Mercurial > libavcodec.hg
annotate mpegvideo.c @ 279:ae5c33165d5c libavcodec
mpeg1 & epzs fix
author | michaelni |
---|---|
date | Fri, 22 Mar 2002 04:35:46 +0000 |
parents | 5cb2978e701f |
children | 3dc1ca4ba717 |
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. | |
18 */ | |
19 #include <stdlib.h> | |
20 #include <stdio.h> | |
21 #include <math.h> | |
22 #include <string.h> | |
23 #include "avcodec.h" | |
24 #include "dsputil.h" | |
25 #include "mpegvideo.h" | |
26 | |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
27 #ifdef USE_FASTMEMCPY |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
28 #include "fastmemcpy.h" |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
29 #endif |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
30 |
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
|
31 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
|
32 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
|
33 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
|
34 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
|
35 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
|
36 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
|
37 DCTELEM *block, int n, int qscale); |
206 | 38 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
220 | 39 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
206 | 40 |
220 | 41 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c; |
206 | 42 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
43 | |
0 | 44 #define EDGE_WIDTH 16 |
45 | |
46 /* enable all paranoid tests for rounding, overflows, etc... */ | |
47 //#define PARANOID | |
48 | |
49 //#define DEBUG | |
50 | |
51 /* for jpeg fast DCT */ | |
52 #define CONST_BITS 14 | |
53 | |
54 static const unsigned short aanscales[64] = { | |
55 /* precomputed values scaled up by 14 bits */ | |
56 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
57 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
58 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
59 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
60 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
61 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
62 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
63 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
64 }; | |
65 | |
66 static UINT8 h263_chroma_roundtab[16] = { | |
67 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
68 }; | |
69 | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
70 static UINT16 default_mv_penalty[MAX_FCODE][MAX_MV*2+1]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
71 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
|
72 |
0 | 73 /* default motion estimation */ |
74 int motion_estimation_method = ME_LOG; | |
75 | |
200 | 76 extern UINT8 zigzag_end[64]; |
77 | |
220 | 78 static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale) |
0 | 79 { |
80 int i; | |
81 | |
82 if (av_fdct == jpeg_fdct_ifast) { | |
83 for(i=0;i<64;i++) { | |
84 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
220 | 85 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
86 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
87 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
0 | 88 |
220 | 89 qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
90 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)])); | |
0 | 91 } |
92 } else { | |
93 for(i=0;i<64;i++) { | |
94 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
220 | 95 So 16 <= qscale * quant_matrix[i] <= 7905 |
96 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
97 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
0 | 98 */ |
220 | 99 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
100 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
0 | 101 } |
102 } | |
103 } | |
104 | |
105 /* init common structure for both encoder and decoder */ | |
106 int MPV_common_init(MpegEncContext *s) | |
107 { | |
108 int c_size, i; | |
109 UINT8 *pict; | |
110 | |
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
|
111 if (s->out_format == FMT_H263) |
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
|
112 s->dct_unquantize = dct_unquantize_h263_c; |
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
|
113 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
|
114 s->dct_unquantize = dct_unquantize_mpeg1_c; |
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 |
0 | 119 s->mb_width = (s->width + 15) / 16; |
120 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
|
121 s->mb_num = s->mb_width * s->mb_height; |
0 | 122 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
123 | |
124 for(i=0;i<3;i++) { | |
125 int w, h, shift, pict_start; | |
126 | |
127 w = s->linesize; | |
128 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
129 shift = (i == 0) ? 0 : 1; | |
130 c_size = (w >> shift) * (h >> shift); | |
131 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
132 | |
133 pict = av_mallocz(c_size); | |
134 if (pict == NULL) | |
135 goto fail; | |
136 s->last_picture_base[i] = pict; | |
137 s->last_picture[i] = pict + pict_start; | |
138 | |
139 pict = av_mallocz(c_size); | |
140 if (pict == NULL) | |
141 goto fail; | |
142 s->next_picture_base[i] = pict; | |
143 s->next_picture[i] = pict + pict_start; | |
144 | |
145 if (s->has_b_frames) { | |
146 pict = av_mallocz(c_size); | |
147 if (pict == NULL) | |
148 goto fail; | |
149 s->aux_picture_base[i] = pict; | |
150 s->aux_picture[i] = pict + pict_start; | |
151 } | |
152 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
153 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
154 if (s->encoding) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
155 /* Allocate MB type table */ |
239 | 156 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
|
157 if (s->mb_type == NULL) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
158 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
159 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
160 } |
239 | 161 |
162 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16)); | |
163 if (s->mb_var == NULL) { | |
164 perror("malloc"); | |
165 goto fail; | |
166 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
167 /* Allocate MV table */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
168 /* By now we just have one MV per MB */ |
239 | 169 s->mv_table[0] = av_mallocz(s->mb_num * sizeof(INT16)); |
170 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
|
171 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
|
172 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
173 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
174 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
175 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
176 |
279 | 177 if (s->out_format == FMT_H263 || s->encoding) { |
0 | 178 int size; |
179 /* MV prediction */ | |
180 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
181 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
182 if (s->motion_val == NULL) | |
183 goto fail; | |
184 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
185 } | |
186 | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
187 if (s->h263_pred || s->h263_plus) { |
0 | 188 int y_size, c_size, i, size; |
189 | |
190 /* dc values */ | |
191 | |
192 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
193 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
194 size = y_size + 2 * c_size; | |
195 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
196 if (s->dc_val[0] == NULL) | |
197 goto fail; | |
198 s->dc_val[1] = s->dc_val[0] + y_size; | |
199 s->dc_val[2] = s->dc_val[1] + c_size; | |
200 for(i=0;i<size;i++) | |
201 s->dc_val[0][i] = 1024; | |
202 | |
203 /* ac values */ | |
204 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
205 if (s->ac_val[0] == NULL) | |
206 goto fail; | |
207 s->ac_val[1] = s->ac_val[0] + y_size; | |
208 s->ac_val[2] = s->ac_val[1] + c_size; | |
209 | |
210 /* cbp values */ | |
211 s->coded_block = av_mallocz(y_size); | |
212 if (!s->coded_block) | |
213 goto fail; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
214 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
215 /* 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
|
216 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
|
217 if (!s->mbintra_table) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
218 goto fail; |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
219 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
|
220 } |
0 | 221 /* default structure is frame */ |
222 s->picture_structure = PICT_FRAME; | |
223 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
224 /* init macroblock skip table */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
225 if (!s->encoding) { |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
226 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
|
227 if (!s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
228 goto fail; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
229 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
230 |
0 | 231 s->context_initialized = 1; |
