Directory: | ../../../ffmpeg/ |
---|---|
File: | src/libavcodec/snow.h |
Date: | 2022-07-07 01:21:54 |
Exec | Total | Coverage | |
---|---|---|---|
Lines: | 292 | 309 | 94.5% |
Branches: | 194 | 210 | 92.4% |
Line | Branch | Exec | Source |
---|---|---|---|
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 SNOW_MAX_PLANES 4 | ||
39 | |||
40 | #define FF_ME_ITER 3 | ||
41 | |||
42 | #define MID_STATE 128 | ||
43 | |||
44 | #define MAX_PLANES 4 | ||
45 | #define QSHIFT 5 | ||
46 | #define QROOT (1<<QSHIFT) | ||
47 | #define LOSSLESS_QLOG -128 | ||
48 | #define FRAC_BITS 4 | ||
49 | #define MAX_REF_FRAMES 8 | ||
50 | |||
51 | #define LOG2_OBMC_MAX 8 | ||
52 | #define OBMC_MAX (1<<(LOG2_OBMC_MAX)) | ||
53 | typedef struct BlockNode{ | ||
54 | int16_t mx; ///< Motion vector component X, see mv_scale | ||
55 | int16_t my; ///< Motion vector component Y, see mv_scale | ||
56 | uint8_t ref; ///< Reference frame index | ||
57 | uint8_t color[3]; ///< Color for intra | ||
58 | uint8_t type; ///< Bitfield of BLOCK_* | ||
59 | //#define TYPE_SPLIT 1 | ||
60 | #define BLOCK_INTRA 1 ///< Intra block, inter otherwise | ||
61 | #define BLOCK_OPT 2 ///< Block needs no checks in this round of iterative motion estiation | ||
62 | //#define TYPE_NOCOLOR 4 | ||
63 | uint8_t level; //FIXME merge into type? | ||
64 | }BlockNode; | ||
65 | |||
66 | static const BlockNode null_block= { //FIXME add border maybe | ||
67 | .color= {128,128,128}, | ||
68 | .mx= 0, | ||
69 | .my= 0, | ||
70 | .ref= 0, | ||
71 | .type= 0, | ||
72 | .level= 0, | ||
73 | }; | ||
74 | |||
75 | #define LOG2_MB_SIZE 4 | ||
76 | #define MB_SIZE (1<<LOG2_MB_SIZE) | ||
77 | #define ENCODER_EXTRA_BITS 4 | ||
78 | #define HTAPS_MAX 8 | ||
79 | |||
80 | typedef struct x_and_coeff{ | ||
81 | int16_t x; | ||
82 | uint16_t coeff; | ||
83 | } x_and_coeff; | ||
84 | |||
85 | typedef struct SubBand{ | ||
86 | int level; | ||
87 | int stride; | ||
88 | int width; | ||
89 | int height; | ||
90 | int qlog; ///< log(qscale)/log[2^(1/6)] | ||
91 | DWTELEM *buf; | ||
92 | IDWTELEM *ibuf; | ||
93 | int buf_x_offset; | ||
94 | int buf_y_offset; | ||
95 | int stride_line; ///< Stride measured in lines, not pixels. | ||
96 | x_and_coeff * x_coeff; | ||
97 | struct SubBand *parent; | ||
98 | uint8_t state[/*7*2*/ 7 + 512][32]; | ||
99 | }SubBand; | ||
100 | |||
101 | typedef struct Plane{ | ||
102 | int width; | ||
103 | int height; | ||
104 | SubBand band[MAX_DECOMPOSITIONS][4]; | ||
105 | |||
106 | int htaps; | ||
107 | int8_t hcoeff[HTAPS_MAX/2]; | ||
108 | int diag_mc; | ||
109 | int fast_mc; | ||
110 | |||
111 | int last_htaps; | ||
112 | int8_t last_hcoeff[HTAPS_MAX/2]; | ||
113 | int last_diag_mc; | ||
114 | }Plane; | ||
115 | |||
116 | typedef struct SnowContext{ | ||
117 | AVClass *class; | ||
118 | AVCodecContext *avctx; | ||
119 | RangeCoder c; | ||
120 | MECmpContext mecc; | ||
121 | HpelDSPContext hdsp; | ||
122 | QpelDSPContext qdsp; | ||
123 | VideoDSPContext vdsp; | ||
124 | H264QpelContext h264qpel; | ||
125 | MpegvideoEncDSPContext mpvencdsp; | ||
126 | SnowDWTContext dwt; | ||
127 | AVFrame *input_picture; ///< new_picture with the internal linesizes | ||
128 | AVFrame *current_picture; | ||
129 | AVFrame *last_picture[MAX_REF_FRAMES]; | ||
130 | uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4]; | ||
131 | AVFrame *mconly_picture; | ||
132 | // uint8_t q_context[16]; | ||
133 | uint8_t header_state[32]; | ||
134 | uint8_t block_state[128 + 32*128]; | ||
135 | int keyframe; | ||
136 | int always_reset; | ||
137 | int version; | ||
138 | int spatial_decomposition_type; | ||
139 | int last_spatial_decomposition_type; | ||
140 | int temporal_decomposition_type; | ||
141 | int spatial_decomposition_count; | ||
142 | int last_spatial_decomposition_count; | ||
143 | int temporal_decomposition_count; | ||
144 | int max_ref_frames; | ||
145 | int ref_frames; | ||
146 | int16_t (*ref_mvs[MAX_REF_FRAMES])[2]; | ||
147 | uint32_t *ref_scores[MAX_REF_FRAMES]; | ||
148 | DWTELEM *spatial_dwt_buffer; | ||
149 | DWTELEM *temp_dwt_buffer; | ||
150 | IDWTELEM *spatial_idwt_buffer; | ||
151 | IDWTELEM *temp_idwt_buffer; | ||
