Mercurial > libavcodec.hg
annotate mpegvideo.c @ 312:8cf5507e6ca5 libavcodec
mpeg4 mpeg quantizer support
author | michaelni |
---|---|
date | Sun, 07 Apr 2002 02:03:32 +0000 |
parents | ac677a84d5df |
children | 40d8092e2ff0 |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest mpeg encoder (well, it was the simplest!) | |
3 * Copyright (c) 2000,2001 Gerard Lantau. | |
4 * | |
5 * This program is free software; you can redistribute it and/or modify | |
6 * it under the terms of the GNU General Public License as published by | |
7 * the Free Software Foundation; either version 2 of the License, or | |
8 * (at your option) any later version. | |
9 * | |
10 * This program is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 * GNU General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU General Public License | |
16 * along with this program; if not, write to the Free Software | |
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
295 | 18 * |
19 * 4MV & hq encoding stuff by Michael Niedermayer <michaelni@gmx.at> | |
0 | 20 */ |
21 #include <stdlib.h> | |
22 #include <stdio.h> | |
23 #include <math.h> | |
24 #include <string.h> | |
25 #include "avcodec.h" | |
26 #include "dsputil.h" | |
27 #include "mpegvideo.h" | |
28 | |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
29 #ifdef USE_FASTMEMCPY |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
30 #include "fastmemcpy.h" |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
31 #endif |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
32 |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
33 static void encode_picture(MpegEncContext *s, int picture_number); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
34 static void rate_control_init(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
35 static int rate_estimate_qscale(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
36 static void dct_unquantize_mpeg1_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
37 DCTELEM *block, int n, int qscale); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
38 static void dct_unquantize_h263_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
39 DCTELEM *block, int n, int qscale); |
206 | 40 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
220 | 41 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
206 | 42 |
220 | 43 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c; |
206 | 44 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
45 | |
0 | 46 #define EDGE_WIDTH 16 |
47 | |
48 /* enable all paranoid tests for rounding, overflows, etc... */ | |
49 //#define PARANOID | |
50 | |
51 //#define DEBUG | |
52 | |
53 /* for jpeg fast DCT */ | |
54 #define CONST_BITS 14 | |
55 | |
56 static const unsigned short aanscales[64] = { | |
57 /* precomputed values scaled up by 14 bits */ | |
58 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
59 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
60 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
61 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
62 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
63 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
64 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
65 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
66 }; | |
67 | |
68 static UINT8 h263_chroma_roundtab[16] = { | |
69 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
70 }; | |
71 | |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
72 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
|
73 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
|
74 |
0 | 75 /* default motion estimation */ |
76 int motion_estimation_method = ME_LOG; | |
77 | |
200 | 78 extern UINT8 zigzag_end[64]; |
79 | |
220 | 80 static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale) |
0 | 81 { |
82 int i; | |
83 | |
84 if (av_fdct == jpeg_fdct_ifast) { | |
85 for(i=0;i<64;i++) { | |
86 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
220 | 87 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
88 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
89 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
0 | 90 |
220 | 91 qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
92 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)])); | |
0 | 93 } |
94 } else { | |
95 for(i=0;i<64;i++) { | |
96 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
220 | 97 So 16 <= qscale * quant_matrix[i] <= 7905 |
98 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
99 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
0 | 100 */ |
220 | 101 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
102 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
0 | 103 } |
104 } | |
105 } | |
106 | |
107 /* init common structure for both encoder and decoder */ | |
108 int MPV_common_init(MpegEncContext *s) | |
109 { | |
110 int c_size, i; | |
111 UINT8 *pict; | |
112 | |
312 | 113 s->dct_unquantize_h263 = dct_unquantize_h263_c; |
114 s->dct_unquantize_mpeg = dct_unquantize_mpeg1_c; | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
115 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
116 #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
|
117 MPV_common_init_mmx(s); |
8 | 118 #endif |
312 | 119 //setup default unquantizers (mpeg4 might change it later) |
120 if(s->out_format == FMT_H263) | |
121 s->dct_unquantize = s->dct_unquantize_h263; | |
122 else | |
123 s->dct_unquantize = s->dct_unquantize_mpeg; | |
124 | |
0 | 125 s->mb_width = (s->width + 15) / 16; |
126 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
|
127 s->mb_num = s->mb_width * s->mb_height; |
0 | 128 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
129 | |
130 for(i=0;i<3;i++) { | |
131 int w, h, shift, pict_start; | |
132 | |
133 w = s->linesize; | |
134 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
135 shift = (i == 0) ? 0 : 1; | |
136 c_size = (w >> shift) * (h >> shift); | |
137 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
138 | |
139 pict = av_mallocz(c_size); | |
140 if (pict == NULL) | |
141 goto fail; | |
142 s->last_picture_base[i] = pict; | |
143 s->last_picture[i] = pict + pict_start; | |
144 | |
145 pict = av_mallocz(c_size); | |
146 if (pict == NULL) | |
147 goto fail; | |
148 s->next_picture_base[i] = pict; | |
149 s->next_picture[i] = pict + pict_start; | |
150 | |
151 if (s->has_b_frames) { | |
152 pict = av_mallocz(c_size); | |
153 if (pict == NULL) | |
154 goto fail; | |
155 s->aux_picture_base[i] = pict; | |
156 s->aux_picture[i] = pict + pict_start; | |
157 } | |
158 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
159 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
160 if (s->encoding) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
161 /* Allocate MB type table */ |
239 | 162 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
|
163 if (s->mb_type == NULL) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
164 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
165 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
166 } |
239 | 167 |
168 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16)); | |
169 if (s->mb_var == NULL) { | |
170 perror("malloc"); | |
171 goto fail; | |
172 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
173 /* Allocate MV table */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
174 /* By now we just have one MV per MB */ |
239 | 175 s->mv_table[0] = av_mallocz(s->mb_num * sizeof(INT16)); |
176 s->mv_table[1] = av_mallocz(s->mb_num * sizeof(INT16)); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
177 if (s->mv_table[1] == NULL || s->mv_table[0] == NULL) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
178 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
179 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
180 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
181 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
182 |
288 | 183 if (s->out_format == FMT_H263 || s->encoding) { |
0 | 184 int size; |
185 /* MV prediction */ | |
186 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
187 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
188 if (s->motion_val == NULL) | |
189 goto fail; | |
190 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
191 } | |
192 | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
193 if (s->h263_pred || s->h263_plus) { |
0 | 194 int y_size, c_size, i, size; |
195 | |
196 /* dc values */ | |
197 | |
198 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
199 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
200 size = y_size + 2 * c_size; | |
201 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
202 if (s->dc_val[0] == NULL) | |
203 goto fail; | |
204 s->dc_val[1] = s->dc_val[0] + y_size; | |
205 s->dc_val[2] = s->dc_val[1] + c_size; | |
206 for(i=0;i<size;i++) | |
207 s->dc_val[0][i] = 1024; | |
208 | |
209 /* ac values */ | |
210 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
211 if (s->ac_val[0] == NULL) | |
212 goto fail; | |
213 s->ac_val[1] = s->ac_val[0] + y_size; | |
214 s->ac_val[2] = s->ac_val[1] + c_size; | |
215 | |
216 /* cbp values */ | |
217 s->coded_block = av_mallocz(y_size); | |
218 if (!s->coded_block) | |
219 goto fail; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
220 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
221 /* 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
|
222 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
|
223 if (!s->mbintra_table) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
224 goto fail; |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
225 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
|
226 } |
0 | 227 /* default structure is frame */ |
228 s->picture_structure = PICT_FRAME; | |
229 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
230 /* init macroblock skip table */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
231 if (!s->encoding) { |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
232 s->mbskip_table = av_mallocz(s->mb_num); |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
233 if (!s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
234 goto fail; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