232 return 0; | |
233 fail: | |
244 | 234 MPV_common_end(s); |
235 return -1; | |
236 } | |
237 | |
238 /* init common structure for both encoder and decoder */ | |
239 void MPV_common_end(MpegEncContext *s) | |
240 { | |
241 int i; | |
242 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
243 if (s->mb_type) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
244 free(s->mb_type); |
239 | 245 if (s->mb_var) |
246 free(s->mb_var); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
247 if (s->mv_table[0]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
248 free(s->mv_table[0]); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
249 if (s->mv_table[1]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
250 free(s->mv_table[1]); |
0 | 251 if (s->motion_val) |
252 free(s->motion_val); | |
253 if (s->dc_val[0]) | |
254 free(s->dc_val[0]); | |
255 if (s->ac_val[0]) | |
256 free(s->ac_val[0]); | |
257 if (s->coded_block) | |
258 free(s->coded_block); | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
259 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
|
260 free(s->mbintra_table); |
244 | 261 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
262 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
263 free(s->mbskip_table); |
0 | 264 for(i=0;i<3;i++) { |
265 if (s->last_picture_base[i]) | |
244 | 266 free(s->last_picture_base[i]); |
267 if (s->next_picture_base[i]) | |
268 free(s->next_picture_base[i]); | |
0 | 269 if (s->has_b_frames) |
270 free(s->aux_picture_base[i]); | |
271 } | |
272 s->context_initialized = 0; | |
273 } | |
274 | |
275 /* init video encoder */ | |
276 int MPV_encode_init(AVCodecContext *avctx) | |
277 { | |
278 MpegEncContext *s = avctx->priv_data; | |
60 | 279 int i; |
0 | 280 |
281 s->bit_rate = avctx->bit_rate; | |
268 | 282 s->bit_rate_tolerance = avctx->bit_rate_tolerance; |
0 | 283 s->frame_rate = avctx->frame_rate; |
284 s->width = avctx->width; | |
285 s->height = avctx->height; | |
286 s->gop_size = avctx->gop_size; | |
162 | 287 s->rtp_mode = avctx->rtp_mode; |
288 s->rtp_payload_size = avctx->rtp_payload_size; | |
231 | 289 if (avctx->rtp_callback) |
290 s->rtp_callback = avctx->rtp_callback; | |
268 | 291 s->qmin= avctx->qmin; |
292 s->qmax= avctx->qmax; | |
293 s->max_qdiff= avctx->max_qdiff; | |
294 s->qcompress= avctx->qcompress; | |
295 s->qblur= avctx->qblur; | |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
296 s->avctx = avctx; |
162 | 297 |
0 | 298 if (s->gop_size <= 1) { |
299 s->intra_only = 1; | |
300 s->gop_size = 12; | |
301 } else { | |
302 s->intra_only = 0; | |
303 } | |
304 s->full_search = motion_estimation_method; | |
305 | |
306 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
|
307 |
0 | 308 switch(avctx->codec->id) { |
309 case CODEC_ID_MPEG1VIDEO: | |
310 s->out_format = FMT_MPEG1; | |
311 break; | |
312 case CODEC_ID_MJPEG: | |
313 s->out_format = FMT_MJPEG; | |
314 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
|
315 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
|
316 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
|
317 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
|
318 s->mjpeg_vsample[2] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
319 s->mjpeg_hsample[0] = 2; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
320 s->mjpeg_hsample[1] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
321 s->mjpeg_hsample[2] = 1; |
0 | 322 if (mjpeg_init(s) < 0) |
323 return -1; | |
324 break; | |
325 case CODEC_ID_H263: | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
326 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
|
327 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
0 | 328 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
329 } |
0 | 330 s->out_format = FMT_H263; |
331 break; | |
332 case CODEC_ID_H263P: | |
333 s->out_format = FMT_H263; | |
162 | 334 s->rtp_mode = 1; |
335 s->rtp_payload_size = 1200; | |
0 | 336 s->h263_plus = 1; |
78 | 337 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
338 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
339 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
340 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
341 s->umvplus_dec = 0; |
0 | 342 break; |
343 case CODEC_ID_RV10: | |
344 s->out_format = FMT_H263; | |
345 s->h263_rv10 = 1; | |
346 break; | |
71 | 347 case CODEC_ID_MPEG4: |
0 | 348 s->out_format = FMT_H263; |
349 s->h263_pred = 1; | |
350 s->unrestricted_mv = 1; | |
351 break; | |
352 case CODEC_ID_MSMPEG4: | |
353 s->out_format = FMT_H263; | |
354 s->h263_msmpeg4 = 1; | |
355 s->h263_pred = 1; | |
356 s->unrestricted_mv = 1; | |
357 break; | |
358 default: | |
359 return -1; | |
360 } | |
361 | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
362 { /* 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
|
363 static int done=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
364 if(!done){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
365 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
366 done=1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
367 memset(default_mv_penalty, 0, sizeof(UINT16)*MAX_FCODE*(2*MAX_MV+1)); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
368 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
|
369 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
370 for(i=-16; i<16; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
371 default_fcode_tab[i + MAX_MV]= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
372 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
373 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
374 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
375 s->mv_penalty= default_mv_penalty; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
376 s->fcode_tab= default_fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
377 |
0 | 378 if (s->out_format == FMT_H263) |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
379 h263_encode_init(s); |
0 | 380 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
381 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
382 |
0 | 383 /* init */ |
384 if (MPV_common_init(s) < 0) | |
385 return -1; | |
386 | |
60 | 387 /* init default q matrix */ |
388 for(i=0;i<64;i++) { | |
389 s->intra_matrix[i] = default_intra_matrix[i]; | |
390 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
391 } | |
392 | |
0 | 393 /* rate control init */ |
394 rate_control_init(s); | |
395 | |
396 s->picture_number = 0; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
397 s->picture_in_gop_number = 0; |
0 | 398 s->fake_picture_number = 0; |
399 /* motion detector init */ | |
400 s->f_code = 1; | |
401 | |
402 return 0; | |
403 } | |
404 | |
405 int MPV_encode_end(AVCodecContext *avctx) | |
406 { | |
407 MpegEncContext *s = avctx->priv_data; | |
408 | |
409 #ifdef STATS | |
410 print_stats(); | |
411 #endif | |
412 MPV_common_end(s); | |
413 if (s->out_format == FMT_MJPEG) | |
414 mjpeg_close(s); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
415 |
0 | 416 return 0; |
417 } | |
418 | |
419 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 420 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 421 { |
422 UINT8 *ptr, *last_line; | |
423 int i; | |
424 | |
425 last_line = buf + (height - 1) * wrap; | |
426 for(i=0;i<w;i++) { | |
427 /* top and bottom */ | |
428 memcpy(buf - (i + 1) * wrap, buf, width); | |
429 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
430 } | |
431 /* left and right */ | |
432 ptr = buf; | |
433 for(i=0;i<height;i++) { | |
434 memset(ptr - w, ptr[0], w); | |
435 memset(ptr + width, ptr[width-1], w); | |
436 ptr += wrap; | |
437 } | |
438 /* corners */ | |
439 for(i=0;i<w;i++) { | |
440 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
441 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
442 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
443 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
444 } | |
445 } | |
446 | |
447 /* generic function for encode/decode called before a frame is coded/decoded */ | |
448 void MPV_frame_start(MpegEncContext *s) | |
449 { | |
450 int i; | |
451 UINT8 *tmp; | |
452 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
453 s->mb_skiped = 0; |
0 | 454 if (s->pict_type == B_TYPE) { |
455 for(i=0;i<3;i++) { | |
456 s->current_picture[i] = s->aux_picture[i]; | |
457 } | |
458 } else { | |
262 | 459 s->last_non_b_pict_type= s->pict_type; |
0 | 460 for(i=0;i<3;i++) { |
461 /* swap next and last */ | |
462 tmp = s->last_picture[i]; | |
463 s->last_picture[i] = s->next_picture[i]; | |
464 s->next_picture[i] = tmp; | |
465 s->current_picture[i] = tmp; | |
466 } | |
467 } | |
468 } | |
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
|
469 |
0 | 470 /* generic function for encode/decode called after a frame has been coded/decoded */ |
471 void MPV_frame_end(MpegEncContext *s) | |
472 { | |
473 /* draw edge for correct motion prediction if outside */ | |
220 | 474 if (s->pict_type != B_TYPE && !s->intra_only) { |
257 | 475 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
|
