Mercurial > libavcodec.hg
annotate mpegvideo.c @ 325:15efd80cf51e libavcodec
mpeg2/mpeg4 dequantizer support (c & mmx)
mpeg1 dequantizer optimizations
author | michaelni |
---|---|
date | Wed, 17 Apr 2002 16:30:31 +0000 |
parents | 9c6f056f0e41 |
children | 6ebb8680885d |
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 * |
325 | 19 * 4MV & hq & b-frame 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); |
325 | 38 static void dct_unquantize_mpeg2_c(MpegEncContext *s, |
39 DCTELEM *block, int n, int qscale); | |
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
|
40 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
|
41 DCTELEM *block, int n, int qscale); |
206 | 42 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
220 | 43 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
206 | 44 |
220 | 45 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c; |
206 | 46 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
47 | |
0 | 48 #define EDGE_WIDTH 16 |
49 | |
50 /* enable all paranoid tests for rounding, overflows, etc... */ | |
51 //#define PARANOID | |
52 | |
53 //#define DEBUG | |
54 | |
321 | 55 |
0 | 56 /* for jpeg fast DCT */ |
57 #define CONST_BITS 14 | |
58 | |
59 static const unsigned short aanscales[64] = { | |
60 /* precomputed values scaled up by 14 bits */ | |
61 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
62 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
63 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
64 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
65 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
66 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
67 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
68 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
69 }; | |
70 | |
71 static UINT8 h263_chroma_roundtab[16] = { | |
72 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
73 }; | |
74 | |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
75 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
|
76 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
|
77 |
321 | 78 extern UINT8 zigzag_end[64]; |
79 | |
0 | 80 /* default motion estimation */ |
321 | 81 int motion_estimation_method = ME_EPZS; |
200 | 82 |
220 | 83 static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale) |
0 | 84 { |
85 int i; | |
86 | |
87 if (av_fdct == jpeg_fdct_ifast) { | |
88 for(i=0;i<64;i++) { | |
89 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
220 | 90 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
91 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
92 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
0 | 93 |
220 | 94 qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
95 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)])); | |
0 | 96 } |
97 } else { | |
98 for(i=0;i<64;i++) { | |
99 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
220 | 100 So 16 <= qscale * quant_matrix[i] <= 7905 |
101 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
102 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
0 | 103 */ |
220 | 104 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
105 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
0 | 106 } |
107 } | |
108 } | |
109 | |
110 /* init common structure for both encoder and decoder */ | |
111 int MPV_common_init(MpegEncContext *s) | |
112 { | |
113 int c_size, i; | |
114 UINT8 *pict; | |
115 | |
312 | 116 s->dct_unquantize_h263 = dct_unquantize_h263_c; |
325 | 117 s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_c; |
118 s->dct_unquantize_mpeg2 = dct_unquantize_mpeg2_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
|
119 |
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
|
120 #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
|
121 MPV_common_init_mmx(s); |
8 | 122 #endif |
312 | 123 //setup default unquantizers (mpeg4 might change it later) |
124 if(s->out_format == FMT_H263) | |
125 s->dct_unquantize = s->dct_unquantize_h263; | |
126 else | |
325 | 127 s->dct_unquantize = s->dct_unquantize_mpeg1; |
312 | 128 |
0 | 129 s->mb_width = (s->width + 15) / 16; |
130 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
|
131 s->mb_num = s->mb_width * s->mb_height; |
0 | 132 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
133 | |
134 for(i=0;i<3;i++) { | |
135 int w, h, shift, pict_start; | |
136 | |
137 w = s->linesize; | |
138 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
139 shift = (i == 0) ? 0 : 1; | |
140 c_size = (w >> shift) * (h >> shift); | |
141 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
142 | |
143 pict = av_mallocz(c_size); | |
144 if (pict == NULL) | |
145 goto fail; | |
146 s->last_picture_base[i] = pict; | |
147 s->last_picture[i] = pict + pict_start; | |
148 | |
149 pict = av_mallocz(c_size); | |
150 if (pict == NULL) | |
151 goto fail; | |
152 s->next_picture_base[i] = pict; | |
153 s->next_picture[i] = pict + pict_start; | |
324 | 154 |
0 | 155 if (s->has_b_frames) { |
156 pict = av_mallocz(c_size); | |
157 if (pict == NULL) | |
158 goto fail; | |
159 s->aux_picture_base[i] = pict; | |
160 s->aux_picture[i] = pict + pict_start; | |
161 } | |
162 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
163 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
164 if (s->encoding) { |
324 | 165 int j; |
166 int mv_table_size= (s->mb_width+2)*(s->mb_height+2); | |
167 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
168 /* Allocate MB type table */ |
239 | 169 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
|
170 if (s->mb_type == NULL) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
171 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
172 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
173 } |
239 | 174 |
175 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16)); | |
176 if (s->mb_var == NULL) { | |
177 perror("malloc"); | |
178 goto fail; | |
179 } | |
324 | 180 |
181 /* Allocate MV tables */ | |
182 s->p_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
183 if (s->p_mv_table == NULL) { | |
184 perror("malloc"); | |
185 goto fail; | |
186 } | |
187 s->last_p_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
188 if (s->last_p_mv_table == NULL) { | |
189 perror("malloc"); | |
190 goto fail; | |
191 } | |
192 s->b_forw_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
193 if (s->b_forw_mv_table == NULL) { | |
194 perror("malloc"); | |
195 goto fail; | |
196 } | |
197 s->b_back_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
198 if (s->b_back_mv_table == NULL) { | |
199 perror("malloc"); | |
200 goto fail; | |
201 } | |
202 s->b_bidir_forw_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
203 if (s->b_bidir_forw_mv_table == NULL) { | |
204 perror("malloc"); | |
205 goto fail; | |
206 } | |
207 s->b_bidir_back_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
208 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
|
209 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
210 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
211 } |
324 | 212 s->b_direct_forw_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); |
213 if (s->b_direct_forw_mv_table == NULL) { | |
214 perror("malloc"); | |
215 goto fail; | |
216 } | |
217 s->b_direct_back_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
218 if (s->b_direct_back_mv_table == NULL) { | |
219 perror("malloc"); | |
220 goto fail; | |
221 } | |
222 s->b_direct_mv_table = av_mallocz(mv_table_size * 2 * sizeof(INT16)); | |
223 if (s->b_direct_mv_table == NULL) { | |
224 perror("malloc"); | |
225 goto fail; | |
226 } | |
227 | |
228 if(s->max_b_frames){ | |
229 for(j=0; j<REORDER_BUFFER_SIZE; j++){ | |
230 int i; | |
231 for(i=0;i<3;i++) { | |
232 int w, h, shift; | |
233 | |
234 w = s->linesize; | |
235 h = s->mb_height * 16; | |
236 shift = (i == 0) ? 0 : 1; | |
237 c_size = (w >> shift) * (h >> shift); | |
238 | |
239 pict = av_mallocz(c_size); | |
240 if (pict == NULL) | |
241 goto fail; | |
242 s->picture_buffer[j][i] = pict; | |
243 } | |
244 } | |
245 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
246 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
247 |
288 | 248 if (s->out_format == FMT_H263 || s->encoding) { |
0 | 249 int size; |
250 /* MV prediction */ | |
251 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
252 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
253 if (s->motion_val == NULL) | |
254 goto fail; | |
255 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
256 } | |
257 | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
258 if (s->h263_pred || s->h263_plus) { |
0 | 259 int y_size, c_size, i, size; |
260 | |
261 /* dc values */ | |
262 | |
263 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
264 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
265 size = y_size + 2 * c_size; | |
266 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
267 if (s->dc_val[0] == NULL) | |
268 goto fail; | |
269 s->dc_val[1] = s->dc_val[0] + y_size; | |
270 s->dc_val[2] = s->dc_val[1] + c_size; | |
271 for(i=0;i<size;i++) | |
272 s->dc_val[0][i] = 1024; | |
273 | |
274 /* ac values */ | |
275 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
276 if (s->ac_val[0] == NULL) | |
277 goto fail; | |
278 s->ac_val[1] = s->ac_val[0] + y_size; | |
279 s->ac_val[2] = s->ac_val[1] + c_size; | |
280 | |
281 /* cbp values */ | |
282 s->coded_block = av_mallocz(y_size); | |
283 if (!s->coded_block) | |
284 goto fail; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
285 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
286 /* 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
|
287 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
|
288 if (!s->mbintra_table) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
289 goto fail; |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
290 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
|
291 } |
0 | 292 /* default structure is frame */ |
293 s->picture_structure = PICT_FRAME; | |
294 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
295 /* init macroblock skip table */ |
