Mercurial > libavcodec.hg
annotate mpegvideo.c @ 86:5e0eb8a361b7 libavcodec
version change
author | glantau |
---|---|
date | Sun, 16 Sep 2001 21:52:17 +0000 |
parents | 82e579c37bc3 |
children | 1e4a4af694d1 |
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; | |
183 } | |
184 /* default structure is frame */ | |
185 s->picture_structure = PICT_FRAME; | |
186 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
187 /* init macroblock skip table */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
188 if (!s->encoding) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
189 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
|
190 if (!s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
191 goto fail; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
192 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
193 |
0 | 194 s->context_initialized = 1; |
195 return 0; | |
196 fail: | |
197 if (s->motion_val) | |
198 free(s->motion_val); | |
199 if (s->dc_val[0]) | |
200 free(s->dc_val[0]); | |
201 if (s->ac_val[0]) | |
202 free(s->ac_val[0]); | |
203 if (s->coded_block) | |
204 free(s->coded_block); | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
205 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
206 free(s->mbskip_table); |
0 | 207 for(i=0;i<3;i++) { |
208 if (s->last_picture_base[i]) | |
209 free(s->last_picture_base[i]); | |
210 if (s->next_picture_base[i]) | |
211 free(s->next_picture_base[i]); | |
212 if (s->aux_picture_base[i]) | |
213 free(s->aux_picture_base[i]); | |
214 } | |
215 return -1; | |
216 } | |
217 | |
218 /* init common structure for both encoder and decoder */ | |
219 void MPV_common_end(MpegEncContext *s) | |
220 { | |
221 int i; | |
222 | |
223 if (s->motion_val) | |
224 free(s->motion_val); | |
225 if (s->h263_pred) { | |
226 free(s->dc_val[0]); | |
227 free(s->ac_val[0]); | |
228 free(s->coded_block); | |
229 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
230 if (s->mbskip_table) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
231 free(s->mbskip_table); |
0 | 232 for(i=0;i<3;i++) { |
233 free(s->last_picture_base[i]); | |
234 free(s->next_picture_base[i]); | |
235 if (s->has_b_frames) | |
236 free(s->aux_picture_base[i]); | |
237 } | |
238 s->context_initialized = 0; | |
239 } | |
240 | |
241 /* init video encoder */ | |
242 int MPV_encode_init(AVCodecContext *avctx) | |
243 { | |
244 MpegEncContext *s = avctx->priv_data; | |
60 | 245 int i; |
0 | 246 |
247 s->bit_rate = avctx->bit_rate; | |
248 s->frame_rate = avctx->frame_rate; | |
249 s->width = avctx->width; | |
250 s->height = avctx->height; | |
251 s->gop_size = avctx->gop_size; | |
252 if (s->gop_size <= 1) { | |
253 s->intra_only = 1; | |
254 s->gop_size = 12; | |
255 } else { | |
256 s->intra_only = 0; | |
257 } | |
258 s->full_search = motion_estimation_method; | |
259 | |
260 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE); | |
261 | |
262 switch(avctx->codec->id) { | |
263 case CODEC_ID_MPEG1VIDEO: | |
264 s->out_format = FMT_MPEG1; | |
265 break; | |
266 case CODEC_ID_MJPEG: | |
267 s->out_format = FMT_MJPEG; | |
268 s->intra_only = 1; /* force intra only for jpeg */ | |
269 if (mjpeg_init(s) < 0) | |
270 return -1; | |
271 break; | |
272 case CODEC_ID_H263: | |
273 if (h263_get_picture_format(s->width, s->height) == 7) | |
274 return -1; | |
275 s->out_format = FMT_H263; | |
276 break; | |
277 case CODEC_ID_H263P: | |
278 s->out_format = FMT_H263; | |
279 s->h263_plus = 1; | |
78 | 280 s->unrestricted_mv = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
281 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
282 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
283 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
284 s->umvplus_dec = 0; |
0 | 285 break; |
286 case CODEC_ID_RV10: | |
287 s->out_format = FMT_H263; | |
288 s->h263_rv10 = 1; | |
289 break; | |
71 | 290 case CODEC_ID_MPEG4: |
0 | 291 s->out_format = FMT_H263; |
292 s->h263_pred = 1; | |
293 s->unrestricted_mv = 1; | |
294 break; | |
295 case CODEC_ID_MSMPEG4: | |
296 s->out_format = FMT_H263; | |
297 s->h263_msmpeg4 = 1; | |
298 s->h263_pred = 1; | |
299 s->unrestricted_mv = 1; | |
300 break; | |
301 default: | |
302 return -1; | |
303 } | |
304 | |
305 if (s->out_format == FMT_H263) | |
306 h263_encode_init_vlc(s); | |
307 | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
308 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
309 |
0 | 310 /* init */ |
311 if (MPV_common_init(s) < 0) | |
312 return -1; | |
313 | |
60 | 314 /* init default q matrix */ |
315 for(i=0;i<64;i++) { | |
316 s->intra_matrix[i] = default_intra_matrix[i]; | |
317 s->non_intra_matrix[i] = default_non_intra_matrix[i]; | |
318 } | |
319 | |
0 | 320 /* rate control init */ |
321 rate_control_init(s); | |
322 | |
323 s->picture_number = 0; | |
324 s->fake_picture_number = 0; | |
325 /* motion detector init */ | |
326 s->f_code = 1; | |
327 | |
328 return 0; | |
329 } | |
330 | |
331 int MPV_encode_end(AVCodecContext *avctx) | |
