FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavcodec/snow.h
Date: 2025-10-10 03:51:19
Exec Total Coverage
Lines: 150 157 95.5%
Functions: 6 6 100.0%
Branches: 96 100 96.0%
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 "avcodec.h"
28 #include "hpeldsp.h"
29 #include "snow_dwt.h"
30
31 #include "rangecoder.h"
32 #include "mathops.h"
33
34 #include "h264qpel.h"
35 #include "videodsp.h"
36
37 #define SNOW_MAX_PLANES 4
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 HpelDSPContext hdsp;
118 VideoDSPContext vdsp;
119 union {
120 /// everything except size 2 are from H.264
121 qpel_mc_func put_snow_qpel_pixels_tab[4][16];
122 H264QpelContext h264qpel;
123 };
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 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 mv_scale;
157 int last_mv_scale;
158 int qbias;
159 int last_qbias;
160 #define QBIAS_SHIFT 3
161 int b_width;
162 int b_height;
163 int block_max_depth;
164 int last_block_max_depth;
165 int nb_planes;
166 Plane plane[MAX_PLANES];
167 BlockNode *block;
168 slice_buffer sb;
169
170 uint8_t *scratchbuf;
171 uint8_t *emu_edge_buffer;
172
173 AVMotionVector *avmv;
174 unsigned avmv_size;
175 int avmv_index;
176 }SnowContext;
177
178 /* Tables */
179 extern const uint8_t * const ff_obmc_tab[4];
180 extern const uint8_t ff_qexp[QROOT];
181 extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES];
182
183 /* common code */
184
185 int ff_snow_common_init(AVCodecContext *avctx);
186 int ff_snow_common_init_after_header(AVCodecContext *avctx);
187 void ff_snow_common_end(SnowContext *s);
188 void ff_snow_reset_contexts(SnowContext *s);
189 int ff_snow_alloc_blocks(SnowContext *s);
190 int ff_snow_frames_prepare(SnowContext *s);
191 void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride,
192 int sx, int sy, int b_w, int b_h, const BlockNode *block,
193 int plane_index, int w, int h);
194 /* common inline functions */
195 //XXX doublecheck all of them should stay inlined
196
197 1912818 static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
198 const BlockNode *left, const BlockNode *top, const BlockNode *tr){
199
1/2
✓ Branch 0 taken 1912818 times.
✗ Branch 1 not taken.
 1912818 if(s->ref_frames == 1){
200 1912818 *mx = mid_pred(left->mx, top->mx, tr->mx);
201 1912818 *my = mid_pred(left->my, top->my, tr->my);
202 }else{
203 ✗ const int *scale = ff_scale_mv_ref[ref];
204 ✗ *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
205 ✗ (top ->mx * scale[top ->ref] + 128) >>8,
206 ✗ (tr ->mx * scale[tr ->ref] + 128) >>8);
207 ✗ *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
208 ✗ (top ->my * scale[top ->ref] + 128) >>8,
209 ✗ (tr ->my * scale[tr ->ref] + 128) >>8);
210 }
211 1912818 }
212
213 16265527 static av_always_inline int same_block(BlockNode *a, BlockNode *b){
214
4/4
✓ Branch 0 taken 1946124 times.
✓ Branch 1 taken 14319403 times.
✓ Branch 2 taken 1087765 times.
✓ Branch 3 taken 858359 times.
 16265527 if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
215 1087765 return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
216 }else{
217 15177762 return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
218 }
219 }
220
221 //FIXME name cleanup (b_w, block_w, b_width stuff)
222 //XXX should we really inline it?
223 3102231 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){
224 3102231 const int b_width = s->b_width << s->block_max_depth;
225 3102231 const int b_height= s->b_height << s->block_max_depth;
226 3102231 const int b_stride= b_width;
227 3102231 BlockNode *lt= &s->block[b_x + b_y*b_stride];
228 3102231 BlockNode *rt= lt+1;
229 3102231 BlockNode *lb= lt+b_stride;
230 3102231 BlockNode *rb= lb+1;
231 uint8_t *block[4];
232 // When src_stride is large enough, it is possible to interleave the blocks.
