Mercurial > libavcodec.hg
annotate mpegvideo.c @ 302:826b179dad50 libavcodec
fixing 1/0
author | michaelni |
---|---|
date | Sat, 30 Mar 2002 21:33:14 +0000 |
parents | e20de99b6295 |
children | 764aeec1320e |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest mpeg encoder (well, it was the simplest!) | |
3 * Copyright (c) 2000,2001 Gerard Lantau. | |
4 * | |
5 * This program is free software; you can redistribute it and/or modify | |
6 * it under the terms of the GNU General Public License as published by | |
7 * the Free Software Foundation; either version 2 of the License, or | |
8 * (at your option) any later version. | |
9 * | |
10 * This program is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 * GNU General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU General Public License | |
16 * along with this program; if not, write to the Free Software | |
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
295 | 18 * |
19 * 4MV & hq encoding stuff by Michael Niedermayer <michaelni@gmx.at> | |
0 | 20 */ |
21 #include <stdlib.h> | |
22 #include <stdio.h> | |
23 #include <math.h> | |
24 #include <string.h> | |
25 #include "avcodec.h" | |
26 #include "dsputil.h" | |
27 #include "mpegvideo.h" | |
28 | |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
29 #ifdef USE_FASTMEMCPY |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
30 #include "fastmemcpy.h" |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
31 #endif |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
32 |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
33 static void encode_picture(MpegEncContext *s, int picture_number); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
34 static void rate_control_init(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
35 static int rate_estimate_qscale(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
36 static void dct_unquantize_mpeg1_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
37 DCTELEM *block, int n, int qscale); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
38 static void dct_unquantize_h263_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
39 DCTELEM *block, int n, int qscale); |
206 | 40 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
220 | 41 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
206 | 42 |
220 | 43 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c; |
206 | 44 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
45 | |
0 | 46 #define EDGE_WIDTH 16 |
47 | |
48 /* enable all paranoid tests for rounding, overflows, etc... */ | |
49 //#define PARANOID | |
50 | |
51 //#define DEBUG | |
52 | |
53 /* for jpeg fast DCT */ | |
54 #define CONST_BITS 14 | |
55 | |
56 static const unsigned short aanscales[64] = { | |
57 /* precomputed values scaled up by 14 bits */ | |
58 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
59 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
60 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
61 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
62 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
63 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
64 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
65 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
66 }; | |
67 | |
68 static UINT8 h263_chroma_roundtab[16] = { | |
69 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
70 }; | |
71 | |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
72 static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1]; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
73 static UINT8 default_fcode_tab[MAX_MV*2+1]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
74 |
0 | 75 /* default motion estimation */ |
76 int motion_estimation_method = ME_LOG; | |
77 | |
200 | 78 extern UINT8 zigzag_end[64]; |
79 | |
220 | 80 static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale) |
0 | 81 { |
82 int i; | |
83 | |
84 if (av_fdct == jpeg_fdct_ifast) { | |
85 for(i=0;i<64;i++) { | |
86 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
220 | 87 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
88 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
89 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
0 | 90 |
220 | 91 qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
92 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)])); | |
0 | 93 } |
94 } else { | |
95 for(i=0;i<64;i++) { | |
96 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
220 | 97 So 16 <= qscale * quant_matrix[i] <= 7905 |
98 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
99 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
0 | 100 */ |
220 | 101 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
102 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
0 | 103 } |
104 } | |
105 } | |
106 | |
107 /* init common structure for both encoder and decoder */ | |
108 int MPV_common_init(MpegEncContext *s) | |
109 { | |
110 int c_size, i; | |
111 UINT8 *pict; | |
112 | |
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
|
113 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
|
114 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
|
115 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
|
116 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
|
117 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
118 #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
|
119 MPV_common_init_mmx(s); |
8 | 120 #endif |
0 | 121 s->mb_width = (s->width + 15) / 16; |
122 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
|
123 s->mb_num = s->mb_width * s->mb_height; |
0 | 124 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
125 | |
126 for(i=0;i<3;i++) { | |
127 int w, h, shift, pict_start; | |
128 | |
129 w = s->linesize; | |
130 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
131 shift = (i == 0) ? 0 : 1; | |
132 c_size = (w >> shift) * (h >> shift); | |
133 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
134 | |
135 pict = av_mallocz(c_size); | |
136 if (pict == NULL) | |
137 goto fail; | |
138 s->last_picture_base[i] = pict; | |
139 s->last_picture[i] = pict + pict_start; | |
140 | |
141 pict = av_mallocz(c_size); | |
142 if (pict == NULL) | |
143 goto fail; | |
144 s->next_picture_base[i] = pict; | |
145 s->next_picture[i] = pict + pict_start; | |
146 | |
147 if (s->has_b_frames) { | |
148 pict = av_mallocz(c_size); | |
149 if (pict == NULL) | |
150 goto fail; | |
151 s->aux_picture_base[i] = pict; | |
152 s->aux_picture[i] = pict + pict_start; | |
153 } | |
154 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
155 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
156 if (s->encoding) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
157 /* Allocate MB type table */ |
239 | 158 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
|
159 if (s->mb_type == NULL) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
160 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
161 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
162 } |
239 | 163 |
164 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16)); | |
165 if (s->mb_var == NULL) { | |
166 perror("malloc"); | |
167 goto fail; | |
168 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
169 /* Allocate MV table */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
170 /* By now we just have one MV per MB */ |
239 | 171 s->mv_table[0] = av_mallocz(s->mb_num * sizeof(INT16)); |
172 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
|
173 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
|
174 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
175 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
176 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
177 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
178 |
288 | 179 if (s->out_format == FMT_H263 || s->encoding) { |
0 | 180 int size; |
181 /* MV prediction */ | |
182 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
183 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
184 if (s->motion_val == NULL) | |
185 goto fail; | |
186 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
187 } | |
188 | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
189 if (s->h263_pred || s->h263_plus) { |
0 | 190 int y_size, c_size, i, size; |
191 | |
192 /* dc values */ | |
193 | |
194 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
195 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
196 size = y_size + 2 * c_size; | |
197 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
198 if (s->dc_val[0] == NULL) | |
199 goto fail; | |
200 s->dc_val[1] = s->dc_val[0] + y_size; | |
201 s->dc_val[2] = s->dc_val[1] + c_size; | |
202 for(i=0;i<size;i++) | |
203 s->dc_val[0][i] = 1024; | |
204 | |
205 /* ac values */ | |
206 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
207 if (s->ac_val[0] == NULL) | |
208 goto fail; | |
209 s->ac_val[1] = s->ac_val[0] + y_size; | |
210 s->ac_val[2] = s->ac_val[1] + c_size; | |
211 | |
212 /* cbp values */ | |
213 s->coded_block = av_mallocz(y_size); | |
214 if (!s->coded_block) | |
215 goto fail; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
216 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
217 /* 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
|
218 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
|
219 if (!s->mbintra_table) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
220 goto fail; |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
221 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
|
222 } |
0 | 223 /* default structure is frame */ |
224 s->picture_structure = PICT_FRAME; | |
225 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
226 /* init macroblock skip table */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
227 if (!s->encoding) { |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
228 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
|
229 if (!s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
230 goto fail; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