332 { | |
333 MpegEncContext *s = avctx->priv_data; | |
334 | |
335 #ifdef STATS | |
336 print_stats(); | |
337 #endif | |
338 MPV_common_end(s); | |
339 if (s->out_format == FMT_MJPEG) | |
340 mjpeg_close(s); | |
341 return 0; | |
342 } | |
343 | |
344 /* draw the edges of width 'w' of an image of size width, height */ | |
345 static void draw_edges(UINT8 *buf, int wrap, int width, int height, int w) | |
346 { | |
347 UINT8 *ptr, *last_line; | |
348 int i; | |
349 | |
350 last_line = buf + (height - 1) * wrap; | |
351 for(i=0;i<w;i++) { | |
352 /* top and bottom */ | |
353 memcpy(buf - (i + 1) * wrap, buf, width); | |
354 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
355 } | |
356 /* left and right */ | |
357 ptr = buf; | |
358 for(i=0;i<height;i++) { | |
359 memset(ptr - w, ptr[0], w); | |
360 memset(ptr + width, ptr[width-1], w); | |
361 ptr += wrap; | |
362 } | |
363 /* corners */ | |
364 for(i=0;i<w;i++) { | |
365 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
366 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
367 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
368 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
369 } | |
370 } | |
371 | |
372 /* generic function for encode/decode called before a frame is coded/decoded */ | |
373 void MPV_frame_start(MpegEncContext *s) | |
374 { | |
375 int i; | |
376 UINT8 *tmp; | |
377 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
378 s->mb_skiped = 0; |
0 | 379 if (s->pict_type == B_TYPE) { |
380 for(i=0;i<3;i++) { | |
381 s->current_picture[i] = s->aux_picture[i]; | |
382 } | |
383 } else { | |
384 for(i=0;i<3;i++) { | |
385 /* swap next and last */ | |
386 tmp = s->last_picture[i]; | |
387 s->last_picture[i] = s->next_picture[i]; | |
388 s->next_picture[i] = tmp; | |
389 s->current_picture[i] = tmp; | |
390 } | |
391 } | |
392 } | |
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
|
393 |
0 | 394 /* generic function for encode/decode called after a frame has been coded/decoded */ |
395 void MPV_frame_end(MpegEncContext *s) | |
396 { | |
397 /* draw edge for correct motion prediction if outside */ | |
398 if (s->pict_type != B_TYPE) { | |
399 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH); | |
400 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
401 draw_edges(s->current_picture[2], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2); | |
402 } | |
403 } | |
404 | |
405 int MPV_encode_picture(AVCodecContext *avctx, | |
406 unsigned char *buf, int buf_size, void *data) | |
407 { | |
408 MpegEncContext *s = avctx->priv_data; | |
409 AVPicture *pict = data; | |
410 int i, j; | |
411 | |
412 if (s->fixed_qscale) | |
413 s->qscale = avctx->quality; | |
414 | |
415 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
416 | |
417 if (!s->intra_only) { | |
418 /* first picture of GOP is intra */ | |
419 if ((s->picture_number % s->gop_size) == 0) | |
420 s->pict_type = I_TYPE; | |
421 else | |
422 s->pict_type = P_TYPE; | |
423 } else { | |
424 s->pict_type = I_TYPE; | |
425 } | |
426 avctx->key_frame = (s->pict_type == I_TYPE); | |
427 | |
428 MPV_frame_start(s); | |
429 | |
430 for(i=0;i<3;i++) { | |
431 UINT8 *src = pict->data[i]; | |
432 UINT8 *dest = s->current_picture[i]; | |
433 int src_wrap = pict->linesize[i]; | |
434 int dest_wrap = s->linesize; | |
435 int w = s->width; | |
436 int h = s->height; | |
437 | |
438 if (i >= 1) { | |
439 dest_wrap >>= 1; | |
440 w >>= 1; | |
441 h >>= 1; | |
442 } | |
443 | |
444 for(j=0;j<h;j++) { | |
445 memcpy(dest, src, w); | |
446 dest += dest_wrap; | |
447 src += src_wrap; | |
448 } | |
449 s->new_picture[i] = s->current_picture[i]; | |
450 } | |
451 | |
452 encode_picture(s, s->picture_number); | |
453 | |
454 MPV_frame_end(s); | |
455 s->picture_number++; | |
456 | |
457 if (s->out_format == FMT_MJPEG) | |
458 mjpeg_picture_trailer(s); | |
459 | |
460 flush_put_bits(&s->pb); | |
461 s->total_bits += (s->pb.buf_ptr - s->pb.buf) * 8; | |
462 avctx->quality = s->qscale; | |
463 return s->pb.buf_ptr - s->pb.buf; | |
464 } | |
465 | |
466 static inline int clip(int a, int amin, int amax) | |
467 { | |
468 if (a < amin) | |
469 return amin; | |
470 else if (a > amax) | |
471 return amax; | |
472 else | |
473 return a; | |
474 } | |
475 | |
476 /* apply one mpeg motion vector to the three components */ | |
477 static inline void mpeg_motion(MpegEncContext *s, | |
478 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
479 int dest_offset, | |
480 UINT8 **ref_picture, int src_offset, | |
481 int field_based, op_pixels_func *pix_op, | |
482 int motion_x, int motion_y, int h) | |
483 { | |
484 UINT8 *ptr; | |
485 int dxy, offset, mx, my, src_x, src_y, height, linesize; | |
486 | |
487 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
488 src_x = s->mb_x * 16 + (motion_x >> 1); | |
489 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
490 | |
491 /* WARNING: do no forget half pels */ | |
492 height = s->height >> field_based; | |
493 src_x = clip(src_x, -16, s->width); | |