152 | int *run_buffer; | ||
153 | int colorspace_type; | ||
154 | int chroma_h_shift; | ||
155 | int chroma_v_shift; | ||
156 | int spatial_scalability; | ||
157 | int qlog; | ||
158 | int last_qlog; | ||
159 | int lambda; | ||
160 | int lambda2; | ||
161 | int pass1_rc; | ||
162 | int mv_scale; | ||
163 | int last_mv_scale; | ||
164 | int qbias; | ||
165 | int last_qbias; | ||
166 | #define QBIAS_SHIFT 3 | ||
167 | int b_width; | ||
168 | int b_height; | ||
169 | int block_max_depth; | ||
170 | int last_block_max_depth; | ||
171 | int nb_planes; | ||
172 | Plane plane[MAX_PLANES]; | ||
173 | BlockNode *block; | ||
174 | #define ME_CACHE_SIZE 1024 | ||
175 | unsigned me_cache[ME_CACHE_SIZE]; | ||
176 | unsigned me_cache_generation; | ||
177 | slice_buffer sb; | ||
178 | int memc_only; | ||
179 | int no_bitstream; | ||
180 | int intra_penalty; | ||
181 | int motion_est; | ||
182 | int iterative_dia_size; | ||
183 | int scenechange_threshold; | ||
184 | |||
185 | 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) | ||
186 | |||
187 | uint8_t *scratchbuf; | ||
188 | uint8_t *emu_edge_buffer; | ||
189 | |||
190 | AVMotionVector *avmv; | ||
191 | unsigned avmv_size; | ||
192 | int avmv_index; | ||
193 | uint64_t encoding_error[SNOW_MAX_PLANES]; | ||
194 | |||
195 | int pred; | ||
196 | }SnowContext; | ||
197 | |||
198 | /* Tables */ | ||
199 | extern const uint8_t * const ff_obmc_tab[4]; | ||
200 | extern const uint8_t ff_qexp[QROOT]; | ||
201 | extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES]; | ||
202 | |||
203 | /* C bits used by mmx/sse2/altivec */ | ||
204 | |||
205 | 248 | static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){ | |
206 | 248 | (*i) = (width) - 2; | |
207 | |||
208 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 248 times.
|
248 | if (width & 1){ |
209 | ✗ | low[(*i)+1] = low[((*i)+1)>>1]; | |
210 | ✗ | (*i)--; | |
211 | } | ||
212 | 248 | } | |
213 | |||
214 | static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){ | ||
215 | for (; (*i)>=0; (*i)-=2){ | ||
216 | low[(*i)+1] = high[(*i)>>1]; | ||
217 | low[*i] = low[(*i)>>1]; | ||
218 | } | ||
219 | } | ||
220 | |||
221 | 744 | 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){ | |
222 |
2/2✓ Branch 0 taken 4992 times.
✓ Branch 1 taken 744 times.
|
5736 | for(; i<w; i++){ |
223 | 4992 | dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift); | |
224 | } | ||
225 | |||
226 |
2/2✓ Branch 0 taken 496 times.
✓ Branch 1 taken 248 times.
|
744 | if((width^lift_high)&1){ |
227 | 496 | dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift); | |
228 | } | ||
229 | 744 | } | |
230 | |||
231 | 248 | static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){ | |
232 |
2/2✓ Branch 0 taken 1664 times.
✓ Branch 1 taken 248 times.
|
1912 | for(; i<w; i++){ |
233 | 1664 | dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS); | |
234 | } | ||
235 | |||
236 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 248 times.
|
248 | if(width&1){ |
237 | ✗ | dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS); | |
238 | } | ||
239 | 248 | } | |
240 | |||
241 | /* common code */ | ||
242 | |||
243 | int ff_snow_common_init(AVCodecContext *avctx); | ||
244 | int ff_snow_common_init_after_header(AVCodecContext *avctx); | ||
245 | void ff_snow_common_end(SnowContext *s); | ||
246 | void ff_snow_release_buffer(AVCodecContext *avctx); | ||
247 | void ff_snow_reset_contexts(SnowContext *s); | ||
248 | int ff_snow_alloc_blocks(SnowContext *s); | ||
249 | int ff_snow_frame_start(SnowContext *s); | ||
250 | void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride, | ||
251 | int sx, int sy, int b_w, int b_h, const BlockNode *block, | ||
252 | int plane_index, int w, int h); | ||
253 | int ff_snow_get_buffer(SnowContext *s, AVFrame *frame); | ||
254 | /* common inline functions */ | ||
255 | //XXX doublecheck all of them should stay inlined | ||
256 | |||
257 | 1565346 | static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref, | |
258 | const BlockNode *left, const BlockNode *top, const BlockNode *tr){ | ||
259 |
1/2✓ Branch 0 taken 1565346 times.
