Mercurial > libavcodec.hg
annotate mpegvideo.c @ 324:9c6f056f0e41 libavcodec
fixed mpeg4 time stuff on encoding
mpeg4 b-frame enoding support
removed old, out-commented ratecontrol
reuse motion compensation code between encoding & decoding
prefix newly added global functions with ff_ to reduce namespace polution
b-frame ME (unfinished, but working)
added some comments to mpegvideo.h
do MC on encoding only once if possible
bugs? ;)
author | michaelni |
---|---|
date | Wed, 17 Apr 2002 04:32:12 +0000 |
parents | 2b00e171b1d4 |
children | 15efd80cf51e |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest mpeg encoder (well, it was the simplest!) | |
3 * Copyright (c) 2000,2001 Gerard Lantau. | |
4 * | |
5 * This program is free software; you can redistribute it and/or modify | |
6 * it under the terms of the GNU General Public License as published by | |
7 * the Free Software Foundation; either version 2 of the License, or | |
8 * (at your option) any later version. | |
9 * | |
10 * This program is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 * GNU General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU General Public License | |
16 * along with this program; if not, write to the Free Software | |
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
295 | 18 * |
19 * 4MV & hq encoding stuff by Michael Niedermayer <michaelni@gmx.at> | |
0 | 20 */ |
21 #include <stdlib.h> | |
22 #include <stdio.h> | |
23 #include <math.h> | |
24 #include <string.h> | |
25 #include "avcodec.h" | |
26 #include "dsputil.h" | |
27 #include "mpegvideo.h" | |
28 | |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
29 #ifdef USE_FASTMEMCPY |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
30 #include "fastmemcpy.h" |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
31 #endif |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
32 |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
33 static void encode_picture(MpegEncContext *s, int picture_number); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
34 static void rate_control_init(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
35 static int rate_estimate_qscale(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
36 static void dct_unquantize_mpeg1_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
37 DCTELEM *block, int n, int qscale); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
38 static void dct_unquantize_h263_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
39 DCTELEM *block, int n, int qscale); |
206 | 40 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
220 | 41 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
206 | 42 |
220 | 43 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c; |
206 | 44 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
45 | |
0 | 46 #define EDGE_WIDTH 16 |
47 | |
48 /* enable all paranoid tests for rounding, overflows, etc... */ | |
49 //#define PARANOID | |
50 | |
51 //#define DEBUG | |
52 | |
321 | 53 |
0 | 54 /* for jpeg fast DCT */ |
55 #define CONST_BITS 14 | |
56 | |
57 static const unsigned short aanscales[64] = { | |
58 /* precomputed values scaled up by 14 bits */ | |
59 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
60 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
61 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
62 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
63 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
64 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
65 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
66 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
67 }; | |
68 | |
69 static UINT8 h263_chroma_roundtab[16] = { | |
70 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
71 }; | |
72 | |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
73 static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1]; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
74 static UINT8 default_fcode_tab[MAX_MV*2+1]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
75 |
321 | 76 extern UINT8 zigzag_end[64]; |
77 | |
0 | 78 /* default motion estimation */ |
321 | 79 int motion_estimation_method = ME_EPZS; |
200 | 80 |
220 | 81 static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale) |
0 | 82 { |
83 int i; | |
84 | |
85 if (av_fdct == jpeg_fdct_ifast) { | |
86 for(i=0;i<64;i++) { | |
87 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
220 | 88 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
89 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
90 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
0 | 91 |
220 | 92 qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
93 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)])); | |
0 | 94 } |
95 } else { | |
96 for(i=0;i<64;i++) { | |
97 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
220 | 98 So 16 <= qscale * quant_matrix[i] <= 7905 |
99 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
100 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
0 | 101 */ |
220 | 102 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
103 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
0 | 104 } |
105 } | |
106 } | |
107 | |
108 /* init common structure for both encoder and decoder */ | |
109 int MPV_common_init(MpegEncContext *s) | |
110 { | |
111 int c_size, i; | |
112 UINT8 *pict; | |
113 | |
312 | 114 s->dct_unquantize_h263 = dct_unquantize_h263_c; |
115 s->dct_unquantize_mpeg = dct_unquantize_mpeg1_c; | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
116 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
117 #ifdef HAVE_MMX |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
118 MPV_common_init_mmx(s); |
8 | 119 #endif |
312 | 120 //setup default unquantizers (mpeg4 might change it later) |
121 if(s->out_format == FMT_H263) | |
122 s->dct_unquantize = s->dct_unquantize_h263; | |
123 else | |
124 s->dct_unquantize = s->dct_unquantize_mpeg; | |
125 | |
0 | 126 s->mb_width = (s->width + 15) / 16; |
127 s->mb_height = (s->height + 15) / 16; | |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
128 s->mb_num = s->mb_width * s->mb_height; |
0 | 129 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
130 | |
131 for(i=0;i<3;i++) { | |
132 int w, h, shift, pict_start; | |
133 | |
134 w = s->linesize; | |
135 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
136 shift = (i == 0) ? 0 : 1; | |
137 c_size = (w >> shift) * (h >> shift); | |
138 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
139 | |
140 pict = av_mallocz(c_size); | |
141 if (pict == NULL) | |
142 goto fail; | |
143 s->last_picture_base[i] = pict; | |
144 s->last_picture[i] = pict + pict_start; | |
145 | |
146 pict = av_mallocz(c_size); | |
147 if (pict == NULL) | |
148 goto fail; | |
149 s->next_picture_base[i] = pict; | |
150 s->next_picture[i] = pict + pict_start; | |
324 | 151 |
0 | 152 if (s->has_b_frames) { |
153 pict = av_mallocz(c_size); | |
154 if (pict == NULL) | |
155 goto fail; | |
156 s->aux_picture_base[i] = pict; | |
157 s->aux_picture[i] = pict + pict_start; | |
158 } | |
159 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
160 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
161 if (s->encoding) { |
324 | 162 int j; |
163 int mv_table_size= (s->mb_width+2)*(s->mb_height+2); | |
164 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
165 /* Allocate MB type table */ |
239 | 166 s->mb_type = av_mallocz(s->mb_num * sizeof(char)); |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
167 if (s->mb_type == NULL) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
168 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
169 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
170 } |
239 | 171 |
172 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16)); | |
173 if (s->mb_var == NULL) { | |
174 perror("malloc"); | |
175 goto fail; | |
176 } | |
324 | 177 |
178 /* Allocate MV tables */ | |
179 s->p_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
180 if (s->p_mv_table == NULL) { | |
181 perror("malloc"); | |
182 goto fail; | |
183 } | |
184 s->last_p_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
185 if (s->last_p_mv_table == NULL) { | |
186 perror("malloc"); | |
187 goto fail; | |
188 } | |
189 s->b_forw_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
190 if (s->b_forw_mv_table == NULL) { | |
191 perror("malloc"); | |
192 goto fail; | |
193 } | |
194 s->b_back_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
195 if (s->b_back_mv_table == NULL) { | |
196 perror("malloc"); | |
197 goto fail; | |
198 } | |
199 s->b_bidir_forw_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
200 if (s->b_bidir_forw_mv_table == NULL) { | |
201 perror("malloc"); | |
202 goto fail; | |
203 } | |
204 s->b_bidir_back_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
205 if (s->b_bidir_back_mv_table == NULL) { | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
206 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
207 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
208 } |
324 | 209 s->b_direct_forw_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); |
210 if (s->b_direct_forw_mv_table == NULL) { | |
211 perror("malloc"); | |
212 goto fail; | |
213 } | |
214 s->b_direct_back_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
215 if (s->b_direct_back_mv_table == NULL) { | |
216 perror("malloc"); | |
217 goto fail; | |
218 } | |
219 s->b_direct_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
220 if (s->b_direct_mv_table == NULL) { | |
221 perror("malloc"); | |
222 goto fail; | |
223 } | |
224 | |
225 if(s->max_b_frames){ | |
226 for(j=0; j<REORDER_BUFFER_SIZE; j++){ | |
227 int i; | |
228 for(i=0;i<3;i++) { | |
229 int w, h, shift; | |
230 | |
231 w = s->linesize; | |
232 h = s->mb_height * 16; | |
233 shift = (i == 0) ? 0 : 1; | |
234 c_size = (w >> shift) * (h >> shift); | |
235 | |
236 pict = av_mallocz(c_size); | |
237 if (pict == NULL) | |
238 goto fail; | |
239 s->picture_buffer[j][i] = pict; | |
240 } | |
241 } | |
242 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
243 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
244 |
288 | 245 if (s->out_format == FMT_H263 || s->encoding) { |
0 | 246 int size; |
247 /* MV prediction */ | |
248 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
249 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
250 if (s->motion_val == NULL) | |
251 goto fail; | |
252 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
253 } | |
254 | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
255 if (s->h263_pred || s->h263_plus) { |
