blob: 59c710b5f9fdc459f92c830993c2aea0b83cf384
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 | #define FF_MPV_OFFSET(x) (offsetof(MpegEncContext, x) + offsetof(SnowContext, m)) |
36 | #include "mpegvideo.h" |
37 | #include "h264qpel.h" |
38 | |
39 | #define MID_STATE 128 |
40 | |
41 | #define MAX_PLANES 4 |
42 | #define QSHIFT 5 |
43 | #define QROOT (1<<QSHIFT) |
44 | #define LOSSLESS_QLOG -128 |
45 | #define FRAC_BITS 4 |
46 | #define MAX_REF_FRAMES 8 |
47 | |
48 | #define LOG2_OBMC_MAX 8 |
49 | #define OBMC_MAX (1<<(LOG2_OBMC_MAX)) |
50 | typedef struct BlockNode{ |
51 | int16_t mx; ///< Motion vector component X, see mv_scale |
52 | int16_t my; ///< Motion vector component Y, see mv_scale |
53 | uint8_t ref; ///< Reference frame index |
54 | uint8_t color[3]; ///< Color for intra |
55 | uint8_t type; ///< Bitfield of BLOCK_* |
56 | //#define TYPE_SPLIT 1 |
57 | #define BLOCK_INTRA 1 ///< Intra block, inter otherwise |
58 | #define BLOCK_OPT 2 ///< Block needs no checks in this round of iterative motion estiation |
59 | //#define TYPE_NOCOLOR 4 |
60 | uint8_t level; //FIXME merge into type? |
61 | }BlockNode; |
62 | |
63 | static const BlockNode null_block= { //FIXME add border maybe |
64 | .color= {128,128,128}, |
65 | .mx= 0, |
66 | .my= 0, |
67 | .ref= 0, |
68 | .type= 0, |
69 | .level= 0, |
70 | }; |
71 | |
72 | #define LOG2_MB_SIZE 4 |
73 | #define MB_SIZE (1<<LOG2_MB_SIZE) |
74 | #define ENCODER_EXTRA_BITS 4 |
75 | #define HTAPS_MAX 8 |
76 | |
77 | typedef struct x_and_coeff{ |
78 | int16_t x; |
79 | uint16_t coeff; |
80 | } x_and_coeff; |
81 | |
82 | typedef struct SubBand{ |
83 | int level; |
84 | int stride; |
85 | int width; |
86 | int height; |
87 | int qlog; ///< log(qscale)/log[2^(1/6)] |
88 | DWTELEM *buf; |
89 | IDWTELEM *ibuf; |
90 | int buf_x_offset; |
91 | int buf_y_offset; |
92 | int stride_line; ///< Stride measured in lines, not pixels. |
93 | x_and_coeff * x_coeff; |
94 | struct SubBand *parent; |
95 | uint8_t state[/*7*2*/ 7 + 512][32]; |
96 | }SubBand; |
97 | |
98 | typedef struct Plane{ |
99 | int width; |
100 | int height; |
101 | SubBand band[MAX_DECOMPOSITIONS][4]; |
102 | |
103 | int htaps; |
104 | int8_t hcoeff[HTAPS_MAX/2]; |
105 | int diag_mc; |
106 | int fast_mc; |
107 | |
108 | int last_htaps; |
109 | int8_t last_hcoeff[HTAPS_MAX/2]; |
110 | int last_diag_mc; |
111 | }Plane; |
112 | |
113 | typedef struct SnowContext{ |
114 | AVClass *class; |
115 | AVCodecContext *avctx; |
116 | RangeCoder c; |
117 | MECmpContext mecc; |
118 | HpelDSPContext hdsp; |
119 | QpelDSPContext qdsp; |
120 | VideoDSPContext vdsp; |
121 | H264QpelContext h264qpel; |
122 | MpegvideoEncDSPContext mpvencdsp; |
123 | SnowDWTContext dwt; |
124 | AVFrame *input_picture; ///< new_picture with the internal linesizes |
125 | AVFrame *current_picture; |
126 | AVFrame *last_picture[MAX_REF_FRAMES]; |
127 | uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4]; |
128 | AVFrame *mconly_picture; |
129 | // uint8_t q_context[16]; |
130 | uint8_t header_state[32]; |
131 | uint8_t block_state[128 + 32*128]; |
132 | int keyframe; |
133 | int always_reset; |
134 | int version; |
135 | int spatial_decomposition_type; |
136 | int last_spatial_decomposition_type; |