324 | 296 s->mbskip_table = av_mallocz(s->mb_num); |
297 if (!s->mbskip_table) | |
298 goto fail; | |
294 | 299 |
300 s->block= s->intra_block; | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
301 |
0 | 302 s->context_initialized = 1; |
303 return 0; | |
304 fail: | |
244 | 305 MPV_common_end(s); |
306 return -1; | |
307 } | |
308 | |
324 | 309 #define CHECK_FREE(p)\ |
310 {\ | |
311 if(p) free(p);\ | |
312 p= NULL;\ | |
313 } | |
314 | |
244 | 315 /* init common structure for both encoder and decoder */ |
316 void MPV_common_end(MpegEncContext *s) | |
317 { | |
318 int i; | |
319 | |
324 | 320 CHECK_FREE(s->mb_type); |
321 CHECK_FREE(s->mb_var); | |
322 CHECK_FREE(s->p_mv_table); | |
323 CHECK_FREE(s->last_p_mv_table); | |
324 CHECK_FREE(s->b_forw_mv_table); | |
325 CHECK_FREE(s->b_back_mv_table); | |
326 CHECK_FREE(s->b_bidir_forw_mv_table); | |
327 CHECK_FREE(s->b_bidir_back_mv_table); | |
328 CHECK_FREE(s->b_direct_forw_mv_table); | |
329 CHECK_FREE(s->b_direct_back_mv_table); | |
330 CHECK_FREE(s->b_direct_mv_table); | |
331 CHECK_FREE(s->motion_val); | |
332 CHECK_FREE(s->dc_val[0]); | |
333 CHECK_FREE(s->ac_val[0]); | |
334 CHECK_FREE(s->coded_block); | |
335 CHECK_FREE(s->mbintra_table); | |
244 | 336 |
324 | 337 CHECK_FREE(s->mbskip_table); |
0 | 338 for(i=0;i<3;i++) { |
324 | 339 int j; |
340 CHECK_FREE(s->last_picture_base[i]); | |
341 CHECK_FREE(s->next_picture_base[i]); | |
342 CHECK_FREE(s->aux_picture_base[i]); | |
343 for(j=0; j<REORDER_BUFFER_SIZE; j++){ | |
344 CHECK_FREE(s->picture_buffer[j][i]); | |
345 } | |
0 | 346 } |
347 s->context_initialized = 0; | |
348 } | |
349 | |
350 /* init video encoder */ | |
351 int MPV_encode_init(AVCodecContext *avctx) | |
352 { | |
353 MpegEncContext *s = avctx->priv_data; | |
60 | 354 int i; |
0 | 355 |
315 | 356 avctx->pix_fmt = PIX_FMT_YUV420P; |
357 | |
0 | 358 s->bit_rate = avctx->bit_rate; |
268 | 359 s->bit_rate_tolerance = avctx->bit_rate_tolerance; |
0 | 360 s->frame_rate = avctx->frame_rate; |
361 s->width = avctx->width; | |
362 s->height = avctx->height; | |
363 s->gop_size = avctx->gop_size; | |
162 | 364 s->rtp_mode = avctx->rtp_mode; |
365 s->rtp_payload_size = avctx->rtp_payload_size; | |
231 | 366 if (avctx->rtp_callback) |
367 s->rtp_callback = avctx->rtp_callback; | |
268 | 368 s->qmin= avctx->qmin; |
369 s->qmax= avctx->qmax; | |
370 s->max_qdiff= avctx->max_qdiff; | |
371 s->qcompress= avctx->qcompress; | |
372 s->qblur= avctx->qblur; | |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
373 s->avctx = avctx; |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
374 s->aspect_ratio_info= avctx->aspect_ratio_info; |
294 | 375 s->flags= avctx->flags; |
324 | 376 s->max_b_frames= avctx->max_b_frames; |
162 | 377 |
0 | 378 if (s->gop_size <= 1) { |
379 s->intra_only = 1; | |
380 s->gop_size = 12; | |
381 } else { | |
382 s->intra_only = 0; | |
383 } | |
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
384 |
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
385 /* ME algorithm */ |
321 | 386 if (avctx->me_method == 0) |
387 /* For compatibility */ | |
388 s->me_method = motion_estimation_method; | |
389 else | |
390 s->me_method = avctx->me_method; | |
391 | |
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
392 /* Fixed QSCALE */ |
0 | 393 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
|
394 |
0 | 395 switch(avctx->codec->id) { |
396 case CODEC_ID_MPEG1VIDEO: | |
397 s->out_format = FMT_MPEG1; | |
398 break; | |
399 case CODEC_ID_MJPEG: | |
400 s->out_format = FMT_MJPEG; | |
401 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
|
402 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
|
403 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
|
404 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
|
405 s->mjpeg_vsample[2] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
406 s->mjpeg_hsample[0] = 2; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
407 s->mjpeg_hsample[1] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
408 s->mjpeg_hsample[2] = 1; |
0 | 409 if (mjpeg_init(s) < 0) |
410 return -1; | |
411 break; | |
412 case CODEC_ID_H263: | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
413 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
|
414 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
0 | 415 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
416 } |
0 | 417 s->out_format = FMT_H263; |
418 break; | |
419 case CODEC_ID_H263P: | |
420 s->out_format = FMT_H263; | |
162 | 421 s->rtp_mode = 1; |
422 s->rtp_payload_size = 1200; | |
0 | 423 s->h263_plus = 1; |
78 | 424 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
425 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
426 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
427 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
428 s->umvplus_dec = 0; |
0 | 429 break; |
430 case CODEC_ID_RV10: | |
431 s->out_format = FMT_H263; | |
432 s->h263_rv10 = 1; | |
433 break; | |
71 | 434 case CODEC_ID_MPEG4: |
0 | 435 s->out_format = FMT_H263; |
436 s->h263_pred = 1; | |
437 s->unrestricted_mv = 1; | |
324 | 438 s->has_b_frames= s->max_b_frames ? 1 : 0; |
0 | 439 break; |
307 | 440 case CODEC_ID_MSMPEG4V1: |
0 | 441 s->out_format = FMT_H263; |
442 s->h263_msmpeg4 = 1; | |
443 s->h263_pred = 1; | |
444 s->unrestricted_mv = 1; | |
307 | 445 s->msmpeg4_version= 1; |
446 break; | |
447 case CODEC_ID_MSMPEG4V2: | |
448 s->out_format = FMT_H263; | |
449 s->h263_msmpeg4 = 1; | |
450 s->h263_pred = 1; | |
451 s->unrestricted_mv = 1; | |
452 s->msmpeg4_version= 2; | |
453 break; | |
454 case CODEC_ID_MSMPEG4V3: | |
455 s->out_format = FMT_H263; | |
456 s->h263_msmpeg4 = 1; | |
457 s->h263_pred = 1; | |
458 s->unrestricted_mv = 1; | |
459 s->msmpeg4_version= 3; | |
0 | 460 break; |
461 default: | |
462 return -1; | |
463 } | |
295 | 464 |
465 if((s->flags&CODEC_FLAG_4MV) && !(s->flags&CODEC_FLAG_HQ)){ | |
466 printf("4MV is currently only supported in HQ mode\n"); | |
467 return -1; | |
468 } | |
0 | 469 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
470 { /* 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
|
471 static int done=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
472 if(!done){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
473 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
474 done=1; |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
475 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
|
476 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
|
477 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
478 for(i=-16; i<16; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
479 default_fcode_tab[i + MAX_MV]= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
480 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
481 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
482 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
483 s->mv_penalty= default_mv_penalty; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
484 s->fcode_tab= default_fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
485 |
0 | 486 if (s->out_format == FMT_H263) |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
487 h263_encode_init(s); |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
488 else if (s->out_format == FMT_MPEG1) |
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
489 mpeg1_encode_init(s); |
0 | 490 |
287 | 491 /* dont use mv_penalty table for crap MV as it would be confused */ |
324 | 492 if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty; |
287 | 493 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
494 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
495 |
0 | 496 /* init */ |
497 if (MPV_common_init(s) < 0) | |
498 return -1; | |
499 | |
60 | 500 /* init default q matrix */ |
501 for(i=0;i<64;i++) { | |
502 s->intra_matrix[i] = default_intra_matrix[i]; | |
503 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
504 } | |
505 | |
0 | 506 /* rate control init */ |
507 rate_control_init(s); | |
508 | |
509 s->picture_number = 0; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
510 s->picture_in_gop_number = 0; |
0 | 511 s->fake_picture_number = 0; |
512 /* motion detector init */ | |
513 s->f_code = 1; | |
324 | 514 s->b_code = 1; |
0 | 515 |
516 return 0; | |
517 } | |
518 | |
519 int MPV_encode_end(AVCodecContext *avctx) | |
520 { | |
521 MpegEncContext *s = avctx->priv_data; | |
522 | |
523 #ifdef STATS | |
524 print_stats(); | |
525 #endif | |
526 MPV_common_end(s); | |
527 if (s->out_format == FMT_MJPEG) | |
528 mjpeg_close(s); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
529 |
0 | 530 return 0; |
531 } | |
532 | |
533 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 534 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 535 { |
536 UINT8 *ptr, *last_line; | |
537 int i; | |
538 | |
539 last_line = buf + (height - 1) * wrap; | |
540 for(i=0;i<w;i++) { | |
541 /* top and bottom */ | |
542 memcpy(buf - (i + 1) * wrap, buf, width); | |
543 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
544 } | |
545 /* left and right */ | |
546 ptr = buf; | |
547 for(i=0;i<height;i++) { | |
548 memset(ptr - w, ptr[0], w); | |
549 memset(ptr + width, ptr[width-1], w); | |
550 ptr += wrap; | |
551 } | |
552 /* corners */ | |
553 for(i=0;i<w;i++) { | |
554 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
555 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
556 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
557 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
558 } | |
559 } | |
560 | |
561 /* generic function for encode/decode called before a frame is coded/decoded */ | |
562 void MPV_frame_start(MpegEncContext *s) | |
563 { | |
564 int i; | |
565 UINT8 *tmp; | |
566 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
567 s->mb_skiped = 0; |
0 | 568 if (s->pict_type == B_TYPE) { |
569 for(i=0;i<3;i++) { | |
570 s->current_picture[i] = s->aux_picture[i]; | |
571 } | |
572 } else { | |
262 | 573 s->last_non_b_pict_type= s->pict_type; |
0 | 574 for(i=0;i<3;i++) { |
575 /* swap next and last */ | |
576 tmp = s->last_picture[i]; | |