0 | 256 int y_size, c_size, i, size; |
257 | |
258 /* dc values */ | |
259 | |
260 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
261 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
262 size = y_size + 2 * c_size; | |
263 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
264 if (s->dc_val[0] == NULL) | |
265 goto fail; | |
266 s->dc_val[1] = s->dc_val[0] + y_size; | |
267 s->dc_val[2] = s->dc_val[1] + c_size; | |
268 for(i=0;i<size;i++) | |
269 s->dc_val[0][i] = 1024; | |
270 | |
271 /* ac values */ | |
272 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
273 if (s->ac_val[0] == NULL) | |
274 goto fail; | |
275 s->ac_val[1] = s->ac_val[0] + y_size; | |
276 s->ac_val[2] = s->ac_val[1] + c_size; | |
277 | |
278 /* cbp values */ | |
279 s->coded_block = av_mallocz(y_size); | |
280 if (!s->coded_block) | |
281 goto fail; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
282 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
283 /* which mb is a intra block */ |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
284 s->mbintra_table = av_mallocz(s->mb_num); |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
285 if (!s->mbintra_table) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
286 goto fail; |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
287 memset(s->mbintra_table, 1, s->mb_num); |
197
21abf1b20016
different fix, s->mbintra_table used only if h263_pred set. - patch by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
196
diff
changeset
|
288 } |
0 | 289 /* default structure is frame */ |
290 s->picture_structure = PICT_FRAME; | |
291 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
292 /* init macroblock skip table */ |
324 | 293 s->mbskip_table = av_mallocz(s->mb_num); |
294 if (!s->mbskip_table) | |
295 goto fail; | |
294 | 296 |
297 s->block= s->intra_block; | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
298 |
0 | 299 s->context_initialized = 1; |
300 return 0; | |
301 fail: | |
244 | 302 MPV_common_end(s); |
303 return -1; | |
304 } | |
305 | |
324 | 306 #define CHECK_FREE(p)\ |
307 {\ | |
308 if(p) free(p);\ | |
309 p= NULL;\ | |
310 } | |
311 | |
244 | 312 /* init common structure for both encoder and decoder */ |
313 void MPV_common_end(MpegEncContext *s) | |
314 { | |
315 int i; | |
316 | |
324 | 317 CHECK_FREE(s->mb_type); |
318 CHECK_FREE(s->mb_var); | |
319 CHECK_FREE(s->p_mv_table); | |
320 CHECK_FREE(s->last_p_mv_table); | |
321 CHECK_FREE(s->b_forw_mv_table); | |
322 CHECK_FREE(s->b_back_mv_table); | |
323 CHECK_FREE(s->b_bidir_forw_mv_table); | |
324 CHECK_FREE(s->b_bidir_back_mv_table); | |
325 CHECK_FREE(s->b_direct_forw_mv_table); | |
326 CHECK_FREE(s->b_direct_back_mv_table); | |
327 CHECK_FREE(s->b_direct_mv_table); | |
328 CHECK_FREE(s->motion_val); | |
329 CHECK_FREE(s->dc_val[0]); | |
330 CHECK_FREE(s->ac_val[0]); | |
331 CHECK_FREE(s->coded_block); | |
332 CHECK_FREE(s->mbintra_table); | |
244 | 333 |
324 | 334 CHECK_FREE(s->mbskip_table); |
0 | 335 for(i=0;i<3;i++) { |
324 | 336 int j; |
337 CHECK_FREE(s->last_picture_base[i]); | |
338 CHECK_FREE(s->next_picture_base[i]); | |
339 CHECK_FREE(s->aux_picture_base[i]); | |
340 for(j=0; j<REORDER_BUFFER_SIZE; j++){ | |
341 CHECK_FREE(s->picture_buffer[j][i]); | |
342 } | |
0 | 343 } |
344 s->context_initialized = 0; | |
345 } | |
346 | |
347 /* init video encoder */ | |
348 int MPV_encode_init(AVCodecContext *avctx) | |
349 { | |
350 MpegEncContext *s = avctx->priv_data; | |
60 | 351 int i; |
0 | 352 |
315 | 353 avctx->pix_fmt = PIX_FMT_YUV420P; |
354 | |
0 | 355 s->bit_rate = avctx->bit_rate; |
268 | 356 s->bit_rate_tolerance = avctx->bit_rate_tolerance; |
0 | 357 s->frame_rate = avctx->frame_rate; |
358 s->width = avctx->width; | |
359 s->height = avctx->height; | |
360 s->gop_size = avctx->gop_size; | |
162 | 361 s->rtp_mode = avctx->rtp_mode; |
362 s->rtp_payload_size = avctx->rtp_payload_size; | |
231 | 363 if (avctx->rtp_callback) |
364 s->rtp_callback = avctx->rtp_callback; | |
268 | 365 s->qmin= avctx->qmin; |
366 s->qmax= avctx->qmax; | |
367 s->max_qdiff= avctx->max_qdiff; | |
368 s->qcompress= avctx->qcompress; | |
369 s->qblur= avctx->qblur; | |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
370 s->avctx = avctx; |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
371 s->aspect_ratio_info= avctx->aspect_ratio_info; |
294 | 372 s->flags= avctx->flags; |
324 | 373 s->max_b_frames= avctx->max_b_frames; |
162 | 374 |
0 | 375 if (s->gop_size <= 1) { |
376 s->intra_only = 1; | |
377 s->gop_size = 12; | |
378 } else { | |
379 s->intra_only = 0; | |
380 } | |
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
381 |
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
382 /* ME algorithm */ |
321 | 383 if (avctx->me_method == 0) |
384 /* For compatibility */ | |
385 s->me_method = motion_estimation_method; | |
386 else | |
387 s->me_method = avctx->me_method; | |
388 | |
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
389 /* Fixed QSCALE */ |
0 | 390 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE); |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
391 |
0 | 392 switch(avctx->codec->id) { |
393 case CODEC_ID_MPEG1VIDEO: | |
394 s->out_format = FMT_MPEG1; | |
395 break; | |
396 case CODEC_ID_MJPEG: | |
397 s->out_format = FMT_MJPEG; | |
398 s->intra_only = 1; /* force intra only for jpeg */ | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
399 s->mjpeg_write_tables = 1; /* write all tables */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
400 s->mjpeg_vsample[0] = 2; /* set up default sampling factors */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
401 s->mjpeg_vsample[1] = 1; /* the only currently supported values */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
402 s->mjpeg_vsample[2] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
403 s->mjpeg_hsample[0] = 2; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
404 s->mjpeg_hsample[1] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
405 s->mjpeg_hsample[2] = 1; |
0 | 406 if (mjpeg_init(s) < 0) |
407 return -1; | |
408 break; | |
409 case CODEC_ID_H263: | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
410 if (h263_get_picture_format(s->width, s->height) == 7) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
411 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
0 | 412 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
413 } |
0 | 414 s->out_format = FMT_H263; |
415 break; | |
416 case CODEC_ID_H263P: | |
417 s->out_format = FMT_H263; | |
162 | 418 s->rtp_mode = 1; |
419 s->rtp_payload_size = 1200; | |
0 | 420 s->h263_plus = 1; |
78 | 421 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
422 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
423 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
424 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
425 s->umvplus_dec = 0; |
0 | 426 break; |
427 case CODEC_ID_RV10: | |
428 s->out_format = FMT_H263; | |
429 s->h263_rv10 = 1; | |
430 break; | |
71 | 431 case CODEC_ID_MPEG4: |
0 | 432 s->out_format = FMT_H263; |
433 s->h263_pred = 1; | |
434 s->unrestricted_mv = 1; | |
324 | 435 s->has_b_frames= s->max_b_frames ? 1 : 0; |
0 | 436 break; |
307 | 437 case CODEC_ID_MSMPEG4V1: |
0 | 438 s->out_format = FMT_H263; |
439 s->h263_msmpeg4 = 1; | |
440 s->h263_pred = 1; | |
441 s->unrestricted_mv = 1; | |
307 | 442 s->msmpeg4_version= 1; |
443 break; | |
444 case CODEC_ID_MSMPEG4V2: | |
445 s->out_format = FMT_H263; | |
446 s->h263_msmpeg4 = 1; | |
447 s->h263_pred = 1; | |
448 s->unrestricted_mv = 1; | |
449 s->msmpeg4_version= 2; | |
450 break; | |
451 case CODEC_ID_MSMPEG4V3: | |
452 s->out_format = FMT_H263; | |
453 s->h263_msmpeg4 = 1; | |
454 s->h263_pred = 1; | |
455 s->unrestricted_mv = 1; | |
456 s->msmpeg4_version= 3; | |
0 | 457 break; |
458 default: | |
459 return -1; | |
460 } | |
295 | 461 |
462 if((s->flags&CODEC_FLAG_4MV) && !(s->flags&CODEC_FLAG_HQ)){ | |
463 printf("4MV is currently only supported in HQ mode\n"); | |
464 return -1; | |
465 } | |
0 | 466 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
467 { /* set up some save defaults, some codecs might override them later */ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
468 static int done=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
469 if(!done){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
470 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
471 done=1; |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
472 memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1)); |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
473 memset(default_fcode_tab , 0, sizeof(UINT8)*(2*MAX_MV+1)); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
474 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
475 for(i=-16; i<16; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
476 default_fcode_tab[i + MAX_MV]= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
477 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
478 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
479 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
480 s->mv_penalty= default_mv_penalty; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
481 s->fcode_tab= default_fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
482 |
0 | 483 if (s->out_format == FMT_H263) |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
484 h263_encode_init(s); |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
485 else if (s->out_format == FMT_MPEG1) |
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
486 mpeg1_encode_init(s); |
0 | 487 |
287 | 488 /* dont use mv_penalty table for crap MV as it would be confused */ |
324 | 489 if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty; |
287 | 490 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
491 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
492 |
0 | 493 /* init */ |
494 if (MPV_common_init(s) < 0) | |
495 return -1; | |
496 | |
60 | 497 /* init default q matrix */ |
498 for(i=0;i<64;i++) { | |
499 s->intra_matrix[i] = default_intra_matrix[i]; | |
500 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
501 } | |
502 | |
0 | 503 /* rate control init */ |
504 rate_control_init(s); | |
505 | |
506 s->picture_number = 0; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
507 s->picture_in_gop_number = 0; |
0 | 508 s->fake_picture_number = 0; |
509 /* motion detector init */ | |
510 s->f_code = 1; | |
324 | 511 s->b_code = 1; |
0 | 512 |
513 return 0; | |
514 } | |
515 | |
516 int MPV_encode_end(AVCodecContext *avctx) | |
