Mercurial > libavcodec.hg
annotate mpegvideo.c @ 255:db20b987c32d libavcodec
divx5-gmc support
q-pel mc support
neither is totally bugfree yet though :(
author | michaelni |
---|---|
date | Sat, 09 Mar 2002 13:01:16 +0000 |
parents | ddb1a0e94cf4 |
children | b8d11794f675 |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest mpeg encoder (well, it was the simplest!) | |
3 * Copyright (c) 2000,2001 Gerard Lantau. | |
4 * | |
5 * This program is free software; you can redistribute it and/or modify | |
6 * it under the terms of the GNU General Public License as published by | |
7 * the Free Software Foundation; either version 2 of the License, or | |
8 * (at your option) any later version. | |
9 * | |
10 * This program is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 * GNU General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU General Public License | |
16 * along with this program; if not, write to the Free Software | |
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
18 */ | |
19 #include <stdlib.h> | |
20 #include <stdio.h> | |
21 #include <math.h> | |
22 #include <string.h> | |
23 #include "avcodec.h" | |
24 #include "dsputil.h" | |
25 #include "mpegvideo.h" | |
26 | |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
27 #ifdef USE_FASTMEMCPY |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
28 #include "fastmemcpy.h" |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
29 #endif |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
30 |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
31 static void encode_picture(MpegEncContext *s, int picture_number); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
32 static void rate_control_init(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
33 static int rate_estimate_qscale(MpegEncContext *s); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
34 static void dct_unquantize_mpeg1_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
35 DCTELEM *block, int n, int qscale); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
36 static void dct_unquantize_h263_c(MpegEncContext *s, |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
37 DCTELEM *block, int n, int qscale); |
206 | 38 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
220 | 39 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
206 | 40 |
220 | 41 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c; |
206 | 42 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
43 | |
0 | 44 #define EDGE_WIDTH 16 |
45 | |
46 /* enable all paranoid tests for rounding, overflows, etc... */ | |
47 //#define PARANOID | |
48 | |
49 //#define DEBUG | |
50 | |
51 /* for jpeg fast DCT */ | |
52 #define CONST_BITS 14 | |
53 | |
54 static const unsigned short aanscales[64] = { | |
55 /* precomputed values scaled up by 14 bits */ | |
56 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
57 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
58 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
59 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
60 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
61 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
62 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
63 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
64 }; | |
65 | |
66 static UINT8 h263_chroma_roundtab[16] = { | |
67 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
68 }; | |
69 | |
70 /* default motion estimation */ | |
71 int motion_estimation_method = ME_LOG; | |
72 | |
200 | 73 extern UINT8 zigzag_end[64]; |
74 | |
220 | 75 static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale) |
0 | 76 { |
77 int i; | |
78 | |
79 if (av_fdct == jpeg_fdct_ifast) { | |
80 for(i=0;i<64;i++) { | |
81 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
220 | 82 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
83 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
84 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
0 | 85 |
220 | 86 qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
87 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)])); | |
0 | 88 } |
89 } else { | |
90 for(i=0;i<64;i++) { | |
91 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
220 | 92 So 16 <= qscale * quant_matrix[i] <= 7905 |
93 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
94 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
0 | 95 */ |
220 | 96 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
97 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
0 | 98 } |
99 } | |
100 } | |
101 | |
102 /* init common structure for both encoder and decoder */ | |
103 int MPV_common_init(MpegEncContext *s) | |
104 { | |
105 int c_size, i; | |
106 UINT8 *pict; | |
107 | |
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
|
108 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
|
109 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
|
110 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
|
111 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
|
112 |
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 #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
|
114 MPV_common_init_mmx(s); |
8 | 115 #endif |
0 | 116 s->mb_width = (s->width + 15) / 16; |
117 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
|
118 s->mb_num = s->mb_width * s->mb_height; |
0 | 119 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
120 | |
121 for(i=0;i<3;i++) { | |
122 int w, h, shift, pict_start; | |
123 | |
124 w = s->linesize; | |
125 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
126 shift = (i == 0) ? 0 : 1; | |
127 c_size = (w >> shift) * (h >> shift); | |
128 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
129 | |
130 pict = av_mallocz(c_size); | |
131 if (pict == NULL) | |
132 goto fail; | |
133 s->last_picture_base[i] = pict; | |
134 s->last_picture[i] = pict + pict_start; | |
135 | |
136 pict = av_mallocz(c_size); | |
137 if (pict == NULL) | |
138 goto fail; | |
139 s->next_picture_base[i] = pict; | |
140 s->next_picture[i] = pict + pict_start; | |
141 | |
142 if (s->has_b_frames) { | |
143 pict = av_mallocz(c_size); | |
144 if (pict == NULL) | |
145 goto fail; | |
146 s->aux_picture_base[i] = pict; | |
147 s->aux_picture[i] = pict + pict_start; | |
148 } | |
149 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
150 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
151 if (s->encoding) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
152 /* Allocate MB type table */ |
239 | 153 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
|
154 if (s->mb_type == NULL) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
155 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
156 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
157 } |
239 | 158 |
159 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16)); | |
160 if (s->mb_var == NULL) { | |
161 perror("malloc"); | |
162 goto fail; | |
163 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
164 /* Allocate MV table */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
165 /* By now we just have one MV per MB */ |
239 | 166 s->mv_table[0] = av_mallocz(s->mb_num * sizeof(INT16)); |
167 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
|
168 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
|
169 perror("malloc"); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
170 goto fail; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
171 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
172 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
173 |
0 | 174 if (s->out_format == FMT_H263) { |
175 int size; | |
176 /* MV prediction */ | |
177 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
178 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
179 if (s->motion_val == NULL) | |
180 goto fail; | |
181 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
182 } | |
183 | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
