Mercurial > libavcodec.hg
annotate mpegvideo.c @ 198:82ba367b1827 libavcodec
rv10 sig11 fix by Michael Niedermayer <michaelni@gmx.at>
| author | arpi_esp |
|---|---|
| date | Sat, 12 Jan 2002 14:36:40 +0000 |
| parents | 21abf1b20016 |
| children | 6ab301aaa652 |
| 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); |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
38 static int dct_quantize(MpegEncContext *s, 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
|
39 static int dct_quantize_mmx(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
|
40 DCTELEM *block, int n, |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
41 int qscale); |
| 0 | 42 #define EDGE_WIDTH 16 |
| 43 | |
| 44 /* enable all paranoid tests for rounding, overflows, etc... */ | |
| 45 //#define PARANOID | |
| 46 | |
| 47 //#define DEBUG | |
| 48 | |
| 49 /* for jpeg fast DCT */ | |
| 50 #define CONST_BITS 14 | |
| 51 | |
| 52 static const unsigned short aanscales[64] = { | |
| 53 /* precomputed values scaled up by 14 bits */ | |
| 54 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
| 55 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
| 56 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
| 57 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
| 58 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
| 59 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
| 60 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
| 61 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
| 62 }; | |
| 63 | |
| 64 static UINT8 h263_chroma_roundtab[16] = { | |
| 65 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
| 66 }; | |
| 67 | |
| 68 /* default motion estimation */ | |
| 69 int motion_estimation_method = ME_LOG; | |
| 70 | |
| 71 /* XXX: should use variable shift ? */ | |
| 72 #define QMAT_SHIFT_MMX 19 | |
| 73 #define QMAT_SHIFT 25 | |
| 74 | |
| 75 static void convert_matrix(int *qmat, const UINT16 *quant_matrix, int qscale) | |
| 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 */ | |
| 82 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ | |
| 83 | |
| 64 | 84 qmat[i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
| 85 (aanscales[i] * qscale * quant_matrix[i])); | |
| 0 | 86 } |
| 87 } else { | |
| 88 for(i=0;i<64;i++) { | |
| 89 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
| 90 So 16 <= qscale * quant_matrix[i] <= 7905 | |
| 91 so (1 << QMAT_SHIFT) / 16 >= qmat[i] >= (1 << QMAT_SHIFT) / 7905 | |
| 92 */ | |
| 93 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); | |
| 94 } | |
| 95 } | |
| 96 } | |
| 97 | |
| 98 /* init common structure for both encoder and decoder */ | |
| 99 int MPV_common_init(MpegEncContext *s) | |
| 100 { | |
| 101 int c_size, i; | |
| 102 UINT8 *pict; | |
| 103 | |
|
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
|
104 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
|
105 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
|
106 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
|
107 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
|
108 |
|
174ef88f619a
use block[] in structure to have it 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 #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
|
110 MPV_common_init_mmx(s); |
| 8 | 111 #endif |
| 0 | 112 s->mb_width = (s->width + 15) / 16; |
| 113 s->mb_height = (s->height + 15) / 16; | |
| 114 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; | |
| 115 | |
| 116 for(i=0;i<3;i++) { | |
| 117 int w, h, shift, pict_start; | |
| 118 | |
| 119 w = s->linesize; | |
| 120 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
| 121 shift = (i == 0) ? 0 : 1; | |
| 122 c_size = (w >> shift) * (h >> shift); | |
| 123 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
| 124 | |
| 125 pict = av_mallocz(c_size); | |
| 126 if (pict == NULL) | |
| 127 goto fail; | |
| 128 s->last_picture_base[i] = pict; | |
| 129 s->last_picture[i] = pict + pict_start; | |
| 130 | |
| 131 pict = av_mallocz(c_size); | |
| 132 if (pict == NULL) | |
| 133 goto fail; | |
| 134 s->next_picture_base[i] = pict; | |
| 135 s->next_picture[i] = pict + pict_start; | |
| 136 | |
| 137 if (s->has_b_frames) { | |
| 138 pict = av_mallocz(c_size); | |
| 139 if (pict == NULL) | |
| 140 goto fail; | |
| 141 s->aux_picture_base[i] = pict; | |
| 142 s->aux_picture[i] = pict + pict_start; | |
| 143 } | |
| 144 } | |
| 145 | |
| 146 if (s->out_format == FMT_H263) { | |
| 147 int size; | |
| 148 /* MV prediction */ | |
| 149 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
| 150 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
| 151 if (s->motion_val == NULL) | |
| 152 goto fail; | |
| 153 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
| 154 } | |
| 155 | |
| 156 if (s->h263_pred) { | |
| 157 int y_size, c_size, i, size; | |
| 158 | |
| 159 /* dc values */ | |
| 160 | |
| 161 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
| 162 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
| 163 size = y_size + 2 * c_size; | |
| 164 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
| 165 if (s->dc_val[0] == NULL) | |
| 166 goto fail; | |
| 167 s->dc_val[1] = s->dc_val[0] + y_size; | |
| 168 s->dc_val[2] = s->dc_val[1] + c_size; | |
| 169 for(i=0;i<size;i++) | |
| 170 s->dc_val[0][i] = 1024; | |
| 171 | |
| 172 /* ac values */ | |
| 173 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
| 174 if (s->ac_val[0] == NULL) | |
| 175 goto fail; | |
| 176 s->ac_val[1] = s->ac_val[0] + y_size; | |
| 177 s->ac_val[2] = s->ac_val[1] + c_size; | |
| 178 | |
| 179 /* cbp values */ | |
| 180 s->coded_block = av_mallocz(y_size); | |
| 181 if (!s->coded_block) | |
| 182 goto fail; | |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
183 |
|
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
184 /* which mb is a intra block */ |
|
196
9ffa69cd4ed6
This fixes segfaults because of uninitialized s->mbintra_table variable - patch by Vladimir Dergachev <volodya@mindspring.com>
arpi_esp
parents:
194
diff
changeset
|
185 s->mbintra_table = av_mallocz(s->mb_width * s->mb_height); |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
186 if (!s->mbintra_table) |
|
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
187 goto fail; |
|
196
9ffa69cd4ed6
This fixes segfaults because of uninitialized s->mbintra_table variable - patch by Vladimir Dergachev <volodya@mindspring.com>
arpi_esp
parents:
194
diff
changeset
|
188 memset(s->mbintra_table, 1, s->mb_width * s->mb_height); |
|
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
|
189 } |
| 0 | 190 /* default structure is frame */ |
