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 |
|
|
int avmv_index; |
190 |
|
|
uint64_t encoding_error[AV_NUM_DATA_POINTERS]; |
191 |
|
|
|
192 |
|
|
int pred; |
193 |
|
|
}SnowContext; |
194 |
|
|
|
195 |
|
|
/* Tables */ |
196 |
|
|
extern const uint8_t * const ff_obmc_tab[4]; |
197 |
|
|
extern uint8_t ff_qexp[QROOT]; |
198 |
|
|
extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES]; |
199 |
|
|
|
200 |
|
|
/* C bits used by mmx/sse2/altivec */ |
201 |
|
|
|
202 |
|
248 |
static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){ |
203 |
|
248 |
(*i) = (width) - 2; |
204 |
|
|
|
205 |
✗✓ |
248 |
if (width & 1){ |
206 |
|
|
low[(*i)+1] = low[((*i)+1)>>1]; |
207 |
|
|
(*i)--; |
208 |
|
|
} |
209 |
|
248 |
} |
210 |
|
|
|
211 |
|
|
static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){ |
212 |
|
|
for (; (*i)>=0; (*i)-=2){ |
213 |
|
|
low[(*i)+1] = high[(*i)>>1]; |
214 |
|
|
low[*i] = low[(*i)>>1]; |
215 |
|
|
} |
216 |
|
|
} |
217 |
|
|
|
218 |
|
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){ |
219 |
✓✓ |
5736 |
for(; i<w; i++){ |
220 |
|
4992 |
dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift); |
221 |
|
|
} |
222 |
|
|
|
223 |
✓✓ |
744 |
if((width^lift_high)&1){ |
224 |
|
496 |
dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift); |
225 |
|
|
} |
226 |
|
744 |
} |
227 |
|
|
|
228 |
|
248 |
static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){ |
229 |
✓✓ |
1912 |
for(; i<w; i++){ |
230 |
|
1664 |
dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS); |
231 |
|
|
} |
232 |
|
|
|
233 |
✗✓ |
248 |
if(width&1){ |
234 |
|
|
dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS); |
235 |
|
|
} |
236 |
|
248 |
} |
237 |
|
|
|
238 |
|
|
/* common code */ |
239 |
|
|
|
240 |
|
|
int ff_snow_common_init(AVCodecContext *avctx); |
241 |
|
|
int ff_snow_common_init_after_header(AVCodecContext *avctx); |
242 |
|
|
void ff_snow_common_end(SnowContext *s); |
243 |
|
|
void ff_snow_release_buffer(AVCodecContext *avctx); |
244 |
|
|
void ff_snow_reset_contexts(SnowContext *s); |
245 |
|
|
int ff_snow_alloc_blocks(SnowContext *s); |
246 |
|
|
int ff_snow_frame_start(SnowContext *s); |
247 |
|
|
void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride, |
248 |
|
|
int sx, int sy, int b_w, int b_h, const BlockNode *block, |
249 |
|
|
int plane_index, int w, int h); |
250 |
|
|
int ff_snow_get_buffer(SnowContext *s, AVFrame *frame); |
251 |
|
|
/* common inline functions */ |
252 |
|
|
//XXX doublecheck all of them should stay inlined |
253 |
|
|
|
254 |
|
1929303 |
static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref, |
255 |
|
|
const BlockNode *left, const BlockNode *top, const BlockNode *tr){ |
256 |
✓✗ |
1929303 |
if(s->ref_frames == 1){ |
257 |
|
1929303 |
*mx = mid_pred(left->mx, top->mx, tr->mx); |
258 |
|
1929303 |
*my = mid_pred(left->my, top->my, tr->my); |
259 |
|
|