231 } |
294 | 232 |
233 s->block= s->intra_block; | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
234 |
0 | 235 s->context_initialized = 1; |
236 return 0; | |
237 fail: | |
244 | 238 MPV_common_end(s); |
239 return -1; | |
240 } | |
241 | |
242 /* init common structure for both encoder and decoder */ | |
243 void MPV_common_end(MpegEncContext *s) | |
244 { | |
245 int i; | |
246 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
247 if (s->mb_type) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
248 free(s->mb_type); |
239 | 249 if (s->mb_var) |
250 free(s->mb_var); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
251 if (s->mv_table[0]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
252 free(s->mv_table[0]); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
253 if (s->mv_table[1]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
254 free(s->mv_table[1]); |
0 | 255 if (s->motion_val) |
256 free(s->motion_val); | |
257 if (s->dc_val[0]) | |
258 free(s->dc_val[0]); | |
259 if (s->ac_val[0]) | |
260 free(s->ac_val[0]); | |
261 if (s->coded_block) | |
262 free(s->coded_block); | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
263 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
|
264 free(s->mbintra_table); |
244 | 265 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
266 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
267 free(s->mbskip_table); |
0 | 268 for(i=0;i<3;i++) { |
269 if (s->last_picture_base[i]) | |
244 | 270 free(s->last_picture_base[i]); |
271 if (s->next_picture_base[i]) | |
272 free(s->next_picture_base[i]); | |
0 | 273 if (s->has_b_frames) |
274 free(s->aux_picture_base[i]); | |
275 } | |
276 s->context_initialized = 0; | |
277 } | |
278 | |
279 /* init video encoder */ | |
280 int MPV_encode_init(AVCodecContext *avctx) | |
281 { | |
282 MpegEncContext *s = avctx->priv_data; | |
60 | 283 int i; |
0 | 284 |
285 s->bit_rate = avctx->bit_rate; | |
268 | 286 s->bit_rate_tolerance = avctx->bit_rate_tolerance; |
0 | 287 s->frame_rate = avctx->frame_rate; |
288 s->width = avctx->width; | |
289 s->height = avctx->height; | |
290 s->gop_size = avctx->gop_size; | |
162 | 291 s->rtp_mode = avctx->rtp_mode; |
292 s->rtp_payload_size = avctx->rtp_payload_size; | |
231 | 293 if (avctx->rtp_callback) |
294 s->rtp_callback = avctx->rtp_callback; | |
268 | 295 s->qmin= avctx->qmin; |
296 s->qmax= avctx->qmax; | |
297 s->max_qdiff= avctx->max_qdiff; | |
298 s->qcompress= avctx->qcompress; | |
299 s->qblur= avctx->qblur; | |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
300 s->avctx = avctx; |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
301 s->aspect_ratio_info= avctx->aspect_ratio_info; |
294 | 302 s->flags= avctx->flags; |
162 | 303 |
0 | 304 if (s->gop_size <= 1) { |
305 s->intra_only = 1; | |
306 s->gop_size = 12; | |
307 } else { | |
308 s->intra_only = 0; | |
309 } | |
310 s->full_search = motion_estimation_method; | |
311 | |
312 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
|
313 |
0 | 314 switch(avctx->codec->id) { |
315 case CODEC_ID_MPEG1VIDEO: | |
316 s->out_format = FMT_MPEG1; | |
317 break; | |
318 case CODEC_ID_MJPEG: | |
319 s->out_format = FMT_MJPEG; | |
320 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
|
321 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
|
322 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
|
323 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
|
324 s->mjpeg_vsample[2] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
325 s->mjpeg_hsample[0] = 2; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
326 s->mjpeg_hsample[1] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
327 s->mjpeg_hsample[2] = 1; |
0 | 328 if (mjpeg_init(s) < 0) |
329 return -1; | |
330 break; | |
331 case CODEC_ID_H263: | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
332 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
|
333 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
0 | 334 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
335 } |
0 | 336 s->out_format = FMT_H263; |
337 break; | |
338 case CODEC_ID_H263P: | |
339 s->out_format = FMT_H263; | |
162 | 340 s->rtp_mode = 1; |
341 s->rtp_payload_size = 1200; | |
0 | 342 s->h263_plus = 1; |
78 | 343 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
344 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
345 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
346 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
347 s->umvplus_dec = 0; |
0 | 348 break; |
349 case CODEC_ID_RV10: | |
350 s->out_format = FMT_H263; | |
351 s->h263_rv10 = 1; | |
352 break; | |
71 | 353 case CODEC_ID_MPEG4: |
0 | 354 s->out_format = FMT_H263; |
355 s->h263_pred = 1; | |
356 s->unrestricted_mv = 1; | |
357 break; | |
358 case CODEC_ID_MSMPEG4: | |
359 s->out_format = FMT_H263; | |
360 s->h263_msmpeg4 = 1; | |
361 s->h263_pred = 1; | |
362 s->unrestricted_mv = 1; | |
363 break; | |
364 default: | |
365 return -1; | |
366 } | |
295 | 367 |
368 if((s->flags&CODEC_FLAG_4MV) && !(s->flags&CODEC_FLAG_HQ)){ | |
369 printf("4MV is currently only supported in HQ mode\n"); | |
370 return -1; | |
371 } | |
0 | 372 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
373 { /* 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
|
374 static int done=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
375 if(!done){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
376 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
377 done=1; |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
378 memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1)); |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
379 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
|
380 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
381 for(i=-16; i<16; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
382 default_fcode_tab[i + MAX_MV]= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
383 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
384 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
385 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
386 s->mv_penalty= default_mv_penalty; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
387 s->fcode_tab= default_fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
388 |
0 | 389 if (s->out_format == FMT_H263) |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
390 h263_encode_init(s); |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
391 else if (s->out_format == FMT_MPEG1) |
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
392 mpeg1_encode_init(s); |
0 | 393 |
287 | 394 /* dont use mv_penalty table for crap MV as it would be confused */ |
395 if(s->full_search<4) s->mv_penalty= default_mv_penalty; | |
396 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
397 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
398 |
0 | 399 /* init */ |
400 if (MPV_common_init(s) < 0) | |
401 return -1; | |
402 | |
60 | 403 /* init default q matrix */ |
404 for(i=0;i<64;i++) { | |
405 s->intra_matrix[i] = default_intra_matrix[i]; | |
406 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
407 } | |
408 | |
0 | 409 /* rate control init */ |
410 rate_control_init(s); | |
411 | |
412 s->picture_number = 0; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
413 s->picture_in_gop_number = 0; |
0 | 414 s->fake_picture_number = 0; |
415 /* motion detector init */ | |
416 s->f_code = 1; | |
417 | |
418 return 0; | |
419 } | |
420 | |
421 int MPV_encode_end(AVCodecContext *avctx) | |
422 { | |
423 MpegEncContext *s = avctx->priv_data; | |
424 | |
425 #ifdef STATS | |
426 print_stats(); | |
427 #endif | |
428 MPV_common_end(s); | |
429 if (s->out_format == FMT_MJPEG) | |
430 mjpeg_close(s); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
431 |
0 | 432 return 0; |
433 } | |
434 | |
435 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 436 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 437 { |
438 UINT8 *ptr, *last_line; | |
439 int i; | |
440 | |
441 last_line = buf + (height - 1) * wrap; | |
442 for(i=0;i<w;i++) { | |
443 /* top and bottom */ | |
444 memcpy(buf - (i + 1) * wrap, buf, width); | |
445 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
446 } | |
447 /* left and right */ | |
448 ptr = buf; | |
449 for(i=0;i<height;i++) { | |
450 memset(ptr - w, ptr[0], w); | |
451 memset(ptr + width, ptr[width-1], w); | |
452 ptr += wrap; | |
453 } | |
454 /* corners */ | |
455 for(i=0;i<w;i++) { | |
456 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
457 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
458 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
459 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
460 } | |
461 } | |
462 | |
463 /* generic function for encode/decode called before a frame is coded/decoded */ | |
464 void MPV_frame_start(MpegEncContext *s) | |
465 { | |
466 int i; | |
467 UINT8 *tmp; | |
468 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
469 s->mb_skiped = 0; |
0 | 470 if (s->pict_type == B_TYPE) { |
471 for(i=0;i<3;i++) { | |
472 s->current_picture[i] = s->aux_picture[i]; | |
473 } | |
474 } else { | |
262 | 475 s->last_non_b_pict_type= s->pict_type; |
0 | 476 for(i=0;i<3;i++) { |
477 /* swap next and last */ | |
478 tmp = s->last_picture[i]; | |
479 s->last_picture[i] = s->next_picture[i]; | |
480 s->next_picture[i] = tmp; | |
481 s->current_picture[i] = tmp; | |
482 } | |
483 } | |
484 } | |
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
|
485 |
0 | 486 /* generic function for encode/decode called after a frame has been coded/decoded */ |
487 void MPV_frame_end(MpegEncContext *s) | |
488 { | |
489 /* draw edge for correct motion prediction if outside */ | |
220 | 490 if (s->pict_type != B_TYPE && !s->intra_only) { |
257 | 491 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
|
492 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
|
493 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
|
494 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
|
495 }else{ |
