const int mb_h= s->b_height << s->block_max_depth;
43 int block_size = MB_SIZE >> s->block_max_depth;
44 int block_w = plane_index ? block_size/2 : block_size;
45 const uint8_t *obmc = plane_index ? ff_obmc_tab[s->block_max_depth+1] : ff_obmc_tab[s->block_max_depth];
46 int obmc_stride= plane_index ? block_size : 2*block_size;
47 int ref_stride= s->current_picture.linesize[plane_index];
48 uint8_t *dst8= s->current_picture.data[plane_index];
52 if(s->keyframe || (s->avctx->debug&512)){
57 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
58 // DWTELEM * line = slice_buffer_get_line(sb, y);
59 IDWTELEM * line = sb->line[y];
61 // int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
62 int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
64 if(v&(~255)) v= ~(v>>31);
65 dst8[x + y*ref_stride]= v;
69 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
70 // DWTELEM * line = slice_buffer_get_line(sb, y);
71 IDWTELEM * line = sb->line[y];
73 line[x] -= 128 << FRAC_BITS;
74 // buf[x + y*w]-= 128<<FRAC_BITS;
82 for(mb_x=0; mb_x<=mb_w; mb_x++){
83 add_yblock(s, 1, sb, old_buffer, dst8, obmc,
84 block_w*mb_x - block_w/2,
85 block_w*mb_y - block_w/2,
88 w, ref_stride, obmc_stride,
94 static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){
95 const int w= b->width;
97 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
98 int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
99 int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
102 if(b->ibuf == s->spatial_idwt_buffer || s->qlog == LOSSLESS_QLOG){
107 /* If we are on the second or later slice, restore our index. */
109 new_index = save_state[0];
112 for(y=start_y; y<h; y++){
115 IDWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
116 memset(line, 0, b->width*sizeof(IDWTELEM));
117 v = b->x_coeff[new_index].coeff;
118 x = b->x_coeff[new_index++].x;
120 register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT;
121 register int u= -(v&1);
124 v = b->x_coeff[new_index].coeff;
125 x = b->x_coeff[new_index++].x;
129 /* Save our variables for the next slice. */
130 save_state[0] = new_index;
135 static int decode_q_branch(SnowContext *s, int level, int x, int y){
136 const int w= s->b_width << s->block_max_depth;
137 const int rem_depth= s->block_max_depth - level;
138 const int index= (x + y*w) << rem_depth;
139 int trx= (x+1)<<rem_depth;
140 const BlockNode *left = x ? &s->block[index-1] : &null_block;
141 const BlockNode *top = y ? &s->block[index-w] : &null_block;
142 const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
143 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
144 int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
148 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);
152 if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
154 int l = left->color[0];
155 int cb= left->color[1];
156 int cr= left->color[2];
158 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
159 int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx));
160 int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my));
162 type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;
165 pred_mv(s, &mx, &my, 0, left, top, tr);
166 l += get_symbol(&s->c, &s->block_state[32], 1);
167 cb+= get_symbol(&s->c, &s->block_state[64], 1);
168 cr+= get_symbol(&s->c, &s->block_state[96], 1);
170 if(s->ref_frames > 1)
171 ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0);
172 if (ref >= s->ref_frames) {
173 av_log(s->avctx, AV_LOG_ERROR, "Invalid ref\n");
174 return AVERROR_INVALIDDATA;
176 pred_mv(s, &mx, &my, ref, left, top, tr);
177 mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1);
178 my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1);
180 set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type);
182 if ((res = decode_q_branch(s, level+1, 2*x+0, 2*y+0)) < 0 ||
183 (res = decode_q_branch(s, level+1, 2*x+1, 2*y+0)) < 0 ||
184 (res = decode_q_branch(s, level+1, 2*x+0, 2*y+1)) < 0 ||
185 (res = decode_q_branch(s, level+1, 2*x+1, 2*y+1)) < 0)
191 static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){
192 const int w= b->width;
193 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
194 const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
195 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
198 if(s->qlog == LOSSLESS_QLOG) return;
200 for(y=start_y; y<end_y; y++){
201 // DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride));