476 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
|
477 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
|
478 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
|
479 }else{ |
257 | 480 /* mpeg4? / opendivx / xvid */ |
0 | 481 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
482 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
483 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
|
484 } |
0 | 485 } |
207 | 486 emms_c(); |
0 | 487 } |
488 | |
489 int MPV_encode_picture(AVCodecContext *avctx, | |
490 unsigned char *buf, int buf_size, void *data) | |
491 { | |
492 MpegEncContext *s = avctx->priv_data; | |
493 AVPicture *pict = data; | |
494 int i, j; | |
495 | |
496 if (s->fixed_qscale) | |
497 s->qscale = avctx->quality; | |
498 | |
499 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
500 | |
501 if (!s->intra_only) { | |
502 /* first picture of GOP is intra */ | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
503 if (s->picture_in_gop_number >= s->gop_size){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
504 s->picture_in_gop_number=0; |
0 | 505 s->pict_type = I_TYPE; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
506 }else |
0 | 507 s->pict_type = P_TYPE; |
508 } else { | |
509 s->pict_type = I_TYPE; | |
510 } | |
511 | |
512 MPV_frame_start(s); | |
220 | 513 |
0 | 514 for(i=0;i<3;i++) { |
515 UINT8 *src = pict->data[i]; | |
516 UINT8 *dest = s->current_picture[i]; | |
517 int src_wrap = pict->linesize[i]; | |
518 int dest_wrap = s->linesize; | |
519 int w = s->width; | |
520 int h = s->height; | |
521 | |
522 if (i >= 1) { | |
523 dest_wrap >>= 1; | |
524 w >>= 1; | |
525 h >>= 1; | |
526 } | |
527 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
528 if(dest_wrap==src_wrap){ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
529 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
|
530 } else { |
220 | 531 for(j=0;j<h;j++) { |
532 memcpy(dest, src, w); | |
533 dest += dest_wrap; | |
534 src += src_wrap; | |
535 } | |
227
ec1bc02a0a47
avoid copying input when encoding non intra stuff too
michaelni
parents:
220
diff
changeset
|
536 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
|
537 } |
0 | 538 } |
539 | |
540 encode_picture(s, s->picture_number); | |
271 | 541 avctx->key_frame = (s->pict_type == I_TYPE); |
0 | 542 |
543 MPV_frame_end(s); | |
544 s->picture_number++; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
545 s->picture_in_gop_number++; |
0 | 546 |
547 if (s->out_format == FMT_MJPEG) | |
548 mjpeg_picture_trailer(s); | |
549 | |
550 flush_put_bits(&s->pb); | |
268 | 551 s->last_frame_bits= s->frame_bits; |
552 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8; | |
553 s->total_bits += s->frame_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
|
554 |
0 | 555 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
|
556 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
|
557 /* 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
|
558 /* 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
|
559 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
|
560 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
|
561 } |
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
|
562 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 563 } |
564 | |
565 static inline int clip(int a, int amin, int amax) | |
566 { | |
567 if (a < amin) | |
568 return amin; | |
569 else if (a > amax) | |
570 return amax; | |
571 else | |
572 return a; | |
573 } | |
574 | |
255 | 575 static inline void gmc1_motion(MpegEncContext *s, |
576 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
577 int dest_offset, | |
578 UINT8 **ref_picture, int src_offset, | |
579 int h) | |
580 { | |
581 UINT8 *ptr; | |
582 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
583 int motion_x, motion_y; | |
584 | |
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
585 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n"); |
255 | 586 motion_x= s->sprite_offset[0][0]; |
587 motion_y= s->sprite_offset[0][1]; | |
588 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
589 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
590 motion_x<<=(3-s->sprite_warping_accuracy); | |
591 motion_y<<=(3-s->sprite_warping_accuracy); | |
592 src_x = clip(src_x, -16, s->width); | |
593 if (src_x == s->width) | |
594 motion_x =0; | |
595 src_y = clip(src_y, -16, s->height); | |
596 if (src_y == s->height) | |
597 motion_y =0; | |
598 | |
599 linesize = s->linesize; | |
600 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
601 | |
602 dest_y+=dest_offset; | |
603 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
604 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
605 | |
606 motion_x= s->sprite_offset[1][0]; | |
607 motion_y= s->sprite_offset[1][1]; | |
608 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
609 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
610 motion_x<<=(3-s->sprite_warping_accuracy); | |
611 motion_y<<=(3-s->sprite_warping_accuracy); | |
612 src_x = clip(src_x, -8, s->width>>1); | |
613 if (src_x == s->width>>1) | |
614 motion_x =0; | |
615 src_y = clip(src_y, -8, s->height>>1); | |
616 if (src_y == s->height>>1) | |
617 motion_y =0; | |
618 | |
619 offset = (src_y * linesize>>1) + src_x + (src_offset>>1); | |
620 ptr = ref_picture[1] + offset; | |
621 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
622 ptr = ref_picture[2] + offset; | |
623 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
624 | |
625 return; | |
626 } | |
627 | |
0 | 628 /* apply one mpeg motion vector to the three components */ |
629 static inline void mpeg_motion(MpegEncContext *s, | |
630 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
631 int dest_offset, | |
632 UINT8 **ref_picture, int src_offset, | |
633 int field_based, op_pixels_func *pix_op, | |
634 int motion_x, int motion_y, int h) | |
635 { | |
636 UINT8 *ptr; | |
637 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
255 | 638 if(s->quarter_sample) |
639 { | |
640 motion_x>>=1; | |
641 motion_y>>=1; | |
642 } | |
0 | 643 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
644 src_x = s->mb_x * 16 + (motion_x >> 1); | |
645 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
646 | |
647 /* WARNING: do no forget half pels */ | |
648 height = s->height >> field_based; | |
649 src_x = clip(src_x, -16, s->width); | |
650 if (src_x == s->width) | |
651 dxy &= ~1; | |
652 src_y = clip(src_y, -16, height); | |
653 if (src_y == height) | |
654 dxy &= ~2; | |
655 linesize = s->linesize << field_based; | |
656 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
657 dest_y += dest_offset; | |
658 pix_op[dxy](dest_y, ptr, linesize, h); | |
659 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
660 | |
661 if (s->out_format == FMT_H263) { | |
662 dxy = 0; | |
663 if ((motion_x & 3) != 0) | |
664 dxy |= 1; | |
665 if ((motion_y & 3) != 0) | |
666 dxy |= 2; | |
667 mx = motion_x >> 2; | |
668 my = motion_y >> 2; | |
669 } else { | |
670 mx = motion_x / 2; | |
671 my = motion_y / 2; | |
672 dxy = ((my & 1) << 1) | (mx & 1); | |
673 mx >>= 1; | |
674 my >>= 1; | |
675 } | |
676 | |
677 src_x = s->mb_x * 8 + mx; | |
678 src_y = s->mb_y * (8 >> field_based) + my; | |
679 src_x = clip(src_x, -8, s->width >> 1); | |
680 if (src_x == (s->width >> 1)) | |
681 dxy &= ~1; | |
682 src_y = clip(src_y, -8, height >> 1); | |
683 if (src_y == (height >> 1)) | |
684 dxy &= ~2; | |
685 | |
686 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
687 ptr = ref_picture[1] + offset; | |
688 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
689 ptr = ref_picture[2] + offset; | |
690 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
691 } | |
692 | |
255 | 693 static inline void qpel_motion(MpegEncContext *s, |
694 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
695 int dest_offset, | |
696 UINT8 **ref_picture, int src_offset, | |
697 int field_based, op_pixels_func *pix_op, | |
698 qpel_mc_func *qpix_op, | |
699 int motion_x, int motion_y, int h) | |
700 { | |
701 UINT8 *ptr; | |
702 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
703 | |
704 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
705 src_x = s->mb_x * 16 + (motion_x >> 2); | |
706 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
707 | |
708 height = s->height >> field_based; | |
709 src_x = clip(src_x, -16, s->width); | |
710 if (src_x == s->width) | |
711 dxy &= ~3; | |
712 src_y = clip(src_y, -16, height); | |
713 if (src_y == height) | |
714 dxy &= ~12; | |
715 linesize = s->linesize << field_based; | |
716 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
717 dest_y += dest_offset; | |
718 //printf("%d %d %d\n", src_x, src_y, dxy); | |
719 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3); | |
720 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3); | |
721 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3); | |
722 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3); | |
723 | |
724 mx= (motion_x>>1) | (motion_x&1); | |
725 my= (motion_y>>1) | (motion_y&1); | |
726 | |
727 dxy = 0; | |
728 if ((mx & 3) != 0) | |