494 if (src_x == s->width) | |
495 dxy &= ~1; | |
496 src_y = clip(src_y, -16, height); | |
497 if (src_y == height) | |
498 dxy &= ~2; | |
499 linesize = s->linesize << field_based; | |
500 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; | |
501 dest_y += dest_offset; | |
502 pix_op[dxy](dest_y, ptr, linesize, h); | |
503 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
504 | |
505 if (s->out_format == FMT_H263) { | |
506 dxy = 0; | |
507 if ((motion_x & 3) != 0) | |
508 dxy |= 1; | |
509 if ((motion_y & 3) != 0) | |
510 dxy |= 2; | |
511 mx = motion_x >> 2; | |
512 my = motion_y >> 2; | |
513 } else { | |
514 mx = motion_x / 2; | |
515 my = motion_y / 2; | |
516 dxy = ((my & 1) << 1) | (mx & 1); | |
517 mx >>= 1; | |
518 my >>= 1; | |
519 } | |
520 | |
521 src_x = s->mb_x * 8 + mx; | |
522 src_y = s->mb_y * (8 >> field_based) + my; | |
523 src_x = clip(src_x, -8, s->width >> 1); | |
524 if (src_x == (s->width >> 1)) | |
525 dxy &= ~1; | |
526 src_y = clip(src_y, -8, height >> 1); | |
527 if (src_y == (height >> 1)) | |
528 dxy &= ~2; | |
529 | |
530 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1); | |
531 ptr = ref_picture[1] + offset; | |
532 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
533 ptr = ref_picture[2] + offset; | |
534 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1); | |
535 } | |
536 | |
537 static inline void MPV_motion(MpegEncContext *s, | |
538 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
539 int dir, UINT8 **ref_picture, | |
540 op_pixels_func *pix_op) | |
541 { | |
542 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
543 int mb_x, mb_y, i; | |
544 UINT8 *ptr, *dest; | |
545 | |
546 mb_x = s->mb_x; | |
547 mb_y = s->mb_y; | |
548 | |
549 switch(s->mv_type) { | |
550 case MV_TYPE_16X16: | |
551 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
552 ref_picture, 0, | |
553 0, pix_op, | |
554 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
555 break; | |
556 case MV_TYPE_8X8: | |
557 for(i=0;i<4;i++) { | |
558 motion_x = s->mv[dir][i][0]; | |
559 motion_y = s->mv[dir][i][1]; | |
560 | |
561 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
562 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
563 src_y = mb_y * 16 + (motion_y >> 1) + ((i >> 1) & 1) * 8; | |
564 | |
565 /* WARNING: do no forget half pels */ | |
566 src_x = clip(src_x, -16, s->width); | |
567 if (src_x == s->width) | |
568 dxy &= ~1; | |
569 src_y = clip(src_y, -16, s->height); | |
570 if (src_y == s->height) | |
571 dxy &= ~2; | |
572 | |
573 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
574 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
575 pix_op[dxy](dest, ptr, s->linesize, 8); | |
576 } | |
577 /* In case of 8X8, we construct a single chroma motion vector | |
578 with a special rounding */ | |
579 mx = 0; | |
580 my = 0; | |
581 for(i=0;i<4;i++) { | |
582 mx += s->mv[dir][i][0]; | |
583 my += s->mv[dir][i][1]; | |
584 } | |
585 if (mx >= 0) | |
586 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
587 else { | |
588 mx = -mx; | |
589 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
590 } | |
591 if (my >= 0) | |
592 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
593 else { | |
594 my = -my; | |
595 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
596 } | |
597 dxy = ((my & 1) << 1) | (mx & 1); | |
598 mx >>= 1; | |
599 my >>= 1; | |
600 | |
601 src_x = mb_x * 8 + mx; | |
602 src_y = mb_y * 8 + my; | |
603 src_x = clip(src_x, -8, s->width/2); | |
604 if (src_x == s->width/2) | |
605 dxy &= ~1; | |
606 src_y = clip(src_y, -8, s->height/2); | |
607 if (src_y == s->height/2) | |
608 dxy &= ~2; | |
609 | |
610 offset = (src_y * (s->linesize >> 1)) + src_x; | |
611 ptr = ref_picture[1] + offset; | |
612 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8); | |
613 ptr = ref_picture[2] + offset; | |
614 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8); | |
615 break; | |
616 case MV_TYPE_FIELD: | |
617 if (s->picture_structure == PICT_FRAME) { | |
618 /* top field */ | |
619 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
620 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
621 1, pix_op, | |
622 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
623 /* bottom field */ | |
624 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
625 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
626 1, pix_op, | |
627 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
628 } else { | |
629 | |
630 | |
631 } | |
632 break; | |
633 } | |
634 } | |
635 | |
636 | |
637 /* put block[] to dest[] */ | |
638 static inline void put_dct(MpegEncContext *s, | |
639 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
640 { | |
641 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
|
642 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
|
643 ff_idct (block); |
0 | 644 put_pixels_clamped(block, dest, line_size); |
645 } | |
646 | |
647 /* add block[] to dest[] */ | |
648 static inline void add_dct(MpegEncContext *s, | |