✗ Branch 1 not taken.
|
1565346 | if(s->ref_frames == 1){ |
260 | 1565346 | *mx = mid_pred(left->mx, top->mx, tr->mx); | |
261 | 1565346 | *my = mid_pred(left->my, top->my, tr->my); | |
262 | }else{ | ||
263 | ✗ | const int *scale = ff_scale_mv_ref[ref]; | |
264 | ✗ | *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8, | |
265 | ✗ | (top ->mx * scale[top ->ref] + 128) >>8, | |
266 | ✗ | (tr ->mx * scale[tr ->ref] + 128) >>8); | |
267 | ✗ | *my = mid_pred((left->my * scale[left->ref] + 128) >>8, | |
268 | ✗ | (top ->my * scale[top ->ref] + 128) >>8, | |
269 | ✗ | (tr ->my * scale[tr ->ref] + 128) >>8); | |
270 | } | ||
271 | 1565346 | } | |
272 | |||
273 | 12015986 | static av_always_inline int same_block(BlockNode *a, BlockNode *b){ | |
274 |
4/4✓ Branch 0 taken 398711 times.
✓ Branch 1 taken 11617275 times.
✓ Branch 2 taken 86127 times.
✓ Branch 3 taken 312584 times.
|
12015986 | if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){ |
275 | 86127 | return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2])); | |
276 | }else{ | ||
277 | 11929859 | return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA)); | |
278 | } | ||
279 | } | ||
280 | |||
281 | //FIXME name cleanup (b_w, block_w, b_width stuff) | ||
282 | //XXX should we really inline it? | ||
283 | 2357745 | 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){ | |
284 | 2357745 | const int b_width = s->b_width << s->block_max_depth; | |
285 | 2357745 | const int b_height= s->b_height << s->block_max_depth; | |
286 | 2357745 | const int b_stride= b_width; | |
287 | 2357745 | BlockNode *lt= &s->block[b_x + b_y*b_stride]; | |
288 | 2357745 | BlockNode *rt= lt+1; | |
289 | 2357745 | BlockNode *lb= lt+b_stride; | |
290 | 2357745 | BlockNode *rb= lb+1; | |
291 | uint8_t *block[4]; | ||
292 | // When src_stride is large enough, it is possible to interleave the blocks. | ||
293 | // Otherwise the blocks are written sequentially in the tmp buffer. | ||
294 |
2/2✓ Branch 0 taken 24300 times.
✓ Branch 1 taken 2333445 times.
|
2357745 | int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; |
295 | 2357745 | uint8_t *tmp = s->scratchbuf; | |
296 | uint8_t *ptmp; | ||
297 | int x,y; | ||
298 | |||
299 |
2/2✓ Branch 0 taken 70551 times.
✓ Branch 1 taken 2287194 times.
|
2357745 | if(b_x<0){ |
300 | 70551 | lt= rt; | |
301 | 70551 | lb= rb; | |
302 |
2/2✓ Branch 0 taken 68187 times.
✓ Branch 1 taken 2219007 times.
|
2287194 | }else if(b_x + 1 >= b_width){ |
303 | 68187 | rt= lt; | |
304 | 68187 | rb= lb; | |
305 | } | ||
306 |
2/2✓ Branch 0 taken 106485 times.
✓ Branch 1 taken 2251260 times.
|
2357745 | if(b_y<0){ |
307 | 106485 | lt= lb; | |
308 | 106485 | rt= rb; | |
309 |
2/2✓ Branch 0 taken 98685 times.
✓ Branch 1 taken 2152575 times.
|
2251260 | }else if(b_y + 1 >= b_height){ |
310 | 98685 | lb= lt; | |
311 | 98685 | rb= rt; | |
312 | } | ||
313 | |||
314 |
2/2✓ Branch 0 taken 70551 times.
✓ Branch 1 taken 2287194 times.
|
2357745 | if(src_x<0){ //FIXME merge with prev & always round internal width up to *16 |
315 | 70551 | obmc -= src_x; | |
316 | 70551 | b_w += src_x; | |
317 |
4/4✓ Branch 0 taken 51516 times.
✓ Branch 1 taken 19035 times.
✓ Branch 2 taken 13446 times.
✓ Branch 3 taken 38070 times.
|
70551 | if(!sliced && !offset_dst) |
318 | 13446 | dst -= src_x; | |
319 | 70551 | src_x=0; | |
320 | } | ||
321 |
2/2✓ Branch 0 taken 68187 times.
✓ Branch 1 taken 2289558 times.
|
2357745 | if(src_x + b_w > w){ |
322 | 68187 | b_w = w - src_x; | |
323 | } | ||
324 |
2/2✓ Branch 0 taken 106485 times.
✓ Branch 1 taken 2251260 times.
|
2357745 | if(src_y<0){ |
325 | 106485 | obmc -= src_y*obmc_stride; | |
326 | 106485 | b_h += src_y; | |
327 |
4/4✓ Branch 0 taken 80970 times.