517 { | |
518 MpegEncContext *s = avctx->priv_data; | |
519 | |
520 #ifdef STATS | |
521 print_stats(); | |
522 #endif | |
523 MPV_common_end(s); | |
524 if (s->out_format == FMT_MJPEG) | |
525 mjpeg_close(s); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
526 |
0 | 527 return 0; |
528 } | |
529 | |
530 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 531 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 532 { |
533 UINT8 *ptr, *last_line; | |
534 int i; | |
535 | |
536 last_line = buf + (height - 1) * wrap; | |
537 for(i=0;i<w;i++) { | |
538 /* top and bottom */ | |
539 memcpy(buf - (i + 1) * wrap, buf, width); | |
540 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
541 } | |
542 /* left and right */ | |
543 ptr = buf; | |
544 for(i=0;i<height;i++) { | |
545 memset(ptr - w, ptr[0], w); | |
546 memset(ptr + width, ptr[width-1], w); | |
547 ptr += wrap; | |
548 } | |
549 /* corners */ | |
550 for(i=0;i<w;i++) { | |
551 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
552 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
553 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
554 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
555 } | |
556 } | |
557 | |
558 /* generic function for encode/decode called before a frame is coded/decoded */ | |
559 void MPV_frame_start(MpegEncContext *s) | |
560 { | |
561 int i; | |
562 UINT8 *tmp; | |
563 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
564 s->mb_skiped = 0; |
0 | 565 if (s->pict_type == B_TYPE) { |
566 for(i=0;i<3;i++) { | |
567 s->current_picture[i] = s->aux_picture[i]; | |
568 } | |
569 } else { | |
262 | 570 s->last_non_b_pict_type= s->pict_type; |
0 | 571 for(i=0;i<3;i++) { |
572 /* swap next and last */ | |
573 tmp = s->last_picture[i]; | |
574 s->last_picture[i] = s->next_picture[i]; | |
575 s->next_picture[i] = tmp; | |
576 s->current_picture[i] = tmp; | |
577 } | |
578 } | |
579 } | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
580 |
0 | 581 /* generic function for encode/decode called after a frame has been coded/decoded */ |
582 void MPV_frame_end(MpegEncContext *s) | |
583 { | |
584 /* draw edge for correct motion prediction if outside */ | |
220 | 585 if (s->pict_type != B_TYPE && !s->intra_only) { |
257 | 586 if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version==500){ |
176
9ce215ee9216
unrestricted MC fixed - thanks to Michael Niedermayer for idea
arpi_esp
parents:
162
diff
changeset
|
587 draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH); |
9ce215ee9216
unrestricted MC fixed - thanks to Michael Niedermayer for idea
arpi_esp
parents:
162
diff
changeset
|
588 draw_edges(s->current_picture[1], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2); |
9ce215ee9216
unrestricted MC fixed - thanks to Michael Niedermayer for idea
arpi_esp
parents:
162
diff
changeset
|
589 draw_edges(s->current_picture[2], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2); |
189
42552c1cf360
fix different UMV handling for mpeg4 vs. h263 - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
187
diff
changeset
|
590 }else{ |
257 | 591 /* mpeg4? / opendivx / xvid */ |
0 | 592 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
593 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
594 draw_edges(s->current_picture[2], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
189
42552c1cf360
fix different UMV handling for mpeg4 vs. h263 - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
187
diff
changeset
|
595 } |
0 | 596 } |
207 | 597 emms_c(); |
0 | 598 } |
599 | |
324 | 600 /* reorder input for encoding */ |
601 void reorder_input(MpegEncContext *s, AVPicture *pict) | |
602 { | |
603 int i, j, index; | |
604 | |
605 if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES; | |
606 | |
607 // delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff) | |
608 | |
609 for(j=0; j<REORDER_BUFFER_SIZE-1; j++){ | |
610 s->coded_order[j]= s->coded_order[j+1]; | |
611 } | |
612 s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers | |
613 | |
614 switch(s->input_pict_type){ | |
615 default: | |
616 case I_TYPE: | |
617 case S_TYPE: | |
618 case P_TYPE: | |
619 index= s->max_b_frames - s->b_frames_since_non_b; | |
620 s->b_frames_since_non_b=0; | |
621 break; | |
622 case B_TYPE: | |
623 index= s->max_b_frames + 1; | |
624 s->b_frames_since_non_b++; | |
625 break; | |
626 } | |
627 //printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize); | |
628 if( (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED)) | |
629 && pict->linesize[0] == s->linesize | |
630 && pict->linesize[1] == s->linesize>>1 | |
631 && pict->linesize[2] == s->linesize>>1){ | |
632 //printf("ptr\n"); | |
633 for(i=0; i<3; i++){ | |
634 s->coded_order[index].picture[i]= pict->data[i]; | |
635 } | |
636 }else{ | |
637 //printf("copy\n"); | |
638 for(i=0; i<3; i++){ | |
639 uint8_t *src = pict->data[i]; | |
640 uint8_t *dest; | |
641 int src_wrap = pict->linesize[i]; | |
642 int dest_wrap = s->linesize; | |
643 int w = s->width; | |
644 int h = s->height; | |
645 | |
646 if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering | |
647 else dest= s->picture_buffer[s->picture_buffer_index][i]; | |
648 | |
649 if (i >= 1) { | |
650 dest_wrap >>= 1; | |
651 w >>= 1; | |
652 h >>= 1; | |
653 } | |
654 | |
655 s->coded_order[index].picture[i]= dest; | |
656 for(j=0;j<h;j++) { | |
657 memcpy(dest, src, w); | |
658 dest += dest_wrap; | |
659 src += src_wrap; | |
660 } | |
661 } | |
662 if(index!=0){ | |
663 s->picture_buffer_index++; | |
664 if(s->picture_buffer_index >= REORDER_BUFFER_SIZE-1) s->picture_buffer_index=0; | |
665 } | |
666 } | |
667 s->coded_order[index].pict_type = s->input_pict_type; | |
668 s->coded_order[index].qscale = s->input_qscale; | |
669 s->coded_order[index].force_type= s->force_input_type; | |
670 s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number; | |
671 s->coded_order[index].picture_number= s->input_picture_number; | |
672 | |
673 for(i=0; i<3; i++){ | |
674 s->new_picture[i]= s->coded_order[0].picture[i]; | |
675 } | |
676 } | |
677 | |
0 | 678 int MPV_encode_picture(AVCodecContext *avctx, |
679 unsigned char *buf, int buf_size, void *data) | |
680 { | |
681 MpegEncContext *s = avctx->priv_data; | |
682 AVPicture *pict = data; | |
683 | |
324 | 684 s->input_qscale = avctx->quality; |
0 | 685 |
686 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
687 | |
324 | 688 s->force_input_type= (avctx->flags&CODEC_FLAG_TYPE) ? |
298
e20de99b6295
using Juanjo's way to pass forced type - especially as I_TYPE/P_TYPE defines aren't exported in the public interface, and the flagged method matches the way as qscale forcing work
arpi_esp
parents:
297
diff
changeset
|
689 (avctx->key_frame ? I_TYPE : P_TYPE) : 0; |
0 | 690 if (!s->intra_only) { |
691 /* first picture of GOP is intra */ | |
324 | 692 if (s->input_picture_in_gop_number % s->gop_size==0 || s->force_input_type==I_TYPE){ |
693 s->input_picture_in_gop_number=0; | |
694 s->input_pict_type = I_TYPE; | |
695 }else if(s->max_b_frames==0){ | |
696 s->input_pict_type = P_TYPE; | |
697 }else{ | |
698 if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ | |
699 s->input_pict_type = B_TYPE; | |
700 else | |
701 s->input_pict_type = P_TYPE; | |
702 } | |
0 | 703 } else { |
324 | 704 s->input_pict_type = I_TYPE; |
0 | 705 } |
706 | |
324 | 707 reorder_input(s, pict); |
708 | |
709 /* output? */ | |
710 if(s->coded_order[0].picture[0]){ | |
711 | |
712 s->pict_type= s->coded_order[0].pict_type; | |
713 if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */ | |
714 s->qscale= s->coded_order[0].qscale; | |
715 s->force_type= s->coded_order[0].force_type; | |
716 s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number; | |
717 s->picture_number= s->coded_order[0].picture_number; | |
718 | |
719 MPV_frame_start(s); | |
286 | 720 |
324 | 721 encode_picture(s, s->picture_number); |
722 avctx->key_frame = (s->pict_type == I_TYPE); | |
723 avctx->header_bits = s->header_bits; | |
724 avctx->mv_bits = s->mv_bits; | |
725 avctx->misc_bits = s->misc_bits; | |
726 avctx->i_tex_bits = s->i_tex_bits; | |
727 avctx->p_tex_bits = s->p_tex_bits; | |
728 avctx->i_count = s->i_count; | |
729 avctx->p_count = s->p_count; | |
730 avctx->skip_count = s->skip_count; | |
0 | 731 |
324 | 732 MPV_frame_end(s); |
733 | |
734 if (s->out_format == FMT_MJPEG) | |
735 mjpeg_picture_trailer(s); | |
736 | |
737 avctx->quality = s->qscale; | |
738 } | |
739 | |
740 s->input_picture_number++; | |
741 s->input_picture_in_gop_number++; | |
0 | 742 |
743 flush_put_bits(&s->pb); | |
268 | 744 s->last_frame_bits= s->frame_bits; |
745 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8; | |
746 s->total_bits += s->frame_bits; | |
286 | 747 avctx->frame_bits = s->frame_bits; |
748 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n", | |
749 //s->f_code, avctx->key_frame, s->header_bits, s->mv_bits, s->misc_bits, s->frame_bits, s->i_tex_bits, s->p_tex_bits); | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
750 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
751 if (avctx->get_psnr) { |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
752 /* At this point pict->data should have the original frame */ |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
753 /* an s->current_picture should have the coded/decoded frame */ |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
754 get_psnr(pict->data, s->current_picture, |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
755 pict->linesize, s->linesize, avctx); |
324 | 756 // printf("%f\n", avctx->psnr_y); |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
757 } |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
758 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 759 } |
760 | |
761 static inline int clip(int a, int amin, int amax) | |
762 { | |
763 if (a < amin) | |
764 return amin; | |
765 else if (a > amax) | |
766 return amax; | |
767 else | |
768 return a; | |
769 } | |
770 | |
255 | 771 static inline void gmc1_motion(MpegEncContext *s, |
772 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
773 int dest_offset, | |
774 UINT8 **ref_picture, int src_offset, | |
775 int h) | |
776 { | |
777 UINT8 *ptr; | |
324 | 778 int offset, src_x, src_y, linesize; |
255 | 779 int motion_x, motion_y; |
780 | |
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
781 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n"); |
255 | 782 motion_x= s->sprite_offset[0][0]; |
783 motion_y= s->sprite_offset[0][1]; | |
784 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