235 } |
294 | 236 |
237 s->block= s->intra_block; | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
238 |
0 | 239 s->context_initialized = 1; |
240 return 0; | |
241 fail: | |
244 | 242 MPV_common_end(s); |
243 return -1; | |
244 } | |
245 | |
246 /* init common structure for both encoder and decoder */ | |
247 void MPV_common_end(MpegEncContext *s) | |
248 { | |
249 int i; | |
250 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
251 if (s->mb_type) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
252 free(s->mb_type); |
239 | 253 if (s->mb_var) |
254 free(s->mb_var); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
255 if (s->mv_table[0]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
256 free(s->mv_table[0]); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
257 if (s->mv_table[1]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
258 free(s->mv_table[1]); |
0 | 259 if (s->motion_val) |
260 free(s->motion_val); | |
261 if (s->dc_val[0]) | |
262 free(s->dc_val[0]); | |
263 if (s->ac_val[0]) | |
264 free(s->ac_val[0]); | |
265 if (s->coded_block) | |
266 free(s->coded_block); | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
267 if (s->mbintra_table) |
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
|
268 free(s->mbintra_table); |
244 | 269 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
270 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
271 free(s->mbskip_table); |
0 | 272 for(i=0;i<3;i++) { |
273 if (s->last_picture_base[i]) | |
244 | 274 free(s->last_picture_base[i]); |
275 if (s->next_picture_base[i]) | |
276 free(s->next_picture_base[i]); | |
0 | 277 if (s->has_b_frames) |
278 free(s->aux_picture_base[i]); | |
279 } | |
280 s->context_initialized = 0; | |
281 } | |
282 | |
283 /* init video encoder */ | |
284 int MPV_encode_init(AVCodecContext *avctx) | |
285 { | |
286 MpegEncContext *s = avctx->priv_data; | |
60 | 287 int i; |
0 | 288 |
289 s->bit_rate = avctx->bit_rate; | |
268 | 290 s->bit_rate_tolerance = avctx->bit_rate_tolerance; |
0 | 291 s->frame_rate = avctx->frame_rate; |
292 s->width = avctx->width; | |
293 s->height = avctx->height; | |
294 s->gop_size = avctx->gop_size; | |
162 | 295 s->rtp_mode = avctx->rtp_mode; |
296 s->rtp_payload_size = avctx->rtp_payload_size; | |
231 | 297 if (avctx->rtp_callback) |
298 s->rtp_callback = avctx->rtp_callback; | |
268 | 299 s->qmin= avctx->qmin; |
300 s->qmax= avctx->qmax; | |
301 s->max_qdiff= avctx->max_qdiff; | |
302 s->qcompress= avctx->qcompress; | |
303 s->qblur= avctx->qblur; | |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
304 s->avctx = avctx; |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
305 s->aspect_ratio_info= avctx->aspect_ratio_info; |
294 | 306 s->flags= avctx->flags; |
162 | 307 |
0 | 308 if (s->gop_size <= 1) { |
309 s->intra_only = 1; | |
310 s->gop_size = 12; | |
311 } else { | |
312 s->intra_only = 0; | |
313 } | |
314 s->full_search = motion_estimation_method; | |
315 | |
316 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
|
317 |
0 | 318 switch(avctx->codec->id) { |
319 case CODEC_ID_MPEG1VIDEO: | |
320 s->out_format = FMT_MPEG1; | |
321 break; | |
322 case CODEC_ID_MJPEG: | |
323 s->out_format = FMT_MJPEG; | |
324 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
|
325 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
|
326 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
|
327 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
|
328 s->mjpeg_vsample[2] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
329 s->mjpeg_hsample[0] = 2; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
330 s->mjpeg_hsample[1] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
331 s->mjpeg_hsample[2] = 1; |
0 | 332 if (mjpeg_init(s) < 0) |
333 return -1; | |
334 break; | |
335 case CODEC_ID_H263: | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
336 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
|
337 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
0 | 338 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
339 } |
0 | 340 s->out_format = FMT_H263; |
341 break; | |
342 case CODEC_ID_H263P: | |
343 s->out_format = FMT_H263; | |
162 | 344 s->rtp_mode = 1; |
345 s->rtp_payload_size = 1200; | |
0 | 346 s->h263_plus = 1; |
78 | 347 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
348 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
349 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
350 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
351 s->umvplus_dec = 0; |
0 | 352 break; |
353 case CODEC_ID_RV10: | |
354 s->out_format = FMT_H263; | |
355 s->h263_rv10 = 1; | |
356 break; | |
71 | 357 case CODEC_ID_MPEG4: |
0 | 358 s->out_format = FMT_H263; |
359 s->h263_pred = 1; | |
360 s->unrestricted_mv = 1; | |
361 break; | |
307 | 362 case CODEC_ID_MSMPEG4V1: |
0 | 363 s->out_format = FMT_H263; |
364 s->h263_msmpeg4 = 1; | |
365 s->h263_pred = 1; | |
366 s->unrestricted_mv = 1; | |
307 | 367 s->msmpeg4_version= 1; |
368 break; | |
369 case CODEC_ID_MSMPEG4V2: | |
370 s->out_format = FMT_H263; | |
371 s->h263_msmpeg4 = 1; | |
372 s->h263_pred = 1; | |
373 s->unrestricted_mv = 1; | |
374 s->msmpeg4_version= 2; | |
375 break; | |
376 case CODEC_ID_MSMPEG4V3: | |
377 s->out_format = FMT_H263; | |
378 s->h263_msmpeg4 = 1; | |
379 s->h263_pred = 1; | |
380 s->unrestricted_mv = 1; | |
381 s->msmpeg4_version= 3; | |
0 | 382 break; |
383 default: | |
384 return -1; | |
385 } | |
295 | 386 |
387 if((s->flags&CODEC_FLAG_4MV) && !(s->flags&CODEC_FLAG_HQ)){ | |
388 printf("4MV is currently only supported in HQ mode\n"); | |
389 return -1; | |
390 } | |
0 | 391 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
392 { /* 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
|
393 static int done=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
394 if(!done){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
395 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
396 done=1; |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
397 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
|
398 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
|
399 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
400 for(i=-16; i<16; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
401 default_fcode_tab[i + MAX_MV]= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
402 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
403 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
404 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
405 s->mv_penalty= default_mv_penalty; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
406 s->fcode_tab= default_fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
407 |
0 | 408 if (s->out_format == FMT_H263) |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
409 h263_encode_init(s); |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
410 else if (s->out_format == FMT_MPEG1) |
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
411 mpeg1_encode_init(s); |
0 | 412 |
287 | 413 /* dont use mv_penalty table for crap MV as it would be confused */ |
414 if(s->full_search<4) s->mv_penalty= default_mv_penalty; | |
415 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
416 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
417 |
0 | 418 /* init */ |
419 if (MPV_common_init(s) < 0) | |
420 return -1; | |
421 | |
60 | 422 /* init default q matrix */ |
423 for(i=0;i<64;i++) { | |
424 s->intra_matrix[i] = default_intra_matrix[i]; | |
425 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
426 } | |
427 | |
0 | 428 /* rate control init */ |
429 rate_control_init(s); | |
430 | |
431 s->picture_number = 0; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
432 s->picture_in_gop_number = 0; |
0 | 433 s->fake_picture_number = 0; |
434 /* motion detector init */ | |
435 s->f_code = 1; | |
436 | |
437 return 0; | |
438 } | |
439 | |
440 int MPV_encode_end(AVCodecContext *avctx) | |
441 { | |
442 MpegEncContext *s = avctx->priv_data; | |
443 | |
444 #ifdef STATS | |
445 print_stats(); | |
446 #endif | |
447 MPV_common_end(s); | |
448 if (s->out_format == FMT_MJPEG) | |
449 mjpeg_close(s); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
450 |
0 | 451 return 0; |
452 } | |
453 | |
454 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 455 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 456 { |
457 UINT8 *ptr, *last_line; | |
458 int i; | |
459 | |
460 last_line = buf + (height - 1) * wrap; | |
461 for(i=0;i<w;i++) { | |
462 /* top and bottom */ | |
463 memcpy(buf - (i + 1) * wrap, buf, width); | |
464 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
465 } | |
466 /* left and right */ | |
467 ptr = buf; | |
468 for(i=0;i<height;i++) { | |
469 memset(ptr - w, ptr[0], w); | |
470 memset(ptr + width, ptr[width-1], w); | |
471 ptr += wrap; | |
472 } | |
473 /* corners */ | |
474 for(i=0;i<w;i++) { | |
475 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
476 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
477 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
478 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
479 } | |
480 } | |
481 | |
482 /* generic function for encode/decode called before a frame is coded/decoded */ | |
483 void MPV_frame_start(MpegEncContext *s) | |
484 { | |
485 int i; | |
486 UINT8 *tmp; | |
487 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
488 s->mb_skiped = 0; |
0 | 489 if (s->pict_type == B_TYPE) { |
490 for(i=0;i<3;i++) { | |
491 s->current_picture[i] = s->aux_picture[i]; | |
492 } | |
493 } else { | |
262 | 494 s->last_non_b_pict_type= s->pict_type; |
0 | 495 for(i=0;i<3;i++) { |
496 /* swap next and last */ | |
497 tmp = s->last_picture[i]; | |
498 s->last_picture[i] = s->next_picture[i]; | |
499 s->next_picture[i] = tmp; | |
500 s->current_picture[i] = tmp; | |
501 } | |
502 } | |
503 } | |
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
|
504 |
0 | 505 /* generic function for encode/decode called after a frame has been coded/decoded */ |
506 void MPV_frame_end(MpegEncContext *s) | |
507 { | |
508 /* draw edge for correct motion prediction if outside */ | |
220 | 509 if (s->pict_type != B_TYPE && !s->intra_only) { |