649 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
650 { | |
651 if (s->block_last_index[i] >= 0) { | |
652 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
|
653 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
|
654 ff_idct (block); |
0 | 655 add_pixels_clamped(block, dest, line_size); |
656 } | |
657 } | |
658 | |
659 /* generic function called after a macroblock has been parsed by the | |
660 decoder or after it has been encoded by the encoder. | |
661 | |
662 Important variables used: | |
663 s->mb_intra : true if intra macroblock | |
664 s->mv_dir : motion vector direction | |
665 s->mv_type : motion vector type | |
666 s->mv : motion vector | |
667 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
668 */ | |
669 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
670 { | |
671 int mb_x, mb_y, motion_x, motion_y; | |
672 int dct_linesize, dct_offset; | |
673 op_pixels_func *op_pix; | |
674 | |
675 mb_x = s->mb_x; | |
676 mb_y = s->mb_y; | |
677 | |
678 /* update DC predictors for P macroblocks */ | |
679 if (!s->mb_intra) { | |
680 if (s->h263_pred) { | |
681 int wrap, x, y, v; | |
682 wrap = 2 * s->mb_width + 2; | |
683 v = 1024; | |
684 x = 2 * mb_x + 1; | |
685 y = 2 * mb_y + 1; | |
686 s->dc_val[0][(x) + (y) * wrap] = v; | |
687 s->dc_val[0][(x + 1) + (y) * wrap] = v; | |
688 s->dc_val[0][(x) + (y + 1) * wrap] = v; | |
689 s->dc_val[0][(x + 1) + (y + 1) * wrap] = v; | |
690 /* ac pred */ | |
691 memset(s->ac_val[0][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
692 memset(s->ac_val[0][(x + 1) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
693 memset(s->ac_val[0][(x) + (y + 1) * wrap], 0, 16 * sizeof(INT16)); | |
694 memset(s->ac_val[0][(x + 1) + (y + 1) * wrap], 0, 16 * sizeof(INT16)); | |
695 if (s->h263_msmpeg4) { | |
696 s->coded_block[(x) + (y) * wrap] = 0; | |
697 s->coded_block[(x + 1) + (y) * wrap] = 0; | |
698 s->coded_block[(x) + (y + 1) * wrap] = 0; | |
699 s->coded_block[(x + 1) + (y + 1) * wrap] = 0; | |
700 } | |
701 /* chroma */ | |
702 wrap = s->mb_width + 2; | |
703 x = mb_x + 1; | |
704 y = mb_y + 1; | |
705 s->dc_val[1][(x) + (y) * wrap] = v; | |
706 s->dc_val[2][(x) + (y) * wrap] = v; | |
707 /* ac pred */ | |
708 memset(s->ac_val[1][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
709 memset(s->ac_val[2][(x) + (y) * wrap], 0, 16 * sizeof(INT16)); | |
710 } else { | |
711 s->last_dc[0] = 128 << s->intra_dc_precision; | |
712 s->last_dc[1] = 128 << s->intra_dc_precision; | |
713 s->last_dc[2] = 128 << s->intra_dc_precision; | |
714 } | |
715 } | |
716 | |
717 /* update motion predictor */ | |
718 if (s->out_format == FMT_H263) { | |
719 int x, y, wrap; | |
720 | |
721 x = 2 * mb_x + 1; | |
722 y = 2 * mb_y + 1; | |
723 wrap = 2 * s->mb_width + 2; | |
724 if (s->mb_intra) { | |
725 motion_x = 0; | |
726 motion_y = 0; | |
727 goto motion_init; | |
728 } else if (s->mv_type == MV_TYPE_16X16) { | |
729 motion_x = s->mv[0][0][0]; | |
730 motion_y = s->mv[0][0][1]; | |
731 motion_init: | |
732 /* no update if 8X8 because it has been done during parsing */ | |
733 s->motion_val[(x) + (y) * wrap][0] = motion_x; | |
734 s->motion_val[(x) + (y) * wrap][1] = motion_y; | |
735 s->motion_val[(x + 1) + (y) * wrap][0] = motion_x; | |
736 s->motion_val[(x + 1) + (y) * wrap][1] = motion_y; | |
737 s->motion_val[(x) + (y + 1) * wrap][0] = motion_x; | |
738 s->motion_val[(x) + (y + 1) * wrap][1] = motion_y; | |
739 s->motion_val[(x + 1) + (y + 1) * wrap][0] = motion_x; | |
740 s->motion_val[(x + 1) + (y + 1) * wrap][1] = motion_y; | |
741 } | |
742 } | |
743 | |
744 if (!s->intra_only) { | |
745 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
|
746 UINT8 *mbskip_ptr; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
747 |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
748 /* 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
|
749 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
|
750 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
|
751 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
752 s->mb_skiped = 0; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
753 /* 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
|
754 if (*mbskip_ptr != 0) |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
755 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
756 *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
|
757 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
758 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
759 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
760 } |
0 | 761 |
762 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16; | |
763 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
764 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8; | |
765 | |
766 if (s->interlaced_dct) { | |
767 dct_linesize = s->linesize * 2; | |
768 dct_offset = s->linesize; | |
769 } else { | |
770 dct_linesize = s->linesize; | |
771 dct_offset = s->linesize * 8; | |
772 } | |
773 | |