785 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
786 motion_x<<=(3-s->sprite_warping_accuracy); | |
787 motion_y<<=(3-s->sprite_warping_accuracy); | |
788 src_x = clip(src_x, -16, s->width); | |
789 if (src_x == s->width) | |
790 motion_x =0; | |
791 src_y = clip(src_y, -16, s->height); | |
792 if (src_y == s->height) | |
793 motion_y =0; | |
794 | |
795 linesize = s->linesize; | |
796 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
797 | |
798 dest_y+=dest_offset; | |
799 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
800 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
801 | |
802 motion_x= s->sprite_offset[1][0]; | |
803 motion_y= s->sprite_offset[1][1]; | |
804 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
805 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
806 motion_x<<=(3-s->sprite_warping_accuracy); | |
807 motion_y<<=(3-s->sprite_warping_accuracy); | |
808 src_x = clip(src_x, -8, s->width>>1); | |
809 if (src_x == s->width>>1) | |
810 motion_x =0; | |
811 src_y = clip(src_y, -8, s->height>>1); | |
812 if (src_y == s->height>>1) | |
813 motion_y =0; | |
814 | |
815 offset = (src_y * linesize>>1) + src_x + (src_offset>>1); | |
816 ptr = ref_picture[1] + offset; | |
817 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
818 ptr = ref_picture[2] + offset; | |
819 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
820 | |
821 return; | |
822 } | |
823 | |
0 | 824 /* apply one mpeg motion vector to the three components */ |
825 static inline void mpeg_motion(MpegEncContext *s, | |
826 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
827 int dest_offset, | |
828 UINT8 **ref_picture, int src_offset, | |
829 int field_based, op_pixels_func *pix_op, | |
830 int motion_x, int motion_y, int h) | |
831 { | |
832 UINT8 *ptr; | |
833 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
255 | 834 if(s->quarter_sample) |
835 { | |
836 motion_x>>=1; | |
837 motion_y>>=1; | |
838 } | |
324 | 839 |
0 | 840 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
841 src_x = s->mb_x * 16 + (motion_x >> 1); | |
842 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
843 | |
844 /* WARNING: do no forget half pels */ | |
845 height = s->height >> field_based; | |
846 src_x = clip(src_x, -16, s->width); | |
847 if (src_x == s->width) | |
848 dxy &= ~1; | |
849 src_y = clip(src_y, -16, height); | |
850 if (src_y == height) | |
851 dxy &= ~2; | |
852 linesize = s->linesize << field_based; | |
853 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
854 dest_y += dest_offset; | |
855 pix_op[dxy](dest_y, ptr, linesize, h); | |
856 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
857 | |
858 if (s->out_format == FMT_H263) { | |
859 dxy = 0; | |
860 if ((motion_x & 3) != 0) | |
861 dxy |= 1; | |
862 if ((motion_y & 3) != 0) | |
863 dxy |= 2; | |
864 mx = motion_x >> 2; | |
865 my = motion_y >> 2; | |
866 } else { | |
867 mx = motion_x / 2; | |
868 my = motion_y / 2; | |
869 dxy = ((my & 1) << 1) | (mx & 1); | |
870 mx >>= 1; | |
871 my >>= 1; | |
872 } | |
873 | |
874 src_x = s->mb_x * 8 + mx; | |
875 src_y = s->mb_y * (8 >> field_based) + my; | |
876 src_x = clip(src_x, -8, s->width >> 1); | |
877 if (src_x == (s->width >> 1)) | |
878 dxy &= ~1; | |
879 src_y = clip(src_y, -8, height >> 1); | |
880 if (src_y == (height >> 1)) | |
881 dxy &= ~2; | |
882 | |
883 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
884 ptr = ref_picture[1] + offset; | |
885 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
886 ptr = ref_picture[2] + offset; | |
887 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
888 } | |
889 | |
255 | 890 static inline void qpel_motion(MpegEncContext *s, |
891 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
892 int dest_offset, | |
893 UINT8 **ref_picture, int src_offset, | |
894 int field_based, op_pixels_func *pix_op, | |
895 qpel_mc_func *qpix_op, | |
896 int motion_x, int motion_y, int h) | |
897 { | |
898 UINT8 *ptr; | |
899 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
900 | |
901 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
902 src_x = s->mb_x * 16 + (motion_x >> 2); | |
903 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
904 | |
905 height = s->height >> field_based; | |
906 src_x = clip(src_x, -16, s->width); | |
907 if (src_x == s->width) | |
908 dxy &= ~3; | |
909 src_y = clip(src_y, -16, height); | |
910 if (src_y == height) | |
911 dxy &= ~12; | |
912 linesize = s->linesize << field_based; | |
913 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
914 dest_y += dest_offset; | |
915 //printf("%d %d %d\n", src_x, src_y, dxy); | |
916 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3); | |
917 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3); | |
918 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3); | |
919 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3); | |
920 | |
921 mx= (motion_x>>1) | (motion_x&1); | |
922 my= (motion_y>>1) | (motion_y&1); | |
923 | |
924 dxy = 0; | |
925 if ((mx & 3) != 0) | |
926 dxy |= 1; | |
927 if ((my & 3) != 0) | |
928 dxy |= 2; | |
929 mx = mx >> 2; | |
930 my = my >> 2; | |
931 | |
932 src_x = s->mb_x * 8 + mx; | |
933 src_y = s->mb_y * (8 >> field_based) + my; | |
934 src_x = clip(src_x, -8, s->width >> 1); | |
935 if (src_x == (s->width >> 1)) | |
936 dxy &= ~1; | |
937 src_y = clip(src_y, -8, height >> 1); | |
938 if (src_y == (height >> 1)) | |
939 dxy &= ~2; | |
940 | |
941 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
942 ptr = ref_picture[1] + offset; | |
943 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
944 ptr = ref_picture[2] + offset; | |
945 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
946 } | |
947 | |
948 | |
0 | 949 static inline void MPV_motion(MpegEncContext *s, |
950 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
951 int dir, UINT8 **ref_picture, | |
255 | 952 op_pixels_func *pix_op, qpel_mc_func *qpix_op) |
0 | 953 { |
954 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
955 int mb_x, mb_y, i; | |
956 UINT8 *ptr, *dest; | |
957 | |
958 mb_x = s->mb_x; | |
959 mb_y = s->mb_y; | |
960 | |
961 switch(s->mv_type) { | |
962 case MV_TYPE_16X16: | |
255 | 963 if(s->mcsel){ |
964 #if 0 | |
965 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
966 ref_picture, 0, | |
967 0, pix_op, | |
968 s->sprite_offset[0][0]>>3, | |
969 s->sprite_offset[0][1]>>3, | |
970 16); | |
971 #else | |
972 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
973 ref_picture, 0, | |
974 16); | |
975 #endif | |
262 | 976 }else if(s->quarter_sample && dir==0){ //FIXME |
255 | 977 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, |
978 ref_picture, 0, | |
979 0, pix_op, qpix_op, | |
980 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
981 }else{ | |
982 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
983 ref_picture, 0, | |
984 0, pix_op, | |
985 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
986 } | |
0 | 987 break; |
988 case MV_TYPE_8X8: | |
989 for(i=0;i<4;i++) { | |
990 motion_x = s->mv[dir][i][0]; | |
991 motion_y = s->mv[dir][i][1]; | |
992 | |
993 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
994 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
295 | 995 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; |
0 | 996 |
997 /* WARNING: do no forget half pels */ | |
998 src_x = clip(src_x, -16, s->width); | |
999 if (src_x == s->width) | |
1000 dxy &= ~1; | |
1001 src_y = clip(src_y, -16, s->height); | |
1002 if (src_y == s->height) | |
1003 dxy &= ~2; | |
1004 | |
1005 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
1006 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
1007 pix_op[dxy](dest, ptr, s->linesize, 8); | |
1008 } | |
1009 /* In case of 8X8, we construct a single chroma motion vector | |
1010 with a special rounding */ | |
1011 mx = 0; | |
1012 my = 0; | |
1013 for(i=0;i<4;i++) { | |
1014 mx += s->mv[dir][i][0]; | |
1015 my += s->mv[dir][i][1]; | |
1016 } | |
1017 if (mx >= 0) | |
1018 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1019 else { | |
1020 mx = -mx; | |
1021 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1022 } | |
1023 if (my >= 0) | |
1024 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1025 else { | |
1026 my = -my; | |
1027 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1028 } | |
1029 dxy = ((my & 1) << 1) | (mx & 1); | |
1030 mx >>= 1; | |
1031 my >>= 1; | |
1032 | |
1033 src_x = mb_x * 8 + mx; | |
1034 src_y = mb_y * 8 + my; | |
1035 src_x = clip(src_x, -8, s->width/2); | |
1036 if (src_x == s->width/2) | |
1037 dxy &= ~1; | |
1038 src_y = clip(src_y, -8, s->height/2); | |
1039 if (src_y == s->height/2) | |
1040 dxy &= ~2; | |
1041 | |
1042 offset = (src_y * (s->linesize >> 1)) + src_x; | |
1043 ptr = ref_picture[1] + offset; | |
1044 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
1045 ptr = ref_picture[2] + offset; | |
1046 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
1047 break; | |
1048 case MV_TYPE_FIELD: | |
1049 if (s->picture_structure == PICT_FRAME) { | |
1050 /* top field */ | |
1051 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
1052 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
1053 1, pix_op, | |
1054 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
1055 /* bottom field */ | |
1056 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
1057 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
1058 1, pix_op, | |
1059 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
1060 } else { | |
1061 | |
1062 | |
1063 } | |
1064 break; | |
1065 } | |
1066 } | |
1067 | |
1068 | |
1069 /* put block[] to dest[] */ | |
1070 static inline void put_dct(MpegEncContext *s, | |
1071 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
1072 { | |
1073 if (!s->mpeg2) | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1074 s->dct_unquantize(s, block, i, s->qscale); |
19
82d4c9be9873
MMX/MMXEXT iDCT support, using external functions currently defined in libmpeg2
arpi_esp
parents:
18
diff
changeset
|
1075 ff_idct (block); |
0 | 1076 put_pixels_clamped(block, dest, line_size); |
1077 } | |
1078 | |
1079 /* add block[] to dest[] */ | |
1080 static inline void add_dct(MpegEncContext *s, | |
1081 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
1082 { | |
1083 if (s->block_last_index[i] >= 0) { | |
1084 if (!s->mpeg2) | |
206 | 1085 if(s->encoding || (!s->h263_msmpeg4)) |
200 | 1086 s->dct_unquantize(s, block, i, s->qscale); |
324 | 1087 |
19
82d4c9be9873
MMX/MMXEXT iDCT support, using external functions currently defined in libmpeg2
arpi_esp
parents:
18
diff
changeset