257 | 496 /* mpeg4? / opendivx / xvid */ |
0 | 497 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
498 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
499 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
|
500 } |
0 | 501 } |
207 | 502 emms_c(); |
0 | 503 } |
504 | |
505 int MPV_encode_picture(AVCodecContext *avctx, | |
506 unsigned char *buf, int buf_size, void *data) | |
507 { | |
508 MpegEncContext *s = avctx->priv_data; | |
509 AVPicture *pict = data; | |
510 int i, j; | |
511 | |
512 if (s->fixed_qscale) | |
513 s->qscale = avctx->quality; | |
514 | |
515 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
516 | |
298
e20de99b6295
using Juanjo's way to pass forced type - especially as I_TYPE/P_TYPE defines aren't exported in the public interface, and the flagged method matches the way as qscale forcing work
arpi_esp
parents:
297
diff
changeset
|
517 s->force_type= (avctx->flags&CODEC_FLAG_TYPE) ? |
e20de99b6295
using Juanjo's way to pass forced type - especially as I_TYPE/P_TYPE defines aren't exported in the public interface, and the flagged method matches the way as qscale forcing work
arpi_esp
parents:
297
diff
changeset
|
518 (avctx->key_frame ? I_TYPE : P_TYPE) : 0; |
0 | 519 if (!s->intra_only) { |
520 /* first picture of GOP is intra */ | |
297 | 521 if (s->picture_in_gop_number % s->gop_size==0 || s->force_type==I_TYPE){ |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
522 s->picture_in_gop_number=0; |
0 | 523 s->pict_type = I_TYPE; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
524 }else |
0 | 525 s->pict_type = P_TYPE; |
526 } else { | |
527 s->pict_type = I_TYPE; | |
528 } | |
529 | |
530 MPV_frame_start(s); | |
220 | 531 |
0 | 532 for(i=0;i<3;i++) { |
533 UINT8 *src = pict->data[i]; | |
534 UINT8 *dest = s->current_picture[i]; | |
535 int src_wrap = pict->linesize[i]; | |
536 int dest_wrap = s->linesize; | |
537 int w = s->width; | |
538 int h = s->height; | |
539 | |
540 if (i >= 1) { | |
541 dest_wrap >>= 1; | |
542 w >>= 1; | |
543 h >>= 1; | |
544 } | |
545 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
546 if(dest_wrap==src_wrap){ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
547 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
|
548 } else { |
220 | 549 for(j=0;j<h;j++) { |
550 memcpy(dest, src, w); | |
551 dest += dest_wrap; | |
552 src += src_wrap; | |
553 } | |
227
ec1bc02a0a47
avoid copying input when encoding non intra stuff too
michaelni
parents:
220
diff
changeset
|
554 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
|
555 } |
0 | 556 } |
557 | |
558 encode_picture(s, s->picture_number); | |
271 | 559 avctx->key_frame = (s->pict_type == I_TYPE); |
286 | 560 avctx->header_bits = s->header_bits; |
561 avctx->mv_bits = s->mv_bits; | |
562 avctx->misc_bits = s->misc_bits; | |
563 avctx->i_tex_bits = s->i_tex_bits; | |
564 avctx->p_tex_bits = s->p_tex_bits; | |
565 avctx->i_count = s->i_count; | |
566 avctx->p_count = s->p_count; | |
567 avctx->skip_count = s->skip_count; | |
568 | |
0 | 569 MPV_frame_end(s); |
570 s->picture_number++; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
571 s->picture_in_gop_number++; |
0 | 572 |
573 if (s->out_format == FMT_MJPEG) | |
574 mjpeg_picture_trailer(s); | |
575 | |
576 flush_put_bits(&s->pb); | |
268 | 577 s->last_frame_bits= s->frame_bits; |
578 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8; | |
579 s->total_bits += s->frame_bits; | |
286 | 580 avctx->frame_bits = s->frame_bits; |
581 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n", | |
582 //s->f_code, avctx->key_frame, s->header_bits, s->mv_bits, s->misc_bits, s->frame_bits, s->i_tex_bits, s->p_tex_bits); | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
583 |
0 | 584 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
|
585 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
|
586 /* 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
|
587 /* 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
|
588 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
|
589 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
|
590 } |
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
|
591 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 592 } |
593 | |
594 static inline int clip(int a, int amin, int amax) | |
595 { | |
596 if (a < amin) | |
597 return amin; | |
598 else if (a > amax) | |
599 return amax; | |
600 else | |
601 return a; | |
602 } | |
603 | |
255 | 604 static inline void gmc1_motion(MpegEncContext *s, |
605 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
606 int dest_offset, | |
607 UINT8 **ref_picture, int src_offset, | |
608 int h) | |
609 { | |
610 UINT8 *ptr; | |
611 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
612 int motion_x, motion_y; | |
613 | |
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
614 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n"); |
255 | 615 motion_x= s->sprite_offset[0][0]; |
616 motion_y= s->sprite_offset[0][1]; | |
617 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
618 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
619 motion_x<<=(3-s->sprite_warping_accuracy); | |
620 motion_y<<=(3-s->sprite_warping_accuracy); | |
621 src_x = clip(src_x, -16, s->width); | |
622 if (src_x == s->width) | |
623 motion_x =0; | |
624 src_y = clip(src_y, -16, s->height); | |
625 if (src_y == s->height) | |
626 motion_y =0; | |
627 | |
628 linesize = s->linesize; | |
629 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
630 | |
631 dest_y+=dest_offset; | |
632 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
633 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
634 | |
635 motion_x= s->sprite_offset[1][0]; | |
636 motion_y= s->sprite_offset[1][1]; | |
637 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
638 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
639 motion_x<<=(3-s->sprite_warping_accuracy); | |
640 motion_y<<=(3-s->sprite_warping_accuracy); | |
641 src_x = clip(src_x, -8, s->width>>1); | |
642 if (src_x == s->width>>1) | |
643 motion_x =0; | |
644 src_y = clip(src_y, -8, s->height>>1); | |
645 if (src_y == s->height>>1) | |
646 motion_y =0; | |
647 | |
648 offset = (src_y * linesize>>1) + src_x + (src_offset>>1); | |
649 ptr = ref_picture[1] + offset; | |
650 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
651 ptr = ref_picture[2] + offset; | |
652 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
653 | |
654 return; | |
655 } | |
656 | |
0 | 657 /* apply one mpeg motion vector to the three components */ |
658 static inline void mpeg_motion(MpegEncContext *s, | |
659 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
660 int dest_offset, | |
661 UINT8 **ref_picture, int src_offset, | |
662 int field_based, op_pixels_func *pix_op, | |
663 int motion_x, int motion_y, int h) | |
664 { | |
665 UINT8 *ptr; | |
666 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
255 | 667 if(s->quarter_sample) |
668 { | |
669 motion_x>>=1; | |
670 motion_y>>=1; | |
671 } | |
0 | 672 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
673 src_x = s->mb_x * 16 + (motion_x >> 1); | |
674 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
675 | |
676 /* WARNING: do no forget half pels */ | |
677 height = s->height >> field_based; | |
678 src_x = clip(src_x, -16, s->width); | |
679 if (src_x == s->width) | |
680 dxy &= ~1; | |
681 src_y = clip(src_y, -16, height); | |
682 if (src_y == height) | |
683 dxy &= ~2; | |
684 linesize = s->linesize << field_based; | |
685 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
686 dest_y += dest_offset; | |
687 pix_op[dxy](dest_y, ptr, linesize, h); | |
688 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
689 | |
690 if (s->out_format == FMT_H263) { | |
691 dxy = 0; | |
692 if ((motion_x & 3) != 0) | |
693 dxy |= 1; | |
694 if ((motion_y & 3) != 0) | |
695 dxy |= 2; | |
696 mx = motion_x >> 2; | |
697 my = motion_y >> 2; | |
698 } else { | |
699 mx = motion_x / 2; | |
700 my = motion_y / 2; | |
701 dxy = ((my & 1) << 1) | (mx & 1); | |
702 mx >>= 1; | |
703 my >>= 1; | |
704 } | |
705 | |
706 src_x = s->mb_x * 8 + mx; | |
707 src_y = s->mb_y * (8 >> field_based) + my; | |
708 src_x = clip(src_x, -8, s->width >> 1); | |
709 if (src_x == (s->width >> 1)) | |
710 dxy &= ~1; | |
711 src_y = clip(src_y, -8, height >> 1); | |
712 if (src_y == (height >> 1)) | |
713 dxy &= ~2; | |
714 | |
715 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
716 ptr = ref_picture[1] + offset; | |
717 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
718 ptr = ref_picture[2] + offset; | |
719 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
720 } | |
721 | |
255 | 722 static inline void qpel_motion(MpegEncContext *s, |
723 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
724 int dest_offset, | |
725 UINT8 **ref_picture, int src_offset, | |
726 int field_based, op_pixels_func *pix_op, | |
727 qpel_mc_func *qpix_op, | |
728 int motion_x, int motion_y, int h) | |
729 { | |
730 UINT8 *ptr; | |
731 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
732 | |
733 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
734 src_x = s->mb_x * 16 + (motion_x >> 2); | |
735 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
736 | |
737 height = s->height >> field_based; | |
738 src_x = clip(src_x, -16, s->width); | |
739 if (src_x == s->width) | |
740 dxy &= ~3; | |
741 src_y = clip(src_y, -16, height); | |
742 if (src_y == height) | |
743 dxy &= ~12; | |
744 linesize = s->linesize << field_based; | |
745 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
746 dest_y += dest_offset; | |
747 //printf("%d %d %d\n", src_x, src_y, dxy); | |
748 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3); | |
749 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3); | |
750 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3); | |
751 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3); | |
752 | |
753 mx= (motion_x>>1) | (motion_x&1); | |
754 my= (motion_y>>1) | (motion_y&1); | |
755 | |
756 dxy = 0; | |
757 if ((mx & 3) != 0) | |