✓ Branch 1 taken 25515 times.
✓ Branch 2 taken 29940 times.
✓ Branch 3 taken 51030 times.
|
106485 | if(!sliced && !offset_dst) |
328 | 29940 | dst -= src_y*dst_stride; | |
329 | 106485 | src_y=0; | |
330 | } | ||
331 |
2/2✓ Branch 0 taken 98685 times.
✓ Branch 1 taken 2259060 times.
|
2357745 | if(src_y + b_h> h){ |
332 | 98685 | b_h = h - src_y; | |
333 | } | ||
334 | |||
335 |
2/4✓ Branch 0 taken 2357745 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 2357745 times.
|
2357745 | if(b_w<=0 || b_h<=0) return; |
336 | |||
337 |
4/4✓ Branch 0 taken 1646970 times.
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 1421550 times.
✓ Branch 3 taken 225420 times.
|
2357745 | if(!sliced && offset_dst) |
338 | 1421550 | dst += src_x + src_y*dst_stride; | |
339 | 2357745 | dst8+= src_x + src_y*src_stride; | |
340 | // src += src_x + src_y*src_stride; | ||
341 | |||
342 | 2357745 | ptmp= tmp + 3*tmp_step; | |
343 | 2357745 | block[0]= ptmp; | |
344 | 2357745 | ptmp+=tmp_step; | |
345 | 2357745 | ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); | |
346 | |||
347 |
2/2✓ Branch 1 taken 785313 times.
✓ Branch 2 taken 1572432 times.
|
2357745 | if(same_block(lt, rt)){ |
348 | 785313 | block[1]= block[0]; | |
349 | }else{ | ||
350 | 1572432 | block[1]= ptmp; | |
351 | 1572432 | ptmp+=tmp_step; | |
352 | 1572432 | ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); | |
353 | } | ||
354 | |||
355 |
2/2✓ Branch 1 taken 792018 times.
✓ Branch 2 taken 1565727 times.
|
2357745 | if(same_block(lt, lb)){ |
356 | 792018 | block[2]= block[0]; | |
357 |
2/2✓ Branch 1 taken 198012 times.
✓ Branch 2 taken 1367715 times.
|
1565727 | }else if(same_block(rt, lb)){ |
358 | 198012 | block[2]= block[1]; | |
359 | }else{ | ||
360 | 1367715 | block[2]= ptmp; | |
361 | 1367715 | ptmp+=tmp_step; | |
362 | 1367715 | ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); | |
363 | } | ||
364 | |||
365 |
2/2✓ Branch 1 taken 442506 times.
✓ Branch 2 taken 1915239 times.
|
2357745 | if(same_block(lt, rb) ){ |
366 | 442506 | block[3]= block[0]; | |
367 |
2/2✓ Branch 1 taken 472239 times.
✓ Branch 2 taken 1443000 times.
|
1915239 | }else if(same_block(rt, rb)){ |
368 | 472239 | block[3]= block[1]; | |
369 |
2/2✓ Branch 1 taken 365625 times.
✓ Branch 2 taken 1077375 times.
|
1443000 | }else if(same_block(lb, rb)){ |
370 | 365625 | block[3]= block[2]; | |
371 | }else{ | ||
372 | 1077375 | block[3]= ptmp; | |
373 | 1077375 | ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); | |
374 | } | ||
375 |
2/2✓ Branch 0 taken 710775 times.
✓ Branch 1 taken 1646970 times.
|
2357745 | if(sliced){ |
376 | 710775 | s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); | |
377 | }else{ | ||
378 |
2/2✓ Branch 0 taken 9747200 times.
✓ Branch 1 taken 1646970 times.
|
11394170 | for(y=0; y<b_h; y++){ |
379 | //FIXME ugly misuse of obmc_stride | ||
380 | 9747200 | const uint8_t *obmc1= obmc + y*obmc_stride; | |
381 | 9747200 | const uint8_t *obmc2= obmc1+ (obmc_stride>>1); | |
382 | 9747200 | const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); | |
383 | 9747200 | const uint8_t *obmc4= obmc3+ (obmc_stride>>1); | |
384 |
2/2✓ Branch 0 taken 71821152 times.
✓ Branch 1 taken 9747200 times.
|
81568352 | for(x=0; x<b_w; x++){ |
385 | 71821152 | int v= obmc1[x] * block[3][x + y*src_stride] | |
386 | 71821152 | +obmc2[x] * block[2][x + y*src_stride] | |
387 | 71821152 | +obmc3[x] * block[1][x + y*src_stride] | |
388 | 71821152 | +obmc4[x] * block[0][x + y*src_stride]; | |
389 | |||
390 | 71821152 | v <<= 8 - LOG2_OBMC_MAX; | |
391 | if(FRAC_BITS != 8){ | ||
392 | 71821152 | v >>= 8 - FRAC_BITS; | |
393 | } | ||
394 |
2/2✓ Branch 0 taken 23846400 times.
✓ Branch 1 taken 47974752 times.
|
71821152 | if(add){ |
395 | 23846400 | v += dst[x + y*dst_stride]; | |
396 | 23846400 | v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; | |
397 |
2/2✓ Branch 0 taken 1144 times.