|
1088 ff_idct (block); |
0 | 1089 add_pixels_clamped(block, dest, line_size); |
1090 } | |
1091 } | |
1092 | |
1093 /* generic function called after a macroblock has been parsed by the | |
1094 decoder or after it has been encoded by the encoder. | |
1095 | |
1096 Important variables used: | |
1097 s->mb_intra : true if intra macroblock | |
1098 s->mv_dir : motion vector direction | |
1099 s->mv_type : motion vector type | |
1100 s->mv : motion vector | |
1101 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
1102 */ | |
1103 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
1104 { | |
244 | 1105 int mb_x, mb_y; |
0 | 1106 int dct_linesize, dct_offset; |
1107 op_pixels_func *op_pix; | |
255 | 1108 qpel_mc_func *op_qpix; |
0 | 1109 |
1110 mb_x = s->mb_x; | |
1111 mb_y = s->mb_y; | |
1112 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1113 #ifdef FF_POSTPROCESS |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1114 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1115 //printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale); |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1116 #endif |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1117 |
0 | 1118 /* update DC predictors for P macroblocks */ |
1119 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1120 if (s->h263_pred || s->h263_aic) { |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1121 if(s->mbintra_table[mb_x + mb_y*s->mb_width]) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1122 { |
244 | 1123 int wrap, xy, v; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1124 s->mbintra_table[mb_x + mb_y*s->mb_width]=0; |
0 | 1125 wrap = 2 * s->mb_width + 2; |
244 | 1126 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 1127 v = 1024; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1128 |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1129 s->dc_val[0][xy] = v; |
244 | 1130 s->dc_val[0][xy + 1] = v; |
1131 s->dc_val[0][xy + wrap] = v; | |
1132 s->dc_val[0][xy + 1 + wrap] = v; | |
0 | 1133 /* ac pred */ |
244 | 1134 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16)); |
1135 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16)); | |
1136 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16)); | |
1137 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16)); | |
0 | 1138 if (s->h263_msmpeg4) { |
244 | 1139 s->coded_block[xy] = 0; |
1140 s->coded_block[xy + 1] = 0; | |
1141 s->coded_block[xy + wrap] = 0; | |
1142 s->coded_block[xy + 1 + wrap] = 0; | |
0 | 1143 } |
1144 /* chroma */ | |
1145 wrap = s->mb_width + 2; | |
244 | 1146 xy = mb_x + 1 + (mb_y + 1) * wrap; |
1147 s->dc_val[1][xy] = v; | |
1148 s->dc_val[2][xy] = v; | |
0 | 1149 /* ac pred */ |
244 | 1150 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); |
1151 memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16)); | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1152 } |
0 | 1153 } else { |
1154 s->last_dc[0] = 128 << s->intra_dc_precision; | |
1155 s->last_dc[1] = 128 << s->intra_dc_precision; | |
1156 s->last_dc[2] = 128 << s->intra_dc_precision; | |
1157 } | |
1158 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1159 else if (s->h263_pred || s->h263_aic) |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1160 s->mbintra_table[mb_x + mb_y*s->mb_width]=1; |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1161 |
280 | 1162 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */ |
324 | 1163 if (s->out_format == FMT_H263) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here |
262 | 1164 if(s->pict_type!=B_TYPE){ |
244 | 1165 int xy, wrap, motion_x, motion_y; |
0 | 1166 |
1167 wrap = 2 * s->mb_width + 2; | |
244 | 1168 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 1169 if (s->mb_intra) { |
1170 motion_x = 0; | |
1171 motion_y = 0; | |
1172 goto motion_init; | |
1173 } else if (s->mv_type == MV_TYPE_16X16) { | |
1174 motion_x = s->mv[0][0][0]; | |
1175 motion_y = s->mv[0][0][1]; | |
1176 motion_init: | |
1177 /* no update if 8X8 because it has been done during parsing */ | |
244 | 1178 s->motion_val[xy][0] = motion_x; |
1179 s->motion_val[xy][1] = motion_y; | |
1180 s->motion_val[xy + 1][0] = motion_x; | |
1181 s->motion_val[xy + 1][1] = motion_y; | |
1182 s->motion_val[xy + wrap][0] = motion_x; | |
1183 s->motion_val[xy + wrap][1] = motion_y; | |
1184 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
1185 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
0 | 1186 } |
262 | 1187 } |
0 | 1188 } |
1189 | |
324 | 1190 if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) { |
0 | 1191 UINT8 *dest_y, *dest_cb, *dest_cr; |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1192 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1193 |
324 | 1194 /* avoid copy if macroblock skipped in last frame too |
1195 dont touch it for B-frames as they need the skip info from the next p-frame */ | |
1196 if (s->pict_type != B_TYPE) { | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1197 mbskip_ptr = &s->mbskip_table[s->mb_y * s->mb_width + s->mb_x]; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1198 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1199 s->mb_skiped = 0; |
324 | 1200 /* if previous was skipped too, then nothing to do ! |
1201 skip only during decoding as we might trash the buffers during encoding a bit */ | |
1202 if (*mbskip_ptr != 0 && !s->encoding) | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1203 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1204 *mbskip_ptr = 1; /* indicate that this time we skiped it */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1205 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1206 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1207 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1208 } |
0 | 1209 |
1210 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
1211 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1212 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1213 | |
1214 if (s->interlaced_dct) { | |
1215 dct_linesize = s->linesize * 2; | |
1216 dct_offset = s->linesize; | |
1217 } else { | |
1218 dct_linesize = s->linesize; | |
1219 dct_offset = s->linesize * 8; | |
1220 } | |
1221 | |
1222 if (!s->mb_intra) { | |
1223 /* motion handling */ | |
324 | 1224 if((s->flags&CODEC_FLAG_HQ) || (!s->encoding)){ |
1225 if (!s->no_rounding){ | |
1226 op_pix = put_pixels_tab; | |
1227 op_qpix= qpel_mc_rnd_tab; | |
1228 }else{ | |
1229 op_pix = put_no_rnd_pixels_tab; | |
1230 op_qpix= qpel_mc_no_rnd_tab; | |
1231 } | |
0 | 1232 |
324 | 1233 if (s->mv_dir & MV_DIR_FORWARD) { |
1234 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
1235 if (!s->no_rounding) | |
1236 op_pix = avg_pixels_tab; | |
1237 else | |
1238 op_pix = avg_no_rnd_pixels_tab; | |
1239 } | |
1240 if (s->mv_dir & MV_DIR_BACKWARD) { | |
1241 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
1242 } | |
0 | 1243 } |
1244 | |
1245 /* add dct residue */ | |
1246 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
1247 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1248 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1249 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1250 | |
57 | 1251 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1252 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1253 } else { |
1254 /* dct only in intra block */ | |
1255 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
1256 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1257 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1258 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1259 | |
57 | 1260 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1261 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1262 } |
1263 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1264 the_end: |
294 | 1265 emms_c(); //FIXME remove |
1266 } | |
1267 | |
324 | 1268 |
1269 static void encode_mb(MpegEncContext *s, int motion_x, int motion_y) | |
294 | 1270 { |
1271 const int mb_x= s->mb_x; | |
1272 const int mb_y= s->mb_y; | |
1273 int i; | |
324 | 1274 #if 0 |
1275 if (s->interlaced_dct) { | |
1276 dct_linesize = s->linesize * 2; | |
1277 dct_offset = s->linesize; | |
1278 } else { | |
1279 dct_linesize = s->linesize; | |
1280 dct_offset = s->linesize * 8; | |
1281 } | |
1282 #endif | |
294 | 1283 |
324 | 1284 if (s->mb_intra) { |
1285 UINT8 *ptr; | |
1286 int wrap; | |
294 | 1287 |
324 | 1288 wrap = s->linesize; |
1289 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
1290 get_pixels(s->block[0], ptr , wrap); | |
1291 get_pixels(s->block[1], ptr + 8, wrap); | |
1292 get_pixels(s->block[2], ptr + 8 * wrap , wrap); | |
1293 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); | |
294 | 1294 |
324 | 1295 wrap >>=1; |
1296 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
1297 get_pixels(s->block[4], ptr, wrap); | |
294 | 1298 |
324 | 1299 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; |
1300 get_pixels(s->block[5], ptr, wrap); | |
1301 }else{ | |
1302 op_pixels_func *op_pix; | |
1303 qpel_mc_func *op_qpix; | |
1304 UINT8 *dest_y, *dest_cb, *dest_cr; | |
1305 UINT8 *ptr; | |
1306 int wrap; | |
294 | 1307 |
324 | 1308 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize ) + mb_x * 16; |
1309 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1310 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1311 | |
1312 if (!s->no_rounding){ | |
1313 op_pix = put_pixels_tab; | |
1314 op_qpix= qpel_mc_rnd_tab; | |
295 | 1315 }else{ |
324 | 1316 op_pix = put_no_rnd_pixels_tab; |
1317 op_qpix= qpel_mc_no_rnd_tab; | |
1318 } | |
295 | 1319 |
324 | 1320 if (s->mv_dir & MV_DIR_FORWARD) { |
1321 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
1322 if (!s->no_rounding) | |
1323 op_pix = avg_pixels_tab; | |
1324 else | |
1325 op_pix = avg_no_rnd_pixels_tab; | |
1326 } | |
1327 if (s->mv_dir & MV_DIR_BACKWARD) { | |
1328 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
1329 } | |
1330 wrap = s->linesize; | |
1331 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
1332 diff_pixels(s->block[0], ptr , dest_y , wrap); | |
1333 diff_pixels(s->block[1], ptr + 8, dest_y + 8, wrap); | |
1334 diff_pixels(s->block[2], ptr + 8 * wrap , dest_y + 8 * wrap , wrap); | |
1335 diff_pixels(s->block[3], ptr + 8 * wrap + 8, dest_y + 8 * wrap + 8, wrap); | |
295 | 1336 |
324 | 1337 wrap >>=1; |
1338 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
1339 diff_pixels(s->block[4], ptr, dest_cb, wrap); | |
1340 | |
1341 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; | |