| 191 s->picture_structure = PICT_FRAME; | |
| 192 | |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
193 /* init macroblock skip table */ |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
194 if (!s->encoding) { |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
195 s->mbskip_table = av_mallocz(s->mb_width * s->mb_height); |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
196 if (!s->mbskip_table) |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
197 goto fail; |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
198 } |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
199 |
| 0 | 200 s->context_initialized = 1; |
| 201 return 0; | |
| 202 fail: | |
| 203 if (s->motion_val) | |
| 204 free(s->motion_val); | |
| 205 if (s->dc_val[0]) | |
| 206 free(s->dc_val[0]); | |
| 207 if (s->ac_val[0]) | |
| 208 free(s->ac_val[0]); | |
| 209 if (s->coded_block) | |
| 210 free(s->coded_block); | |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
211 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
|
212 free(s->mbintra_table); |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
213 if (s->mbskip_table) |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
214 free(s->mbskip_table); |
| 0 | 215 for(i=0;i<3;i++) { |
| 216 if (s->last_picture_base[i]) | |
| 217 free(s->last_picture_base[i]); | |
| 218 if (s->next_picture_base[i]) | |
| 219 free(s->next_picture_base[i]); | |
| 220 if (s->aux_picture_base[i]) | |
| 221 free(s->aux_picture_base[i]); | |
| 222 } | |
| 223 return -1; | |
| 224 } | |
| 225 | |
| 226 /* init common structure for both encoder and decoder */ | |
| 227 void MPV_common_end(MpegEncContext *s) | |
| 228 { | |
| 229 int i; | |
| 230 | |
| 231 if (s->motion_val) | |
| 232 free(s->motion_val); | |
| 233 if (s->h263_pred) { | |
| 234 free(s->dc_val[0]); | |
| 235 free(s->ac_val[0]); | |
| 236 free(s->coded_block); | |
|
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
|
237 free(s->mbintra_table); |
| 0 | 238 } |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
239 if (s->mbskip_table) |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
240 free(s->mbskip_table); |
| 0 | 241 for(i=0;i<3;i++) { |
| 242 free(s->last_picture_base[i]); | |
| 243 free(s->next_picture_base[i]); | |
| 244 if (s->has_b_frames) | |
| 245 free(s->aux_picture_base[i]); | |
| 246 } | |
| 247 s->context_initialized = 0; | |
| 248 } | |
| 249 | |
| 250 /* init video encoder */ | |
| 251 int MPV_encode_init(AVCodecContext *avctx) | |
| 252 { | |
| 253 MpegEncContext *s = avctx->priv_data; | |
| 60 | 254 int i; |
| 0 | 255 |
| 256 s->bit_rate = avctx->bit_rate; | |
| 257 s->frame_rate = avctx->frame_rate; | |
| 258 s->width = avctx->width; | |
| 259 s->height = avctx->height; | |
| 260 s->gop_size = avctx->gop_size; | |
| 162 | 261 s->rtp_mode = avctx->rtp_mode; |
| 262 s->rtp_payload_size = avctx->rtp_payload_size; | |
|
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
263 s->avctx = avctx; |
| 162 | 264 |
| 0 | 265 if (s->gop_size <= 1) { |
| 266 s->intra_only = 1; | |
| 267 s->gop_size = 12; | |
| 268 } else { | |
| 269 s->intra_only = 0; | |
| 270 } | |
| 271 s->full_search = motion_estimation_method; | |
| 272 | |
| 273 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE); | |
| 274 | |
| 275 switch(avctx->codec->id) { | |
| 276 case CODEC_ID_MPEG1VIDEO: | |
| 277 s->out_format = FMT_MPEG1; | |
| 278 break; | |
| 279 case CODEC_ID_MJPEG: | |
| 280 s->out_format = FMT_MJPEG; | |
| 281 s->intra_only = 1; /* force intra only for jpeg */ | |
| 282 if (mjpeg_init(s) < 0) | |
| 283 return -1; | |
| 284 break; | |
| 285 case CODEC_ID_H263: | |
| 178 | 286 if (h263_get_picture_format(s->width, s->height) == 7){ |
| 287 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); | |
| 0 | 288 return -1; |
| 178 | 289 } |
| 0 | 290 s->out_format = FMT_H263; |
| 291 break; | |
| 292 case CODEC_ID_H263P: | |
| 293 s->out_format = FMT_H263; | |
| 162 | 294 s->rtp_mode = 1; |
| 295 s->rtp_payload_size = 1200; | |
| 0 | 296 s->h263_plus = 1; |
| 78 | 297 s->unrestricted_mv = 1; |
|
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
298 |
|
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
299 /* These are just to be sure */ |
|
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
300 s->umvplus = 0; |
|
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
301 s->umvplus_dec = 0; |
| 0 | 302 break; |
| 303 case CODEC_ID_RV10: | |
| 304 s->out_format = FMT_H263; | |
| 305 s->h263_rv10 = 1; | |
| 306 break; | |
| 71 | 307 case CODEC_ID_MPEG4: |
| 0 | 308 s->out_format = FMT_H263; |
| 309 s->h263_pred = 1; | |
| 310 s->unrestricted_mv = 1; | |
| 311 break; | |
| 312 case CODEC_ID_MSMPEG4: | |
| 313 s->out_format = FMT_H263; | |
| 314 s->h263_msmpeg4 = 1; | |
| 315 s->h263_pred = 1; | |
| 316 s->unrestricted_mv = 1; | |
| 317 break; | |
| 318 default: | |
| 319 return -1; | |
| 320 } | |
| 321 | |
| 322 if (s->out_format == FMT_H263) | |
| 323 h263_encode_init_vlc(s); | |
| 324 | |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
325 s->encoding = 1; |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
326 |
| 0 | 327 /* init */ |
| 328 if (MPV_common_init(s) < 0) | |
| 329 return -1; | |
| 330 | |
| 60 | 331 /* init default q matrix */ |
| 332 for(i=0;i<64;i++) { | |
| 333 s->intra_matrix[i] = default_intra_matrix[i]; | |
| 334 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
| 335 } | |
| 336 | |
| 0 | 337 /* rate control init */ |
| 338 rate_control_init(s); | |
| 339 | |
| 340 s->picture_number = 0; | |
| 341 s->fake_picture_number = 0; | |
| 342 /* motion detector init */ | |
| 343 s->f_code = 1; | |
| 344 | |
| 345 return 0; | |
| 346 } | |
| 347 | |
| 348 int MPV_encode_end(AVCodecContext *avctx) | |
| 349 { | |
| 350 MpegEncContext *s = avctx->priv_data; | |
| 351 | |
| 352 #ifdef STATS | |
| 353 print_stats(); | |
| 354 #endif | |
| 355 MPV_common_end(s); | |
| 356 if (s->out_format == FMT_MJPEG) | |
| 357 mjpeg_close(s); | |
| 358 return 0; | |
| 359 } | |
| 360 | |
| 361 /* draw the edges of width 'w' of an image of size width, height */ | |
| 362 static void draw_edges(UINT8 *buf, int wrap, int width, int height, int w) | |
| 363 { | |
| 364 UINT8 *ptr, *last_line; | |
| 365 int i; | |
| 366 | |
| 367 last_line = buf + (height - 1) * wrap; | |
| 368 for(i=0;i<w;i++) { | |
| 369 /* top and bottom */ | |
| 370 memcpy(buf - (i + 1) * wrap, buf, width); | |
| 371 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
| 372 } | |
| 373 /* left and right */ | |
| 374 ptr = buf; | |
| 375 for(i=0;i<height;i++) { | |
| 376 memset(ptr - w, ptr[0], w); | |
