• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
3  * Copyright (C) 2006 Robert Edele <yartrebo@earthlink.net>
4  *
5  * This file is part of FFmpeg.
6  *
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.
11  *
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.
16  *
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
20  */
21 
22 #ifndef AVCODEC_SNOW_H
23 #define AVCODEC_SNOW_H
24 
25 #include "libavutil/motion_vector.h"
26 
27 #include "hpeldsp.h"
28 #include "me_cmp.h"
29 #include "qpeldsp.h"
30 #include "snow_dwt.h"
31 
32 #include "rangecoder.h"
33 #include "mathops.h"
34 
35 #include "mpegvideo.h"
36 #include "h264qpel.h"
37 
38 #define FF_ME_ITER 3
39 
40 #define MID_STATE 128
41 
42 #define MAX_PLANES 4
43 #define QSHIFT 5
44 #define QROOT (1<<QSHIFT)
45 #define LOSSLESS_QLOG -128
46 #define FRAC_BITS 4
47 #define MAX_REF_FRAMES 8
48 
49 #define LOG2_OBMC_MAX 8
50 #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
51 typedef struct BlockNode{
52     int16_t mx;                 ///< Motion vector component X, see mv_scale
53     int16_t my;                 ///< Motion vector component Y, see mv_scale
54     uint8_t ref;                ///< Reference frame index
55     uint8_t color[3];           ///< Color for intra
56     uint8_t type;               ///< Bitfield of BLOCK_*
57 //#define TYPE_SPLIT    1
58 #define BLOCK_INTRA   1         ///< Intra block, inter otherwise
59 #define BLOCK_OPT     2         ///< Block needs no checks in this round of iterative motion estiation
60 //#define TYPE_NOCOLOR  4
61     uint8_t level; //FIXME merge into type?
62 }BlockNode;
63 
64 static const BlockNode null_block= { //FIXME add border maybe
65     .color= {128,128,128},
66     .mx= 0,
67     .my= 0,
68     .ref= 0,
69     .type= 0,
70     .level= 0,
71 };
72 
73 #define LOG2_MB_SIZE 4
74 #define MB_SIZE (1<<LOG2_MB_SIZE)
75 #define ENCODER_EXTRA_BITS 4
76 #define HTAPS_MAX 8
77 
78 typedef struct x_and_coeff{
79     int16_t x;
80     uint16_t coeff;
81 } x_and_coeff;
82 
83 typedef struct SubBand{
84     int level;
85     int stride;
86     int width;
87     int height;
88     int qlog;        ///< log(qscale)/log[2^(1/6)]
89     DWTELEM *buf;
90     IDWTELEM *ibuf;
91     int buf_x_offset;
92     int buf_y_offset;
93     int stride_line; ///< Stride measured in lines, not pixels.
94     x_and_coeff * x_coeff;
95     struct SubBand *parent;
96     uint8_t state[/*7*2*/ 7 + 512][32];
97 }SubBand;
98 
99 typedef struct Plane{
100     int width;
101     int height;
102     SubBand band[MAX_DECOMPOSITIONS][4];
103 
104     int htaps;
105     int8_t hcoeff[HTAPS_MAX/2];
106     int diag_mc;
107     int fast_mc;
108 
109     int last_htaps;
110     int8_t last_hcoeff[HTAPS_MAX/2];
111     int last_diag_mc;
112 }Plane;
113 
114 typedef struct SnowContext{
115     AVClass *class;
116     AVCodecContext *avctx;
117     RangeCoder c;
118     MECmpContext mecc;
119     HpelDSPContext hdsp;
120     QpelDSPContext qdsp;
121     VideoDSPContext vdsp;
122     H264QpelContext h264qpel;
123     MpegvideoEncDSPContext mpvencdsp;
124     SnowDWTContext dwt;
125     AVFrame *input_picture;              ///< new_picture with the internal linesizes
126     AVFrame *current_picture;
127     AVFrame *last_picture[MAX_REF_FRAMES];
128     uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4];
129     AVFrame *mconly_picture;
130 //     uint8_t q_context[16];
131     uint8_t header_state[32];
132     uint8_t block_state[128 + 32*128];
133     int keyframe;
134     int always_reset;
135     int version;
136     int spatial_decomposition_type;
137     int last_spatial_decomposition_type;
138     int temporal_decomposition_type;
139     int spatial_decomposition_count;
140     int last_spatial_decomposition_count;
141     int temporal_decomposition_count;
142     int max_ref_frames;
143     int ref_frames;
144     int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
145     uint32_t *ref_scores[MAX_REF_FRAMES];
146     DWTELEM *spatial_dwt_buffer;
147     DWTELEM *temp_dwt_buffer;
148     IDWTELEM *spatial_idwt_buffer;
149     IDWTELEM *temp_idwt_buffer;
150     int *run_buffer;
151     int colorspace_type;
152     int chroma_h_shift;
153     int chroma_v_shift;
154     int spatial_scalability;
155     int qlog;
156     int last_qlog;
157     int lambda;
158     int lambda2;
159     int pass1_rc;
160     int mv_scale;
161     int last_mv_scale;
162     int qbias;
163     int last_qbias;