184 if (s->h263_pred || s->h263_plus) { |
0 | 185 int y_size, c_size, i, size; |
186 | |
187 /* dc values */ | |
188 | |
189 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
190 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
191 size = y_size + 2 * c_size; | |
192 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
193 if (s->dc_val[0] == NULL) | |
194 goto fail; | |
195 s->dc_val[1] = s->dc_val[0] + y_size; | |
196 s->dc_val[2] = s->dc_val[1] + c_size; | |
197 for(i=0;i<size;i++) | |
198 s->dc_val[0][i] = 1024; | |
199 | |
200 /* ac values */ | |
201 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
202 if (s->ac_val[0] == NULL) | |
203 goto fail; | |
204 s->ac_val[1] = s->ac_val[0] + y_size; | |
205 s->ac_val[2] = s->ac_val[1] + c_size; | |
206 | |
207 /* cbp values */ | |
208 s->coded_block = av_mallocz(y_size); | |
209 if (!s->coded_block) | |
210 goto fail; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
211 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
212 /* 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
|
213 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
|
214 if (!s->mbintra_table) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
215 goto fail; |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
216 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
|
217 } |
0 | 218 /* default structure is frame */ |
219 s->picture_structure = PICT_FRAME; | |
220 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
221 /* init macroblock skip table */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
222 if (!s->encoding) { |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
223 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
|
224 if (!s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
225 goto fail; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
226 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
227 |
0 | 228 s->context_initialized = 1; |
229 return 0; | |
230 fail: | |
244 | 231 MPV_common_end(s); |
232 return -1; | |
233 } | |
234 | |
235 /* init common structure for both encoder and decoder */ | |
236 void MPV_common_end(MpegEncContext *s) | |
237 { | |
238 int i; | |
239 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
240 if (s->mb_type) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
241 free(s->mb_type); |
239 | 242 if (s->mb_var) |
243 free(s->mb_var); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
244 if (s->mv_table[0]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
245 free(s->mv_table[0]); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
246 if (s->mv_table[1]) |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
247 free(s->mv_table[1]); |
0 | 248 if (s->motion_val) |
249 free(s->motion_val); | |
250 if (s->dc_val[0]) | |
251 free(s->dc_val[0]); | |
252 if (s->ac_val[0]) | |
253 free(s->ac_val[0]); | |
254 if (s->coded_block) | |
255 free(s->coded_block); | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
256 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
|
257 free(s->mbintra_table); |
244 | 258 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
259 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
260 free(s->mbskip_table); |
0 | 261 for(i=0;i<3;i++) { |
262 if (s->last_picture_base[i]) | |
244 | 263 free(s->last_picture_base[i]); |
264 if (s->next_picture_base[i]) | |
265 free(s->next_picture_base[i]); | |
0 | 266 if (s->has_b_frames) |
267 free(s->aux_picture_base[i]); | |
268 } | |
269 s->context_initialized = 0; | |
270 } | |
271 | |
272 /* init video encoder */ | |
273 int MPV_encode_init(AVCodecContext *avctx) | |
274 { | |
275 MpegEncContext *s = avctx->priv_data; | |
60 | 276 int i; |
0 | 277 |
278 s->bit_rate = avctx->bit_rate; | |
279 s->frame_rate = avctx->frame_rate; | |
280 s->width = avctx->width; | |
281 s->height = avctx->height; | |
282 s->gop_size = avctx->gop_size; | |
162 | 283 s->rtp_mode = avctx->rtp_mode; |
284 s->rtp_payload_size = avctx->rtp_payload_size; | |
231 | 285 if (avctx->rtp_callback) |
286 s->rtp_callback = avctx->rtp_callback; | |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
287 s->avctx = avctx; |
162 | 288 |
0 | 289 if (s->gop_size <= 1) { |
290 s->intra_only = 1; | |
291 s->gop_size = 12; | |
292 } else { | |
293 s->intra_only = 0; | |
294 } | |
295 s->full_search = motion_estimation_method; | |
296 | |
297 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
|
298 |
0 | 299 switch(avctx->codec->id) { |
300 case CODEC_ID_MPEG1VIDEO: | |
301 s->out_format = FMT_MPEG1; | |
302 break; | |
303 case CODEC_ID_MJPEG: | |
304 s->out_format = FMT_MJPEG; | |
305 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
|
306 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
|
307 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
|
308 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
|
309 s->mjpeg_vsample[2] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
310 s->mjpeg_hsample[0] = 2; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
311 s->mjpeg_hsample[1] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
312 s->mjpeg_hsample[2] = 1; |
0 | 313 if (mjpeg_init(s) < 0) |
314 return -1; | |
315 break; | |
316 case CODEC_ID_H263: | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
317 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
|
318 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
0 | 319 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
320 } |
0 | 321 s->out_format = FMT_H263; |
322 break; | |
323 case CODEC_ID_H263P: | |
324 s->out_format = FMT_H263; | |
162 | 325 s->rtp_mode = 1; |
326 s->rtp_payload_size = 1200; | |
0 | 327 s->h263_plus = 1; |
78 | 328 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
329 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
330 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
331 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
332 s->umvplus_dec = 0; |
0 | 333 break; |
334 case CODEC_ID_RV10: | |
335 s->out_format = FMT_H263; | |
336 s->h263_rv10 = 1; | |
337 break; | |
71 | 338 case CODEC_ID_MPEG4: |
0 | 339 s->out_format = FMT_H263; |
340 s->h263_pred = 1; | |
341 s->unrestricted_mv = 1; | |
342 break; | |
343 case CODEC_ID_MSMPEG4: | |
344 s->out_format = FMT_H263; | |
345 s->h263_msmpeg4 = 1; | |
346 s->h263_pred = 1; | |
347 s->unrestricted_mv = 1; | |
348 break; | |
349 default: | |
350 return -1; | |
351 } | |
352 | |
353 if (s->out_format == FMT_H263) | |
354 h263_encode_init_vlc(s); | |
355 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
356 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
357 |
0 | 358 /* init */ |
359 if (MPV_common_init(s) < 0) | |
360 return -1; | |
361 | |
60 | 362 /* init default q matrix */ |
363 for(i=0;i<64;i++) { | |
364 s->intra_matrix[i] = default_intra_matrix[i]; | |
365 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
366 } | |
367 | |
0 | 368 /* rate control init */ |
369 rate_control_init(s); | |
370 | |
371 s->picture_number = 0; | |
372 s->fake_picture_number = 0; | |
373 /* motion detector init */ | |
374 s->f_code = 1; | |
375 | |
376 return 0; | |
377 } | |
378 | |
379 int MPV_encode_end(AVCodecContext *avctx) | |
380 { | |
381 MpegEncContext *s = avctx->priv_data; | |
382 | |
383 #ifdef STATS | |
384 print_stats(); | |
385 #endif | |
386 MPV_common_end(s); | |
387 if (s->out_format == FMT_MJPEG) | |
388 mjpeg_close(s); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
389 |
0 | 390 return 0; |
391 } | |
392 | |
393 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 394 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 395 { |
396 UINT8 *ptr, *last_line; | |
397 int i; | |
398 | |
399 last_line = buf + (height - 1) * wrap; | |
400 for(i=0;i<w;i++) { | |
401 /* top and bottom */ | |
402 memcpy(buf - (i + 1) * wrap, buf, width); | |
403 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
404 } | |
405 /* left and right */ | |
406 ptr = buf; | |