758 dxy |= 1; | |
759 if ((my & 3) != 0) | |
760 dxy |= 2; | |
761 mx = mx >> 2; | |
762 my = my >> 2; | |
763 | |
764 src_x = s->mb_x * 8 + mx; | |
765 src_y = s->mb_y * (8 >> field_based) + my; | |
766 src_x = clip(src_x, -8, s->width >> 1); | |
767 if (src_x == (s->width >> 1)) | |
768 dxy &= ~1; | |
769 src_y = clip(src_y, -8, height >> 1); | |
770 if (src_y == (height >> 1)) | |
771 dxy &= ~2; | |
772 | |
773 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
774 ptr = ref_picture[1] + offset; | |
775 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
776 ptr = ref_picture[2] + offset; | |
777 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
778 } | |
779 | |
780 | |
0 | 781 static inline void MPV_motion(MpegEncContext *s, |
782 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
783 int dir, UINT8 **ref_picture, | |
255 | 784 op_pixels_func *pix_op, qpel_mc_func *qpix_op) |
0 | 785 { |
786 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
787 int mb_x, mb_y, i; | |
788 UINT8 *ptr, *dest; | |
789 | |
790 mb_x = s->mb_x; | |
791 mb_y = s->mb_y; | |
792 | |
793 switch(s->mv_type) { | |
794 case MV_TYPE_16X16: | |
255 | 795 if(s->mcsel){ |
796 #if 0 | |
797 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
798 ref_picture, 0, | |
799 0, pix_op, | |
800 s->sprite_offset[0][0]>>3, | |
801 s->sprite_offset[0][1]>>3, | |
802 16); | |
803 #else | |
804 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
805 ref_picture, 0, | |
806 16); | |
807 #endif | |
262 | 808 }else if(s->quarter_sample && dir==0){ //FIXME |
255 | 809 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, |
810 ref_picture, 0, | |
811 0, pix_op, qpix_op, | |
812 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
813 }else{ | |
814 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
815 ref_picture, 0, | |
816 0, pix_op, | |
817 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
818 } | |
0 | 819 break; |
820 case MV_TYPE_8X8: | |
821 for(i=0;i<4;i++) { | |
822 motion_x = s->mv[dir][i][0]; | |
823 motion_y = s->mv[dir][i][1]; | |
824 | |
825 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
826 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
295 | 827 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; |
0 | 828 |
829 /* WARNING: do no forget half pels */ | |
830 src_x = clip(src_x, -16, s->width); | |
831 if (src_x == s->width) | |
832 dxy &= ~1; | |
833 src_y = clip(src_y, -16, s->height); | |
834 if (src_y == s->height) | |
835 dxy &= ~2; | |
836 | |
837 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
838 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
839 pix_op[dxy](dest, ptr, s->linesize, 8); | |
840 } | |
841 /* In case of 8X8, we construct a single chroma motion vector | |
842 with a special rounding */ | |
843 mx = 0; | |
844 my = 0; | |
845 for(i=0;i<4;i++) { | |
846 mx += s->mv[dir][i][0]; | |
847 my += s->mv[dir][i][1]; | |
848 } | |
849 if (mx >= 0) | |
850 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
851 else { | |
852 mx = -mx; | |
853 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
854 } | |
855 if (my >= 0) | |
856 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
857 else { | |
858 my = -my; | |
859 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
860 } | |
861 dxy = ((my & 1) << 1) | (mx & 1); | |
862 mx >>= 1; | |
863 my >>= 1; | |
864 | |
865 src_x = mb_x * 8 + mx; | |
866 src_y = mb_y * 8 + my; | |
867 src_x = clip(src_x, -8, s->width/2); | |
868 if (src_x == s->width/2) | |
869 dxy &= ~1; | |
870 src_y = clip(src_y, -8, s->height/2); | |
871 if (src_y == s->height/2) | |
872 dxy &= ~2; | |
873 | |
874 offset = (src_y * (s->linesize >> 1)) + src_x; | |
875 ptr = ref_picture[1] + offset; | |
876 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
877 ptr = ref_picture[2] + offset; | |
878 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
879 break; | |
880 case MV_TYPE_FIELD: | |
881 if (s->picture_structure == PICT_FRAME) { | |
882 /* top field */ | |
883 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
884 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
885 1, pix_op, | |
886 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
887 /* bottom field */ | |
888 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
889 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
890 1, pix_op, | |
891 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
892 } else { | |
893 | |
894 | |
895 } | |
896 break; | |
897 } | |
898 } | |
899 | |
900 | |
901 /* put block[] to dest[] */ | |
902 static inline void put_dct(MpegEncContext *s, | |
903 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
904 { | |
905 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
|
906 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
|
907 ff_idct (block); |
0 | 908 put_pixels_clamped(block, dest, line_size); |
909 } | |
910 | |
911 /* add block[] to dest[] */ | |
912 static inline void add_dct(MpegEncContext *s, | |
913 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
914 { | |
915 if (s->block_last_index[i] >= 0) { | |
916 if (!s->mpeg2) | |
206 | 917 if(s->encoding || (!s->h263_msmpeg4)) |
200 | 918 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
|
919 ff_idct (block); |
0 | 920 add_pixels_clamped(block, dest, line_size); |
921 } | |
922 } | |
923 | |
924 /* generic function called after a macroblock has been parsed by the | |
925 decoder or after it has been encoded by the encoder. | |
926 | |
927 Important variables used: | |
928 s->mb_intra : true if intra macroblock | |
929 s->mv_dir : motion vector direction | |
930 s->mv_type : motion vector type | |
931 s->mv : motion vector | |
932 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
933 */ | |
934 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
935 { | |
244 | 936 int mb_x, mb_y; |
0 | 937 int dct_linesize, dct_offset; |
938 op_pixels_func *op_pix; | |
255 | 939 qpel_mc_func *op_qpix; |
0 | 940 |
941 mb_x = s->mb_x; | |
942 mb_y = s->mb_y; | |
943 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
944 #ifdef FF_POSTPROCESS |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
945 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
946 //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
|
947 #endif |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
948 |
0 | 949 /* update DC predictors for P macroblocks */ |
950 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
951 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
|
952 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
|
953 { |
244 | 954 int wrap, xy, v; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
955 s->mbintra_table[mb_x + mb_y*s->mb_width]=0; |
0 | 956 wrap = 2 * s->mb_width + 2; |
244 | 957 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 958 v = 1024; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
959 |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
960 s->dc_val[0][xy] = v; |
244 | 961 s->dc_val[0][xy + 1] = v; |
962 s->dc_val[0][xy + wrap] = v; | |
963 s->dc_val[0][xy + 1 + wrap] = v; | |
0 | 964 /* ac pred */ |
244 | 965 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16)); |
966 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16)); | |
967 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16)); | |
968 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16)); | |
0 | 969 if (s->h263_msmpeg4) { |
244 | 970 s->coded_block[xy] = 0; |
971 s->coded_block[xy + 1] = 0; | |
972 s->coded_block[xy + wrap] = 0; | |
973 s->coded_block[xy + 1 + wrap] = 0; | |
0 | 974 } |
975 /* chroma */ | |
976 wrap = s->mb_width + 2; | |
244 | 977 xy = mb_x + 1 + (mb_y + 1) * wrap; |
978 s->dc_val[1][xy] = v; | |
979 s->dc_val[2][xy] = v; | |
0 | 980 /* ac pred */ |
244 | 981 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); |
982 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
|
983 } |
0 | 984 } else { |
985 s->last_dc[0] = 128 << s->intra_dc_precision; | |
986 s->last_dc[1] = 128 << s->intra_dc_precision; | |
987 s->last_dc[2] = 128 << s->intra_dc_precision; | |
988 } | |
989 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
990 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
|
991 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
|
992 |
280 | 993 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */ |
994 if (s->out_format == FMT_H263) { | |
262 | 995 if(s->pict_type!=B_TYPE){ |
244 | 996 int xy, wrap, motion_x, motion_y; |
0 | 997 |
998 wrap = 2 * s->mb_width + 2; | |
244 | 999 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 1000 if (s->mb_intra) { |
1001 motion_x = 0; | |
1002 motion_y = 0; | |
1003 goto motion_init; | |
1004 } else if (s->mv_type == MV_TYPE_16X16) { | |
1005 motion_x = s->mv[0][0][0]; | |
1006 motion_y = s->mv[0][0][1]; | |
1007 motion_init: | |
1008 /* no update if 8X8 because it has been done during parsing */ | |
244 | 1009 s->motion_val[xy][0] = motion_x; |
1010 s->motion_val[xy][1] = motion_y; | |
1011 s->motion_val[xy + 1][0] = motion_x; | |
1012 s->motion_val[xy + 1][1] = motion_y; | |
1013 s->motion_val[xy + wrap][0] = motion_x; | |
1014 s->motion_val[xy + wrap][1] = motion_y; | |
1015 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
1016 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
0 | 1017 } |
262 | 1018 } |
0 | 1019 } |
1020 | |
1021 if (!s->intra_only) { | |
1022 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
|
1023 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1024 |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1025 /* 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
|
1026 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
|
1027 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
|
1028 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1029 s->mb_skiped = 0; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1030 /* 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
|
1031 if (*mbskip_ptr != 0) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1032 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1033 *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