729 dxy |= 1; | |
730 if ((my & 3) != 0) | |
731 dxy |= 2; | |
732 mx = mx >> 2; | |
733 my = my >> 2; | |
734 | |
735 src_x = s->mb_x * 8 + mx; | |
736 src_y = s->mb_y * (8 >> field_based) + my; | |
737 src_x = clip(src_x, -8, s->width >> 1); | |
738 if (src_x == (s->width >> 1)) | |
739 dxy &= ~1; | |
740 src_y = clip(src_y, -8, height >> 1); | |
741 if (src_y == (height >> 1)) | |
742 dxy &= ~2; | |
743 | |
744 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
745 ptr = ref_picture[1] + offset; | |
746 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
747 ptr = ref_picture[2] + offset; | |
748 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
749 } | |
750 | |
751 | |
0 | 752 static inline void MPV_motion(MpegEncContext *s, |
753 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
754 int dir, UINT8 **ref_picture, | |
255 | 755 op_pixels_func *pix_op, qpel_mc_func *qpix_op) |
0 | 756 { |
757 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
758 int mb_x, mb_y, i; | |
759 UINT8 *ptr, *dest; | |
760 | |
761 mb_x = s->mb_x; | |
762 mb_y = s->mb_y; | |
763 | |
764 switch(s->mv_type) { | |
765 case MV_TYPE_16X16: | |
255 | 766 if(s->mcsel){ |
767 #if 0 | |
768 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
769 ref_picture, 0, | |
770 0, pix_op, | |
771 s->sprite_offset[0][0]>>3, | |
772 s->sprite_offset[0][1]>>3, | |
773 16); | |
774 #else | |
775 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
776 ref_picture, 0, | |
777 16); | |
778 #endif | |
262 | 779 }else if(s->quarter_sample && dir==0){ //FIXME |
255 | 780 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, |
781 ref_picture, 0, | |
782 0, pix_op, qpix_op, | |
783 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
784 }else{ | |
785 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
786 ref_picture, 0, | |
787 0, pix_op, | |
788 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
789 } | |
0 | 790 break; |
791 case MV_TYPE_8X8: | |
792 for(i=0;i<4;i++) { | |
793 motion_x = s->mv[dir][i][0]; | |
794 motion_y = s->mv[dir][i][1]; | |
795 | |
796 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
797 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
798 src_y = mb_y * 16 + (motion_y >> 1) + ((i >> 1) & 1) * 8; | |
799 | |
800 /* WARNING: do no forget half pels */ | |
801 src_x = clip(src_x, -16, s->width); | |
802 if (src_x == s->width) | |
803 dxy &= ~1; | |
804 src_y = clip(src_y, -16, s->height); | |
805 if (src_y == s->height) | |
806 dxy &= ~2; | |
807 | |
808 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
809 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
810 pix_op[dxy](dest, ptr, s->linesize, 8); | |
811 } | |
812 /* In case of 8X8, we construct a single chroma motion vector | |
813 with a special rounding */ | |
814 mx = 0; | |
815 my = 0; | |
816 for(i=0;i<4;i++) { | |
817 mx += s->mv[dir][i][0]; | |
818 my += s->mv[dir][i][1]; | |
819 } | |
820 if (mx >= 0) | |
821 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
822 else { | |
823 mx = -mx; | |
824 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
825 } | |
826 if (my >= 0) | |
827 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
828 else { | |
829 my = -my; | |
830 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
831 } | |
832 dxy = ((my & 1) << 1) | (mx & 1); | |
833 mx >>= 1; | |
834 my >>= 1; | |
835 | |
836 src_x = mb_x * 8 + mx; | |
837 src_y = mb_y * 8 + my; | |
838 src_x = clip(src_x, -8, s->width/2); | |
839 if (src_x == s->width/2) | |
840 dxy &= ~1; | |
841 src_y = clip(src_y, -8, s->height/2); | |
842 if (src_y == s->height/2) | |
843 dxy &= ~2; | |
844 | |
845 offset = (src_y * (s->linesize >> 1)) + src_x; | |
846 ptr = ref_picture[1] + offset; | |
847 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
848 ptr = ref_picture[2] + offset; | |
849 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
850 break; | |
851 case MV_TYPE_FIELD: | |
852 if (s->picture_structure == PICT_FRAME) { | |
853 /* top field */ | |
854 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
855 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
856 1, pix_op, | |
857 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
858 /* bottom field */ | |
859 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
860 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
861 1, pix_op, | |
862 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
863 } else { | |
864 | |
865 | |
866 } | |
867 break; | |
868 } | |
869 } | |
870 | |
871 | |
872 /* put block[] to dest[] */ | |
873 static inline void put_dct(MpegEncContext *s, | |
874 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
875 { | |
876 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
|
877 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
|
878 ff_idct (block); |
0 | 879 put_pixels_clamped(block, dest, line_size); |
880 } | |
881 | |
882 /* add block[] to dest[] */ | |
883 static inline void add_dct(MpegEncContext *s, | |
884 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
885 { | |
886 if (s->block_last_index[i] >= 0) { | |
887 if (!s->mpeg2) | |
206 | 888 if(s->encoding || (!s->h263_msmpeg4)) |
200 | 889 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
|
890 ff_idct (block); |
0 | 891 add_pixels_clamped(block, dest, line_size); |
892 } | |
893 } | |
894 | |
895 /* generic function called after a macroblock has been parsed by the | |
896 decoder or after it has been encoded by the encoder. | |
897 | |
898 Important variables used: | |
899 s->mb_intra : true if intra macroblock | |
900 s->mv_dir : motion vector direction | |
901 s->mv_type : motion vector type | |
902 s->mv : motion vector | |
903 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
904 */ | |
905 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
906 { | |
244 | 907 int mb_x, mb_y; |
0 | 908 int dct_linesize, dct_offset; |
909 op_pixels_func *op_pix; | |
255 | 910 qpel_mc_func *op_qpix; |
0 | 911 |
912 mb_x = s->mb_x; | |
913 mb_y = s->mb_y; | |
914 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
915 #ifdef FF_POSTPROCESS |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
916 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
917 //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
|
918 #endif |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
919 |
0 | 920 /* update DC predictors for P macroblocks */ |
921 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
922 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
|
923 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
|
924 { |
244 | 925 int wrap, xy, v; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
926 s->mbintra_table[mb_x + mb_y*s->mb_width]=0; |
0 | 927 wrap = 2 * s->mb_width + 2; |
244 | 928 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 929 v = 1024; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
930 |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
931 s->dc_val[0][xy] = v; |
244 | 932 s->dc_val[0][xy + 1] = v; |
933 s->dc_val[0][xy + wrap] = v; | |
934 s->dc_val[0][xy + 1 + wrap] = v; | |
0 | 935 /* ac pred */ |
244 | 936 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16)); |
937 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16)); | |
938 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16)); | |
939 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16)); | |
0 | 940 if (s->h263_msmpeg4) { |
244 | 941 s->coded_block[xy] = 0; |
942 s->coded_block[xy + 1] = 0; | |
943 s->coded_block[xy + wrap] = 0; | |
944 s->coded_block[xy + 1 + wrap] = 0; | |
0 | 945 } |
946 /* chroma */ | |
947 wrap = s->mb_width + 2; | |
244 | 948 xy = mb_x + 1 + (mb_y + 1) * wrap; |
949 s->dc_val[1][xy] = v; | |
950 s->dc_val[2][xy] = v; | |
0 | 951 /* ac pred */ |
244 | 952 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); |
953 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
|
954 } |
0 | 955 } else { |
956 s->last_dc[0] = 128 << s->intra_dc_precision; | |
957 s->last_dc[1] = 128 << s->intra_dc_precision; | |
958 s->last_dc[2] = 128 << s->intra_dc_precision; | |
959 } | |
960 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
961 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
|
962 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
|
963 |
279 | 964 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame |
965 motion_val is needed for encodig too, as the ME needs it */ | |
966 if (s->out_format == FMT_H263 || s->encoding) { | |
262 | 967 if(s->pict_type!=B_TYPE){ |
244 | 968 int xy, wrap, motion_x, motion_y; |
0 | 969 |
970 wrap = 2 * s->mb_width + 2; | |
244 | 971 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 972 if (s->mb_intra) { |
973 motion_x = 0; | |
974 motion_y = 0; | |
975 goto motion_init; | |
976 } else if (s->mv_type == MV_TYPE_16X16) { | |
977 motion_x = s->mv[0][0][0]; | |
978 motion_y = s->mv[0][0][1]; | |
979 motion_init: | |
980 /* no update if 8X8 because it has been done during parsing */ | |
244 | 981 s->motion_val[xy][0] = motion_x; |
982 s->motion_val[xy][1] = motion_y; | |
983 s->motion_val[xy + 1][0] = motion_x; | |
