2 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (C) 2006 Robert Edele <yartrebo@earthlink.net>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #ifndef AVCODEC_SNOW_H
23 #define AVCODEC_SNOW_H
28 #include "rangecoder.h"
30 #include "mpegvideo.h"
36 #define QROOT (1<<QSHIFT)
37 #define LOSSLESS_QLOG -128
39 #define MAX_REF_FRAMES 8
41 #define LOG2_OBMC_MAX 8
42 #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
43 typedef struct BlockNode{
49 //#define TYPE_SPLIT 1
52 //#define TYPE_NOCOLOR 4
53 uint8_t level; //FIXME merge into type?
56 static const BlockNode null_block= { //FIXME add border maybe
57 .color= {128,128,128},
65 #define LOG2_MB_SIZE 4
66 #define MB_SIZE (1<<LOG2_MB_SIZE)
67 #define ENCODER_EXTRA_BITS 4
70 typedef struct x_and_coeff{
75 typedef struct SubBand{
80 int qlog; ///< log(qscale)/log[2^(1/6)]
85 int stride_line; ///< Stride measured in lines, not pixels.
86 x_and_coeff * x_coeff;
87 struct SubBand *parent;
88 uint8_t state[/*7*2*/ 7 + 512][32];
94 SubBand band[MAX_DECOMPOSITIONS][4];
97 int8_t hcoeff[HTAPS_MAX/2];
102 int8_t last_hcoeff[HTAPS_MAX/2];
106 typedef struct SnowContext{
108 AVCodecContext *avctx;
113 AVFrame input_picture; ///< new_picture with the internal linesizes
114 AVFrame current_picture;
115 AVFrame last_picture[MAX_REF_FRAMES];
116 uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4];
117 AVFrame mconly_picture;
118 // uint8_t q_context[16];
119 uint8_t header_state[32];
120 uint8_t block_state[128 + 32*128];
124 int spatial_decomposition_type;
125 int last_spatial_decomposition_type;
126 int temporal_decomposition_type;
127 int spatial_decomposition_count;
128 int last_spatial_decomposition_count;
129 int temporal_decomposition_count;
132 int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
133 uint32_t *ref_scores[MAX_REF_FRAMES];
134 DWTELEM *spatial_dwt_buffer;
135 IDWTELEM *spatial_idwt_buffer;
139 int spatial_scalability;
149 #define QBIAS_SHIFT 3
153 int last_block_max_depth;
154 Plane plane[MAX_PLANES];
156 #define ME_CACHE_SIZE 1024
157 unsigned me_cache[ME_CACHE_SIZE];
158 unsigned me_cache_generation;
162 MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)
168 extern const uint8_t * const obmc_tab[4];
170 // Avoid a name clash on SGI IRIX
173 extern uint8_t qexp[QROOT];
174 extern int scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES];
176 /* C bits used by mmx/sse2/altivec */
178 static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
182 low[(*i)+1] = low[((*i)+1)>>1];
187 static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){
188 for (; (*i)>=0; (*i)-=2){
189 low[(*i)+1] = high[(*i)>>1];
190 low[*i] = low[(*i)>>1];
194 static av_always_inline void snow_horizontal_compose_lift_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w, int lift_high, int mul, int add, int shift){
196 dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
199 if((width^lift_high)&1){
200 dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
204 static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){
206 dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS);
210 dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS);
216 int ff_snow_common_init(AVCodecContext *avctx);
217 int ff_snow_common_init_after_header(AVCodecContext *avctx);
218 void ff_snow_common_end(SnowContext *s);
219 void ff_snow_release_buffer(AVCodecContext *avctx);
220 void ff_snow_reset_contexts(SnowContext *s);
221 int ff_snow_alloc_blocks(SnowContext *s);
222 int ff_snow_frame_start(SnowContext *s);
223 void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride,
224 int sx, int sy, int b_w, int b_h, BlockNode *block,
225 int plane_index, int w, int h);
226 /* common inline functions */
227 //XXX doublecheck all of them should stay inlined
229 static inline void snow_set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
230 const int w= s->b_width << s->block_max_depth;
231 const int rem_depth= s->block_max_depth - level;
