Mercurial > libavcodec.hg
annotate mpegvideo.c @ 225:ae145876789d libavcodec
use multiply instead of divides for DC prediction on X86
author | michaelni |
---|---|
date | Tue, 05 Feb 2002 22:51:23 +0000 |
parents | 0b234715e205 |
children | ec1bc02a0a47 |
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; | |
118 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; | |
119 | |
120 for(i=0;i<3;i++) { | |
121 int w, h, shift, pict_start; | |
122 | |
123 w = s->linesize; | |
124 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
125 shift = (i == 0) ? 0 : 1; | |
126 c_size = (w >> shift) * (h >> shift); | |
127 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
128 | |
129 pict = av_mallocz(c_size); | |
130 if (pict == NULL) | |
131 goto fail; | |
132 s->last_picture_base[i] = pict; | |
133 s->last_picture[i] = pict + pict_start; | |
134 | |
135 pict = av_mallocz(c_size); | |
136 if (pict == NULL) | |
137 goto fail; | |
138 s->next_picture_base[i] = pict; | |
139 s->next_picture[i] = pict + pict_start; | |
140 | |
141 if (s->has_b_frames) { | |
142 pict = av_mallocz(c_size); | |
143 if (pict == NULL) | |
144 goto fail; | |
145 s->aux_picture_base[i] = pict; | |
146 s->aux_picture[i] = pict + pict_start; | |
147 } | |
148 } | |
149 | |
150 if (s->out_format == FMT_H263) { | |
151 int size; | |
152 /* MV prediction */ | |
153 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
154 s->motion_val = malloc(size * 2 * sizeof(INT16)); | |
155 if (s->motion_val == NULL) | |
156 goto fail; | |
157 memset(s->motion_val, 0, size * 2 * sizeof(INT16)); | |
158 } | |
159 | |
160 if (s->h263_pred) { | |
161 int y_size, c_size, i, size; | |
162 | |
163 /* dc values */ | |
164 | |
165 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
166 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
167 size = y_size + 2 * c_size; | |
168 s->dc_val[0] = malloc(size * sizeof(INT16)); | |
169 if (s->dc_val[0] == NULL) | |
170 goto fail; | |
171 s->dc_val[1] = s->dc_val[0] + y_size; | |
172 s->dc_val[2] = s->dc_val[1] + c_size; | |
173 for(i=0;i<size;i++) | |
174 s->dc_val[0][i] = 1024; | |
175 | |
176 /* ac values */ | |
177 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16); | |
178 if (s->ac_val[0] == NULL) | |
179 goto fail; | |
180 s->ac_val[1] = s->ac_val[0] + y_size; | |
181 s->ac_val[2] = s->ac_val[1] + c_size; | |
182 | |
183 /* cbp values */ | |
184 s->coded_block = av_mallocz(y_size); | |
185 if (!s->coded_block) | |
186 goto fail; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
187 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
188 /* 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
|
189 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
|
190 if (!s->mbintra_table) |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
191 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
|
192 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
|
193 } |
0 | 194 /* default structure is frame */ |
195 s->picture_structure = PICT_FRAME; | |
196 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
197 /* init macroblock skip table */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
198 if (!s->encoding) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
199 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
|
200 if (!s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
201 goto fail; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
202 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
203 |
0 | 204 s->context_initialized = 1; |
205 return 0; | |
206 fail: | |
207 if (s->motion_val) | |
208 free(s->motion_val); | |
209 if (s->dc_val[0]) | |
210 free(s->dc_val[0]); | |
211 if (s->ac_val[0]) | |
212 free(s->ac_val[0]); | |
213 if (s->coded_block) | |
214 free(s->coded_block); | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
215 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
|
216 free(s->mbintra_table); |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
217 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
218 free(s->mbskip_table); |
0 | 219 for(i=0;i<3;i++) { |
220 if (s->last_picture_base[i]) | |
221 free(s->last_picture_base[i]); | |
222 if (s->next_picture_base[i]) | |
223 free(s->next_picture_base[i]); | |
224 if (s->aux_picture_base[i]) | |
225 free(s->aux_picture_base[i]); | |
226 } | |
227 return -1; | |
228 } | |
229 | |
230 /* init common structure for both encoder and decoder */ | |
231 void MPV_common_end(MpegEncContext *s) | |
232 { | |
233 int i; | |
234 | |
235 if (s->motion_val) | |
236 free(s->motion_val); | |
237 if (s->h263_pred) { | |
238 free(s->dc_val[0]); | |
239 free(s->ac_val[0]); | |
240 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
|
241 free(s->mbintra_table); |
0 | 242 } |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
243 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
244 free(s->mbskip_table); |
0 | 245 for(i=0;i<3;i++) { |
246 free(s->last_picture_base[i]); | |
247 free(s->next_picture_base[i]); | |
248 if (s->has_b_frames) | |
249 free(s->aux_picture_base[i]); | |
250 } | |
251 s->context_initialized = 0; | |
252 } | |
253 | |
254 /* init video encoder */ | |
255 int MPV_encode_init(AVCodecContext *avctx) | |
256 { | |
257 MpegEncContext *s = avctx->priv_data; | |
60 | 258 int i; |
0 | 259 |
260 s->bit_rate = avctx->bit_rate; | |
261 s->frame_rate = avctx->frame_rate; | |
262 s->width = avctx->width; | |
263 s->height = avctx->height; | |
264 s->gop_size = avctx->gop_size; | |
162 | 265 s->rtp_mode = avctx->rtp_mode; |
266 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
