Mercurial > libavcodec.hg
annotate mpegvideo.c @ 591:d3f374218030 libavcodec
fixing mpeg4 mpeg quantizers
author | michaelni |
---|---|
date | Fri, 02 Aug 2002 20:27:46 +0000 |
parents | 9e271424160f |
children | 909e50c67d0e |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest mpeg encoder (well, it was the simplest!) | |
429 | 3 * Copyright (c) 2000,2001 Fabrice Bellard. |
0 | 4 * |
429 | 5 * This library is free software; you can redistribute it and/or |
6 * modify it under the terms of the GNU Lesser General Public | |
7 * License as published by the Free Software Foundation; either | |
8 * version 2 of the License, or (at your option) any later version. | |
0 | 9 * |
429 | 10 * This library is distributed in the hope that it will be useful, |
0 | 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 * Lesser General Public License for more details. | |
0 | 14 * |
429 | 15 * You should have received a copy of the GNU Lesser General Public |
16 * License along with this library; if not, write to the Free Software | |
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
295 | 18 * |
325 | 19 * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at> |
0 | 20 */ |
21 #include "avcodec.h" | |
22 #include "dsputil.h" | |
23 #include "mpegvideo.h" | |
24 | |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
25 #ifdef USE_FASTMEMCPY |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
26 #include "fastmemcpy.h" |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
27 #endif |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
28 |
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
|
29 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
|
30 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
|
31 DCTELEM *block, int n, int qscale); |
325 | 32 static void dct_unquantize_mpeg2_c(MpegEncContext *s, |
33 DCTELEM *block, int n, int qscale); | |
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
|
34 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
|
35 DCTELEM *block, int n, int qscale); |
206 | 36 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
344 | 37 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow); |
206 | 38 |
344 | 39 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow)= dct_quantize_c; |
206 | 40 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
566 | 41 static void emulated_edge_mc(MpegEncContext *s, UINT8 *src, int linesize, int block_w, int block_h, |
42 int src_x, int src_y, int w, int h); | |
206 | 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 | |
321 | 51 |
0 | 52 /* for jpeg fast DCT */ |
53 #define CONST_BITS 14 | |
54 | |
55 static const unsigned short aanscales[64] = { | |
56 /* precomputed values scaled up by 14 bits */ | |
57 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
58 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
59 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
60 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
61 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
62 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
63 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
64 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
65 }; | |
66 | |
67 static UINT8 h263_chroma_roundtab[16] = { | |
68 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
69 }; | |
70 | |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
71 static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1]; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
72 static UINT8 default_fcode_tab[MAX_MV*2+1]; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
73 |
321 | 74 extern UINT8 zigzag_end[64]; |
75 | |
0 | 76 /* default motion estimation */ |
321 | 77 int motion_estimation_method = ME_EPZS; |
200 | 78 |
344 | 79 static void convert_matrix(int (*qmat)[64], uint16_t (*qmat16)[64], uint16_t (*qmat16_bias)[64], |
80 const UINT16 *quant_matrix, int bias) | |
0 | 81 { |
344 | 82 int qscale; |
0 | 83 |
344 | 84 for(qscale=1; qscale<32; qscale++){ |
85 int i; | |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
467
diff
changeset
|
86 if (av_fdct == fdct_ifast) { |
344 | 87 for(i=0;i<64;i++) { |
88 const int j= block_permute_op(i); | |
89 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
90 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ | |
91 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
92 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
93 | |
94 qmat[qscale][j] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / | |
95 (aanscales[i] * qscale * quant_matrix[j])); | |
96 } | |
97 } else { | |
98 for(i=0;i<64;i++) { | |