1342 diff_pixels(s->block[5], ptr, dest_cr, wrap); | |
294 | 1343 } |
1344 | |
1345 #if 0 | |
1346 { | |
1347 float adap_parm; | |
1348 | |
1349 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / | |
1350 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); | |
1351 | |
1352 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", | |
1353 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', | |
1354 s->qscale, adap_parm, s->qscale*adap_parm, | |
1355 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); | |
1356 } | |
1357 #endif | |
1358 /* DCT & quantize */ | |
1359 if (s->h263_msmpeg4) { | |
1360 msmpeg4_dc_scale(s); | |
1361 } else if (s->h263_pred) { | |
1362 h263_dc_scale(s); | |
1363 } else { | |
1364 /* default quantization values */ | |
1365 s->y_dc_scale = 8; | |
1366 s->c_dc_scale = 8; | |
1367 } | |
1368 for(i=0;i<6;i++) { | |
1369 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); | |
1370 } | |
1371 | |
1372 /* huffman encode */ | |
1373 switch(s->out_format) { | |
1374 case FMT_MPEG1: | |
1375 mpeg1_encode_mb(s, s->block, motion_x, motion_y); | |
1376 break; | |
1377 case FMT_H263: | |
1378 if (s->h263_msmpeg4) | |
1379 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1380 else if(s->h263_pred) | |
1381 mpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1382 else | |
1383 h263_encode_mb(s, s->block, motion_x, motion_y); | |
1384 break; | |
1385 case FMT_MJPEG: | |
1386 mjpeg_encode_mb(s, s->block); | |
1387 break; | |
1388 } | |
1389 } | |
1390 | |
1391 static void copy_bits(PutBitContext *pb, UINT8 *src, int length) | |
1392 { | |
1393 int bytes= length>>3; | |
1394 int bits= length&7; | |
1395 int i; | |
1396 | |
1397 for(i=0; i<bytes; i++) put_bits(pb, 8, src[i]); | |
1398 put_bits(pb, bits, src[i]>>(8-bits)); | |
0 | 1399 } |
1400 | |
1401 static void encode_picture(MpegEncContext *s, int picture_number) | |
1402 { | |
294 | 1403 int mb_x, mb_y, last_gob, pdif = 0; |
1404 int i; | |
286 | 1405 int bits; |
294 | 1406 MpegEncContext best_s; |
1407 UINT8 bit_buf[4][3000]; //FIXME check that this is ALLWAYS large enogh for a MB | |
0 | 1408 |
1409 s->picture_number = picture_number; | |
268 | 1410 |
294 | 1411 s->block_wrap[0]= |
1412 s->block_wrap[1]= | |
1413 s->block_wrap[2]= | |
1414 s->block_wrap[3]= s->mb_width*2 + 2; | |
1415 s->block_wrap[4]= | |
1416 s->block_wrap[5]= s->mb_width + 2; | |
1417 | |
268 | 1418 s->last_mc_mb_var = s->mc_mb_var; |
1419 /* Reset the average MB variance */ | |
1420 s->avg_mb_var = 0; | |
1421 s->mc_mb_var = 0; | |
324 | 1422 |
268 | 1423 /* Estimate motion for every MB */ |
324 | 1424 if(s->pict_type != I_TYPE){ |
1425 // int16_t (*tmp)[2]= s->p_mv_table; | |
1426 // s->p_mv_table= s->last_mv_table; | |
1427 // s->last_mv_table= s->mv_table; | |
1428 | |
294 | 1429 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
1430 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; | |
1431 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1432 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1433 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1434 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
1435 s->mb_x = mb_x; | |
1436 s->mb_y = mb_y; | |
1437 s->block_index[0]+=2; | |
1438 s->block_index[1]+=2; | |
1439 s->block_index[2]+=2; | |
1440 s->block_index[3]+=2; | |
268 | 1441 |
294 | 1442 /* compute motion vector & mb_type and store in context */ |
324 | 1443 if(s->pict_type==B_TYPE) |
1444 ff_estimate_b_frame_motion(s, mb_x, mb_y); | |
1445 else | |
1446 ff_estimate_p_frame_motion(s, mb_x, mb_y); | |
294 | 1447 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER; |
268 | 1448 } |
1449 } | |
294 | 1450 emms_c(); |
324 | 1451 }else if(s->pict_type == I_TYPE){ |
294 | 1452 /* I-Frame */ |
1453 //FIXME do we need to zero them? | |
1454 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2); | |
324 | 1455 memset(s->p_mv_table , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2); |
294 | 1456 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
268 | 1457 } |
1458 | |
324 | 1459 if(s->avg_mb_var < s->mc_mb_var && s->pict_type != B_TYPE && (!s->force_type) && s->max_b_frames==0){ //FIXME subtract MV bits |
1460 // FIXME b-frames & scene change detection | |
1461 s->input_pict_type= I_TYPE; | |
271 | 1462 s->pict_type= I_TYPE; |
324 | 1463 s->input_picture_in_gop_number=0; |
294 | 1464 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
1465 //printf("Scene change detected, encoding as I Frame\n"); | |
271 | 1466 } |
324 | 1467 |
1468 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) | |
1469 s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER); | |
1470 ff_fix_long_p_mvs(s); | |
1471 if(s->pict_type==B_TYPE){ | |
1472 s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD); | |
1473 s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD); | |
1474 //FIXME if BIDIR != for&back | |
1475 ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD |MB_TYPE_BIDIR); | |
1476 ff_fix_long_b_mvs(s, s->b_back_mv_table, s->b_code, MB_TYPE_BACKWARD|MB_TYPE_BIDIR); | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1477 } |
324 | 1478 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1479 //printf("f_code %d ///\n", s->f_code); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1480 |
271 | 1481 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var); |
1482 | |
0 | 1483 if (!s->fixed_qscale) |
1484 s->qscale = rate_estimate_qscale(s); | |
1485 | |
1486 /* precompute matrix */ | |
1487 if (s->out_format == FMT_MJPEG) { | |
1488 /* for mjpeg, we do include qscale in the matrix */ | |
1489 s->intra_matrix[0] = default_intra_matrix[0]; | |
1490 for(i=1;i<64;i++) | |
1491 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
220 | 1492 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
0 | 1493 } else { |
220 | 1494 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
1495 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); | |
0 | 1496 } |
1497 | |
286 | 1498 s->last_bits= get_bit_count(&s->pb); |
0 | 1499 switch(s->out_format) { |
1500 case FMT_MJPEG: | |
1501 mjpeg_picture_header(s); | |
1502 break; | |
1503 case FMT_H263: | |
1504 if (s->h263_msmpeg4) | |
1505 msmpeg4_encode_picture_header(s, picture_number); | |
1506 else if (s->h263_pred) | |
1507 mpeg4_encode_picture_header(s, picture_number); | |
1508 else if (s->h263_rv10) | |
1509 rv10_encode_picture_header(s, picture_number); | |
1510 else | |
1511 h263_encode_picture_header(s, picture_number); | |
1512 break; | |
1513 case FMT_MPEG1: | |
1514 mpeg1_encode_picture_header(s, picture_number); | |
1515 break; | |
1516 } | |
286 | 1517 bits= get_bit_count(&s->pb); |
1518 s->header_bits= bits - s->last_bits; | |
1519 s->last_bits= bits; | |
1520 s->mv_bits=0; | |
1521 s->misc_bits=0; | |
1522 s->i_tex_bits=0; | |
1523 s->p_tex_bits=0; | |
1524 s->i_count=0; | |
1525 s->p_count=0; | |
1526 s->skip_count=0; | |
1527 | |
0 | 1528 /* init last dc values */ |
1529 /* note: quant matrix value (8) is implied here */ | |
1530 s->last_dc[0] = 128; | |
1531 s->last_dc[1] = 128; | |
1532 s->last_dc[2] = 128; | |
1533 s->mb_incr = 1; | |
1534 s->last_mv[0][0][0] = 0; | |
1535 s->last_mv[0][0][1] = 0; | |
1536 | |
162 | 1537 /* Get the GOB height based on picture height */ |
231 | 1538 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
162 | 1539 if (s->height <= 400) |
1540 s->gob_index = 1; | |
1541 else if (s->height <= 800) | |
1542 s->gob_index = 2; | |
1543 else | |
1544 s->gob_index = 4; | |
1545 } | |
268 | 1546 |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1547 s->avg_mb_var = s->avg_mb_var / s->mb_num; |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1548 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1549 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1550 /* Put GOB header based on RTP MTU */ |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1551 /* TODO: Put all this stuff in a separate generic function */ |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1552 if (s->rtp_mode) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1553 if (!mb_y) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1554 s->ptr_lastgob = s->pb.buf; |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1555 s->ptr_last_mb_line = s->pb.buf; |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1556 } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1557 last_gob = h263_encode_gob_header(s, mb_y); |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1558 if (last_gob) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1559 s->first_gob_line = 1; |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1560 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1561 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1562 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1563 |
266 | 1564 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; |
1565 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1566 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1567 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1568 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
1569 s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1570 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
324 | 1571 /*const */int mb_type= s->mb_type[mb_y * s->mb_width + mb_x]; |
294 | 1572 PutBitContext pb; |
1573 int d; | |
1574 int dmin=10000000; | |
1575 int best=0; | |
0 | 1576 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1577 s->mb_x = mb_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1578 s->mb_y = mb_y; |
266 | 1579 s->block_index[0]+=2; |
1580 s->block_index[1]+=2; | |
1581 s->block_index[2]+=2; | |
1582 s->block_index[3]+=2; | |
1583 s->block_index[4]++; | |
1584 s->block_index[5]++; | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1585 |
294 | 1586 if(mb_type & (mb_type-1)){ // more than 1 MB type possible |
1587 pb= s->pb; | |
324 | 1588 s->mv_dir = MV_DIR_FORWARD; |
294 | 1589 if(mb_type&MB_TYPE_INTER){ |
324 | 1590 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; |
295 | 1591 s->mv_type = MV_TYPE_16X16; |
294 | 1592 s->mb_intra= 0; |
324 | 1593 s->mv[0][0][0] = s->p_mv_table[xy][0]; |
1594 s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
294 | 1595 init_put_bits(&s->pb, bit_buf[1], 3000, NULL, NULL); |
1596 s->block= s->inter_block; | |
0 | 1597 |
324 | 1598 encode_mb(s, s->mv[0][0][0], s->mv[0][0][1]); |
294 | 1599 d= get_bit_count(&s->pb); |
1600 if(d<dmin){ | |
1601 flush_put_bits(&s->pb); | |
1602 dmin=d; | |
1603 best_s.mv[0][0][0]= s->mv[0][0][0]; | |
1604 best_s.mv[0][0][1]= s->mv[0][0][1]; | |