| 377 memset(ptr + width, ptr[width-1], w); | |
| 378 ptr += wrap; | |
| 379 } | |
| 380 /* corners */ | |
| 381 for(i=0;i<w;i++) { | |
| 382 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
| 383 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
| 384 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
| 385 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
| 386 } | |
| 387 } | |
| 388 | |
| 389 /* generic function for encode/decode called before a frame is coded/decoded */ | |
| 390 void MPV_frame_start(MpegEncContext *s) | |
| 391 { | |
| 392 int i; | |
| 393 UINT8 *tmp; | |
| 394 | |
|
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
395 s->mb_skiped = 0; |
| 0 | 396 if (s->pict_type == B_TYPE) { |
| 397 for(i=0;i<3;i++) { | |
| 398 s->current_picture[i] = s->aux_picture[i]; | |
| 399 } | |
| 400 } else { | |
| 401 for(i=0;i<3;i++) { | |
| 402 /* swap next and last */ | |
| 403 tmp = s->last_picture[i]; | |
| 404 s->last_picture[i] = s->next_picture[i]; | |
| 405 s->next_picture[i] = tmp; | |
| 406 s->current_picture[i] = tmp; | |
| 407 } | |
| 408 } | |
| 409 } | |
|
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
|
410 |
| 0 | 411 /* generic function for encode/decode called after a frame has been coded/decoded */ |
| 412 void MPV_frame_end(MpegEncContext *s) | |
| 413 { | |
| 414 /* draw edge for correct motion prediction if outside */ | |
| 415 if (s->pict_type != B_TYPE) { | |
|
198
82ba367b1827
rv10 sig11 fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
197
diff
changeset
|
416 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
|
417 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
|
418 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
|
419 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
|
420 }else{ |
|
42552c1cf360
fix different UMV handling for mpeg4 vs. h263 - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
187
diff
changeset
|
421 /* OpenDivx, but i dunno how to distinguish it from mpeg4 */ |
| 0 | 422 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
| 423 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
| 424 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
|
425 } |
| 0 | 426 } |
| 427 } | |
| 428 | |
| 429 int MPV_encode_picture(AVCodecContext *avctx, | |
| 430 unsigned char *buf, int buf_size, void *data) | |
| 431 { | |
| 432 MpegEncContext *s = avctx->priv_data; | |
| 433 AVPicture *pict = data; | |
| 434 int i, j; | |
| 435 | |
| 436 if (s->fixed_qscale) | |
| 437 s->qscale = avctx->quality; | |
| 438 | |
| 439 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
| 440 | |
| 441 if (!s->intra_only) { | |
| 442 /* first picture of GOP is intra */ | |
| 443 if ((s->picture_number % s->gop_size) == 0) | |
| 444 s->pict_type = I_TYPE; | |
| 445 else | |
| 446 s->pict_type = P_TYPE; | |
| 447 } else { | |
| 448 s->pict_type = I_TYPE; | |
| 449 } | |
| 450 avctx->key_frame = (s->pict_type == I_TYPE); | |
| 451 | |
| 452 MPV_frame_start(s); | |
| 453 | |
| 454 for(i=0;i<3;i++) { | |
| 455 UINT8 *src = pict->data[i]; | |
| 456 UINT8 *dest = s->current_picture[i]; | |
| 457 int src_wrap = pict->linesize[i]; | |
| 458 int dest_wrap = s->linesize; | |
| 459 int w = s->width; | |
| 460 int h = s->height; | |
| 461 | |
| 462 if (i >= 1) { | |
| 463 dest_wrap >>= 1; | |
| 464 w >>= 1; | |
| 465 h >>= 1; | |
| 466 } | |
| 467 | |
| 468 for(j=0;j<h;j++) { | |
| 469 memcpy(dest, src, w); | |
| 470 dest += dest_wrap; | |
| 471 src += src_wrap; | |
| 472 } | |
| 473 s->new_picture[i] = s->current_picture[i]; | |
| 474 } | |
| 475 | |
| 476 encode_picture(s, s->picture_number); | |
| 477 | |
| 478 MPV_frame_end(s); | |
| 479 s->picture_number++; | |
| 480 | |
| 481 if (s->out_format == FMT_MJPEG) | |
| 482 mjpeg_picture_trailer(s); | |
| 483 | |
| 484 flush_put_bits(&s->pb); | |
| 485 s->total_bits += (s->pb.buf_ptr - s->pb.buf) * 8; | |
| 486 avctx->quality = s->qscale; | |
| 487 return s->pb.buf_ptr - s->pb.buf; | |
| 488 } | |
| 489 | |
| 490 static inline int clip(int a, int amin, int amax) | |
| 491 { | |
| 492 if (a < amin) | |
| 493 return amin; | |
| 494 else if (a > amax) | |
| 495 return amax; | |
| 496 else | |
| 497 return a; | |
| 498 } | |
| 499 | |
| 500 /* apply one mpeg motion vector to the three components */ | |
| 501 static inline void mpeg_motion(MpegEncContext *s, | |
| 502 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
| 503 int dest_offset, | |
| 504 UINT8 **ref_picture, int src_offset, | |
| 505 int field_based, op_pixels_func *pix_op, | |
| 506 int motion_x, int motion_y, int h) | |
| 507 { | |
| 508 UINT8 *ptr; | |
| 509 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
| 510 | |
| 511 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
| 512 src_x = s->mb_x * 16 + (motion_x >> 1); | |
| 513 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
| 514 | |
| 515 /* WARNING: do no forget half pels */ | |
| 516 height = s->height >> field_based; | |
| 517 src_x = clip(src_x, -16, s->width); | |
| 518 if (src_x == s->width) | |
| 519 dxy &= ~1; | |
| 520 src_y = clip(src_y, -16, height); | |
| 521 if (src_y == height) | |
| 522 dxy &= ~2; | |
| 523 linesize = s->linesize << field_based; | |
| 524 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
| 525 dest_y += dest_offset; | |
| 526 pix_op[dxy](dest_y, ptr, linesize, h); | |
| 527 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
| 528 | |
| 529 if (s->out_format == FMT_H263) { | |
| 530 dxy = 0; | |
| 531 if ((motion_x & 3) != 0) | |
| 532 dxy |= 1; | |
| 533 if ((motion_y & 3) != 0) | |
| 534 dxy |= 2; | |
| 535 mx = motion_x >> 2; | |
| 536 my = motion_y >> 2; | |
| 537 } else { | |
| 538 mx = motion_x / 2; | |
| 539 my = motion_y / 2; | |
| 540 dxy = ((my & 1) << 1) | (mx & 1); | |
| 541 mx >>= 1; | |
| 542 my >>= 1; | |
| 543 } | |
| 544 | |
| 545 src_x = s->mb_x * 8 + mx; | |
| 546 src_y = s->mb_y * (8 >> field_based) + my; | |
| 547 src_x = clip(src_x, -8, s->width >> 1); | |
| 548 if (src_x == (s->width >> 1)) | |
| 549 dxy &= ~1; | |
| 550 src_y = clip(src_y, -8, height >> 1); | |
| 551 if (src_y == (height >> 1)) | |
| 552 dxy &= ~2; | |
| 553 | |
| 554 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
| 555 ptr = ref_picture[1] + offset; | |
| 556 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
| 557 ptr = ref_picture[2] + offset; | |
| 558 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
| 559 } | |
| 560 | |
| 561 static inline void MPV_motion(MpegEncContext *s, | |
| 562 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
| 563 int dir, UINT8 **ref_picture, | |