407 for(i=0;i<height;i++) { | |
408 memset(ptr - w, ptr[0], w); | |
409 memset(ptr + width, ptr[width-1], w); | |
410 ptr += wrap; | |
411 } | |
412 /* corners */ | |
413 for(i=0;i<w;i++) { | |
414 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
415 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
416 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
417 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
418 } | |
419 } | |
420 | |
421 /* generic function for encode/decode called before a frame is coded/decoded */ | |
422 void MPV_frame_start(MpegEncContext *s) | |
423 { | |
424 int i; | |
425 UINT8 *tmp; | |
426 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
427 s->mb_skiped = 0; |
0 | 428 if (s->pict_type == B_TYPE) { |
429 for(i=0;i<3;i++) { | |
430 s->current_picture[i] = s->aux_picture[i]; | |
431 } | |
432 } else { | |
433 for(i=0;i<3;i++) { | |
434 /* swap next and last */ | |
435 tmp = s->last_picture[i]; | |
436 s->last_picture[i] = s->next_picture[i]; | |
437 s->next_picture[i] = tmp; | |
438 s->current_picture[i] = tmp; | |
439 } | |
440 } | |
441 } | |
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
|
442 |
0 | 443 /* generic function for encode/decode called after a frame has been coded/decoded */ |
444 void MPV_frame_end(MpegEncContext *s) | |
445 { | |
446 /* draw edge for correct motion prediction if outside */ | |
220 | 447 if (s->pict_type != B_TYPE && !s->intra_only) { |
198
82ba367b1827
rv10 sig11 fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
197
diff
changeset
|
448 if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4){ |
176
9ce215ee9216
unrestricted MC fixed - thanks to Michael Niedermayer for idea
arpi_esp
parents:
162
diff
changeset
|
449 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
|
450 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
|
451 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
|
452 }else{ |
42552c1cf360
fix different UMV handling for mpeg4 vs. h263 - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
187
diff
changeset
|
453 /* OpenDivx, but i dunno how to distinguish it from mpeg4 */ |
0 | 454 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
455 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
456 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
|
457 } |
0 | 458 } |
207 | 459 emms_c(); |
0 | 460 } |
461 | |
462 int MPV_encode_picture(AVCodecContext *avctx, | |
463 unsigned char *buf, int buf_size, void *data) | |
464 { | |
465 MpegEncContext *s = avctx->priv_data; | |
466 AVPicture *pict = data; | |
467 int i, j; | |
468 | |
469 if (s->fixed_qscale) | |
470 s->qscale = avctx->quality; | |
471 | |
472 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
473 | |
474 if (!s->intra_only) { | |
475 /* first picture of GOP is intra */ | |
476 if ((s->picture_number % s->gop_size) == 0) | |
477 s->pict_type = I_TYPE; | |
478 else | |
479 s->pict_type = P_TYPE; | |
480 } else { | |
481 s->pict_type = I_TYPE; | |
482 } | |
483 avctx->key_frame = (s->pict_type == I_TYPE); | |
484 | |
485 MPV_frame_start(s); | |
220 | 486 |
0 | 487 for(i=0;i<3;i++) { |
488 UINT8 *src = pict->data[i]; | |
489 UINT8 *dest = s->current_picture[i]; | |
490 int src_wrap = pict->linesize[i]; | |
491 int dest_wrap = s->linesize; | |
492 int w = s->width; | |
493 int h = s->height; | |
494 | |
495 if (i >= 1) { | |
496 dest_wrap >>= 1; | |
497 w >>= 1; | |
498 h >>= 1; | |
499 } | |
500 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
501 if(dest_wrap==src_wrap){ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
502 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
|
503 } else { |
220 | 504 for(j=0;j<h;j++) { |
505 memcpy(dest, src, w); | |
506 dest += dest_wrap; | |
507 src += src_wrap; | |
508 } | |
227
ec1bc02a0a47
avoid copying input when encoding non intra stuff too
michaelni
parents:
220
diff
changeset
|
509 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
|
510 } |
0 | 511 } |
512 | |
513 encode_picture(s, s->picture_number); | |
514 | |
515 MPV_frame_end(s); | |
516 s->picture_number++; | |
517 | |
518 if (s->out_format == FMT_MJPEG) | |
519 mjpeg_picture_trailer(s); | |
520 | |
521 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
|
522 s->total_bits += (pbBufPtr(&s->pb) - s->pb.buf) * 8; |
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
|
523 |
0 | 524 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
|
525 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
|
526 /* 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
|
527 /* 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
|
528 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
|
529 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
|
530 } |
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
|
531 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 532 } |
533 | |
534 static inline int clip(int a, int amin, int amax) | |
535 { | |
536 if (a < amin) | |
537 return amin; | |
538 else if (a > amax) | |
539 return amax; | |
540 else | |
541 return a; | |
542 } | |
543 | |
255 | 544 static inline void gmc1_motion(MpegEncContext *s, |
545 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
546 int dest_offset, | |
547 UINT8 **ref_picture, int src_offset, | |
548 int h) | |
549 { | |
550 UINT8 *ptr; | |
551 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
552 int motion_x, motion_y; | |
553 | |
554 if(s->real_sprite_warping_points>1) printf("Oops, thats bad, contact the developers\n"); | |
555 motion_x= s->sprite_offset[0][0]; | |
556 motion_y= s->sprite_offset[0][1]; | |
557 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
558 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
559 motion_x<<=(3-s->sprite_warping_accuracy); | |
560 motion_y<<=(3-s->sprite_warping_accuracy); | |
561 src_x = clip(src_x, -16, s->width); | |
562 if (src_x == s->width) | |
563 motion_x =0; | |
564 src_y = clip(src_y, -16, s->height); | |
565 if (src_y == s->height) | |
566 motion_y =0; | |
567 | |
568 linesize = s->linesize; | |
569 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
570 | |
571 dest_y+=dest_offset; | |
572 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
573 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
574 | |
575 motion_x= s->sprite_offset[1][0]; | |
576 motion_y= s->sprite_offset[1][1]; | |
577 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
578 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
579 motion_x<<=(3-s->sprite_warping_accuracy); | |
580 motion_y<<=(3-s->sprite_warping_accuracy); | |
581 src_x = clip(src_x, -8, s->width>>1); | |
582 if (src_x == s->width>>1) | |
583 motion_x =0; | |
584 src_y = clip(src_y, -8, s->height>>1); | |
585 if (src_y == s->height>>1) | |
586 motion_y =0; | |
587 | |
588 offset = (src_y * linesize>>1) + src_x + (src_offset>>1); | |
589 ptr = ref_picture[1] + offset; | |
590 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
591 ptr = ref_picture[2] + offset; | |
592 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding); | |
593 | |
594 return; | |
595 } | |
596 | |
0 | 597 /* apply one mpeg motion vector to the three components */ |
598 static inline void mpeg_motion(MpegEncContext *s, | |
599 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
600 int dest_offset, | |
601 UINT8 **ref_picture, int src_offset, | |
602 int field_based, op_pixels_func *pix_op, | |
603 int motion_x, int motion_y, int h) | |
604 { | |
605 UINT8 *ptr; | |
606 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
255 | 607 if(s->quarter_sample) |
608 { | |
609 motion_x>>=1; | |
610 motion_y>>=1; | |
611 } | |
0 | 612 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
613 src_x = s->mb_x * 16 + (motion_x >> 1); | |
614 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
615 | |
616 /* WARNING: do no forget half pels */ | |
617 height = s->height >> field_based; | |
618 src_x = clip(src_x, -16, s->width); | |
619 if (src_x == s->width) | |