577 s->last_picture[i] = s->next_picture[i]; | |
578 s->next_picture[i] = tmp; | |
579 s->current_picture[i] = tmp; | |
580 } | |
581 } | |
582 } | |
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
|
583 |
0 | 584 /* generic function for encode/decode called after a frame has been coded/decoded */ |
585 void MPV_frame_end(MpegEncContext *s) | |
586 { | |
587 /* draw edge for correct motion prediction if outside */ | |
220 | 588 if (s->pict_type != B_TYPE && !s->intra_only) { |
257 | 589 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
|
590 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
|
591 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
|
592 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
|
593 }else{ |
257 | 594 /* mpeg4? / opendivx / xvid */ |
0 | 595 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
596 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
597 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
|
598 } |
0 | 599 } |
207 | 600 emms_c(); |
0 | 601 } |
602 | |
324 | 603 /* reorder input for encoding */ |
604 void reorder_input(MpegEncContext *s, AVPicture *pict) | |
605 { | |
606 int i, j, index; | |
607 | |
608 if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES; | |
609 | |
610 // delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff) | |
611 | |
612 for(j=0; j<REORDER_BUFFER_SIZE-1; j++){ | |
613 s->coded_order[j]= s->coded_order[j+1]; | |
614 } | |
615 s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers | |
616 | |
617 switch(s->input_pict_type){ | |
618 default: | |
619 case I_TYPE: | |
620 case S_TYPE: | |
621 case P_TYPE: | |
622 index= s->max_b_frames - s->b_frames_since_non_b; | |
623 s->b_frames_since_non_b=0; | |
624 break; | |
625 case B_TYPE: | |
626 index= s->max_b_frames + 1; | |
627 s->b_frames_since_non_b++; | |
628 break; | |
629 } | |
630 //printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize); | |
631 if( (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED)) | |
632 && pict->linesize[0] == s->linesize | |
633 && pict->linesize[1] == s->linesize>>1 | |
634 && pict->linesize[2] == s->linesize>>1){ | |
635 //printf("ptr\n"); | |
636 for(i=0; i<3; i++){ | |
637 s->coded_order[index].picture[i]= pict->data[i]; | |
638 } | |
639 }else{ | |
640 //printf("copy\n"); | |
641 for(i=0; i<3; i++){ | |
642 uint8_t *src = pict->data[i]; | |
643 uint8_t *dest; | |
644 int src_wrap = pict->linesize[i]; | |
645 int dest_wrap = s->linesize; | |
646 int w = s->width; | |
647 int h = s->height; | |
648 | |
649 if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering | |
650 else dest= s->picture_buffer[s->picture_buffer_index][i]; | |
651 | |
652 if (i >= 1) { | |
653 dest_wrap >>= 1; | |
654 w >>= 1; | |
655 h >>= 1; | |
656 } | |
657 | |
658 s->coded_order[index].picture[i]= dest; | |
659 for(j=0;j<h;j++) { | |
660 memcpy(dest, src, w); | |
661 dest += dest_wrap; | |
662 src += src_wrap; | |
663 } | |
664 } | |
665 if(index!=0){ | |
666 s->picture_buffer_index++; | |
667 if(s->picture_buffer_index >= REORDER_BUFFER_SIZE-1) s->picture_buffer_index=0; | |
668 } | |
669 } | |
670 s->coded_order[index].pict_type = s->input_pict_type; | |
671 s->coded_order[index].qscale = s->input_qscale; | |
672 s->coded_order[index].force_type= s->force_input_type; | |
673 s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number; | |
674 s->coded_order[index].picture_number= s->input_picture_number; | |
675 | |
676 for(i=0; i<3; i++){ | |
677 s->new_picture[i]= s->coded_order[0].picture[i]; | |
678 } | |
679 } | |
680 | |
0 | 681 int MPV_encode_picture(AVCodecContext *avctx, |
682 unsigned char *buf, int buf_size, void *data) | |
683 { | |
684 MpegEncContext *s = avctx->priv_data; | |
685 AVPicture *pict = data; | |
686 | |
324 | 687 s->input_qscale = avctx->quality; |
0 | 688 |
689 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
690 | |
324 | 691 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
|
692 (avctx->key_frame ? I_TYPE : P_TYPE) : 0; |
0 | 693 if (!s->intra_only) { |
694 /* first picture of GOP is intra */ | |
324 | 695 if (s->input_picture_in_gop_number % s->gop_size==0 || s->force_input_type==I_TYPE){ |
696 s->input_picture_in_gop_number=0; | |
697 s->input_pict_type = I_TYPE; | |
698 }else if(s->max_b_frames==0){ | |
699 s->input_pict_type = P_TYPE; | |
700 }else{ | |
701 if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ | |
702 s->input_pict_type = B_TYPE; | |
703 else | |
704 s->input_pict_type = P_TYPE; | |
705 } | |
0 | 706 } else { |
324 | 707 s->input_pict_type = I_TYPE; |
0 | 708 } |
709 | |
324 | 710 reorder_input(s, pict); |
711 | |
712 /* output? */ | |
713 if(s->coded_order[0].picture[0]){ | |
714 | |
715 s->pict_type= s->coded_order[0].pict_type; | |
716 if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */ | |
717 s->qscale= s->coded_order[0].qscale; | |
718 s->force_type= s->coded_order[0].force_type; | |
719 s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number; | |
720 s->picture_number= s->coded_order[0].picture_number; | |
721 | |
722 MPV_frame_start(s); | |
286 | 723 |
324 | 724 encode_picture(s, s->picture_number); |
725 avctx->key_frame = (s->pict_type == I_TYPE); | |
726 avctx->header_bits = s->header_bits; | |
727 avctx->mv_bits = s->mv_bits; | |
728 avctx->misc_bits = s->misc_bits; | |
729 avctx->i_tex_bits = s->i_tex_bits; | |
730 avctx->p_tex_bits = s->p_tex_bits; | |
731 avctx->i_count = s->i_count; | |
732 avctx->p_count = s->p_count; | |
733 avctx->skip_count = s->skip_count; | |
0 | 734 |
324 | 735 MPV_frame_end(s); |
736 | |
737 if (s->out_format == FMT_MJPEG) | |
738 mjpeg_picture_trailer(s); | |
739 | |
740 avctx->quality = s->qscale; | |
741 } | |
742 | |
743 s->input_picture_number++; | |
744 s->input_picture_in_gop_number++; | |
0 | 745 |
746 flush_put_bits(&s->pb); | |
268 | 747 s->last_frame_bits= s->frame_bits; |
748 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8; | |
749 s->total_bits += s->frame_bits; | |
286 | 750 avctx->frame_bits = s->frame_bits; |
751 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n", | |
752 //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
|
753 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
754 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
|
755 /* 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
|
756 /* 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
|
757 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
|
758 pict->linesize, s->linesize, avctx); |
324 | 759 // 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
|
760 } |
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
|
761 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 762 } |
763 | |
764 static inline int clip(int a, int amin, int amax) | |
765 { | |
766 if (a < amin) | |
767 return amin; | |
768 else if (a > amax) | |
769 return amax; | |
770 else | |
771 return a; | |
772 } | |
773 | |
255 | 774 static inline void gmc1_motion(MpegEncContext *s, |
775 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
776 int dest_offset, | |
777 UINT8 **ref_picture, int src_offset, | |
778 int h) | |
779 { | |
780 UINT8 *ptr; | |
324 | 781 int offset, src_x, src_y, linesize; |
255 | 782 int motion_x, motion_y; |
783 | |
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
784 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n"); |
255 | 785 motion_x= s->sprite_offset[0][0]; |
786 motion_y= s->sprite_offset[0][1]; | |
787 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
788 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
789 motion_x<<=(3-s->sprite_warping_accuracy); | |
790 motion_y<<=(3-s->sprite_warping_accuracy); | |
791 src_x = clip(src_x, -16, s->width); | |
792 if (src_x == s->width) | |
793 motion_x =0; | |
794 src_y = clip(src_y, -16, s->height); | |
795 if (src_y == s->height) | |
796 motion_y =0; | |
797 | |
798 linesize = s->linesize; | |
799 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
800 | |
801 dest_y+=dest_offset; | |
802 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
803 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
804 | |
805 motion_x= s->sprite_offset[1][0]; | |
806 motion_y= s->sprite_offset[1][1]; | |
807 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
808 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
809 motion_x<<=(3-s->sprite_warping_accuracy); | |
810 motion_y<<=(3-s->sprite_warping_accuracy); | |
811 src_x = clip(src_x, -8, s->width>>1); | |
812 if (src_x == s->width>>1) | |
813 motion_x =0; | |
814 src_y = clip(src_y, -8, s->height>>1); | |
815 if (src_y == s->height>>1) | |
816 motion_y =0; | |
817 | |
818 offset = (src_y * linesize>>1) + src_x + (src_offset>>1); | |
819 ptr = ref_picture[1] + offset; | |
820 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
821 ptr = ref_picture[2] + offset; | |
822 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
823 | |
824 return; | |
825 } | |
826 | |
0 | 827 /* apply one mpeg motion vector to the three components */ |
828 static inline void mpeg_motion(MpegEncContext *s, | |
829 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
830 int dest_offset, | |
831 UINT8 **ref_picture, int src_offset, | |
832 int field_based, op_pixels_func *pix_op, | |
833 int motion_x, int motion_y, int h) | |
834 { | |
835 UINT8 *ptr; | |
836 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
255 | 837 if(s->quarter_sample) |
838 { | |
839 motion_x>>=1; | |
840 motion_y>>=1; | |
841 } | |
324 | 842 |
0 | 843 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
844 src_x = s->mb_x * 16 + (motion_x >> 1); | |
845 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
846 | |
847 /* WARNING: do no forget half pels */ | |
848 height = s->height >> field_based; | |
849 src_x = clip(src_x, -16, s->width); | |
850 if (src_x == s->width) | |
851 dxy &= ~1; | |
852 src_y = clip(src_y, -16, height); | |
853 if (src_y == height) | |
854 dxy &= ~2; | |
855 linesize = s->linesize << field_based; | |
856 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
857 dest_y += dest_offset; | |
858 pix_op[dxy](dest_y, ptr, linesize, h); | |