164 #define QBIAS_SHIFT 3
165     int b_width;
166     int b_height;
167     int block_max_depth;
168     int last_block_max_depth;
169     int nb_planes;
170     Plane plane[MAX_PLANES];
171     BlockNode *block;
172 #define ME_CACHE_SIZE 1024
173     unsigned me_cache[ME_CACHE_SIZE];
174     unsigned me_cache_generation;
175     slice_buffer sb;
176     int memc_only;
177     int no_bitstream;
178     int intra_penalty;
179     int motion_est;
180     int iterative_dia_size;
181     int scenechange_threshold;
182 
183     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)
184 
185     uint8_t *scratchbuf;
186     uint8_t *emu_edge_buffer;
187 
188     AVMotionVector *avmv;
189     unsigned avmv_size;
190     int avmv_index;
191     uint64_t encoding_error[AV_NUM_DATA_POINTERS];
192 
193     int pred;
194 }SnowContext;
195 
196 /* Tables */
197 extern const uint8_t * const ff_obmc_tab[4];
198 extern uint8_t ff_qexp[QROOT];
199 extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES];
200 
201 /* C bits used by mmx/sse2/altivec */
202 
snow_interleave_line_header(int * i,int width,IDWTELEM * low,IDWTELEM * high)203 static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
204     (*i) = (width) - 2;
205 
206     if (width & 1){
207         low[(*i)+1] = low[((*i)+1)>>1];
208         (*i)--;
209     }
210 }
211 
snow_interleave_line_footer(int * i,IDWTELEM * low,IDWTELEM * high)212 static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){
213     for (; (*i)>=0; (*i)-=2){
214         low[(*i)+1] = high[(*i)>>1];
215         low[*i] = low[(*i)>>1];
216     }
217 }
218 
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)219 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){
220     for(; i<w; i++){
221         dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
222     }
223 
224     if((width^lift_high)&1){
225         dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
226     }
227 }
228 
snow_horizontal_compose_liftS_lead_out(int i,IDWTELEM * dst,IDWTELEM * src,IDWTELEM * ref,int width,int w)229 static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){
230         for(; i<w; i++){
231             dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS);
232         }
233 
234         if(width&1){
235             dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS);
236         }
237 }
238 
239 /* common code */
240 
241 int ff_snow_common_init(AVCodecContext *avctx);
242 int ff_snow_common_init_after_header(AVCodecContext *avctx);
243 void ff_snow_common_end(SnowContext *s);
244 void ff_snow_release_buffer(AVCodecContext *avctx);
245 void ff_snow_reset_contexts(SnowContext *s);
246 int ff_snow_alloc_blocks(SnowContext *s);
247 int ff_snow_frame_start(SnowContext *s);
248 void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride,
249                      int sx, int sy, int b_w, int b_h, const BlockNode *block,
250                      int plane_index, int w, int h);
251 int ff_snow_get_buffer(SnowContext *s, AVFrame *frame);
252 /* common inline functions */
253 //XXX doublecheck all of them should stay inlined
254 
pred_mv(SnowContext * s,int * mx,int * my,int ref,const BlockNode * left,const BlockNode * top,const BlockNode * tr)255 static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
256                            const BlockNode *left, const BlockNode *top, const BlockNode *tr){
257     if(s->ref_frames == 1){
258         *mx = mid_pred(left->mx, top->mx, tr->mx);
259         *my = mid_pred(left->my, top->my, tr->my);
260     }else{
261         const int *scale = ff_scale_mv_ref[ref];
262         *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
263                        (top ->mx * scale[top ->ref] + 128) >>8,
264                        (tr  ->mx * scale[tr  ->ref] + 128) >>8);
265         *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
266                        (top ->my * scale[top ->ref] + 128) >>8,
267                        (tr  ->my * scale[tr  ->ref] + 128) >>8);
268     }
269 }
270 
same_block(BlockNode * a,BlockNode * b)271 static av_always_inline int same_block(BlockNode *a, BlockNode *b){
272     if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
273         return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
274     }else{
275         return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
276     }
277 }
278 
279 //FIXME name cleanup (b_w, block_w, b_width stuff)
280 //XXX should we really inline it?