620 dxy &= ~1; | |
621 src_y = clip(src_y, -16, height); | |
622 if (src_y == height) | |
623 dxy &= ~2; | |
624 linesize = s->linesize << field_based; | |
625 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
626 dest_y += dest_offset; | |
627 pix_op[dxy](dest_y, ptr, linesize, h); | |
628 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
629 | |
630 if (s->out_format == FMT_H263) { | |
631 dxy = 0; | |
632 if ((motion_x & 3) != 0) | |
633 dxy |= 1; | |
634 if ((motion_y & 3) != 0) | |
635 dxy |= 2; | |
636 mx = motion_x >> 2; | |
637 my = motion_y >> 2; | |
638 } else { | |
639 mx = motion_x / 2; | |
640 my = motion_y / 2; | |
641 dxy = ((my & 1) << 1) | (mx & 1); | |
642 mx >>= 1; | |
643 my >>= 1; | |
644 } | |
645 | |
646 src_x = s->mb_x * 8 + mx; | |
647 src_y = s->mb_y * (8 >> field_based) + my; | |
648 src_x = clip(src_x, -8, s->width >> 1); | |
649 if (src_x == (s->width >> 1)) | |
650 dxy &= ~1; | |
651 src_y = clip(src_y, -8, height >> 1); | |
652 if (src_y == (height >> 1)) | |
653 dxy &= ~2; | |
654 | |
655 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
656 ptr = ref_picture[1] + offset; | |
657 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
658 ptr = ref_picture[2] + offset; | |
659 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
660 } | |
661 | |
255 | 662 static inline void qpel_motion(MpegEncContext *s, |
663 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
664 int dest_offset, | |
665 UINT8 **ref_picture, int src_offset, | |
666 int field_based, op_pixels_func *pix_op, | |
667 qpel_mc_func *qpix_op, | |
668 int motion_x, int motion_y, int h) | |
669 { | |
670 UINT8 *ptr; | |
671 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
672 | |
673 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
674 src_x = s->mb_x * 16 + (motion_x >> 2); | |
675 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
676 | |
677 height = s->height >> field_based; | |
678 src_x = clip(src_x, -16, s->width); | |
679 if (src_x == s->width) | |
680 dxy &= ~3; | |
681 src_y = clip(src_y, -16, height); | |
682 if (src_y == height) | |
683 dxy &= ~12; | |
684 linesize = s->linesize << field_based; | |
685 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
686 dest_y += dest_offset; | |
687 //printf("%d %d %d\n", src_x, src_y, dxy); | |
688 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3); | |
689 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3); | |
690 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3); | |
691 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3); | |
692 | |
693 mx= (motion_x>>1) | (motion_x&1); | |
694 my= (motion_y>>1) | (motion_y&1); | |
695 | |
696 dxy = 0; | |
697 if ((mx & 3) != 0) | |
698 dxy |= 1; | |
699 if ((my & 3) != 0) | |
700 dxy |= 2; | |
701 mx = mx >> 2; | |
702 my = my >> 2; | |
703 | |
704 src_x = s->mb_x * 8 + mx; | |
705 src_y = s->mb_y * (8 >> field_based) + my; | |
706 src_x = clip(src_x, -8, s->width >> 1); | |
707 if (src_x == (s->width >> 1)) | |
708 dxy &= ~1; | |
709 src_y = clip(src_y, -8, height >> 1); | |
710 if (src_y == (height >> 1)) | |
711 dxy &= ~2; | |
712 | |
713 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
714 ptr = ref_picture[1] + offset; | |
715 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
716 ptr = ref_picture[2] + offset; | |
717 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
718 } | |
719 | |
720 | |
0 | 721 static inline void MPV_motion(MpegEncContext *s, |
722 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
723 int dir, UINT8 **ref_picture, | |
255 | 724 op_pixels_func *pix_op, qpel_mc_func *qpix_op) |
0 | 725 { |
726 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
727 int mb_x, mb_y, i; | |
728 UINT8 *ptr, *dest; | |
729 | |
730 mb_x = s->mb_x; | |
731 mb_y = s->mb_y; | |
732 | |
733 switch(s->mv_type) { | |
734 case MV_TYPE_16X16: | |
255 | 735 if(s->mcsel){ |
736 #if 0 | |
737 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
738 ref_picture, 0, | |
739 0, pix_op, | |
740 s->sprite_offset[0][0]>>3, | |
741 s->sprite_offset[0][1]>>3, | |
742 16); | |
743 #else | |
744 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
745 ref_picture, 0, | |
746 16); | |
747 #endif | |
748 }else if(s->quarter_sample){ | |
749 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, | |
750 ref_picture, 0, | |
751 0, pix_op, qpix_op, | |
752 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
753 }else{ | |
754 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
755 ref_picture, 0, | |
756 0, pix_op, | |
757 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
758 } | |
0 | 759 break; |
760 case MV_TYPE_8X8: | |
761 for(i=0;i<4;i++) { | |
762 motion_x = s->mv[dir][i][0]; | |
763 motion_y = s->mv[dir][i][1]; | |
764 | |
765 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
766 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
767 src_y = mb_y * 16 + (motion_y >> 1) + ((i >> 1) & 1) * 8; | |
768 | |
769 /* WARNING: do no forget half pels */ | |
770 src_x = clip(src_x, -16, s->width); | |
771 if (src_x == s->width) | |
772 dxy &= ~1; | |
773 src_y = clip(src_y, -16, s->height); | |
774 if (src_y == s->height) | |
775 dxy &= ~2; | |
776 | |
777 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
778 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
779 pix_op[dxy](dest, ptr, s->linesize, 8); | |
780 } | |
781 /* In case of 8X8, we construct a single chroma motion vector | |
782 with a special rounding */ | |
783 mx = 0; | |
784 my = 0; | |
785 for(i=0;i<4;i++) { | |
786 mx += s->mv[dir][i][0]; | |
787 my += s->mv[dir][i][1]; | |
788 } | |
789 if (mx >= 0) | |
790 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
791 else { | |
792 mx = -mx; | |
793 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
794 } | |
795 if (my >= 0) | |
796 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
797 else { | |
798 my = -my; | |
799 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
800 } | |
801 dxy = ((my & 1) << 1) | (mx & 1); | |
802 mx >>= 1; | |
803 my >>= 1; | |
804 | |
805 src_x = mb_x * 8 + mx; | |
806 src_y = mb_y * 8 + my; | |
807 src_x = clip(src_x, -8, s->width/2); | |
808 if (src_x == s->width/2) | |
809 dxy &= ~1; | |
810 src_y = clip(src_y, -8, s->height/2); | |
811 if (src_y == s->height/2) | |
812 dxy &= ~2; | |
813 | |
814 offset = (src_y * (s->linesize >> 1)) + src_x; | |
815 ptr = ref_picture[1] + offset; | |
816 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
817 ptr = ref_picture[2] + offset; | |
818 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
819 break; | |
820 case MV_TYPE_FIELD: | |
821 if (s->picture_structure == PICT_FRAME) { | |
822 /* top field */ | |
823 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
824 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
825 1, pix_op, | |
826 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
827 /* bottom field */ | |
828 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
829 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
830 1, pix_op, | |
831 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
832 } else { | |
833 | |
834 | |
835 } | |
836 break; | |
837 } | |
838 } | |
839 | |
840 | |
841 /* put block[] to dest[] */ | |
842 static inline void put_dct(MpegEncContext *s, | |
843 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
844 { | |
845 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
|
846 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
|
847 ff_idct (block); |
0 | 848 put_pixels_clamped(block, dest, line_size); |
849 } | |
850 | |
851 /* add block[] to dest[] */ | |
852 static inline void add_dct(MpegEncContext *s, | |
853 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
854 { | |
855 if (s->block_last_index[i] >= 0) { | |
856 if (!s->mpeg2) | |
206 | 857 if(s->encoding || (!s->h263_msmpeg4)) |
200 | 858 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
|
859 ff_idct (block); |