| 564 op_pixels_func *pix_op) | |
| 565 { | |
| 566 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
| 567 int mb_x, mb_y, i; | |
| 568 UINT8 *ptr, *dest; | |
| 569 | |
| 570 mb_x = s->mb_x; | |
| 571 mb_y = s->mb_y; | |
| 572 | |
| 573 switch(s->mv_type) { | |
| 574 case MV_TYPE_16X16: | |
| 575 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 576 ref_picture, 0, | |
| 577 0, pix_op, | |
| 578 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
| 579 break; | |
| 580 case MV_TYPE_8X8: | |
| 581 for(i=0;i<4;i++) { | |
| 582 motion_x = s->mv[dir][i][0]; | |
| 583 motion_y = s->mv[dir][i][1]; | |
| 584 | |
| 585 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
| 586 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
| 587 src_y = mb_y * 16 + (motion_y >> 1) + ((i >> 1) & 1) * 8; | |
| 588 | |
| 589 /* WARNING: do no forget half pels */ | |
| 590 src_x = clip(src_x, -16, s->width); | |
| 591 if (src_x == s->width) | |
| 592 dxy &= ~1; | |
| 593 src_y = clip(src_y, -16, s->height); | |
| 594 if (src_y == s->height) | |
| 595 dxy &= ~2; | |
| 596 | |
| 597 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
| 598 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
| 599 pix_op[dxy](dest, ptr, s->linesize, 8); | |
| 600 } | |
| 601 /* In case of 8X8, we construct a single chroma motion vector | |
| 602 with a special rounding */ | |
| 603 mx = 0; | |
| 604 my = 0; | |
| 605 for(i=0;i<4;i++) { | |
| 606 mx += s->mv[dir][i][0]; | |
| 607 my += s->mv[dir][i][1]; | |
| 608 } | |
| 609 if (mx >= 0) | |
| 610 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
| 611 else { | |
| 612 mx = -mx; | |
| 613 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
| 614 } | |
| 615 if (my >= 0) | |
| 616 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
| 617 else { | |
| 618 my = -my; | |
| 619 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
| 620 } | |
| 621 dxy = ((my & 1) << 1) | (mx & 1); | |
| 622 mx >>= 1; | |
| 623 my >>= 1; | |
| 624 | |
| 625 src_x = mb_x * 8 + mx; | |
| 626 src_y = mb_y * 8 + my; | |
| 627 src_x = clip(src_x, -8, s->width/2); | |
| 628 if (src_x == s->width/2) | |
| 629 dxy &= ~1; | |
| 630 src_y = clip(src_y, -8, s->height/2); | |
| 631 if (src_y == s->height/2) | |
| 632 dxy &= ~2; | |
| 633 | |
| 634 offset = (src_y * (s->linesize >> 1)) + src_x; | |
| 635 ptr = ref_picture[1] + offset; | |
| 636 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
| 637 ptr = ref_picture[2] + offset; | |
| 638 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
| 639 break; | |
| 640 case MV_TYPE_FIELD: | |
| 641 if (s->picture_structure == PICT_FRAME) { | |
| 642 /* top field */ | |
| 643 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 644 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
| 645 1, pix_op, | |
| 646 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
| 647 /* bottom field */ | |
| 648 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
| 649 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
| 650 1, pix_op, | |
| 651 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
| 652 } else { | |
| 653 | |
| 654 | |
| 655 } | |
| 656 break; | |
| 657 } | |
| 658 } | |
| 659 | |
| 660 | |
| 661 /* put block[] to dest[] */ | |
| 662 static inline void put_dct(MpegEncContext *s, | |
| 663 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
| 664 { | |
| 665 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
|
666 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
|
667 ff_idct (block); |
| 0 | 668 put_pixels_clamped(block, dest, line_size); |
| 669 } | |
| 670 | |
| 671 /* add block[] to dest[] */ | |
| 672 static inline void add_dct(MpegEncContext *s, | |
| 673 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
| 674 { | |
| 675 if (s->block_last_index[i] >= 0) { | |
| 676 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
|
677 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
|
678 ff_idct (block); |
| 0 | 679 add_pixels_clamped(block, dest, line_size); |
| 680 } | |
| 681 } | |
| 682 | |
| 683 /* generic function called after a macroblock has been parsed by the | |
| 684 decoder or after it has been encoded by the encoder. | |
| 685 | |
| 686 Important variables used: | |
| 687 s->mb_intra : true if intra macroblock | |
| 688 s->mv_dir : motion vector direction | |
| 689 s->mv_type : motion vector type | |
| 690 s->mv : motion vector | |
| 691 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
| 692 */ | |
| 693 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
| 694 { | |
| 695 int mb_x, mb_y, motion_x, motion_y; | |
| 696 int dct_linesize, dct_offset; | |
| 697 op_pixels_func *op_pix; | |
| 698 | |
| 699 mb_x = s->mb_x; | |
| 700 mb_y = s->mb_y; | |
| 701 | |
|
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
702 #ifdef FF_POSTPROCESS |
|
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
703 quant_store[mb_y][mb_x]=s->qscale; |
|
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
704 //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
|
705 #endif |
|
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
706 |
| 0 | 707 /* update DC predictors for P macroblocks */ |
| 708 if (!s->mb_intra) { | |
| 709 if (s->h263_pred) { | |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
710 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
|
711 { |
| 0 | 712 int wrap, x, y, v; |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
713 s->mbintra_table[mb_x + mb_y*s->mb_width]=0; |
|
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
714 |
| 0 | 715 wrap = 2 * s->mb_width + 2; |
| 716 v = 1024; | |
| 717 x = 2 * mb_x + 1; | |
| 718 y = 2 * mb_y + 1; | |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
719 |
| 0 | 720 s->dc_val[0][(x) + (y) * wrap] = v; |
| 721 s->dc_val[0][(x + 1) + (y) * wrap] = v; | |
| 722 s->dc_val[0][(x) + (y + 1) * wrap] = v; | |
| 723 s->dc_val[0][(x + 1) + (y + 1) * wrap] = v; | |
| 724 /* ac pred */ | |
| 725 memset(s->ac_val[0][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
| 726 memset(s->ac_val[0][(x + 1) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
| 727 memset(s->ac_val[0][(x) + (y + 1) * wrap], 0, 16 * sizeof(INT16)); | |
| 728 memset(s->ac_val[0][(x + 1) + (y + 1) * wrap], 0, 16 * sizeof(INT16)); | |
| 729 if (s->h263_msmpeg4) { | |
| 730 s->coded_block[(x) + (y) * wrap] = 0; | |
| 731 s->coded_block[(x + 1) + (y) * wrap] = 0; | |
| 732 s->coded_block[(x) + (y + 1) * wrap] = 0; | |
| 733 s->coded_block[(x + 1) + (y + 1) * wrap] = 0; | |
| 734 } | |
| 735 /* chroma */ | |
| 736 wrap = s->mb_width + 2; | |
| 737 x = mb_x + 1; | |