984 s->motion_val[xy + 1][1] = motion_y; | |
985 s->motion_val[xy + wrap][0] = motion_x; | |
986 s->motion_val[xy + wrap][1] = motion_y; | |
987 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
988 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
0 | 989 } |
262 | 990 } |
0 | 991 } |
992 | |
993 if (!s->intra_only) { | |
994 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
|
995 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
996 |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
997 /* 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
|
998 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
|
999 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
|
1000 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1001 s->mb_skiped = 0; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1002 /* 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
|
1003 if (*mbskip_ptr != 0) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1004 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1005 *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
|
1006 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1007 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1008 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1009 } |
0 | 1010 |
1011 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
1012 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1013 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1014 | |
1015 if (s->interlaced_dct) { | |
1016 dct_linesize = s->linesize * 2; | |
1017 dct_offset = s->linesize; | |
1018 } else { | |
1019 dct_linesize = s->linesize; | |
1020 dct_offset = s->linesize * 8; | |
1021 } | |
1022 | |
1023 if (!s->mb_intra) { | |
1024 /* motion handling */ | |
255 | 1025 if (!s->no_rounding){ |
0 | 1026 op_pix = put_pixels_tab; |
255 | 1027 op_qpix= qpel_mc_rnd_tab; |
1028 }else{ | |
0 | 1029 op_pix = put_no_rnd_pixels_tab; |
255 | 1030 op_qpix= qpel_mc_no_rnd_tab; |
1031 } | |
0 | 1032 |
1033 if (s->mv_dir & MV_DIR_FORWARD) { | |
255 | 1034 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); |
0 | 1035 if (!s->no_rounding) |
1036 op_pix = avg_pixels_tab; | |
1037 else | |
1038 op_pix = avg_no_rnd_pixels_tab; | |
1039 } | |
1040 if (s->mv_dir & MV_DIR_BACKWARD) { | |
255 | 1041 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); |
0 | 1042 } |
1043 | |
1044 /* add dct residue */ | |
1045 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
1046 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1047 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1048 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1049 | |
57 | 1050 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1051 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1052 } else { |
1053 /* dct only in intra block */ | |
1054 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
1055 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1056 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1057 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1058 | |
57 | 1059 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1060 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1061 } |
1062 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1063 the_end: |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1064 emms_c(); |
0 | 1065 } |
1066 | |
1067 static void encode_picture(MpegEncContext *s, int picture_number) | |
1068 { | |
231 | 1069 int mb_x, mb_y, wrap, last_gob, pdif = 0; |
0 | 1070 UINT8 *ptr; |
1071 int i, motion_x, motion_y; | |
1072 | |
1073 s->picture_number = picture_number; | |
268 | 1074 |
1075 s->last_mc_mb_var = s->mc_mb_var; | |
1076 /* Reset the average MB variance */ | |
1077 s->avg_mb_var = 0; | |
1078 s->mc_mb_var = 0; | |
1079 | |
1080 /* Estimate motion for every MB */ | |
1081 for(mb_y=0; mb_y < s->mb_height; mb_y++) { | |
1082 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
1083 int xy= mb_y * s->mb_width + mb_x; | |
1084 s->mb_x = mb_x; | |
1085 s->mb_y = mb_y; | |
1086 | |
1087 /* compute motion vector and macro block type (intra or non intra) */ | |
1088 motion_x = 0; | |
1089 motion_y = 0; | |
1090 if (s->pict_type == P_TYPE) { | |
1091 s->mb_intra = estimate_motion(s, mb_x, mb_y, | |
1092 &motion_x, | |
1093 &motion_y); | |
1094 } else { | |
1095 s->mb_intra = 1; | |
1096 } | |
1097 /* Store MB type and MV */ | |
1098 s->mb_type[xy] = s->mb_intra; | |
1099 s->mv_table[0][xy] = motion_x; | |
1100 s->mv_table[1][xy] = motion_y; | |
1101 } | |
1102 } | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1103 emms_c(); |
268 | 1104 |
271 | 1105 if(s->avg_mb_var < s->mc_mb_var && s->pict_type != B_TYPE){ //FIXME subtract MV bits |
1106 int i; | |
1107 s->pict_type= I_TYPE; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1108 s->picture_in_gop_number=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1109 for(i=0; i<s->mb_num; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1110 s->mb_type[i] = 1; |
271 | 1111 s->mv_table[0][i] = 0; |
1112 s->mv_table[1][i] = 0; | |
1113 } | |
1114 } | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1115 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1116 /* find best f_code */ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1117 if(s->pict_type==P_TYPE){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1118 int mv_num[8]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1119 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1120 int loose=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1121 UINT8 * fcode_tab= s->fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1122 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1123 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
|
1124 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1125 for(i=0; i<s->mb_num; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1126 if(s->mb_type[i] == 0){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1127 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
|
1128 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
|
1129 //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
|
1130 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1131 //else printf("I"); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1132 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1133 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1134 for(i=MAX_FCODE; i>1; i--){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1135 loose+= mv_num[i]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1136 if(loose > 4) break; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1137 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1138 s->f_code= i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1139 }else{ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1140 s->f_code= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1141 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1142 //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
|
1143 /* 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
|
1144 if(s->pict_type==P_TYPE){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1145 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1146 const int f_code= s->f_code; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1147 UINT8 * fcode_tab= s->fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1148 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1149 for(i=0; i<s->mb_num; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1150 if(s->mb_type[i] == 0){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1151 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
|
1152 || 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
|
1153 || 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
|
1154 || fcode_tab[s->mv_table[1][i] + MAX_MV] == 0 ){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1155 s->mb_type[i] = 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1156 s->mv_table[0][i] = 0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1157 s->mv_table[1][i] = 0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1158 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1159 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1160 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1161 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1162 |
271 | 1163 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var); |
1164 | |
0 | 1165 if (!s->fixed_qscale) |
1166 s->qscale = rate_estimate_qscale(s); | |
1167 | |
1168 /* precompute matrix */ | |
1169 if (s->out_format == FMT_MJPEG) { | |
1170 /* for mjpeg, we do include qscale in the matrix */ | |
1171 s->intra_matrix[0] = default_intra_matrix[0]; | |
1172 for(i=1;i<64;i++) | |
1173 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
220 | 1174 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
0 | 1175 } else { |
220 | 1176 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
1177 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); | |
0 | 1178 } |
1179 | |
1180 switch(s->out_format) { | |
1181 case FMT_MJPEG: | |