99 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
100 So 16 <= qscale * quant_matrix[i] <= 7905 | |
101 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
102 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
103 */ | |
104 qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); | |
105 qmat16[qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
106 | |
107 if(qmat16[qscale][i]==0 || qmat16[qscale][i]==128*256) qmat16[qscale][i]=128*256-1; | |
108 | |
109 qmat16_bias[qscale][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][i]); | |
110 } | |
0 | 111 } |
112 } | |
113 } | |
456 | 114 // move into common.c perhaps |
115 #define CHECKED_ALLOCZ(p, size)\ | |
116 {\ | |
117 p= av_mallocz(size);\ | |
118 if(p==NULL){\ | |
119 perror("malloc");\ | |
120 goto fail;\ | |
121 }\ | |
122 } | |
0 | 123 |
124 /* init common structure for both encoder and decoder */ | |
125 int MPV_common_init(MpegEncContext *s) | |
126 { | |
127 int c_size, i; | |
128 UINT8 *pict; | |
129 | |
312 | 130 s->dct_unquantize_h263 = dct_unquantize_h263_c; |
325 | 131 s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_c; |
132 s->dct_unquantize_mpeg2 = dct_unquantize_mpeg2_c; | |
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
|
133 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
134 #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
|
135 MPV_common_init_mmx(s); |
8 | 136 #endif |
514
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M«©ns Rullg«©rd
mellum
parents:
506
diff
changeset
|
137 #ifdef ARCH_ALPHA |
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M«©ns Rullg«©rd
mellum
parents:
506
diff
changeset
|
138 MPV_common_init_axp(s); |
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M«©ns Rullg«©rd
mellum
parents:
506
diff
changeset
|
139 #endif |
591 | 140 |
0 | 141 s->mb_width = (s->width + 15) / 16; |
142 s->mb_height = (s->height + 15) / 16; | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
143 |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
144 /* set default edge pos, will be overriden in decode_header if needed */ |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
145 s->h_edge_pos= s->mb_width*16; |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
146 s->v_edge_pos= s->mb_height*16; |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
147 |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
148 s->mb_num = s->mb_width * s->mb_height; |
553 | 149 if(!(s->flags&CODEC_FLAG_DR1)){ |
556 | 150 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
151 s->uvlinesize = s->mb_width * 8 + EDGE_WIDTH; | |
0 | 152 |
553 | 153 for(i=0;i<3;i++) { |
0 | 154 int w, h, shift, pict_start; |
155 | |
156 w = s->linesize; | |
157 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
158 shift = (i == 0) ? 0 : 1; | |
556 | 159 c_size = (s->linesize>>shift) * (h >> shift); |
160 pict_start = (s->linesize>>shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
0 | 161 |
456 | 162 CHECKED_ALLOCZ(pict, c_size) |
0 | 163 s->last_picture_base[i] = pict; |
164 s->last_picture[i] = pict + pict_start; | |
485 | 165 if(i>0) memset(s->last_picture_base[i], 128, c_size); |
0 | 166 |
456 | 167 CHECKED_ALLOCZ(pict, c_size) |
0 | 168 s->next_picture_base[i] = pict; |
169 s->next_picture[i] = pict + pict_start; | |
485 | 170 if(i>0) memset(s->next_picture_base[i], 128, c_size); |
324 | 171 |
365
fdeec2834c79
there are divx5? encoded files without a userdata section but with b-frames :(
michaelni
parents:
350
diff
changeset
|
172 if (s->has_b_frames || s->codec_id==CODEC_ID_MPEG4) { |
fdeec2834c79
there are divx5? encoded files without a userdata section but with b-frames :(
michaelni
parents:
350
diff
changeset
|
173 /* Note the MPEG4 stuff is here cuz of buggy encoders which dont set the low_delay flag but |
fdeec2834c79
there are divx5? encoded files without a userdata section but with b-frames :(
michaelni
parents:
350
diff
changeset
|
174 do low-delay encoding, so we cant allways distinguish b-frame containing streams from low_delay streams */ |
456 | 175 CHECKED_ALLOCZ(pict, c_size) |
0 | 176 s->aux_picture_base[i] = pict; |
177 s->aux_picture[i] = pict + pict_start; | |
485 | 178 if(i>0) memset(s->aux_picture_base[i], 128, c_size); |
0 | 179 } |
553 | 180 } |
556 | 181 s->ip_buffer_count= 2; |
0 | 182 } |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
183 |
567
5a6eaaa9fba2
larger edge emu buffer (might have been too small)
michaelni
parents:
566
diff
changeset
|
184 CHECKED_ALLOCZ(s->edge_emu_buffer, (s->width+64)*2*17*2); //(width + edge + align)*interlaced*MBsize*tolerance |
553 | 185 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
186 if (s->encoding) { |