0 | 860 add_pixels_clamped(block, dest, line_size); |
861 } | |
862 } | |
863 | |
864 /* generic function called after a macroblock has been parsed by the | |
865 decoder or after it has been encoded by the encoder. | |
866 | |
867 Important variables used: | |
868 s->mb_intra : true if intra macroblock | |
869 s->mv_dir : motion vector direction | |
870 s->mv_type : motion vector type | |
871 s->mv : motion vector | |
872 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
873 */ | |
874 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
875 { | |
244 | 876 int mb_x, mb_y; |
0 | 877 int dct_linesize, dct_offset; |
878 op_pixels_func *op_pix; | |
255 | 879 qpel_mc_func *op_qpix; |
0 | 880 |
881 mb_x = s->mb_x; | |
882 mb_y = s->mb_y; | |
883 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
884 #ifdef FF_POSTPROCESS |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
885 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
886 //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
|
887 #endif |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
888 |
0 | 889 /* update DC predictors for P macroblocks */ |
890 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
891 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
|
892 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
|
893 { |
244 | 894 int wrap, xy, v; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
895 s->mbintra_table[mb_x + mb_y*s->mb_width]=0; |
0 | 896 wrap = 2 * s->mb_width + 2; |
244 | 897 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 898 v = 1024; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
899 |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
900 s->dc_val[0][xy] = v; |
244 | 901 s->dc_val[0][xy + 1] = v; |
902 s->dc_val[0][xy + wrap] = v; | |
903 s->dc_val[0][xy + 1 + wrap] = v; | |
0 | 904 /* ac pred */ |
244 | 905 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16)); |
906 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16)); | |
907 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16)); | |
908 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16)); | |
0 | 909 if (s->h263_msmpeg4) { |
244 | 910 s->coded_block[xy] = 0; |
911 s->coded_block[xy + 1] = 0; | |
912 s->coded_block[xy + wrap] = 0; | |
913 s->coded_block[xy + 1 + wrap] = 0; | |
0 | 914 } |
915 /* chroma */ | |
916 wrap = s->mb_width + 2; | |
244 | 917 xy = mb_x + 1 + (mb_y + 1) * wrap; |
918 s->dc_val[1][xy] = v; | |
919 s->dc_val[2][xy] = v; | |
0 | 920 /* ac pred */ |
244 | 921 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); |
922 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
|
923 } |
0 | 924 } else { |
925 s->last_dc[0] = 128 << s->intra_dc_precision; | |
926 s->last_dc[1] = 128 << s->intra_dc_precision; | |
927 s->last_dc[2] = 128 << s->intra_dc_precision; | |
928 } | |
929 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
930 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
|
931 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
|
932 |
0 | 933 /* update motion predictor */ |
934 if (s->out_format == FMT_H263) { | |
244 | 935 int xy, wrap, motion_x, motion_y; |
0 | 936 |
937 wrap = 2 * s->mb_width + 2; | |
244 | 938 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap; |
0 | 939 if (s->mb_intra) { |
940 motion_x = 0; | |
941 motion_y = 0; | |
942 goto motion_init; | |
943 } else if (s->mv_type == MV_TYPE_16X16) { | |
944 motion_x = s->mv[0][0][0]; | |
945 motion_y = s->mv[0][0][1]; | |
946 motion_init: | |
947 /* no update if 8X8 because it has been done during parsing */ | |
244 | 948 s->motion_val[xy][0] = motion_x; |
949 s->motion_val[xy][1] = motion_y; | |
950 s->motion_val[xy + 1][0] = motion_x; | |
951 s->motion_val[xy + 1][1] = motion_y; | |
952 s->motion_val[xy + wrap][0] = motion_x; | |
953 s->motion_val[xy + wrap][1] = motion_y; | |
954 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
955 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
0 | 956 } |
957 } | |
958 | |
959 if (!s->intra_only) { | |
960 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
|
961 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
962 |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
963 /* 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
|
964 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
|
965 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
|
966 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
967 s->mb_skiped = 0; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
968 /* 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
|
969 if (*mbskip_ptr != 0) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
970 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
971 *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
|
972 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
973 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
974 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
975 } |
0 | 976 |
977 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
978 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
979 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
980 | |
981 if (s->interlaced_dct) { | |
982 dct_linesize = s->linesize * 2; | |
983 dct_offset = s->linesize; | |
984 } else { | |
985 dct_linesize = s->linesize; | |
986 dct_offset = s->linesize * 8; | |
987 } | |
988 | |
989 if (!s->mb_intra) { | |
990 /* motion handling */ | |
255 | 991 if (!s->no_rounding){ |
0 | 992 op_pix = put_pixels_tab; |
255 | 993 op_qpix= qpel_mc_rnd_tab; |
994 }else{ | |
0 | 995 op_pix = put_no_rnd_pixels_tab; |
255 | 996 op_qpix= qpel_mc_no_rnd_tab; |
997 } | |
0 | 998 |
999 if (s->mv_dir & MV_DIR_FORWARD) { | |
255 | 1000 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); |
0 | 1001 if (!s->no_rounding) |
1002 op_pix = avg_pixels_tab; | |
1003 else | |
1004 op_pix = avg_no_rnd_pixels_tab; | |
1005 } | |
1006 if (s->mv_dir & MV_DIR_BACKWARD) { | |
255 | 1007 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); |
0 | 1008 } |
1009 | |
1010 /* add dct residue */ | |
1011 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
1012 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1013 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1014 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1015 | |
57 | 1016 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1017 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1018 } else { |
1019 /* dct only in intra block */ | |
1020 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
1021 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1022 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1023 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1024 | |
57 | 1025 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
1026 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 1027 } |
1028 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1029 the_end: |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1030 emms_c(); |
0 | 1031 } |
1032 | |
1033 static void encode_picture(MpegEncContext *s, int picture_number) | |
1034 { | |
231 | 1035 int mb_x, mb_y, wrap, last_gob, pdif = 0; |
0 | 1036 UINT8 *ptr; |
1037 int i, motion_x, motion_y; | |
1038 | |
1039 s->picture_number = picture_number; | |
1040 if (!s->fixed_qscale) | |
1041 s->qscale = rate_estimate_qscale(s); | |
1042 | |
1043 /* precompute matrix */ | |
1044 if (s->out_format == FMT_MJPEG) { | |
1045 /* for mjpeg, we do include qscale in the matrix */ | |
1046 s->intra_matrix[0] = default_intra_matrix[0]; | |
1047 for(i=1;i<64;i++) | |
1048 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
220 | 1049 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
0 | 1050 } else { |
220 | 1051 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
1052 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); | |
0 | 1053 } |
1054 | |
1055 switch(s->out_format) { | |
1056 case FMT_MJPEG: | |
1057 mjpeg_picture_header(s); | |