|
1034 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1035 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1036 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1037 } |
0 | 1038 |
1039 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
1040 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1041 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
1042 | |
1043 if (s->interlaced_dct) { | |
1044 dct_linesize = s->linesize * 2; | |
1045 dct_offset = s->linesize; | |
1046 } else { | |
1047 dct_linesize = s->linesize; | |
1048 dct_offset = s->linesize * 8; | |
1049 } | |
1050 | |
1051 if (!s->mb_intra) { | |
1052 /* motion handling */ | |
255 | 1053 if (!s->no_rounding){ |
0 | 1054 op_pix = put_pixels_tab; |
255 | 1055 op_qpix= qpel_mc_rnd_tab; |
1056 }else{ | |
0 | 1057 op_pix = put_no_rnd_pixels_tab; |
255 | 1058 op_qpix= qpel_mc_no_rnd_tab; |
1059 } | |
0 | 1060 |
1061 if (s->mv_dir & MV_DIR_FORWARD) { | |
255 | 1062 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); |
0 | 1063 if (!s->no_rounding) |
1064 op_pix = avg_pixels_tab; | |
1065 else | |
1066 op_pix = avg_no_rnd_pixels_tab; | |
1067 } | |
1068 if (s->mv_dir & MV_DIR_BACKWARD) { | |
255 | 1069 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); |
0 | 1070 } |
1071 | |
1072 /* add dct residue */ | |
1073 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
1074 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1075 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1076 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1077 | |
57 | 1078 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1079 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1080 } else { |
1081 /* dct only in intra block */ | |
1082 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
1083 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1084 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1085 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1086 | |
57 | 1087 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1088 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1089 } |
1090 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1091 the_end: |
294 | 1092 emms_c(); //FIXME remove |
1093 } | |
1094 | |
1095 static void encode_mb(MpegEncContext *s) | |
1096 { | |
1097 int wrap; | |
1098 const int mb_x= s->mb_x; | |
1099 const int mb_y= s->mb_y; | |
1100 UINT8 *ptr; | |
1101 const int motion_x= s->mv[0][0][0]; | |
1102 const int motion_y= s->mv[0][0][1]; | |
1103 int i; | |
1104 | |
1105 /* get the pixels */ | |
1106 wrap = s->linesize; | |
1107 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
1108 get_pixels(s->block[0], ptr, wrap); | |
1109 get_pixels(s->block[1], ptr + 8, wrap); | |
1110 get_pixels(s->block[2], ptr + 8 * wrap, wrap); | |
1111 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); | |
1112 wrap = s->linesize >> 1; | |
1113 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
1114 get_pixels(s->block[4], ptr, wrap); | |
1115 | |
1116 wrap = s->linesize >> 1; | |
1117 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; | |
1118 get_pixels(s->block[5], ptr, wrap); | |
1119 | |
1120 /* subtract previous frame if non intra */ | |
1121 if (!s->mb_intra) { | |
1122 int dxy, offset, mx, my; | |
295 | 1123 |
1124 if(s->mv_type==MV_TYPE_16X16){ | |
1125 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
1126 ptr = s->last_picture[0] + | |
1127 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) + | |
1128 (mb_x * 16 + (motion_x >> 1)); | |
294 | 1129 |
295 | 1130 sub_pixels_2(s->block[0], ptr, s->linesize, dxy); |
1131 sub_pixels_2(s->block[1], ptr + 8, s->linesize, dxy); | |
1132 sub_pixels_2(s->block[2], ptr + s->linesize * 8, s->linesize, dxy); | |
1133 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy); | |
294 | 1134 |
295 | 1135 if (s->out_format == FMT_H263) { |
1136 /* special rounding for h263 */ | |
1137 dxy = 0; | |
1138 if ((motion_x & 3) != 0) | |
1139 dxy |= 1; | |
1140 if ((motion_y & 3) != 0) | |
1141 dxy |= 2; | |
1142 mx = motion_x >> 2; | |
1143 my = motion_y >> 2; | |
1144 } else { | |
1145 mx = motion_x / 2; | |
1146 my = motion_y / 2; | |
1147 dxy = ((my & 1) << 1) | (mx & 1); | |
1148 mx >>= 1; | |
1149 my >>= 1; | |
1150 } | |
1151 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx); | |
1152 ptr = s->last_picture[1] + offset; | |
1153 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); | |
1154 ptr = s->last_picture[2] + offset; | |
1155 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); | |
1156 }else{ | |
1157 int src_x, src_y; | |
1158 | |
1159 for(i=0;i<4;i++) { | |
1160 int motion_x = s->mv[0][i][0]; | |
1161 int motion_y = s->mv[0][i][1]; | |
1162 | |
1163 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
1164 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
1165 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; | |
1166 | |
1167 ptr = s->last_picture[0] + (src_y * s->linesize) + (src_x); | |
1168 sub_pixels_2(s->block[i], ptr, s->linesize, dxy); | |
1169 } | |
1170 /* In case of 8X8, we construct a single chroma motion vector | |
1171 with a special rounding */ | |
1172 mx = 0; | |
1173 my = 0; | |
1174 for(i=0;i<4;i++) { | |
1175 mx += s->mv[0][i][0]; | |
1176 my += s->mv[0][i][1]; | |
1177 } | |
1178 if (mx >= 0) | |
1179 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1180 else { | |
1181 mx = -mx; | |
1182 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1183 } | |
1184 if (my >= 0) | |
1185 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1186 else { | |
1187 my = -my; | |
1188 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1189 } | |
294 | 1190 dxy = ((my & 1) << 1) | (mx & 1); |
1191 mx >>= 1; | |
1192 my >>= 1; | |
295 | 1193 |
1194 src_x = mb_x * 8 + mx; | |
1195 src_y = mb_y * 8 + my; | |
1196 src_x = clip(src_x, -8, s->width/2); | |
1197 if (src_x == s->width/2) | |
1198 dxy &= ~1; | |
1199 src_y = clip(src_y, -8, s->height/2); | |
1200 if (src_y == s->height/2) | |
1201 dxy &= ~2; | |
1202 | |
1203 offset = (src_y * (s->linesize >> 1)) + src_x; | |
1204 ptr = s->last_picture[1] + offset; | |
1205 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); | |
1206 ptr = s->last_picture[2] + offset; | |
1207 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); | |
294 | 1208 } |
1209 } | |
1210 | |
1211 #if 0 | |
1212 { | |
1213 float adap_parm; | |
1214 | |
1215 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / | |
1216 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); | |
1217 | |
1218 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", | |
1219 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', | |
1220 s->qscale, adap_parm, s->qscale*adap_parm, | |
1221 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); | |
1222 } | |
1223 #endif | |
1224 /* DCT & quantize */ | |
1225 if (s->h263_msmpeg4) { | |
1226 msmpeg4_dc_scale(s); | |
1227 } else if (s->h263_pred) { | |
1228 h263_dc_scale(s); | |
1229 } else { | |
1230 /* default quantization values */ | |
1231 s->y_dc_scale = 8; | |
1232 s->c_dc_scale = 8; | |
1233 } | |
1234 for(i=0;i<6;i++) { | |
1235 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); | |
1236 } | |
1237 | |
1238 /* huffman encode */ | |
1239 switch(s->out_format) { | |
1240 case FMT_MPEG1: | |
1241 mpeg1_encode_mb(s, s->block, motion_x, motion_y); | |
1242 break; | |
1243 case FMT_H263: | |
1244 if (s->h263_msmpeg4) | |
1245 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1246 else if(s->h263_pred) | |
1247 mpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1248 else | |
1249 h263_encode_mb(s, s->block, motion_x, motion_y); | |
1250 break; | |
1251 case FMT_MJPEG: | |
1252 mjpeg_encode_mb(s, s->block); | |
1253 break; | |
1254 } | |
1255 } | |
1256 | |
1257 static void copy_bits(PutBitContext *pb, UINT8 *src, int length) | |
1258 { | |
1259 int bytes= length>>3; | |
1260 int bits= length&7; | |
1261 int i; | |
1262 | |
1263 for(i=0; i<bytes; i++) put_bits(pb, 8, src[i]); | |
1264 put_bits(pb, bits, src[i]>>(8-bits)); | |
0 | 1265 } |
1266 | |
1267 static void encode_picture(MpegEncContext *s, int picture_number) | |
1268 { | |
294 | 1269 int mb_x, mb_y, last_gob, pdif = 0; |
1270 int i; | |
286 | 1271 int bits; |
294 | 1272 MpegEncContext best_s; |
1273 UINT8 bit_buf[4][3000]; //FIXME check that this is ALLWAYS large enogh for a MB | |
0 | 1274 |
1275 s->picture_number = picture_number; | |
268 | 1276 |
294 | 1277 s->block_wrap[0]= |
1278 s->block_wrap[1]= | |
1279 s->block_wrap[2]= | |
1280 s->block_wrap[3]= s->mb_width*2 + 2; | |
1281 s->block_wrap[4]= | |
1282 s->block_wrap[5]= s->mb_width + 2; | |
1283 | |
268 | 1284 s->last_mc_mb_var = s->mc_mb_var; |
1285 /* Reset the average MB variance */ | |
1286 s->avg_mb_var = 0; | |
1287 s->mc_mb_var = 0; | |
1288 /* Estimate motion for every MB */ | |
294 | 1289 if(s->pict_type == P_TYPE){ |
1290 for(mb_y=0; mb_y < s->mb_height; mb_y++) { | |
1291 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; | |
1292 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1293 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1294 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1295 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
1296 s->mb_x = mb_x; | |
1297 s->mb_y = mb_y; | |
1298 s->block_index[0]+=2; | |
1299 s->block_index[1]+=2; | |
1300 s->block_index[2]+=2; | |
1301 s->block_index[3]+=2; | |
268 | 1302 |
294 | 1303 /* compute motion vector & mb_type and store in context */ |
1304 estimate_motion(s, mb_x, mb_y); | |
1305 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER; | |
268 | 1306 } |
1307 } | |
294 | 1308 emms_c(); |
1309 }else{ | |
1310 /* I-Frame */ | |
1311 //FIXME do we need to zero them? | |
1312 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2); | |
1313 memset(s->mv_table[0] , 0, sizeof(INT16)*s->mb_width*s->mb_height); | |
1314 memset(s->mv_table[1] , 0, sizeof(INT16)*s->mb_width*s->mb_height); | |
1315 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); | |