859 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
860 | |
861 if (s->out_format == FMT_H263) { | |
862 dxy = 0; | |
863 if ((motion_x & 3) != 0) | |
864 dxy |= 1; | |
865 if ((motion_y & 3) != 0) | |
866 dxy |= 2; | |
867 mx = motion_x >> 2; | |
868 my = motion_y >> 2; | |
869 } else { | |
870 mx = motion_x / 2; | |
871 my = motion_y / 2; | |
872 dxy = ((my & 1) << 1) | (mx & 1); | |
873 mx >>= 1; | |
874 my >>= 1; | |
875 } | |
876 | |
877 src_x = s->mb_x * 8 + mx; | |
878 src_y = s->mb_y * (8 >> field_based) + my; | |
879 src_x = clip(src_x, -8, s->width >> 1); | |
880 if (src_x == (s->width >> 1)) | |
881 dxy &= ~1; | |
882 src_y = clip(src_y, -8, height >> 1); | |
883 if (src_y == (height >> 1)) | |
884 dxy &= ~2; | |
885 | |
886 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
887 ptr = ref_picture[1] + offset; | |
888 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
889 ptr = ref_picture[2] + offset; | |
890 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
891 } | |
892 | |
255 | 893 static inline void qpel_motion(MpegEncContext *s, |
894 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
895 int dest_offset, | |
896 UINT8 **ref_picture, int src_offset, | |
897 int field_based, op_pixels_func *pix_op, | |
898 qpel_mc_func *qpix_op, | |
899 int motion_x, int motion_y, int h) | |
900 { | |
901 UINT8 *ptr; | |
902 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
903 | |
904 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
905 src_x = s->mb_x * 16 + (motion_x >> 2); | |
906 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
907 | |
908 height = s->height >> field_based; | |
909 src_x = clip(src_x, -16, s->width); | |
910 if (src_x == s->width) | |
911 dxy &= ~3; | |
912 src_y = clip(src_y, -16, height); | |
913 if (src_y == height) | |
914 dxy &= ~12; | |
915 linesize = s->linesize << field_based; | |
916 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
917 dest_y += dest_offset; | |
918 //printf("%d %d %d\n", src_x, src_y, dxy); | |
919 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3); | |
920 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3); | |
921 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3); | |
922 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3); | |
923 | |
924 mx= (motion_x>>1) | (motion_x&1); | |
925 my= (motion_y>>1) | (motion_y&1); | |
926 | |
927 dxy = 0; | |
928 if ((mx & 3) != 0) | |
929 dxy |= 1; | |
930 if ((my & 3) != 0) | |
931 dxy |= 2; | |
932 mx = mx >> 2; | |
933 my = my >> 2; | |
934 | |
935 src_x = s->mb_x * 8 + mx; | |
936 src_y = s->mb_y * (8 >> field_based) + my; | |
937 src_x = clip(src_x, -8, s->width >> 1); | |
938 if (src_x == (s->width >> 1)) | |
939 dxy &= ~1; | |
940 src_y = clip(src_y, -8, height >> 1); | |
941 if (src_y == (height >> 1)) | |
942 dxy &= ~2; | |
943 | |
944 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
945 ptr = ref_picture[1] + offset; | |
946 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
947 ptr = ref_picture[2] + offset; | |
948 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
949 } | |
950 | |
951 | |
0 | 952 static inline void MPV_motion(MpegEncContext *s, |
953 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
954 int dir, UINT8 **ref_picture, | |
255 | 955 op_pixels_func *pix_op, qpel_mc_func *qpix_op) |
0 | 956 { |
957 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
958 int mb_x, mb_y, i; | |
959 UINT8 *ptr, *dest; | |
960 | |
961 mb_x = s->mb_x; | |
962 mb_y = s->mb_y; | |
963 | |
964 switch(s->mv_type) { | |
965 case MV_TYPE_16X16: | |
255 | 966 if(s->mcsel){ |
967 #if 0 | |
968 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
969 ref_picture, 0, | |
970 0, pix_op, | |
971 s->sprite_offset[0][0]>>3, | |
972 s->sprite_offset[0][1]>>3, | |
973 16); | |
974 #else | |
975 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
976 ref_picture, 0, | |
977 16); | |
978 #endif | |
262 | 979 }else if(s->quarter_sample && dir==0){ //FIXME |
255 | 980 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, |
981 ref_picture, 0, | |
982 0, pix_op, qpix_op, | |
983 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
984 }else{ | |
985 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
986 ref_picture, 0, | |
987 0, pix_op, | |
988 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
989 } | |
0 | 990 break; |
991 case MV_TYPE_8X8: | |
992 for(i=0;i<4;i++) { | |
993 motion_x = s->mv[dir][i][0]; | |
994 motion_y = s->mv[dir][i][1]; | |
995 | |
996 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
997 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
295 | 998 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; |
0 | 999 |
1000 /* WARNING: do no forget half pels */ | |
1001 src_x = clip(src_x, -16, s->width); | |
1002 if (src_x == s->width) | |
1003 dxy &= ~1; | |
1004 src_y = clip(src_y, -16, s->height); | |
1005 if (src_y == s->height) | |
1006 dxy &= ~2; | |
1007 | |
1008 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
1009 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
1010 pix_op[dxy](dest, ptr, s->linesize, 8); | |
1011 } | |
1012 /* In case of 8X8, we construct a single chroma motion vector | |
1013 with a special rounding */ | |
1014 mx = 0; | |
1015 my = 0; | |
1016 for(i=0;i<4;i++) { | |
1017 mx += s->mv[dir][i][0]; | |
1018 my += s->mv[dir][i][1]; | |
1019 } | |
1020 if (mx >= 0) | |
1021 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1022 else { | |
1023 mx = -mx; | |
1024 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1025 } | |
1026 if (my >= 0) | |
1027 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1028 else { | |
1029 my = -my; | |
1030 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1031 } | |
1032 dxy = ((my & 1) << 1) | (mx & 1); | |
1033 mx >>= 1; | |
1034 my >>= 1; | |
1035 | |
1036 src_x = mb_x * 8 + mx; | |
1037 src_y = mb_y * 8 + my; | |
1038 src_x = clip(src_x, -8, s->width/2); | |
1039 if (src_x == s->width/2) | |
1040 dxy &= ~1; | |
1041 src_y = clip(src_y, -8, s->height/2); | |
1042 if (src_y == s->height/2) | |
1043 dxy &= ~2; | |
1044 | |
1045 offset = (src_y * (s->linesize >> 1)) + src_x; | |
1046 ptr = ref_picture[1] + offset; | |
1047 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
1048 ptr = ref_picture[2] + offset; | |
1049 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
1050 break; | |
1051 case MV_TYPE_FIELD: | |
1052 if (s->picture_structure == PICT_FRAME) { | |
1053 /* top field */ | |
1054 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
1055 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
1056 1, pix_op, | |
1057 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
1058 /* bottom field */ | |
1059 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
1060 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
1061 1, pix_op, | |
1062 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
1063 } else { | |
1064 | |
1065 | |
1066 } | |
1067 break; | |
1068 } | |
1069 } | |
1070 | |
1071 | |
1072 /* put block[] to dest[] */ | |
1073 static inline void put_dct(MpegEncContext *s, | |
1074 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
1075 { | |
1076 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
|
1077 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
|
1078 ff_idct (block); |
0 | 1079 put_pixels_clamped(block, dest, line_size); |
1080 } | |
1081 | |
1082 /* add block[] to dest[] */ | |
1083 static inline void add_dct(MpegEncContext *s, | |
1084 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
1085 { | |
1086 if (s->block_last_index[i] >= 0) { | |
1087 if (!s->mpeg2) | |
206 | 1088 if(s->encoding || (!s->h263_msmpeg4)) |
200 | 1089 s->dct_unquantize(s, block, i, s->qscale); |
324 | 1090 |
19
82d4c9be9873
MMX/MMXEXT iDCT support, using external functions currently defined in libmpeg2
arpi_esp
parents:
18
diff
changeset
|
1091 ff_idct (block); |
0 | 1092 add_pixels_clamped(block, dest, line_size); |
1093 } | |
1094 } | |
1095 | |
1096 /* generic function called after a macroblock has been parsed by the | |
1097 decoder or after it has been encoded by the encoder. | |
1098 | |
1099 Important variables used: | |
1100 s->mb_intra : true if intra macroblock | |
1101 s->mv_dir : motion vector direction | |
1102 s->mv_type : motion vector type | |
1103 s->mv : motion vector | |
1104 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
1105 */ | |
1106 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
1107 { | |
244 | 1108 int mb_x, mb_y; |
0 | 1109 int dct_linesize, dct_offset; |
1110 op_pixels_func *op_pix; | |
255 | 1111 qpel_mc_func *op_qpix; |
0 | 1112 |
1113 mb_x = s->mb_x; | |
1114 mb_y = s->mb_y; | |
1115 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1116 #ifdef FF_POSTPROCESS |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1117 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1118 //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
|
1119 #endif |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1120 |
0 | 1121 /* update DC predictors for P macroblocks */ |
1122 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1123 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
|
1124 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
|
1125 { |
244 | 1126 int wrap, xy, v; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1127 s->mbintra_table[mb_x + mb_y*s->mb_width]=0; |
0 | 1128 wrap = 2 * s->mb_width + 2; |
244 | 1129 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 1130 v = 1024; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1131 |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1132 s->dc_val[0][xy] = v; |
244 | 1133 s->dc_val[0][xy + 1] = v; |
1134 s->dc_val[0][xy + wrap] = v; | |
1135 s->dc_val[0][xy + 1 + wrap] = v; | |
0 | 1136 /* ac pred */ |
244 | 1137 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16)); |
1138 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16)); | |
1139 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16)); | |
1140 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16)); | |
0 | 1141 if (s->h263_msmpeg4) { |