257 | 510 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
|
511 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
|
512 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
|
513 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
|
514 }else{ |
257 | 515 /* mpeg4? / opendivx / xvid */ |
0 | 516 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
517 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
518 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
|
519 } |
0 | 520 } |
207 | 521 emms_c(); |
0 | 522 } |
523 | |
524 int MPV_encode_picture(AVCodecContext *avctx, | |
525 unsigned char *buf, int buf_size, void *data) | |
526 { | |
527 MpegEncContext *s = avctx->priv_data; | |
528 AVPicture *pict = data; | |
529 int i, j; | |
530 | |
531 if (s->fixed_qscale) | |
532 s->qscale = avctx->quality; | |
533 | |
534 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
535 | |
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
|
536 s->force_type= (avctx->flags&CODEC_FLAG_TYPE) ? |
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
|
537 (avctx->key_frame ? I_TYPE : P_TYPE) : 0; |
0 | 538 if (!s->intra_only) { |
539 /* first picture of GOP is intra */ | |
297 | 540 if (s->picture_in_gop_number % s->gop_size==0 || s->force_type==I_TYPE){ |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
541 s->picture_in_gop_number=0; |
0 | 542 s->pict_type = I_TYPE; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
543 }else |
0 | 544 s->pict_type = P_TYPE; |
545 } else { | |
546 s->pict_type = I_TYPE; | |
547 } | |
548 | |
549 MPV_frame_start(s); | |
220 | 550 |
0 | 551 for(i=0;i<3;i++) { |
552 UINT8 *src = pict->data[i]; | |
553 UINT8 *dest = s->current_picture[i]; | |
554 int src_wrap = pict->linesize[i]; | |
555 int dest_wrap = s->linesize; | |
556 int w = s->width; | |
557 int h = s->height; | |
558 | |
559 if (i >= 1) { | |
560 dest_wrap >>= 1; | |
561 w >>= 1; | |
562 h >>= 1; | |
563 } | |
564 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
565 if(dest_wrap==src_wrap){ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
566 s->new_picture[i] = pict->data[i]; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
567 } else { |
220 | 568 for(j=0;j<h;j++) { |
569 memcpy(dest, src, w); | |
570 dest += dest_wrap; | |
571 src += src_wrap; | |
572 } | |
227
ec1bc02a0a47
avoid copying input when encoding non intra stuff too
michaelni
parents:
220
diff
changeset
|
573 s->new_picture[i] = s->current_picture[i]; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
574 } |
0 | 575 } |
576 | |
577 encode_picture(s, s->picture_number); | |
271 | 578 avctx->key_frame = (s->pict_type == I_TYPE); |
286 | 579 avctx->header_bits = s->header_bits; |
580 avctx->mv_bits = s->mv_bits; | |
581 avctx->misc_bits = s->misc_bits; | |
582 avctx->i_tex_bits = s->i_tex_bits; | |
583 avctx->p_tex_bits = s->p_tex_bits; | |
584 avctx->i_count = s->i_count; | |
585 avctx->p_count = s->p_count; | |
586 avctx->skip_count = s->skip_count; | |
587 | |
0 | 588 MPV_frame_end(s); |
589 s->picture_number++; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
590 s->picture_in_gop_number++; |
0 | 591 |
592 if (s->out_format == FMT_MJPEG) | |
593 mjpeg_picture_trailer(s); | |
594 | |
595 flush_put_bits(&s->pb); | |
268 | 596 s->last_frame_bits= s->frame_bits; |
597 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8; | |
598 s->total_bits += s->frame_bits; | |
286 | 599 avctx->frame_bits = s->frame_bits; |
600 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n", | |
601 //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
|
602 |
0 | 603 avctx->quality = s->qscale; |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
604 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
|
605 /* 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
|
606 /* 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
|
607 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
|
608 pict->linesize, s->linesize, avctx); |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
609 } |
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
|
610 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 611 } |
612 | |
613 static inline int clip(int a, int amin, int amax) | |
614 { | |
615 if (a < amin) | |
616 return amin; | |
617 else if (a > amax) | |
618 return amax; | |
619 else | |
620 return a; | |
621 } | |
622 | |
255 | 623 static inline void gmc1_motion(MpegEncContext *s, |
624 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
625 int dest_offset, | |
626 UINT8 **ref_picture, int src_offset, | |
627 int h) | |
628 { | |
629 UINT8 *ptr; | |
630 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
631 int motion_x, motion_y; | |
632 | |
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
633 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n"); |
255 | 634 motion_x= s->sprite_offset[0][0]; |
635 motion_y= s->sprite_offset[0][1]; | |
636 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
637 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
638 motion_x<<=(3-s->sprite_warping_accuracy); | |
639 motion_y<<=(3-s->sprite_warping_accuracy); | |
640 src_x = clip(src_x, -16, s->width); | |
641 if (src_x == s->width) | |
642 motion_x =0; | |
643 src_y = clip(src_y, -16, s->height); | |
644 if (src_y == s->height) | |
645 motion_y =0; | |
646 | |
647 linesize = s->linesize; | |
648 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
649 | |
650 dest_y+=dest_offset; | |
651 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
652 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
653 | |
654 motion_x= s->sprite_offset[1][0]; | |
655 motion_y= s->sprite_offset[1][1]; | |
656 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
657 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
658 motion_x<<=(3-s->sprite_warping_accuracy); | |
659 motion_y<<=(3-s->sprite_warping_accuracy); | |
660 src_x = clip(src_x, -8, s->width>>1); | |
661 if (src_x == s->width>>1) | |
662 motion_x =0; | |
663 src_y = clip(src_y, -8, s->height>>1); | |
664 if (src_y == s->height>>1) | |
665 motion_y =0; | |
666 | |
667 offset = (src_y * linesize>>1) + src_x + (src_offset>>1); | |
668 ptr = ref_picture[1] + offset; | |
669 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
670 ptr = ref_picture[2] + offset; | |
671 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
672 | |
673 return; | |
674 } | |
675 | |
0 | 676 /* apply one mpeg motion vector to the three components */ |
677 static inline void mpeg_motion(MpegEncContext *s, | |
678 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
679 int dest_offset, | |
680 UINT8 **ref_picture, int src_offset, | |
681 int field_based, op_pixels_func *pix_op, | |
682 int motion_x, int motion_y, int h) | |
683 { | |
684 UINT8 *ptr; | |
685 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
255 | 686 if(s->quarter_sample) |
687 { | |
688 motion_x>>=1; | |
689 motion_y>>=1; | |
690 } | |
0 | 691 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
692 src_x = s->mb_x * 16 + (motion_x >> 1); | |
693 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
694 | |
695 /* WARNING: do no forget half pels */ | |
696 height = s->height >> field_based; | |
697 src_x = clip(src_x, -16, s->width); | |
698 if (src_x == s->width) | |
699 dxy &= ~1; | |
700 src_y = clip(src_y, -16, height); | |
701 if (src_y == height) | |
702 dxy &= ~2; | |
703 linesize = s->linesize << field_based; | |
704 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
705 dest_y += dest_offset; | |
706 pix_op[dxy](dest_y, ptr, linesize, h); | |
707 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
708 | |
709 if (s->out_format == FMT_H263) { | |
710 dxy = 0; | |
711 if ((motion_x & 3) != 0) | |
712 dxy |= 1; | |
713 if ((motion_y & 3) != 0) | |
714 dxy |= 2; | |
715 mx = motion_x >> 2; | |
716 my = motion_y >> 2; | |
717 } else { | |
718 mx = motion_x / 2; | |
719 my = motion_y / 2; | |
720 dxy = ((my & 1) << 1) | (mx & 1); | |
721 mx >>= 1; | |
722 my >>= 1; | |
723 } | |
724 | |
725 src_x = s->mb_x * 8 + mx; | |
726 src_y = s->mb_y * (8 >> field_based) + my; | |
727 src_x = clip(src_x, -8, s->width >> 1); | |
728 if (src_x == (s->width >> 1)) | |
729 dxy &= ~1; | |
730 src_y = clip(src_y, -8, height >> 1); | |
731 if (src_y == (height >> 1)) | |
732 dxy &= ~2; | |
733 | |
734 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
735 ptr = ref_picture[1] + offset; | |
736 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
737 ptr = ref_picture[2] + offset; | |
738 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
739 } | |
740 | |
255 | 741 static inline void qpel_motion(MpegEncContext *s, |
742 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
743 int dest_offset, | |
744 UINT8 **ref_picture, int src_offset, | |
745 int field_based, op_pixels_func *pix_op, | |
746 qpel_mc_func *qpix_op, | |
747 int motion_x, int motion_y, int h) | |
748 { | |
749 UINT8 *ptr; | |
750 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
751 | |
752 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
753 src_x = s->mb_x * 16 + (motion_x >> 2); | |
754 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
755 | |
756 height = s->height >> field_based; | |
757 src_x = clip(src_x, -16, s->width); | |
758 if (src_x == s->width) | |
759 dxy &= ~3; | |
760 src_y = clip(src_y, -16, height); | |
761 if (src_y == height) | |
762 dxy &= ~12; | |
763 linesize = s->linesize << field_based; | |
764 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
765 dest_y += dest_offset; | |
766 //printf("%d %d %d\n", src_x, src_y, dxy); | |
767 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3); | |
768 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3); | |
769 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3); | |
770 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3); | |
771 | |
772 mx= (motion_x>>1) | (motion_x&1); | |
773 my= (motion_y>>1) | (motion_y&1); | |
774 | |
775 dxy = 0; | |
776 if ((mx & 3) != 0) | |
777 dxy |= 1; | |
778 if ((my & 3) != 0) | |
779 dxy |= 2; | |
780 mx = mx >> 2; | |
781 my = my >> 2; | |
782 | |
783 src_x = s->mb_x * 8 + mx; | |
784 src_y = s->mb_y * (8 >> field_based) + my; | |