|
267 s->avctx = avctx; |
162 | 268 |
0 | 269 if (s->gop_size <= 1) { |
270 s->intra_only = 1; | |
271 s->gop_size = 12; | |
272 } else { | |
273 s->intra_only = 0; | |
274 } | |
275 s->full_search = motion_estimation_method; | |
276 | |
277 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE); | |
278 | |
279 switch(avctx->codec->id) { | |
280 case CODEC_ID_MPEG1VIDEO: | |
281 s->out_format = FMT_MPEG1; | |
282 break; | |
283 case CODEC_ID_MJPEG: | |
284 s->out_format = FMT_MJPEG; | |
285 s->intra_only = 1; /* force intra only for jpeg */ | |
286 if (mjpeg_init(s) < 0) | |
287 return -1; | |
288 break; | |
289 case CODEC_ID_H263: | |
178 | 290 if (h263_get_picture_format(s->width, s->height) == 7){ |
291 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); | |
0 | 292 return -1; |
178 | 293 } |
0 | 294 s->out_format = FMT_H263; |
295 break; | |
296 case CODEC_ID_H263P: | |
297 s->out_format = FMT_H263; | |
162 | 298 s->rtp_mode = 1; |
299 s->rtp_payload_size = 1200; | |
0 | 300 s->h263_plus = 1; |
78 | 301 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
302 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
303 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
304 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
305 s->umvplus_dec = 0; |
0 | 306 break; |
307 case CODEC_ID_RV10: | |
308 s->out_format = FMT_H263; | |
309 s->h263_rv10 = 1; | |
310 break; | |
71 | 311 case CODEC_ID_MPEG4: |
0 | 312 s->out_format = FMT_H263; |
313 s->h263_pred = 1; | |
314 s->unrestricted_mv = 1; | |
315 break; | |
316 case CODEC_ID_MSMPEG4: | |
317 s->out_format = FMT_H263; | |
318 s->h263_msmpeg4 = 1; | |
319 s->h263_pred = 1; | |
320 s->unrestricted_mv = 1; | |
321 break; | |
322 default: | |
323 return -1; | |
324 } | |
325 | |
326 if (s->out_format == FMT_H263) | |
327 h263_encode_init_vlc(s); | |
328 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
329 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
330 |
0 | 331 /* init */ |
332 if (MPV_common_init(s) < 0) | |
333 return -1; | |
334 | |
60 | 335 /* init default q matrix */ |
336 for(i=0;i<64;i++) { | |
337 s->intra_matrix[i] = default_intra_matrix[i]; | |
338 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
339 } | |
340 | |
0 | 341 /* rate control init */ |
342 rate_control_init(s); | |
343 | |
344 s->picture_number = 0; | |
345 s->fake_picture_number = 0; | |
346 /* motion detector init */ | |
347 s->f_code = 1; | |
348 | |
349 return 0; | |
350 } | |
351 | |
352 int MPV_encode_end(AVCodecContext *avctx) | |
353 { | |
354 MpegEncContext *s = avctx->priv_data; | |
355 | |
356 #ifdef STATS | |
357 print_stats(); | |
358 #endif | |
359 MPV_common_end(s); | |
360 if (s->out_format == FMT_MJPEG) | |
361 mjpeg_close(s); | |
362 return 0; | |
363 } | |
364 | |
365 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 366 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 367 { |
368 UINT8 *ptr, *last_line; | |
369 int i; | |
370 | |
371 last_line = buf + (height - 1) * wrap; | |
372 for(i=0;i<w;i++) { | |
373 /* top and bottom */ | |
374 memcpy(buf - (i + 1) * wrap, buf, width); | |
375 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
376 } | |
377 /* left and right */ | |
378 ptr = buf; | |
379 for(i=0;i<height;i++) { | |
380 memset(ptr - w, ptr[0], w); | |
381 memset(ptr + width, ptr[width-1], w); | |
382 ptr += wrap; | |
383 } | |
384 /* corners */ | |
385 for(i=0;i<w;i++) { | |
386 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
387 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
388 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
389 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
390 } | |
391 } | |
392 | |
393 /* generic function for encode/decode called before a frame is coded/decoded */ | |
394 void MPV_frame_start(MpegEncContext *s) | |
395 { | |
396 int i; | |
397 UINT8 *tmp; | |
398 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
399 s->mb_skiped = 0; |
0 | 400 if (s->pict_type == B_TYPE) { |
401 for(i=0;i<3;i++) { | |
402 s->current_picture[i] = s->aux_picture[i]; | |
403 } | |
404 } else { | |
405 for(i=0;i<3;i++) { | |
406 /* swap next and last */ | |
407 tmp = s->last_picture[i]; | |
408 s->last_picture[i] = s->next_picture[i]; | |
409 s->next_picture[i] = tmp; | |
410 s->current_picture[i] = tmp; | |
411 } | |
412 } | |
413 } | |
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
|
414 |
0 | 415 /* generic function for encode/decode called after a frame has been coded/decoded */ |
416 void MPV_frame_end(MpegEncContext *s) | |
417 { | |
418 /* draw edge for correct motion prediction if outside */ | |
220 | 419 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
|
420 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
|
421 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
|
422 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
|
423 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
|
424 }else{ |
42552c1cf360
fix different UMV handling for mpeg4 vs. h263 - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
187
diff
changeset
|
425 /* OpenDivx, but i dunno how to distinguish it from mpeg4 */ |
0 | 426 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); |
427 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
428 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
|
429 } |
0 | 430 } |
207 | 431 emms_c(); |
0 | 432 } |
433 | |
434 int MPV_encode_picture(AVCodecContext *avctx, | |
435 unsigned char *buf, int buf_size, void *data) | |
436 { | |
437 MpegEncContext *s = avctx->priv_data; | |