268 | 1316 } |
1317 | |
297 | 1318 if(s->avg_mb_var < s->mc_mb_var && s->pict_type != B_TYPE && (!s->force_type)){ //FIXME subtract MV bits |
271 | 1319 s->pict_type= I_TYPE; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1320 s->picture_in_gop_number=0; |
294 | 1321 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
1322 //printf("Scene change detected, encoding as I Frame\n"); | |
271 | 1323 } |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1324 |
287 | 1325 /* find best f_code for ME which do unlimited searches */ |
1326 if(s->pict_type==P_TYPE && s->full_search>3){ | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1327 int mv_num[8]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1328 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1329 int loose=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1330 UINT8 * fcode_tab= s->fcode_tab; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1331 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1332 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
|
1333 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1334 for(i=0; i<s->mb_num; i++){ |
295 | 1335 if(s->mb_type[i] & MB_TYPE_INTER){ |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1336 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
|
1337 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
|
1338 //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
|
1339 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1340 //else printf("I"); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1341 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1342 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1343 for(i=MAX_FCODE; i>1; i--){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1344 loose+= mv_num[i]; |
286 | 1345 if(loose > 10) break; //FIXME this is pretty ineffective |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1346 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1347 s->f_code= i; |
287 | 1348 /* for(i=0; i<=MAX_FCODE; i++){ |
1349 printf("%d ", mv_num[i]); | |
1350 } | |
1351 printf("\n");*/ | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1352 }else{ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1353 s->f_code= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1354 } |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
1355 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1356 //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
|
1357 /* 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
|
1358 if(s->pict_type==P_TYPE){ |
295 | 1359 int i, x, y; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1360 const int f_code= s->f_code; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1361 UINT8 * fcode_tab= s->fcode_tab; |
295 | 1362 //FIXME try to clip instead of intra izing ;) |
1363 /* clip / convert to intra 16x16 type MVs */ | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1364 for(i=0; i<s->mb_num; i++){ |
294 | 1365 if(s->mb_type[i]&MB_TYPE_INTER){ |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1366 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
|
1367 || 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
|
1368 || 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
|
1369 || fcode_tab[s->mv_table[1][i] + MAX_MV] == 0 ){ |
294 | 1370 s->mb_type[i] &= ~MB_TYPE_INTER; |
1371 s->mb_type[i] |= MB_TYPE_INTRA; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1372 s->mv_table[0][i] = 0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1373 s->mv_table[1][i] = 0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1374 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1375 } |
295 | 1376 } |
1377 | |
1378 if(s->flags&CODEC_FLAG_4MV){ | |
1379 int wrap= 2+ s->mb_width*2; | |
1380 | |
1381 /* clip / convert to intra 8x8 type MVs */ | |
1382 for(y=0; y<s->mb_height; y++){ | |
1383 int xy= (y*2 + 1)*wrap + 1; | |
1384 i= y*s->mb_width; | |
1385 | |
1386 for(x=0; x<s->mb_width; x++){ | |
1387 if(s->mb_type[i]&MB_TYPE_INTER4V){ | |
1388 int block; | |
1389 for(block=0; block<4; block++){ | |
1390 int off= (block& 1) + (block>>1)*wrap; | |
1391 int mx= s->motion_val[ xy + off ][0]; | |
1392 int my= s->motion_val[ xy + off ][1]; | |
1393 | |
1394 if( fcode_tab[mx + MAX_MV] > f_code | |
1395 || fcode_tab[mx + MAX_MV] == 0 | |
1396 || fcode_tab[my + MAX_MV] > f_code | |
1397 || fcode_tab[my + MAX_MV] == 0 ){ | |
1398 s->mb_type[i] &= ~MB_TYPE_INTER4V; | |
1399 s->mb_type[i] |= MB_TYPE_INTRA; | |
1400 } | |
1401 } | |
1402 xy+=2; | |
1403 i++; | |
1404 } | |
1405 } | |
294 | 1406 } |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1407 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1408 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
1409 |
271 | 1410 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var); |
1411 | |
0 | 1412 if (!s->fixed_qscale) |
1413 s->qscale = rate_estimate_qscale(s); | |
1414 | |
1415 /* precompute matrix */ | |
1416 if (s->out_format == FMT_MJPEG) { | |
1417 /* for mjpeg, we do include qscale in the matrix */ | |
1418 s->intra_matrix[0] = default_intra_matrix[0]; | |
1419 for(i=1;i<64;i++) | |
1420 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
220 | 1421 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
0 | 1422 } else { |
220 | 1423 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
1424 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); | |
0 | 1425 } |
1426 | |
286 | 1427 s->last_bits= get_bit_count(&s->pb); |
0 | 1428 switch(s->out_format) { |
1429 case FMT_MJPEG: | |
1430 mjpeg_picture_header(s); | |
1431 break; | |
1432 case FMT_H263: | |
1433 if (s->h263_msmpeg4) | |
1434 msmpeg4_encode_picture_header(s, picture_number); | |
1435 else if (s->h263_pred) | |
1436 mpeg4_encode_picture_header(s, picture_number); | |
1437 else if (s->h263_rv10) | |
1438 rv10_encode_picture_header(s, picture_number); | |
1439 else | |
1440 h263_encode_picture_header(s, picture_number); | |
1441 break; | |
1442 case FMT_MPEG1: | |
1443 mpeg1_encode_picture_header(s, picture_number); | |
1444 break; | |
1445 } | |
286 | 1446 bits= get_bit_count(&s->pb); |
1447 s->header_bits= bits - s->last_bits; | |
1448 s->last_bits= bits; | |
1449 s->mv_bits=0; | |
1450 s->misc_bits=0; | |
1451 s->i_tex_bits=0; | |
1452 s->p_tex_bits=0; | |
1453 s->i_count=0; | |
1454 s->p_count=0; | |
1455 s->skip_count=0; | |
1456 | |
0 | 1457 /* init last dc values */ |
1458 /* note: quant matrix value (8) is implied here */ | |
1459 s->last_dc[0] = 128; | |
1460 s->last_dc[1] = 128; | |
1461 s->last_dc[2] = 128; | |
1462 s->mb_incr = 1; | |
1463 s->last_mv[0][0][0] = 0; | |
1464 s->last_mv[0][0][1] = 0; | |
1465 | |
162 | 1466 /* Get the GOB height based on picture height */ |
231 | 1467 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
162 | 1468 if (s->height <= 400) |
1469 s->gob_index = 1; | |
1470 else if (s->height <= 800) | |
1471 s->gob_index = 2; | |
1472 else | |
1473 s->gob_index = 4; | |
1474 } | |
268 | 1475 |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1476 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
|
1477 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1478 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
|
1479 /* 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
|
1480 /* 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
|
1481 if (s->rtp_mode) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1482 if (!mb_y) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1483 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
|
1484 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
|
1485 } 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
|
1486 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
|
1487 if (last_gob) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1488 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
|
1489 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1490 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1491 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1492 |
266 | 1493 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; |
1494 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1495 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1496 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1497 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
1498 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
|
1499 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
294 | 1500 const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x]; |
1501 PutBitContext pb; | |
1502 int d; | |
1503 int dmin=10000000; | |
1504 int best=0; | |
0 | 1505 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1506 s->mb_x = mb_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1507 s->mb_y = mb_y; |
266 | 1508 s->block_index[0]+=2; |
1509 s->block_index[1]+=2; | |
1510 s->block_index[2]+=2; | |
1511 s->block_index[3]+=2; | |
1512 s->block_index[4]++; | |
1513 s->block_index[5]++; | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1514 |
294 | 1515 s->mv_dir = MV_DIR_FORWARD; |
1516 if(mb_type & (mb_type-1)){ // more than 1 MB type possible | |
1517 pb= s->pb; | |
1518 if(mb_type&MB_TYPE_INTER){ | |
295 | 1519 s->mv_type = MV_TYPE_16X16; |
294 | 1520 s->mb_intra= 0; |
1521 s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x]; | |
1522 s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x]; | |
1523 init_put_bits(&s->pb, bit_buf[1], 3000, NULL, NULL); | |
1524 s->block= s->inter_block; | |
0 | 1525 |
294 | 1526 encode_mb(s); |
1527 d= get_bit_count(&s->pb); | |
1528 if(d<dmin){ | |
1529 flush_put_bits(&s->pb); | |
1530 dmin=d; | |
1531 best_s.mv[0][0][0]= s->mv[0][0][0]; | |
1532 best_s.mv[0][0][1]= s->mv[0][0][1]; | |
1533 best_s.mb_intra= 0; | |
295 | 1534 best_s.mv_type = MV_TYPE_16X16; |
294 | 1535 best_s.pb=s->pb; |
1536 best_s.block= s->block; | |
1537 best=1; | |
1538 for(i=0; i<6; i++) | |
1539 best_s.block_last_index[i]= s->block_last_index[i]; | |
1540 } | |
0 | 1541 } |
295 | 1542 if(mb_type&MB_TYPE_INTER4V){ |