}else{ |
260 |
|
|
const int *scale = ff_scale_mv_ref[ref]; |
261 |
|
|
*mx = mid_pred((left->mx * scale[left->ref] + 128) >>8, |
262 |
|
|
(top ->mx * scale[top ->ref] + 128) >>8, |
263 |
|
|
(tr ->mx * scale[tr ->ref] + 128) >>8); |
264 |
|
|
*my = mid_pred((left->my * scale[left->ref] + 128) >>8, |
265 |
|
|
(top ->my * scale[top ->ref] + 128) >>8, |
266 |
|
|
(tr ->my * scale[tr ->ref] + 128) >>8); |
267 |
|
|
} |
268 |
|
1929303 |
} |
269 |
|
|
|
270 |
|
16154477 |
static av_always_inline int same_block(BlockNode *a, BlockNode *b){ |
271 |
✓✓✓✓
|
16154477 |
if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){ |
272 |
|
1029639 |
return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2])); |
273 |
|
|
}else{ |
274 |
|
15124838 |
return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA)); |
275 |
|
|
} |
276 |
|
|
} |
277 |
|
|
|
278 |
|
|
//FIXME name cleanup (b_w, block_w, b_width stuff) |
279 |
|
|
//XXX should we really inline it? |
280 |
|
3083760 |
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){ |
281 |
|
3083760 |
const int b_width = s->b_width << s->block_max_depth; |
282 |
|
3083760 |
const int b_height= s->b_height << s->block_max_depth; |
283 |
|
3083760 |
const int b_stride= b_width; |
284 |
|
3083760 |
BlockNode *lt= &s->block[b_x + b_y*b_stride]; |
285 |
|
3083760 |
BlockNode *rt= lt+1; |
286 |
|
3083760 |
BlockNode *lb= lt+b_stride; |
287 |
|
3083760 |
BlockNode *rb= lb+1; |
288 |
|
|
uint8_t *block[4]; |
289 |
|
|
// When src_stride is large enough, it is possible to interleave the blocks. |
290 |
|
|
// Otherwise the blocks are written sequentially in the tmp buffer. |
291 |
✓✓ |
3083760 |
int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; |
292 |
|
3083760 |
uint8_t *tmp = s->scratchbuf; |
293 |
|
|
uint8_t *ptmp; |
294 |
|
|
int x,y; |
295 |
|
|
|
296 |
✓✓ |
3083760 |
if(b_x<0){ |
297 |
|
80121 |
lt= rt; |
298 |
|
80121 |
lb= rb; |
299 |
✓✓ |
3003639 |
}else if(b_x + 1 >= b_width){ |
300 |
|
77757 |
rt= lt; |
301 |
|
77757 |
rb= lb; |
302 |
|
|
} |
303 |
✓✓ |
3083760 |
if(b_y<0){ |
304 |
|
118404 |
lt= lb; |
305 |
|
118404 |
rt= rb; |
306 |
✓✓ |
2965356 |
}else if(b_y + 1 >= b_height){ |
307 |
|
110604 |
lb= lt; |
308 |
|
110604 |
rb= rt; |
309 |
|
|
} |
310 |
|
|
|
311 |
✓✓ |
3083760 |
if(src_x<0){ //FIXME merge with prev & always round internal width up to *16 |
312 |
|
80121 |
obmc -= src_x; |
313 |
|
80121 |
b_w += src_x; |
314 |
✓✓✓✓
|
80121 |
if(!sliced && !offset_dst) |
315 |
|
13446 |
dst -= src_x; |
316 |
|
80121 |
src_x=0; |
317 |
|
|
} |
318 |
✓✓ |
3083760 |
if(src_x + b_w > w){ |
319 |
|
77757 |
b_w = w - src_x; |
320 |
|
|
} |
321 |
✓✓ |
3083760 |
if(src_y<0){ |
322 |
|
118404 |
obmc -= src_y*obmc_stride; |
323 |
|
118404 |
b_h += src_y; |
324 |
✓✓✓✓
|
118404 |
if(!sliced && !offset_dst) |
325 |
|
29940 |
dst -= src_y*dst_stride; |
326 |
|
118404 |
src_y=0; |
327 |
|
|
} |