1182 mjpeg_picture_header(s); | |
1183 break; | |
1184 case FMT_H263: | |
1185 if (s->h263_msmpeg4) | |
1186 msmpeg4_encode_picture_header(s, picture_number); | |
1187 else if (s->h263_pred) | |
1188 mpeg4_encode_picture_header(s, picture_number); | |
1189 else if (s->h263_rv10) | |
1190 rv10_encode_picture_header(s, picture_number); | |
1191 else | |
1192 h263_encode_picture_header(s, picture_number); | |
1193 break; | |
1194 case FMT_MPEG1: | |
1195 mpeg1_encode_picture_header(s, picture_number); | |
1196 break; | |
1197 } | |
1198 | |
1199 /* init last dc values */ | |
1200 /* note: quant matrix value (8) is implied here */ | |
1201 s->last_dc[0] = 128; | |
1202 s->last_dc[1] = 128; | |
1203 s->last_dc[2] = 128; | |
1204 s->mb_incr = 1; | |
1205 s->last_mv[0][0][0] = 0; | |
1206 s->last_mv[0][0][1] = 0; | |
1207 s->mv_type = MV_TYPE_16X16; | |
1208 s->mv_dir = MV_DIR_FORWARD; | |
1209 | |
162 | 1210 /* Get the GOB height based on picture height */ |
231 | 1211 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
162 | 1212 if (s->height <= 400) |
1213 s->gob_index = 1; | |
1214 else if (s->height <= 800) | |
1215 s->gob_index = 2; | |
1216 else | |
1217 s->gob_index = 4; | |
1218 } | |
268 | 1219 |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1220 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
|
1221 |
266 | 1222 s->block_wrap[0]= |
1223 s->block_wrap[1]= | |
1224 s->block_wrap[2]= | |
1225 s->block_wrap[3]= s->mb_width*2 + 2; | |
1226 s->block_wrap[4]= | |
1227 s->block_wrap[5]= s->mb_width + 2; | |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1228 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
|
1229 /* 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
|
1230 /* 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
|
1231 if (s->rtp_mode) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1232 if (!mb_y) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1233 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
|
1234 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
|
1235 } 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
|
1236 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
|
1237 if (last_gob) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1238 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
|
1239 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1240 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1241 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1242 |
266 | 1243 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; |
1244 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1245 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1246 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1247 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
1248 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
|
1249 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
0 | 1250 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1251 s->mb_x = mb_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1252 s->mb_y = mb_y; |
266 | 1253 s->block_index[0]+=2; |
1254 s->block_index[1]+=2; | |
1255 s->block_index[2]+=2; | |
1256 s->block_index[3]+=2; | |
1257 s->block_index[4]++; | |
1258 s->block_index[5]++; | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1259 #if 0 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1260 /* compute motion vector and macro block type (intra or non intra) */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1261 motion_x = 0; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1262 motion_y = 0; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1263 if (s->pict_type == P_TYPE) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1264 s->mb_intra = estimate_motion(s, mb_x, mb_y, |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1265 &motion_x, |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1266 &motion_y); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1267 } else { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1268 s->mb_intra = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1269 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1270 #endif |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1271 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1272 s->mb_intra = s->mb_type[mb_y * s->mb_width + mb_x]; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1273 motion_x = s->mv_table[0][mb_y * s->mb_width + mb_x]; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1274 motion_y = s->mv_table[1][mb_y * s->mb_width + mb_x]; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1275 |
0 | 1276 /* get the pixels */ |
1277 wrap = s->linesize; | |
1278 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
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
|
1279 get_pixels(s->block[0], ptr, wrap); |
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
|
1280 get_pixels(s->block[1], ptr + 8, wrap); |
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
|
1281 get_pixels(s->block[2], ptr + 8 * wrap, wrap); |
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
|
1282 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); |
0 | 1283 wrap = s->linesize >> 1; |
1284 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
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
|
1285 get_pixels(s->block[4], ptr, wrap); |
0 | 1286 |
1287 wrap = s->linesize >> 1; | |
1288 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; | |
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
|
1289 get_pixels(s->block[5], ptr, wrap); |
0 | 1290 |
1291 /* subtract previous frame if non intra */ | |
1292 if (!s->mb_intra) { | |
1293 int dxy, offset, mx, my; | |
1294 | |
1295 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
1296 ptr = s->last_picture[0] + | |
1297 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) + | |
1298 (mb_x * 16 + (motion_x >> 1)); | |
1299 | |
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
|
1300 sub_pixels_2(s->block[0], ptr, s->linesize, dxy); |
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
|
1301 sub_pixels_2(s->block[1], ptr + 8, s->linesize, dxy); |
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
|
1302 sub_pixels_2(s->block[2], ptr + s->linesize * 8, s->linesize, dxy); |
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
|
1303 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy); |
0 | 1304 |
1305 if (s->out_format == FMT_H263) { | |
1306 /* special rounding for h263 */ | |
1307 dxy = 0; | |
1308 if ((motion_x & 3) != 0) | |
1309 dxy |= 1; | |
1310 if ((motion_y & 3) != 0) | |
1311 dxy |= 2; | |
1312 mx = motion_x >> 2; | |
1313 my = motion_y >> 2; | |
1314 } else { | |
1315 mx = motion_x / 2; | |
1316 my = motion_y / 2; | |
1317 dxy = ((my & 1) << 1) | (mx & 1); | |
1318 mx >>= 1; | |
1319 my >>= 1; | |
1320 } | |
1321 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx); | |
1322 ptr = s->last_picture[1] + offset; | |
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
|
1323 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); |
0 | 1324 ptr = s->last_picture[2] + offset; |
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
|
1325 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); |
0 | 1326 } |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1327 emms_c(); |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1328 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1329 #if 0 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1330 { |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1331 float adap_parm; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1332 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1333 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1334 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1335 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1336 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1337 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1338 s->qscale, adap_parm, s->qscale*adap_parm, |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1339 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1340 } |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1341 #endif |
0 | 1342 /* DCT & quantize */ |
1343 if (s->h263_msmpeg4) { | |
1344 msmpeg4_dc_scale(s); | |
1345 } else if (s->h263_pred) { | |
1346 h263_dc_scale(s); | |
1347 } else { | |
1348 /* default quantization values */ | |
1349 s->y_dc_scale = 8; | |
1350 s->c_dc_scale = 8; | |
1351 } | |
1352 for(i=0;i<6;i++) { | |
220 | 1353 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); |
0 | 1354 } |
1355 | |
1356 /* huffman encode */ | |
1357 switch(s->out_format) { | |
1358 case FMT_MPEG1: | |
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
|
1359 mpeg1_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1360 break; |
1361 case FMT_H263: | |
1362 if (s->h263_msmpeg4) | |
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
|
1363 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); |
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
1364 else if(s->h263_pred) |