785 src_x = clip(src_x, -8, s->width >> 1); | |
786 if (src_x == (s->width >> 1)) | |
787 dxy &= ~1; | |
788 src_y = clip(src_y, -8, height >> 1); | |
789 if (src_y == (height >> 1)) | |
790 dxy &= ~2; | |
791 | |
792 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
793 ptr = ref_picture[1] + offset; | |
794 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
795 ptr = ref_picture[2] + offset; | |
796 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
797 } | |
798 | |
799 | |
0 | 800 static inline void MPV_motion(MpegEncContext *s, |
801 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
802 int dir, UINT8 **ref_picture, | |
255 | 803 op_pixels_func *pix_op, qpel_mc_func *qpix_op) |
0 | 804 { |
805 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
806 int mb_x, mb_y, i; | |
807 UINT8 *ptr, *dest; | |
808 | |
809 mb_x = s->mb_x; | |
810 mb_y = s->mb_y; | |
811 | |
812 switch(s->mv_type) { | |
813 case MV_TYPE_16X16: | |
255 | 814 if(s->mcsel){ |
815 #if 0 | |
816 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
817 ref_picture, 0, | |
818 0, pix_op, | |
819 s->sprite_offset[0][0]>>3, | |
820 s->sprite_offset[0][1]>>3, | |
821 16); | |
822 #else | |
823 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
824 ref_picture, 0, | |
825 16); | |
826 #endif | |
262 | 827 }else if(s->quarter_sample && dir==0){ //FIXME |
255 | 828 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, |
829 ref_picture, 0, | |
830 0, pix_op, qpix_op, | |
831 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
832 }else{ | |
833 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
834 ref_picture, 0, | |
835 0, pix_op, | |
836 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
837 } | |
0 | 838 break; |
839 case MV_TYPE_8X8: | |
840 for(i=0;i<4;i++) { | |
841 motion_x = s->mv[dir][i][0]; | |
842 motion_y = s->mv[dir][i][1]; | |
843 | |
844 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
845 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
295 | 846 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; |
0 | 847 |
848 /* WARNING: do no forget half pels */ | |
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, s->height); | |
853 if (src_y == s->height) | |
854 dxy &= ~2; | |
855 | |
856 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
857 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
858 pix_op[dxy](dest, ptr, s->linesize, 8); | |
859 } | |
860 /* In case of 8X8, we construct a single chroma motion vector | |
861 with a special rounding */ | |
862 mx = 0; | |
863 my = 0; | |
864 for(i=0;i<4;i++) { | |
865 mx += s->mv[dir][i][0]; | |
866 my += s->mv[dir][i][1]; | |
867 } | |
868 if (mx >= 0) | |
869 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
870 else { | |
871 mx = -mx; | |
872 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
873 } | |
874 if (my >= 0) | |
875 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
876 else { | |
877 my = -my; | |
878 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
879 } | |
880 dxy = ((my & 1) << 1) | (mx & 1); | |
881 mx >>= 1; | |
882 my >>= 1; | |
883 | |
884 src_x = mb_x * 8 + mx; | |
885 src_y = mb_y * 8 + my; | |
886 src_x = clip(src_x, -8, s->width/2); | |
887 if (src_x == s->width/2) | |
888 dxy &= ~1; | |
889 src_y = clip(src_y, -8, s->height/2); | |
890 if (src_y == s->height/2) | |
891 dxy &= ~2; | |
892 | |
893 offset = (src_y * (s->linesize >> 1)) + src_x; | |
894 ptr = ref_picture[1] + offset; | |
895 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
896 ptr = ref_picture[2] + offset; | |
897 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
898 break; | |
899 case MV_TYPE_FIELD: | |
900 if (s->picture_structure == PICT_FRAME) { | |
901 /* top field */ | |
902 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
903 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
904 1, pix_op, | |
905 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
906 /* bottom field */ | |
907 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
908 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
909 1, pix_op, | |
910 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
911 } else { | |
912 | |
913 | |
914 } | |
915 break; | |
916 } | |
917 } | |
918 | |
919 | |
920 /* put block[] to dest[] */ | |
921 static inline void put_dct(MpegEncContext *s, | |
922 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
923 { | |
924 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
|
925 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
|
926 ff_idct (block); |
0 | 927 put_pixels_clamped(block, dest, line_size); |
928 } | |
929 | |
930 /* add block[] to dest[] */ | |
931 static inline void add_dct(MpegEncContext *s, | |
932 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
933 { | |
934 if (s->block_last_index[i] >= 0) { | |
935 if (!s->mpeg2) | |
206 | 936 if(s->encoding || (!s->h263_msmpeg4)) |
200 | 937 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
|
938 ff_idct (block); |
0 | 939 add_pixels_clamped(block, dest, line_size); |
940 } | |
941 } | |
942 | |
943 /* generic function called after a macroblock has been parsed by the | |
944 decoder or after it has been encoded by the encoder. | |
945 | |
946 Important variables used: | |
947 s->mb_intra : true if intra macroblock | |
948 s->mv_dir : motion vector direction | |
949 s->mv_type : motion vector type | |
950 s->mv : motion vector | |
951 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
952 */ | |
953 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
954 { | |
244 | 955 int mb_x, mb_y; |
0 | 956 int dct_linesize, dct_offset; |
957 op_pixels_func *op_pix; | |
255 | 958 qpel_mc_func *op_qpix; |
0 | 959 |
960 mb_x = s->mb_x; | |
961 mb_y = s->mb_y; | |
962 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
963 #ifdef FF_POSTPROCESS |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
964 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
965 //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
|
966 #endif |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
967 |
0 | 968 /* update DC predictors for P macroblocks */ |
969 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
970 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
|
971 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
|
972 { |
244 | 973 int wrap, xy, v; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
974 s->mbintra_table[mb_x + mb_y*s->mb_width]=0; |
0 | 975 wrap = 2 * s->mb_width + 2; |
244 | 976 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 977 v = 1024; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
978 |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
979 s->dc_val[0][xy] = v; |
244 | 980 s->dc_val[0][xy + 1] = v; |
981 s->dc_val[0][xy + wrap] = v; | |
982 s->dc_val[0][xy + 1 + wrap] = v; | |
0 | 983 /* ac pred */ |
244 | 984 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16)); |
985 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16)); | |
986 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16)); | |
987 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16)); | |
0 | 988 if (s->h263_msmpeg4) { |
244 | 989 s->coded_block[xy] = 0; |
990 s->coded_block[xy + 1] = 0; | |
991 s->coded_block[xy + wrap] = 0; | |
992 s->coded_block[xy + 1 + wrap] = 0; | |
0 | 993 } |
994 /* chroma */ | |
995 wrap = s->mb_width + 2; | |
244 | 996 xy = mb_x + 1 + (mb_y + 1) * wrap; |
997 s->dc_val[1][xy] = v; | |
998 s->dc_val[2][xy] = v; | |
0 | 999 /* ac pred */ |
244 | 1000 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); |
1001 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
|
1002 } |
0 | 1003 } else { |
1004 s->last_dc[0] = 128 << s->intra_dc_precision; | |
1005 s->last_dc[1] = 128 << s->intra_dc_precision; | |
1006 s->last_dc[2] = 128 << s->intra_dc_precision; | |
1007 } | |
1008 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1009 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
|
1010 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
|
1011 |
280 | 1012 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */ |
1013 if (s->out_format == FMT_H263) { | |
262 | 1014 if(s->pict_type!=B_TYPE){ |
244 | 1015 int xy, wrap, motion_x, motion_y; |
0 | 1016 |
1017 wrap = 2 * s->mb_width + 2; | |
244 | 1018 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 1019 if (s->mb_intra) { |
1020 motion_x = 0; | |
1021 motion_y = 0; | |
1022 goto motion_init; | |
1023 } else if (s->mv_type == MV_TYPE_16X16) { | |
1024 motion_x = s->mv[0][0][0]; | |
1025 motion_y = s->mv[0][0][1]; | |
1026 motion_init: | |
1027 /* no update if 8X8 because it has been done during parsing */ | |
244 | 1028 s->motion_val[xy][0] = motion_x; |
1029 s->motion_val[xy][1] = motion_y; | |
1030 s->motion_val[xy + 1][0] = motion_x; | |
1031 s->motion_val[xy + 1][1] = motion_y; | |
1032 s->motion_val[xy + wrap][0] = motion_x; | |
1033 s->motion_val[xy + wrap][1] = motion_y; | |
1034 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
1035 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
0 | 1036 } |
262 | 1037 } |
0 | 1038 } |
1039 | |
1040 if (!s->intra_only) { | |
1041 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
|
1042 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1043 |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1044 /* avoid copy if macroblock skipped in last frame too */ |
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
|
1045 if (!s->encoding && s->pict_type != B_TYPE) { |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1046 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
|
1047 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1048 s->mb_skiped = 0; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1049 /* if previous was skipped too, then nothing to do ! */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1050 if (*mbskip_ptr != 0) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1051 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1052 *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
|
1053 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1054 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1055 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1056 } |
0 | 1057 |