324 | 187 int j; |
188 int mv_table_size= (s->mb_width+2)*(s->mb_height+2); | |
239 | 189 |
456 | 190 CHECKED_ALLOCZ(s->mb_var , s->mb_num * sizeof(INT16)) |
191 CHECKED_ALLOCZ(s->mc_mb_var, s->mb_num * sizeof(INT16)) | |
324 | 192 |
193 /* Allocate MV tables */ | |
456 | 194 CHECKED_ALLOCZ(s->p_mv_table , mv_table_size * 2 * sizeof(INT16)) |
195 CHECKED_ALLOCZ(s->b_forw_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
196 CHECKED_ALLOCZ(s->b_back_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
197 CHECKED_ALLOCZ(s->b_bidir_forw_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
198 CHECKED_ALLOCZ(s->b_bidir_back_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
199 CHECKED_ALLOCZ(s->b_direct_forw_mv_table, mv_table_size * 2 * sizeof(INT16)) | |
200 CHECKED_ALLOCZ(s->b_direct_back_mv_table, mv_table_size * 2 * sizeof(INT16)) | |
201 CHECKED_ALLOCZ(s->b_direct_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
324 | 202 |
456 | 203 CHECKED_ALLOCZ(s->me_scratchpad, s->linesize*16*3*sizeof(uint8_t)) |
204 | |
205 CHECKED_ALLOCZ(s->me_map , ME_MAP_SIZE*sizeof(uint32_t)) | |
206 CHECKED_ALLOCZ(s->me_score_map, ME_MAP_SIZE*sizeof(uint16_t)) | |
327 | 207 |
324 | 208 if(s->max_b_frames){ |
209 for(j=0; j<REORDER_BUFFER_SIZE; j++){ | |
210 int i; | |
211 for(i=0;i<3;i++) { | |
212 int w, h, shift; | |
213 | |
214 w = s->linesize; | |
215 h = s->mb_height * 16; | |
216 shift = (i == 0) ? 0 : 1; | |
217 c_size = (w >> shift) * (h >> shift); | |
218 | |
456 | 219 CHECKED_ALLOCZ(pict, c_size); |
324 | 220 s->picture_buffer[j][i] = pict; |
221 } | |
222 } | |
223 } | |
456 | 224 |
225 if(s->codec_id==CODEC_ID_MPEG4){ | |
226 CHECKED_ALLOCZ(s->tex_pb_buffer, PB_BUFFER_SIZE); | |
227 CHECKED_ALLOCZ( s->pb2_buffer, PB_BUFFER_SIZE); | |
228 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
229 } |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
230 |
288 | 231 if (s->out_format == FMT_H263 || s->encoding) { |
0 | 232 int size; |
456 | 233 /* Allocate MB type table */ |
234 CHECKED_ALLOCZ(s->mb_type , s->mb_num * sizeof(UINT8)) | |
235 | |
0 | 236 /* MV prediction */ |
237 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
456 | 238 CHECKED_ALLOCZ(s->motion_val, size * 2 * sizeof(INT16)); |
0 | 239 } |
240 | |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
241 if (s->h263_pred || s->h263_plus) { |
0 | 242 int y_size, c_size, i, size; |
243 | |
244 /* dc values */ | |
245 | |
246 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
247 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
248 size = y_size + 2 * c_size; | |
456 | 249 CHECKED_ALLOCZ(s->dc_val[0], size * sizeof(INT16)); |
0 | 250 s->dc_val[1] = s->dc_val[0] + y_size; |
251 s->dc_val[2] = s->dc_val[1] + c_size; | |
252 for(i=0;i<size;i++) | |
253 s->dc_val[0][i] = 1024; | |
254 | |
255 /* ac values */ | |
456 | 256 CHECKED_ALLOCZ(s->ac_val[0], size * sizeof(INT16) * 16); |
0 | 257 s->ac_val[1] = s->ac_val[0] + y_size; |
258 s->ac_val[2] = s->ac_val[1] + c_size; | |
259 | |
260 /* cbp values */ | |
456 | 261 CHECKED_ALLOCZ(s->coded_block, y_size); |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
262 |
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
263 /* which mb is a intra block */ |
456 | 264 CHECKED_ALLOCZ(s->mbintra_table, s->mb_num); |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
265 memset(s->mbintra_table, 1, s->mb_num); |
333 | 266 |
267 /* divx501 bitstream reorder buffer */ | |
456 | 268 CHECKED_ALLOCZ(s->bitstream_buffer, BITSTREAM_BUFFER_SIZE); |
269 | |
270 /* cbp, ac_pred, pred_dir */ | |
271 CHECKED_ALLOCZ(s->cbp_table , s->mb_num * sizeof(UINT8)) | |
272 CHECKED_ALLOCZ(s->pred_dir_table, s->mb_num * sizeof(UINT8)) | |
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
|
273 } |
575 | 274 CHECKED_ALLOCZ(s->qscale_table , s->mb_num * sizeof(UINT8)) |
275 | |
0 | 276 /* default structure is frame */ |
277 s->picture_structure = PICT_FRAME; | |
553 | 278 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
279 /* init macroblock skip table */ |
456 | 280 CHECKED_ALLOCZ(s->mbskip_table, s->mb_num); |
294 | 281 |
327 | 282 s->block= s->blocks[0]; |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
283 |
0 | 284 s->context_initialized = 1; |
285 return 0; | |
286 fail: | |
244 | 287 MPV_common_end(s); |
288 return -1; | |
289 } | |
290 | |
456 | 291 |
292 //extern int sads; | |
293 | |
244 | 294 /* init common structure for both encoder and decoder */ |
295 void MPV_common_end(MpegEncContext *s) | |
296 { | |
297 int i; | |
298 | |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
299 av_freep(&s->mb_type); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