| 738 y = mb_y + 1; | |
| 739 s->dc_val[1][(x) + (y) * wrap] = v; | |
| 740 s->dc_val[2][(x) + (y) * wrap] = v; | |
| 741 /* ac pred */ | |
| 742 memset(s->ac_val[1][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
| 743 memset(s->ac_val[2][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
744 } |
| 0 | 745 } else { |
| 746 s->last_dc[0] = 128 << s->intra_dc_precision; | |
| 747 s->last_dc[1] = 128 << s->intra_dc_precision; | |
| 748 s->last_dc[2] = 128 << s->intra_dc_precision; | |
| 749 } | |
| 750 } | |
|
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
|
751 else if (s->h263_pred) |
|
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
752 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
|
753 |
| 0 | 754 /* update motion predictor */ |
| 755 if (s->out_format == FMT_H263) { | |
| 756 int x, y, wrap; | |
| 757 | |
| 758 x = 2 * mb_x + 1; | |
| 759 y = 2 * mb_y + 1; | |
| 760 wrap = 2 * s->mb_width + 2; | |
| 761 if (s->mb_intra) { | |
| 762 motion_x = 0; | |
| 763 motion_y = 0; | |
| 764 goto motion_init; | |
| 765 } else if (s->mv_type == MV_TYPE_16X16) { | |
| 766 motion_x = s->mv[0][0][0]; | |
| 767 motion_y = s->mv[0][0][1]; | |
| 768 motion_init: | |
| 769 /* no update if 8X8 because it has been done during parsing */ | |
| 770 s->motion_val[(x) + (y) * wrap][0] = motion_x; | |
| 771 s->motion_val[(x) + (y) * wrap][1] = motion_y; | |
| 772 s->motion_val[(x + 1) + (y) * wrap][0] = motion_x; | |
| 773 s->motion_val[(x + 1) + (y) * wrap][1] = motion_y; | |
| 774 s->motion_val[(x) + (y + 1) * wrap][0] = motion_x; | |
| 775 s->motion_val[(x) + (y + 1) * wrap][1] = motion_y; | |
| 776 s->motion_val[(x + 1) + (y + 1) * wrap][0] = motion_x; | |
| 777 s->motion_val[(x + 1) + (y + 1) * wrap][1] = motion_y; | |
| 778 } | |
| 779 } | |
| 780 | |
| 781 if (!s->intra_only) { | |
| 782 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
|
783 UINT8 *mbskip_ptr; |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
784 |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
785 /* 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
|
786 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
|
787 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
|
788 if (s->mb_skiped) { |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
789 s->mb_skiped = 0; |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
790 /* 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
|
791 if (*mbskip_ptr != 0) |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
792 goto the_end; |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
793 *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
|
794 } else { |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
795 *mbskip_ptr = 0; /* not skipped */ |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
796 } |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
797 } |
| 0 | 798 |
| 799 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
| 800 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
| 801 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
| 802 | |
| 803 if (s->interlaced_dct) { | |
| 804 dct_linesize = s->linesize * 2; | |
| 805 dct_offset = s->linesize; | |
| 806 } else { | |
| 807 dct_linesize = s->linesize; | |
| 808 dct_offset = s->linesize * 8; | |
| 809 } | |
| 810 | |
| 811 if (!s->mb_intra) { | |
| 812 /* motion handling */ | |
| 813 if (!s->no_rounding) | |
| 814 op_pix = put_pixels_tab; | |
| 815 else | |
| 816 op_pix = put_no_rnd_pixels_tab; | |
| 817 | |
| 818 if (s->mv_dir & MV_DIR_FORWARD) { | |
| 819 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix); | |
| 820 if (!s->no_rounding) | |
| 821 op_pix = avg_pixels_tab; | |
| 822 else | |
| 823 op_pix = avg_no_rnd_pixels_tab; | |
| 824 } | |
| 825 if (s->mv_dir & MV_DIR_BACKWARD) { | |
| 826 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix); | |
| 827 } | |
| 828 | |
| 829 /* add dct residue */ | |
| 830 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
| 831 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
| 832 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
| 833 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
| 834 | |
| 57 | 835 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
| 836 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
| 0 | 837 } else { |
| 838 /* dct only in intra block */ | |
| 839 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
| 840 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
| 841 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
| 842 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
| 843 | |
| 57 | 844 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
| 845 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
| 0 | 846 } |
| 847 } | |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
848 the_end: |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
849 emms_c(); |
| 0 | 850 } |
| 851 | |
| 852 static void encode_picture(MpegEncContext *s, int picture_number) | |
| 853 { | |
| 162 | 854 int mb_x, mb_y, wrap, last_gob; |
| 0 | 855 UINT8 *ptr; |
| 856 int i, motion_x, motion_y; | |
| 857 | |
| 858 s->picture_number = picture_number; | |
| 859 if (!s->fixed_qscale) | |
| 860 s->qscale = rate_estimate_qscale(s); | |
| 861 | |
| 862 /* precompute matrix */ | |
| 863 if (s->out_format == FMT_MJPEG) { | |
| 864 /* for mjpeg, we do include qscale in the matrix */ | |
| 865 s->intra_matrix[0] = default_intra_matrix[0]; | |
| 866 for(i=1;i<64;i++) | |
| 867 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
| 868 convert_matrix(s->q_intra_matrix, s->intra_matrix, 8); | |
| 869 } else { | |
| 870 convert_matrix(s->q_intra_matrix, s->intra_matrix, s->qscale); | |
| 871 convert_matrix(s->q_non_intra_matrix, s->non_intra_matrix, s->qscale); | |
| 872 } | |
| 873 | |
| 874 switch(s->out_format) { | |
| 875 case FMT_MJPEG: | |
| 876 mjpeg_picture_header(s); | |
| 877 break; | |
| 878 case FMT_H263: | |
| 879 if (s->h263_msmpeg4) | |
| 880 msmpeg4_encode_picture_header(s, picture_number); | |
| 881 else if (s->h263_pred) | |
| 882 mpeg4_encode_picture_header(s, picture_number); | |
| 883 else if (s->h263_rv10) | |
| 884 rv10_encode_picture_header(s, picture_number); | |
| 885 else | |
| 886 h263_encode_picture_header(s, picture_number); | |
| 887 break; | |
| 888 case FMT_MPEG1: | |
| 889 mpeg1_encode_picture_header(s, picture_number); | |
| 890 break; | |
| 891 } | |
| 892 | |
| 893 /* init last dc values */ | |
| 894 /* note: quant matrix value (8) is implied here */ | |
| 895 s->last_dc[0] = 128; | |
| 896 s->last_dc[1] = 128; | |
| 897 s->last_dc[2] = 128; | |