1605 best_s.mb_intra= 0; | |
295 | 1606 best_s.mv_type = MV_TYPE_16X16; |
294 | 1607 best_s.pb=s->pb; |
1608 best_s.block= s->block; | |
1609 best=1; | |
1610 for(i=0; i<6; i++) | |
1611 best_s.block_last_index[i]= s->block_last_index[i]; | |
1612 } | |
0 | 1613 } |
295 | 1614 if(mb_type&MB_TYPE_INTER4V){ |
1615 s->mv_type = MV_TYPE_8X8; | |
1616 s->mb_intra= 0; | |
1617 for(i=0; i<4; i++){ | |
1618 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
1619 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
1620 } | |
1621 init_put_bits(&s->pb, bit_buf[2], 3000, NULL, NULL); | |
1622 s->block= s->inter4v_block; | |
1623 | |
324 | 1624 encode_mb(s, 0, 0); |
295 | 1625 d= get_bit_count(&s->pb); |
324 | 1626 if(d<dmin && 0){ |
295 | 1627 flush_put_bits(&s->pb); |
1628 dmin=d; | |
1629 for(i=0; i<4; i++){ | |
1630 best_s.mv[0][i][0] = s->mv[0][i][0]; | |
1631 best_s.mv[0][i][1] = s->mv[0][i][1]; | |
1632 } | |
1633 best_s.mb_intra= 0; | |
1634 best_s.mv_type = MV_TYPE_8X8; | |
1635 best_s.pb=s->pb; | |
1636 best_s.block= s->block; | |
1637 best=2; | |
1638 for(i=0; i<6; i++) | |
1639 best_s.block_last_index[i]= s->block_last_index[i]; | |
1640 } | |
1641 } | |
294 | 1642 if(mb_type&MB_TYPE_INTRA){ |
295 | 1643 s->mv_type = MV_TYPE_16X16; |
294 | 1644 s->mb_intra= 1; |
1645 s->mv[0][0][0] = 0; | |
1646 s->mv[0][0][1] = 0; | |
1647 init_put_bits(&s->pb, bit_buf[0], 3000, NULL, NULL); | |
1648 s->block= s->intra_block; | |
1649 | |
324 | 1650 encode_mb(s, 0, 0); |
294 | 1651 d= get_bit_count(&s->pb); |
1652 if(d<dmin){ | |
1653 flush_put_bits(&s->pb); | |
1654 dmin=d; | |
1655 best_s.mv[0][0][0]= 0; | |
1656 best_s.mv[0][0][1]= 0; | |
1657 best_s.mb_intra= 1; | |
295 | 1658 best_s.mv_type = MV_TYPE_16X16; |
294 | 1659 best_s.pb=s->pb; |
1660 best_s.block= s->block; | |
1661 for(i=0; i<6; i++) | |
1662 best_s.block_last_index[i]= s->block_last_index[i]; | |
1663 best=0; | |
1664 } | |
1665 /* force cleaning of ac/dc if needed ... */ | |
1666 s->mbintra_table[mb_x + mb_y*s->mb_width]=1; | |
1667 } | |
295 | 1668 for(i=0; i<4; i++){ |
1669 s->mv[0][i][0] = best_s.mv[0][i][0]; | |
1670 s->mv[0][i][1] = best_s.mv[0][i][1]; | |
1671 } | |
294 | 1672 s->mb_intra= best_s.mb_intra; |
295 | 1673 s->mv_type= best_s.mv_type; |
294 | 1674 for(i=0; i<6; i++) |
1675 s->block_last_index[i]= best_s.block_last_index[i]; | |
1676 copy_bits(&pb, bit_buf[best], dmin); | |
1677 s->block= best_s.block; | |
1678 s->pb= pb; | |
1679 } else { | |
324 | 1680 int motion_x, motion_y; |
1681 s->mv_type=MV_TYPE_16X16; | |
294 | 1682 // only one MB-Type possible |
324 | 1683 //FIXME convert to swicth() |
294 | 1684 if(mb_type&MB_TYPE_INTRA){ |
324 | 1685 s->mv_dir = MV_DIR_FORWARD; |
294 | 1686 s->mb_intra= 1; |
324 | 1687 motion_x= s->mv[0][0][0] = 0; |
1688 motion_y= s->mv[0][0][1] = 0; | |
1689 }else if(mb_type&MB_TYPE_INTER){ | |
1690 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1691 s->mv_dir = MV_DIR_FORWARD; | |
1692 s->mb_intra= 0; | |
1693 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0]; | |
1694 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
1695 }else if(mb_type&MB_TYPE_DIRECT){ | |
1696 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1697 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; | |
1698 s->mb_intra= 0; | |
1699 motion_x=0; | |
1700 motion_y=0; | |
294 | 1701 s->mv[0][0][0] = 0; |
1702 s->mv[0][0][1] = 0; | |
324 | 1703 s->mv[1][0][0] = 0; |
1704 s->mv[1][0][1] = 0; | |
1705 }else if(mb_type&MB_TYPE_BIDIR){ | |
1706 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1707 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; | |
294 | 1708 s->mb_intra= 0; |
324 | 1709 motion_x=0; |
1710 motion_y=0; | |
1711 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
1712 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
1713 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
1714 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
1715 }else if(mb_type&MB_TYPE_BACKWARD){ | |
1716 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1717 s->mv_dir = MV_DIR_BACKWARD; | |
1718 s->mb_intra= 0; | |
1719 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
1720 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
1721 }else if(mb_type&MB_TYPE_FORWARD){ | |
1722 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1723 s->mv_dir = MV_DIR_FORWARD; | |
1724 s->mb_intra= 0; | |
1725 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
1726 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
1727 // printf(" %d %d ", motion_x, motion_y); | |
1728 }else{ | |
1729 motion_x=motion_y=0; //gcc warning fix | |
1730 printf("illegal MB type\n"); | |
294 | 1731 } |
324 | 1732 encode_mb(s, motion_x, motion_y); |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1733 } |
0 | 1734 |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1735 MPV_decode_mb(s, s->block); |
0 | 1736 } |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
1737 |
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
1738 |
231 | 1739 /* Obtain average GOB size for RTP */ |
1740 if (s->rtp_mode) { | |
1741 if (!mb_y) | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
1742 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
231 | 1743 else if (!(mb_y % s->gob_index)) { |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
1744 s->mb_line_avgsize = (s->mb_line_avgsize + pbBufPtr(&s->pb) - s->ptr_last_mb_line) >> 1; |
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
1745 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
231 | 1746 } |
1747 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
1748 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
1749 s->first_gob_line = 0; | |
1750 } | |
0 | 1751 } |
294 | 1752 emms_c(); |
286 | 1753 |
311 | 1754 if (s->h263_msmpeg4 && s->msmpeg4_version<4 && s->pict_type == I_TYPE) |
208 | 1755 msmpeg4_encode_ext_header(s); |
1756 | |
162 | 1757 //if (s->gob_number) |
1758 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
231 | 1759 |
1760 /* Send the last GOB if RTP */ | |
1761 if (s->rtp_mode) { | |
1762 flush_put_bits(&s->pb); | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
1763 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 1764 /* Call the RTP callback to send the last GOB */ |
1765 if (s->rtp_callback) | |
1766 s->rtp_callback(s->ptr_lastgob, pdif, s->gob_number); | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
1767 s->ptr_lastgob = pbBufPtr(&s->pb); |
231 | 1768 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
1769 } | |
0 | 1770 } |
1771 | |
220 | 1772 static int dct_quantize_c(MpegEncContext *s, |
0 | 1773 DCTELEM *block, int n, |
1774 int qscale) | |
1775 { | |
1776 int i, j, level, last_non_zero, q; | |
1777 const int *qmat; | |
216 | 1778 int minLevel, maxLevel; |
1779 | |
1780 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ | |
1781 /* mpeg4 */ | |
1782 minLevel= -2048; | |
1783 maxLevel= 2047; | |
1784 }else if(s->out_format==FMT_MPEG1){ | |
1785 /* mpeg1 */ | |
1786 minLevel= -255; | |
1787 maxLevel= 255; | |
230 | 1788 }else if(s->out_format==FMT_MJPEG){ |
1789 /* (m)jpeg */ | |
1790 minLevel= -1023; | |
1791 maxLevel= 1023; | |
216 | 1792 }else{ |
1793 /* h263 / msmpeg4 */ | |
1794 minLevel= -128; | |
1795 maxLevel= 127; | |
1796 } | |
0 | 1797 |
1798 av_fdct (block); | |
1799 | |
64 | 1800 /* we need this permutation so that we correct the IDCT |
1801 permutation. will be moved into DCT code */ | |
1802 block_permute(block); | |
1803 | |
0 | 1804 if (s->mb_intra) { |
1805 if (n < 4) | |
1806 q = s->y_dc_scale; | |
1807 else | |
1808 q = s->c_dc_scale; | |
1809 q = q << 3; | |
1810 | |
1811 /* note: block[0] is assumed to be positive */ | |
1812 block[0] = (block[0] + (q >> 1)) / q; | |
1813 i = 1; | |
1814 last_non_zero = 0; | |
1815 if (s->out_format == FMT_H263) { | |
1816 qmat = s->q_non_intra_matrix; | |
1817 } else { | |
1818 qmat = s->q_intra_matrix; | |
1819 } | |
1820 } else { | |
1821 i = 0; | |
1822 last_non_zero = -1; | |
1823 qmat = s->q_non_intra_matrix; | |
1824 } | |
1825 | |
1826 for(;i<64;i++) { | |
1827 j = zigzag_direct[i]; | |
1828 level = block[j]; | |
1829 level = level * qmat[j]; | |
1830 #ifdef PARANOID | |
1831 { | |
1832 static int count = 0; | |
1833 int level1, level2, qmat1; | |
1834 double val; | |
1835 if (qmat == s->q_non_intra_matrix) { | |
1836 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1837 } else { | |
1838 qmat1 = default_intra_matrix[j] * s->qscale; | |
1839 } | |
1840 if (av_fdct != jpeg_fdct_ifast) | |
1841 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1842 else | |
1843 val = ((double)block[j] * 8.0 * 2048.0) / | |
1844 ((double)qmat1 * aanscales[j]); | |
1845 level1 = (int)val; | |
1846 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1847 if (level1 != level2) { | |
1848 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1849 count, level2, level1, block[j], qmat1, qmat[j], | |
1850 val); | |
1851 count++; | |
1852 } | |
1853 | |
1854 } | |
1855 #endif | |
1856 /* XXX: slight error for the low range. Test should be equivalent to | |
1857 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1858 (QMAT_SHIFT - 3))) | |
1859 */ | |
1860 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1861 level) { | |
1862 level = level / (1 << (QMAT_SHIFT - 3)); | |
1863 /* XXX: currently, this code is not optimal. the range should be: | |
1864 mpeg1: -255..255 | |
1865 mpeg2: -2048..2047 | |
1866 h263: -128..127 | |
1867 mpeg4: -2048..2047 | |
1868 */ | |
216 | 1869 if (level > maxLevel) |
1870 level = maxLevel; | |
1871 else if (level < minLevel) | |
1872 level = minLevel; | |
0 | 1873 |
1874 block[j] = level; | |
1875 last_non_zero = i; | |
1876 } else { | |
1877 block[j] = 0; | |
1878 } | |
1879 } | |
1880 return last_non_zero; | |
1881 } | |
1882 | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1883 static void dct_unquantize_mpeg1_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1884 DCTELEM *block, int n, int qscale) |
0 | 1885 { |
200 | 1886 int i, level, nCoeffs; |
0 | 1887 const UINT16 *quant_matrix; |
1888 | |
200 | 1889 if(s->alternate_scan) nCoeffs= 64; |
1890 else nCoeffs= s->block_last_index[n]+1; | |
1891 | |
0 | 1892 if (s->mb_intra) { |
1893 if (n < 4) | |
1894 block[0] = block[0] * s->y_dc_scale; | |
1895 else | |
1896 block[0] = block[0] * s->c_dc_scale; | |
1897 /* XXX: only mpeg1 */ | |
1898 quant_matrix = s->intra_matrix; | |
200 | 1899 for(i=1;i<nCoeffs;i++) { |
1900 int j= zigzag_direct[i]; | |
1901 level = block[j]; | |
0 | 1902 if (level) { |
1903 if (level < 0) { | |
1904 level = -level; | |