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)281 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){
282     const int b_width = s->b_width  << s->block_max_depth;
283     const int b_height= s->b_height << s->block_max_depth;
284     const int b_stride= b_width;
285     BlockNode *lt= &s->block[b_x + b_y*b_stride];
286     BlockNode *rt= lt+1;
287     BlockNode *lb= lt+b_stride;
288     BlockNode *rb= lb+1;
289     uint8_t *block[4];
290     // When src_stride is large enough, it is possible to interleave the blocks.
291     // Otherwise the blocks are written sequentially in the tmp buffer.
292     int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
293     uint8_t *tmp = s->scratchbuf;
294     uint8_t *ptmp;
295     int x,y;
296 
297     if(b_x<0){
298         lt= rt;
299         lb= rb;
300     }else if(b_x + 1 >= b_width){
301         rt= lt;
302         rb= lb;
303     }
304     if(b_y<0){
305         lt= lb;
306         rt= rb;
307     }else if(b_y + 1 >= b_height){
308         lb= lt;
309         rb= rt;
310     }
311 
312     if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
313         obmc -= src_x;
314         b_w += src_x;
315         if(!sliced && !offset_dst)
316             dst -= src_x;
317         src_x=0;
318     }
319     if(src_x + b_w > w){
320         b_w = w - src_x;
321     }
322     if(src_y<0){
323         obmc -= src_y*obmc_stride;
324         b_h += src_y;
325         if(!sliced && !offset_dst)
326             dst -= src_y*dst_stride;
327         src_y=0;
328     }
329     if(src_y + b_h> h){
330         b_h = h - src_y;
331     }
332 
333     if(b_w<=0 || b_h<=0) return;
334 
335     if(!sliced && offset_dst)
336         dst += src_x + src_y*dst_stride;
337     dst8+= src_x + src_y*src_stride;
338 //    src += src_x + src_y*src_stride;
339 
340     ptmp= tmp + 3*tmp_step;
341     block[0]= ptmp;
342     ptmp+=tmp_step;
343     ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
344 
345     if(same_block(lt, rt)){
346         block[1]= block[0];
347     }else{
348         block[1]= ptmp;
349         ptmp+=tmp_step;
350         ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
351     }
352 
353     if(same_block(lt, lb)){
354         block[2]= block[0];
355     }else if(same_block(rt, lb)){
356         block[2]= block[1];
357     }else{
358         block[2]= ptmp;
359         ptmp+=tmp_step;
360         ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
361     }
362 
363     if(same_block(lt, rb) ){
364         block[3]= block[0];
365     }else if(same_block(rt, rb)){
366         block[3]= block[1];
367     }else if(same_block(lb, rb)){
368         block[3]= block[2];
369     }else{
370         block[3]= ptmp;
371         ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
372     }
373     if(sliced){
374         s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
375     }else{
376         for(y=0; y<b_h; y++){
377             //FIXME ugly misuse of obmc_stride
378             const uint8_t *obmc1= obmc + y*obmc_stride;
379             const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
380             const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
381             const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
382             for(x=0; x<b_w; x++){
383                 int v=   obmc1[x] * block[3][x + y*src_stride]
384                         +obmc2[x] * block[2][x + y*src_stride]
385                         +obmc3[x] * block[1][x + y*src_stride]
386                         +obmc4[x] * block[0][x + y*src_stride];
387 
388                 v <<= 8 - LOG2_OBMC_MAX;
389                 if(FRAC_BITS != 8){
390                     v >>= 8 - FRAC_BITS;
391                 }
392                 if(add){
393                     v += dst[x + y*dst_stride];
394                     v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
395                     if(v&(~255)) v= ~(v>>31);
396                     dst8[x + y*src_stride] = v;
397                 }else{
398                     dst[x + y*dst_stride] -= v;
399                 }
400             }
401         }
402     }
403 }
404 
predict_slice(SnowContext * s,IDWTELEM * buf,int plane_index,int add,int mb_y)405 static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
406     Plane *p= &s->plane[plane_index];
407     const int mb_w= s->b_width  << s->block_max_depth;
408     const int mb_h= s->b_height << s->block_max_depth;
409     int x, y, mb_x;
410     int block_size = MB_SIZE >> s->block_max_depth;
411     int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
412     int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
413     const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
414     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
415     int ref_stride= s->current_picture->linesize[plane_index];
416     uint8_t *dst8= s->current_picture->data[plane_index];