202 IDWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
206 line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
208 line[x]= (( i*qmul + qadd)>>(QEXPSHIFT));
214 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){
215 const int w= b->width;
218 IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning
222 line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
224 for(y=start_y; y<end_y; y++){
226 // line = slice_buffer_get_line_from_address(sb, src + (y * stride));
227 line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
231 if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]);
232 else line[x] += line[x - 1];
234 if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]);
235 else line[x] += line[x - 1];
238 if(y) line[x] += prev[x];
244 static void decode_qlogs(SnowContext *s){
245 int plane_index, level, orientation;
247 for(plane_index=0; plane_index<3; plane_index++){
248 for(level=0; level<s->spatial_decomposition_count; level++){
249 for(orientation=level ? 1:0; orientation<4; orientation++){
251 if (plane_index==2) q= s->plane[1].band[level][orientation].qlog;
252 else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog;
253 else q= get_symbol(&s->c, s->header_state, 1);
254 s->plane[plane_index].band[level][orientation].qlog= q;
260 #define GET_S(dst, check) \
261 tmp= get_symbol(&s->c, s->header_state, 0);\
263 av_log(s->avctx, AV_LOG_ERROR, "Error " #dst " is %d\n", tmp);\
268 static int decode_header(SnowContext *s){
269 int plane_index, tmp;
272 memset(kstate, MID_STATE, sizeof(kstate));
274 s->keyframe= get_rac(&s->c, kstate);
275 if(s->keyframe || s->always_reset){
276 ff_snow_reset_contexts(s);
277 s->spatial_decomposition_type=
281 s->block_max_depth= 0;
284 GET_S(s->version, tmp <= 0U)
285 s->always_reset= get_rac(&s->c, s->header_state);
286 s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0);
287 s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0);
288 GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
289 s->colorspace_type= get_symbol(&s->c, s->header_state, 0);
290 s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0);
291 s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0);
292 s->spatial_scalability= get_rac(&s->c, s->header_state);
293 // s->rate_scalability= get_rac(&s->c, s->header_state);
294 GET_S(s->max_ref_frames, tmp < (unsigned)MAX_REF_FRAMES)
301 if(get_rac(&s->c, s->header_state)){
302 for(plane_index=0; plane_index<2; plane_index++){
304 Plane *p= &s->plane[plane_index];
305 p->diag_mc= get_rac(&s->c, s->header_state);
306 htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
307 if((unsigned)htaps > HTAPS_MAX || htaps==0)
310 for(i= htaps/2; i; i--){
311 p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
314 p->hcoeff[0]= 32-sum;
316 s->plane[2].diag_mc= s->plane[1].diag_mc;
317 s->plane[2].htaps = s->plane[1].htaps;
318 memcpy(s->plane[2].hcoeff, s->plane[1].hcoeff, sizeof(s->plane[1].hcoeff));
320 if(get_rac(&s->c, s->header_state)){
321 GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
326 s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1);
327 if(s->spatial_decomposition_type > 1U){
328 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
331 if(FFMIN(s->avctx-> width>>s->chroma_h_shift,
332 s->avctx->height>>s->chroma_v_shift) >> (s->spatial_decomposition_count-1) <= 0){
333 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_count %d too large for size", s->spatial_decomposition_count);
337 if (s->chroma_h_shift != 1 || s->chroma_v_shift != 1) {
338 av_log(s->avctx, AV_LOG_ERROR, "Invalid chroma shift\n");
339 return AVERROR_PATCHWELCOME;
342 s->qlog += get_symbol(&s->c, s->header_state, 1);
343 s->mv_scale += get_symbol(&s->c, s->header_state, 1);
344 s->qbias += get_symbol(&s->c, s->header_state, 1);
345 s->block_max_depth+= get_symbol(&s->c, s->header_state, 1);
346 if(s->block_max_depth > 1 || s->block_max_depth < 0){
347 av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
348 s->block_max_depth= 0;
355 static av_cold int decode_init(AVCodecContext *avctx)
357 avctx->pix_fmt= PIX_FMT_YUV420P;
359 ff_snow_common_init(avctx);
364 static int decode_blocks(SnowContext *s){
372 if ((res = decode_q_branch(s, 0, x, y)) < 0)
379 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt){
380 const uint8_t *buf = avpkt->data;
381 int buf_size = avpkt->size;
382 SnowContext *s = avctx->priv_data;
383 RangeCoder * const c= &s->c;
385 AVFrame *picture = data;
386 int level, orientation, plane_index;