232 const int index= (x + y*w) << rem_depth;
233 const int block_w= 1<<rem_depth;
246 for(j=0; j<block_w; j++){
247 for(i=0; i<block_w; i++){
248 s->block[index + i + j*w]= block;
253 static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
254 const BlockNode *left, const BlockNode *top, const BlockNode *tr){
255 if(s->ref_frames == 1){
256 *mx = mid_pred(left->mx, top->mx, tr->mx);
257 *my = mid_pred(left->my, top->my, tr->my);
259 const int *scale = scale_mv_ref[ref];
260 *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
261 (top ->mx * scale[top ->ref] + 128) >>8,
262 (tr ->mx * scale[tr ->ref] + 128) >>8);
263 *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
264 (top ->my * scale[top ->ref] + 128) >>8,
265 (tr ->my * scale[tr ->ref] + 128) >>8);
269 static av_always_inline int same_block(BlockNode *a, BlockNode *b){
270 if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
271 return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
273 return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
277 //FIXME name cleanup (b_w, block_w, b_width stuff)
278 //XXX should we really inline it?
279 static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, IDWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){
280 const int b_width = s->b_width << s->block_max_depth;
281 const int b_height= s->b_height << s->block_max_depth;
282 const int b_stride= b_width;
283 BlockNode *lt= &s->block[b_x + b_y*b_stride];
285 BlockNode *lb= lt+b_stride;
288 int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
289 uint8_t *tmp = s->scratchbuf;
296 }else if(b_x + 1 >= b_width){
303 }else if(b_y + 1 >= b_height){
308 if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
311 if(!sliced && !offset_dst)
314 }else if(src_x + b_w > w){
318 obmc -= src_y*obmc_stride;
320 if(!sliced && !offset_dst)
321 dst -= src_y*dst_stride;
323 }else if(src_y + b_h> h){
327 if(b_w<=0 || b_h<=0) return;
329 assert(src_stride > 2*MB_SIZE + 5);
331 if(!sliced && offset_dst)
332 dst += src_x + src_y*dst_stride;
333 dst8+= src_x + src_y*src_stride;
334 // src += src_x + src_y*src_stride;
336 ptmp= tmp + 3*tmp_step;
339 ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
341 if(same_block(lt, rt)){
346 ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
349 if(same_block(lt, lb)){
351 }else if(same_block(rt, lb)){
356 ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
359 if(same_block(lt, rb) ){
361 }else if(same_block(rt, rb)){
363 }else if(same_block(lb, rb)){
367 ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
370 s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
372 for(y=0; y<b_h; y++){
373 //FIXME ugly misuse of obmc_stride
374 const uint8_t *obmc1= obmc + y*obmc_stride;
375 const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
376 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
377 const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
378 for(x=0; x<b_w; x++){
379 int v= obmc1[x] * block[3][x + y*src_stride]
380 +obmc2[x] * block[2][x + y*src_stride]
381 +obmc3[x] * block[1][x + y*src_stride]
382 +obmc4[x] * block[0][x + y*src_stride];
384 v <<= 8 - LOG2_OBMC_MAX;
389 v += dst[x + y*dst_stride];
390 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
391 if(v&(~255)) v= ~(v>>31);
392 dst8[x + y*src_stride] = v;
394 dst[x + y*dst_stride] -= v;
401 static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
402 Plane *p= &s->plane[plane_index];
403 const int mb_w= s->b_width << s->block_max_depth;
404 const int mb_h= s->b_height << s->block_max_depth;
406 int block_size = MB_SIZE >> s->block_max_depth;
407 int block_w = plane_index ? block_size/2 : block_size;
408 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
409 const int obmc_stride= plane_index ? block_size : 2*block_size;
410 int ref_stride= s->current_picture.linesize[plane_index];
411 uint8_t *dst8= s->current_picture.data[plane_index];