✓ Branch 1 taken 23845256 times.
|
23846400 | if(v&(~255)) v= ~(v>>31); |
398 | 23846400 | dst8[x + y*src_stride] = v; | |
399 | }else{ | ||
400 | 47974752 | dst[x + y*dst_stride] -= v; | |
401 | } | ||
402 | } | ||
403 | } | ||
404 | } | ||
405 | } | ||
406 | |||
407 | 42300 | static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){ | |
408 | 42300 | Plane *p= &s->plane[plane_index]; | |
409 | 42300 | const int mb_w= s->b_width << s->block_max_depth; | |
410 | 42300 | const int mb_h= s->b_height << s->block_max_depth; | |
411 | int x, y, mb_x; | ||
412 | 42300 | int block_size = MB_SIZE >> s->block_max_depth; | |
413 |
2/2✓ Branch 0 taken 28200 times.
✓ Branch 1 taken 14100 times.
|
42300 | int block_w = plane_index ? block_size>>s->chroma_h_shift : block_size; |
414 |
2/2✓ Branch 0 taken 28200 times.
✓ Branch 1 taken 14100 times.
|
42300 | int block_h = plane_index ? block_size>>s->chroma_v_shift : block_size; |
415 |
2/2✓ Branch 0 taken 28200 times.
✓ Branch 1 taken 14100 times.
|
42300 | const uint8_t *obmc = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth]; |
416 |
2/2✓ Branch 0 taken 28200 times.
✓ Branch 1 taken 14100 times.
|
42300 | const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size; |
417 | 42300 | int ref_stride= s->current_picture->linesize[plane_index]; | |
418 | 42300 | uint8_t *dst8= s->current_picture->data[plane_index]; | |
419 | 42300 | int w= p->width; | |
420 | 42300 | int h= p->height; | |
421 | av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares | ||
422 |
3/4✓ Branch 0 taken 38070 times.
✓ Branch 1 taken 4230 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 38070 times.
|
42300 | if(s->keyframe || (s->avctx->debug&512)){ |
423 |
2/2✓ Branch 0 taken 270 times.
✓ Branch 1 taken 3960 times.
|
4230 | if(mb_y==mb_h) |
424 | 270 | return; | |
425 | |||
426 |
2/2✓ Branch 0 taken 1980 times.
✓ Branch 1 taken 1980 times.
|
3960 | if(add){ |
427 |
2/2✓ Branch 0 taken 12480 times.
✓ Branch 1 taken 1980 times.
|
14460 | for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){ |
428 |
2/2✓ Branch 0 taken 2649600 times.
✓ Branch 1 taken 12480 times.
|
2662080 | for(x=0; x<w; x++){ |
429 | 2649600 | int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); | |
430 | 2649600 | v >>= FRAC_BITS; | |
431 |
2/2✓ Branch 0 taken 104 times.
✓ Branch 1 taken 2649496 times.
|
2649600 | if(v&(~255)) v= ~(v>>31); |
432 | 2649600 | dst8[x + y*ref_stride]= v; | |
433 | } | ||
434 | } | ||
435 | }else{ | ||
436 |
2/2✓ Branch 0 taken 12480 times.
✓ Branch 1 taken 1980 times.
|
14460 | for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){ |
437 |
2/2✓ Branch 0 taken 2649600 times.
✓ Branch 1 taken 12480 times.
|
2662080 | for(x=0; x<w; x++){ |
438 | 2649600 | buf[x + y*w]-= 128<<FRAC_BITS; | |
439 | } | ||
440 | } | ||
441 | } | ||
442 | |||
443 | 3960 | return; | |
444 | } | ||
445 | |||
446 |
2/2✓ Branch 0 taken 1421550 times.
✓ Branch 1 taken 38070 times.
|
1459620 | for(mb_x=0; mb_x<=mb_w; mb_x++){ |
447 | 1421550 | add_yblock(s, 0, NULL, buf, dst8, obmc, | |
448 | 1421550 | block_w*mb_x - block_w/2, | |
449 | 1421550 | block_h*mb_y - block_h/2, | |
450 | block_w, block_h, | ||
451 | w, h, | ||
452 | w, ref_stride, obmc_stride, | ||
453 | mb_x - 1, mb_y - 1, | ||
454 | add, 1, plane_index); | ||
455 | } | ||
456 | } | ||
457 | |||
458 | 2700 | static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){ | |
459 | 2700 | const int mb_h= s->b_height << s->block_max_depth; | |
460 | int mb_y; | ||
461 |
2/2✓ Branch 0 taken 42300 times.