1058 break; | |
1059 case FMT_H263: | |
1060 if (s->h263_msmpeg4) | |
1061 msmpeg4_encode_picture_header(s, picture_number); | |
1062 else if (s->h263_pred) | |
1063 mpeg4_encode_picture_header(s, picture_number); | |
1064 else if (s->h263_rv10) | |
1065 rv10_encode_picture_header(s, picture_number); | |
1066 else | |
1067 h263_encode_picture_header(s, picture_number); | |
1068 break; | |
1069 case FMT_MPEG1: | |
1070 mpeg1_encode_picture_header(s, picture_number); | |
1071 break; | |
1072 } | |
1073 | |
1074 /* init last dc values */ | |
1075 /* note: quant matrix value (8) is implied here */ | |
1076 s->last_dc[0] = 128; | |
1077 s->last_dc[1] = 128; | |
1078 s->last_dc[2] = 128; | |
1079 s->mb_incr = 1; | |
1080 s->last_mv[0][0][0] = 0; | |
1081 s->last_mv[0][0][1] = 0; | |
1082 s->mv_type = MV_TYPE_16X16; | |
1083 s->mv_dir = MV_DIR_FORWARD; | |
1084 | |
162 | 1085 /* Get the GOB height based on picture height */ |
231 | 1086 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
162 | 1087 if (s->height <= 400) |
1088 s->gob_index = 1; | |
1089 else if (s->height <= 800) | |
1090 s->gob_index = 2; | |
1091 else | |
1092 s->gob_index = 4; | |
1093 } | |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1094 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1095 /* Reset the average MB variance */ |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1096 s->avg_mb_var = 0; |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1097 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1098 /* Estimate motion for every MB */ |
0 | 1099 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
1100 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
1101 s->mb_x = mb_x; | |
1102 s->mb_y = mb_y; | |
1103 | |
1104 /* compute motion vector and macro block type (intra or non intra) */ | |
1105 motion_x = 0; | |
1106 motion_y = 0; | |
1107 if (s->pict_type == P_TYPE) { | |
1108 s->mb_intra = estimate_motion(s, mb_x, mb_y, | |
1109 &motion_x, | |
1110 &motion_y); | |
1111 } else { | |
1112 s->mb_intra = 1; | |
1113 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1114 /* Store MB type and MV */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1115 s->mb_type[mb_y * s->mb_width + mb_x] = s->mb_intra; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1116 s->mv_table[0][mb_y * s->mb_width + mb_x] = motion_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1117 s->mv_table[1][mb_y * s->mb_width + mb_x] = motion_y; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1118 } |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1119 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1120 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1121 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
|
1122 |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1123 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
|
1124 /* 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
|
1125 /* 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
|
1126 if (s->rtp_mode) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1127 if (!mb_y) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1128 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
|
1129 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
|
1130 } 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
|
1131 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
|
1132 if (last_gob) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1133 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
|
1134 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1135 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1136 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
1137 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1138 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
0 | 1139 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1140 s->mb_x = mb_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1141 s->mb_y = mb_y; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1142 #if 0 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1143 /* compute motion vector and macro block type (intra or non intra) */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1144 motion_x = 0; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1145 motion_y = 0; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1146 if (s->pict_type == P_TYPE) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1147 s->mb_intra = estimate_motion(s, mb_x, mb_y, |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1148 &motion_x, |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1149 &motion_y); |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1150 } else { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1151 s->mb_intra = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1152 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1153 #endif |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1154 |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1155 s->mb_intra = s->mb_type[mb_y * s->mb_width + mb_x]; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1156 motion_x = s->mv_table[0][mb_y * s->mb_width + mb_x]; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1157 motion_y = s->mv_table[1][mb_y * s->mb_width + mb_x]; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1158 |
0 | 1159 /* get the pixels */ |
1160 wrap = s->linesize; | |
1161 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1162 get_pixels(s->block[0], ptr, wrap); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1163 get_pixels(s->block[1], ptr + 8, wrap); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1164 get_pixels(s->block[2], ptr + 8 * wrap, wrap); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1165 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); |
0 | 1166 wrap = s->linesize >> 1; |
1167 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1168 get_pixels(s->block[4], ptr, wrap); |
0 | 1169 |
1170 wrap = s->linesize >> 1; | |
1171 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1172 get_pixels(s->block[5], ptr, wrap); |
0 | 1173 |
1174 /* subtract previous frame if non intra */ | |
1175 if (!s->mb_intra) { | |
1176 int dxy, offset, mx, my; | |
1177 | |
1178 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
1179 ptr = s->last_picture[0] + | |
1180 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) + | |
1181 (mb_x * 16 + (motion_x >> 1)); | |
1182 | |
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
|
1183 sub_pixels_2(s->block[0], ptr, s->linesize, dxy); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1184 sub_pixels_2(s->block[1], ptr + 8, s->linesize, dxy); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1185 sub_pixels_2(s->block[2], ptr + s->linesize * 8, s->linesize, dxy); |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1186 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy); |
0 | 1187 |
1188 if (s->out_format == FMT_H263) { | |
1189 /* special rounding for h263 */ | |
1190 dxy = 0; | |
1191 if ((motion_x & 3) != 0) | |
1192 dxy |= 1; | |
1193 if ((motion_y & 3) != 0) | |
1194 dxy |= 2; | |
1195 mx = motion_x >> 2; | |
1196 my = motion_y >> 2; | |
1197 } else { | |
1198 mx = motion_x / 2; | |
1199 my = motion_y / 2; | |
1200 dxy = ((my & 1) << 1) | (mx & 1); | |
1201 mx >>= 1; | |
1202 my >>= 1; | |
1203 } | |
1204 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx); | |
1205 ptr = s->last_picture[1] + offset; | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1206 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); |
0 | 1207 ptr = s->last_picture[2] + offset; |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1208 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); |