415 if(s->keyframe || (s->avctx->debug&512)){
420 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
422 int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
424 if(v&(~255)) v= ~(v>>31);
425 dst8[x + y*ref_stride]= v;
429 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
431 buf[x + y*w]-= 128<<FRAC_BITS;
439 for(mb_x=0; mb_x<=mb_w; mb_x++){
440 add_yblock(s, 0, NULL, buf, dst8, obmc,
441 block_w*mb_x - block_w/2,
442 block_w*mb_y - block_w/2,
445 w, ref_stride, obmc_stride,
447 add, 1, plane_index);
451 static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
452 const int mb_h= s->b_height << s->block_max_depth;
454 for(mb_y=0; mb_y<=mb_h; mb_y++)
455 predict_slice(s, buf, plane_index, add, mb_y);
458 static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
459 const int w= s->b_width << s->block_max_depth;
460 const int rem_depth= s->block_max_depth - level;
461 const int index= (x + y*w) << rem_depth;
462 const int block_w= 1<<rem_depth;
475 for(j=0; j<block_w; j++){
476 for(i=0; i<block_w; i++){
477 s->block[index + i + j*w]= block;
482 static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
483 const int offset[3]= {
485 ((y*c->uvstride + x)>>1),
486 ((y*c->uvstride + x)>>1),
490 c->src[0][i]= src [i];
491 c->ref[0][i]= ref [i] + offset[i];
497 /* bitstream functions */
499 extern const int8_t quant3bA[256];
501 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
503 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
507 const int a= FFABS(v);
508 const int e= av_log2(a);
509 const int el= FFMIN(e, 10);
510 put_rac(c, state+0, 0);
513 put_rac(c, state+1+i, 1); //1..10
516 put_rac(c, state+1+9, 1); //1..10
518 put_rac(c, state+1+FFMIN(i,9), 0);
520 for(i=e-1; i>=el; i--){
521 put_rac(c, state+22+9, (a>>i)&1); //22..31
524 put_rac(c, state+22+i, (a>>i)&1); //22..31
528 put_rac(c, state+11 + el, v < 0); //11..21
530 put_rac(c, state+0, 1);
534 static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
535 if(get_rac(c, state+0))
540 while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
545 for(i=e-1; i>=0; i--){
546 a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
549 e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21
554 static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
556 int r= log2>=0 ? 1<<log2 : 1;
562 put_rac(c, state+4+log2, 1);
567 put_rac(c, state+4+log2, 0);
569 for(i=log2-1; i>=0; i--){
570 put_rac(c, state+31-i, (v>>i)&1);
574 static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
576 int r= log2>=0 ? 1<<log2 : 1;
581 while(get_rac(c, state+4+log2)){
587 for(i=log2-1; i>=0; i--){
588 v+= get_rac(c, state+31-i)<<i;
594 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
595 const int w= b->width;
596 const int h= b->height;
600 x_and_coeff *xc= b->x_coeff;
601 x_and_coeff *prev_xc= NULL;
602 x_and_coeff *prev2_xc= xc;
603 x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
604 x_and_coeff *prev_parent_xc= parent_xc;
606 runs= get_symbol2(&s->c, b->state[30], 0);
607 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
614 if(y && prev_xc->x == 0){
626 if(prev_xc->x == x + 1)
632 if(x>>1 > parent_xc->x){
635 if(x>>1 == parent_xc->x){
639 if(/*ll|*/l|lt|t|rt|p){
640 int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
642 v=get_rac(&s->c, &b->state[0][context]);
644 v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
645 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
652 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
654 v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
655 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
664 if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
665 else max_run= FFMIN(run, w-x-1);
667 max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
673 (xc++)->x= w+1; //end marker
679 while(parent_xc->x != parent->width+1)
682 prev_parent_xc= parent_xc;
684 parent_xc= prev_parent_xc;
689 (xc++)->x= w+1; //end marker
692 #endif /* AVCODEC_SNOW_H */