200 | 1905 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1906 level = (level - 1) | 1; |
1907 level = -level; | |
1908 } else { | |
200 | 1909 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1910 level = (level - 1) | 1; |
1911 } | |
1912 #ifdef PARANOID | |
1913 if (level < -2048 || level > 2047) | |
1914 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1915 #endif | |
200 | 1916 block[j] = level; |
0 | 1917 } |
1918 } | |
1919 } else { | |
1920 i = 0; | |
1921 quant_matrix = s->non_intra_matrix; | |
217 | 1922 for(;i<nCoeffs;i++) { |
200 | 1923 int j= zigzag_direct[i]; |
1924 level = block[j]; | |
0 | 1925 if (level) { |
1926 if (level < 0) { | |
1927 level = -level; | |
1928 level = (((level << 1) + 1) * qscale * | |
200 | 1929 ((int) (quant_matrix[j]))) >> 4; |
0 | 1930 level = (level - 1) | 1; |
1931 level = -level; | |
1932 } else { | |
1933 level = (((level << 1) + 1) * qscale * | |
200 | 1934 ((int) (quant_matrix[j]))) >> 4; |
0 | 1935 level = (level - 1) | 1; |
1936 } | |
1937 #ifdef PARANOID | |
1938 if (level < -2048 || level > 2047) | |
1939 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1940 #endif | |
200 | 1941 block[j] = level; |
0 | 1942 } |
1943 } | |
1944 } | |
1945 } | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1946 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1947 static void dct_unquantize_h263_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1948 DCTELEM *block, int n, int qscale) |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1949 { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1950 int i, level, qmul, qadd; |
200 | 1951 int nCoeffs; |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1952 |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1953 if (s->mb_intra) { |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1954 if (!s->h263_aic) { |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1955 if (n < 4) |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1956 block[0] = block[0] * s->y_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1957 else |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1958 block[0] = block[0] * s->c_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1959 } |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1960 i = 1; |
200 | 1961 nCoeffs= 64; //does not allways use zigzag table |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1962 } else { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1963 i = 0; |
200 | 1964 nCoeffs= zigzag_end[ s->block_last_index[n] ]; |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1965 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1966 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1967 qmul = s->qscale << 1; |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1968 if (s->h263_aic && s->mb_intra) |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1969 qadd = 0; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1970 else |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1971 qadd = (s->qscale - 1) | 1; |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1972 |
200 | 1973 for(;i<nCoeffs;i++) { |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1974 level = block[i]; |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1975 if (level) { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1976 if (level < 0) { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1977 level = level * qmul - qadd; |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1978 } else { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1979 level = level * qmul + qadd; |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1980 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1981 #ifdef PARANOID |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1982 if (level < -2048 || level > 2047) |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1983 fprintf(stderr, "unquant error %d %d\n", i, level); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1984 #endif |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1985 block[i] = level; |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1986 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1987 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1988 } |
0 | 1989 |
1990 /* rate control */ | |
1991 | |
1992 static void rate_control_init(MpegEncContext *s) | |
1993 { | |
268 | 1994 emms_c(); |
1995 | |
1996 //initial values, they dont really matter as they will be totally different within a few frames | |
1997 s->i_pred.coeff= s->p_pred.coeff= 7.0; | |
1998 s->i_pred.count= s->p_pred.count= 1.0; | |
1999 | |
2000 s->i_pred.decay= s->p_pred.decay= 0.4; | |
2001 | |
2002 // use more bits at the beginning, otherwise high motion at the begin will look like shit | |
2003 s->qsum=100; | |
2004 s->qcount=100; | |
2005 | |
2006 s->short_term_qsum=0.001; | |
2007 s->short_term_qcount=0.001; | |
2008 } | |
2009 | |
2010 static double predict(Predictor *p, double q, double var) | |
2011 { | |
2012 return p->coeff*var / (q*p->count); | |
2013 } | |
2014 | |
2015 static void update_predictor(Predictor *p, double q, double var, double size) | |
2016 { | |
2017 double new_coeff= size*q / (var + 1); | |
2018 if(var<1000) return; | |
2019 /*{ | |
2020 int pred= predict(p, q, var); | |
2021 int error= abs(pred-size); | |
2022 static double sum=0; | |
2023 static int count=0; | |
2024 if(count>5) sum+=error; | |
2025 count++; | |
2026 if(256*256*256*64%count==0){ | |
2027 printf("%d %f %f\n", count, sum/count, p->coeff); | |
2028 } | |
2029 }*/ | |
2030 p->count*= p->decay; | |
2031 p->coeff*= p->decay; | |
2032 p->count++; | |
2033 p->coeff+= new_coeff; | |
0 | 2034 } |
2035 | |
2036 static int rate_estimate_qscale(MpegEncContext *s) | |
2037 { | |
268 | 2038 int qmin= s->qmin; |
2039 int qmax= s->qmax; | |
2040 int rate_q=5; | |
2041 float q; | |
2042 int qscale; | |
2043 float br_compensation; | |
2044 double diff; | |
2045 double short_term_q; | |
2046 double long_term_q; | |
2047 int last_qscale= s->qscale; | |
2048 double fps; | |
2049 INT64 wanted_bits; | |
2050 emms_c(); | |
297 | 2051 |
268 | 2052 fps= (double)s->frame_rate / FRAME_RATE_BASE; |
2053 wanted_bits= s->bit_rate*(double)s->picture_number/fps; | |
2054 | |
2055 | |
2056 if(s->picture_number>2){ | |
2057 /* update predictors */ | |
2058 if(s->last_pict_type == I_TYPE){ | |
2059 //FIXME | |
2060 }else{ //P Frame | |
2061 //printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff); | |
2062 update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits); | |
2063 } | |
2064 } | |
2065 | |
2066 if(s->pict_type == I_TYPE){ | |
2067 //FIXME | |
2068 rate_q= s->qsum/s->qcount; | |
2069 }else{ //P Frame | |
2070 int i; | |
2071 int diff, best_diff=1000000000; | |
2072 for(i=1; i<=31; i++){ | |
2073 diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps; | |
2074 if(diff<0) diff= -diff; | |
2075 if(diff<best_diff){ | |
2076 best_diff= diff; | |
2077 rate_q= i; | |
2078 } | |
2079 } | |
2080 } | |
2081 | |
2082 s->short_term_qsum*=s->qblur; | |
2083 s->short_term_qcount*=s->qblur; | |
2084 | |
2085 s->short_term_qsum+= rate_q; | |
2086 s->short_term_qcount++; | |
2087 short_term_q= s->short_term_qsum/s->short_term_qcount; | |
2088 | |
302 | 2089 long_term_q= s->qsum/s->qcount*(s->total_bits+1)/(wanted_bits+1); //+1 to avoid nan & 0 |
268 | 2090 |
2091 // q= (long_term_q - short_term_q)*s->qcompress + short_term_q; | |
2092 q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q); | |
2093 | |
2094 diff= s->total_bits - wanted_bits; | |
2095 br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance; | |
270 | 2096 if(br_compensation<=0.0) br_compensation=0.001; |
268 | 2097 q/=br_compensation; |
2098 | |
2099 qscale= (int)(q + 0.5); | |
2100 if (qscale<qmin) qscale=qmin; | |
2101 else if(qscale>qmax) qscale=qmax; | |
2102 | |
2103 if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff; | |
2104 else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff; | |
2105 | |
2106 s->qsum+= qscale; | |
2107 s->qcount++; | |
2108 | |
2109 s->last_pict_type= s->pict_type; | |
302 | 2110 //printf("q:%d diff:%d comp:%f rate_q:%d st_q:%f fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation, |
2111 // rate_q, short_term_q, s->mc_mb_var, s->frame_bits); | |
268 | 2112 //printf("%d %d\n", s->bit_rate, (int)fps); |
2113 return qscale; | |
0 | 2114 } |
2115 | |
2116 AVCodec mpeg1video_encoder = { | |
2117 "mpeg1video", | |
2118 CODEC_TYPE_VIDEO, | |
2119 CODEC_ID_MPEG1VIDEO, | |
2120 sizeof(MpegEncContext), | |
2121 MPV_encode_init, | |
2122 MPV_encode_picture, | |
2123 MPV_encode_end, | |
2124 }; | |
2125 | |
2126 AVCodec h263_encoder = { | |
2127 "h263", | |
2128 CODEC_TYPE_VIDEO, | |
2129 CODEC_ID_H263, | |
2130 sizeof(MpegEncContext), | |
2131 MPV_encode_init, | |
2132 MPV_encode_picture, | |
2133 MPV_encode_end, | |
2134 }; | |
2135 | |
2136 AVCodec h263p_encoder = { | |
2137 "h263p", | |
2138 CODEC_TYPE_VIDEO, | |
2139 CODEC_ID_H263P, | |
2140 sizeof(MpegEncContext), | |
2141 MPV_encode_init, | |
2142 MPV_encode_picture, | |
2143 MPV_encode_end, | |
2144 }; | |
2145 | |
2146 AVCodec rv10_encoder = { | |
2147 "rv10", | |
2148 CODEC_TYPE_VIDEO, | |
2149 CODEC_ID_RV10, | |
2150 sizeof(MpegEncContext), | |
2151 MPV_encode_init, | |
2152 MPV_encode_picture, | |
2153 MPV_encode_end, | |
2154 }; | |
2155 | |
2156 AVCodec mjpeg_encoder = { | |
2157 "mjpeg", | |
2158 CODEC_TYPE_VIDEO, | |
2159 CODEC_ID_MJPEG, | |
2160 sizeof(MpegEncContext), | |
2161 MPV_encode_init, | |
2162 MPV_encode_picture, | |
2163 MPV_encode_end, | |
2164 }; | |
2165 | |
71 | 2166 AVCodec mpeg4_encoder = { |
2167 "mpeg4", | |
0 | 2168 CODEC_TYPE_VIDEO, |
71 | 2169 CODEC_ID_MPEG4, |
0 | 2170 sizeof(MpegEncContext), |
2171 MPV_encode_init, | |
2172 MPV_encode_picture, | |
2173 MPV_encode_end, | |
2174 }; | |
2175 | |
307 | 2176 AVCodec msmpeg4v1_encoder = { |
2177 "msmpeg4v1", | |
0 | 2178 CODEC_TYPE_VIDEO, |
307 | 2179 CODEC_ID_MSMPEG4V1, |
0 | 2180 sizeof(MpegEncContext), |
2181 MPV_encode_init, | |
2182 MPV_encode_picture, | |
2183 MPV_encode_end, | |
2184 }; | |
307 | 2185 |
2186 AVCodec msmpeg4v2_encoder = { | |
2187 "msmpeg4v2", | |
2188 CODEC_TYPE_VIDEO, | |
2189 CODEC_ID_MSMPEG4V2, | |
2190 sizeof(MpegEncContext), | |
2191 MPV_encode_init, | |
2192 MPV_encode_picture, | |
2193 MPV_encode_end, | |
2194 }; | |
2195 | |
2196 AVCodec msmpeg4v3_encoder = { | |
2197 "msmpeg4", | |
2198 CODEC_TYPE_VIDEO, | |
2199 CODEC_ID_MSMPEG4V3, | |
2200 sizeof(MpegEncContext), | |
2201 MPV_encode_init, | |
2202 MPV_encode_picture, | |
2203 MPV_encode_end, | |
2204 }; |