300 av_freep(&s->mb_var); |
456 | 301 av_freep(&s->mc_mb_var); |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
302 av_freep(&s->p_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
303 av_freep(&s->b_forw_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
304 av_freep(&s->b_back_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
305 av_freep(&s->b_bidir_forw_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
306 av_freep(&s->b_bidir_back_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
307 av_freep(&s->b_direct_forw_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
308 av_freep(&s->b_direct_back_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
309 av_freep(&s->b_direct_mv_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
310 av_freep(&s->motion_val); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
311 av_freep(&s->dc_val[0]); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
312 av_freep(&s->ac_val[0]); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
313 av_freep(&s->coded_block); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
314 av_freep(&s->mbintra_table); |
456 | 315 av_freep(&s->cbp_table); |
316 av_freep(&s->pred_dir_table); | |
317 av_freep(&s->qscale_table); | |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
318 av_freep(&s->me_scratchpad); |
456 | 319 av_freep(&s->me_map); |
320 av_freep(&s->me_score_map); | |
321 | |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
322 av_freep(&s->mbskip_table); |
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
323 av_freep(&s->bitstream_buffer); |
456 | 324 av_freep(&s->tex_pb_buffer); |
325 av_freep(&s->pb2_buffer); | |
553 | 326 av_freep(&s->edge_emu_buffer); |
327 | |
0 | 328 for(i=0;i<3;i++) { |
324 | 329 int j; |
553 | 330 if(!(s->flags&CODEC_FLAG_DR1)){ |
331 av_freep(&s->last_picture_base[i]); | |
332 av_freep(&s->next_picture_base[i]); | |
333 av_freep(&s->aux_picture_base[i]); | |
334 } | |
335 s->last_picture_base[i]= | |
336 s->next_picture_base[i]= | |
337 s->aux_picture_base [i] = NULL; | |
338 s->last_picture[i]= | |
339 s->next_picture[i]= | |
340 s->aux_picture [i] = NULL; | |
341 | |
324 | 342 for(j=0; j<REORDER_BUFFER_SIZE; j++){ |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
343 av_freep(&s->picture_buffer[j][i]); |
324 | 344 } |
0 | 345 } |
346 s->context_initialized = 0; | |
347 } | |
348 | |
349 /* init video encoder */ | |
350 int MPV_encode_init(AVCodecContext *avctx) | |
351 { | |
352 MpegEncContext *s = avctx->priv_data; | |
60 | 353 int i; |
0 | 354 |
315 | 355 avctx->pix_fmt = PIX_FMT_YUV420P; |
356 | |
0 | 357 s->bit_rate = avctx->bit_rate; |
268 | 358 s->bit_rate_tolerance = avctx->bit_rate_tolerance; |
0 | 359 s->frame_rate = avctx->frame_rate; |
360 s->width = avctx->width; | |
361 s->height = avctx->height; | |
456 | 362 if(avctx->gop_size > 600){ |
462
3fdd0627f6bd
typo (found by Bohdan Horst <nexus at hoth.amu.edu.pl>)
michaelni
parents:
461
diff
changeset
|
363 fprintf(stderr, "Warning keyframe interval too large! reducing it ...\n"); |
456 | 364 avctx->gop_size=600; |
365 } | |
0 | 366 s->gop_size = avctx->gop_size; |
162 | 367 s->rtp_mode = avctx->rtp_mode; |
368 s->rtp_payload_size = avctx->rtp_payload_size; | |
231 | 369 if (avctx->rtp_callback) |
370 s->rtp_callback = avctx->rtp_callback; | |
268 | 371 s->qmin= avctx->qmin; |
372 s->qmax= avctx->qmax; | |
373 s->max_qdiff= avctx->max_qdiff; | |
374 s->qcompress= avctx->qcompress; | |
375 s->qblur= avctx->qblur; | |
329 | 376 s->b_quant_factor= avctx->b_quant_factor; |
456 | 377 s->b_quant_offset= avctx->b_quant_offset; |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
378 s->avctx = avctx; |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
379 s->aspect_ratio_info= avctx->aspect_ratio_info; |
294 | 380 s->flags= avctx->flags; |
324 | 381 s->max_b_frames= avctx->max_b_frames; |
329 | 382 s->rc_strategy= avctx->rc_strategy; |
383 s->b_frame_strategy= avctx->b_frame_strategy; | |
344 | 384 s->codec_id= avctx->codec->id; |
456 | 385 s->luma_elim_threshold = avctx->luma_elim_threshold; |
386 s->chroma_elim_threshold= avctx->chroma_elim_threshold; | |
387 s->strict_std_compliance= avctx->strict_std_compliance; | |
388 s->data_partitioning= avctx->flags & CODEC_FLAG_PART; | |
344 | 389 |
0 | 390 if (s->gop_size <= 1) { |
391 s->intra_only = 1; | |
392 s->gop_size = 12; | |
393 } else { | |
394 s->intra_only = 0; | |
395 } | |
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
396 |
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
397 /* ME algorithm */ |
321 | 398 if (avctx->me_method == 0) |