244 | 1142 s->coded_block[xy] = 0; |
1143 s->coded_block[xy + 1] = 0; | |
1144 s->coded_block[xy + wrap] = 0; | |
1145 s->coded_block[xy + 1 + wrap] = 0; | |
0 | 1146 } |
1147 /* chroma */ | |
1148 wrap = s->mb_width + 2; | |
244 | 1149 xy = mb_x + 1 + (mb_y + 1) * wrap; |
1150 s->dc_val[1][xy] = v; | |
1151 s->dc_val[2][xy] = v; | |
0 | 1152 /* ac pred */ |
244 | 1153 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); |
1154 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
|
1155 } |
0 | 1156 } else { |
1157 s->last_dc[0] = 128 << s->intra_dc_precision; | |
1158 s->last_dc[1] = 128 << s->intra_dc_precision; | |
1159 s->last_dc[2] = 128 << s->intra_dc_precision; | |
1160 } | |
1161 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1162 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
|
1163 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
|
1164 |
280 | 1165 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */ |
324 | 1166 if (s->out_format == FMT_H263) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here |
262 | 1167 if(s->pict_type!=B_TYPE){ |
244 | 1168 int xy, wrap, motion_x, motion_y; |
0 | 1169 |
1170 wrap = 2 * s->mb_width + 2; | |
244 | 1171 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 1172 if (s->mb_intra) { |
1173 motion_x = 0; | |
1174 motion_y = 0; | |
1175 goto motion_init; | |
1176 } else if (s->mv_type == MV_TYPE_16X16) { | |
1177 motion_x = s->mv[0][0][0]; | |
1178 motion_y = s->mv[0][0][1]; | |
1179 motion_init: | |
1180 /* no update if 8X8 because it has been done during parsing */ | |
244 | 1181 s->motion_val[xy][0] = motion_x; |
1182 s->motion_val[xy][1] = motion_y; | |
1183 s->motion_val[xy + 1][0] = motion_x; | |
1184 s->motion_val[xy + 1][1] = motion_y; | |
1185 s->motion_val[xy + wrap][0] = motion_x; | |
1186 s->motion_val[xy + wrap][1] = motion_y; | |
1187 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
1188 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
0 | 1189 } |
262 | 1190 } |
0 | 1191 } |
1192 | |
324 | 1193 if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) { |
0 | 1194 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
|
1195 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1196 |
324 | 1197 /* avoid copy if macroblock skipped in last frame too |
1198 dont touch it for B-frames as they need the skip info from the next p-frame */ | |
1199 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
|
1200 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
|
1201 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1202 s->mb_skiped = 0; |
324 | 1203 /* if previous was skipped too, then nothing to do ! |
1204 skip only during decoding as we might trash the buffers during encoding a bit */ | |
1205 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
|
1206 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1207 *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
|
1208 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1209 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1210 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1211 } |
0 | 1212 |
1213 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
1214 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1215 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1216 | |
1217 if (s->interlaced_dct) { | |
1218 dct_linesize = s->linesize * 2; | |
1219 dct_offset = s->linesize; | |
1220 } else { | |
1221 dct_linesize = s->linesize; | |
1222 dct_offset = s->linesize * 8; | |
1223 } | |
1224 | |
1225 if (!s->mb_intra) { | |
1226 /* motion handling */ | |
324 | 1227 if((s->flags&CODEC_FLAG_HQ) || (!s->encoding)){ |
1228 if (!s->no_rounding){ | |
1229 op_pix = put_pixels_tab; | |
1230 op_qpix= qpel_mc_rnd_tab; | |
1231 }else{ | |
1232 op_pix = put_no_rnd_pixels_tab; | |
1233 op_qpix= qpel_mc_no_rnd_tab; | |
1234 } | |
0 | 1235 |
324 | 1236 if (s->mv_dir & MV_DIR_FORWARD) { |
1237 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
1238 if (!s->no_rounding) | |
1239 op_pix = avg_pixels_tab; | |
1240 else | |
1241 op_pix = avg_no_rnd_pixels_tab; | |
1242 } | |
1243 if (s->mv_dir & MV_DIR_BACKWARD) { | |
1244 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
1245 } | |
0 | 1246 } |
1247 | |
1248 /* add dct residue */ | |
1249 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
1250 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1251 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1252 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1253 | |
57 | 1254 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1255 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1256 } else { |
1257 /* dct only in intra block */ | |
1258 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
1259 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1260 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1261 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1262 | |
57 | 1263 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1264 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1265 } |
1266 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1267 the_end: |
294 | 1268 emms_c(); //FIXME remove |
1269 } | |
1270 | |
324 | 1271 |
1272 static void encode_mb(MpegEncContext *s, int motion_x, int motion_y) | |
294 | 1273 { |
1274 const int mb_x= s->mb_x; | |
1275 const int mb_y= s->mb_y; | |
1276 int i; | |
324 | 1277 #if 0 |
1278 if (s->interlaced_dct) { | |
1279 dct_linesize = s->linesize * 2; | |
1280 dct_offset = s->linesize; | |
1281 } else { | |
1282 dct_linesize = s->linesize; | |
1283 dct_offset = s->linesize * 8; | |
1284 } | |
1285 #endif | |
294 | 1286 |
324 | 1287 if (s->mb_intra) { |
1288 UINT8 *ptr; | |
1289 int wrap; | |
294 | 1290 |
324 | 1291 wrap = s->linesize; |
1292 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
1293 get_pixels(s->block[0], ptr , wrap); | |
1294 get_pixels(s->block[1], ptr + 8, wrap); | |
1295 get_pixels(s->block[2], ptr + 8 * wrap , wrap); | |
1296 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); | |
294 | 1297 |
324 | 1298 wrap >>=1; |
1299 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
1300 get_pixels(s->block[4], ptr, wrap); | |
294 | 1301 |
324 | 1302 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; |
1303 get_pixels(s->block[5], ptr, wrap); | |
1304 }else{ | |
1305 op_pixels_func *op_pix; | |
1306 qpel_mc_func *op_qpix; | |
1307 UINT8 *dest_y, *dest_cb, *dest_cr; | |
1308 UINT8 *ptr; | |
1309 int wrap; | |
294 | 1310 |
324 | 1311 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize ) + mb_x * 16; |
1312 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1313 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1314 | |
1315 if (!s->no_rounding){ | |
1316 op_pix = put_pixels_tab; | |
1317 op_qpix= qpel_mc_rnd_tab; | |
295 | 1318 }else{ |
324 | 1319 op_pix = put_no_rnd_pixels_tab; |
1320 op_qpix= qpel_mc_no_rnd_tab; | |
1321 } | |
295 | 1322 |
324 | 1323 if (s->mv_dir & MV_DIR_FORWARD) { |
1324 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
1325 if (!s->no_rounding) | |
1326 op_pix = avg_pixels_tab; | |
1327 else | |
1328 op_pix = avg_no_rnd_pixels_tab; | |
1329 } | |
1330 if (s->mv_dir & MV_DIR_BACKWARD) { | |
1331 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
1332 } | |
1333 wrap = s->linesize; | |
1334 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
1335 diff_pixels(s->block[0], ptr , dest_y , wrap); | |
1336 diff_pixels(s->block[1], ptr + 8, dest_y + 8, wrap); | |
1337 diff_pixels(s->block[2], ptr + 8 * wrap , dest_y + 8 * wrap , wrap); | |
1338 diff_pixels(s->block[3], ptr + 8 * wrap + 8, dest_y + 8 * wrap + 8, wrap); | |
295 | 1339 |
324 | 1340 wrap >>=1; |
1341 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
1342 diff_pixels(s->block[4], ptr, dest_cb, wrap); | |
1343 | |
1344 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; | |
1345 diff_pixels(s->block[5], ptr, dest_cr, wrap); | |
294 | 1346 } |
1347 | |
1348 #if 0 | |
1349 { | |
1350 float adap_parm; | |
1351 | |
1352 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / | |
1353 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); | |
1354 | |
1355 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", | |
1356 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', | |
1357 s->qscale, adap_parm, s->qscale*adap_parm, | |
1358 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); | |
1359 } | |
1360 #endif | |
1361 /* DCT & quantize */ | |
1362 if (s->h263_msmpeg4) { | |
1363 msmpeg4_dc_scale(s); | |
1364 } else if (s->h263_pred) { | |
1365 h263_dc_scale(s); | |
1366 } else { | |
1367 /* default quantization values */ | |
1368 s->y_dc_scale = 8; | |
1369 s->c_dc_scale = 8; | |
1370 } | |
1371 for(i=0;i<6;i++) { | |
1372 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); | |
1373 } | |
1374 | |
1375 /* huffman encode */ | |
1376 switch(s->out_format) { | |
1377 case FMT_MPEG1: | |
1378 mpeg1_encode_mb(s, s->block, motion_x, motion_y); | |
1379 break; | |
1380 case FMT_H263: | |
1381 if (s->h263_msmpeg4) | |
1382 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1383 else if(s->h263_pred) | |
1384 mpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1385 else | |
1386 h263_encode_mb(s, s->block, motion_x, motion_y); | |
1387 break; | |
1388 case FMT_MJPEG: | |
1389 mjpeg_encode_mb(s, s->block); | |
1390 break; | |
1391 } | |
1392 } | |
1393 | |
1394 static void copy_bits(PutBitContext *pb, UINT8 *src, int length) | |
1395 { | |
1396 int bytes= length>>3; | |
1397 int bits= length&7; | |
1398 int i; | |
1399 | |
1400 for(i=0; i<bytes; i++) put_bits(pb, 8, src[i]); | |
1401 put_bits(pb, bits, src[i]>>(8-bits)); | |
0 | 1402 } |
1403 | |
1404 static void encode_picture(MpegEncContext *s, int picture_number) | |
1405 { | |
294 | 1406 int mb_x, mb_y, last_gob, pdif = 0; |
1407 int i; | |
286 | 1408 int bits; |
294 | 1409 MpegEncContext best_s; |
1410 UINT8 bit_buf[4][3000]; //FIXME check that this is ALLWAYS large enogh for a MB | |
0 | 1411 |
1412 s->picture_number = picture_number; | |
268 | 1413 |
294 | 1414 s->block_wrap[0]= |
1415 s->block_wrap[1]= | |
1416 s->block_wrap[2]= | |