774 if (!s->mb_intra) { | |
775 /* motion handling */ | |
776 if (!s->no_rounding) | |
777 op_pix = put_pixels_tab; | |
778 else | |
779 op_pix = put_no_rnd_pixels_tab; | |
780 | |
781 if (s->mv_dir & MV_DIR_FORWARD) { | |
782 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix); | |
783 if (!s->no_rounding) | |
784 op_pix = avg_pixels_tab; | |
785 else | |
786 op_pix = avg_no_rnd_pixels_tab; | |
787 } | |
788 if (s->mv_dir & MV_DIR_BACKWARD) { | |
789 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix); | |
790 } | |
791 | |
792 /* add dct residue */ | |
793 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
794 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
795 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
796 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
797 | |
57 | 798 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
799 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 800 } else { |
801 /* dct only in intra block */ | |
802 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
803 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
804 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
805 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
806 | |
57 | 807 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1); |
808 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1); | |
0 | 809 } |
810 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
811 the_end: |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
812 emms_c(); |
0 | 813 } |
814 | |
815 static void encode_picture(MpegEncContext *s, int picture_number) | |
816 { | |
817 int mb_x, mb_y, wrap; | |
818 UINT8 *ptr; | |
819 int i, motion_x, motion_y; | |
820 | |
821 s->picture_number = picture_number; | |
822 if (!s->fixed_qscale) | |
823 s->qscale = rate_estimate_qscale(s); | |
824 | |
825 /* precompute matrix */ | |
826 if (s->out_format == FMT_MJPEG) { | |
827 /* for mjpeg, we do include qscale in the matrix */ | |
828 s->intra_matrix[0] = default_intra_matrix[0]; | |
829 for(i=1;i<64;i++) | |
830 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; | |
831 convert_matrix(s->q_intra_matrix, s->intra_matrix, 8); | |
832 } else { | |
833 convert_matrix(s->q_intra_matrix, s->intra_matrix, s->qscale); | |
834 convert_matrix(s->q_non_intra_matrix, s->non_intra_matrix, s->qscale); | |
835 } | |
836 | |
837 switch(s->out_format) { | |
838 case FMT_MJPEG: | |
839 mjpeg_picture_header(s); | |
840 break; | |
841 case FMT_H263: | |
842 if (s->h263_msmpeg4) | |
843 msmpeg4_encode_picture_header(s, picture_number); | |
844 else if (s->h263_pred) | |
845 mpeg4_encode_picture_header(s, picture_number); | |
846 else if (s->h263_rv10) | |
847 rv10_encode_picture_header(s, picture_number); | |
848 else | |
849 h263_encode_picture_header(s, picture_number); | |
850 break; | |
851 case FMT_MPEG1: | |
852 mpeg1_encode_picture_header(s, picture_number); | |
853 break; | |
854 } | |
855 | |
856 /* init last dc values */ | |
857 /* note: quant matrix value (8) is implied here */ | |
858 s->last_dc[0] = 128; | |
859 s->last_dc[1] = 128; | |
860 s->last_dc[2] = 128; | |
861 s->mb_incr = 1; | |
862 s->last_mv[0][0][0] = 0; | |
863 s->last_mv[0][0][1] = 0; | |
864 s->mv_type = MV_TYPE_16X16; | |
865 s->mv_dir = MV_DIR_FORWARD; | |
866 | |
867 for(mb_y=0; mb_y < s->mb_height; mb_y++) { | |
868 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
869 | |
870 s->mb_x = mb_x; | |
871 s->mb_y = mb_y; | |
872 | |
873 /* compute motion vector and macro block type (intra or non intra) */ | |
874 motion_x = 0; | |
875 motion_y = 0; | |
876 if (s->pict_type == P_TYPE) { | |
877 s->mb_intra = estimate_motion(s, mb_x, mb_y, | |
878 &motion_x, | |
879 &motion_y); | |
880 } else { | |
881 s->mb_intra = 1; | |
882 } | |
883 | |
884 /* get the pixels */ | |
885 wrap = s->linesize; | |
886 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
|
887 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
|
888 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
|
889 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
|
890 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); |
0 | 891 wrap = s->linesize >> 1; |
892 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
|
893 get_pixels(s->block[4], ptr, wrap); |
0 | 894 |
895 wrap = s->linesize >> 1; | |
896 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
|
897 get_pixels(s->block[5], ptr, wrap); |
0 | 898 |
899 /* subtract previous frame if non intra */ | |
900 if (!s->mb_intra) { | |
901 int dxy, offset, mx, my; | |
902 | |
903 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
904 ptr = s->last_picture[0] + | |
905 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) + | |
906 (mb_x * 16 + (motion_x >> 1)); | |
907 | |
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
|
908 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
|
909 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
|
910 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
|