399 /* For compatibility */ | |
400 s->me_method = motion_estimation_method; | |
401 else | |
402 s->me_method = avctx->me_method; | |
403 | |
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
404 /* Fixed QSCALE */ |
0 | 405 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE); |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
406 |
0 | 407 switch(avctx->codec->id) { |
408 case CODEC_ID_MPEG1VIDEO: | |
409 s->out_format = FMT_MPEG1; | |
336 | 410 avctx->delay=0; //FIXME not sure, should check the spec |
0 | 411 break; |
412 case CODEC_ID_MJPEG: | |
413 s->out_format = FMT_MJPEG; | |
414 s->intra_only = 1; /* force intra only for jpeg */ | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
415 s->mjpeg_write_tables = 1; /* write all tables */ |
370
0eca28d16cbd
clamp intra matrix to 8bit for mjpeg (workaround for qscale>=25)
al3x
parents:
365
diff
changeset
|
416 s->mjpeg_data_only_frames = 0; /* write all the needed headers */ |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
417 s->mjpeg_vsample[0] = 2; /* set up default sampling factors */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
418 s->mjpeg_vsample[1] = 1; /* the only currently supported values */ |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
419 s->mjpeg_vsample[2] = 1; |
370
0eca28d16cbd
clamp intra matrix to 8bit for mjpeg (workaround for qscale>=25)
al3x
parents:
365
diff
changeset
|
420 s->mjpeg_hsample[0] = 2; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
421 s->mjpeg_hsample[1] = 1; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
422 s->mjpeg_hsample[2] = 1; |
0 | 423 if (mjpeg_init(s) < 0) |
424 return -1; | |
336 | 425 avctx->delay=0; |
0 | 426 break; |
427 case CODEC_ID_H263: | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
428 if (h263_get_picture_format(s->width, s->height) == 7) { |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
429 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
0 | 430 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
431 } |
0 | 432 s->out_format = FMT_H263; |
336 | 433 avctx->delay=0; |
0 | 434 break; |
435 case CODEC_ID_H263P: | |
436 s->out_format = FMT_H263; | |
162 | 437 s->rtp_mode = 1; |
438 s->rtp_payload_size = 1200; | |
0 | 439 s->h263_plus = 1; |
78 | 440 s->unrestricted_mv = 1; |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
441 s->h263_aic = 1; |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
442 |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
443 /* These are just to be sure */ |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
444 s->umvplus = 0; |
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
445 s->umvplus_dec = 0; |
336 | 446 avctx->delay=0; |
0 | 447 break; |
448 case CODEC_ID_RV10: | |
449 s->out_format = FMT_H263; | |
450 s->h263_rv10 = 1; | |
336 | 451 avctx->delay=0; |
0 | 452 break; |
71 | 453 case CODEC_ID_MPEG4: |
0 | 454 s->out_format = FMT_H263; |
455 s->h263_pred = 1; | |
456 s->unrestricted_mv = 1; | |
324 | 457 s->has_b_frames= s->max_b_frames ? 1 : 0; |
336 | 458 s->low_delay=0; |
459 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); | |
0 | 460 break; |
307 | 461 case CODEC_ID_MSMPEG4V1: |
0 | 462 s->out_format = FMT_H263; |
463 s->h263_msmpeg4 = 1; | |
464 s->h263_pred = 1; | |
465 s->unrestricted_mv = 1; | |
307 | 466 s->msmpeg4_version= 1; |
336 | 467 avctx->delay=0; |
307 | 468 break; |
469 case CODEC_ID_MSMPEG4V2: | |
470 s->out_format = FMT_H263; | |
471 s->h263_msmpeg4 = 1; | |
472 s->h263_pred = 1; | |
473 s->unrestricted_mv = 1; | |
474 s->msmpeg4_version= 2; | |
336 | 475 avctx->delay=0; |
307 | 476 break; |
477 case CODEC_ID_MSMPEG4V3: | |
478 s->out_format = FMT_H263; | |
479 s->h263_msmpeg4 = 1; | |
480 s->h263_pred = 1; | |
481 s->unrestricted_mv = 1; | |
482 s->msmpeg4_version= 3; | |
336 | 483 avctx->delay=0; |
0 | 484 break; |
499 | 485 case CODEC_ID_WMV1: |
486 s->out_format = FMT_H263; | |
487 s->h263_msmpeg4 = 1; | |
488 s->h263_pred = 1; | |
489 s->unrestricted_mv = 1; | |
490 s->msmpeg4_version= 4; | |
491 avctx->delay=0; | |
492 break; | |
493 case CODEC_ID_WMV2: | |
494 s->out_format = FMT_H263; | |
495 s->h263_msmpeg4 = 1; | |
496 s->h263_pred = 1; | |
497 s->unrestricted_mv = 1; | |
498 s->msmpeg4_version= 5; | |
499 avctx->delay=0; | |
500 break; | |
0 | 501 default: |
502 return -1; | |
503 } | |
295 | 504 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
505 { /* set up some save defaults, some codecs might override them later */ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
506 static int done=0; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
507 if(!done){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