1417 s->block_wrap[3]= s->mb_width*2 + 2; | |
1418 s->block_wrap[4]= | |
1419 s->block_wrap[5]= s->mb_width + 2; | |
1420 | |
268 | 1421 s->last_mc_mb_var = s->mc_mb_var; |
1422 /* Reset the average MB variance */ | |
1423 s->avg_mb_var = 0; | |
1424 s->mc_mb_var = 0; | |
324 | 1425 |
268 | 1426 /* Estimate motion for every MB */ |
324 | 1427 if(s->pict_type != I_TYPE){ |
1428 // int16_t (*tmp)[2]= s->p_mv_table; | |
1429 // s->p_mv_table= s->last_mv_table; | |
1430 // s->last_mv_table= s->mv_table; | |
1431 | |
294 | 1432 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
1433 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; | |
1434 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1435 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1436 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1437 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
1438 s->mb_x = mb_x; | |
1439 s->mb_y = mb_y; | |
1440 s->block_index[0]+=2; | |
1441 s->block_index[1]+=2; | |
1442 s->block_index[2]+=2; | |
1443 s->block_index[3]+=2; | |
268 | 1444 |
294 | 1445 /* compute motion vector & mb_type and store in context */ |
324 | 1446 if(s->pict_type==B_TYPE) |
1447 ff_estimate_b_frame_motion(s, mb_x, mb_y); | |
1448 else | |
1449 ff_estimate_p_frame_motion(s, mb_x, mb_y); | |
294 | 1450 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER; |
268 | 1451 } |
1452 } | |
294 | 1453 emms_c(); |
324 | 1454 }else if(s->pict_type == I_TYPE){ |
294 | 1455 /* I-Frame */ |
1456 //FIXME do we need to zero them? | |
1457 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2); | |
324 | 1458 memset(s->p_mv_table , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2); |
294 | 1459 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
268 | 1460 } |
1461 | |
324 | 1462 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 |
1463 // FIXME b-frames & scene change detection | |
1464 s->input_pict_type= I_TYPE; | |
271 | 1465 s->pict_type= I_TYPE; |
324 | 1466 s->input_picture_in_gop_number=0; |
294 | 1467 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
1468 //printf("Scene change detected, encoding as I Frame\n"); | |
271 | 1469 } |
324 | 1470 |
1471 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) | |
1472 s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER); | |
1473 ff_fix_long_p_mvs(s); | |
1474 if(s->pict_type==B_TYPE){ | |
1475 s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD); | |
1476 s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD); | |
1477 //FIXME if BIDIR != for&back | |
1478 ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD |MB_TYPE_BIDIR); | |
1479 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
|
1480 } |
324 | 1481 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1482 //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
|
1483 |
271 | 1484 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var); |
1485 | |
0 | 1486 if (!s->fixed_qscale) |
1487 s->qscale = rate_estimate_qscale(s); | |
1488 | |
1489 /* precompute matrix */ | |
1490 if (s->out_format == FMT_MJPEG) { | |
1491 /* for mjpeg, we do include qscale in the matrix */ | |
1492 s->intra_matrix[0] = default_intra_matrix[0]; | |
1493 for(i=1;i<64;i++) | |
1494 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
220 | 1495 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
0 | 1496 } else { |
220 | 1497 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
1498 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); | |
0 | 1499 } |
1500 | |
286 | 1501 s->last_bits= get_bit_count(&s->pb); |
0 | 1502 switch(s->out_format) { |
1503 case FMT_MJPEG: | |
1504 mjpeg_picture_header(s); | |
1505 break; | |
1506 case FMT_H263: | |
1507 if (s->h263_msmpeg4) | |
1508 msmpeg4_encode_picture_header(s, picture_number); | |
1509 else if (s->h263_pred) | |
1510 mpeg4_encode_picture_header(s, picture_number); | |
1511 else if (s->h263_rv10) | |
1512 rv10_encode_picture_header(s, picture_number); | |
1513 else | |
1514 h263_encode_picture_header(s, picture_number); | |
1515 break; | |
1516 case FMT_MPEG1: | |
1517 mpeg1_encode_picture_header(s, picture_number); | |
1518 break; | |
1519 } | |
286 | 1520 bits= get_bit_count(&s->pb); |
1521 s->header_bits= bits - s->last_bits; | |
1522 s->last_bits= bits; | |
1523 s->mv_bits=0; | |
1524 s->misc_bits=0; | |
1525 s->i_tex_bits=0; | |
1526 s->p_tex_bits=0; | |
1527 s->i_count=0; | |
1528 s->p_count=0; | |
1529 s->skip_count=0; | |
1530 | |
0 | 1531 /* init last dc values */ |
1532 /* note: quant matrix value (8) is implied here */ | |
1533 s->last_dc[0] = 128; | |
1534 s->last_dc[1] = 128; | |
1535 s->last_dc[2] = 128; | |
1536 s->mb_incr = 1; | |
1537 s->last_mv[0][0][0] = 0; | |
1538 s->last_mv[0][0][1] = 0; | |
1539 | |
162 | 1540 /* Get the GOB height based on picture height */ |
231 | 1541 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
162 | 1542 if (s->height <= 400) |
1543 s->gob_index = 1; | |
1544 else if (s->height <= 800) | |
1545 s->gob_index = 2; | |
1546 else | |
1547 s->gob_index = 4; | |
1548 } | |
268 | 1549 |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1550 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
|
1551 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1552 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
|
1553 /* 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
|
1554 /* 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
|
1555 if (s->rtp_mode) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1556 if (!mb_y) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1557 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
|
1558 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
|
1559 } 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
|
1560 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
|
1561 if (last_gob) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1562 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
|
1563 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1564 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1565 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1566 |
266 | 1567 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; |
1568 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1569 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1570 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1571 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
1572 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
|
1573 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
324 | 1574 /*const */int mb_type= s->mb_type[mb_y * s->mb_width + mb_x]; |
294 | 1575 PutBitContext pb; |
1576 int d; | |
1577 int dmin=10000000; | |
1578 int best=0; | |
0 | 1579 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1580 s->mb_x = mb_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1581 s->mb_y = mb_y; |
266 | 1582 s->block_index[0]+=2; |
1583 s->block_index[1]+=2; | |
1584 s->block_index[2]+=2; | |
1585 s->block_index[3]+=2; | |
1586 s->block_index[4]++; | |
1587 s->block_index[5]++; | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1588 |
294 | 1589 if(mb_type & (mb_type-1)){ // more than 1 MB type possible |
1590 pb= s->pb; | |
324 | 1591 s->mv_dir = MV_DIR_FORWARD; |
294 | 1592 if(mb_type&MB_TYPE_INTER){ |
324 | 1593 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; |
295 | 1594 s->mv_type = MV_TYPE_16X16; |
294 | 1595 s->mb_intra= 0; |
324 | 1596 s->mv[0][0][0] = s->p_mv_table[xy][0]; |
1597 s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
294 | 1598 init_put_bits(&s->pb, bit_buf[1], 3000, NULL, NULL); |
1599 s->block= s->inter_block; | |
0 | 1600 |
324 | 1601 encode_mb(s, s->mv[0][0][0], s->mv[0][0][1]); |
294 | 1602 d= get_bit_count(&s->pb); |
1603 if(d<dmin){ | |
1604 flush_put_bits(&s->pb); | |
1605 dmin=d; | |
1606 best_s.mv[0][0][0]= s->mv[0][0][0]; | |
1607 best_s.mv[0][0][1]= s->mv[0][0][1]; | |
1608 best_s.mb_intra= 0; | |
295 | 1609 best_s.mv_type = MV_TYPE_16X16; |
294 | 1610 best_s.pb=s->pb; |
1611 best_s.block= s->block; | |
1612 best=1; | |
1613 for(i=0; i<6; i++) | |
1614 best_s.block_last_index[i]= s->block_last_index[i]; | |
1615 } | |
0 | 1616 } |
295 | 1617 if(mb_type&MB_TYPE_INTER4V){ |
1618 s->mv_type = MV_TYPE_8X8; | |
1619 s->mb_intra= 0; | |
1620 for(i=0; i<4; i++){ | |
1621 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
1622 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
1623 } | |
1624 init_put_bits(&s->pb, bit_buf[2], 3000, NULL, NULL); | |
1625 s->block= s->inter4v_block; | |
1626 | |
324 | 1627 encode_mb(s, 0, 0); |
295 | 1628 d= get_bit_count(&s->pb); |
324 | 1629 if(d<dmin && 0){ |
295 | 1630 flush_put_bits(&s->pb); |
1631 dmin=d; | |
1632 for(i=0; i<4; i++){ | |
1633 best_s.mv[0][i][0] = s->mv[0][i][0]; | |
1634 best_s.mv[0][i][1] = s->mv[0][i][1]; | |
1635 } | |
1636 best_s.mb_intra= 0; | |
1637 best_s.mv_type = MV_TYPE_8X8; | |
1638 best_s.pb=s->pb; | |
1639 best_s.block= s->block; | |
1640 best=2; | |
1641 for(i=0; i<6; i++) | |
1642 best_s.block_last_index[i]= s->block_last_index[i]; | |
1643 } | |
1644 } | |
294 | 1645 if(mb_type&MB_TYPE_INTRA){ |
295 | 1646 s->mv_type = MV_TYPE_16X16; |
294 | 1647 s->mb_intra= 1; |
1648 s->mv[0][0][0] = 0; | |
1649 s->mv[0][0][1] = 0; | |
1650 init_put_bits(&s->pb, bit_buf[0], 3000, NULL, NULL); | |
1651 s->block= s->intra_block; | |
1652 | |
324 | 1653 encode_mb(s, 0, 0); |
294 | 1654 d= get_bit_count(&s->pb); |
1655 if(d<dmin){ | |
1656 flush_put_bits(&s->pb); | |
1657 dmin=d; | |
1658 best_s.mv[0][0][0]= 0; | |
1659 best_s.mv[0][0][1]= 0; | |
1660 best_s.mb_intra= 1; | |
295 | 1661 best_s.mv_type = MV_TYPE_16X16; |
294 | 1662 best_s.pb=s->pb; |
1663 best_s.block= s->block; | |
1664 for(i=0; i<6; i++) | |
1665 best_s.block_last_index[i]= s->block_last_index[i]; | |
1666 best=0; | |
1667 } | |
1668 /* force cleaning of ac/dc if needed ... */ | |
1669 s->mbintra_table[mb_x + mb_y*s->mb_width]=1; | |
1670 } | |
295 | 1671 for(i=0; i<4; i++){ |
1672 s->mv[0][i][0] = best_s.mv[0][i][0]; | |
1673 s->mv[0][i][1] = best_s.mv[0][i][1]; | |
1674 } | |
294 | 1675 s->mb_intra= best_s.mb_intra; |