1543 s->mv_type = MV_TYPE_8X8; | |
1544 s->mb_intra= 0; | |
1545 for(i=0; i<4; i++){ | |
1546 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
1547 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
1548 } | |
1549 init_put_bits(&s->pb, bit_buf[2], 3000, NULL, NULL); | |
1550 s->block= s->inter4v_block; | |
1551 | |
1552 encode_mb(s); | |
1553 d= get_bit_count(&s->pb); | |
1554 if(d<dmin){ | |
1555 flush_put_bits(&s->pb); | |
1556 dmin=d; | |
1557 for(i=0; i<4; i++){ | |
1558 best_s.mv[0][i][0] = s->mv[0][i][0]; | |
1559 best_s.mv[0][i][1] = s->mv[0][i][1]; | |
1560 } | |
1561 best_s.mb_intra= 0; | |
1562 best_s.mv_type = MV_TYPE_8X8; | |
1563 best_s.pb=s->pb; | |
1564 best_s.block= s->block; | |
1565 best=2; | |
1566 for(i=0; i<6; i++) | |
1567 best_s.block_last_index[i]= s->block_last_index[i]; | |
1568 } | |
1569 } | |
294 | 1570 if(mb_type&MB_TYPE_INTRA){ |
295 | 1571 s->mv_type = MV_TYPE_16X16; |
294 | 1572 s->mb_intra= 1; |
1573 s->mv[0][0][0] = 0; | |
1574 s->mv[0][0][1] = 0; | |
1575 init_put_bits(&s->pb, bit_buf[0], 3000, NULL, NULL); | |
1576 s->block= s->intra_block; | |
1577 | |
1578 encode_mb(s); | |
1579 d= get_bit_count(&s->pb); | |
1580 if(d<dmin){ | |
1581 flush_put_bits(&s->pb); | |
1582 dmin=d; | |
1583 best_s.mv[0][0][0]= 0; | |
1584 best_s.mv[0][0][1]= 0; | |
1585 best_s.mb_intra= 1; | |
295 | 1586 best_s.mv_type = MV_TYPE_16X16; |
294 | 1587 best_s.pb=s->pb; |
1588 best_s.block= s->block; | |
1589 for(i=0; i<6; i++) | |
1590 best_s.block_last_index[i]= s->block_last_index[i]; | |
1591 best=0; | |
1592 } | |
1593 /* force cleaning of ac/dc if needed ... */ | |
1594 s->mbintra_table[mb_x + mb_y*s->mb_width]=1; | |
1595 } | |
295 | 1596 for(i=0; i<4; i++){ |
1597 s->mv[0][i][0] = best_s.mv[0][i][0]; | |
1598 s->mv[0][i][1] = best_s.mv[0][i][1]; | |
1599 } | |
294 | 1600 s->mb_intra= best_s.mb_intra; |
295 | 1601 s->mv_type= best_s.mv_type; |
294 | 1602 for(i=0; i<6; i++) |
1603 s->block_last_index[i]= best_s.block_last_index[i]; | |
1604 copy_bits(&pb, bit_buf[best], dmin); | |
1605 s->block= best_s.block; | |
1606 s->pb= pb; | |
1607 } else { | |
1608 // only one MB-Type possible | |
1609 if(mb_type&MB_TYPE_INTRA){ | |
1610 s->mb_intra= 1; | |
1611 s->mv[0][0][0] = 0; | |
1612 s->mv[0][0][1] = 0; | |
1613 }else{ | |
1614 s->mb_intra= 0; | |
1615 s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x]; | |
1616 s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x]; | |
1617 } | |
1618 encode_mb(s); | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1619 } |
0 | 1620 |
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
|
1621 MPV_decode_mb(s, s->block); |
0 | 1622 } |
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
|
1623 |
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
|
1624 |
231 | 1625 /* Obtain average GOB size for RTP */ |
1626 if (s->rtp_mode) { | |
1627 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
|
1628 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
231 | 1629 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
|
1630 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
|
1631 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
231 | 1632 } |
1633 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
1634 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
1635 s->first_gob_line = 0; | |
1636 } | |
0 | 1637 } |
294 | 1638 emms_c(); |
286 | 1639 |
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
|
1640 if (s->h263_msmpeg4 && s->pict_type == I_TYPE) |
208 | 1641 msmpeg4_encode_ext_header(s); |
1642 | |
162 | 1643 //if (s->gob_number) |
1644 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
231 | 1645 |
1646 /* Send the last GOB if RTP */ | |
1647 if (s->rtp_mode) { | |
1648 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
|
1649 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 1650 /* Call the RTP callback to send the last GOB */ |
1651 if (s->rtp_callback) | |
1652 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
|
1653 s->ptr_lastgob = pbBufPtr(&s->pb); |
231 | 1654 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
1655 } | |
0 | 1656 } |
1657 | |
220 | 1658 static int dct_quantize_c(MpegEncContext *s, |
0 | 1659 DCTELEM *block, int n, |
1660 int qscale) | |
1661 { | |
1662 int i, j, level, last_non_zero, q; | |
1663 const int *qmat; | |
216 | 1664 int minLevel, maxLevel; |
1665 | |
1666 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ | |
1667 /* mpeg4 */ | |
1668 minLevel= -2048; | |
1669 maxLevel= 2047; | |
1670 }else if(s->out_format==FMT_MPEG1){ | |
1671 /* mpeg1 */ | |
1672 minLevel= -255; | |
1673 maxLevel= 255; | |
230 | 1674 }else if(s->out_format==FMT_MJPEG){ |
1675 /* (m)jpeg */ | |
1676 minLevel= -1023; | |
1677 maxLevel= 1023; | |
216 | 1678 }else{ |
1679 /* h263 / msmpeg4 */ | |
1680 minLevel= -128; | |
1681 maxLevel= 127; | |
1682 } | |
0 | 1683 |
1684 av_fdct (block); | |
1685 | |
64 | 1686 /* we need this permutation so that we correct the IDCT |
1687 permutation. will be moved into DCT code */ | |
1688 block_permute(block); | |
1689 | |
0 | 1690 if (s->mb_intra) { |
1691 if (n < 4) | |
1692 q = s->y_dc_scale; | |
1693 else | |
1694 q = s->c_dc_scale; | |
1695 q = q << 3; | |
1696 | |
1697 /* note: block[0] is assumed to be positive */ | |
1698 block[0] = (block[0] + (q >> 1)) / q; | |
1699 i = 1; | |
1700 last_non_zero = 0; | |
1701 if (s->out_format == FMT_H263) { | |
1702 qmat = s->q_non_intra_matrix; | |
1703 } else { | |
1704 qmat = s->q_intra_matrix; | |
1705 } | |
1706 } else { | |
1707 i = 0; | |
1708 last_non_zero = -1; | |
1709 qmat = s->q_non_intra_matrix; | |
1710 } | |
1711 | |
1712 for(;i<64;i++) { | |
1713 j = zigzag_direct[i]; | |
1714 level = block[j]; | |
1715 level = level * qmat[j]; | |
1716 #ifdef PARANOID | |
1717 { | |
1718 static int count = 0; | |
1719 int level1, level2, qmat1; | |
1720 double val; | |
1721 if (qmat == s->q_non_intra_matrix) { | |
1722 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1723 } else { | |
1724 qmat1 = default_intra_matrix[j] * s->qscale; | |
1725 } | |
1726 if (av_fdct != jpeg_fdct_ifast) | |
1727 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1728 else | |
1729 val = ((double)block[j] * 8.0 * 2048.0) / | |
1730 ((double)qmat1 * aanscales[j]); | |
1731 level1 = (int)val; | |
1732 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1733 if (level1 != level2) { | |
1734 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1735 count, level2, level1, block[j], qmat1, qmat[j], | |
1736 val); | |
1737 count++; | |
1738 } | |
1739 | |
1740 } | |
1741 #endif | |
1742 /* XXX: slight error for the low range. Test should be equivalent to | |
1743 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1744 (QMAT_SHIFT - 3))) | |
1745 */ | |
1746 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1747 level) { | |
1748 level = level / (1 << (QMAT_SHIFT - 3)); | |
1749 /* XXX: currently, this code is not optimal. the range should be: | |
1750 mpeg1: -255..255 | |
1751 mpeg2: -2048..2047 | |
1752 h263: -128..127 | |
1753 mpeg4: -2048..2047 | |
1754 */ | |
216 | 1755 if (level > maxLevel) |
1756 level = maxLevel; | |
1757 else if (level < minLevel) | |
1758 level = minLevel; | |
0 | 1759 |
1760 block[j] = level; | |
1761 last_non_zero = i; | |
1762 } else { | |
1763 block[j] = 0; | |
1764 } | |
1765 } | |
1766 return last_non_zero; | |
1767 } | |
1768 | |
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
|
1769 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
|
1770 DCTELEM *block, int n, int qscale) |
0 | 1771 { |
200 | 1772 int i, level, nCoeffs; |
0 | 1773 const UINT16 *quant_matrix; |
1774 | |
200 | 1775 if(s->alternate_scan) nCoeffs= 64; |
1776 else nCoeffs= s->block_last_index[n]+1; | |
1777 | |
0 | 1778 if (s->mb_intra) { |
1779 if (n < 4) | |
1780 block[0] = block[0] * s->y_dc_scale; | |
1781 else | |
1782 block[0] = block[0] * s->c_dc_scale; | |
1783 /* XXX: only mpeg1 */ | |
1784 quant_matrix = s->intra_matrix; | |
200 | 1785 for(i=1;i<nCoeffs;i++) { |
1786 int j= zigzag_direct[i]; | |
1787 level = block[j]; | |
0 | 1788 if (level) { |
1789 if (level < 0) { | |
1790 level = -level; | |
200 | 1791 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1792 level = (level - 1) | 1; |
1793 level = -level; | |
1794 } else { | |
200 | 1795 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1796 level = (level - 1) | 1; |
1797 } | |
1798 #ifdef PARANOID | |
1799 if (level < -2048 || level > 2047) | |
1800 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1801 #endif | |
200 | 1802 block[j] = level; |
0 | 1803 } |
1804 } | |
1805 } else { | |
1806 i = 0; | |
1807 quant_matrix = s->non_intra_matrix; | |
217 | 1808 for(;i<nCoeffs;i++) { |
200 | 1809 int j= zigzag_direct[i]; |
1810 level = block[j]; | |
0 | 1811 if (level) { |
1812 if (level < 0) { | |
1813 level = -level; | |
1814 level = (((level << 1) + 1) * qscale * | |
200 | 1815 ((int) (quant_matrix[j]))) >> 4; |
0 | 1816 level = (level - 1) | 1; |
1817 level = -level; | |
1818 } else { | |
1819 level = (((level << 1) + 1) * qscale * | |
200 | 1820 ((int) (quant_matrix[j]))) >> 4; |
0 | 1821 level = (level - 1) | 1; |
1822 } | |
1823 #ifdef PARANOID | |
1824 if (level < -2048 || level > 2047) | |
1825 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1826 #endif | |
200 | 1827 block[j] = level; |
0 | 1828 } |
1829 } | |
1830 } | |
1831 } | |
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
|
1832 |
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
|
1833 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
|
1834 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
|
1835 { |
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
|
1836 int i, level, qmul, qadd; |
200 | 1837 int nCoeffs; |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1838 |
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
|