✓ Branch 1 taken 2700 times.
|
45000 | for(mb_y=0; mb_y<=mb_h; mb_y++) |
462 | 42300 | predict_slice(s, buf, plane_index, add, mb_y); | |
463 | 2700 | } | |
464 | |||
465 | 452314 | 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){ | |
466 | 452314 | const int w= s->b_width << s->block_max_depth; | |
467 | 452314 | const int rem_depth= s->block_max_depth - level; | |
468 | 452314 | const int index= (x + y*w) << rem_depth; | |
469 | 452314 | const int block_w= 1<<rem_depth; | |
470 | 452314 | const int block_h= 1<<rem_depth; //FIXME "w!=h" | |
471 | BlockNode block; | ||
472 | int i,j; | ||
473 | |||
474 | 452314 | block.color[0]= l; | |
475 | 452314 | block.color[1]= cb; | |
476 | 452314 | block.color[2]= cr; | |
477 | 452314 | block.mx= mx; | |
478 | 452314 | block.my= my; | |
479 | 452314 | block.ref= ref; | |
480 | 452314 | block.type= type; | |
481 | 452314 | block.level= level; | |
482 | |||
483 |
2/2✓ Branch 0 taken 482672 times.
✓ Branch 1 taken 452314 times.
|
934986 | for(j=0; j<block_h; j++){ |
484 |
2/2✓ Branch 0 taken 543388 times.
✓ Branch 1 taken 482672 times.
|
1026060 | for(i=0; i<block_w; i++){ |
485 | 543388 | s->block[index + i + j*w]= block; | |
486 | } | ||
487 | } | ||
488 | 452314 | } | |
489 | |||
490 | 275940 | 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){ | |
491 | 275940 | SnowContext *s = c->avctx->priv_data; | |
492 | 275940 | const int offset[3]= { | |
493 | 275940 | y*c-> stride + x, | |
494 | 275940 | ((y*c->uvstride + x)>>s->chroma_h_shift), | |
495 | 275940 | ((y*c->uvstride + x)>>s->chroma_h_shift), | |
496 | }; | ||
497 | int i; | ||
498 |
2/2✓ Branch 0 taken 827820 times.
✓ Branch 1 taken 275940 times.
|
1103760 | for(i=0; i<3; i++){ |
499 | 827820 | c->src[0][i]= src [i]; | |
500 | 827820 | c->ref[0][i]= ref [i] + offset[i]; | |
501 | } | ||
502 | av_assert2(!ref_index); | ||
503 | 275940 | } | |
504 | |||
505 | |||
506 | /* bitstream functions */ | ||
507 | |||
508 | extern const int8_t ff_quant3bA[256]; | ||
509 | |||
510 | #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0 | ||
511 | |||
512 | 1400980 | static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){ | |
513 | int i; | ||
514 | |||
515 |
2/2✓ Branch 0 taken 1122246 times.
✓ Branch 1 taken 278734 times.
|
1400980 | if(v){ |
516 | 1122246 | const int a= FFABS(v); | |
517 | 1122246 | const int e= av_log2(a); | |
518 | 1122246 | const int el= FFMIN(e, 10); | |
519 | 1122246 | put_rac(c, state+0, 0); | |
520 | |||
521 |
2/2✓ Branch 0 taken 3458123 times.
✓ Branch 1 taken 1122246 times.
|
4580369 | for(i=0; i<el; i++){ |
522 | 3458123 | put_rac(c, state+1+i, 1); //1..10 | |
523 | } | ||
524 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1122246 times.
|
1122246 | for(; i<e; i++){ |
525 | ✗ | put_rac(c, state+1+9, 1); //1..10 | |
526 | } | ||
527 |
1/2✓ Branch 0 taken 1122246 times.
✗ Branch 1 not taken.
|
1122246 | put_rac(c, state+1+FFMIN(i,9), 0); |
528 | |||
529 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1122246 times.
|
1122246 | for(i=e-1; i>=el; i--){ |
530 | ✗ | put_rac(c, state+22+9, (a>>i)&1); //22..31 | |
531 | } | ||
532 |
2/2✓ Branch 0 taken 3458123 times.
✓ Branch 1 taken 1122246 times.
|
4580369 | for(; i>=0; i--){ |
533 | 3458123 | put_rac(c, state+22+i, (a>>i)&1); //22..31 | |
534 | } | ||
535 | |||
536 |
2/2✓ Branch 0 taken 1121841 times.
✓ Branch 1 taken 405 times.
|
1122246 | if(is_signed) |
537 | 1121841 | put_rac(c, state+11 + el, v < 0); //11..21 | |
538 | }else{ | ||
539 | 278734 | put_rac(c, state+0, 1); | |
540 | } | ||
541 | 1400980 | } | |
542 | |||
543 | 398667 | static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ | |
544 |
2/2✓ Branch 1 taken 181336 times.
✓ Branch 2 taken 217331 times.
|
398667 | if(get_rac(c, state+0)) |
545 | 181336 | return 0; | |
546 | else{ | ||
547 | int i, e; | ||
548 | unsigned a; | ||
549 | 217331 | e= 0; | |
550 |
3/4✓ Branch 0 taken 396432 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 179101 times.
✓ Branch 4 taken 217331 times.
|
396432 | while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10 |
551 | 179101 | e++; | |
552 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 179101 times.
|
179101 | if (e > 31) |
553 | ✗ | return AVERROR_INVALIDDATA; | |
554 | } | ||
555 | |||
556 | 217331 | a= 1; | |
557 |
2/2✓ Branch 0 taken 179101 times.
✓ Branch 1 taken 217331 times.
|
396432 | for(i=e-1; i>=0; i--){ |
558 |
1/2✓ Branch 0 taken 179101 times.