| 898 s->mb_incr = 1; | |
| 899 s->last_mv[0][0][0] = 0; | |
| 900 s->last_mv[0][0][1] = 0; | |
| 901 s->mv_type = MV_TYPE_16X16; | |
| 902 s->mv_dir = MV_DIR_FORWARD; | |
| 903 | |
| 162 | 904 /* Get the GOB height based on picture height */ |
| 905 if (s->out_format == FMT_H263 && s->h263_plus) { | |
| 906 if (s->height <= 400) | |
| 907 s->gob_index = 1; | |
| 908 else if (s->height <= 800) | |
| 909 s->gob_index = 2; | |
| 910 else | |
| 911 s->gob_index = 4; | |
| 912 } | |
| 913 | |
| 0 | 914 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
| 162 | 915 /* Put GOB header based on RTP MTU */ |
| 916 if (!mb_y) { | |
| 917 s->ptr_lastgob = s->pb.buf_ptr; | |
| 918 s->ptr_last_mb_line = s->pb.buf_ptr; | |
| 919 } else if (s->out_format == FMT_H263 && s->h263_plus) { | |
| 920 last_gob = h263_encode_gob_header(s, mb_y); | |
| 921 if (last_gob) { | |
| 922 //fprintf(stderr,"\nLast GOB size: %d", last_gob); | |
| 923 s->first_gob_line = 1; | |
| 924 } else | |
| 925 s->first_gob_line = 0; | |
| 926 } | |
| 0 | 927 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
| 928 | |
| 929 s->mb_x = mb_x; | |
| 930 s->mb_y = mb_y; | |
| 931 | |
| 932 /* compute motion vector and macro block type (intra or non intra) */ | |
| 933 motion_x = 0; | |
| 934 motion_y = 0; | |
| 935 if (s->pict_type == P_TYPE) { | |
| 936 s->mb_intra = estimate_motion(s, mb_x, mb_y, | |
| 937 &motion_x, | |
| 938 &motion_y); | |
| 939 } else { | |
| 940 s->mb_intra = 1; | |
| 941 } | |
| 942 | |
| 943 /* get the pixels */ | |
| 944 wrap = s->linesize; | |
| 945 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
|
946 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
|
947 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
|
948 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
|
949 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); |
| 0 | 950 wrap = s->linesize >> 1; |
| 951 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
|
952 get_pixels(s->block[4], ptr, wrap); |
| 0 | 953 |
| 954 wrap = s->linesize >> 1; | |
| 955 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
|
956 get_pixels(s->block[5], ptr, wrap); |
| 0 | 957 |
| 958 /* subtract previous frame if non intra */ | |
| 959 if (!s->mb_intra) { | |
| 960 int dxy, offset, mx, my; | |
| 961 | |
| 962 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
| 963 ptr = s->last_picture[0] + | |
| 964 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) + | |
| 965 (mb_x * 16 + (motion_x >> 1)); | |
| 966 | |
|
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
|
967 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
|
968 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
|
969 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
|
970 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy); |
| 0 | 971 |
| 972 if (s->out_format == FMT_H263) { | |
| 973 /* special rounding for h263 */ | |
| 974 dxy = 0; | |
| 975 if ((motion_x & 3) != 0) | |
| 976 dxy |= 1; | |
| 977 if ((motion_y & 3) != 0) | |
| 978 dxy |= 2; | |
| 979 mx = motion_x >> 2; | |
| 980 my = motion_y >> 2; | |
| 981 } else { | |
| 982 mx = motion_x / 2; | |
| 983 my = motion_y / 2; | |
| 984 dxy = ((my & 1) << 1) | (mx & 1); | |
| 985 mx >>= 1; | |
| 986 my >>= 1; | |
| 987 } | |
| 988 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx); | |
| 989 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
|
990 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); |
| 0 | 991 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
|
992 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); |
| 0 | 993 } |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
994 emms_c(); |
| 0 | 995 |
| 996 /* DCT & quantize */ | |
| 997 if (s->h263_msmpeg4) { | |
| 998 msmpeg4_dc_scale(s); | |
| 999 } else if (s->h263_pred) { | |
| 1000 h263_dc_scale(s); | |
| 1001 } else { | |
| 1002 /* default quantization values */ | |
| 1003 s->y_dc_scale = 8; | |
| 1004 s->c_dc_scale = 8; | |
| 1005 } | |
| 1006 | |
| 1007 for(i=0;i<6;i++) { | |
| 1008 int last_index; | |
| 1009 if (av_fdct == jpeg_fdct_ifast) | |
|
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
|
1010 last_index = dct_quantize(s, s->block[i], i, s->qscale); |
| 0 | 1011 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
|
1012 last_index = dct_quantize_mmx(s, s->block[i], i, s->qscale); |
| 0 | 1013 s->block_last_index[i] = last_index; |
| 1014 } | |
| 1015 | |
| 1016 /* huffman encode */ | |
| 1017 switch(s->out_format) { | |
| 1018 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
|
1019 mpeg1_encode_mb(s, s->block, motion_x, motion_y); |
| 0 | 1020 break; |
| 1021 case FMT_H263: | |
| 1022 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
|
1023 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); |
| 0 | 1024 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
|
1025 h263_encode_mb(s, s->block, motion_x, motion_y); |
| 0 | 1026 break; |
| 1027 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
|
1028 mjpeg_encode_mb(s, s->block); |
| 0 | 1029 break; |
| 1030 } | |
| 1031 | |
| 1032 /* decompress blocks so that we keep the state of the decoder */ | |
| 1033 s->mv[0][0][0] = motion_x; | |
| 1034 s->mv[0][0][1] = motion_y; | |
| 1035 | |
|
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
|
1036 MPV_decode_mb(s, s->block); |
| 0 | 1037 } |
| 162 | 1038 /* Obtain average MB line size for RTP */ |
| 1039 if (!mb_y) | |
| 1040 s->mb_line_avgsize = s->pb.buf_ptr - s->ptr_last_mb_line; | |
| 1041 else | |
| 1042 s->mb_line_avgsize = (s->mb_line_avgsize + s->pb.buf_ptr - s->ptr_last_mb_line) >> 1; | |
| 1043 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
| 1044 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
| 1045 s->ptr_last_mb_line = s->pb.buf_ptr; | |
| 0 | 1046 } |
| 162 | 1047 //if (s->gob_number) |
| 1048 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
| 0 | 1049 } |
| 1050 | |
| 1051 static int dct_quantize(MpegEncContext *s, | |
| 1052 DCTELEM *block, int n, | |
| 1053 int qscale) | |
| 1054 { | |
| 1055 int i, j, level, last_non_zero, q; | |
| 1056 const int *qmat; | |
| 1057 | |
| 1058 av_fdct (block); | |
| 1059 | |
| 64 | 1060 /* we need this permutation so that we correct the IDCT |
| 1061 permutation. will be moved into DCT code */ | |
| 1062 block_permute(block); | |
| 1063 | |
| 0 | 1064 if (s->mb_intra) { |
| 1065 if (n < 4) | |
| 1066 q = s->y_dc_scale; | |
| 1067 else | |
| 1068 q = s->c_dc_scale; | |
| 1069 q = q << 3; | |
| 1070 | |
| 1071 /* note: block[0] is assumed to be positive */ | |
| 1072 block[0] = (block[0] + (q >> 1)) / q; | |
| 1073 i = 1; | |