1839 if (s->mb_intra) { |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1840 if (!s->h263_aic) { |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1841 if (n < 4) |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1842 block[0] = block[0] * s->y_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1843 else |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1844 block[0] = block[0] * s->c_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1845 } |
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
|
1846 i = 1; |
200 | 1847 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
|
1848 } 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
|
1849 i = 0; |
200 | 1850 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
|
1851 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1852 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1853 qmul = s->qscale << 1; |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1854 if (s->h263_aic && s->mb_intra) |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1855 qadd = 0; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1856 else |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1857 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
|
1858 |
200 | 1859 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
|
1860 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
|
1861 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
|
1862 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
|
1863 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
|
1864 } 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
|
1865 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
|
1866 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1867 #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
|
1868 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
|
1869 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
|
1870 #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
|
1871 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
|
1872 } |
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
|
1873 } |
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
|
1874 } |
0 | 1875 |
1876 /* rate control */ | |
1877 | |
1878 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */ | |
1879 #define I_FRAME_SIZE_RATIO 3.0 | |
1880 #define QSCALE_K 20 | |
1881 | |
1882 static void rate_control_init(MpegEncContext *s) | |
1883 { | |
268 | 1884 #if 1 |
1885 emms_c(); | |
1886 | |
1887 //initial values, they dont really matter as they will be totally different within a few frames | |
1888 s->i_pred.coeff= s->p_pred.coeff= 7.0; | |
1889 s->i_pred.count= s->p_pred.count= 1.0; | |
1890 | |
1891 s->i_pred.decay= s->p_pred.decay= 0.4; | |
1892 | |
1893 // use more bits at the beginning, otherwise high motion at the begin will look like shit | |
1894 s->qsum=100; | |
1895 s->qcount=100; | |
1896 | |
1897 s->short_term_qsum=0.001; | |
1898 s->short_term_qcount=0.001; | |
1899 #else | |
0 | 1900 s->wanted_bits = 0; |
1901 | |
1902 if (s->intra_only) { | |
1903 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate; | |
1904 s->P_frame_bits = s->I_frame_bits; | |
1905 } else { | |
1906 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) / | |
1907 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1))); | |
1908 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO); | |
1909 } | |
268 | 1910 |
0 | 1911 #if defined(DEBUG) |
1912 printf("I_frame_size=%d P_frame_size=%d\n", | |
1913 s->I_frame_bits, s->P_frame_bits); | |
1914 #endif | |
268 | 1915 #endif |
1916 } | |
1917 | |
1918 static double predict(Predictor *p, double q, double var) | |
1919 { | |
1920 return p->coeff*var / (q*p->count); | |
1921 } | |
1922 | |
1923 static void update_predictor(Predictor *p, double q, double var, double size) | |
1924 { | |
1925 double new_coeff= size*q / (var + 1); | |
1926 if(var<1000) return; | |
1927 /*{ | |
1928 int pred= predict(p, q, var); | |
1929 int error= abs(pred-size); | |
1930 static double sum=0; | |
1931 static int count=0; | |
1932 if(count>5) sum+=error; | |
1933 count++; | |
1934 if(256*256*256*64%count==0){ | |
1935 printf("%d %f %f\n", count, sum/count, p->coeff); | |
1936 } | |
1937 }*/ | |
1938 p->count*= p->decay; | |
1939 p->coeff*= p->decay; | |
1940 p->count++; | |
1941 p->coeff+= new_coeff; | |
0 | 1942 } |
1943 | |
1944 static int rate_estimate_qscale(MpegEncContext *s) | |
1945 { | |
268 | 1946 #if 1 |
1947 int qmin= s->qmin; | |
1948 int qmax= s->qmax; | |
1949 int rate_q=5; | |
1950 float q; | |
1951 int qscale; | |
1952 float br_compensation; | |
1953 double diff; | |
1954 double short_term_q; | |
1955 double long_term_q; | |
1956 int last_qscale= s->qscale; | |
1957 double fps; | |
1958 INT64 wanted_bits; | |
1959 emms_c(); | |
297 | 1960 |
268 | 1961 fps= (double)s->frame_rate / FRAME_RATE_BASE; |
1962 wanted_bits= s->bit_rate*(double)s->picture_number/fps; | |
1963 | |
1964 | |
1965 if(s->picture_number>2){ | |
1966 /* update predictors */ | |
1967 if(s->last_pict_type == I_TYPE){ | |
1968 //FIXME | |
1969 }else{ //P Frame | |
1970 //printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff); | |
1971 update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits); | |
1972 } | |
1973 } | |
1974 | |
1975 if(s->pict_type == I_TYPE){ | |
1976 //FIXME | |
1977 rate_q= s->qsum/s->qcount; | |
1978 }else{ //P Frame | |
1979 int i; | |
1980 int diff, best_diff=1000000000; | |
1981 for(i=1; i<=31; i++){ | |
1982 diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps; | |
1983 if(diff<0) diff= -diff; | |
1984 if(diff<best_diff){ | |
1985 best_diff= diff; | |
1986 rate_q= i; | |
1987 } | |
1988 } | |
1989 } | |
1990 | |
1991 s->short_term_qsum*=s->qblur; | |
1992 s->short_term_qcount*=s->qblur; | |
1993 | |
1994 s->short_term_qsum+= rate_q; | |
1995 s->short_term_qcount++; | |
1996 short_term_q= s->short_term_qsum/s->short_term_qcount; | |
1997 | |
302 | 1998 long_term_q= s->qsum/s->qcount*(s->total_bits+1)/(wanted_bits+1); //+1 to avoid nan & 0 |
268 | 1999 |
2000 // q= (long_term_q - short_term_q)*s->qcompress + short_term_q; | |
2001 q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q); | |
2002 | |
2003 diff= s->total_bits - wanted_bits; | |
2004 br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance; | |
270 | 2005 if(br_compensation<=0.0) br_compensation=0.001; |
268 | 2006 q/=br_compensation; |
2007 | |
2008 qscale= (int)(q + 0.5); | |
2009 if (qscale<qmin) qscale=qmin; | |
2010 else if(qscale>qmax) qscale=qmax; | |
2011 | |
2012 if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff; | |
2013 else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff; | |
2014 | |
2015 s->qsum+= qscale; | |
2016 s->qcount++; | |
2017 | |
2018 s->last_pict_type= s->pict_type; | |
302 | 2019 //printf("q:%d diff:%d comp:%f rate_q:%d st_q:%f fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation, |
2020 // rate_q, short_term_q, s->mc_mb_var, s->frame_bits); | |
268 | 2021 //printf("%d %d\n", s->bit_rate, (int)fps); |
2022 return qscale; | |
2023 #else | |
187
3f3b14d3a23d
qscale estimate fix, diff extended to 64bit wide. patch by Stephen Davies <steve@daviesfam.org>
arpi_esp
parents:
178
diff
changeset
|
2024 INT64 diff, total_bits = s->total_bits; |
0 | 2025 float q; |
268 | 2026 int qscale; |
0 | 2027 if (s->pict_type == I_TYPE) { |
2028 s->wanted_bits += s->I_frame_bits; | |
2029 } else { | |
2030 s->wanted_bits += s->P_frame_bits; | |
2031 } | |
2032 diff = s->wanted_bits - total_bits; | |
2033 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width); | |
2034 /* adjust for I frame */ | |
2035 if (s->pict_type == I_TYPE && !s->intra_only) { | |
2036 q /= I_FRAME_SIZE_RATIO; | |
2037 } | |
2038 | |
2039 /* using a too small Q scale leeds to problems in mpeg1 and h263 | |
2040 because AC coefficients are clamped to 255 or 127 */ | |
2041 qmin = 3; | |
2042 if (q < qmin) | |
2043 q = qmin; | |
2044 else if (q > 31) | |
2045 q = 31; | |
2046 qscale = (int)(q + 0.5); | |
2047 #if defined(DEBUG) | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2048 printf("\n%d: total=%0.0f wanted=%0.0f br=%0.1f diff=%d qest=%2.1f\n", |
0 | 2049 s->picture_number, |
64 | 2050 (double)total_bits, |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2051 (double)s->wanted_bits, |
0 | 2052 (float)s->frame_rate / FRAME_RATE_BASE * |
2053 total_bits / s->picture_number, | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2054 (int)diff, q); |
0 | 2055 #endif |
2056 return qscale; | |
268 | 2057 #endif |
0 | 2058 } |
2059 | |
2060 AVCodec mpeg1video_encoder = { | |
2061 "mpeg1video", | |
2062 CODEC_TYPE_VIDEO, | |
2063 CODEC_ID_MPEG1VIDEO, | |
2064 sizeof(MpegEncContext), | |
2065 MPV_encode_init, | |
2066 MPV_encode_picture, | |
2067 MPV_encode_end, | |
2068 }; | |
2069 | |
2070 AVCodec h263_encoder = { | |
2071 "h263", | |
2072 CODEC_TYPE_VIDEO, | |
2073 CODEC_ID_H263, | |
2074 sizeof(MpegEncContext), | |
2075 MPV_encode_init, | |
2076 MPV_encode_picture, | |
2077 MPV_encode_end, | |
2078 }; | |
2079 | |
2080 AVCodec h263p_encoder = { | |
2081 "h263p", | |
2082 CODEC_TYPE_VIDEO, | |
2083 CODEC_ID_H263P, | |
2084 sizeof(MpegEncContext), | |
2085 MPV_encode_init, | |
2086 MPV_encode_picture, | |
2087 MPV_encode_end, | |
2088 }; | |
2089 | |
2090 AVCodec rv10_encoder = { | |
2091 "rv10", | |
2092 CODEC_TYPE_VIDEO, | |
2093 CODEC_ID_RV10, | |
2094 sizeof(MpegEncContext), | |
2095 MPV_encode_init, | |
2096 MPV_encode_picture, | |
2097 MPV_encode_end, | |
2098 }; | |
2099 | |
2100 AVCodec mjpeg_encoder = { | |
2101 "mjpeg", | |
2102 CODEC_TYPE_VIDEO, | |
2103 CODEC_ID_MJPEG, | |
2104 sizeof(MpegEncContext), | |
2105 MPV_encode_init, | |
2106 MPV_encode_picture, | |
2107 MPV_encode_end, | |
2108 }; | |
2109 | |
71 | 2110 AVCodec mpeg4_encoder = { |
2111 "mpeg4", | |
0 | 2112 CODEC_TYPE_VIDEO, |
71 | 2113 CODEC_ID_MPEG4, |
0 | 2114 sizeof(MpegEncContext), |
2115 MPV_encode_init, | |
2116 MPV_encode_picture, | |
2117 MPV_encode_end, | |
2118 }; | |
2119 | |
2120 AVCodec msmpeg4_encoder = { | |
2121 "msmpeg4", | |
2122 CODEC_TYPE_VIDEO, | |
2123 CODEC_ID_MSMPEG4, | |
2124 sizeof(MpegEncContext), | |
2125 MPV_encode_init, | |
2126 MPV_encode_picture, | |
2127 MPV_encode_end, | |
2128 }; |