911 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy); |
0 | 912 |
913 if (s->out_format == FMT_H263) { | |
914 /* special rounding for h263 */ | |
915 dxy = 0; | |
916 if ((motion_x & 3) != 0) | |
917 dxy |= 1; | |
918 if ((motion_y & 3) != 0) | |
919 dxy |= 2; | |
920 mx = motion_x >> 2; | |
921 my = motion_y >> 2; | |
922 } else { | |
923 mx = motion_x / 2; | |
924 my = motion_y / 2; | |
925 dxy = ((my & 1) << 1) | (mx & 1); | |
926 mx >>= 1; | |
927 my >>= 1; | |
928 } | |
929 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx); | |
930 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
|
931 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy); |
0 | 932 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
|
933 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy); |
0 | 934 } |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
935 emms_c(); |
0 | 936 |
937 /* DCT & quantize */ | |
938 if (s->h263_msmpeg4) { | |
939 msmpeg4_dc_scale(s); | |
940 } else if (s->h263_pred) { | |
941 h263_dc_scale(s); | |
942 } else { | |
943 /* default quantization values */ | |
944 s->y_dc_scale = 8; | |
945 s->c_dc_scale = 8; | |
946 } | |
947 | |
948 for(i=0;i<6;i++) { | |
949 int last_index; | |
950 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
|
951 last_index = dct_quantize(s, s->block[i], i, s->qscale); |
0 | 952 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
|
953 last_index = dct_quantize_mmx(s, s->block[i], i, s->qscale); |
0 | 954 s->block_last_index[i] = last_index; |
955 } | |
956 | |
957 /* huffman encode */ | |
958 switch(s->out_format) { | |
959 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
|
960 mpeg1_encode_mb(s, s->block, motion_x, motion_y); |
0 | 961 break; |
962 case FMT_H263: | |
963 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
|
964 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); |
0 | 965 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
|
966 h263_encode_mb(s, s->block, motion_x, motion_y); |
0 | 967 break; |
968 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
|
969 mjpeg_encode_mb(s, s->block); |
0 | 970 break; |
971 } | |
972 | |
973 /* decompress blocks so that we keep the state of the decoder */ | |
974 s->mv[0][0][0] = motion_x; | |
975 s->mv[0][0][1] = motion_y; | |
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 MPV_decode_mb(s, s->block); |
0 | 978 } |
979 } | |
980 } | |
981 | |
982 static int dct_quantize(MpegEncContext *s, | |
983 DCTELEM *block, int n, | |
984 int qscale) | |
985 { | |
986 int i, j, level, last_non_zero, q; | |
987 const int *qmat; | |
988 | |
989 av_fdct (block); | |
990 | |
64 | 991 /* we need this permutation so that we correct the IDCT |
992 permutation. will be moved into DCT code */ | |
993 block_permute(block); | |
994 | |
0 | 995 if (s->mb_intra) { |
996 if (n < 4) | |
997 q = s->y_dc_scale; | |
998 else | |
999 q = s->c_dc_scale; | |
1000 q = q << 3; | |
1001 | |
1002 /* note: block[0] is assumed to be positive */ | |
1003 block[0] = (block[0] + (q >> 1)) / q; | |
1004 i = 1; | |
1005 last_non_zero = 0; | |
1006 if (s->out_format == FMT_H263) { | |
1007 qmat = s->q_non_intra_matrix; | |
1008 } else { | |
1009 qmat = s->q_intra_matrix; | |
1010 } | |
1011 } else { | |
1012 i = 0; | |
1013 last_non_zero = -1; | |
1014 qmat = s->q_non_intra_matrix; | |
1015 } | |
1016 | |
1017 for(;i<64;i++) { | |
1018 j = zigzag_direct[i]; | |
1019 level = block[j]; | |
1020 level = level * qmat[j]; | |
1021 #ifdef PARANOID | |
1022 { | |
1023 static int count = 0; | |
1024 int level1, level2, qmat1; | |
1025 double val; | |
1026 if (qmat == s->q_non_intra_matrix) { | |
1027 qmat1 = default_non_intra_matrix[j] * s->qscale; | |
1028 } else { | |
1029 qmat1 = default_intra_matrix[j] * s->qscale; | |
1030 } | |
1031 if (av_fdct != jpeg_fdct_ifast) | |
1032 val = ((double)block[j] * 8.0) / (double)qmat1; | |
1033 else | |
1034 val = ((double)block[j] * 8.0 * 2048.0) / | |
1035 ((double)qmat1 * aanscales[j]); | |
1036 level1 = (int)val; | |
1037 level2 = level / (1 << (QMAT_SHIFT - 3)); | |
1038 if (level1 != level2) { | |
1039 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", | |
1040 count, level2, level1, block[j], qmat1, qmat[j], | |
1041 val); | |
1042 count++; | |
1043 } | |
1044 | |
1045 } | |
1046 #endif | |
1047 /* XXX: slight error for the low range. Test should be equivalent to | |
1048 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << | |
1049 (QMAT_SHIFT - 3))) | |
1050 */ | |
1051 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != | |
1052 level) { | |
1053 level = level / (1 << (QMAT_SHIFT - 3)); | |
1054 /* XXX: currently, this code is not optimal. the range should be: | |
1055 mpeg1: -255..255 | |
1056 mpeg2: -2048..2047 | |
1057 h263: -128..127 | |
1058 mpeg4: -2048..2047 | |
1059 */ | |
1060 if (level > 127) | |
1061 level = 127; | |
1062 else if (level < -128) | |
1063 level = -128; | |
1064 block[j] = level; | |
1065 last_non_zero = i; | |
1066 } else { | |
1067 block[j] = 0; | |
1068 } | |