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
1365 mpeg4_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1366 else |
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
|
1367 h263_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1368 break; |
1369 case FMT_MJPEG: | |
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
|
1370 mjpeg_encode_mb(s, s->block); |
0 | 1371 break; |
1372 } | |
1373 | |
1374 /* decompress blocks so that we keep the state of the decoder */ | |
1375 s->mv[0][0][0] = motion_x; | |
1376 s->mv[0][0][1] = motion_y; | |
1377 | |
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
|
1378 MPV_decode_mb(s, s->block); |
0 | 1379 } |
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
|
1380 |
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
|
1381 |
231 | 1382 /* Obtain average GOB size for RTP */ |
1383 if (s->rtp_mode) { | |
1384 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
|
1385 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
231 | 1386 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
|
1387 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
|
1388 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
231 | 1389 } |
1390 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
1391 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
1392 s->first_gob_line = 0; | |
1393 } | |
0 | 1394 } |
208 | 1395 |
251
75091bfc577b
fixing msmpeg4 decoding if fps < 16 (i thought it was a indicator for the ext header, its the fps indeed)
michaelni
parents:
249
diff
changeset
|
1396 if (s->h263_msmpeg4 && s->pict_type == I_TYPE) |
208 | 1397 msmpeg4_encode_ext_header(s); |
1398 | |
162 | 1399 //if (s->gob_number) |
1400 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
231 | 1401 |
1402 /* Send the last GOB if RTP */ | |
1403 if (s->rtp_mode) { | |
1404 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
|
1405 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 1406 /* Call the RTP callback to send the last GOB */ |
1407 if (s->rtp_callback) | |
1408 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
|
1409 s->ptr_lastgob = pbBufPtr(&s->pb); |
231 | 1410 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
1411 } | |
1412 | |
0 | 1413 } |
1414 | |
220 | 1415 static int dct_quantize_c(MpegEncContext *s, |
0 | 1416 DCTELEM *block, int n, |
1417 int qscale) | |
1418 { | |
1419 int i, j, level, last_non_zero, q; | |
1420 const int *qmat; | |
216 | 1421 int minLevel, maxLevel; |
1422 | |
1423 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ | |
1424 /* mpeg4 */ | |
1425 minLevel= -2048; | |
1426 maxLevel= 2047; | |
1427 }else if(s->out_format==FMT_MPEG1){ | |
1428 /* mpeg1 */ | |
1429 minLevel= -255; | |
1430 maxLevel= 255; | |
230 | 1431 }else if(s->out_format==FMT_MJPEG){ |
1432 /* (m)jpeg */ | |
1433 minLevel= -1023; | |
1434 maxLevel= 1023; | |
216 | 1435 }else{ |
1436 /* h263 / msmpeg4 */ | |
1437 minLevel= -128; | |
1438 maxLevel= 127; | |
1439 } | |
0 | 1440 |
1441 av_fdct (block); | |
1442 | |
64 | 1443 /* we need this permutation so that we correct the IDCT |
1444 permutation. will be moved into DCT code */ | |
1445 block_permute(block); | |
1446 | |
0 | 1447 if (s->mb_intra) { |
1448 if (n < 4) | |
1449 q = s->y_dc_scale; | |
1450 else | |
1451 q = s->c_dc_scale; | |
1452 q = q << 3; | |
1453 | |
1454 /* note: block[0] is assumed to be positive */ | |
1455 block[0] = (block[0] + (q >> 1)) / q; | |
1456 i = 1; | |
1457 last_non_zero = 0; | |
1458 if (s->out_format == FMT_H263) { | |
1459 qmat = s->q_non_intra_matrix; | |
1460 } else { | |
1461 qmat = s->q_intra_matrix; | |
1462 } | |
1463 } else { | |
1464 i = 0; | |
1465 last_non_zero = -1; | |
1466 qmat = s->q_non_intra_matrix; | |
1467 } | |
1468 | |
1469 for(;i<64;i++) { | |
1470 j = zigzag_direct[i]; | |
1471 level = block[j]; | |
1472 level = level * qmat[j]; | |
1473 #ifdef PARANOID | |
1474 { | |
1475 static int count = 0; | |
1476 int level1, level2, qmat1; | |
1477 double val; | |
1478 if (qmat == s->q_non_intra_matrix) { | |
1479 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1480 } else { | |
1481 qmat1 = default_intra_matrix[j] * s->qscale; | |
1482 } | |
1483 if (av_fdct != jpeg_fdct_ifast) | |
1484 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1485 else | |
1486 val = ((double)block[j] * 8.0 * 2048.0) / | |
1487 ((double)qmat1 * aanscales[j]); | |
1488 level1 = (int)val; | |
1489 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1490 if (level1 != level2) { | |
1491 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1492 count, level2, level1, block[j], qmat1, qmat[j], | |
1493 val); | |
1494 count++; | |
1495 } | |
1496 | |
1497 } | |
1498 #endif | |
1499 /* XXX: slight error for the low range. Test should be equivalent to | |
1500 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1501 (QMAT_SHIFT - 3))) | |
1502 */ | |
1503 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1504 level) { | |
1505 level = level / (1 << (QMAT_SHIFT - 3)); | |
1506 /* XXX: currently, this code is not optimal. the range should be: | |
1507 mpeg1: -255..255 | |
1508 mpeg2: -2048..2047 | |
1509 h263: -128..127 | |
1510 mpeg4: -2048..2047 | |
1511 */ | |
216 | 1512 if (level > maxLevel) |
1513 level = maxLevel; | |
1514 else if (level < minLevel) | |
1515 level = minLevel; | |
0 | 1516 |
1517 block[j] = level; | |
1518 last_non_zero = i; | |
1519 } else { | |
1520 block[j] = 0; | |
1521 } | |
1522 } | |
1523 return last_non_zero; | |
1524 } | |
1525 | |
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
|
1526 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
|
1527 DCTELEM *block, int n, int qscale) |
0 | 1528 { |
200 | 1529 int i, level, nCoeffs; |
0 | 1530 const UINT16 *quant_matrix; |
1531 | |
200 | 1532 if(s->alternate_scan) nCoeffs= 64; |
1533 else nCoeffs= s->block_last_index[n]+1; | |
1534 | |
0 | 1535 if (s->mb_intra) { |
1536 if (n < 4) | |
1537 block[0] = block[0] * s->y_dc_scale; | |
1538 else | |
1539 block[0] = block[0] * s->c_dc_scale; | |
1540 /* XXX: only mpeg1 */ | |
1541 quant_matrix = s->intra_matrix; | |
200 | 1542 for(i=1;i<nCoeffs;i++) { |
1543 int j= zigzag_direct[i]; | |
1544 level = block[j]; | |
0 | 1545 if (level) { |
1546 if (level < 0) { | |
1547 level = -level; | |
200 | 1548 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1549 level = (level - 1) | 1; |
1550 level = -level; | |
1551 } else { | |
200 | 1552 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1553 level = (level - 1) | 1; |
1554 } | |
1555 #ifdef PARANOID | |
1556 if (level < -2048 || level > 2047) | |
1557 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1558 #endif | |
200 | 1559 block[j] = level; |
0 | 1560 } |
1561 } | |
1562 } else { | |
1563 i = 0; | |
1564 quant_matrix = s->non_intra_matrix; | |
217 | 1565 for(;i<nCoeffs;i++) { |
200 | 1566 int j= zigzag_direct[i]; |
1567 level = block[j]; | |
0 | 1568 if (level) { |
1569 if (level < 0) { | |
1570 level = -level; | |
1571 level = (((level << 1) + 1) * qscale * | |
200 | 1572 ((int) (quant_matrix[j]))) >> 4; |
0 | 1573 level = (level - 1) | 1; |
1574 level = -level; | |
1575 } else { | |
1576 level = (((level << 1) + 1) * qscale * | |
200 | 1577 ((int) (quant_matrix[j]))) >> 4; |
0 | 1578 level = (level - 1) | 1; |
1579 } | |
1580 #ifdef PARANOID | |
1581 if (level < -2048 || level > 2047) | |
1582 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1583 #endif | |
200 | 1584 block[j] = level; |
0 | 1585 } |
1586 } | |
1587 } | |
1588 } | |
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
|
1589 |
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
|
1590 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
|
1591 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
|
1592 { |
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
|
1593 int i, level, qmul, qadd; |
200 | 1594 int nCoeffs; |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1595 |
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
|
1596 if (s->mb_intra) { |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1597 if (!s->h263_aic) { |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1598 if (n < 4) |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1599 block[0] = block[0] * s->y_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1600 else |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1601 block[0] = block[0] * s->c_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1602 } |
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
|
1603 i = 1; |
200 | 1604 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
|
1605 } 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
|
1606 i = 0; |
200 | 1607 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
|
1608 } |
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
|
1609 |
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
|
1610 qmul = s->qscale << 1; |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1611 if (s->h263_aic && s->mb_intra) |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1612 qadd = 0; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1613 else |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1614 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