508 int i; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
509 done=1; |
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
510 memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1)); |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
511 memset(default_fcode_tab , 0, sizeof(UINT8)*(2*MAX_MV+1)); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
512 |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
513 for(i=-16; i<16; i++){ |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
514 default_fcode_tab[i + MAX_MV]= 1; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
515 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
516 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
517 } |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
518 s->mv_penalty= default_mv_penalty; |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
519 s->fcode_tab= default_fcode_tab; |
506
b9ed0ae72d51
init dc_scale tables to defaults (fixes mjpeg sig11)
michaelni
parents:
499
diff
changeset
|
520 s->y_dc_scale_table= |
b9ed0ae72d51
init dc_scale tables to defaults (fixes mjpeg sig11)
michaelni
parents:
499
diff
changeset
|
521 s->c_dc_scale_table= ff_mpeg1_dc_scale_table; |
b9ed0ae72d51
init dc_scale tables to defaults (fixes mjpeg sig11)
michaelni
parents:
499
diff
changeset
|
522 |
0 | 523 if (s->out_format == FMT_H263) |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
524 h263_encode_init(s); |
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
525 else if (s->out_format == FMT_MPEG1) |
499 | 526 ff_mpeg1_encode_init(s); |
527 if(s->msmpeg4_version) | |
528 ff_msmpeg4_encode_init(s); | |
0 | 529 |
287 | 530 /* dont use mv_penalty table for crap MV as it would be confused */ |
324 | 531 if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty; |
287 | 532 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
533 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
534 |
0 | 535 /* init */ |
536 if (MPV_common_init(s) < 0) | |
537 return -1; | |
538 | |
60 | 539 /* init default q matrix */ |
540 for(i=0;i<64;i++) { | |
344 | 541 if(s->out_format == FMT_H263) |
533
3c07cf9595de
adding ff prefix to avoid global name conficts with xvid (patch by Marko Kreen <marko at l-t.ee>)
michaelni
parents:
514
diff
changeset
|
542 s->intra_matrix[i] = ff_mpeg1_default_non_intra_matrix[i]; |
344 | 543 else |
533
3c07cf9595de
adding ff prefix to avoid global name conficts with xvid (patch by Marko Kreen <marko at l-t.ee>)
michaelni
parents:
514
diff
changeset
|
544 s->intra_matrix[i] = ff_mpeg1_default_intra_matrix[i]; |
344 | 545 |
533
3c07cf9595de
adding ff prefix to avoid global name conficts with xvid (patch by Marko Kreen <marko at l-t.ee>)
michaelni
parents:
514
diff
changeset
|
546 s->inter_matrix[i] = ff_mpeg1_default_non_intra_matrix[i]; |
344 | 547 } |
548 | |
549 /* precompute matrix */ | |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
550 /* for mjpeg, we do include qscale in the matrix */ |
344 | 551 if (s->out_format != FMT_MJPEG) { |
552 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias, | |
553 s->intra_matrix, s->intra_quant_bias); | |
554 convert_matrix(s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias, | |
555 s->inter_matrix, s->inter_quant_bias); | |
60 | 556 } |
557 | |
329 | 558 if(ff_rate_control_init(s) < 0) |
559 return -1; | |
0 | 560 |
561 s->picture_number = 0; | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
562 s->picture_in_gop_number = 0; |
0 | 563 s->fake_picture_number = 0; |
564 /* motion detector init */ | |
565 s->f_code = 1; | |
324 | 566 s->b_code = 1; |
0 | 567 |
568 return 0; | |
569 } | |
570 | |
571 int MPV_encode_end(AVCodecContext *avctx) | |
572 { | |
573 MpegEncContext *s = avctx->priv_data; | |
574 | |
575 #ifdef STATS | |
576 print_stats(); | |
577 #endif | |
329 | 578 |
579 ff_rate_control_uninit(s); | |
580 | |
0 | 581 MPV_common_end(s); |
582 if (s->out_format == FMT_MJPEG) | |
583 mjpeg_close(s); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
584 |
0 | 585 return 0; |
586 } | |
587 | |
588 /* draw the edges of width 'w' of an image of size width, height */ | |
206 | 589 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
0 | 590 { |
591 UINT8 *ptr, *last_line; | |
592 int i; | |
593 | |
594 last_line = buf + (height - 1) * wrap; | |
595 for(i=0;i<w;i++) { | |
596 /* top and bottom */ | |
597 memcpy(buf - (i + 1) * wrap, buf, width); | |
598 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
599 } | |
600 /* left and right */ | |
601 ptr = buf; | |
602 for(i=0;i<height;i++) { | |
603 memset(ptr - w, ptr[0], w); | |
604 memset(ptr + width, ptr[width-1], w); | |