389 ff_init_range_decoder(c, buf, buf_size);
390 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
392 s->current_picture.pict_type= AV_PICTURE_TYPE_I; //FIXME I vs. P
393 if(decode_header(s)<0)
395 ff_snow_common_init_after_header(avctx);
397 // realloc slice buffer for the case that spatial_decomposition_count changed
398 ff_slice_buffer_destroy(&s->sb);
399 ff_slice_buffer_init(&s->sb, s->plane[0].height, (MB_SIZE >> s->block_max_depth) + s->spatial_decomposition_count * 8 + 1, s->plane[0].width, s->spatial_idwt_buffer);
401 for(plane_index=0; plane_index<3; plane_index++){
402 Plane *p= &s->plane[plane_index];
403 p->fast_mc= p->diag_mc && p->htaps==6 && p->hcoeff[0]==40
408 ff_snow_alloc_blocks(s);
410 if(ff_snow_frame_start(s) < 0)
412 //keyframe flag duplication mess FIXME
413 if(avctx->debug&FF_DEBUG_PICT_INFO)
414 av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog);
416 if ((res = decode_blocks(s)) < 0)
419 for(plane_index=0; plane_index<3; plane_index++){
420 Plane *p= &s->plane[plane_index];
424 int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */
426 if(s->avctx->debug&2048){
427 memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
428 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
432 int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x];
433 s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v;
439 for(level=0; level<s->spatial_decomposition_count; level++){
440 for(orientation=level ? 1 : 0; orientation<4; orientation++){
441 SubBand *b= &p->band[level][orientation];
442 unpack_coeffs(s, b, b->parent, orientation);
448 const int mb_h= s->b_height << s->block_max_depth;
449 const int block_size = MB_SIZE >> s->block_max_depth;
450 const int block_w = plane_index ? block_size/2 : block_size;
452 DWTCompose cs[MAX_DECOMPOSITIONS];
457 ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count);
458 for(mb_y=0; mb_y<=mb_h; mb_y++){
460 int slice_starty = block_w*mb_y;
461 int slice_h = block_w*(mb_y+1);
462 if (!(s->keyframe || s->avctx->debug&512)){
463 slice_starty = FFMAX(0, slice_starty - (block_w >> 1));
464 slice_h -= (block_w >> 1);
467 for(level=0; level<s->spatial_decomposition_count; level++){
468 for(orientation=level ? 1 : 0; orientation<4; orientation++){
469 SubBand *b= &p->band[level][orientation];
472 int our_mb_start = mb_y;
473 int our_mb_end = (mb_y + 1);
475 start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0);
476 end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra);
477 if (!(s->keyframe || s->avctx->debug&512)){
478 start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level)));
479 end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level)));
481 start_y = FFMIN(b->height, start_y);
482 end_y = FFMIN(b->height, end_y);
484 if (start_y != end_y){
485 if (orientation == 0){
486 SubBand * correlate_band = &p->band[0][0];
487 int correlate_end_y = FFMIN(b->height, end_y + 1);
488 int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0));
489 decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]);
490 correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
491 dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y);
494 decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
499 for(; yd<slice_h; yd+=4){
500 ff_spatial_idwt_buffered_slice(&s->dwt, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
503 if(s->qlog == LOSSLESS_QLOG){
504 for(; yq<slice_h && yq<h; yq++){
505 IDWTELEM * line = slice_buffer_get_line(&s->sb, yq);
507 line[x] <<= FRAC_BITS;
512 predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y);
514 y = FFMIN(p->height, slice_starty);
515 end_y = FFMIN(p->height, slice_h);
517 ff_slice_buffer_release(&s->sb, y++);
520 ff_slice_buffer_flush(&s->sb);
527 ff_snow_release_buffer(avctx);
529 if(!(s->avctx->debug&2048))
530 *picture= s->current_picture;
532 *picture= s->mconly_picture;
534 *data_size = sizeof(AVFrame);
536 bytes_read= c->bytestream - c->bytestream_start;
537 if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME
542 static av_cold int decode_end(AVCodecContext *avctx)
544 SnowContext *s = avctx->priv_data;
546 ff_slice_buffer_destroy(&s->sb);
548 ff_snow_common_end(s);
553 AVCodec ff_snow_decoder = {
555 .type = AVMEDIA_TYPE_VIDEO,
557 .priv_data_size = sizeof(SnowContext),
560 .decode = decode_frame,
561 .capabilities = CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
562 .long_name = NULL_IF_CONFIG_SMALL("Snow"),