| 1074 last_non_zero = 0; | |
| 1075 if (s->out_format == FMT_H263) { | |
| 1076 qmat = s->q_non_intra_matrix; | |
| 1077 } else { | |
| 1078 qmat = s->q_intra_matrix; | |
| 1079 } | |
| 1080 } else { | |
| 1081 i = 0; | |
| 1082 last_non_zero = -1; | |
| 1083 qmat = s->q_non_intra_matrix; | |
| 1084 } | |
| 1085 | |
| 1086 for(;i<64;i++) { | |
| 1087 j = zigzag_direct[i]; | |
| 1088 level = block[j]; | |
| 1089 level = level * qmat[j]; | |
| 1090 #ifdef PARANOID | |
| 1091 { | |
| 1092 static int count = 0; | |
| 1093 int level1, level2, qmat1; | |
| 1094 double val; | |
| 1095 if (qmat == s->q_non_intra_matrix) { | |
| 1096 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
| 1097 } else { | |
| 1098 qmat1 = default_intra_matrix[j] * s->qscale; | |
| 1099 } | |
| 1100 if (av_fdct != jpeg_fdct_ifast) | |
| 1101 val = ((double)block[j] * 8.0) / (double)qmat1; | |
| 1102 else | |
| 1103 val = ((double)block[j] * 8.0 * 2048.0) / | |
| 1104 ((double)qmat1 * aanscales[j]); | |
| 1105 level1 = (int)val; | |
| 1106 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
| 1107 if (level1 != level2) { | |
| 1108 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
| 1109 count, level2, level1, block[j], qmat1, qmat[j], | |
| 1110 val); | |
| 1111 count++; | |
| 1112 } | |
| 1113 | |
| 1114 } | |
| 1115 #endif | |
| 1116 /* XXX: slight error for the low range. Test should be equivalent to | |
| 1117 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
| 1118 (QMAT_SHIFT - 3))) | |
| 1119 */ | |
| 1120 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
| 1121 level) { | |
| 1122 level = level / (1 << (QMAT_SHIFT - 3)); | |
| 1123 /* XXX: currently, this code is not optimal. the range should be: | |
| 1124 mpeg1: -255..255 | |
| 1125 mpeg2: -2048..2047 | |
| 1126 h263: -128..127 | |
| 1127 mpeg4: -2048..2047 | |
| 1128 */ | |
| 1129 if (level > 127) | |
| 1130 level = 127; | |
| 1131 else if (level < -128) | |
| 1132 level = -128; | |
| 1133 block[j] = level; | |
| 1134 last_non_zero = i; | |
| 1135 } else { | |
| 1136 block[j] = 0; | |
| 1137 } | |
| 1138 } | |
| 1139 return last_non_zero; | |
| 1140 } | |
| 1141 | |
| 1142 static int dct_quantize_mmx(MpegEncContext *s, | |
| 1143 DCTELEM *block, int n, | |
| 1144 int qscale) | |
| 1145 { | |
| 1146 int i, j, level, last_non_zero, q; | |
| 1147 const int *qmat; | |
| 1148 | |
| 1149 av_fdct (block); | |
|
40
f324d9047ca3
fixed block permutation in encoder (not optimal - should move it in forward DCT code)
glantau
parents:
19
diff
changeset
|
1150 |
|
f324d9047ca3
fixed block permutation in encoder (not optimal - should move it in forward DCT code)
glantau
parents:
19
diff
changeset
|
1151 /* we need this permutation so that we correct the IDCT |
|
f324d9047ca3
fixed block permutation in encoder (not optimal - should move it in forward DCT code)
glantau
parents:
19
diff
changeset
|
1152 permutation. will be moved into DCT code */ |
|
f324d9047ca3
fixed block permutation in encoder (not optimal - should move it in forward DCT code)
glantau
parents:
19
diff
changeset
|
1153 block_permute(block); |
| 0 | 1154 |
| 1155 if (s->mb_intra) { | |
| 1156 if (n < 4) | |
| 1157 q = s->y_dc_scale; | |
| 1158 else | |
| 1159 q = s->c_dc_scale; | |
| 1160 | |
| 1161 /* note: block[0] is assumed to be positive */ | |
| 1162 block[0] = (block[0] + (q >> 1)) / q; | |
| 1163 i = 1; | |
| 1164 last_non_zero = 0; | |
| 1165 if (s->out_format == FMT_H263) { | |
| 1166 qmat = s->q_non_intra_matrix; | |
| 1167 } else { | |
| 1168 qmat = s->q_intra_matrix; | |
| 1169 } | |
| 1170 } else { | |
| 1171 i = 0; | |
| 1172 last_non_zero = -1; | |
| 1173 qmat = s->q_non_intra_matrix; | |
| 1174 } | |
| 1175 | |
| 1176 for(;i<64;i++) { | |
| 1177 j = zigzag_direct[i]; | |
| 1178 level = block[j]; | |
| 1179 level = level * qmat[j]; | |
| 1180 /* XXX: slight error for the low range. Test should be equivalent to | |
| 1181 (level <= -(1 << (QMAT_SHIFT_MMX - 3)) || level >= (1 << | |
| 1182 (QMAT_SHIFT_MMX - 3))) | |
| 1183 */ | |
| 1184 if (((level << (31 - (QMAT_SHIFT_MMX - 3))) >> (31 - (QMAT_SHIFT_MMX - 3))) != | |
| 1185 level) { | |
| 1186 level = level / (1 << (QMAT_SHIFT_MMX - 3)); | |
| 1187 /* XXX: currently, this code is not optimal. the range should be: | |
| 1188 mpeg1: -255..255 | |
| 1189 mpeg2: -2048..2047 | |
| 1190 h263: -128..127 | |
| 1191 mpeg4: -2048..2047 | |
| 1192 */ | |
| 1193 if (level > 127) | |
| 1194 level = 127; | |
| 1195 else if (level < -128) | |
| 1196 level = -128; | |
| 1197 block[j] = level; | |
| 1198 last_non_zero = i; | |
| 1199 } else { | |
| 1200 block[j] = 0; | |
| 1201 } | |
| 1202 } | |
| 1203 return last_non_zero; | |
| 1204 } | |
| 1205 | |
|
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 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
|
1207 DCTELEM *block, int n, int qscale) |
| 0 | 1208 { |
| 1209 int i, level; | |
| 1210 const UINT16 *quant_matrix; | |
| 1211 | |
| 1212 if (s->mb_intra) { | |
| 1213 if (n < 4) | |
| 1214 block[0] = block[0] * s->y_dc_scale; | |
| 1215 else | |
| 1216 block[0] = block[0] * s->c_dc_scale; | |
| 1217 /* XXX: only mpeg1 */ | |
| 1218 quant_matrix = s->intra_matrix; | |
| 1219 for(i=1;i<64;i++) { | |
| 1220 level = block[i]; | |
| 1221 if (level) { | |
| 1222 if (level < 0) { | |
| 1223 level = -level; | |
| 1224 level = (int)(level * qscale * quant_matrix[i]) >> 3; | |
| 1225 level = (level - 1) | 1; | |
| 1226 level = -level; | |
| 1227 } else { | |
| 1228 level = (int)(level * qscale * quant_matrix[i]) >> 3; | |
| 1229 level = (level - 1) | 1; | |
| 1230 } | |
| 1231 #ifdef PARANOID | |
| 1232 if (level < -2048 || level > 2047) | |
| 1233 fprintf(stderr, "unquant error %d %d\n", i, level); | |
| 1234 #endif | |
| 1235 block[i] = level; | |
| 1236 } | |
| 1237 } | |
| 1238 } else { | |
| 1239 i = 0; | |
| 1240 quant_matrix = s->non_intra_matrix; | |
| 1241 for(;i<64;i++) { | |
| 1242 level = block[i]; | |
| 1243 if (level) { | |
| 1244 if (level < 0) { | |
| 1245 level = -level; | |
| 1246 level = (((level << 1) + 1) * qscale * | |
| 1247 ((int) (quant_matrix[i]))) >> 4; | |
| 1248 level = (level - 1) | 1; | |
| 1249 level = -level; | |
| 1250 } else { | |
| 1251 level = (((level << 1) + 1) * qscale * | |
| 1252 ((int) (quant_matrix[i]))) >> 4; | |
| 1253 level = (level - 1) | 1; | |
| 1254 } | |
| 1255 #ifdef PARANOID | |
| 1256 if (level < -2048 || level > 2047) | |
| 1257 fprintf(stderr, "unquant error %d %d\n", i, level); | |
| 1258 #endif | |
| 1259 block[i] = level; | |
| 1260 } | |
| 1261 } | |
| 1262 } | |
| 1263 } | |
|
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