438 AVPicture *pict = data; | |
439 int i, j; | |
440 | |
441 if (s->fixed_qscale) | |
442 s->qscale = avctx->quality; | |
443 | |
444 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
445 | |
446 if (!s->intra_only) { | |
447 /* first picture of GOP is intra */ | |
448 if ((s->picture_number % s->gop_size) == 0) | |
449 s->pict_type = I_TYPE; | |
450 else | |
451 s->pict_type = P_TYPE; | |
452 } else { | |
453 s->pict_type = I_TYPE; | |
454 } | |
455 avctx->key_frame = (s->pict_type == I_TYPE); | |
456 | |
457 MPV_frame_start(s); | |
220 | 458 |
0 | 459 for(i=0;i<3;i++) { |
460 UINT8 *src = pict->data[i]; | |
461 UINT8 *dest = s->current_picture[i]; | |
462 int src_wrap = pict->linesize[i]; | |
463 int dest_wrap = s->linesize; | |
464 int w = s->width; | |
465 int h = s->height; | |
466 | |
467 if (i >= 1) { | |
468 dest_wrap >>= 1; | |
469 w >>= 1; | |
470 h >>= 1; | |
471 } | |
472 | |
220 | 473 if(s->intra_only && dest_wrap==src_wrap){ |
474 s->current_picture[i] = pict->data[i]; | |
475 }else { | |
476 for(j=0;j<h;j++) { | |
477 memcpy(dest, src, w); | |
478 dest += dest_wrap; | |
479 src += src_wrap; | |
480 } | |
481 } | |
0 | 482 s->new_picture[i] = s->current_picture[i]; |
483 } | |
484 | |
485 encode_picture(s, s->picture_number); | |
486 | |
487 MPV_frame_end(s); | |
488 s->picture_number++; | |
489 | |
490 if (s->out_format == FMT_MJPEG) | |
491 mjpeg_picture_trailer(s); | |
492 | |
493 flush_put_bits(&s->pb); | |
494 s->total_bits += (s->pb.buf_ptr - s->pb.buf) * 8; | |
495 avctx->quality = s->qscale; | |
496 return s->pb.buf_ptr - s->pb.buf; | |
497 } | |
498 | |
499 static inline int clip(int a, int amin, int amax) | |
500 { | |
501 if (a < amin) | |
502 return amin; | |
503 else if (a > amax) | |
504 return amax; | |
505 else | |
506 return a; | |
507 } | |
508 | |
509 /* apply one mpeg motion vector to the three components */ | |
510 static inline void mpeg_motion(MpegEncContext *s, | |
511 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
512 int dest_offset, | |
513 UINT8 **ref_picture, int src_offset, | |
514 int field_based, op_pixels_func *pix_op, | |
515 int motion_x, int motion_y, int h) | |
516 { | |
517 UINT8 *ptr; | |
518 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
519 | |
520 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
521 src_x = s->mb_x * 16 + (motion_x >> 1); | |
522 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
523 | |
524 /* WARNING: do no forget half pels */ | |
525 height = s->height >> field_based; | |
526 src_x = clip(src_x, -16, s->width); | |
527 if (src_x == s->width) | |
528 dxy &= ~1; | |
529 src_y = clip(src_y, -16, height); | |
530 if (src_y == height) | |
531 dxy &= ~2; | |
532 linesize = s->linesize << field_based; | |
533 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
534 dest_y += dest_offset; | |
535 pix_op[dxy](dest_y, ptr, linesize, h); | |
536 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
537 | |
538 if (s->out_format == FMT_H263) { | |
539 dxy = 0; | |
540 if ((motion_x & 3) != 0) | |
541 dxy |= 1; | |
542 if ((motion_y & 3) != 0) | |
543 dxy |= 2; | |
544 mx = motion_x >> 2; | |
545 my = motion_y >> 2; | |
546 } else { | |
547 mx = motion_x / 2; | |
548 my = motion_y / 2; | |
549 dxy = ((my & 1) << 1) | (mx & 1); | |
550 mx >>= 1; | |
551 my >>= 1; | |
552 } | |
553 | |
554 src_x = s->mb_x * 8 + mx; | |
555 src_y = s->mb_y * (8 >> field_based) + my; | |
556 src_x = clip(src_x, -8, s->width >> 1); | |
557 if (src_x == (s->width >> 1)) | |
558 dxy &= ~1; | |
559 src_y = clip(src_y, -8, height >> 1); | |
560 if (src_y == (height >> 1)) | |
561 dxy &= ~2; | |
562 | |
563 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
564 ptr = ref_picture[1] + offset; | |
565 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
566 ptr = ref_picture[2] + offset; | |
567 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
568 } | |
569 | |
570 static inline void MPV_motion(MpegEncContext *s, | |
571 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
572 int dir, UINT8 **ref_picture, | |
573 op_pixels_func *pix_op) | |
574 { | |
575 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
576 int mb_x, mb_y, i; | |
577 UINT8 *ptr, *dest; | |
578 | |
579 mb_x = s->mb_x; | |
580 mb_y = s->mb_y; | |
581 | |
582 switch(s->mv_type) { | |
583 case MV_TYPE_16X16: | |
584 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
585 ref_picture, 0, | |
586 0, pix_op, | |
587 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
588 break; | |
589 case MV_TYPE_8X8: | |
590 for(i=0;i<4;i++) { | |
591 motion_x = s->mv[dir][i][0]; | |
592 motion_y = s->mv[dir][i][1]; | |
593 | |
594 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
595 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
596 src_y = mb_y * 16 + (motion_y >> 1) + ((i >> 1) & 1) * 8; | |
597 | |
598 /* WARNING: do no forget half pels */ | |
599 src_x = clip(src_x, -16, s->width); | |
600 if (src_x == s->width) | |
601 dxy &= ~1; | |
602 src_y = clip(src_y, -16, s->height); | |
603 if (src_y == s->height) | |
604 dxy &= ~2; | |
605 | |
606 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
607 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
608 pix_op[dxy](dest, ptr, s->linesize, 8); | |
609 } | |
610 /* In case of 8X8, we construct a single chroma motion vector | |
611 with a special rounding */ | |
612 mx = 0; | |
613 my = 0; | |
614 for(i=0;i<4;i++) { | |
615 mx += s->mv[dir][i][0]; | |
616 my += s->mv[dir][i][1]; | |
617 } | |
618 if (mx >= 0) | |
619 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