0 | 1209 } |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1210 emms_c(); |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1211 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1212 #if 0 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1213 { |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1214 float adap_parm; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1215 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1216 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1217 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1218 |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1219 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1220 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1221 s->qscale, adap_parm, s->qscale*adap_parm, |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1222 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1223 } |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1224 #endif |
0 | 1225 /* DCT & quantize */ |
1226 if (s->h263_msmpeg4) { | |
1227 msmpeg4_dc_scale(s); | |
1228 } else if (s->h263_pred) { | |
1229 h263_dc_scale(s); | |
1230 } else { | |
1231 /* default quantization values */ | |
1232 s->y_dc_scale = 8; | |
1233 s->c_dc_scale = 8; | |
1234 } | |
1235 for(i=0;i<6;i++) { | |
220 | 1236 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); |
0 | 1237 } |
1238 | |
1239 /* huffman encode */ | |
1240 switch(s->out_format) { | |
1241 case FMT_MPEG1: | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1242 mpeg1_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1243 break; |
1244 case FMT_H263: | |
1245 if (s->h263_msmpeg4) | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1246 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1247 else |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1248 h263_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1249 break; |
1250 case FMT_MJPEG: | |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1251 mjpeg_encode_mb(s, s->block); |
0 | 1252 break; |
1253 } | |
1254 | |
1255 /* decompress blocks so that we keep the state of the decoder */ | |
1256 s->mv[0][0][0] = motion_x; | |
1257 s->mv[0][0][1] = motion_y; | |
1258 | |
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
|
1259 MPV_decode_mb(s, s->block); |
0 | 1260 } |
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
|
1261 |
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
|
1262 |
231 | 1263 /* Obtain average GOB size for RTP */ |
1264 if (s->rtp_mode) { | |
1265 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
|
1266 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
231 | 1267 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
|
1268 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
|
1269 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
231 | 1270 } |
1271 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
1272 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
1273 s->first_gob_line = 0; | |
1274 } | |
0 | 1275 } |
208 | 1276 |
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
|
1277 if (s->h263_msmpeg4 && s->pict_type == I_TYPE) |
208 | 1278 msmpeg4_encode_ext_header(s); |
1279 | |
162 | 1280 //if (s->gob_number) |
1281 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
231 | 1282 |
1283 /* Send the last GOB if RTP */ | |
1284 if (s->rtp_mode) { | |
1285 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
|
1286 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 1287 /* Call the RTP callback to send the last GOB */ |
1288 if (s->rtp_callback) | |
1289 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
|
1290 s->ptr_lastgob = pbBufPtr(&s->pb); |
231 | 1291 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
1292 } | |
1293 | |
0 | 1294 } |
1295 | |
220 | 1296 static int dct_quantize_c(MpegEncContext *s, |
0 | 1297 DCTELEM *block, int n, |
1298 int qscale) | |
1299 { | |
1300 int i, j, level, last_non_zero, q; | |
1301 const int *qmat; | |
216 | 1302 int minLevel, maxLevel; |
1303 | |
1304 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ | |
1305 /* mpeg4 */ | |
1306 minLevel= -2048; | |
1307 maxLevel= 2047; | |
1308 }else if(s->out_format==FMT_MPEG1){ | |
1309 /* mpeg1 */ | |
1310 minLevel= -255; | |
1311 maxLevel= 255; | |
230 | 1312 }else if(s->out_format==FMT_MJPEG){ |
1313 /* (m)jpeg */ | |
1314 minLevel= -1023; | |
1315 maxLevel= 1023; | |
216 | 1316 }else{ |
1317 /* h263 / msmpeg4 */ | |
1318 minLevel= -128; | |
1319 maxLevel= 127; | |
1320 } | |
0 | 1321 |
1322 av_fdct (block); | |
1323 | |
64 | 1324 /* we need this permutation so that we correct the IDCT |
1325 permutation. will be moved into DCT code */ | |
1326 block_permute(block); | |
1327 | |
0 | 1328 if (s->mb_intra) { |
1329 if (n < 4) | |
1330 q = s->y_dc_scale; | |
1331 else | |
1332 q = s->c_dc_scale; | |
1333 q = q << 3; | |
1334 | |
1335 /* note: block[0] is assumed to be positive */ | |
1336 block[0] = (block[0] + (q >> 1)) / q; | |
1337 i = 1; | |
1338 last_non_zero = 0; | |
1339 if (s->out_format == FMT_H263) { | |
1340 qmat = s->q_non_intra_matrix; | |
1341 } else { | |
1342 qmat = s->q_intra_matrix; | |
1343 } | |
1344 } else { | |
1345 i = 0; | |
1346 last_non_zero = -1; | |
1347 qmat = s->q_non_intra_matrix; | |
1348 } | |
1349 | |
1350 for(;i<64;i++) { | |
1351 j = zigzag_direct[i]; | |
1352 level = block[j]; | |
1353 level = level * qmat[j]; | |
1354 #ifdef PARANOID | |
1355 { | |
1356 static int count = 0; | |
1357 int level1, level2, qmat1; | |
1358 double val; | |
1359 if (qmat == s->q_non_intra_matrix) { | |
1360 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1361 } else { | |
1362 qmat1 = default_intra_matrix[j] * s->qscale; | |
1363 } | |
1364 if (av_fdct != jpeg_fdct_ifast) | |
1365 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1366 else | |
1367 val = ((double)block[j] * 8.0 * 2048.0) / | |
1368 ((double)qmat1 * aanscales[j]); | |
1369 level1 = (int)val; | |
1370 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1371 if (level1 != level2) { | |
1372 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1373 count, level2, level1, block[j], qmat1, qmat[j], | |
1374 val); | |
1375 count++; | |
1376 } | |
1377 | |
1378 } | |
1379 #endif | |
1380 /* XXX: slight error for the low range. Test should be equivalent to | |
1381 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1382 (QMAT_SHIFT - 3))) | |
1383 */ | |
1384 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1385 level) { | |
1386 level = level / (1 << (QMAT_SHIFT - 3)); | |
1387 /* XXX: currently, this code is not optimal. the range should be: | |
1388 mpeg1: -255..255 | |
1389 mpeg2: -2048..2047 | |
1390 h263: -128..127 | |
1391 mpeg4: -2048..2047 | |
1392 */ | |
216 | 1393 if (level > maxLevel) |
1394 level = maxLevel; | |
1395 else if (level < minLevel) | |
1396 level = minLevel; | |
0 | 1397 |
1398 block[j] = level; | |
1399 last_non_zero = i; | |
1400 } else { | |
1401 block[j] = 0; | |
1402 } | |
1403 } | |
1404 return last_non_zero; | |
1405 } | |
1406 | |
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
|
1407 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
|
1408 DCTELEM *block, int n, int qscale) |
0 | 1409 { |
200 | 1410 int i, level, nCoeffs; |
0 | 1411 const UINT16 *quant_matrix; |
1412 | |
200 | 1413 if(s->alternate_scan) nCoeffs= 64; |
1414 else nCoeffs= s->block_last_index[n]+1; | |
1415 | |
0 | 1416 if (s->mb_intra) { |
1417 if (n < 4) | |
1418 block[0] = block[0] * s->y_dc_scale; | |
1419 else | |
1420 block[0] = block[0] * s->c_dc_scale; | |
1421 /* XXX: only mpeg1 */ | |
1422 quant_matrix = s->intra_matrix; | |
200 | 1423 for(i=1;i<nCoeffs;i++) { |
1424 int j= zigzag_direct[i]; | |
1425 level = block[j]; | |
0 | 1426 if (level) { |
1427 if (level < 0) { | |
1428 level = -level; | |
200 | 1429 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1430 level = (level - 1) | 1; |
1431 level = -level; | |
1432 } else { | |
200 | 1433 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1434 level = (level - 1) | 1; |
1435 } | |
1436 #ifdef PARANOID | |
1437 if (level < -2048 || level > 2047) | |