|
1264 |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1265 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
|
1266 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
|
1267 { |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1268 int i, 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
|
1269 |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1270 if (s->mb_intra) { |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1271 if (n < 4) |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1272 block[0] = block[0] * s->y_dc_scale; |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1273 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
|
1274 block[0] = block[0] * s->c_dc_scale; |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1275 i = 1; |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1276 } 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
|
1277 i = 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
|
1278 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1279 |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1280 qmul = s->qscale << 1; |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1281 qadd = (s->qscale - 1) | 1; |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1282 |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1283 for(;i<64;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
|
1284 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
|
1285 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
|
1286 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
|
1287 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
|
1288 } 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
|
1289 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
|
1290 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1291 #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
|
1292 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
|
1293 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
|
1294 #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
|
1295 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
|
1296 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1297 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1298 } |
| 0 | 1299 |
| 1300 /* rate control */ | |
| 1301 | |
| 1302 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */ | |
| 1303 #define I_FRAME_SIZE_RATIO 3.0 | |
| 1304 #define QSCALE_K 20 | |
| 1305 | |
| 1306 static void rate_control_init(MpegEncContext *s) | |
| 1307 { | |
| 1308 s->wanted_bits = 0; | |
| 1309 | |
| 1310 if (s->intra_only) { | |
| 1311 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate; | |
| 1312 s->P_frame_bits = s->I_frame_bits; | |
| 1313 } else { | |
| 1314 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) / | |
| 1315 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1))); | |
| 1316 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO); | |
| 1317 } | |
| 1318 | |
| 1319 #if defined(DEBUG) | |
| 1320 printf("I_frame_size=%d P_frame_size=%d\n", | |
| 1321 s->I_frame_bits, s->P_frame_bits); | |
| 1322 #endif | |
| 1323 } | |
| 1324 | |
| 1325 | |
| 1326 /* | |
| 1327 * This heuristic is rather poor, but at least we do not have to | |
| 1328 * change the qscale at every macroblock. | |
| 1329 */ | |
| 1330 static int rate_estimate_qscale(MpegEncContext *s) | |
| 1331 { | |
|
187
3f3b14d3a23d
qscale estimate fix, diff extended to 64bit wide. patch by Stephen Davies <steve@daviesfam.org>
arpi_esp
parents:
178
diff
changeset
|
1332 INT64 diff, total_bits = s->total_bits; |
| 0 | 1333 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
|
1334 int qscale, qmin; |
| 0 | 1335 |
| 1336 if (s->pict_type == I_TYPE) { | |
| 1337 s->wanted_bits += s->I_frame_bits; | |
| 1338 } else { | |
| 1339 s->wanted_bits += s->P_frame_bits; | |
| 1340 } | |
| 1341 diff = s->wanted_bits - total_bits; | |
| 1342 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width); | |
| 1343 /* adjust for I frame */ | |
| 1344 if (s->pict_type == I_TYPE && !s->intra_only) { | |
| 1345 q /= I_FRAME_SIZE_RATIO; | |
| 1346 } | |
| 1347 | |
| 1348 /* using a too small Q scale leeds to problems in mpeg1 and h263 | |
| 1349 because AC coefficients are clamped to 255 or 127 */ | |
| 1350 qmin = 3; | |
| 1351 if (q < qmin) | |
| 1352 q = qmin; | |
| 1353 else if (q > 31) | |
| 1354 q = 31; | |
| 1355 qscale = (int)(q + 0.5); | |
| 1356 #if defined(DEBUG) | |
| 64 | 1357 printf("%d: total=%0.0f br=%0.1f diff=%d qest=%0.1f\n", |
| 0 | 1358 s->picture_number, |
| 64 | 1359 (double)total_bits, |
| 0 | 1360 (float)s->frame_rate / FRAME_RATE_BASE * |
| 1361 total_bits / s->picture_number, | |
| 1362 diff, q); | |
| 1363 #endif | |
| 1364 return qscale; | |
| 1365 } | |
| 1366 | |
| 1367 AVCodec mpeg1video_encoder = { | |
| 1368 "mpeg1video", | |
| 1369 CODEC_TYPE_VIDEO, | |
| 1370 CODEC_ID_MPEG1VIDEO, | |
| 1371 sizeof(MpegEncContext), | |
| 1372 MPV_encode_init, | |
| 1373 MPV_encode_picture, | |
| 1374 MPV_encode_end, | |
| 1375 }; | |
| 1376 | |
| 1377 AVCodec h263_encoder = { | |
| 1378 "h263", | |
| 1379 CODEC_TYPE_VIDEO, | |
| 1380 CODEC_ID_H263, | |
| 1381 sizeof(MpegEncContext), | |
| 1382 MPV_encode_init, | |
| 1383 MPV_encode_picture, | |
| 1384 MPV_encode_end, | |
| 1385 }; | |
| 1386 | |
| 1387 AVCodec h263p_encoder = { | |
| 1388 "h263p", | |
| 1389 CODEC_TYPE_VIDEO, | |
| 1390 CODEC_ID_H263P, | |
| 1391 sizeof(MpegEncContext), | |
| 1392 MPV_encode_init, | |
| 1393 MPV_encode_picture, | |
| 1394 MPV_encode_end, | |
| 1395 }; | |
| 1396 | |
| 1397 AVCodec rv10_encoder = { | |
| 1398 "rv10", | |
| 1399 CODEC_TYPE_VIDEO, | |
| 1400 CODEC_ID_RV10, | |
| 1401 sizeof(MpegEncContext), | |
| 1402 MPV_encode_init, | |
| 1403 MPV_encode_picture, | |
| 1404 MPV_encode_end, | |
| 1405 }; | |
| 1406 | |
| 1407 AVCodec mjpeg_encoder = { | |
| 1408 "mjpeg", | |
| 1409 CODEC_TYPE_VIDEO, | |
| 1410 CODEC_ID_MJPEG, | |
| 1411 sizeof(MpegEncContext), | |
| 1412 MPV_encode_init, | |
| 1413 MPV_encode_picture, | |
| 1414 MPV_encode_end, | |
| 1415 }; | |
| 1416 | |
| 71 | 1417 AVCodec mpeg4_encoder = { |
| 1418 "mpeg4", | |
| 0 | 1419 CODEC_TYPE_VIDEO, |
| 71 | 1420 CODEC_ID_MPEG4, |
| 0 | 1421 sizeof(MpegEncContext), |
| 1422 MPV_encode_init, | |
| 1423 MPV_encode_picture, | |
| 1424 MPV_encode_end, | |
| 1425 }; | |
| 1426 | |
| 1427 AVCodec msmpeg4_encoder = { | |
| 1428 "msmpeg4", | |
| 1429 CODEC_TYPE_VIDEO, | |
| 1430 CODEC_ID_MSMPEG4, | |
| 1431 sizeof(MpegEncContext), | |
| 1432 MPV_encode_init, | |
| 1433 MPV_encode_picture, | |
| 1434 MPV_encode_end, | |
| 1435 }; |