1058 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
1059 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1060 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1061 | |
1062 if (s->interlaced_dct) { | |
1063 dct_linesize = s->linesize * 2; | |
1064 dct_offset = s->linesize; | |
1065 } else { | |
1066 dct_linesize = s->linesize; | |
1067 dct_offset = s->linesize * 8; | |
1068 } | |
1069 | |
1070 if (!s->mb_intra) { | |
1071 /* motion handling */ | |
255 | 1072 if (!s->no_rounding){ |
0 | 1073 op_pix = put_pixels_tab; |
255 | 1074 op_qpix= qpel_mc_rnd_tab; |
1075 }else{ | |
0 | 1076 op_pix = put_no_rnd_pixels_tab; |
255 | 1077 op_qpix= qpel_mc_no_rnd_tab; |
1078 } | |
0 | 1079 |
1080 if (s->mv_dir & MV_DIR_FORWARD) { | |
255 | 1081 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); |
0 | 1082 if (!s->no_rounding) |
1083 op_pix = avg_pixels_tab; | |
1084 else | |
1085 op_pix = avg_no_rnd_pixels_tab; | |
1086 } | |
1087 if (s->mv_dir & MV_DIR_BACKWARD) { | |
255 | 1088 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); |
0 | 1089 } |
1090 | |
1091 /* add dct residue */ | |
1092 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
1093 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1094 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1095 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1096 | |
57 | 1097 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1098 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1099 } else { |
1100 /* dct only in intra block */ | |
1101 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
1102 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1103 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1104 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1105 | |
57 | 1106 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1107 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1108 } |
1109 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1110 the_end: |
294 | 1111 emms_c(); //FIXME remove |
1112 } | |
1113 | |
1114 static void encode_mb(MpegEncContext *s) | |
1115 { | |
1116 int wrap; | |
1117 const int mb_x= s->mb_x; | |
1118 const int mb_y= s->mb_y; | |
1119 UINT8 *ptr; | |
1120 const int motion_x= s->mv[0][0][0]; | |
1121 const int motion_y= s->mv[0][0][1]; | |
1122 int i; | |
1123 | |
1124 /* get the pixels */ | |
1125 wrap = s->linesize; | |
1126 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
1127 get_pixels(s->block[0], ptr, wrap); | |
1128 get_pixels(s->block[1], ptr + 8, wrap); | |
1129 get_pixels(s->block[2], ptr + 8 * wrap, wrap); | |
1130 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); | |
1131 wrap = s->linesize >> 1; | |
1132 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
1133 get_pixels(s->block[4], ptr, wrap); | |
1134 | |
1135 wrap = s->linesize >> 1; | |
1136 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; | |
1137 get_pixels(s->block[5], ptr, wrap); | |
1138 | |
1139 /* subtract previous frame if non intra */ | |
1140 if (!s->mb_intra) { | |
1141 int dxy, offset, mx, my; | |
295 | 1142 |
1143 if(s->mv_type==MV_TYPE_16X16){ | |
1144 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
1145 ptr = s->last_picture[0] + | |
1146 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) + | |
1147 (mb_x * 16 + (motion_x >> 1)); | |
294 | 1148 |
295 | 1149 sub_pixels_2(s->block[0], ptr, s->linesize, dxy); |
1150 sub_pixels_2(s->block[1], ptr + 8, s->linesize, dxy); | |
1151 sub_pixels_2(s->block[2], ptr + s->linesize * 8, s->linesize, dxy); | |
1152 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy); | |
294 | 1153 |
295 | 1154 if (s->out_format == FMT_H263) { |
1155 /* special rounding for h263 */ | |
1156 dxy = 0; | |
1157 if ((motion_x & 3) != 0) | |
1158 dxy |= 1; | |
1159 if ((motion_y & 3) != 0) | |
1160 dxy |= 2; | |
1161 mx = motion_x >> 2; | |
1162 my = motion_y >> 2; | |
1163 } else { | |
1164 mx = motion_x / 2; | |
1165 my = motion_y / 2; | |
1166 dxy = ((my & 1) << 1) | (mx & 1); | |
1167 mx >>= 1; | |
1168 my >>= 1; | |
1169 } | |
1170 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx); | |
1171 ptr = s->last_picture[1] + offset; | |
1172 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); | |
1173 ptr = s->last_picture[2] + offset; | |
1174 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); | |
1175 }else{ | |
1176 int src_x, src_y; | |
1177 | |
1178 for(i=0;i<4;i++) { | |
1179 int motion_x = s->mv[0][i][0]; | |
1180 int motion_y = s->mv[0][i][1]; | |
1181 | |
1182 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
1183 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
1184 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; | |
1185 | |
1186 ptr = s->last_picture[0] + (src_y * s->linesize) + (src_x); | |
1187 sub_pixels_2(s->block[i], ptr, s->linesize, dxy); | |
1188 } | |
1189 /* In case of 8X8, we construct a single chroma motion vector | |
1190 with a special rounding */ | |
1191 mx = 0; | |
1192 my = 0; | |
1193 for(i=0;i<4;i++) { | |
1194 mx += s->mv[0][i][0]; | |
1195 my += s->mv[0][i][1]; | |
1196 } | |
1197 if (mx >= 0) | |
1198 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1199 else { | |
1200 mx = -mx; | |
1201 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1202 } | |
1203 if (my >= 0) | |
1204 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1205 else { | |
1206 my = -my; | |
1207 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1208 } | |
294 | 1209 dxy = ((my & 1) << 1) | (mx & 1); |
1210 mx >>= 1; | |
1211 my >>= 1; | |
295 | 1212 |
1213 src_x = mb_x * 8 + mx; | |
1214 src_y = mb_y * 8 + my; | |
1215 src_x = clip(src_x, -8, s->width/2); | |
1216 if (src_x == s->width/2) | |
1217 dxy &= ~1; | |
1218 src_y = clip(src_y, -8, s->height/2); | |
1219 if (src_y == s->height/2) | |
1220 dxy &= ~2; | |
1221 | |
1222 offset = (src_y * (s->linesize >> 1)) + src_x; | |
1223 ptr = s->last_picture[1] + offset; | |
1224 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); | |
1225 ptr = s->last_picture[2] + offset; | |
1226 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); | |
294 | 1227 } |
1228 } | |
1229 | |
1230 #if 0 | |
1231 { | |
1232 float adap_parm; | |
1233 | |
1234 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / | |
1235 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); | |
1236 | |
1237 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", | |
1238 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', | |
1239 s->qscale, adap_parm, s->qscale*adap_parm, | |
1240 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); | |
1241 } | |
1242 #endif | |
1243 /* DCT & quantize */ | |
1244 if (s->h263_msmpeg4) { | |
1245 msmpeg4_dc_scale(s); | |
1246 } else if (s->h263_pred) { | |
1247 h263_dc_scale(s); | |
1248 } else { | |
1249 /* default quantization values */ | |
1250 s->y_dc_scale = 8; | |
1251 s->c_dc_scale = 8; | |
1252 } | |
1253 for(i=0;i<6;i++) { | |
1254 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); | |
1255 } | |
1256 | |
1257 /* huffman encode */ | |
1258 switch(s->out_format) { | |
1259 case FMT_MPEG1: | |
1260 mpeg1_encode_mb(s, s->block, motion_x, motion_y); | |
1261 break; | |
1262 case FMT_H263: | |
1263 if (s->h263_msmpeg4) | |
1264 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1265 else if(s->h263_pred) | |
1266 mpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1267 else | |
1268 h263_encode_mb(s, s->block, motion_x, motion_y); | |
1269 break; | |
1270 case FMT_MJPEG: | |
1271 mjpeg_encode_mb(s, s->block); | |
1272 break; | |
1273 } | |
1274 } | |
1275 | |
1276 static void copy_bits(PutBitContext *pb, UINT8 *src, int length) | |
1277 { | |
1278 int bytes= length>>3; | |
1279 int bits= length&7; | |
1280 int i; | |
1281 | |
1282 for(i=0; i<bytes; i++) put_bits(pb, 8, src[i]); | |
1283 put_bits(pb, bits, src[i]>>(8-bits)); | |
0 | 1284 } |
1285 | |
1286 static void encode_picture(MpegEncContext *s, int picture_number) | |
1287 { | |
294 | 1288 int mb_x, mb_y, last_gob, pdif = 0; |
1289 int i; | |
286 | 1290 int bits; |
294 | 1291 MpegEncContext best_s; |
1292 UINT8 bit_buf[4][3000]; //FIXME check that this is ALLWAYS large enogh for a MB | |
0 | 1293 |
1294 s->picture_number = picture_number; | |
268 | 1295 |
294 | 1296 s->block_wrap[0]= |
1297 s->block_wrap[1]= | |
1298 s->block_wrap[2]= | |
1299 s->block_wrap[3]= s->mb_width*2 + 2; | |
1300 s->block_wrap[4]= | |
1301 s->block_wrap[5]= s->mb_width + 2; | |
1302 | |
268 | 1303 s->last_mc_mb_var = s->mc_mb_var; |
1304 /* Reset the average MB variance */ | |
1305 s->avg_mb_var = 0; | |
1306 s->mc_mb_var = 0; | |
1307 /* Estimate motion for every MB */ | |
294 | 1308 if(s->pict_type == P_TYPE){ |
1309 for(mb_y=0; mb_y < s->mb_height; mb_y++) { | |
1310 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; | |
1311 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1312 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1313 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1314 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
1315 s->mb_x = mb_x; | |
1316 s->mb_y = mb_y; | |
1317 s->block_index[0]+=2; | |
1318 s->block_index[1]+=2; | |
1319 s->block_index[2]+=2; | |
1320 s->block_index[3]+=2; | |
268 | 1321 |
294 | 1322 /* compute motion vector & mb_type and store in context */ |
1323 estimate_motion(s, mb_x, mb_y); | |
1324 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER; | |
268 | 1325 } |
1326 } | |
294 | 1327 emms_c(); |
1328 }else{ | |
1329 /* I-Frame */ | |
1330 //FIXME do we need to zero them? | |
1331 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2); | |
1332 memset(s->mv_table[0] , 0, sizeof(INT16)*s->mb_width*s->mb_height); | |
1333 memset(s->mv_table[1] , 0, sizeof(INT16)*s->mb_width*s->mb_height); | |
1334 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); | |
268 | 1335 } |
1336 | |
297 | 1337 if(s->avg_mb_var < s->mc_mb_var && s->pict_type != B_TYPE && (!s->force_type)){ //FIXME subtract MV bits |
271 | 1338 s->pict_type= I_TYPE; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1339 s->picture_in_gop_number=0; |
294 | 1340 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
1341 //printf("Scene change detected, encoding as I Frame\n"); | |