417     int w= p->width;
418     int h= p->height;
419     av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares
420     if(s->keyframe || (s->avctx->debug&512)){
421         if(mb_y==mb_h)
422             return;
423 
424         if(add){
425             for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
426                 for(x=0; x<w; x++){
427                     int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
428                     v >>= FRAC_BITS;
429                     if(v&(~255)) v= ~(v>>31);
430                     dst8[x + y*ref_stride]= v;
431                 }
432             }
433         }else{
434             for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
435                 for(x=0; x<w; x++){
436                     buf[x + y*w]-= 128<<FRAC_BITS;
437                 }
438             }
439         }
440 
441         return;
442     }
443 
444     for(mb_x=0; mb_x<=mb_w; mb_x++){
445         add_yblock(s, 0, NULL, buf, dst8, obmc,
446                    block_w*mb_x - block_w/2,
447                    block_h*mb_y - block_h/2,
448                    block_w, block_h,
449                    w, h,
450                    w, ref_stride, obmc_stride,
451                    mb_x - 1, mb_y - 1,
452                    add, 1, plane_index);
453     }
454 }
455 
predict_plane(SnowContext * s,IDWTELEM * buf,int plane_index,int add)456 static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
457     const int mb_h= s->b_height << s->block_max_depth;
458     int mb_y;
459     for(mb_y=0; mb_y<=mb_h; mb_y++)
460         predict_slice(s, buf, plane_index, add, mb_y);
461 }
462 
set_blocks(SnowContext * s,int level,int x,int y,int l,int cb,int cr,int mx,int my,int ref,int type)463 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){
464     const int w= s->b_width << s->block_max_depth;
465     const int rem_depth= s->block_max_depth - level;
466     const int index= (x + y*w) << rem_depth;
467     const int block_w= 1<<rem_depth;
468     const int block_h= 1<<rem_depth; //FIXME "w!=h"
469     BlockNode block;
470     int i,j;
471 
472     block.color[0]= l;
473     block.color[1]= cb;
474     block.color[2]= cr;
475     block.mx= mx;
476     block.my= my;
477     block.ref= ref;
478     block.type= type;
479     block.level= level;
480 
481     for(j=0; j<block_h; j++){
482         for(i=0; i<block_w; i++){
483             s->block[index + i + j*w]= block;
484         }
485     }
486 }
487 
init_ref(MotionEstContext * c,uint8_t * src[3],uint8_t * ref[3],uint8_t * ref2[3],int x,int y,int ref_index)488 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){
489     SnowContext *s = c->avctx->priv_data;
490     const int offset[3]= {
491           y*c->  stride + x,
492         ((y*c->uvstride + x)>>s->chroma_h_shift),
493         ((y*c->uvstride + x)>>s->chroma_h_shift),
494     };
495     int i;
496     for(i=0; i<3; i++){
497         c->src[0][i]= src [i];
498         c->ref[0][i]= ref [i] + offset[i];
499     }
500     av_assert2(!ref_index);
501 }
502 
503 
504 /* bitstream functions */
505 
506 extern const int8_t ff_quant3bA[256];
507 
508 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
509 
put_symbol(RangeCoder * c,uint8_t * state,int v,int is_signed)510 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
511     int i;
512 
513     if(v){
514         const int a= FFABS(v);
515         const int e= av_log2(a);
516         const int el= FFMIN(e, 10);
517         put_rac(c, state+0, 0);
518 
519         for(i=0; i<el; i++){
520             put_rac(c, state+1+i, 1);  //1..10
521         }
522         for(; i<e; i++){
523             put_rac(c, state+1+9, 1);  //1..10
524         }
525         put_rac(c, state+1+FFMIN(i,9), 0);
526 
527         for(i=e-1; i>=el; i--){
528             put_rac(c, state+22+9, (a>>i)&1); //22..31
529         }
530         for(; i>=0; i--){
531             put_rac(c, state+22+i, (a>>i)&1); //22..31
532         }
533 
534         if(is_signed)
535             put_rac(c, state+11 + el, v < 0); //11..21
536     }else{
537         put_rac(c, state+0, 1);
538     }
539 }
540 
get_symbol(RangeCoder * c,uint8_t * state,int is_signed)541 static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
542     if(get_rac(c, state+0))
543         return 0;
544     else{
545         int i, e;
546         unsigned a;
547         e= 0;
548         while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
549             e++;
550             if (e > 31)
551                 return AVERROR_INVALIDDATA;
552         }
553 
554         a= 1;
555         for(i=e-1; i>=0; i--){
556             a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
557         }
558 
559         e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21
560         return (a^e)-e;
561     }
562 }
563 