1069 } | |
1070 return last_non_zero; | |
1071 } | |
1072 | |
1073 static int dct_quantize_mmx(MpegEncContext *s, | |
1074 DCTELEM *block, int n, | |
1075 int qscale) | |
1076 { | |
1077 int i, j, level, last_non_zero, q; | |
1078 const int *qmat; | |
1079 | |
1080 av_fdct (block); | |
40
f324d9047ca3
fixed block permutation in encoder (not optimal - should move it in forward DCT code)
glantau
parents:
19
diff
changeset
|
1081 |
f324d9047ca3
fixed block permutation in encoder (not optimal - should move it in forward DCT code)
glantau
parents:
19
diff
changeset
|
1082 /* 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
|
1083 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
|
1084 block_permute(block); |
0 | 1085 |
1086 if (s->mb_intra) { | |
1087 if (n < 4) | |
1088 q = s->y_dc_scale; | |
1089 else | |
1090 q = s->c_dc_scale; | |
1091 | |
1092 /* note: block[0] is assumed to be positive */ | |
1093 block[0] = (block[0] + (q >> 1)) / q; | |
1094 i = 1; | |
1095 last_non_zero = 0; | |
1096 if (s->out_format == FMT_H263) { | |
1097 qmat = s->q_non_intra_matrix; | |
1098 } else { | |
1099 qmat = s->q_intra_matrix; | |
1100 } | |
1101 } else { | |
1102 i = 0; | |
1103 last_non_zero = -1; | |
1104 qmat = s->q_non_intra_matrix; | |
1105 } | |
1106 | |
1107 for(;i<64;i++) { | |
1108 j = zigzag_direct[i]; | |
1109 level = block[j]; | |
1110 level = level * qmat[j]; | |
1111 /* XXX: slight error for the low range. Test should be equivalent to | |
1112 (level <= -(1 << (QMAT_SHIFT_MMX - 3)) || level >= (1 << | |
1113 (QMAT_SHIFT_MMX - 3))) | |
1114 */ | |
1115 if (((level << (31 - (QMAT_SHIFT_MMX - 3))) >> (31 - (QMAT_SHIFT_MMX - 3))) != | |
1116 level) { | |
1117 level = level / (1 << (QMAT_SHIFT_MMX - 3)); | |
1118 /* XXX: currently, this code is not optimal. the range should be: | |
1119 mpeg1: -255..255 | |
1120 mpeg2: -2048..2047 | |
1121 h263: -128..127 | |
1122 mpeg4: -2048..2047 | |
1123 */ | |
1124 if (level > 127) | |
1125 level = 127; | |
1126 else if (level < -128) | |
1127 level = -128; | |
1128 block[j] = level; | |
1129 last_non_zero = i; | |
1130 } else { | |
1131 block[j] = 0; | |
1132 } | |
1133 } | |
1134 return last_non_zero; | |
1135 } | |
1136 | |
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
|
1137 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
|
1138 DCTELEM *block, int n, int qscale) |
0 | 1139 { |
1140 int i, level; | |
1141 const UINT16 *quant_matrix; | |
1142 | |
1143 if (s->mb_intra) { | |
1144 if (n < 4) | |
1145 block[0] = block[0] * s->y_dc_scale; | |
1146 else | |
1147 block[0] = block[0] * s->c_dc_scale; | |
1148 /* XXX: only mpeg1 */ | |
1149 quant_matrix = s->intra_matrix; | |
1150 for(i=1;i<64;i++) { | |
1151 level = block[i]; | |
1152 if (level) { | |
1153 if (level < 0) { | |
1154 level = -level; | |
1155 level = (int)(level * qscale * quant_matrix[i]) >> 3; | |
1156 level = (level - 1) | 1; | |
1157 level = -level; | |
1158 } else { | |
1159 level = (int)(level * qscale * quant_matrix[i]) >> 3; | |
1160 level = (level - 1) | 1; | |
1161 } | |
1162 #ifdef PARANOID | |
1163 if (level < -2048 || level > 2047) | |
1164 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1165 #endif | |
1166 block[i] = level; | |
1167 } | |
1168 } | |
1169 } else { | |
1170 i = 0; | |
1171 quant_matrix = s->non_intra_matrix; | |
1172 for(;i<64;i++) { | |
1173 level = block[i]; | |
1174 if (level) { | |
1175 if (level < 0) { | |
1176 level = -level; | |
1177 level = (((level << 1) + 1) * qscale * | |
1178 ((int) (quant_matrix[i]))) >> 4; | |
1179 level = (level - 1) | 1; | |
1180 level = -level; | |
1181 } else { | |
1182 level = (((level << 1) + 1) * qscale * | |
1183 ((int) (quant_matrix[i]))) >> 4; | |
1184 level = (level - 1) | 1; | |
1185 } | |
1186 #ifdef PARANOID | |
1187 if (level < -2048 || level > 2047) | |
1188 fprintf(stderr, "unquant error %d %d\n", i, level); | |
1189 #endif | |
1190 block[i] = level; | |
1191 } | |
1192 } | |
1193 } | |
1194 } | |
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
|
1195 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1196 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
|
1197 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
|
1198 { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1199 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
|
1200 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1201 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
|
1202 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
|
1203 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
|
1204 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
|
1205 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
|
1206 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
|
1207 } 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
|
1208 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
|
1209 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1210 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1211 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