328 |
✓✓ |
3083760 |
if(src_y + b_h> h){ |
329 |
|
110604 |
b_h = h - src_y; |
330 |
|
|
} |
331 |
|
|
|
332 |
✓✗✗✓
|
3083760 |
if(b_w<=0 || b_h<=0) return; |
333 |
|
|
|
334 |
✓✓✓✓
|
3083760 |
if(!sliced && offset_dst) |
335 |
|
2147565 |
dst += src_x + src_y*dst_stride; |
336 |
|
3083760 |
dst8+= src_x + src_y*src_stride; |
337 |
|
|
// src += src_x + src_y*src_stride; |
338 |
|
|
|
339 |
|
3083760 |
ptmp= tmp + 3*tmp_step; |
340 |
|
3083760 |
block[0]= ptmp; |
341 |
|
3083760 |
ptmp+=tmp_step; |
342 |
|
3083760 |
ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); |
343 |
|
|
|
344 |
✓✓ |
3083760 |
if(same_block(lt, rt)){ |
345 |
|
846306 |
block[1]= block[0]; |
346 |
|
|
}else{ |
347 |
|
2237454 |
block[1]= ptmp; |
348 |
|
2237454 |
ptmp+=tmp_step; |
349 |
|
2237454 |
ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); |
350 |
|
|
} |
351 |
|
|
|
352 |
✓✓ |
3083760 |
if(same_block(lt, lb)){ |
353 |
|
874395 |
block[2]= block[0]; |
354 |
✓✓ |
2209365 |
}else if(same_block(rt, lb)){ |
355 |
|
208647 |
block[2]= block[1]; |
356 |
|
|
}else{ |
357 |
|
2000718 |
block[2]= ptmp; |
358 |
|
2000718 |
ptmp+=tmp_step; |
359 |
|
2000718 |
ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); |
360 |
|
|
} |
361 |
|
|
|
362 |
✓✓ |
3083760 |
if(same_block(lt, rb) ){ |
363 |
|
481845 |
block[3]= block[0]; |
364 |
✓✓ |
2601915 |
}else if(same_block(rt, rb)){ |
365 |
|
528783 |
block[3]= block[1]; |
366 |
✓✓ |
2073132 |
}else if(same_block(lb, rb)){ |
367 |
|
396921 |
block[3]= block[2]; |
368 |
|
|
}else{ |
369 |
|
1676211 |
block[3]= ptmp; |
370 |
|
1676211 |
ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); |
371 |
|
|
} |
372 |
✓✓ |
3083760 |
if(sliced){ |
373 |
|
710775 |
s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); |
374 |
|
|
}else{ |
375 |
✓✓ |
16697081 |
for(y=0; y<b_h; y++){ |
376 |
|
|
//FIXME ugly misuse of obmc_stride |
377 |
|
14324096 |
const uint8_t *obmc1= obmc + y*obmc_stride; |
378 |
|
14324096 |
const uint8_t *obmc2= obmc1+ (obmc_stride>>1); |
379 |
|
14324096 |
const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); |
380 |
|
14324096 |
const uint8_t *obmc4= obmc3+ (obmc_stride>>1); |
381 |
✓✓ |
122225888 |
for(x=0; x<b_w; x++){ |
382 |
|
107901792 |
int v= obmc1[x] * block[3][x + y*src_stride] |
383 |
|
107901792 |
+obmc2[x] * block[2][x + y*src_stride] |
384 |
|
107901792 |
+obmc3[x] * block[1][x + y*src_stride] |
385 |
|
107901792 |
+obmc4[x] * block[0][x + y*src_stride]; |
386 |
|
|
|
387 |
|
107901792 |
v <<= 8 - LOG2_OBMC_MAX; |
388 |
|
|
if(FRAC_BITS != 8){ |
389 |
|
107901792 |
v >>= 8 - FRAC_BITS; |
390 |
|
|
} |
391 |
✓✓ |
107901792 |
if(add){ |
392 |
|
59927040 |
v += dst[x + y*dst_stride]; |
393 |
|
59927040 |
v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
394 |
✓✓ |
59927040 |
if(v&(~255)) v= ~(v>>31); |
395 |
|
59927040 |
dst8[x + y*src_stride] = v; |
396 |
|
|
}else{ |
397 |