620 else { | |
621 mx = -mx; | |
622 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
623 } | |
624 if (my >= 0) | |
625 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
626 else { | |
627 my = -my; | |
628 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
629 } | |
630 dxy = ((my & 1) << 1) | (mx & 1); | |
631 mx >>= 1; | |
632 my >>= 1; | |
633 | |
634 src_x = mb_x * 8 + mx; | |
635 src_y = mb_y * 8 + my; | |
636 src_x = clip(src_x, -8, s->width/2); | |
637 if (src_x == s->width/2) | |
638 dxy &= ~1; | |
639 src_y = clip(src_y, -8, s->height/2); | |
640 if (src_y == s->height/2) | |
641 dxy &= ~2; | |
642 | |
643 offset = (src_y * (s->linesize >> 1)) + src_x; | |
644 ptr = ref_picture[1] + offset; | |
645 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
646 ptr = ref_picture[2] + offset; | |
647 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
648 break; | |
649 case MV_TYPE_FIELD: | |
650 if (s->picture_structure == PICT_FRAME) { | |
651 /* top field */ | |
652 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
653 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
654 1, pix_op, | |
655 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
656 /* bottom field */ | |
657 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
658 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
659 1, pix_op, | |
660 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
661 } else { | |
662 | |
663 | |
664 } | |
665 break; | |
666 } | |
667 } | |
668 | |
669 | |
670 /* put block[] to dest[] */ | |
671 static inline void put_dct(MpegEncContext *s, | |
672 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
673 { | |
674 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
|
675 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
|
676 ff_idct (block); |
0 | 677 put_pixels_clamped(block, dest, line_size); |
678 } | |
679 | |
680 /* add block[] to dest[] */ | |
681 static inline void add_dct(MpegEncContext *s, | |
682 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
683 { | |
684 if (s->block_last_index[i] >= 0) { | |
685 if (!s->mpeg2) | |
206 | 686 if(s->encoding || (!s->h263_msmpeg4)) |
200 | 687 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
|
688 ff_idct (block); |
0 | 689 add_pixels_clamped(block, dest, line_size); |
690 } | |
691 } | |
692 | |
693 /* generic function called after a macroblock has been parsed by the | |
694 decoder or after it has been encoded by the encoder. | |
695 | |
696 Important variables used: | |
697 s->mb_intra : true if intra macroblock | |
698 s->mv_dir : motion vector direction | |
699 s->mv_type : motion vector type | |
700 s->mv : motion vector | |
701 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
702 */ | |
703 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
704 { | |
705 int mb_x, mb_y, motion_x, motion_y; | |
706 int dct_linesize, dct_offset; | |
707 op_pixels_func *op_pix; | |
708 | |
709 mb_x = s->mb_x; | |
710 mb_y = s->mb_y; | |
711 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
712 #ifdef FF_POSTPROCESS |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
713 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
714 //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
|
715 #endif |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
716 |
0 | 717 /* update DC predictors for P macroblocks */ |
718 if (!s->mb_intra) { | |
719 if (s->h263_pred) { | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
720 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
|
721 { |
0 | 722 int wrap, x, y, v; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
723 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
|
724 |
0 | 725 wrap = 2 * s->mb_width + 2; |
726 v = 1024; | |
727 x = 2 * mb_x + 1; | |
728 y = 2 * mb_y + 1; | |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
729 |
0 | 730 s->dc_val[0][(x) + (y) * wrap] = v; |
731 s->dc_val[0][(x + 1) + (y) * wrap] = v; | |
732 s->dc_val[0][(x) + (y + 1) * wrap] = v; | |
733 s->dc_val[0][(x + 1) + (y + 1) * wrap] = v; | |
734 /* ac pred */ | |
735 memset(s->ac_val[0][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
736 memset(s->ac_val[0][(x + 1) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
737 memset(s->ac_val[0][(x) + (y + 1) * wrap], 0, 16 * sizeof(INT16)); | |
738 memset(s->ac_val[0][(x + 1) + (y + 1) * wrap], 0, 16 * sizeof(INT16)); | |
739 if (s->h263_msmpeg4) { | |
740 s->coded_block[(x) + (y) * wrap] = 0; | |
741 s->coded_block[(x + 1) + (y) * wrap] = 0; | |
742 s->coded_block[(x) + (y + 1) * wrap] = 0; | |
743 s->coded_block[(x + 1) + (y + 1) * wrap] = 0; | |
744 } | |
745 /* chroma */ | |
746 wrap = s->mb_width + 2; | |
747 x = mb_x + 1; | |
748 y = mb_y + 1; | |
749 s->dc_val[1][(x) + (y) * wrap] = v; | |
750 s->dc_val[2][(x) + (y) * wrap] = v; | |
751 /* ac pred */ | |
752 memset(s->ac_val[1][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
753 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
|
754 } |
0 | 755 } else { |
756 s->last_dc[0] = 128 << s->intra_dc_precision; | |
757 s->last_dc[1] = 128 << s->intra_dc_precision; | |
758 s->last_dc[2] = 128 << s->intra_dc_precision; | |
759 } | |
760 } | |
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
|
761 else if (s->h263_pred) |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
762 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
|
763 |
0 | 764 /* update motion predictor */ |
765 if (s->out_format == FMT_H263) { | |
766 int x, y, wrap; | |
767 | |
768 x = 2 * mb_x + 1; | |
769 y = 2 * mb_y + 1; | |