271 | 1342 } |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1343 |
287 | 1344 /* find best f_code for ME which do unlimited searches */ |
1345 if(s->pict_type==P_TYPE && s->full_search>3){ | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1346 int mv_num[8]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1347 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1348 int loose=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1349 UINT8 * fcode_tab= s->fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1350 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1351 for(i=0; i<8; i++) mv_num[i]=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1352 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1353 for(i=0; i<s->mb_num; i++){ |
295 | 1354 if(s->mb_type[i] & MB_TYPE_INTER){ |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1355 mv_num[ fcode_tab[s->mv_table[0][i] + MAX_MV] ]++; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1356 mv_num[ fcode_tab[s->mv_table[1][i] + MAX_MV] ]++; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1357 //printf("%d %d %d\n", s->mv_table[0][i], fcode_tab[s->mv_table[0][i] + MAX_MV], i); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1358 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1359 //else printf("I"); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1360 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1361 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1362 for(i=MAX_FCODE; i>1; i--){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1363 loose+= mv_num[i]; |
286 | 1364 if(loose > 10) break; //FIXME this is pretty ineffective |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1365 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1366 s->f_code= i; |
287 | 1367 /* for(i=0; i<=MAX_FCODE; i++){ |
1368 printf("%d ", mv_num[i]); | |
1369 } | |
1370 printf("\n");*/ | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1371 }else{ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1372 s->f_code= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1373 } |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
1374 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1375 //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
|
1376 /* convert MBs with too long MVs to I-Blocks */ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1377 if(s->pict_type==P_TYPE){ |
295 | 1378 int i, x, y; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1379 const int f_code= s->f_code; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1380 UINT8 * fcode_tab= s->fcode_tab; |
295 | 1381 //FIXME try to clip instead of intra izing ;) |
1382 /* clip / convert to intra 16x16 type MVs */ | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1383 for(i=0; i<s->mb_num; i++){ |
294 | 1384 if(s->mb_type[i]&MB_TYPE_INTER){ |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1385 if( fcode_tab[s->mv_table[0][i] + MAX_MV] > f_code |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1386 || fcode_tab[s->mv_table[0][i] + MAX_MV] == 0 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1387 || fcode_tab[s->mv_table[1][i] + MAX_MV] > f_code |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1388 || fcode_tab[s->mv_table[1][i] + MAX_MV] == 0 ){ |
294 | 1389 s->mb_type[i] &= ~MB_TYPE_INTER; |
1390 s->mb_type[i] |= MB_TYPE_INTRA; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1391 s->mv_table[0][i] = 0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1392 s->mv_table[1][i] = 0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1393 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1394 } |
295 | 1395 } |
1396 | |
1397 if(s->flags&CODEC_FLAG_4MV){ | |
1398 int wrap= 2+ s->mb_width*2; | |
1399 | |
1400 /* clip / convert to intra 8x8 type MVs */ | |
1401 for(y=0; y<s->mb_height; y++){ | |
1402 int xy= (y*2 + 1)*wrap + 1; | |
1403 i= y*s->mb_width; | |
1404 | |
1405 for(x=0; x<s->mb_width; x++){ | |
1406 if(s->mb_type[i]&MB_TYPE_INTER4V){ | |
1407 int block; | |
1408 for(block=0; block<4; block++){ | |
1409 int off= (block& 1) + (block>>1)*wrap; | |
1410 int mx= s->motion_val[ xy + off ][0]; | |
1411 int my= s->motion_val[ xy + off ][1]; | |
1412 | |
1413 if( fcode_tab[mx + MAX_MV] > f_code | |
1414 || fcode_tab[mx + MAX_MV] == 0 | |
1415 || fcode_tab[my + MAX_MV] > f_code | |
1416 || fcode_tab[my + MAX_MV] == 0 ){ | |
1417 s->mb_type[i] &= ~MB_TYPE_INTER4V; | |
1418 s->mb_type[i] |= MB_TYPE_INTRA; | |
1419 } | |
1420 } | |
1421 xy+=2; | |
1422 i++; | |
1423 } | |
1424 } | |
294 | 1425 } |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1426 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1427 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1428 |
271 | 1429 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var); |
1430 | |
0 | 1431 if (!s->fixed_qscale) |
1432 s->qscale = rate_estimate_qscale(s); | |
1433 | |
1434 /* precompute matrix */ | |
1435 if (s->out_format == FMT_MJPEG) { | |
1436 /* for mjpeg, we do include qscale in the matrix */ | |
1437 s->intra_matrix[0] = default_intra_matrix[0]; | |
1438 for(i=1;i<64;i++) | |
1439 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
220 | 1440 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
0 | 1441 } else { |
220 | 1442 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
1443 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); | |
0 | 1444 } |
1445 | |
286 | 1446 s->last_bits= get_bit_count(&s->pb); |
0 | 1447 switch(s->out_format) { |
1448 case FMT_MJPEG: | |
1449 mjpeg_picture_header(s); | |
1450 break; | |
1451 case FMT_H263: | |
1452 if (s->h263_msmpeg4) | |
1453 msmpeg4_encode_picture_header(s, picture_number); | |
1454 else if (s->h263_pred) | |
1455 mpeg4_encode_picture_header(s, picture_number); | |
1456 else if (s->h263_rv10) | |
1457 rv10_encode_picture_header(s, picture_number); | |
1458 else | |
1459 h263_encode_picture_header(s, picture_number); | |
1460 break; | |
1461 case FMT_MPEG1: | |
1462 mpeg1_encode_picture_header(s, picture_number); | |
1463 break; | |
1464 } | |
286 | 1465 bits= get_bit_count(&s->pb); |
1466 s->header_bits= bits - s->last_bits; | |
1467 s->last_bits= bits; | |
1468 s->mv_bits=0; | |
1469 s->misc_bits=0; | |
1470 s->i_tex_bits=0; | |
1471 s->p_tex_bits=0; | |
1472 s->i_count=0; | |
1473 s->p_count=0; | |
1474 s->skip_count=0; | |
1475 | |
0 | 1476 /* init last dc values */ |
1477 /* note: quant matrix value (8) is implied here */ | |
1478 s->last_dc[0] = 128; | |
1479 s->last_dc[1] = 128; | |
1480 s->last_dc[2] = 128; | |
1481 s->mb_incr = 1; | |
1482 s->last_mv[0][0][0] = 0; | |
1483 s->last_mv[0][0][1] = 0; | |
1484 | |
162 | 1485 /* Get the GOB height based on picture height */ |
231 | 1486 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
162 | 1487 if (s->height <= 400) |
1488 s->gob_index = 1; | |
1489 else if (s->height <= 800) | |
1490 s->gob_index = 2; | |
1491 else | |
1492 s->gob_index = 4; | |
1493 } | |
268 | 1494 |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1495 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
|
1496 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1497 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
|
1498 /* 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
|
1499 /* 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
|
1500 if (s->rtp_mode) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1501 if (!mb_y) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1502 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
|
1503 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
|
1504 } 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
|
1505 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
|
1506 if (last_gob) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1507 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
|
1508 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1509 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1510 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1511 |
266 | 1512 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; |
1513 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1514 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1515 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1516 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
1517 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
|
1518 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
294 | 1519 const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x]; |
1520 PutBitContext pb; | |
1521 int d; | |
1522 int dmin=10000000; | |
1523 int best=0; | |
0 | 1524 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1525 s->mb_x = mb_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1526 s->mb_y = mb_y; |
266 | 1527 s->block_index[0]+=2; |
1528 s->block_index[1]+=2; | |
1529 s->block_index[2]+=2; | |
1530 s->block_index[3]+=2; | |
1531 s->block_index[4]++; | |
1532 s->block_index[5]++; | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1533 |
294 | 1534 s->mv_dir = MV_DIR_FORWARD; |
1535 if(mb_type & (mb_type-1)){ // more than 1 MB type possible | |
1536 pb= s->pb; | |
1537 if(mb_type&MB_TYPE_INTER){ | |
295 | 1538 s->mv_type = MV_TYPE_16X16; |
294 | 1539 s->mb_intra= 0; |
1540 s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x]; | |
1541 s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x]; | |
1542 init_put_bits(&s->pb, bit_buf[1], 3000, NULL, NULL); | |
1543 s->block= s->inter_block; | |
0 | 1544 |
294 | 1545 encode_mb(s); |
1546 d= get_bit_count(&s->pb); | |
1547 if(d<dmin){ | |
1548 flush_put_bits(&s->pb); | |
1549 dmin=d; | |
1550 best_s.mv[0][0][0]= s->mv[0][0][0]; | |
1551 best_s.mv[0][0][1]= s->mv[0][0][1]; | |
1552 best_s.mb_intra= 0; | |
295 | 1553 best_s.mv_type = MV_TYPE_16X16; |
294 | 1554 best_s.pb=s->pb; |
1555 best_s.block= s->block; | |
1556 best=1; | |
1557 for(i=0; i<6; i++) | |
1558 best_s.block_last_index[i]= s->block_last_index[i]; | |
1559 } | |
0 | 1560 } |
295 | 1561 if(mb_type&MB_TYPE_INTER4V){ |
1562 s->mv_type = MV_TYPE_8X8; | |
1563 s->mb_intra= 0; | |
1564 for(i=0; i<4; i++){ | |
1565 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
1566 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