|
47974752 |
dst[x + y*dst_stride] -= v; |
398 |
|
|
} |
399 |
|
|
} |
400 |
|
|
} |
401 |
|
|
} |
402 |
|
|
} |
403 |
|
|
|
404 |
|
51870 |
static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){ |
405 |
|
51870 |
Plane *p= &s->plane[plane_index]; |
406 |
|
51870 |
const int mb_w= s->b_width << s->block_max_depth; |
407 |
|
51870 |
const int mb_h= s->b_height << s->block_max_depth; |
408 |
|
|
int x, y, mb_x; |
409 |
|
51870 |
int block_size = MB_SIZE >> s->block_max_depth; |
410 |
✓✓ |
51870 |
int block_w = plane_index ? block_size>>s->chroma_h_shift : block_size; |
411 |
✓✓ |
51870 |
int block_h = plane_index ? block_size>>s->chroma_v_shift : block_size; |
412 |
✓✓ |
51870 |
const uint8_t *obmc = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth]; |
413 |
✓✓ |
51870 |
const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size; |
414 |
|
51870 |
int ref_stride= s->current_picture->linesize[plane_index]; |
415 |
|
51870 |
uint8_t *dst8= s->current_picture->data[plane_index]; |
416 |
|
51870 |
int w= p->width; |
417 |
|
51870 |
int h= p->height; |
418 |
|
|
av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares |
419 |
✓✓✗✓
|
51870 |
if(s->keyframe || (s->avctx->debug&512)){ |
420 |
✓✓ |
4230 |
if(mb_y==mb_h) |
421 |
|
270 |
return; |
422 |
|
|
|
423 |
✓✓ |
3960 |
if(add){ |
424 |
✓✓ |
14460 |
for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){ |
425 |
✓✓ |
2662080 |
for(x=0; x<w; x++){ |
426 |
|
2649600 |
int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
427 |
|
2649600 |
v >>= FRAC_BITS; |
428 |
✓✓ |
2649600 |
if(v&(~255)) v= ~(v>>31); |
429 |
|
2649600 |
dst8[x + y*ref_stride]= v; |
430 |
|
|
} |
431 |
|
|
} |
432 |
|
|
}else{ |
433 |
✓✓ |
14460 |
for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){ |
434 |
✓✓ |
2662080 |
for(x=0; x<w; x++){ |
435 |
|
2649600 |
buf[x + y*w]-= 128<<FRAC_BITS; |
436 |
|
|
} |
437 |
|
|
} |
438 |
|
|
} |
439 |
|
|
|
440 |
|
3960 |
return; |
441 |
|
|
} |
442 |
|
|
|
443 |
✓✓ |
2195205 |
for(mb_x=0; mb_x<=mb_w; mb_x++){ |
444 |
|
2147565 |
add_yblock(s, 0, NULL, buf, dst8, obmc, |
445 |
|
2147565 |
block_w*mb_x - block_w/2, |
446 |
|
2147565 |
block_h*mb_y - block_h/2, |
447 |
|
|
block_w, block_h, |
448 |
|
|
w, h, |
449 |
|
|
w, ref_stride, obmc_stride, |
450 |
|
|
mb_x - 1, mb_y - 1, |
451 |
|
|
add, 1, plane_index); |
452 |
|
|
} |
453 |
|
|
} |
454 |
|
|
|
455 |
|
2874 |
static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){ |
456 |
|
2874 |
const int mb_h= s->b_height << s->block_max_depth; |
457 |
|
|
int mb_y; |
458 |
✓✓ |
54744 |
for(mb_y=0; mb_y<=mb_h; mb_y++) |
459 |
|
51870 |
predict_slice(s, buf, plane_index, add, mb_y); |
460 |
|
2874 |
} |
461 |
|
|
|
462 |
|
693867 |
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){ |
463 |
|
693867 |
const int w= s->b_width << s->block_max_depth; |
464 |
|
693867 |
const int rem_depth= s->block_max_depth - level; |