770 wrap = 2 * s->mb_width + 2; | |
771 if (s->mb_intra) { | |
772 motion_x = 0; | |
773 motion_y = 0; | |
774 goto motion_init; | |
775 } else if (s->mv_type == MV_TYPE_16X16) { | |
776 motion_x = s->mv[0][0][0]; | |
777 motion_y = s->mv[0][0][1]; | |
778 motion_init: | |
779 /* no update if 8X8 because it has been done during parsing */ | |
780 s->motion_val[(x) + (y) * wrap][0] = motion_x; | |
781 s->motion_val[(x) + (y) * wrap][1] = motion_y; | |
782 s->motion_val[(x + 1) + (y) * wrap][0] = motion_x; | |
783 s->motion_val[(x + 1) + (y) * wrap][1] = motion_y; | |
784 s->motion_val[(x) + (y + 1) * wrap][0] = motion_x; | |
785 s->motion_val[(x) + (y + 1) * wrap][1] = motion_y; | |
786 s->motion_val[(x + 1) + (y + 1) * wrap][0] = motion_x; | |
787 s->motion_val[(x + 1) + (y + 1) * wrap][1] = motion_y; | |
788 } | |
789 } | |
790 | |
791 if (!s->intra_only) { | |
792 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
|
793 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
794 |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
795 /* 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
|
796 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
|
797 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
|
798 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
799 s->mb_skiped = 0; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
800 /* 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
|
801 if (*mbskip_ptr != 0) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
802 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
803 *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
|
804 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
805 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
806 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
807 } |
0 | 808 |
809 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
810 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
811 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
812 | |
813 if (s->interlaced_dct) { | |
814 dct_linesize = s->linesize * 2; | |
815 dct_offset = s->linesize; | |
816 } else { | |
817 dct_linesize = s->linesize; | |
818 dct_offset = s->linesize * 8; | |
819 } | |
820 | |
821 if (!s->mb_intra) { | |
822 /* motion handling */ | |
823 if (!s->no_rounding) | |
824 op_pix = put_pixels_tab; | |
825 else | |
826 op_pix = put_no_rnd_pixels_tab; | |
827 | |
828 if (s->mv_dir & MV_DIR_FORWARD) { | |
829 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix); | |
830 if (!s->no_rounding) | |
831 op_pix = avg_pixels_tab; | |
832 else | |
833 op_pix = avg_no_rnd_pixels_tab; | |
834 } | |
835 if (s->mv_dir & MV_DIR_BACKWARD) { | |
836 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix); | |
837 } | |
838 | |
839 /* add dct residue */ | |
840 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
841 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
842 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
843 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
844 | |
57 | 845 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
846 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 847 } else { |
848 /* dct only in intra block */ | |
849 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
850 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
851 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
852 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
853 | |
57 | 854 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
855 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 856 } |
857 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
858 the_end: |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
859 emms_c(); |
0 | 860 } |
861 | |
862 static void encode_picture(MpegEncContext *s, int picture_number) | |
863 { | |
162 | 864 int mb_x, mb_y, wrap, last_gob; |
0 | 865 UINT8 *ptr; |
866 int i, motion_x, motion_y; | |
867 | |
868 s->picture_number = picture_number; | |
869 if (!s->fixed_qscale) | |
870 s->qscale = rate_estimate_qscale(s); | |
871 | |
872 /* precompute matrix */ | |
873 if (s->out_format == FMT_MJPEG) { | |
874 /* for mjpeg, we do include qscale in the matrix */ | |
875 s->intra_matrix[0] = default_intra_matrix[0]; | |
876 for(i=1;i<64;i++) | |
877 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
220 | 878 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
0 | 879 } else { |
220 | 880 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
881 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); | |
0 | 882 } |
883 | |
884 switch(s->out_format) { | |
885 case FMT_MJPEG: | |
886 mjpeg_picture_header(s); | |
887 break; | |
888 case FMT_H263: | |
889 if (s->h263_msmpeg4) | |
890 msmpeg4_encode_picture_header(s, picture_number); | |
891 else if (s->h263_pred) | |
892 mpeg4_encode_picture_header(s, picture_number); | |
893 else if (s->h263_rv10) | |
894 rv10_encode_picture_header(s, picture_number); | |
895 else | |
896 h263_encode_picture_header(s, picture_number); | |
897 break; | |
898 case FMT_MPEG1: | |
899 mpeg1_encode_picture_header(s, picture_number); | |
900 break; | |
901 } | |
902 | |
903 /* init last dc values */ | |
904 /* note: quant matrix value (8) is implied here */ | |
905 s->last_dc[0] = 128; | |
906 s->last_dc[1] = 128; | |
907 s->last_dc[2] = 128; | |
908 s->mb_incr = 1; | |
909 s->last_mv[0][0][0] = 0; | |
910 s->last_mv[0][0][1] = 0; | |
911 s->mv_type = MV_TYPE_16X16; | |
912 s->mv_dir = MV_DIR_FORWARD; | |
913 | |
162 | 914 /* Get the GOB height based on picture height */ |