1438 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1439 #endif | |
200 | 1440 block[j] = level; |
0 | 1441 } |
1442 } | |
1443 } else { | |
1444 i = 0; | |
1445 quant_matrix = s->non_intra_matrix; | |
217 | 1446 for(;i<nCoeffs;i++) { |
200 | 1447 int j= zigzag_direct[i]; |
1448 level = block[j]; | |
0 | 1449 if (level) { |
1450 if (level < 0) { | |
1451 level = -level; | |
1452 level = (((level << 1) + 1) * qscale * | |
200 | 1453 ((int) (quant_matrix[j]))) >> 4; |
0 | 1454 level = (level - 1) | 1; |
1455 level = -level; | |
1456 } else { | |
1457 level = (((level << 1) + 1) * qscale * | |
200 | 1458 ((int) (quant_matrix[j]))) >> 4; |
0 | 1459 level = (level - 1) | 1; |
1460 } | |
1461 #ifdef PARANOID | |
1462 if (level < -2048 || level > 2047) | |
1463 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1464 #endif | |
200 | 1465 block[j] = level; |
0 | 1466 } |
1467 } | |
1468 } | |
1469 } | |
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
|
1470 |
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
|
1471 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
|
1472 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
|
1473 { |
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
|
1474 int i, level, qmul, qadd; |
200 | 1475 int nCoeffs; |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1476 |
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
|
1477 if (s->mb_intra) { |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1478 if (!s->h263_aic) { |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1479 if (n < 4) |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1480 block[0] = block[0] * s->y_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1481 else |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1482 block[0] = block[0] * s->c_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1483 } |
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
|
1484 i = 1; |
200 | 1485 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
|
1486 } 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
|
1487 i = 0; |
200 | 1488 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
|
1489 } |
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
|
1490 |
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
|
1491 qmul = s->qscale << 1; |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1492 if (s->h263_aic && s->mb_intra) |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1493 qadd = 0; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1494 else |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1495 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
|
1496 |
200 | 1497 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
|
1498 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
|
1499 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
|
1500 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
|
1501 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
|
1502 } 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
|
1503 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
|
1504 } |
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
|
1505 #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
|
1506 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
|
1507 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
|
1508 #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
|
1509 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
|
1510 } |
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
|
1511 } |
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
|
1512 } |
0 | 1513 |
1514 /* rate control */ | |
1515 | |
1516 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */ | |
1517 #define I_FRAME_SIZE_RATIO 3.0 | |
1518 #define QSCALE_K 20 | |
1519 | |
1520 static void rate_control_init(MpegEncContext *s) | |
1521 { | |
1522 s->wanted_bits = 0; | |
1523 | |
1524 if (s->intra_only) { | |
1525 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate; | |
1526 s->P_frame_bits = s->I_frame_bits; | |
1527 } else { | |
1528 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) / | |
1529 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1))); | |
1530 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO); | |
1531 } | |
1532 | |
1533 #if defined(DEBUG) | |
1534 printf("I_frame_size=%d P_frame_size=%d\n", | |
1535 s->I_frame_bits, s->P_frame_bits); | |
1536 #endif | |
1537 } | |
1538 | |
1539 | |
1540 /* | |
1541 * This heuristic is rather poor, but at least we do not have to | |
1542 * change the qscale at every macroblock. | |
1543 */ | |
1544 static int rate_estimate_qscale(MpegEncContext *s) | |
1545 { | |
187
3f3b14d3a23d
qscale estimate fix, diff extended to 64bit wide. patch by Stephen Davies <steve@daviesfam.org>
arpi_esp
parents:
178
diff
changeset
|
1546 INT64 diff, total_bits = s->total_bits; |
0 | 1547 float q; |
187
3f3b14d3a23d
qscale estimate fix, diff extended to 64bit wide. patch by Stephen Davies <steve@daviesfam.org>
arpi_esp
parents:
178
diff
changeset
|
1548 int qscale, qmin; |
0 | 1549 |
1550 if (s->pict_type == I_TYPE) { | |
1551 s->wanted_bits += s->I_frame_bits; | |
1552 } else { | |
1553 s->wanted_bits += s->P_frame_bits; | |
1554 } | |
1555 diff = s->wanted_bits - total_bits; | |
1556 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width); | |
1557 /* adjust for I frame */ | |
1558 if (s->pict_type == I_TYPE && !s->intra_only) { | |
1559 q /= I_FRAME_SIZE_RATIO; | |
1560 } | |
1561 | |
1562 /* using a too small Q scale leeds to problems in mpeg1 and h263 | |
1563 because AC coefficients are clamped to 255 or 127 */ | |
1564 qmin = 3; | |
1565 if (q < qmin) | |
1566 q = qmin; | |
1567 else if (q > 31) | |
1568 q = 31; | |
1569 qscale = (int)(q + 0.5); | |
1570 #if defined(DEBUG) | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1571 printf("\n%d: total=%0.0f wanted=%0.0f br=%0.1f diff=%d qest=%2.1f\n", |
0 | 1572 s->picture_number, |
64 | 1573 (double)total_bits, |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1574 (double)s->wanted_bits, |
0 | 1575 (float)s->frame_rate / FRAME_RATE_BASE * |
1576 total_bits / s->picture_number, | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
1577 (int)diff, q); |
0 | 1578 #endif |
1579 return qscale; | |
1580 } | |
1581 | |
1582 AVCodec mpeg1video_encoder = { | |
1583 "mpeg1video", | |
1584 CODEC_TYPE_VIDEO, | |
1585 CODEC_ID_MPEG1VIDEO, | |
1586 sizeof(MpegEncContext), | |
1587 MPV_encode_init, | |
1588 MPV_encode_picture, | |
1589 MPV_encode_end, | |
1590 }; | |
1591 | |
1592 AVCodec h263_encoder = { | |
1593 "h263", | |
1594 CODEC_TYPE_VIDEO, | |
1595 CODEC_ID_H263, | |
1596 sizeof(MpegEncContext), | |
1597 MPV_encode_init, | |
1598 MPV_encode_picture, | |
1599 MPV_encode_end, | |
1600 }; | |
1601 | |
1602 AVCodec h263p_encoder = { | |
1603 "h263p", | |
1604 CODEC_TYPE_VIDEO, | |
1605 CODEC_ID_H263P, | |
1606 sizeof(MpegEncContext), | |
1607 MPV_encode_init, | |
1608 MPV_encode_picture, | |
1609 MPV_encode_end, | |
1610 }; | |
1611 | |
1612 AVCodec rv10_encoder = { | |
1613 "rv10", | |
1614 CODEC_TYPE_VIDEO, | |
1615 CODEC_ID_RV10, | |
1616 sizeof(MpegEncContext), | |
1617 MPV_encode_init, | |
1618 MPV_encode_picture, | |
1619 MPV_encode_end, | |
1620 }; | |
1621 | |
1622 AVCodec mjpeg_encoder = { | |
1623 "mjpeg", | |
1624 CODEC_TYPE_VIDEO, | |
1625 CODEC_ID_MJPEG, | |
1626 sizeof(MpegEncContext), | |
1627 MPV_encode_init, | |
1628 MPV_encode_picture, | |
1629 MPV_encode_end, | |
1630 }; | |
1631 | |
71 | 1632 AVCodec mpeg4_encoder = { |
1633 "mpeg4", | |
0 | 1634 CODEC_TYPE_VIDEO, |
71 | 1635 CODEC_ID_MPEG4, |
0 | 1636 sizeof(MpegEncContext), |
1637 MPV_encode_init, | |
1638 MPV_encode_picture, | |
1639 MPV_encode_end, | |
1640 }; | |
1641 | |
1642 AVCodec msmpeg4_encoder = { | |
1643 "msmpeg4", | |
1644 CODEC_TYPE_VIDEO, | |
1645 CODEC_ID_MSMPEG4, | |
1646 sizeof(MpegEncContext), | |
1647 MPV_encode_init, | |
1648 MPV_encode_picture, | |
1649 MPV_encode_end, | |
1650 }; |