137 | int temporal_decomposition_type; |
138 | int spatial_decomposition_count; |
139 | int last_spatial_decomposition_count; |
140 | int temporal_decomposition_count; |
141 | int max_ref_frames; |
142 | int ref_frames; |
143 | int16_t (*ref_mvs[MAX_REF_FRAMES])[2]; |
144 | uint32_t *ref_scores[MAX_REF_FRAMES]; |
145 | DWTELEM *spatial_dwt_buffer; |
146 | DWTELEM *temp_dwt_buffer; |
147 | IDWTELEM *spatial_idwt_buffer; |
148 | IDWTELEM *temp_idwt_buffer; |
149 | int *run_buffer; |
150 | int colorspace_type; |
151 | int chroma_h_shift; |
152 | int chroma_v_shift; |
153 | int spatial_scalability; |
154 | int qlog; |
155 | int last_qlog; |
156 | int lambda; |
157 | int lambda2; |
158 | int pass1_rc; |
159 | int mv_scale; |
160 | int last_mv_scale; |
161 | int qbias; |
162 | int last_qbias; |
163 | #define QBIAS_SHIFT 3 |
164 | int b_width; |
165 | int b_height; |
166 | int block_max_depth; |
167 | int last_block_max_depth; |
168 | int nb_planes; |
169 | Plane plane[MAX_PLANES]; |
170 | BlockNode *block; |
171 | #define ME_CACHE_SIZE 1024 |
172 | unsigned me_cache[ME_CACHE_SIZE]; |
173 | unsigned me_cache_generation; |
174 | slice_buffer sb; |
175 | int memc_only; |
176 | int no_bitstream; |
177 | int intra_penalty; |
178 | int motion_est; |
179 | int iterative_dia_size; |
180 | int scenechange_threshold; |
181 | |
182 | 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) |
183 | |
184 | uint8_t *scratchbuf; |
185 | uint8_t *emu_edge_buffer; |
186 | |
187 | AVMotionVector *avmv; |
188 | int avmv_index; |
189 | uint64_t encoding_error[AV_NUM_DATA_POINTERS]; |
190 | |
191 | int pred; |
192 | }SnowContext; |
193 | |
194 | /* Tables */ |
195 | extern const uint8_t * const ff_obmc_tab[4]; |
196 | extern uint8_t ff_qexp[QROOT]; |
197 | extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES]; |
198 | |
199 | /* C bits used by mmx/sse2/altivec */ |
200 | |
201 | static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){ |
202 | (*i) = (width) - 2; |
203 | |
204 | if (width & 1){ |
205 | low[(*i)+1] = low[((*i)+1)>>1]; |
206 | (*i)--; |
207 | } |
208 | } |
209 | |
210 | static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){ |
211 | for (; (*i)>=0; (*i)-=2){ |
212 | low[(*i)+1] = high[(*i)>>1]; |
213 | low[*i] = low[(*i)>>1]; |
214 | } |
215 | } |
216 | |
217 | 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){ |
218 | for(; i<w; i++){ |
219 | dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift); |
220 | } |
221 | |
222 | if((width^lift_high)&1){ |
223 | dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift); |
224 | } |
225 | } |
226 | |
227 | static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){ |
228 | for(; i<w; i++){ |
229 | dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS); |
230 | } |
231 | |
232 | if(width&1){ |
233 | dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS); |
234 | } |
235 | } |
236 | |
237 | /* common code */ |
238 | |
239 | int ff_snow_common_init(AVCodecContext *avctx); |
240 | int ff_snow_common_init_after_header(AVCodecContext *avctx); |
241 | void ff_snow_common_end(SnowContext *s); |
242 | void ff_snow_release_buffer(AVCodecContext *avctx); |
243 | void ff_snow_reset_contexts(SnowContext *s); |