✗ Branch 1 not taken.
|
179101 | a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31 |
559 | } | ||
560 | |||
561 |
5/6✓ Branch 0 taken 216893 times.
✓ Branch 1 taken 438 times.
✓ Branch 2 taken 216893 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 84755 times.
✓ Branch 6 taken 132138 times.
|
217331 | e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21 |
562 | 217331 | return (a^e)-e; | |
563 | } | ||
564 | } | ||
565 | |||
566 | 16306397 | static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){ | |
567 | int i; | ||
568 |
2/2✓ Branch 0 taken 6740816 times.
✓ Branch 1 taken 9565581 times.
|
16306397 | int r= log2>=0 ? 1<<log2 : 1; |
569 | |||
570 | av_assert2(v>=0); | ||
571 | av_assert2(log2>=-4); | ||
572 | |||
573 |
2/2✓ Branch 0 taken 11892573 times.
✓ Branch 1 taken 16306397 times.
|
28198970 | while(v >= r){ |
574 | 11892573 | put_rac(c, state+4+log2, 1); | |
575 | 11892573 | v -= r; | |
576 | 11892573 | log2++; | |
577 |
2/2✓ Branch 0 taken 8609899 times.
✓ Branch 1 taken 3282674 times.
|
11892573 | if(log2>0) r+=r; |
578 | } | ||
579 | 16306397 | put_rac(c, state+4+log2, 0); | |
580 | |||
581 |
2/2✓ Branch 0 taken 16638975 times.
✓ Branch 1 taken 16306397 times.
|
32945372 | for(i=log2-1; i>=0; i--){ |
582 | 16638975 | put_rac(c, state+31-i, (v>>i)&1); | |
583 | } | ||
584 | 16306397 | } | |
585 | |||
586 | 16813073 | static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){ | |
587 | int i; | ||
588 |
2/2✓ Branch 0 taken 7065615 times.
✓ Branch 1 taken 9747458 times.
|
16813073 | int r= log2>=0 ? 1<<log2 : 1; |
589 | 16813073 | int v=0; | |
590 | |||
591 | av_assert2(log2>=-4); | ||
592 | |||
593 |
3/4✓ Branch 0 taken 29171987 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 12358914 times.
✓ Branch 4 taken 16813073 times.
|
29171987 | while(log2<28 && get_rac(c, state+4+log2)){ |
594 | 12358914 | v+= r; | |
595 | 12358914 | log2++; | |
596 |
2/2✓ Branch 0 taken 9002672 times.
✓ Branch 1 taken 3356242 times.
|
12358914 | if(log2>0) r+=r; |
597 | } | ||
598 | |||
599 |
2/2✓ Branch 0 taken 17511318 times.
✓ Branch 1 taken 16813073 times.
|
34324391 | for(i=log2-1; i>=0; i--){ |
600 | 17511318 | v+= get_rac(c, state+31-i)<<i; | |
601 | } | ||
602 | |||
603 | 16813073 | return v; | |
604 | } | ||
605 | |||
606 | 22128 | static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){ | |
607 | 22128 | const int w= b->width; | |
608 | 22128 | const int h= b->height; | |
609 | int x,y; | ||
610 | |||
611 | int run, runs; | ||
612 | 22128 | x_and_coeff *xc= b->x_coeff; | |
613 | 22128 | x_and_coeff *prev_xc= NULL; | |
614 | 22128 | x_and_coeff *prev2_xc= xc; | |
615 |
2/2✓ Branch 0 taken 16596 times.
✓ Branch 1 taken 5532 times.
|
22128 | x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL; |
616 | 22128 | x_and_coeff *prev_parent_xc= parent_xc; | |
617 | |||
618 | 22128 | runs= get_symbol2(&s->c, b->state[30], 0); | |
619 |
2/2✓ Branch 0 taken 17236 times.
✓ Branch 1 taken 4892 times.
|
22128 | if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
620 | 4892 | else run= INT_MAX; | |
621 | |||
622 |
2/2✓ Branch 0 taken 376000 times.
✓ Branch 1 taken 22128 times.
|
398128 | for(y=0; y<h; y++){ |
623 | 376000 | int v=0; | |
624 | 376000 | int lt=0, t=0, rt=0; | |
625 | |||
626 |
4/4✓ Branch 0 taken 353872 times.
✓ Branch 1 taken 22128 times.
✓ Branch 2 taken 214402 times.
✓ Branch 3 taken 139470 times.
|
376000 | if(y && prev_xc->x == 0){ |
627 | 214402 | rt= prev_xc->coeff; | |
628 | } | ||
629 |
2/2✓ Branch 0 taken 25205367 times.
✓ Branch 1 taken 376000 times.
|
25581367 | for(x=0; x<w; x++){ |
630 | 25205367 | int p=0; | |
631 | 25205367 | const int l= v; | |
632 | |||
633 | 25205367 | lt= t; t= rt; | |
634 | |||
635 |
2/2✓ Branch 0 taken 24753695 times.
✓ Branch 1 taken 451672 times.
|
25205367 | if(y){ |
636 |
2/2✓ Branch 0 taken 16207850 times.