915 if (s->out_format == FMT_H263 && s->h263_plus) { | |
916 if (s->height <= 400) | |
917 s->gob_index = 1; | |
918 else if (s->height <= 800) | |
919 s->gob_index = 2; | |
920 else | |
921 s->gob_index = 4; | |
922 } | |
923 | |
0 | 924 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
162 | 925 /* Put GOB header based on RTP MTU */ |
926 if (!mb_y) { | |
927 s->ptr_lastgob = s->pb.buf_ptr; | |
928 s->ptr_last_mb_line = s->pb.buf_ptr; | |
929 } else if (s->out_format == FMT_H263 && s->h263_plus) { | |
930 last_gob = h263_encode_gob_header(s, mb_y); | |
931 if (last_gob) { | |
932 //fprintf(stderr,"\nLast GOB size: %d", last_gob); | |
933 s->first_gob_line = 1; | |
934 } else | |
935 s->first_gob_line = 0; | |
936 } | |
0 | 937 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
938 | |
939 s->mb_x = mb_x; | |
940 s->mb_y = mb_y; | |
941 | |
942 /* compute motion vector and macro block type (intra or non intra) */ | |
943 motion_x = 0; | |
944 motion_y = 0; | |
945 if (s->pict_type == P_TYPE) { | |
946 s->mb_intra = estimate_motion(s, mb_x, mb_y, | |
947 &motion_x, | |
948 &motion_y); | |
949 } else { | |
950 s->mb_intra = 1; | |
951 } | |
952 | |
953 /* get the pixels */ | |
954 wrap = s->linesize; | |
955 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
|
956 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
|
957 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
|
958 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
|
959 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); |
0 | 960 wrap = s->linesize >> 1; |
961 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
|
962 get_pixels(s->block[4], ptr, wrap); |
0 | 963 |
964 wrap = s->linesize >> 1; | |
965 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
|
966 get_pixels(s->block[5], ptr, wrap); |
0 | 967 |
968 /* subtract previous frame if non intra */ | |
969 if (!s->mb_intra) { | |
970 int dxy, offset, mx, my; | |
971 | |
972 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
973 ptr = s->last_picture[0] + | |
974 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) + | |
975 (mb_x * 16 + (motion_x >> 1)); | |
976 | |
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
|
977 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
|
978 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
|
979 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
|
980 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy); |
0 | 981 |
982 if (s->out_format == FMT_H263) { | |
983 /* special rounding for h263 */ | |
984 dxy = 0; | |
985 if ((motion_x & 3) != 0) | |
986 dxy |= 1; | |
987 if ((motion_y & 3) != 0) | |
988 dxy |= 2; | |
989 mx = motion_x >> 2; | |
990 my = motion_y >> 2; | |
991 } else { | |
992 mx = motion_x / 2; | |
993 my = motion_y / 2; | |
994 dxy = ((my & 1) << 1) | (mx & 1); | |
995 mx >>= 1; | |
996 my >>= 1; | |
997 } | |
998 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx); | |
999 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
|
1000 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); |
0 | 1001 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
|
1002 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); |
0 | 1003 } |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1004 emms_c(); |
0 | 1005 |
1006 /* DCT & quantize */ | |
1007 if (s->h263_msmpeg4) { | |
1008 msmpeg4_dc_scale(s); | |
1009 } else if (s->h263_pred) { | |
1010 h263_dc_scale(s); | |
1011 } else { | |
1012 /* default quantization values */ | |
1013 s->y_dc_scale = 8; | |
1014 s->c_dc_scale = 8; | |
1015 } | |
1016 for(i=0;i<6;i++) { | |
220 | 1017 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); |
0 | 1018 } |
1019 | |
1020 /* huffman encode */ | |
1021 switch(s->out_format) { | |
1022 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
|
1023 mpeg1_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1024 break; |
1025 case FMT_H263: | |
1026 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
|
1027 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1028 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
|
1029 h263_encode_mb(s, s->block, motion_x, motion_y); |
0 | 1030 break; |
1031 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
|
1032 mjpeg_encode_mb(s, s->block); |
0 | 1033 break; |
1034 } | |
1035 | |
1036 /* decompress blocks so that we keep the state of the decoder */ | |
1037 s->mv[0][0][0] = motion_x; | |
1038 s->mv[0][0][1] = motion_y; | |
1039 | |
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
|
1040 MPV_decode_mb(s, s->block); |
0 | 1041 } |
162 | 1042 /* Obtain average MB line size for RTP */ |
1043 if (!mb_y) | |
1044 s->mb_line_avgsize = s->pb.buf_ptr - s->ptr_last_mb_line; | |
1045 else | |
1046 s->mb_line_avgsize = (s->mb_line_avgsize + s->pb.buf_ptr - s->ptr_last_mb_line) >> 1; | |
1047 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
1048 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
1049 s->ptr_last_mb_line = s->pb.buf_ptr; | |
0 | 1050 } |
208 | 1051 |
1052 if (s->h263_msmpeg4) | |
1053 msmpeg4_encode_ext_header(s); | |
1054 | |
162 | 1055 //if (s->gob_number) |
1056 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
0 | 1057 } |
1058 | |
220 | 1059 static int dct_quantize_c(MpegEncContext *s, |
0 | 1060 DCTELEM *block, int n, |
1061 int qscale) | |
1062 { | |
1063 int i, j, level, last_non_zero, q; | |
1064 const int *qmat; | |
216 | 1065 int minLevel, maxLevel; |
1066 | |
1067 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ | |
1068 /* mpeg4 */ | |
1069 minLevel= -2048; | |
1070 maxLevel= 2047; | |