465 |
|
693867 |
const int index= (x + y*w) << rem_depth; |
466 |
|
693867 |
const int block_w= 1<<rem_depth; |
467 |
|
693867 |
const int block_h= 1<<rem_depth; //FIXME "w!=h" |
468 |
|
|
BlockNode block; |
469 |
|
|
int i,j; |
470 |
|
|
|
471 |
|
693867 |
block.color[0]= l; |
472 |
|
693867 |
block.color[1]= cb; |
473 |
|
693867 |
block.color[2]= cr; |
474 |
|
693867 |
block.mx= mx; |
475 |
|
693867 |
block.my= my; |
476 |
|
693867 |
block.ref= ref; |
477 |
|
693867 |
block.type= type; |
478 |
|
693867 |
block.level= level; |
479 |
|
|
|
480 |
✓✓ |
1423125 |
for(j=0; j<block_h; j++){ |
481 |
✓✓ |
1529298 |
for(i=0; i<block_w; i++){ |
482 |
|
800040 |
s->block[index + i + j*w]= block; |
483 |
|
|
} |
484 |
|
|
} |
485 |
|
693867 |
} |
486 |
|
|
|
487 |
|
557820 |
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){ |
488 |
|
557820 |
SnowContext *s = c->avctx->priv_data; |
489 |
|
557820 |
const int offset[3]= { |
490 |
|
557820 |
y*c-> stride + x, |
491 |
|
557820 |
((y*c->uvstride + x)>>s->chroma_h_shift), |
492 |
|
557820 |
((y*c->uvstride + x)>>s->chroma_h_shift), |
493 |
|
|
}; |
494 |
|
|
int i; |
495 |
✓✓ |
2231280 |
for(i=0; i<3; i++){ |
496 |
|
1673460 |
c->src[0][i]= src [i]; |
497 |
|
1673460 |
c->ref[0][i]= ref [i] + offset[i]; |
498 |
|
|
} |
499 |
|
|
av_assert2(!ref_index); |
500 |
|
557820 |
} |
501 |
|
|
|
502 |
|
|
|
503 |
|
|
/* bitstream functions */ |
504 |
|
|
|
505 |
|
|
extern const int8_t ff_quant3bA[256]; |
506 |
|
|
|
507 |
|
|
#define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0 |
508 |
|
|
|
509 |
|
2810756 |
static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){ |
510 |
|
|
int i; |
511 |
|
|
|
512 |
✓✓ |
2810756 |
if(v){ |
513 |
|
2212871 |
const int a= FFABS(v); |
514 |
|
2212871 |
const int e= av_log2(a); |
515 |
|
2212871 |
const int el= FFMIN(e, 10); |
516 |
|
2212871 |
put_rac(c, state+0, 0); |
517 |
|
|
|
518 |
✓✓ |
7535665 |
for(i=0; i<el; i++){ |
519 |
|
5322794 |
put_rac(c, state+1+i, 1); //1..10 |
520 |
|
|
} |
521 |
✗✓ |
2212871 |
for(; i<e; i++){ |
522 |
|
|
put_rac(c, state+1+9, 1); //1..10 |
523 |
|
|
} |
524 |
✓✗ |
2212871 |
put_rac(c, state+1+FFMIN(i,9), 0); |
525 |
|
|
|
526 |
✗✓ |
2212871 |
for(i=e-1; i>=el; i--){ |
527 |
|
|
put_rac(c, state+22+9, (a>>i)&1); //22..31 |
528 |
|
|
} |
529 |
✓✓ |
7535665 |
for(; i>=0; i--){ |
530 |
|
5322794 |
put_rac(c, state+22+i, (a>>i)&1); //22..31 |
531 |
|
|
} |
532 |
|
|
|
533 |
✓✓ |
2212871 |
if(is_signed) |
534 |
|
2212448 |
put_rac(c, state+11 + el, v < 0); //11..21 |
535 |
|
|
}else{ |
536 |
|
597885 |
put_rac(c, state+0, 1); |
537 |
|
|
} |
538 |
|
2810756 |
} |
539 |
|
|
|
540 |
|
398667 |
static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ |
541 |
✓✓ |
398667 |
if(get_rac(c, state+0)) |
542 |
|
181336 |
return 0; |
543 |
|
|
else{ |
544 |
|
|
int i, e; |
545 |
|
|
unsigned a; |
546 |
|
217331 |
e= 0; |