295 | 1676 s->mv_type= best_s.mv_type; |
294 | 1677 for(i=0; i<6; i++) |
1678 s->block_last_index[i]= best_s.block_last_index[i]; | |
1679 copy_bits(&pb, bit_buf[best], dmin); | |
1680 s->block= best_s.block; | |
1681 s->pb= pb; | |
1682 } else { | |
324 | 1683 int motion_x, motion_y; |
1684 s->mv_type=MV_TYPE_16X16; | |
294 | 1685 // only one MB-Type possible |
324 | 1686 //FIXME convert to swicth() |
294 | 1687 if(mb_type&MB_TYPE_INTRA){ |
324 | 1688 s->mv_dir = MV_DIR_FORWARD; |
294 | 1689 s->mb_intra= 1; |
324 | 1690 motion_x= s->mv[0][0][0] = 0; |
1691 motion_y= s->mv[0][0][1] = 0; | |
1692 }else if(mb_type&MB_TYPE_INTER){ | |
1693 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1694 s->mv_dir = MV_DIR_FORWARD; | |
1695 s->mb_intra= 0; | |
1696 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0]; | |
1697 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
1698 }else if(mb_type&MB_TYPE_DIRECT){ | |
1699 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1700 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; | |
1701 s->mb_intra= 0; | |
1702 motion_x=0; | |
1703 motion_y=0; | |
294 | 1704 s->mv[0][0][0] = 0; |
1705 s->mv[0][0][1] = 0; | |
324 | 1706 s->mv[1][0][0] = 0; |
1707 s->mv[1][0][1] = 0; | |
1708 }else if(mb_type&MB_TYPE_BIDIR){ | |
1709 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1710 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; | |
294 | 1711 s->mb_intra= 0; |
324 | 1712 motion_x=0; |
1713 motion_y=0; | |
1714 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
1715 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
1716 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
1717 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
1718 }else if(mb_type&MB_TYPE_BACKWARD){ | |
1719 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1720 s->mv_dir = MV_DIR_BACKWARD; | |
1721 s->mb_intra= 0; | |
1722 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
1723 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
1724 }else if(mb_type&MB_TYPE_FORWARD){ | |
1725 int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
1726 s->mv_dir = MV_DIR_FORWARD; | |
1727 s->mb_intra= 0; | |
1728 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
1729 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
1730 // printf(" %d %d ", motion_x, motion_y); | |
1731 }else{ | |
1732 motion_x=motion_y=0; //gcc warning fix | |
1733 printf("illegal MB type\n"); | |
294 | 1734 } |
324 | 1735 encode_mb(s, motion_x, motion_y); |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1736 } |
0 | 1737 |
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
|
1738 MPV_decode_mb(s, s->block); |
0 | 1739 } |
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
|
1740 |
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
|
1741 |
231 | 1742 /* Obtain average GOB size for RTP */ |
1743 if (s->rtp_mode) { | |
1744 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
|
1745 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
231 | 1746 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
|
1747 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
|
1748 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
231 | 1749 } |
1750 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
1751 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
1752 s->first_gob_line = 0; | |
1753 } | |
0 | 1754 } |
294 | 1755 emms_c(); |
286 | 1756 |
311 | 1757 if (s->h263_msmpeg4 && s->msmpeg4_version<4 && s->pict_type == I_TYPE) |
208 | 1758 msmpeg4_encode_ext_header(s); |
1759 | |
162 | 1760 //if (s->gob_number) |
1761 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
231 | 1762 |
1763 /* Send the last GOB if RTP */ | |
1764 if (s->rtp_mode) { | |
1765 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
|
1766 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 1767 /* Call the RTP callback to send the last GOB */ |
1768 if (s->rtp_callback) | |
1769 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
|
1770 s->ptr_lastgob = pbBufPtr(&s->pb); |
231 | 1771 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
1772 } | |
0 | 1773 } |
1774 | |
220 | 1775 static int dct_quantize_c(MpegEncContext *s, |
0 | 1776 DCTELEM *block, int n, |
1777 int qscale) | |
1778 { | |
1779 int i, j, level, last_non_zero, q; | |
1780 const int *qmat; | |
216 | 1781 int minLevel, maxLevel; |
1782 | |
1783 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ | |
1784 /* mpeg4 */ | |
1785 minLevel= -2048; | |
1786 maxLevel= 2047; | |
1787 }else if(s->out_format==FMT_MPEG1){ | |
1788 /* mpeg1 */ | |
1789 minLevel= -255; | |
1790 maxLevel= 255; | |
230 | 1791 }else if(s->out_format==FMT_MJPEG){ |
1792 /* (m)jpeg */ | |
1793 minLevel= -1023; | |
1794 maxLevel= 1023; | |
216 | 1795 }else{ |
1796 /* h263 / msmpeg4 */ | |
1797 minLevel= -128; | |
1798 maxLevel= 127; | |
1799 } | |
0 | 1800 |
1801 av_fdct (block); | |
1802 | |
64 | 1803 /* we need this permutation so that we correct the IDCT |
1804 permutation. will be moved into DCT code */ | |
1805 block_permute(block); | |
1806 | |
0 | 1807 if (s->mb_intra) { |
1808 if (n < 4) | |
1809 q = s->y_dc_scale; | |
1810 else | |
1811 q = s->c_dc_scale; | |
1812 q = q << 3; | |
1813 | |
1814 /* note: block[0] is assumed to be positive */ | |
1815 block[0] = (block[0] + (q >> 1)) / q; | |
1816 i = 1; | |
1817 last_non_zero = 0; | |
1818 if (s->out_format == FMT_H263) { | |
1819 qmat = s->q_non_intra_matrix; | |
1820 } else { | |
1821 qmat = s->q_intra_matrix; | |
1822 } | |
1823 } else { | |
1824 i = 0; | |
1825 last_non_zero = -1; | |
1826 qmat = s->q_non_intra_matrix; | |
1827 } | |
1828 | |
1829 for(;i<64;i++) { | |
1830 j = zigzag_direct[i]; | |
1831 level = block[j]; | |
1832 level = level * qmat[j]; | |
1833 #ifdef PARANOID | |
1834 { | |
1835 static int count = 0; | |
1836 int level1, level2, qmat1; | |
1837 double val; | |
1838 if (qmat == s->q_non_intra_matrix) { | |
1839 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1840 } else { | |
1841 qmat1 = default_intra_matrix[j] * s->qscale; | |
1842 } | |
1843 if (av_fdct != jpeg_fdct_ifast) | |
1844 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1845 else | |
1846 val = ((double)block[j] * 8.0 * 2048.0) / | |
1847 ((double)qmat1 * aanscales[j]); | |
1848 level1 = (int)val; | |
1849 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1850 if (level1 != level2) { | |
1851 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1852 count, level2, level1, block[j], qmat1, qmat[j], | |
1853 val); | |
1854 count++; | |
1855 } | |
1856 | |
1857 } | |
1858 #endif | |
1859 /* XXX: slight error for the low range. Test should be equivalent to | |
1860 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1861 (QMAT_SHIFT - 3))) | |
1862 */ | |
1863 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1864 level) { | |
1865 level = level / (1 << (QMAT_SHIFT - 3)); | |
1866 /* XXX: currently, this code is not optimal. the range should be: | |
1867 mpeg1: -255..255 | |
1868 mpeg2: -2048..2047 | |
1869 h263: -128..127 | |
1870 mpeg4: -2048..2047 | |
1871 */ | |
216 | 1872 if (level > maxLevel) |
1873 level = maxLevel; | |
1874 else if (level < minLevel) | |
1875 level = minLevel; | |
0 | 1876 |
1877 block[j] = level; | |
1878 last_non_zero = i; | |
1879 } else { | |
1880 block[j] = 0; | |
1881 } | |
1882 } | |
1883 return last_non_zero; | |
1884 } | |
1885 | |
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
|
1886 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
|
1887 DCTELEM *block, int n, int qscale) |
0 | 1888 { |
200 | 1889 int i, level, nCoeffs; |
0 | 1890 const UINT16 *quant_matrix; |
1891 | |
200 | 1892 if(s->alternate_scan) nCoeffs= 64; |
1893 else nCoeffs= s->block_last_index[n]+1; | |
1894 | |
0 | 1895 if (s->mb_intra) { |
1896 if (n < 4) | |
1897 block[0] = block[0] * s->y_dc_scale; | |
1898 else | |
1899 block[0] = block[0] * s->c_dc_scale; | |
1900 /* XXX: only mpeg1 */ | |
1901 quant_matrix = s->intra_matrix; | |
200 | 1902 for(i=1;i<nCoeffs;i++) { |
1903 int j= zigzag_direct[i]; | |
1904 level = block[j]; | |
0 | 1905 if (level) { |
1906 if (level < 0) { | |
1907 level = -level; | |
200 | 1908 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1909 level = (level - 1) | 1; |
1910 level = -level; | |
1911 } else { | |
200 | 1912 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1913 level = (level - 1) | 1; |
1914 } | |
1915 #ifdef PARANOID | |
1916 if (level < -2048 || level > 2047) | |
1917 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1918 #endif | |
200 | 1919 block[j] = level; |
0 | 1920 } |
1921 } | |
1922 } else { | |
1923 i = 0; | |
1924 quant_matrix = s->non_intra_matrix; | |
217 | 1925 for(;i<nCoeffs;i++) { |
200 | 1926 int j= zigzag_direct[i]; |
1927 level = block[j]; | |
0 | 1928 if (level) { |
1929 if (level < 0) { | |
1930 level = -level; | |
1931 level = (((level << 1) + 1) * qscale * | |
200 | 1932 ((int) (quant_matrix[j]))) >> 4; |
0 | 1933 level = (level - 1) | 1; |
1934 level = -level; | |
1935 } else { | |
1936 level = (((level << 1) + 1) * qscale * | |
200 | 1937 ((int) (quant_matrix[j]))) >> 4; |
0 | 1938 level = (level - 1) | 1; |
1939 } | |
1940 #ifdef PARANOID | |
1941 if (level < -2048 || level > 2047) | |
1942 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1943 #endif | |
200 | 1944 block[j] = level; |
0 | 1945 } |
1946 } | |
1947 } | |
1948 } | |
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
|
1949 |
325 | 1950 static void dct_unquantize_mpeg2_c(MpegEncContext *s, |
1951 DCTELEM *block, int n, int qscale) | |
1952 { | |
1953 int i, level, nCoeffs; | |
1954 const UINT16 *quant_matrix; | |
1955 | |
1956 if(s->alternate_scan) nCoeffs= 64; | |
1957 else nCoeffs= s->block_last_index[n]+1; | |
1958 | |
1959 if (s->mb_intra) { | |
1960 if (n < 4) | |
1961 block[0] = block[0] * s->y_dc_scale; | |
1962 else | |
1963 block[0] = block[0] * s->c_dc_scale; | |
1964 quant_matrix = s->intra_matrix; | |
1965 for(i=1;i<nCoeffs;i++) { | |
1966 int j= zigzag_direct[i]; | |
1967 level = block[j]; | |
1968 if (level) { | |
1969 if (level < 0) { | |
1970 level = -level; | |