1071 }else if(s->out_format==FMT_MPEG1){ | |
1072 /* mpeg1 */ | |
1073 minLevel= -255; | |
1074 maxLevel= 255; | |
1075 }else{ | |
1076 /* h263 / msmpeg4 */ | |
1077 minLevel= -128; | |
1078 maxLevel= 127; | |
1079 } | |
0 | 1080 |
1081 av_fdct (block); | |
1082 | |
64 | 1083 /* we need this permutation so that we correct the IDCT |
1084 permutation. will be moved into DCT code */ | |
1085 block_permute(block); | |
1086 | |
0 | 1087 if (s->mb_intra) { |
1088 if (n < 4) | |
1089 q = s->y_dc_scale; | |
1090 else | |
1091 q = s->c_dc_scale; | |
1092 q = q << 3; | |
1093 | |
1094 /* note: block[0] is assumed to be positive */ | |
1095 block[0] = (block[0] + (q >> 1)) / q; | |
1096 i = 1; | |
1097 last_non_zero = 0; | |
1098 if (s->out_format == FMT_H263) { | |
1099 qmat = s->q_non_intra_matrix; | |
1100 } else { | |
1101 qmat = s->q_intra_matrix; | |
1102 } | |
1103 } else { | |
1104 i = 0; | |
1105 last_non_zero = -1; | |
1106 qmat = s->q_non_intra_matrix; | |
1107 } | |
1108 | |
1109 for(;i<64;i++) { | |
1110 j = zigzag_direct[i]; | |
1111 level = block[j]; | |
1112 level = level * qmat[j]; | |
1113 #ifdef PARANOID | |
1114 { | |
1115 static int count = 0; | |
1116 int level1, level2, qmat1; | |
1117 double val; | |
1118 if (qmat == s->q_non_intra_matrix) { | |
1119 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1120 } else { | |
1121 qmat1 = default_intra_matrix[j] * s->qscale; | |
1122 } | |
1123 if (av_fdct != jpeg_fdct_ifast) | |
1124 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1125 else | |
1126 val = ((double)block[j] * 8.0 * 2048.0) / | |
1127 ((double)qmat1 * aanscales[j]); | |
1128 level1 = (int)val; | |
1129 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1130 if (level1 != level2) { | |
1131 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1132 count, level2, level1, block[j], qmat1, qmat[j], | |
1133 val); | |
1134 count++; | |
1135 } | |
1136 | |
1137 } | |
1138 #endif | |
1139 /* XXX: slight error for the low range. Test should be equivalent to | |
1140 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1141 (QMAT_SHIFT - 3))) | |
1142 */ | |
1143 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1144 level) { | |
1145 level = level / (1 << (QMAT_SHIFT - 3)); | |
1146 /* XXX: currently, this code is not optimal. the range should be: | |
1147 mpeg1: -255..255 | |
1148 mpeg2: -2048..2047 | |
1149 h263: -128..127 | |
1150 mpeg4: -2048..2047 | |
1151 */ | |
216 | 1152 if (level > maxLevel) |
1153 level = maxLevel; | |
1154 else if (level < minLevel) | |
1155 level = minLevel; | |
0 | 1156 |
1157 block[j] = level; | |
1158 last_non_zero = i; | |
1159 } else { | |
1160 block[j] = 0; | |
1161 } | |
1162 } | |
1163 return last_non_zero; | |
1164 } | |
1165 | |
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
|
1166 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
|
1167 DCTELEM *block, int n, int qscale) |
0 | 1168 { |
200 | 1169 int i, level, nCoeffs; |
0 | 1170 const UINT16 *quant_matrix; |
1171 | |
200 | 1172 if(s->alternate_scan) nCoeffs= 64; |
1173 else nCoeffs= s->block_last_index[n]+1; | |
1174 | |
0 | 1175 if (s->mb_intra) { |
1176 if (n < 4) | |
1177 block[0] = block[0] * s->y_dc_scale; | |
1178 else | |
1179 block[0] = block[0] * s->c_dc_scale; | |
1180 /* XXX: only mpeg1 */ | |
1181 quant_matrix = s->intra_matrix; | |
200 | 1182 for(i=1;i<nCoeffs;i++) { |
1183 int j= zigzag_direct[i]; | |
1184 level = block[j]; | |
0 | 1185 if (level) { |
1186 if (level < 0) { | |
1187 level = -level; | |
200 | 1188 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1189 level = (level - 1) | 1; |
1190 level = -level; | |
1191 } else { | |
200 | 1192 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 1193 level = (level - 1) | 1; |
1194 } | |
1195 #ifdef PARANOID | |
1196 if (level < -2048 || level > 2047) | |
1197 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1198 #endif | |
200 | 1199 block[j] = level; |
0 | 1200 } |
1201 } | |
1202 } else { | |
1203 i = 0; | |
1204 quant_matrix = s->non_intra_matrix; | |
217 | 1205 for(;i<nCoeffs;i++) { |
200 | 1206 int j= zigzag_direct[i]; |
1207 level = block[j]; | |
0 | 1208 if (level) { |
1209 if (level < 0) { | |
1210 level = -level; | |
1211 level = (((level << 1) + 1) * qscale * | |
200 | 1212 ((int) (quant_matrix[j]))) >> 4; |
0 | 1213 level = (level - 1) | 1; |
1214 level = -level; | |
1215 } else { | |
1216 level = (((level << 1) + 1) * qscale * | |
200 | 1217 ((int) (quant_matrix[j]))) >> 4; |
0 | 1218 level = (level - 1) | 1; |
1219 } | |
1220 #ifdef PARANOID | |
1221 if (level < -2048 || level > 2047) | |
1222 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1223 #endif | |
200 | 1224 block[j] = level; |
0 | 1225 } |
1226 } | |
1227 } | |
1228 } | |
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
|
1229 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1230 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
|
1231 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
|
1232 { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1233 int i, level, qmul, qadd; |
200 | 1234 int nCoeffs; |
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
|
1235 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1236 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
|
1237 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
|
1238 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
|
1239 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
|
1240 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
|
1241 i = 1; |