233 // Otherwise the blocks are written sequentially in the tmp buffer.
234
2/2
✓ Branch 0 taken 24300 times.
✓ Branch 1 taken 3077931 times.
 3102231 int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
235 3102231 uint8_t *tmp = s->scratchbuf;
236 uint8_t *ptmp;
237 int x,y;
238
239
2/2
✓ Branch 0 taken 80451 times.
✓ Branch 1 taken 3021780 times.
 3102231 if(b_x<0){
240 80451 lt= rt;
241 80451 lb= rb;
242
2/2
✓ Branch 0 taken 77583 times.
✓ Branch 1 taken 2944197 times.
 3021780 }else if(b_x + 1 >= b_width){
243 77583 rt= lt;
244 77583 rb= lb;
245 }
246
2/2
✓ Branch 0 taken 118527 times.
✓ Branch 1 taken 2983704 times.
 3102231 if(b_y<0){
247 118527 lt= lb;
248 118527 rt= rb;
249
2/2
✓ Branch 0 taken 109827 times.
✓ Branch 1 taken 2873877 times.
 2983704 }else if(b_y + 1 >= b_height){
250 109827 lb= lt;
251 109827 rb= rt;
252 }
253
254
2/2
✓ Branch 0 taken 80451 times.
✓ Branch 1 taken 3021780 times.
 3102231 if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
255 80451 obmc -= src_x;
256 80451 b_w += src_x;
257
4/4
✓ Branch 0 taken 61416 times.
✓ Branch 1 taken 19035 times.
✓ Branch 2 taken 13446 times.
✓ Branch 3 taken 47970 times.
 80451 if(!sliced && !offset_dst)
258 13446 dst -= src_x;
259 80451 src_x=0;
260 }
261
2/2
✓ Branch 0 taken 77583 times.
✓ Branch 1 taken 3024648 times.
 3102231 if(src_x + b_w > w){
262 77583 b_w = w - src_x;
263 }
264
2/2
✓ Branch 0 taken 118527 times.
✓ Branch 1 taken 2983704 times.
 3102231 if(src_y<0){
265 118527 obmc -= src_y*obmc_stride;
266 118527 b_h += src_y;
267
4/4
✓ Branch 0 taken 93012 times.
✓ Branch 1 taken 25515 times.
✓ Branch 2 taken 29652 times.
✓ Branch 3 taken 63360 times.
 118527 if(!sliced && !offset_dst)
268 29652 dst -= src_y*dst_stride;
269 118527 src_y=0;
270 }
271
2/2
✓ Branch 0 taken 109827 times.
✓ Branch 1 taken 2992404 times.
 3102231 if(src_y + b_h> h){
272 109827 b_h = h - src_y;
273 }
274
275
2/4
✓ Branch 0 taken 3102231 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 3102231 times.
 3102231 if(b_w<=0 || b_h<=0) return;
276
277
4/4
✓ Branch 0 taken 2391456 times.
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 2172600 times.
✓ Branch 3 taken 218856 times.
 3102231 if(!sliced && offset_dst)
278 2172600 dst += src_x + src_y*dst_stride;
279
4/4
✓ Branch 0 taken 2391456 times.
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 1461825 times.
✓ Branch 3 taken 929631 times.
 3102231 if (sliced || add)
280 2172600 dst8+= src_x + src_y*src_stride;
281 // src += src_x + src_y*src_stride;
282
283 3102231 ptmp= tmp + 3*tmp_step;
284 3102231 block[0]= ptmp;
285 3102231 ptmp+=tmp_step;
286 3102231 ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
287
288
2/2
✓ Branch 1 taken 846966 times.
✓ Branch 2 taken 2255265 times.