1567 } | |
1568 init_put_bits(&s->pb, bit_buf[2], 3000, NULL, NULL); | |
1569 s->block= s->inter4v_block; | |
1570 | |
1571 encode_mb(s); | |
1572 d= get_bit_count(&s->pb); | |
1573 if(d<dmin){ | |
1574 flush_put_bits(&s->pb); | |
1575 dmin=d; | |
1576 for(i=0; i<4; i++){ | |
1577 best_s.mv[0][i][0] = s->mv[0][i][0]; | |
1578 best_s.mv[0][i][1] = s->mv[0][i][1]; | |
1579 } | |
1580 best_s.mb_intra= 0; | |
1581 best_s.mv_type = MV_TYPE_8X8; | |
1582 best_s.pb=s->pb; | |
1583 best_s.block= s->block; | |
1584 best=2; | |
1585 for(i=0; i<6; i++) | |
1586 best_s.block_last_index[i]= s->block_last_index[i]; | |
1587 } | |
1588 } | |
294 | 1589 if(mb_type&MB_TYPE_INTRA){ |
295 | 1590 s->mv_type = MV_TYPE_16X16; |
294 | 1591 s->mb_intra= 1; |
1592 s->mv[0][0][0] = 0; | |
1593 s->mv[0][0][1] = 0; | |
1594 init_put_bits(&s->pb, bit_buf[0], 3000, NULL, NULL); | |
1595 s->block= s->intra_block; | |
1596 | |
1597 encode_mb(s); | |
1598 d= get_bit_count(&s->pb); | |
1599 if(d<dmin){ | |
1600 flush_put_bits(&s->pb); | |
1601 dmin=d; | |
1602 best_s.mv[0][0][0]= 0; | |
1603 best_s.mv[0][0][1]= 0; | |
1604 best_s.mb_intra= 1; | |
295 | 1605 best_s.mv_type = MV_TYPE_16X16; |
294 | 1606 best_s.pb=s->pb; |
1607 best_s.block= s->block; | |
1608 for(i=0; i<6; i++) | |
1609 best_s.block_last_index[i]= s->block_last_index[i]; | |
1610 best=0; | |
1611 } | |
1612 /* force cleaning of ac/dc if needed ... */ | |
1613 s->mbintra_table[mb_x + mb_y*s->mb_width]=1; | |
1614 } | |
295 | 1615 for(i=0; i<4; i++){ |
1616 s->mv[0][i][0] = best_s.mv[0][i][0]; | |
1617 s->mv[0][i][1] = best_s.mv[0][i][1]; | |
1618 } | |
294 | 1619 s->mb_intra= best_s.mb_intra; |
295 | 1620 s->mv_type= best_s.mv_type; |
294 | 1621 for(i=0; i<6; i++) |
1622 s->block_last_index[i]= best_s.block_last_index[i]; | |
1623 copy_bits(&pb, bit_buf[best], dmin); | |
1624 s->block= best_s.block; | |
1625 s->pb= pb; | |
1626 } else { | |
1627 // only one MB-Type possible | |
1628 if(mb_type&MB_TYPE_INTRA){ | |
1629 s->mb_intra= 1; | |
1630 s->mv[0][0][0] = 0; | |
1631 s->mv[0][0][1] = 0; | |
1632 }else{ | |
1633 s->mb_intra= 0; | |
1634 s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x]; | |
1635 s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x]; | |
1636 } | |
1637 encode_mb(s); | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1638 } |
0 | 1639 |
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
|
1640 MPV_decode_mb(s, s->block); |
0 | 1641 } |
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
|
1642 |
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
|
1643 |
231 | 1644 /* Obtain average GOB size for RTP */ |
1645 if (s->rtp_mode) { | |
1646 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
|
1647 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
231 | 1648 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
|
1649 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
|
1650 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
231 | 1651 } |
1652 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
1653 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
1654 s->first_gob_line = 0; | |
1655 } | |
0 | 1656 } |
294 | 1657 emms_c(); |
286 | 1658 |
311 | 1659 if (s->h263_msmpeg4 && s->msmpeg4_version<4 && s->pict_type == I_TYPE) |
208 | 1660 msmpeg4_encode_ext_header(s); |
1661 | |
162 | 1662 //if (s->gob_number) |
1663 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
231 | 1664 |
1665 /* Send the last GOB if RTP */ | |
1666 if (s->rtp_mode) { | |
1667 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
|
1668 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 1669 /* Call the RTP callback to send the last GOB */ |
1670 if (s->rtp_callback) | |
1671 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
|
1672 s->ptr_lastgob = pbBufPtr(&s->pb); |
231 | 1673 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
1674 } | |
0 | 1675 } |
1676 | |
220 | 1677 static int dct_quantize_c(MpegEncContext *s, |
0 | 1678 DCTELEM *block, int n, |
1679 int qscale) | |
1680 { | |
1681 int i, j, level, last_non_zero, q; | |
1682 const int *qmat; | |
216 | 1683 int minLevel, maxLevel; |
1684 | |
1685 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ | |
1686 /* mpeg4 */ | |
1687 minLevel= -2048; | |
1688 maxLevel= 2047; | |
1689 }else if(s->out_format==FMT_MPEG1){ | |
1690 /* mpeg1 */ | |
1691 minLevel= -255; | |
1692 maxLevel= 255; | |
230 | 1693 }else if(s->out_format==FMT_MJPEG){ |
1694 /* (m)jpeg */ | |
1695 minLevel= -1023; | |
1696 maxLevel= 1023; | |
216 | 1697 }else{ |
1698 /* h263 / msmpeg4 */ | |
1699 minLevel= -128; | |
1700 maxLevel= 127; | |
1701 } | |
0 | 1702 |
1703 av_fdct (block); | |
1704 | |
64 | 1705 /* we need this permutation so that we correct the IDCT |
1706 permutation. will be moved into DCT code */ | |
1707 block_permute(block); | |
1708 | |
0 | 1709 if (s->mb_intra) { |
1710 if (n < 4) | |
1711 q = s->y_dc_scale; | |
1712 else | |
1713 q = s->c_dc_scale; | |
1714 q = q << 3; | |
1715 | |
1716 /* note: block[0] is assumed to be positive */ | |
1717 block[0] = (block[0] + (q >> 1)) / q; | |
1718 i = 1; | |
1719 last_non_zero = 0; | |
1720 if (s->out_format == FMT_H263) { | |
1721 qmat = s->q_non_intra_matrix; | |
1722 } else { | |
1723 qmat = s->q_intra_matrix; | |
1724 } | |
1725 } else { | |
1726 i = 0; | |
1727 last_non_zero = -1; | |
1728 qmat = s->q_non_intra_matrix; | |
1729 } | |
1730 | |
1731 for(;i<64;i++) { | |
1732 j = zigzag_direct[i]; | |
1733 level = block[j]; | |
1734 level = level * qmat[j]; | |
1735 #ifdef PARANOID | |
1736 { | |
1737 static int count = 0; | |
1738 int level1, level2, qmat1; | |
1739 double val; | |
1740 if (qmat == s->q_non_intra_matrix) { | |
1741 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1742 } else { | |
1743 qmat1 = default_intra_matrix[j] * s->qscale; | |
1744 } | |
1745 if (av_fdct != jpeg_fdct_ifast) | |
1746 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1747 else | |
1748 val = ((double)block[j] * 8.0 * 2048.0) / | |
1749 ((double)qmat1 * aanscales[j]); | |
1750 level1 = (int)val; | |
1751 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1752 if (level1 != level2) { | |
1753 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1754 count, level2, level1, block[j], qmat1, qmat[j], | |
1755 val); | |
1756 count++; | |
1757 } | |
1758 | |
1759 } | |
1760 #endif | |
1761 /* XXX: slight error for the low range. Test should be equivalent to | |
1762 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1763 (QMAT_SHIFT - 3))) | |
1764 */ | |
1765 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1766 level) { | |
1767 level = level / (1 << (QMAT_SHIFT - 3)); | |
1768 /* XXX: currently, this code is not optimal. the range should be: | |
1769 mpeg1: -255..255 | |
1770 mpeg2: -2048..2047 | |
1771 h263: -128..127 | |
1772 mpeg4: -2048..2047 | |
1773 */ | |
216 | 1774 if (level > maxLevel) |
1775 level = maxLevel; | |
1776 else if (level < minLevel) | |
1777 level = minLevel; | |
0 | 1778 |
1779 block[j] = level; | |
1780 last_non_zero = i; | |
1781 } else { | |
1782 block[j] = 0; | |
1783 } | |
1784 } | |
1785 return last_non_zero; | |
1786 } | |
1787 | |
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
|
1788 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
|
1789 DCTELEM *block, int n, int qscale) |
0 | 1790 { |
200 | 1791 int i, level, nCoeffs; |
0 | 1792 const UINT16 *quant_matrix; |
1793 | |
200 | 1794 if(s->alternate_scan) nCoeffs= 64; |
1795 else nCoeffs= s->block_last_index[n]+1; | |
1796 | |
0 | 1797 if (s->mb_intra) { |
1798 if (n < 4) | |
1799 block[0] = block[0] * s->y_dc_scale; | |
1800 else | |
1801 block[0] = block[0] * s->c_dc_scale; | |
1802 /* XXX: only mpeg1 */ | |
1803 quant_matrix = s->intra_matrix; | |
200 | 1804 for(i=1;i<nCoeffs;i++) { |
1805 int j= zigzag_direct[i]; | |
1806 level = block[j]; | |
0 | 1807 if (level) { |
1808 if (level < 0) { | |
1809 level = -level; | |
200 | 1810 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1811 level = (level - 1) | 1; |
1812 level = -level; | |
1813 } else { | |
200 | 1814 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1815 level = (level - 1) | 1; |
1816 } | |
1817 #ifdef PARANOID | |
1818 if (level < -2048 || level > 2047) | |
1819 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1820 #endif | |
200 | 1821 block[j] = level; |
0 | 1822 } |
1823 } | |
1824 } else { | |
1825 i = 0; | |
1826 quant_matrix = s->non_intra_matrix; | |
217 | 1827 for(;i<nCoeffs;i++) { |
200 | 1828 int j= zigzag_direct[i]; |
1829 level = block[j]; | |
0 | 1830 if (level) { |
1831 if (level < 0) { | |
1832 level = -level; | |
1833 level = (((level << 1) + 1) * qscale * | |
200 | 1834 ((int) (quant_matrix[j]))) >> 4; |
0 | 1835 level = (level - 1) | 1; |
1836 level = -level; | |
1837 } else { | |
1838 level = (((level << 1) + 1) * qscale * | |
200 | 1839 ((int) (quant_matrix[j]))) >> 4; |
0 | 1840 level = (level - 1) | 1; |
1841 } | |
1842 #ifdef PARANOID | |
1843 if (level < -2048 || level > 2047) | |
1844 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1845 #endif | |
200 | 1846 block[j] = level; |
0 | 1847 } |
1848 } | |
1849 } | |
1850 } | |
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
|
1851 |
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
|
1852 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
|
1853 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
|
1854 { |
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
|
1855 int i, level, qmul, qadd; |
200 | 1856 int nCoeffs; |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1857 |
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
|
1858 if (s->mb_intra) { |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1859 if (!s->h263_aic) { |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1860 if (n < 4) |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1861 block[0] = block[0] * s->y_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1862 else |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1863 block[0] = block[0] * s->c_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1864 } |
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
|
1865 i = 1; |
200 | 1866 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