200 | 1242 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
|
1243 } 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
|
1244 i = 0; |
200 | 1245 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
|
1246 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1247 |
174ef88f619a
use block[] in structure to have it 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 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
|
1249 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
|
1250 |
200 | 1251 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
|
1252 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
|
1253 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
|
1254 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
|
1255 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
|
1256 } 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
|
1257 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
|
1258 } |
174ef88f619a
use block[] in structure to have it 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 #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
|
1260 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
|
1261 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
|
1262 #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
|
1263 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
|
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 } |
174ef88f619a
use block[] in structure to have it 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 } |
0 | 1267 |
1268 /* rate control */ | |
1269 | |
1270 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */ | |
1271 #define I_FRAME_SIZE_RATIO 3.0 | |
1272 #define QSCALE_K 20 | |
1273 | |
1274 static void rate_control_init(MpegEncContext *s) | |
1275 { | |
1276 s->wanted_bits = 0; | |
1277 | |
1278 if (s->intra_only) { | |
1279 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate; | |
1280 s->P_frame_bits = s->I_frame_bits; | |
1281 } else { | |
1282 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) / | |
1283 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1))); | |
1284 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO); | |
1285 } | |
1286 | |
1287 #if defined(DEBUG) | |
1288 printf("I_frame_size=%d P_frame_size=%d\n", | |
1289 s->I_frame_bits, s->P_frame_bits); | |
1290 #endif | |
1291 } | |
1292 | |
1293 | |
1294 /* | |
1295 * This heuristic is rather poor, but at least we do not have to | |
1296 * change the qscale at every macroblock. | |
1297 */ | |
1298 static int rate_estimate_qscale(MpegEncContext *s) | |
1299 { | |
187
3f3b14d3a23d
qscale estimate fix, diff extended to 64bit wide. patch by Stephen Davies <steve@daviesfam.org>
arpi_esp
parents:
178
diff
changeset
|
1300 INT64 diff, total_bits = s->total_bits; |
0 | 1301 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
|
1302 int qscale, qmin; |
0 | 1303 |
1304 if (s->pict_type == I_TYPE) { | |
1305 s->wanted_bits += s->I_frame_bits; | |
1306 } else { | |
1307 s->wanted_bits += s->P_frame_bits; | |
1308 } | |
1309 diff = s->wanted_bits - total_bits; | |
1310 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width); | |
1311 /* adjust for I frame */ | |
1312 if (s->pict_type == I_TYPE && !s->intra_only) { | |
1313 q /= I_FRAME_SIZE_RATIO; | |
1314 } | |
1315 | |
1316 /* using a too small Q scale leeds to problems in mpeg1 and h263 | |
1317 because AC coefficients are clamped to 255 or 127 */ | |
1318 qmin = 3; | |
1319 if (q < qmin) | |
1320 q = qmin; | |
1321 else if (q > 31) | |
1322 q = 31; | |
1323 qscale = (int)(q + 0.5); | |
1324 #if defined(DEBUG) | |
64 | 1325 printf("%d: total=%0.0f br=%0.1f diff=%d qest=%0.1f\n", |
0 | 1326 s->picture_number, |
64 | 1327 (double)total_bits, |
0 | 1328 (float)s->frame_rate / FRAME_RATE_BASE * |
1329 total_bits / s->picture_number, | |
1330 diff, q); | |
1331 #endif | |
1332 return qscale; | |
1333 } | |
1334 | |
1335 AVCodec mpeg1video_encoder = { | |
1336 "mpeg1video", | |
1337 CODEC_TYPE_VIDEO, | |
1338 CODEC_ID_MPEG1VIDEO, | |
1339 sizeof(MpegEncContext), | |
1340 MPV_encode_init, | |
1341 MPV_encode_picture, | |
1342 MPV_encode_end, | |
1343 }; | |
1344 | |
1345 AVCodec h263_encoder = { | |
1346 "h263", | |
1347 CODEC_TYPE_VIDEO, | |
1348 CODEC_ID_H263, | |
1349 sizeof(MpegEncContext), | |
1350 MPV_encode_init, | |
1351 MPV_encode_picture, | |
1352 MPV_encode_end, | |
1353 }; | |
1354 | |
1355 AVCodec h263p_encoder = { | |
1356 "h263p", | |
1357 CODEC_TYPE_VIDEO, | |
1358 CODEC_ID_H263P, | |
1359 sizeof(MpegEncContext), | |
1360 MPV_encode_init, | |
1361 MPV_encode_picture, | |
1362 MPV_encode_end, | |
1363 }; | |
1364 | |
1365 AVCodec rv10_encoder = { | |
1366 "rv10", | |
1367 CODEC_TYPE_VIDEO, | |
1368 CODEC_ID_RV10, | |
1369 sizeof(MpegEncContext), | |
1370 MPV_encode_init, | |
1371 MPV_encode_picture, | |
1372 MPV_encode_end, | |
1373 }; | |
1374 | |
1375 AVCodec mjpeg_encoder = { | |
1376 "mjpeg", | |
1377 CODEC_TYPE_VIDEO, | |
1378 CODEC_ID_MJPEG, | |
1379 sizeof(MpegEncContext), | |
1380 MPV_encode_init, | |
1381 MPV_encode_picture, | |
1382 MPV_encode_end, | |
1383 }; | |
1384 | |
71 | 1385 AVCodec mpeg4_encoder = { |
1386 "mpeg4", | |
0 | 1387 CODEC_TYPE_VIDEO, |
71 | 1388 CODEC_ID_MPEG4, |
0 | 1389 sizeof(MpegEncContext), |
1390 MPV_encode_init, | |
1391 MPV_encode_picture, | |
1392 MPV_encode_end, | |
1393 }; | |
1394 | |
1395 AVCodec msmpeg4_encoder = { | |
1396 "msmpeg4", | |
1397 CODEC_TYPE_VIDEO, | |
1398 CODEC_ID_MSMPEG4, | |
1399 sizeof(MpegEncContext), | |
1400 MPV_encode_init, | |
1401 MPV_encode_picture, | |
1402 MPV_encode_end, | |
1403 }; |