 3102231 if(same_block(lt, rt)){
289 846966 block[1]= block[0];
290 }else{
291 2255265 block[1]= ptmp;
292 2255265 ptmp+=tmp_step;
293 2255265 ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
294 }
295
296
2/2
✓ Branch 1 taken 873945 times.
✓ Branch 2 taken 2228286 times.
 3102231 if(same_block(lt, lb)){
297 873945 block[2]= block[0];
298
2/2
✓ Branch 1 taken 208740 times.
✓ Branch 2 taken 2019546 times.
 2228286 }else if(same_block(rt, lb)){
299 208740 block[2]= block[1];
300 }else{
301 2019546 block[2]= ptmp;
302 2019546 ptmp+=tmp_step;
303 2019546 ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
304 }
305
306
2/2
✓ Branch 1 taken 482331 times.
✓ Branch 2 taken 2619900 times.
 3102231 if(same_block(lt, rb) ){
307 482331 block[3]= block[0];
308
2/2
✓ Branch 1 taken 527490 times.
✓ Branch 2 taken 2092410 times.
 2619900 }else if(same_block(rt, rb)){
309 527490 block[3]= block[1];
310
2/2
✓ Branch 1 taken 397176 times.
✓ Branch 2 taken 1695234 times.
 2092410 }else if(same_block(lb, rb)){
311 397176 block[3]= block[2];
312 }else{
313 1695234 block[3]= ptmp;
314 1695234 ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
315 }
316
2/2
✓ Branch 0 taken 710775 times.
✓ Branch 1 taken 2391456 times.
 3102231 if(sliced){
317 710775 s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
318 }else{
319
2/2
✓ Branch 0 taken 14419776 times.
✓ Branch 1 taken 2391456 times.
 16811232 for(y=0; y<b_h; y++){
320 //FIXME ugly misuse of obmc_stride
321 14419776 const uint8_t *obmc1= obmc + y*obmc_stride;
322 14419776 const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
323 14419776 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
324 14419776 const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
325
2/2
✓ Branch 0 taken 108428832 times.
✓ Branch 1 taken 14419776 times.
 122848608 for(x=0; x<b_w; x++){
326 108428832 int v= obmc1[x] * block[3][x + y*src_stride]
327 108428832 +obmc2[x] * block[2][x + y*src_stride]
328 108428832 +obmc3[x] * block[1][x + y*src_stride]
329 108428832 +obmc4[x] * block[0][x + y*src_stride];
330
331 108428832 v <<= 8 - LOG2_OBMC_MAX;
332 if(FRAC_BITS != 8){
333 108428832 v >>= 8 - FRAC_BITS;
334 }
335
2/2
✓ Branch 0 taken 61171200 times.
✓ Branch 1 taken 47257632 times.
 108428832 if(add){
336 61171200 v += dst[x + y*dst_stride];
337 61171200 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
338
2/2
✓ Branch 0 taken 1126 times.
✓ Branch 1 taken 61170074 times.
 61171200 if(v&(~255)) v= ~(v>>31);
339 61171200 dst8[x + y*src_stride] = v;
340 }else{
341 47257632 dst[x + y*dst_stride] -= v;
342 }
343 }
344 }
345 }
346 }
347
348 52200 static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
349 52200 Plane *p= &s->plane[plane_index];
350 52200 const int mb_w= s->b_width << s->block_max_depth;
351 52200 const int mb_h= s->b_height << s->block_max_depth;
352 int x, y, mb_x;
353 52200 int block_size = MB_SIZE >> s->block_max_depth;
354
2/2
✓ Branch 0 taken 34800 times.
✓ Branch 1 taken 17400 times.
 52200 int block_w = plane_index ? block_size>>s->chroma_h_shift : block_size;
355
2/2
✓ Branch 0 taken 34800 times.
✓ Branch 1 taken 17400 times.
 52200 int block_h = plane_index ? block_size>>s->chroma_v_shift : block_size;
356
2/2
✓ Branch 0 taken 34800 times.