|
1615 |
200 | 1616 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
|
1617 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
|
1618 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
|
1619 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
|
1620 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
|
1621 } 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
|
1622 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
|
1623 } |
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
|
1624 #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
|
1625 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
|
1626 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
|
1627 #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
|
1628 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
|
1629 } |
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
|
1630 } |
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
|
1631 } |
0 | 1632 |
1633 /* rate control */ | |
1634 | |
1635 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */ | |
1636 #define I_FRAME_SIZE_RATIO 3.0 | |
1637 #define QSCALE_K 20 | |
1638 | |
1639 static void rate_control_init(MpegEncContext *s) | |
1640 { | |
268 | 1641 #if 1 |
1642 emms_c(); | |
1643 | |
1644 //initial values, they dont really matter as they will be totally different within a few frames | |
1645 s->i_pred.coeff= s->p_pred.coeff= 7.0; | |
1646 s->i_pred.count= s->p_pred.count= 1.0; | |
1647 | |
1648 s->i_pred.decay= s->p_pred.decay= 0.4; | |
1649 | |
1650 // use more bits at the beginning, otherwise high motion at the begin will look like shit | |
1651 s->qsum=100; | |
1652 s->qcount=100; | |
1653 | |
1654 s->short_term_qsum=0.001; | |
1655 s->short_term_qcount=0.001; | |
1656 #else | |
0 | 1657 s->wanted_bits = 0; |
1658 | |
1659 if (s->intra_only) { | |
1660 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate; | |
1661 s->P_frame_bits = s->I_frame_bits; | |
1662 } else { | |
1663 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) / | |
1664 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1))); | |
1665 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO); | |
1666 } | |
268 | 1667 |
0 | 1668 #if defined(DEBUG) |
1669 printf("I_frame_size=%d P_frame_size=%d\n", | |
1670 s->I_frame_bits, s->P_frame_bits); | |
1671 #endif | |
268 | 1672 #endif |
1673 } | |
1674 | |
1675 static double predict(Predictor *p, double q, double var) | |
1676 { | |
1677 return p->coeff*var / (q*p->count); | |
1678 } | |
1679 | |
1680 static void update_predictor(Predictor *p, double q, double var, double size) | |
1681 { | |
1682 double new_coeff= size*q / (var + 1); | |
1683 if(var<1000) return; | |
1684 /*{ | |
1685 int pred= predict(p, q, var); | |
1686 int error= abs(pred-size); | |
1687 static double sum=0; | |
1688 static int count=0; | |
1689 if(count>5) sum+=error; | |
1690 count++; | |
1691 if(256*256*256*64%count==0){ | |
1692 printf("%d %f %f\n", count, sum/count, p->coeff); | |
1693 } | |
1694 }*/ | |
1695 p->count*= p->decay; | |
1696 p->coeff*= p->decay; | |
1697 p->count++; | |
1698 p->coeff+= new_coeff; | |
0 | 1699 } |
1700 | |
1701 static int rate_estimate_qscale(MpegEncContext *s) | |
1702 { | |
268 | 1703 #if 1 |
1704 int qmin= s->qmin; | |
1705 int qmax= s->qmax; | |
1706 int rate_q=5; | |
1707 float q; | |
1708 int qscale; | |
1709 float br_compensation; | |
1710 double diff; | |
1711 double short_term_q; | |
1712 double long_term_q; | |
1713 int last_qscale= s->qscale; | |
1714 double fps; | |
1715 INT64 wanted_bits; | |
1716 emms_c(); | |
1717 | |
1718 fps= (double)s->frame_rate / FRAME_RATE_BASE; | |
1719 wanted_bits= s->bit_rate*(double)s->picture_number/fps; | |
1720 | |
1721 | |
1722 if(s->picture_number>2){ | |
1723 /* update predictors */ | |
1724 if(s->last_pict_type == I_TYPE){ | |
1725 //FIXME | |
1726 }else{ //P Frame | |
1727 //printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff); | |
1728 update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits); | |
1729 } | |
1730 } | |
1731 | |
1732 if(s->pict_type == I_TYPE){ | |
1733 //FIXME | |
1734 rate_q= s->qsum/s->qcount; | |
1735 }else{ //P Frame | |
1736 int i; | |
1737 int diff, best_diff=1000000000; | |
1738 for(i=1; i<=31; i++){ | |
1739 diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps; | |
1740 if(diff<0) diff= -diff; | |
1741 if(diff<best_diff){ | |
1742 best_diff= diff; | |
1743 rate_q= i; | |
1744 } | |
1745 } | |
1746 } | |
1747 | |
1748 s->short_term_qsum*=s->qblur; | |
1749 s->short_term_qcount*=s->qblur; | |
1750 | |
1751 s->short_term_qsum+= rate_q; | |
1752 s->short_term_qcount++; | |
1753 short_term_q= s->short_term_qsum/s->short_term_qcount; | |
1754 | |
1755 long_term_q= s->qsum/s->qcount*s->total_bits/wanted_bits; | |
1756 | |
1757 // q= (long_term_q - short_term_q)*s->qcompress + short_term_q; | |
1758 q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q); | |
1759 | |
1760 diff= s->total_bits - wanted_bits; | |
1761 br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance; | |
270 | 1762 if(br_compensation<=0.0) br_compensation=0.001; |
268 | 1763 q/=br_compensation; |
1764 | |
1765 qscale= (int)(q + 0.5); | |
1766 if (qscale<qmin) qscale=qmin; | |
1767 else if(qscale>qmax) qscale=qmax; | |
1768 | |
1769 if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff; | |
1770 else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff; | |
1771 | |
1772 s->qsum+= qscale; | |
1773 s->qcount++; | |
1774 | |
1775 s->last_pict_type= s->pict_type; | |
1776 //printf("q:%d diff:%d comp:%f rate_q:%d st_q:%d fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation, | |
1777 // rate_q, (int)short_term_q, s->mc_mb_var, s->frame_bits); | |
1778 //printf("%d %d\n", s->bit_rate, (int)fps); | |
1779 return qscale; | |
1780 #else | |
187
3f3b14d3a23d
qscale estimate fix, diff extended to 64bit wide. patch by Stephen Davies <steve@daviesfam.org>
arpi_esp
parents:
178
diff
changeset
|
1781 INT64 diff, total_bits = s->total_bits; |
0 | 1782 float q; |
268 | 1783 int qscale; |
0 | 1784 if (s->pict_type == I_TYPE) { |
1785 s->wanted_bits += s->I_frame_bits; | |
1786 } else { | |
1787 s->wanted_bits += s->P_frame_bits; | |
1788 } | |
1789 diff = s->wanted_bits - total_bits; | |
1790 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width); | |
1791 /* adjust for I frame */ | |
1792 if (s->pict_type == I_TYPE && !s->intra_only) { | |
1793 q /= I_FRAME_SIZE_RATIO; | |
1794 } | |
1795 | |
1796 /* using a too small Q scale leeds to problems in mpeg1 and h263 | |
1797 because AC coefficients are clamped to 255 or 127 */ | |
1798 qmin = 3; | |
1799 if (q < qmin) | |
1800 q = qmin; | |
1801 else if (q > 31) | |
1802 q = 31; | |
1803 qscale = (int)(q + 0.5); | |
1804 #if defined(DEBUG) | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1805 printf("\n%d: total=%0.0f wanted=%0.0f br=%0.1f diff=%d qest=%2.1f\n", |
0 | 1806 s->picture_number, |
64 | 1807 (double)total_bits, |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1808 (double)s->wanted_bits, |
0 | 1809 (float)s->frame_rate / FRAME_RATE_BASE * |
1810 total_bits / s->picture_number, | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1811 (int)diff, q); |
0 | 1812 #endif |
1813 return qscale; | |
268 | 1814 #endif |
0 | 1815 } |
1816 | |
1817 AVCodec mpeg1video_encoder = { | |
1818 "mpeg1video", | |
1819 CODEC_TYPE_VIDEO, | |
1820 CODEC_ID_MPEG1VIDEO, | |
1821 sizeof(MpegEncContext), | |
1822 MPV_encode_init, | |
1823 MPV_encode_picture, | |
1824 MPV_encode_end, | |
1825 }; | |
1826 | |
1827 AVCodec h263_encoder = { | |
1828 "h263", | |
1829 CODEC_TYPE_VIDEO, | |
1830 CODEC_ID_H263, | |
1831 sizeof(MpegEncContext), | |
1832 MPV_encode_init, | |
1833 MPV_encode_picture, | |
1834 MPV_encode_end, | |
1835 }; | |
1836 | |
1837 AVCodec h263p_encoder = { | |
1838 "h263p", | |
1839 CODEC_TYPE_VIDEO, | |
1840 CODEC_ID_H263P, | |
1841 sizeof(MpegEncContext), | |
1842 MPV_encode_init, | |
1843 MPV_encode_picture, | |
1844 MPV_encode_end, | |
1845 }; | |
1846 | |
1847 AVCodec rv10_encoder = { | |
1848 "rv10", | |
1849 CODEC_TYPE_VIDEO, | |
1850 CODEC_ID_RV10, | |
1851 sizeof(MpegEncContext), | |
1852 MPV_encode_init, | |
1853 MPV_encode_picture, | |
1854 MPV_encode_end, | |
1855 }; | |
1856 | |
1857 AVCodec mjpeg_encoder = { | |
1858 "mjpeg", | |
1859 CODEC_TYPE_VIDEO, | |
1860 CODEC_ID_MJPEG, | |
1861 sizeof(MpegEncContext), | |
1862 MPV_encode_init, | |
1863 MPV_encode_picture, | |
1864 MPV_encode_end, | |
1865 }; | |
1866 | |
71 | 1867 AVCodec mpeg4_encoder = { |
1868 "mpeg4", | |
0 | 1869 CODEC_TYPE_VIDEO, |
71 | 1870 CODEC_ID_MPEG4, |
0 | 1871 sizeof(MpegEncContext), |
1872 MPV_encode_init, | |
1873 MPV_encode_picture, | |
1874 MPV_encode_end, | |
1875 }; | |
1876 | |
1877 AVCodec msmpeg4_encoder = { | |
1878 "msmpeg4", | |
1879 CODEC_TYPE_VIDEO, | |
1880 CODEC_ID_MSMPEG4, | |
1881 sizeof(MpegEncContext), | |
1882 MPV_encode_init, | |
1883 MPV_encode_picture, | |
1884 MPV_encode_end, | |
1885 }; |