244 | int ff_snow_alloc_blocks(SnowContext *s); |
245 | int ff_snow_frame_start(SnowContext *s); |
246 | void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride, |
247 | int sx, int sy, int b_w, int b_h, const BlockNode *block, |
248 | int plane_index, int w, int h); |
249 | int ff_snow_get_buffer(SnowContext *s, AVFrame *frame); |
250 | /* common inline functions */ |
251 | //XXX doublecheck all of them should stay inlined |
252 | |
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); |
258 | }else{ |
259 | const int *scale = ff_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); |
266 | } |
267 | } |
268 | |
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])); |
272 | }else{ |
273 | return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA)); |
274 | } |
275 | } |
276 | |
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]; |
284 | BlockNode *rt= lt+1; |
285 | BlockNode *lb= lt+b_stride; |
286 | BlockNode *rb= lb+1; |
287 | uint8_t *block[4]; |
288 | // When src_stride is large enough, it is possible to interleave the blocks. |
289 | // Otherwise the blocks are written sequentially in the tmp buffer. |
290 | int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; |
291 | uint8_t *tmp = s->scratchbuf; |
292 | uint8_t *ptmp; |
293 | int x,y; |
294 | |
295 | if(b_x<0){ |
296 | lt= rt; |
297 | lb= rb; |
298 | }else if(b_x + 1 >= b_width){ |
299 | rt= lt; |
300 | rb= lb; |
301 | } |
302 | if(b_y<0){ |
303 | lt= lb; |
304 | rt= rb; |
305 | }else if(b_y + 1 >= b_height){ |
306 | lb= lt; |
307 | rb= rt; |
308 | } |
309 | |
310 | if(src_x<0){ //FIXME merge with prev & always round internal width up to *16 |
311 | obmc -= src_x; |
312 | b_w += src_x; |
313 | if(!sliced && !offset_dst) |
314 | dst -= src_x; |
315 | src_x=0; |
316 | } |
317 | if(src_x + b_w > w){ |
318 | b_w = w - src_x; |
319 | } |
320 | if(src_y<0){ |
321 | obmc -= src_y*obmc_stride; |
322 | b_h += src_y; |
323 | if(!sliced && !offset_dst) |
324 | dst -= src_y*dst_stride; |
325 | src_y=0; |
326 | } |
327 | if(src_y + b_h> h){ |
328 | b_h = h - src_y; |
329 | } |
330 | |
331 | if(b_w<=0 || b_h<=0) return; |
332 | |
333 | if(!sliced && offset_dst) |
334 | dst += src_x + src_y*dst_stride; |
335 | dst8+= src_x + src_y*src_stride; |
336 | // src += src_x + src_y*src_stride; |
337 | |
338 | ptmp= tmp + 3*tmp_step; |
339 | block[0]= ptmp; |
340 | ptmp+=tmp_step; |
341 | ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); |
342 | |
343 | if(same_block(lt, rt)){ |
344 | block[1]= block[0]; |
345 | }else{ |
346 | block[1]= ptmp; |
347 | ptmp+=tmp_step; |
348 | ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); |
349 | } |
350 | |
351 | if(same_block(lt, lb)){ |
352 | block[2]= block[0]; |
353 | }else if(same_block(rt, lb)){ |
354 | block[2]= block[1]; |
355 | }else{ |
356 | block[2]= ptmp; |
357 | ptmp+=tmp_step; |
358 | ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); |
359 | } |
360 | |
361 | if(same_block(lt, rb) ){ |
362 | block[3]= block[0]; |
363 | }else if(same_block(rt, rb)){ |
364 | block[3]= block[1]; |
365 | }else if(same_block(lb, rb)){ |
366 | block[3]= block[2]; |
367 | }else{ |
368 | block[3]= ptmp; |
369 | ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); |