✓ Branch 1 taken 17400 times.
 52200 const uint8_t *obmc = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
357
2/2
✓ Branch 0 taken 34800 times.
✓ Branch 1 taken 17400 times.
 52200 const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
358 52200 int ref_stride= s->current_picture->linesize[plane_index];
359 52200 uint8_t *dst8= s->current_picture->data[plane_index];
360 52200 int w= p->width;
361 52200 int h= p->height;
362 av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares
363
3/4
✓ Branch 0 taken 47970 times.
✓ Branch 1 taken 4230 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 47970 times.
 52200 if(s->keyframe || (s->avctx->debug&512)){
364
2/2
✓ Branch 0 taken 270 times.
✓ Branch 1 taken 3960 times.
 4230 if(mb_y==mb_h)
365 270 return;
366
367
2/2
✓ Branch 0 taken 1980 times.
✓ Branch 1 taken 1980 times.
 3960 if(add){
368
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++){
369
2/2
✓ Branch 0 taken 2649600 times.
✓ Branch 1 taken 12480 times.
 2662080 for(x=0; x<w; x++){
370 2649600 int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
371 2649600 v >>= FRAC_BITS;
372
2/2
✓ Branch 0 taken 104 times.
✓ Branch 1 taken 2649496 times.
 2649600 if(v&(~255)) v= ~(v>>31);
373 2649600 dst8[x + y*ref_stride]= v;
374 }
375 }
376 }else{
377
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++){
378
2/2
✓ Branch 0 taken 2649600 times.
✓ Branch 1 taken 12480 times.
 2662080 for(x=0; x<w; x++){
379 2649600 buf[x + y*w]-= 128<<FRAC_BITS;
380 }
381 }
382 }
383
384 3960 return;
385 }
386
387
2/2
✓ Branch 0 taken 2172600 times.
✓ Branch 1 taken 47970 times.
 2220570 for(mb_x=0; mb_x<=mb_w; mb_x++){
388 2172600 add_yblock(s, 0, NULL, buf, dst8, obmc,
389 2172600 block_w*mb_x - block_w/2,
390 2172600 block_h*mb_y - block_h/2,
391 block_w, block_h,
392 w, h,
393 w, ref_stride, obmc_stride,
394 mb_x - 1, mb_y - 1,
395 add, 1, plane_index);
396 }
397 }
398
399 2880 static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
400 2880 const int mb_h= s->b_height << s->block_max_depth;
401 int mb_y;
402
2/2
✓ Branch 0 taken 52200 times.
✓ Branch 1 taken 2880 times.
 55080 for(mb_y=0; mb_y<=mb_h; mb_y++)
403 52200 predict_slice(s, buf, plane_index, add, mb_y);
404 2880 }
405
406 702006 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){
407 702006 const int w= s->b_width << s->block_max_depth;
408 702006 const int rem_depth= s->block_max_depth - level;
409 702006 const int index= (x + y*w) << rem_depth;
410 702006 const int block_w= 1<<rem_depth;
411 702006 const int block_h= 1<<rem_depth; //FIXME "w!=h"
412 BlockNode block;
413 int i,j;
414
415 702006 block.color[0]= l;
416 702006 block.color[1]= cb;
417 702006 block.color[2]= cr;
418 702006 block.mx= mx;
419 702006 block.my= my;
420 702006 block.ref= ref;
421 702006 block.type= type;
422 702006 block.level= level;
423
424
2/2
✓ Branch 0 taken 737436 times.
✓ Branch 1 taken 702006 times.
 1439442 for(j=0; j<block_h; j++){
425
2/2
✓ Branch 0 taken 808296 times.
✓ Branch 1 taken 737436 times.
 1545732 for(i=0; i<block_w; i++){
426 808296 s->block[index + i + j*w]= block;
427 }
428 }
429 702006 }
430
431 extern const int8_t ff_quant3bA[256];
432
433 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
434
435 #endif /* AVCODEC_SNOW_H */
436