1971 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
1972 level = -level; | |
1973 } else { | |
1974 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
1975 } | |
1976 #ifdef PARANOID | |
1977 if (level < -2048 || level > 2047) | |
1978 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1979 #endif | |
1980 block[j] = level; | |
1981 } | |
1982 } | |
1983 } else { | |
1984 int sum=-1; | |
1985 i = 0; | |
1986 quant_matrix = s->non_intra_matrix; | |
1987 for(;i<nCoeffs;i++) { | |
1988 int j= zigzag_direct[i]; | |
1989 level = block[j]; | |
1990 if (level) { | |
1991 if (level < 0) { | |
1992 level = -level; | |
1993 level = (((level << 1) + 1) * qscale * | |
1994 ((int) (quant_matrix[j]))) >> 4; | |
1995 level = -level; | |
1996 } else { | |
1997 level = (((level << 1) + 1) * qscale * | |
1998 ((int) (quant_matrix[j]))) >> 4; | |
1999 } | |
2000 #ifdef PARANOID | |
2001 if (level < -2048 || level > 2047) | |
2002 fprintf(stderr, "unquant error %d %d\n", i, level); | |
2003 #endif | |
2004 block[j] = level; | |
2005 sum+=level; | |
2006 } | |
2007 } | |
2008 block[63]^=sum&1; | |
2009 } | |
2010 } | |
2011 | |
2012 | |
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
|
2013 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
|
2014 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
|
2015 { |
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
|
2016 int i, level, qmul, qadd; |
200 | 2017 int nCoeffs; |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2018 |
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
|
2019 if (s->mb_intra) { |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2020 if (!s->h263_aic) { |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2021 if (n < 4) |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2022 block[0] = block[0] * s->y_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2023 else |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2024 block[0] = block[0] * s->c_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2025 } |
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
|
2026 i = 1; |
200 | 2027 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
|
2028 } 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
|
2029 i = 0; |
200 | 2030 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
|
2031 } |
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
|
2032 |
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
|
2033 qmul = s->qscale << 1; |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2034 if (s->h263_aic && s->mb_intra) |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2035 qadd = 0; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2036 else |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2037 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
|
2038 |
200 | 2039 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
|
2040 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
|
2041 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
|
2042 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
|
2043 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
|
2044 } 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
|
2045 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
|
2046 } |
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
|
2047 #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
|
2048 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
|
2049 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
|
2050 #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
|
2051 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
|
2052 } |
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
|
2053 } |
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
|
2054 } |
0 | 2055 |
2056 /* rate control */ | |
2057 | |
2058 static void rate_control_init(MpegEncContext *s) | |
2059 { | |
268 | 2060 emms_c(); |
2061 | |
2062 //initial values, they dont really matter as they will be totally different within a few frames | |
2063 s->i_pred.coeff= s->p_pred.coeff= 7.0; | |
2064 s->i_pred.count= s->p_pred.count= 1.0; | |
2065 | |
2066 s->i_pred.decay= s->p_pred.decay= 0.4; | |
2067 | |
2068 // use more bits at the beginning, otherwise high motion at the begin will look like shit | |
2069 s->qsum=100; | |
2070 s->qcount=100; | |
2071 | |
2072 s->short_term_qsum=0.001; | |
2073 s->short_term_qcount=0.001; | |
2074 } | |
2075 | |
2076 static double predict(Predictor *p, double q, double var) | |
2077 { | |
2078 return p->coeff*var / (q*p->count); | |
2079 } | |
2080 | |
2081 static void update_predictor(Predictor *p, double q, double var, double size) | |
2082 { | |
2083 double new_coeff= size*q / (var + 1); | |
2084 if(var<1000) return; | |
2085 /*{ | |
2086 int pred= predict(p, q, var); | |
2087 int error= abs(pred-size); | |
2088 static double sum=0; | |
2089 static int count=0; | |
2090 if(count>5) sum+=error; | |
2091 count++; | |
2092 if(256*256*256*64%count==0){ | |
2093 printf("%d %f %f\n", count, sum/count, p->coeff); | |
2094 } | |
2095 }*/ | |
2096 p->count*= p->decay; | |
2097 p->coeff*= p->decay; | |
2098 p->count++; | |
2099 p->coeff+= new_coeff; | |
0 | 2100 } |
2101 | |
2102 static int rate_estimate_qscale(MpegEncContext *s) | |
2103 { | |
268 | 2104 int qmin= s->qmin; |
2105 int qmax= s->qmax; | |
2106 int rate_q=5; | |
2107 float q; | |
2108 int qscale; | |
2109 float br_compensation; | |
2110 double diff; | |
2111 double short_term_q; | |
2112 double long_term_q; | |
2113 int last_qscale= s->qscale; | |
2114 double fps; | |
2115 INT64 wanted_bits; | |
2116 emms_c(); | |
297 | 2117 |
268 | 2118 fps= (double)s->frame_rate / FRAME_RATE_BASE; |
2119 wanted_bits= s->bit_rate*(double)s->picture_number/fps; | |
2120 | |
2121 | |
2122 if(s->picture_number>2){ | |
2123 /* update predictors */ | |
2124 if(s->last_pict_type == I_TYPE){ | |
2125 //FIXME | |
2126 }else{ //P Frame | |
2127 //printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff); | |
2128 update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits); | |
2129 } | |
2130 } | |
2131 | |
2132 if(s->pict_type == I_TYPE){ | |
2133 //FIXME | |
2134 rate_q= s->qsum/s->qcount; | |
2135 }else{ //P Frame | |
2136 int i; | |
2137 int diff, best_diff=1000000000; | |
2138 for(i=1; i<=31; i++){ | |
2139 diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps; | |
2140 if(diff<0) diff= -diff; | |
2141 if(diff<best_diff){ | |
2142 best_diff= diff; | |
2143 rate_q= i; | |
2144 } | |
2145 } | |
2146 } | |
2147 | |
2148 s->short_term_qsum*=s->qblur; | |
2149 s->short_term_qcount*=s->qblur; | |
2150 | |
2151 s->short_term_qsum+= rate_q; | |
2152 s->short_term_qcount++; | |
2153 short_term_q= s->short_term_qsum/s->short_term_qcount; | |
2154 | |
302 | 2155 long_term_q= s->qsum/s->qcount*(s->total_bits+1)/(wanted_bits+1); //+1 to avoid nan & 0 |
268 | 2156 |
2157 // q= (long_term_q - short_term_q)*s->qcompress + short_term_q; | |
2158 q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q); | |
2159 | |
2160 diff= s->total_bits - wanted_bits; | |
2161 br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance; | |
270 | 2162 if(br_compensation<=0.0) br_compensation=0.001; |
268 | 2163 q/=br_compensation; |
2164 | |
2165 qscale= (int)(q + 0.5); | |
2166 if (qscale<qmin) qscale=qmin; | |
2167 else if(qscale>qmax) qscale=qmax; | |
2168 | |
2169 if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff; | |
2170 else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff; | |
2171 | |
2172 s->qsum+= qscale; | |
2173 s->qcount++; | |
2174 | |
2175 s->last_pict_type= s->pict_type; | |
302 | 2176 //printf("q:%d diff:%d comp:%f rate_q:%d st_q:%f fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation, |
2177 // rate_q, short_term_q, s->mc_mb_var, s->frame_bits); | |
268 | 2178 //printf("%d %d\n", s->bit_rate, (int)fps); |
2179 return qscale; | |
0 | 2180 } |
2181 | |
2182 AVCodec mpeg1video_encoder = { | |
2183 "mpeg1video", | |
2184 CODEC_TYPE_VIDEO, | |
2185 CODEC_ID_MPEG1VIDEO, | |
2186 sizeof(MpegEncContext), | |
2187 MPV_encode_init, | |
2188 MPV_encode_picture, | |
2189 MPV_encode_end, | |
2190 }; | |
2191 | |
2192 AVCodec h263_encoder = { | |
2193 "h263", | |
2194 CODEC_TYPE_VIDEO, | |
2195 CODEC_ID_H263, | |
2196 sizeof(MpegEncContext), | |
2197 MPV_encode_init, | |
2198 MPV_encode_picture, | |
2199 MPV_encode_end, | |
2200 }; | |
2201 | |
2202 AVCodec h263p_encoder = { | |
2203 "h263p", | |
2204 CODEC_TYPE_VIDEO, | |
2205 CODEC_ID_H263P, | |
2206 sizeof(MpegEncContext), | |
2207 MPV_encode_init, | |
2208 MPV_encode_picture, | |
2209 MPV_encode_end, | |
2210 }; | |
2211 | |
2212 AVCodec rv10_encoder = { | |
2213 "rv10", | |
2214 CODEC_TYPE_VIDEO, | |
2215 CODEC_ID_RV10, | |
2216 sizeof(MpegEncContext), | |
2217 MPV_encode_init, | |
2218 MPV_encode_picture, | |
2219 MPV_encode_end, | |
2220 }; | |
2221 | |
2222 AVCodec mjpeg_encoder = { | |
2223 "mjpeg", | |
2224 CODEC_TYPE_VIDEO, | |
2225 CODEC_ID_MJPEG, | |
2226 sizeof(MpegEncContext), | |
2227 MPV_encode_init, | |
2228 MPV_encode_picture, | |
2229 MPV_encode_end, | |
2230 }; | |
2231 | |
71 | 2232 AVCodec mpeg4_encoder = { |
2233 "mpeg4", | |
0 | 2234 CODEC_TYPE_VIDEO, |
71 | 2235 CODEC_ID_MPEG4, |
0 | 2236 sizeof(MpegEncContext), |
2237 MPV_encode_init, | |
2238 MPV_encode_picture, | |
2239 MPV_encode_end, | |
2240 }; | |
2241 | |
307 | 2242 AVCodec msmpeg4v1_encoder = { |
2243 "msmpeg4v1", | |
0 | 2244 CODEC_TYPE_VIDEO, |
307 | 2245 CODEC_ID_MSMPEG4V1, |
0 | 2246 sizeof(MpegEncContext), |
2247 MPV_encode_init, | |
2248 MPV_encode_picture, | |
2249 MPV_encode_end, | |
2250 }; | |
307 | 2251 |
2252 AVCodec msmpeg4v2_encoder = { | |
2253 "msmpeg4v2", | |
2254 CODEC_TYPE_VIDEO, | |
2255 CODEC_ID_MSMPEG4V2, | |
2256 sizeof(MpegEncContext), | |
2257 MPV_encode_init, | |
2258 MPV_encode_picture, | |
2259 MPV_encode_end, | |
2260 }; | |
2261 | |
2262 AVCodec msmpeg4v3_encoder = { | |
2263 "msmpeg4", | |
2264 CODEC_TYPE_VIDEO, | |
2265 CODEC_ID_MSMPEG4V3, | |
2266 sizeof(MpegEncContext), | |
2267 MPV_encode_init, | |
2268 MPV_encode_picture, | |
2269 MPV_encode_end, | |
2270 }; |