370 | } |
371 | if(sliced){ |
372 | s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); |
373 | }else{ |
374 | for(y=0; y<b_h; y++){ |
375 | //FIXME ugly misuse of obmc_stride |
376 | const uint8_t *obmc1= obmc + y*obmc_stride; |
377 | const uint8_t *obmc2= obmc1+ (obmc_stride>>1); |
378 | const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); |
379 | const uint8_t *obmc4= obmc3+ (obmc_stride>>1); |
380 | for(x=0; x<b_w; x++){ |
381 | int v= obmc1[x] * block[3][x + y*src_stride] |
382 | +obmc2[x] * block[2][x + y*src_stride] |
383 | +obmc3[x] * block[1][x + y*src_stride] |
384 | +obmc4[x] * block[0][x + y*src_stride]; |
385 | |
386 | v <<= 8 - LOG2_OBMC_MAX; |
387 | if(FRAC_BITS != 8){ |
388 | v >>= 8 - FRAC_BITS; |
389 | } |
390 | if(add){ |
391 | v += dst[x + y*dst_stride]; |
392 | v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
393 | if(v&(~255)) v= ~(v>>31); |
394 | dst8[x + y*src_stride] = v; |
395 | }else{ |
396 | dst[x + y*dst_stride] -= v; |
397 | } |
398 | } |
399 | } |
400 | } |
401 | } |
402 | |
403 | static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){ |
404 | Plane *p= &s->plane[plane_index]; |
405 | const int mb_w= s->b_width << s->block_max_depth; |
406 | const int mb_h= s->b_height << s->block_max_depth; |
407 | int x, y, mb_x; |
408 | int block_size = MB_SIZE >> s->block_max_depth; |
409 | int block_w = plane_index ? block_size>>s->chroma_h_shift : block_size; |
410 | int block_h = plane_index ? block_size>>s->chroma_v_shift : block_size; |
411 | const uint8_t *obmc = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth]; |
412 | const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size; |
413 | int ref_stride= s->current_picture->linesize[plane_index]; |
414 | uint8_t *dst8= s->current_picture->data[plane_index]; |
415 | int w= p->width; |
416 | int h= p->height; |
417 | av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares |
418 | if(s->keyframe || (s->avctx->debug&512)){ |
419 | if(mb_y==mb_h) |
420 | return; |
421 | |
422 | if(add){ |
423 | for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){ |
424 | for(x=0; x<w; x++){ |
425 | int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
426 | v >>= FRAC_BITS; |
427 | if(v&(~255)) v= ~(v>>31); |
428 | dst8[x + y*ref_stride]= v; |
429 | } |
430 | } |
431 | }else{ |
432 | for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){ |
433 | for(x=0; x<w; x++){ |
434 | buf[x + y*w]-= 128<<FRAC_BITS; |
435 | } |
436 | } |
437 | } |
438 | |
439 | return; |
440 | } |
441 | |
442 | for(mb_x=0; mb_x<=mb_w; mb_x++){ |
443 | add_yblock(s, 0, NULL, buf, dst8, obmc, |
444 | block_w*mb_x - block_w/2, |
445 | block_h*mb_y - block_h/2, |
446 | block_w, block_h, |
447 | w, h, |
448 | w, ref_stride, obmc_stride, |
449 | mb_x - 1, mb_y - 1, |
450 | add, 1, plane_index); |
451 | } |
452 | } |
453 | |
454 | static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){ |
455 | const int mb_h= s->b_height << s->block_max_depth; |
456 | int mb_y; |
457 | for(mb_y=0; mb_y<=mb_h; mb_y++) |
458 | predict_slice(s, buf, plane_index, add, mb_y); |
459 | } |
460 | |
461 | 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){ |
462 | const int w= s->b_width << s->block_max_depth; |