|
1212 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
|
1213 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1214 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
|
1215 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
|
1216 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
|
1217 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
|
1218 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
|
1219 } 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
|
1220 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
|
1221 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1222 #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
|
1223 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
|
1224 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
|
1225 #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
|
1226 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
|
1227 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
1228 } |
174ef88f619a
use block[] in structure to have it 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 } |
0 | 1230 |
1231 /* rate control */ | |
1232 | |
1233 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */ | |
1234 #define I_FRAME_SIZE_RATIO 3.0 | |
1235 #define QSCALE_K 20 | |
1236 | |
1237 static void rate_control_init(MpegEncContext *s) | |
1238 { | |
1239 s->wanted_bits = 0; | |
1240 | |
1241 if (s->intra_only) { | |
1242 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate; | |
1243 s->P_frame_bits = s->I_frame_bits; | |
1244 } else { | |
1245 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) / | |
1246 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1))); | |
1247 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO); | |
1248 } | |
1249 | |
1250 #if defined(DEBUG) | |
1251 printf("I_frame_size=%d P_frame_size=%d\n", | |
1252 s->I_frame_bits, s->P_frame_bits); | |
1253 #endif | |
1254 } | |
1255 | |
1256 | |
1257 /* | |
1258 * This heuristic is rather poor, but at least we do not have to | |
1259 * change the qscale at every macroblock. | |
1260 */ | |
1261 static int rate_estimate_qscale(MpegEncContext *s) | |
1262 { | |
64 | 1263 INT64 total_bits = s->total_bits; |
0 | 1264 float q; |
1265 int qscale, diff, qmin; | |
1266 | |
1267 if (s->pict_type == I_TYPE) { | |
1268 s->wanted_bits += s->I_frame_bits; | |
1269 } else { | |
1270 s->wanted_bits += s->P_frame_bits; | |
1271 } | |
1272 diff = s->wanted_bits - total_bits; | |
1273 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width); | |
1274 /* adjust for I frame */ | |
1275 if (s->pict_type == I_TYPE && !s->intra_only) { | |
1276 q /= I_FRAME_SIZE_RATIO; | |
1277 } | |
1278 | |
1279 /* using a too small Q scale leeds to problems in mpeg1 and h263 | |
1280 because AC coefficients are clamped to 255 or 127 */ | |
1281 qmin = 3; | |
1282 if (q < qmin) | |
1283 q = qmin; | |
1284 else if (q > 31) | |
1285 q = 31; | |
1286 qscale = (int)(q + 0.5); | |
1287 #if defined(DEBUG) | |
64 | 1288 printf("%d: total=%0.0f br=%0.1f diff=%d qest=%0.1f\n", |
0 | 1289 s->picture_number, |
64 | 1290 (double)total_bits, |
0 | 1291 (float)s->frame_rate / FRAME_RATE_BASE * |
1292 total_bits / s->picture_number, | |
1293 diff, q); | |
1294 #endif | |
1295 return qscale; | |
1296 } | |
1297 | |
1298 AVCodec mpeg1video_encoder = { | |
1299 "mpeg1video", | |
1300 CODEC_TYPE_VIDEO, | |
1301 CODEC_ID_MPEG1VIDEO, | |
1302 sizeof(MpegEncContext), | |
1303 MPV_encode_init, | |
1304 MPV_encode_picture, | |
1305 MPV_encode_end, | |
1306 }; | |
1307 | |
1308 AVCodec h263_encoder = { | |
1309 "h263", | |
1310 CODEC_TYPE_VIDEO, | |
1311 CODEC_ID_H263, | |
1312 sizeof(MpegEncContext), | |
1313 MPV_encode_init, | |
1314 MPV_encode_picture, | |
1315 MPV_encode_end, | |
1316 }; | |
1317 | |
1318 AVCodec h263p_encoder = { | |
1319 "h263p", | |
1320 CODEC_TYPE_VIDEO, | |
1321 CODEC_ID_H263P, | |
1322 sizeof(MpegEncContext), | |
1323 MPV_encode_init, | |
1324 MPV_encode_picture, | |
1325 MPV_encode_end, | |
1326 }; | |
1327 | |
1328 AVCodec rv10_encoder = { | |
1329 "rv10", | |
1330 CODEC_TYPE_VIDEO, | |
1331 CODEC_ID_RV10, | |
1332 sizeof(MpegEncContext), | |
1333 MPV_encode_init, | |
1334 MPV_encode_picture, | |
1335 MPV_encode_end, | |
1336 }; | |
1337 | |
1338 AVCodec mjpeg_encoder = { | |
1339 "mjpeg", | |
1340 CODEC_TYPE_VIDEO, | |
1341 CODEC_ID_MJPEG, | |
1342 sizeof(MpegEncContext), | |
1343 MPV_encode_init, | |
1344 MPV_encode_picture, | |
1345 MPV_encode_end, | |
1346 }; | |
1347 | |
71 | 1348 AVCodec mpeg4_encoder = { |
1349 "mpeg4", | |
0 | 1350 CODEC_TYPE_VIDEO, |
71 | 1351 CODEC_ID_MPEG4, |
0 | 1352 sizeof(MpegEncContext), |
1353 MPV_encode_init, | |
1354 MPV_encode_picture, | |
1355 MPV_encode_end, | |
1356 }; | |
1357 | |
1358 AVCodec msmpeg4_encoder = { | |
1359 "msmpeg4", | |
1360 CODEC_TYPE_VIDEO, | |
1361 CODEC_ID_MSMPEG4, | |
1362 sizeof(MpegEncContext), | |
1363 MPV_encode_init, | |
1364 MPV_encode_picture, | |
1365 MPV_encode_end, | |
1366 }; |