|
1867 } 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
|
1868 i = 0; |
200 | 1869 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
|
1870 } |
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
|
1871 |
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
|
1872 qmul = s->qscale << 1; |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1873 if (s->h263_aic && s->mb_intra) |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1874 qadd = 0; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1875 else |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1876 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
|
1877 |
200 | 1878 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
|
1879 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
|
1880 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
|
1881 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
|
1882 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
|
1883 } 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
|
1884 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
|
1885 } |
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 #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
|
1887 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
|
1888 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
|
1889 #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
|
1890 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
|
1891 } |
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
|
1892 } |
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
|
1893 } |
0 | 1894 |
1895 /* rate control */ | |
1896 | |
1897 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */ | |
1898 #define I_FRAME_SIZE_RATIO 3.0 | |
1899 #define QSCALE_K 20 | |
1900 | |
1901 static void rate_control_init(MpegEncContext *s) | |
1902 { | |
268 | 1903 #if 1 |
1904 emms_c(); | |
1905 | |
1906 //initial values, they dont really matter as they will be totally different within a few frames | |
1907 s->i_pred.coeff= s->p_pred.coeff= 7.0; | |
1908 s->i_pred.count= s->p_pred.count= 1.0; | |
1909 | |
1910 s->i_pred.decay= s->p_pred.decay= 0.4; | |
1911 | |
1912 // use more bits at the beginning, otherwise high motion at the begin will look like shit | |
1913 s->qsum=100; | |
1914 s->qcount=100; | |
1915 | |
1916 s->short_term_qsum=0.001; | |
1917 s->short_term_qcount=0.001; | |
1918 #else | |
0 | 1919 s->wanted_bits = 0; |
1920 | |
1921 if (s->intra_only) { | |
1922 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate; | |
1923 s->P_frame_bits = s->I_frame_bits; | |
1924 } else { | |
1925 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) / | |
1926 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1))); | |
1927 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO); | |
1928 } | |
268 | 1929 |
0 | 1930 #if defined(DEBUG) |
1931 printf("I_frame_size=%d P_frame_size=%d\n", | |
1932 s->I_frame_bits, s->P_frame_bits); | |
1933 #endif | |
268 | 1934 #endif |
1935 } | |
1936 | |
1937 static double predict(Predictor *p, double q, double var) | |
1938 { | |
1939 return p->coeff*var / (q*p->count); | |
1940 } | |
1941 | |
1942 static void update_predictor(Predictor *p, double q, double var, double size) | |
1943 { | |
1944 double new_coeff= size*q / (var + 1); | |
1945 if(var<1000) return; | |
1946 /*{ | |
1947 int pred= predict(p, q, var); | |
1948 int error= abs(pred-size); | |
1949 static double sum=0; | |
1950 static int count=0; | |
1951 if(count>5) sum+=error; | |
1952 count++; | |
1953 if(256*256*256*64%count==0){ | |
1954 printf("%d %f %f\n", count, sum/count, p->coeff); | |
1955 } | |
1956 }*/ | |
1957 p->count*= p->decay; | |
1958 p->coeff*= p->decay; | |
1959 p->count++; | |
1960 p->coeff+= new_coeff; | |
0 | 1961 } |
1962 | |
1963 static int rate_estimate_qscale(MpegEncContext *s) | |
1964 { | |
268 | 1965 #if 1 |
1966 int qmin= s->qmin; | |
1967 int qmax= s->qmax; | |
1968 int rate_q=5; | |
1969 float q; | |
1970 int qscale; | |
1971 float br_compensation; | |
1972 double diff; | |
1973 double short_term_q; | |
1974 double long_term_q; | |
1975 int last_qscale= s->qscale; | |
1976 double fps; | |
1977 INT64 wanted_bits; | |
1978 emms_c(); | |
297 | 1979 |
268 | 1980 fps= (double)s->frame_rate / FRAME_RATE_BASE; |
1981 wanted_bits= s->bit_rate*(double)s->picture_number/fps; | |
1982 | |
1983 | |
1984 if(s->picture_number>2){ | |
1985 /* update predictors */ | |
1986 if(s->last_pict_type == I_TYPE){ | |
1987 //FIXME | |
1988 }else{ //P Frame | |
1989 //printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff); | |
1990 update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits); | |
1991 } | |
1992 } | |
1993 | |
1994 if(s->pict_type == I_TYPE){ | |
1995 //FIXME | |
1996 rate_q= s->qsum/s->qcount; | |
1997 }else{ //P Frame | |
1998 int i; | |
1999 int diff, best_diff=1000000000; | |
2000 for(i=1; i<=31; i++){ | |
2001 diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps; | |
2002 if(diff<0) diff= -diff; | |
2003 if(diff<best_diff){ | |
2004 best_diff= diff; | |
2005 rate_q= i; | |
2006 } | |
2007 } | |
2008 } | |
2009 | |
2010 s->short_term_qsum*=s->qblur; | |
2011 s->short_term_qcount*=s->qblur; | |
2012 | |
2013 s->short_term_qsum+= rate_q; | |
2014 s->short_term_qcount++; | |
2015 short_term_q= s->short_term_qsum/s->short_term_qcount; | |
2016 | |
302 | 2017 long_term_q= s->qsum/s->qcount*(s->total_bits+1)/(wanted_bits+1); //+1 to avoid nan & 0 |
268 | 2018 |
2019 // q= (long_term_q - short_term_q)*s->qcompress + short_term_q; | |
2020 q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q); | |
2021 | |
2022 diff= s->total_bits - wanted_bits; | |
2023 br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance; | |
270 | 2024 if(br_compensation<=0.0) br_compensation=0.001; |
268 | 2025 q/=br_compensation; |
2026 | |
2027 qscale= (int)(q + 0.5); | |
2028 if (qscale<qmin) qscale=qmin; | |
2029 else if(qscale>qmax) qscale=qmax; | |
2030 | |
2031 if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff; | |
2032 else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff; | |
2033 | |
2034 s->qsum+= qscale; | |
2035 s->qcount++; | |
2036 | |
2037 s->last_pict_type= s->pict_type; | |
302 | 2038 //printf("q:%d diff:%d comp:%f rate_q:%d st_q:%f fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation, |
2039 // rate_q, short_term_q, s->mc_mb_var, s->frame_bits); | |
268 | 2040 //printf("%d %d\n", s->bit_rate, (int)fps); |
2041 return qscale; | |
2042 #else | |
187
3f3b14d3a23d
qscale estimate fix, diff extended to 64bit wide. patch by Stephen Davies <steve@daviesfam.org>
arpi_esp
parents:
178
diff
changeset
|
2043 INT64 diff, total_bits = s->total_bits; |
0 | 2044 float q; |
268 | 2045 int qscale; |
0 | 2046 if (s->pict_type == I_TYPE) { |
2047 s->wanted_bits += s->I_frame_bits; | |
2048 } else { | |
2049 s->wanted_bits += s->P_frame_bits; | |
2050 } | |
2051 diff = s->wanted_bits - total_bits; | |
2052 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width); | |
2053 /* adjust for I frame */ | |
2054 if (s->pict_type == I_TYPE && !s->intra_only) { | |
2055 q /= I_FRAME_SIZE_RATIO; | |
2056 } | |
2057 | |
2058 /* using a too small Q scale leeds to problems in mpeg1 and h263 | |
2059 because AC coefficients are clamped to 255 or 127 */ | |
2060 qmin = 3; | |
2061 if (q < qmin) | |
2062 q = qmin; | |
2063 else if (q > 31) | |
2064 q = 31; | |
2065 qscale = (int)(q + 0.5); | |
2066 #if defined(DEBUG) | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2067 printf("\n%d: total=%0.0f wanted=%0.0f br=%0.1f diff=%d qest=%2.1f\n", |
0 | 2068 s->picture_number, |
64 | 2069 (double)total_bits, |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2070 (double)s->wanted_bits, |
0 | 2071 (float)s->frame_rate / FRAME_RATE_BASE * |
2072 total_bits / s->picture_number, | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2073 (int)diff, q); |
0 | 2074 #endif |
2075 return qscale; | |
268 | 2076 #endif |
0 | 2077 } |
2078 | |
2079 AVCodec mpeg1video_encoder = { | |
2080 "mpeg1video", | |
2081 CODEC_TYPE_VIDEO, | |
2082 CODEC_ID_MPEG1VIDEO, | |
2083 sizeof(MpegEncContext), | |
2084 MPV_encode_init, | |
2085 MPV_encode_picture, | |
2086 MPV_encode_end, | |
2087 }; | |
2088 | |
2089 AVCodec h263_encoder = { | |
2090 "h263", | |
2091 CODEC_TYPE_VIDEO, | |
2092 CODEC_ID_H263, | |
2093 sizeof(MpegEncContext), | |
2094 MPV_encode_init, | |
2095 MPV_encode_picture, | |
2096 MPV_encode_end, | |
2097 }; | |
2098 | |
2099 AVCodec h263p_encoder = { | |
2100 "h263p", | |
2101 CODEC_TYPE_VIDEO, | |
2102 CODEC_ID_H263P, | |
2103 sizeof(MpegEncContext), | |
2104 MPV_encode_init, | |
2105 MPV_encode_picture, | |
2106 MPV_encode_end, | |
2107 }; | |
2108 | |
2109 AVCodec rv10_encoder = { | |
2110 "rv10", | |
2111 CODEC_TYPE_VIDEO, | |
2112 CODEC_ID_RV10, | |
2113 sizeof(MpegEncContext), | |
2114 MPV_encode_init, | |
2115 MPV_encode_picture, | |
2116 MPV_encode_end, | |
2117 }; | |
2118 | |
2119 AVCodec mjpeg_encoder = { | |
2120 "mjpeg", | |
2121 CODEC_TYPE_VIDEO, | |
2122 CODEC_ID_MJPEG, | |
2123 sizeof(MpegEncContext), | |
2124 MPV_encode_init, | |
2125 MPV_encode_picture, | |
2126 MPV_encode_end, | |
2127 }; | |
2128 | |
71 | 2129 AVCodec mpeg4_encoder = { |
2130 "mpeg4", | |
0 | 2131 CODEC_TYPE_VIDEO, |
71 | 2132 CODEC_ID_MPEG4, |
0 | 2133 sizeof(MpegEncContext), |
2134 MPV_encode_init, | |
2135 MPV_encode_picture, | |
2136 MPV_encode_end, | |
2137 }; | |
2138 | |
307 | 2139 AVCodec msmpeg4v1_encoder = { |
2140 "msmpeg4v1", | |
0 | 2141 CODEC_TYPE_VIDEO, |
307 | 2142 CODEC_ID_MSMPEG4V1, |
0 | 2143 sizeof(MpegEncContext), |
2144 MPV_encode_init, | |
2145 MPV_encode_picture, | |
2146 MPV_encode_end, | |
2147 }; | |
307 | 2148 |
2149 AVCodec msmpeg4v2_encoder = { | |
2150 "msmpeg4v2", | |
2151 CODEC_TYPE_VIDEO, | |
2152 CODEC_ID_MSMPEG4V2, | |
2153 sizeof(MpegEncContext), | |
2154 MPV_encode_init, | |
2155 MPV_encode_picture, | |
2156 MPV_encode_end, | |
2157 }; | |
2158 | |
2159 AVCodec msmpeg4v3_encoder = { | |
2160 "msmpeg4", | |
2161 CODEC_TYPE_VIDEO, | |
2162 CODEC_ID_MSMPEG4V3, | |
2163 sizeof(MpegEncContext), | |
2164 MPV_encode_init, | |
2165 MPV_encode_picture, | |
2166 MPV_encode_end, | |
2167 }; |