✓ Branch 1 taken 8545845 times.
|
24753695 | if(prev_xc->x <= x) |
637 | 16207850 | prev_xc++; | |
638 |
2/2✓ Branch 0 taken 15993448 times.
✓ Branch 1 taken 8760247 times.
|
24753695 | if(prev_xc->x == x + 1) |
639 | 15993448 | rt= prev_xc->coeff; | |
640 | else | ||
641 | 8760247 | rt=0; | |
642 | } | ||
643 |
2/2✓ Branch 0 taken 25101348 times.
✓ Branch 1 taken 104019 times.
|
25205367 | if(parent_xc){ |
644 |
2/2✓ Branch 0 taken 8437610 times.
✓ Branch 1 taken 16663738 times.
|
25101348 | if(x>>1 > parent_xc->x){ |
645 | 8437610 | parent_xc++; | |
646 | } | ||
647 |
2/2✓ Branch 0 taken 17312556 times.
✓ Branch 1 taken 7788792 times.
|
25101348 | if(x>>1 == parent_xc->x){ |
648 | 17312556 | p= parent_xc->coeff; | |
649 | } | ||
650 | } | ||
651 |
2/2✓ Branch 0 taken 24345882 times.
✓ Branch 1 taken 859485 times.
|
25205367 | if(/*ll|*/l|lt|t|rt|p){ |
652 | 24345882 | int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); | |
653 | |||
654 | 24345882 | v=get_rac(&s->c, &b->state[0][context]); | |
655 |
2/2✓ Branch 0 taken 16203935 times.
✓ Branch 1 taken 8141947 times.
|
24345882 | if(v){ |
656 | 16203935 | v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); | |
657 | 16203935 | v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l&0xFF] + 3*ff_quant3bA[t&0xFF]]); | |
658 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 16203935 times.
|
16203935 | if ((uint16_t)v != v) { |
659 | ✗ | av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n"); | |
660 | ✗ | v = 1; | |
661 | } | ||
662 | 16203935 | xc->x=x; | |
663 | 16203935 | (xc++)->coeff= v; | |
664 | } | ||
665 | }else{ | ||
666 |
2/2✓ Branch 0 taken 293505 times.
✓ Branch 1 taken 565980 times.
|
859485 | if(!run){ |
667 |
2/2✓ Branch 0 taken 276269 times.
✓ Branch 1 taken 17236 times.
|
293505 | if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
668 | 17236 | else run= INT_MAX; | |
669 | 293505 | v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); | |
670 | 293505 | v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); | |
671 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 293505 times.
|
293505 | if ((uint16_t)v != v) { |
672 | ✗ | av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n"); | |
673 | ✗ | v = 1; | |
674 | } | ||
675 | |||
676 | 293505 | xc->x=x; | |
677 | 293505 | (xc++)->coeff= v; | |
678 | }else{ | ||
679 | int max_run; | ||
680 | 565980 | run--; | |
681 | 565980 | v=0; | |
682 | av_assert2(run >= 0); | ||
683 |
4/4✓ Branch 0 taken 534418 times.
✓ Branch 1 taken 31562 times.
✓ Branch 2 taken 314315 times.
✓ Branch 3 taken 220103 times.
|
565980 | if(y) max_run= FFMIN(run, prev_xc->x - x - 2); |
684 |
2/2✓ Branch 0 taken 8180 times.
✓ Branch 1 taken 23382 times.
|
31562 | else max_run= FFMIN(run, w-x-1); |
685 |
2/2✓ Branch 0 taken 562134 times.
✓ Branch 1 taken 3846 times.
|
565980 | if(parent_xc) |
686 |
2/2✓ Branch 0 taken 118384 times.
✓ Branch 1 taken 443750 times.
|
562134 | max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); |
687 | av_assert2(max_run >= 0 && max_run <= run); | ||
688 | |||
689 | 565980 | x+= max_run; | |
690 | 565980 | run-= max_run; | |
691 | } | ||
692 | } | ||
693 | } | ||
694 | 376000 | (xc++)->x= w+1; //end marker | |
695 | 376000 | prev_xc= prev2_xc; | |
696 | 376000 | prev2_xc= xc; | |
697 | |||
698 |
2/2✓ Branch 0 taken 360000 times.
✓ Branch 1 taken 16000 times.
|
376000 | if(parent_xc){ |
699 |
2/2✓ Branch 0 taken 179538 times.
✓ Branch 1 taken 180462 times.
|
360000 | if(y&1){ |
700 |
2/2✓ Branch 0 taken 109742 times.
✓ Branch 1 taken 179538 times.
|
289280 | while(parent_xc->x != parent->width+1) |
701 | 109742 | parent_xc++; | |
702 | 179538 | parent_xc++; | |
703 | 179538 | prev_parent_xc= parent_xc; | |
704 | }else{ | ||
705 | 180462 | parent_xc= prev_parent_xc; | |
706 | } | ||
707 | } | ||
708 | } | ||
709 | |||
710 | 22128 | (xc++)->x= w+1; //end marker | |
711 | 22128 | } | |
712 | |||
713 | #endif /* AVCODEC_SNOW_H */ | ||
714 |