463 | const int rem_depth= s->block_max_depth - level; |
464 | const int index= (x + y*w) << rem_depth; |
465 | const int block_w= 1<<rem_depth; |
466 | const int block_h= 1<<rem_depth; //FIXME "w!=h" |
467 | BlockNode block; |
468 | int i,j; |
469 | |
470 | block.color[0]= l; |
471 | block.color[1]= cb; |
472 | block.color[2]= cr; |
473 | block.mx= mx; |
474 | block.my= my; |
475 | block.ref= ref; |
476 | block.type= type; |
477 | block.level= level; |
478 | |
479 | for(j=0; j<block_h; j++){ |
480 | for(i=0; i<block_w; i++){ |
481 | s->block[index + i + j*w]= block; |
482 | } |
483 | } |
484 | } |
485 | |
486 | 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){ |
487 | SnowContext *s = c->avctx->priv_data; |
488 | const int offset[3]= { |
489 | y*c-> stride + x, |
490 | ((y*c->uvstride + x)>>s->chroma_h_shift), |
491 | ((y*c->uvstride + x)>>s->chroma_h_shift), |
492 | }; |
493 | int i; |
494 | for(i=0; i<3; i++){ |
495 | c->src[0][i]= src [i]; |
496 | c->ref[0][i]= ref [i] + offset[i]; |
497 | } |
498 | av_assert2(!ref_index); |
499 | } |
500 | |
501 | |
502 | /* bitstream functions */ |
503 | |
504 | extern const int8_t ff_quant3bA[256]; |
505 | |
506 | #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0 |
507 | |
508 | static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){ |
509 | int i; |
510 | |
511 | if(v){ |
512 | const int a= FFABS(v); |
513 | const int e= av_log2(a); |
514 | const int el= FFMIN(e, 10); |
515 | put_rac(c, state+0, 0); |
516 | |
517 | for(i=0; i<el; i++){ |
518 | put_rac(c, state+1+i, 1); //1..10 |
519 | } |
520 | for(; i<e; i++){ |
521 | put_rac(c, state+1+9, 1); //1..10 |
522 | } |
523 | put_rac(c, state+1+FFMIN(i,9), 0); |
524 | |
525 | for(i=e-1; i>=el; i--){ |
526 | put_rac(c, state+22+9, (a>>i)&1); //22..31 |
527 | } |
528 | for(; i>=0; i--){ |
529 | put_rac(c, state+22+i, (a>>i)&1); //22..31 |
530 | } |
531 | |
532 | if(is_signed) |
533 | put_rac(c, state+11 + el, v < 0); //11..21 |
534 | }else{ |
535 | put_rac(c, state+0, 1); |
536 | } |
537 | } |
538 | |
539 | static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ |
540 | if(get_rac(c, state+0)) |
541 | return 0; |
542 | else{ |
543 | int i, e, a; |
544 | e= 0; |
545 | while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10 |
546 | e++; |
547 | if (e > 31) |
548 | return AVERROR_INVALIDDATA; |
549 | } |
550 | |
551 | a= 1; |
552 | for(i=e-1; i>=0; i--){ |
553 | a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31 |
554 | } |
555 | |
556 | e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21 |
557 | return (a^e)-e; |
558 | } |
559 | } |
560 | |
561 | static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){ |
562 | int i; |
563 | int r= log2>=0 ? 1<<log2 : 1; |
564 | |
565 | av_assert2(v>=0); |
566 | av_assert2(log2>=-4); |
567 | |
568 | while(v >= r){ |
569 | put_rac(c, state+4+log2, 1); |
570 | v -= r; |
571 | log2++; |
572 | if(log2>0) r+=r; |
573 | } |
574 | put_rac(c, state+4+log2, 0); |
575 | |
576 | for(i=log2-1; i>=0; i--){ |
577 | put_rac(c, state+31-i, (v>>i)&1); |
578 | } |
579 | } |
580 | |
581 | static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){ |
582 | int i; |
583 | int r= log2>=0 ? 1<<log2 : 1; |
584 | int v=0; |
585 | |