547 |
✓✗✓✓
|
396432 |
while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10 |
548 |
|
179101 |
e++; |
549 |
✗✓ |
179101 |
if (e > 31) |
550 |
|
|
return AVERROR_INVALIDDATA; |
551 |
|
|
} |
552 |
|
|
|
553 |
|
217331 |
a= 1; |
554 |
✓✓ |
396432 |
for(i=e-1; i>=0; i--){ |
555 |
✓✗ |
179101 |
a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31 |
556 |
|
|
} |
557 |
|
|
|
558 |
✓✓✓✗ ✓✓ |
217331 |
e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21 |
559 |
|
217331 |
return (a^e)-e; |
560 |
|
|
} |
561 |
|
|
} |
562 |
|
|
|
563 |
|
16306397 |
static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){ |
564 |
|
|
int i; |
565 |
✓✓ |
16306397 |
int r= log2>=0 ? 1<<log2 : 1; |
566 |
|
|
|
567 |
|
|
av_assert2(v>=0); |
568 |
|
|
av_assert2(log2>=-4); |
569 |
|
|
|
570 |
✓✓ |
28198970 |
while(v >= r){ |
571 |
|
11892573 |
put_rac(c, state+4+log2, 1); |
572 |
|
11892573 |
v -= r; |
573 |
|
11892573 |
log2++; |
574 |
✓✓ |
11892573 |
if(log2>0) r+=r; |
575 |
|
|
} |
576 |
|
16306397 |
put_rac(c, state+4+log2, 0); |
577 |
|
|
|
578 |
✓✓ |
32945372 |
for(i=log2-1; i>=0; i--){ |
579 |
|
16638975 |
put_rac(c, state+31-i, (v>>i)&1); |
580 |
|
|
} |
581 |
|
16306397 |
} |
582 |
|
|
|
583 |
|
16813073 |
static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){ |
584 |
|
|
int i; |
585 |
✓✓ |
16813073 |
int r= log2>=0 ? 1<<log2 : 1; |
586 |
|
16813073 |
int v=0; |
587 |
|
|
|
588 |
|
|
av_assert2(log2>=-4); |
589 |
|
|
|
590 |
✓✗✓✓
|
29171987 |
while(log2<28 && get_rac(c, state+4+log2)){ |
591 |
|
12358914 |
v+= r; |
592 |
|
12358914 |
log2++; |
593 |
✓✓ |
12358914 |
if(log2>0) r+=r; |
594 |
|
|
} |
595 |
|
|
|
596 |
✓✓ |
34324391 |
for(i=log2-1; i>=0; i--){ |
597 |
|
17511318 |
v+= get_rac(c, state+31-i)<<i; |
598 |
|
|
} |
599 |
|
|
|
600 |
|
16813073 |
return v; |
601 |
|
|
} |
602 |
|
|
|
603 |
|
22128 |
static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){ |
604 |
|
22128 |
const int w= b->width; |
605 |
|
22128 |
const int h= b->height; |
606 |
|
|
int x,y; |
607 |
|
|
|
608 |
|
|
int run, runs; |
609 |
|
22128 |
x_and_coeff *xc= b->x_coeff; |
610 |
|
22128 |
x_and_coeff *prev_xc= NULL; |
611 |
|
22128 |
x_and_coeff *prev2_xc= xc; |
612 |
✓✓ |
22128 |
x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL; |
613 |
|
22128 |
x_and_coeff *prev_parent_xc= parent_xc; |
614 |
|
|
|
615 |
|
22128 |
runs= get_symbol2(&s->c, b->state[30], 0); |
616 |
✓✓ |
22128 |
if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
617 |
|
4892 |
else run= INT_MAX; |
618 |
|
|
|
619 |
✓✓ |
398128 |
for(y=0; y<h; y++){ |
620 |
|
376000 |
int v=0; |
621 |
|
376000 |
int lt=0, t=0, rt=0; |
622 |
|
|
|
623 |
✓✓✓✓
|
376000 |
if(y && prev_xc->x == 0){ |
624 |
|
214402 |
rt= prev_xc->coeff; |
625 |
|
|
} |
626 |
✓✓ |
25581367 |
for(x=0; x<w; x++){ |
627 |
|
25205367 |
int p=0; |
628 |
|
25205367 |
const int l= v; |
629 |
|
|
|
630 |
|
25205367 |