put_symbol2(RangeCoder * c,uint8_t * state,int v,int log2)564 static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
565     int i;
566     int r= log2>=0 ? 1<<log2 : 1;
567 
568     av_assert2(v>=0);
569     av_assert2(log2>=-4);
570 
571     while(v >= r){
572         put_rac(c, state+4+log2, 1);
573         v -= r;
574         log2++;
575         if(log2>0) r+=r;
576     }
577     put_rac(c, state+4+log2, 0);
578 
579     for(i=log2-1; i>=0; i--){
580         put_rac(c, state+31-i, (v>>i)&1);
581     }
582 }
583 
get_symbol2(RangeCoder * c,uint8_t * state,int log2)584 static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
585     int i;
586     int r= log2>=0 ? 1<<log2 : 1;
587     int v=0;
588 
589     av_assert2(log2>=-4);
590 
591     while(log2<28 && get_rac(c, state+4+log2)){
592         v+= r;
593         log2++;
594         if(log2>0) r+=r;
595     }
596 
597     for(i=log2-1; i>=0; i--){
598         v+= get_rac(c, state+31-i)<<i;
599     }
600 
601     return v;
602 }
603 
unpack_coeffs(SnowContext * s,SubBand * b,SubBand * parent,int orientation)604 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
605     const int w= b->width;
606     const int h= b->height;
607     int x,y;
608 
609     int run, runs;
610     x_and_coeff *xc= b->x_coeff;
611     x_and_coeff *prev_xc= NULL;
612     x_and_coeff *prev2_xc= xc;
613     x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
614     x_and_coeff *prev_parent_xc= parent_xc;
615 
616     runs= get_symbol2(&s->c, b->state[30], 0);
617     if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
618     else           run= INT_MAX;
619 
620     for(y=0; y<h; y++){
621         int v=0;
622         int lt=0, t=0, rt=0;
623 
624         if(y && prev_xc->x == 0){
625             rt= prev_xc->coeff;
626         }
627         for(x=0; x<w; x++){
628             int p=0;
629             const int l= v;
630 
631             lt= t; t= rt;
632 
633             if(y){
634                 if(prev_xc->x <= x)
635                     prev_xc++;
636                 if(prev_xc->x == x + 1)
637                     rt= prev_xc->coeff;
638                 else
639                     rt=0;
640             }
641             if(parent_xc){
642                 if(x>>1 > parent_xc->x){
643                     parent_xc++;
644                 }
645                 if(x>>1 == parent_xc->x){
646                     p= parent_xc->coeff;
647                 }
648             }
649             if(/*ll|*/l|lt|t|rt|p){
650                 int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
651 
652                 v=get_rac(&s->c, &b->state[0][context]);
653                 if(v){
654                     v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
655                     v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l&0xFF] + 3*ff_quant3bA[t&0xFF]]);
656                     if ((uint16_t)v != v) {
657                         av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n");
658                         v = 1;
659                     }
660                     xc->x=x;
661                     (xc++)->coeff= v;
662                 }
663             }else{
664                 if(!run){
665                     if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
666                     else           run= INT_MAX;
667                     v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
668                     v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
669                     if ((uint16_t)v != v) {
670                         av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n");
671                         v = 1;
672                     }
673 
674                     xc->x=x;
675                     (xc++)->coeff= v;
676                 }else{
677                     int max_run;
678                     run--;
679                     v=0;
680                     av_assert2(run >= 0);
681                     if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
682                     else  max_run= FFMIN(run, w-x-1);
683                     if(parent_xc)
684                         max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
685                     av_assert2(max_run >= 0 && max_run <= run);
686 
687                     x+= max_run;
688                     run-= max_run;
689                 }
690             }
691         }
692         (xc++)->x= w+1; //end marker
693         prev_xc= prev2_xc;
694         prev2_xc= xc;
695 
696         if(parent_xc){
697             if(y&1){
698                 while(parent_xc->x != parent->width+1)
699                     parent_xc++;
700                 parent_xc++;
701                 prev_parent_xc= parent_xc;
702             }else{
703                 parent_xc= prev_parent_xc;
704             }
705         }
706     }
707 
708     (xc++)->x= w+1; //end marker
709 }
710 
711 #endif /* AVCODEC_SNOW_H */
712