586 | av_assert2(log2>=-4); |
587 | |
588 | while(log2<28 && get_rac(c, state+4+log2)){ |
589 | v+= r; |
590 | log2++; |
591 | if(log2>0) r+=r; |
592 | } |
593 | |
594 | for(i=log2-1; i>=0; i--){ |
595 | v+= get_rac(c, state+31-i)<<i; |
596 | } |
597 | |
598 | return v; |
599 | } |
600 | |
601 | static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){ |
602 | const int w= b->width; |
603 | const int h= b->height; |
604 | int x,y; |
605 | |
606 | int run, runs; |
607 | x_and_coeff *xc= b->x_coeff; |
608 | x_and_coeff *prev_xc= NULL; |
609 | x_and_coeff *prev2_xc= xc; |
610 | x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL; |
611 | x_and_coeff *prev_parent_xc= parent_xc; |
612 | |
613 | runs= get_symbol2(&s->c, b->state[30], 0); |
614 | if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
615 | else run= INT_MAX; |
616 | |
617 | for(y=0; y<h; y++){ |
618 | int v=0; |
619 | int lt=0, t=0, rt=0; |
620 | |
621 | if(y && prev_xc->x == 0){ |
622 | rt= prev_xc->coeff; |
623 | } |
624 | for(x=0; x<w; x++){ |
625 | int p=0; |
626 | const int l= v; |
627 | |
628 | lt= t; t= rt; |
629 | |
630 | if(y){ |
631 | if(prev_xc->x <= x) |
632 | prev_xc++; |
633 | if(prev_xc->x == x + 1) |
634 | rt= prev_xc->coeff; |
635 | else |
636 | rt=0; |
637 | } |
638 | if(parent_xc){ |
639 | if(x>>1 > parent_xc->x){ |
640 | parent_xc++; |
641 | } |
642 | if(x>>1 == parent_xc->x){ |
643 | p= parent_xc->coeff; |
644 | } |
645 | } |
646 | if(/*ll|*/l|lt|t|rt|p){ |
647 | int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); |
648 | |
649 | v=get_rac(&s->c, &b->state[0][context]); |
650 | if(v){ |
651 | v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); |
652 | v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l&0xFF] + 3*ff_quant3bA[t&0xFF]]); |
653 | if ((uint16_t)v != v) { |
654 | av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n"); |
655 | v = 1; |
656 | } |
657 | xc->x=x; |
658 | (xc++)->coeff= v; |
659 | } |
660 | }else{ |
661 | if(!run){ |
662 | if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
663 | else run= INT_MAX; |
664 | v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); |
665 | v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); |
666 | if ((uint16_t)v != v) { |
667 | av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n"); |
668 | v = 1; |
669 | } |
670 | |
671 | xc->x=x; |
672 | (xc++)->coeff= v; |
673 | }else{ |
674 | int max_run; |
675 | run--; |
676 | v=0; |
677 | av_assert2(run >= 0); |
678 | if(y) max_run= FFMIN(run, prev_xc->x - x - 2); |
679 | else max_run= FFMIN(run, w-x-1); |
680 | if(parent_xc) |
681 | max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); |
682 | av_assert2(max_run >= 0 && max_run <= run); |
683 | |
684 | x+= max_run; |
685 | run-= max_run; |
686 | } |
687 | } |
688 | } |
689 | (xc++)->x= w+1; //end marker |
690 | prev_xc= prev2_xc; |
691 | prev2_xc= xc; |
692 | |
693 | if(parent_xc){ |
694 | if(y&1){ |
695 | while(parent_xc->x != parent->width+1) |
696 | parent_xc++; |
697 | parent_xc++; |
698 | prev_parent_xc= parent_xc; |
699 | }else{ |
700 | parent_xc= prev_parent_xc; |
701 | } |
702 | } |
703 | } |
704 | |
705 | (xc++)->x= w+1; //end marker |
706 | } |
707 | |
708 | #endif /* AVCODEC_SNOW_H */ |
709 |