lt= t; t= rt; |
631 |
|
|
|
632 |
✓✓ |
25205367 |
if(y){ |
633 |
✓✓ |
24753695 |
if(prev_xc->x <= x) |
634 |
|
16207850 |
prev_xc++; |
635 |
✓✓ |
24753695 |
if(prev_xc->x == x + 1) |
636 |
|
15993448 |
rt= prev_xc->coeff; |
637 |
|
|
else |
638 |
|
8760247 |
rt=0; |
639 |
|
|
} |
640 |
✓✓ |
25205367 |
if(parent_xc){ |
641 |
✓✓ |
25101348 |
if(x>>1 > parent_xc->x){ |
642 |
|
8437610 |
parent_xc++; |
643 |
|
|
} |
644 |
✓✓ |
25101348 |
if(x>>1 == parent_xc->x){ |
645 |
|
17312556 |
p= parent_xc->coeff; |
646 |
|
|
} |
647 |
|
|
} |
648 |
✓✓ |
25205367 |
if(/*ll|*/l|lt|t|rt|p){ |
649 |
|
24345882 |
int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); |
650 |
|
|
|
651 |
|
24345882 |
v=get_rac(&s->c, &b->state[0][context]); |
652 |
✓✓ |
24345882 |
if(v){ |
653 |
|
16203935 |
v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); |
654 |
|
16203935 |
v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l&0xFF] + 3*ff_quant3bA[t&0xFF]]); |
655 |
✗✓ |
16203935 |
if ((uint16_t)v != v) { |
656 |
|
|
av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n"); |
657 |
|
|
v = 1; |
658 |
|
|
} |
659 |
|
16203935 |
xc->x=x; |
660 |
|
16203935 |
(xc++)->coeff= v; |
661 |
|
|
} |
662 |
|
|
}else{ |
663 |
✓✓ |
859485 |
if(!run){ |
664 |
✓✓ |
293505 |
if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
665 |
|
17236 |
else run= INT_MAX; |
666 |
|
293505 |
v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); |
667 |
|
293505 |
v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); |
668 |
✗✓ |
293505 |
if ((uint16_t)v != v) { |
669 |
|
|
av_log(s->avctx, AV_LOG_ERROR, "Coefficient damaged\n"); |
670 |
|
|
v = 1; |
671 |
|
|
} |
672 |
|
|
|
673 |
|
293505 |
xc->x=x; |
674 |
|
293505 |
(xc++)->coeff= v; |
675 |
|
|
}else{ |
676 |
|
|
int max_run; |
677 |
|
565980 |
run--; |
678 |
|
565980 |
v=0; |
679 |
|
|
av_assert2(run >= 0); |
680 |
✓✓✓✓
|
565980 |
if(y) max_run= FFMIN(run, prev_xc->x - x - 2); |
681 |
✓✓ |
31562 |
else max_run= FFMIN(run, w-x-1); |
682 |
✓✓ |
565980 |
if(parent_xc) |
683 |
✓✓ |
562134 |
max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); |
684 |
|
|
av_assert2(max_run >= 0 && max_run <= run); |
685 |
|
|
|
686 |
|
565980 |
x+= max_run; |
687 |
|
565980 |
run-= max_run; |
688 |
|
|
} |
689 |
|
|
} |
690 |
|
|
} |
691 |
|
376000 |
(xc++)->x= w+1; //end marker |
692 |
|
376000 |
prev_xc= prev2_xc; |
693 |
|
376000 |
prev2_xc= xc; |
694 |
|
|
|
695 |
✓✓ |
376000 |
if(parent_xc){ |
696 |
✓✓ |
360000 |
if(y&1){ |
697 |
✓✓ |
289280 |
while(parent_xc->x != parent->width+1) |
698 |
|
109742 |
parent_xc++; |
699 |
|
179538 |
parent_xc++; |
700 |
|
179538 |
prev_parent_xc= parent_xc; |
701 |
|
|
}else{ |
702 |
|
180462 |
parent_xc= prev_parent_xc; |
703 |
|
|
} |
704 |
|
|
} |
705 |
|
|
} |
706 |
|
|
|
707 |
|
22128 |
(xc++)->x= w+1; //end marker |
708 |
|
22128 |
} |
709 |
|
|
|
710 |
|
|
#endif /* AVCODEC_SNOW_H */ |