605 ptr += wrap; | |
606 } | |
607 /* corners */ | |
608 for(i=0;i<w;i++) { | |
609 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
610 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
611 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
612 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
613 } | |
614 } | |
615 | |
616 /* generic function for encode/decode called before a frame is coded/decoded */ | |
553 | 617 void MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx) |
0 | 618 { |
619 int i; | |
620 UINT8 *tmp; | |
621 | |
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
622 s->mb_skiped = 0; |
456 | 623 s->decoding_error=0; |
562
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
624 avctx->mbskip_table= s->mbskip_table; |
456 | 625 |
553 | 626 if(avctx->flags&CODEC_FLAG_DR1){ |
627 avctx->get_buffer_callback(avctx, s->width, s->height, s->pict_type); | |
628 | |
556 | 629 s->linesize = avctx->dr_stride; |
630 s->uvlinesize= avctx->dr_uvstride; | |
631 s->ip_buffer_count= avctx->dr_ip_buffer_count; | |
553 | 632 } |
562
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
633 avctx->dr_ip_buffer_count= s->ip_buffer_count; |
553 | 634 |
0 | 635 if (s->pict_type == B_TYPE) { |
636 for(i=0;i<3;i++) { | |
553 | 637 if(avctx->flags&CODEC_FLAG_DR1) |
638 s->aux_picture[i]= avctx->dr_buffer[i]; | |
569 | 639 |
640 //FIXME the following should never be needed, the decoder should drop b frames if no reference is available | |
641 if(s->next_picture[i]==NULL) | |
642 s->next_picture[i]= s->aux_picture[i]; | |
643 if(s->last_picture[i]==NULL) | |
644 s->last_picture[i]= s->next_picture[i]; | |
553 | 645 |
0 | 646 s->current_picture[i] = s->aux_picture[i]; |
647 } | |
648 } else { | |
649 for(i=0;i<3;i++) { | |
650 /* swap next and last */ | |
553 | 651 if(avctx->flags&CODEC_FLAG_DR1) |
652 tmp= avctx->dr_buffer[i]; | |
653 else | |
654 tmp = s->last_picture[i]; | |
655 | |
0 | 656 s->last_picture[i] = s->next_picture[i]; |
657 s->next_picture[i] = tmp; | |
658 s->current_picture[i] = tmp; | |
553 | 659 |
660 s->last_dr_opaque= s->next_dr_opaque; | |
661 s->next_dr_opaque= avctx->dr_opaque_frame; | |
662 | |
557 | 663 if(s->has_b_frames && s->last_dr_opaque && s->codec_id!=CODEC_ID_SVQ1) |
553 | 664 avctx->dr_opaque_frame= s->last_dr_opaque; |
665 else | |
666 avctx->dr_opaque_frame= s->next_dr_opaque; | |
0 | 667 } |
668 } | |
591 | 669 |
670 /* set dequantizer, we cant do it during init as it might change for mpeg4 | |
671 and we cant do it in the header decode as init isnt called for mpeg4 there yet */ | |
672 if(s->out_format == FMT_H263){ | |
673 if(s->mpeg_quant) | |
674 s->dct_unquantize = s->dct_unquantize_mpeg2; | |
675 else | |
676 s->dct_unquantize = s->dct_unquantize_h263; | |
677 }else | |
678 s->dct_unquantize = s->dct_unquantize_mpeg1; | |
0 | 679 } |
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
|
680 |
0 | 681 /* generic function for encode/decode called after a frame has been coded/decoded */ |
682 void MPV_frame_end(MpegEncContext *s) | |
683 { | |
456 | 684 // if((s->picture_number%100)==0 && s->encoding) printf("sads:%d //\n", sads); |
685 | |
0 | 686 /* draw edge for correct motion prediction if outside */ |
553 | 687 if (s->pict_type != B_TYPE && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) { |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
688 draw_edges(s->current_picture[0], s->linesize , s->h_edge_pos , s->v_edge_pos , EDGE_WIDTH ); |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
689 draw_edges(s->current_picture[1], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2); |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
690 draw_edges(s->current_picture[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2); |
0 | 691 } |
207 | 692 emms_c(); |
329 | 693 |
694 if(s->pict_type!=B_TYPE){ | |
695 s->last_non_b_pict_type= s->pict_type; | |
696 s->last_non_b_qscale= s->qscale; | |
456 | 697 s->last_non_b_mc_mb_var= s->mc_mb_var_sum; |
341 | 698 s->num_available_buffers++; |
699 if(s->num_available_buffers>2) s->num_available_buffers= 2; | |
329 | 700 } |
0 | 701 } |
702 | |
324 | 703 /* reorder input for encoding */ |
704 void reorder_input(MpegEncContext *s, AVPicture *pict) | |
705 { | |
706 int i, j, index; | |
707 | |
708 if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES; | |
709 | |
710 // delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff) | |
711 | |
712 for(j=0; j<REORDER_BUFFER_SIZE-1; j++){ | |
713 s->coded_order[j]= s->coded_order[j+1]; | |
714 } | |
715 s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers | |
339 | 716 s->coded_order[j].pict_type=0; |
324 | 717 |
718 switch(s->input_pict_type){ | |
719 default: | |
720 case I_TYPE: | |
721 case S_TYPE: | |
722 case P_TYPE: | |
723 index= s->max_b_frames - s->b_frames_since_non_b; | |
724 s->b_frames_since_non_b=0; | |
725 break; | |
726 case B_TYPE: | |
727 index= s->max_b_frames + 1; | |
728 s->b_frames_since_non_b++; | |
729 break; | |
730 } | |
731 //printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize); | |
732 if( (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED)) | |
733 && pict->linesize[0] == s->linesize | |
556 | 734 && pict->linesize[1] == s->uvlinesize |
735 && pict->linesize[2] == s->uvlinesize){ | |
324 | 736 //printf("ptr\n"); |
737 for(i=0; i<3; i++){ | |
738 s->coded_order[index].picture[i]= pict->data[i]; | |
739 } | |
740 }else{ | |
741 //printf("copy\n"); | |
742 for(i=0; i<3; i++){ | |
743 uint8_t *src = pict->data[i]; | |
744 uint8_t *dest; | |
745 int src_wrap = pict->linesize[i]; | |
746 int dest_wrap = s->linesize; | |
747 int w = s->width; | |
748 int h = s->height; | |
749 | |
750 if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering | |
751 else dest= s->picture_buffer[s->picture_buffer_index][i]; | |
752 | |
753 if (i >= 1) { | |
754 dest_wrap >>= 1; | |
755 w >>= 1; | |
756 h >>= 1; | |
757 } | |
758 | |
759 s->coded_order[index].picture[i]= dest; | |
760 for(j=0;j<h;j++) { | |
761 memcpy(dest, src, w); | |
762 dest += dest_wrap; | |
763 src += src_wrap; | |
764 } | |
765 } | |
766 if(index!=0){ | |
767 s->picture_buffer_index++; | |
588 | 768 if(s->picture_buffer_index >= REORDER_BUFFER_SIZE) s->picture_buffer_index=0; |
324 | 769 } |
770 } | |
771 s->coded_order[index].pict_type = s->input_pict_type; | |
772 s->coded_order[index].qscale = s->input_qscale; | |
773 s->coded_order[index].force_type= s->force_input_type; | |
774 s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number; | |
775 s->coded_order[index].picture_number= s->input_picture_number; | |
776 | |
777 for(i=0; i<3; i++){ | |
778 s->new_picture[i]= s->coded_order[0].picture[i]; | |
779 } | |
780 } | |
781 | |
0 | 782 int MPV_encode_picture(AVCodecContext *avctx, |
783 unsigned char *buf, int buf_size, void *data) | |
784 { | |
785 MpegEncContext *s = avctx->priv_data; | |
786 AVPicture *pict = data; | |
787 | |
324 | 788 s->input_qscale = avctx->quality; |
0 | 789 |
790 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
791 | |
329 | 792 if(avctx->flags&CODEC_FLAG_TYPE){ |
793 s->input_pict_type= | |
794 s->force_input_type= avctx->key_frame ? I_TYPE : P_TYPE; | |
795 }else if(s->flags&CODEC_FLAG_PASS2){ | |
796 s->input_pict_type= | |
797 s->force_input_type= s->rc_context.entry[s->input_picture_number].new_pict_type; | |
798 }else{ | |
799 s->force_input_type=0; | |
800 if (!s->intra_only) { | |
801 /* first picture of GOP is intra */ | |
802 if (s->input_picture_in_gop_number % s->gop_size==0){ | |
803 s->input_pict_type = I_TYPE; | |
804 }else if(s->max_b_frames==0){ | |
805 s->input_pict_type = P_TYPE; | |
806 }else{ | |
807 if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ | |
808 s->input_pict_type = B_TYPE; | |
809 else | |
810 s->input_pict_type = P_TYPE; | |
811 } | |
812 } else { | |
324 | 813 s->input_pict_type = I_TYPE; |
814 } | |
0 | 815 } |
816 | |
329 | 817 if(s->input_pict_type==I_TYPE) |
818 s->input_picture_in_gop_number=0; | |
819 | |
324 | 820 reorder_input(s, pict); |
821 | |
822 /* output? */ | |
823 if(s->coded_order[0].picture[0]){ | |
824 | |
825 s->pict_type= s->coded_order[0].pict_type; | |
826 if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */ | |
827 s->qscale= s->coded_order[0].qscale; | |
828 s->force_type= s->coded_order[0].force_type; | |
829 s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number; | |
830 s->picture_number= s->coded_order[0].picture_number; | |
831 | |
553 | 832 MPV_frame_start(s, avctx); |
286 | 833 |
324 | 834 encode_picture(s, s->picture_number); |
376 | 835 avctx->key_frame = (s->pict_type == I_TYPE); |
836 avctx->pict_type = s->pict_type; | |
837 avctx->real_pict_num = s->picture_number; | |
324 | 838 avctx->header_bits = s->header_bits; |
839 avctx->mv_bits = s->mv_bits; | |
840 avctx->misc_bits = s->misc_bits; | |
841 avctx->i_tex_bits = s->i_tex_bits; | |
842 avctx->p_tex_bits = s->p_tex_bits; | |
843 avctx->i_count = s->i_count; | |
844 avctx->p_count = s->p_count; | |
845 avctx->skip_count = s->skip_count; | |
0 | 846 |
324 | 847 MPV_frame_end(s); |
848 | |
849 if (s->out_format == FMT_MJPEG) | |
850 mjpeg_picture_trailer(s); | |
851 | |
852 avctx->quality = s->qscale; | |
329 | 853 |
854 if(s->flags&CODEC_FLAG_PASS1) | |
855 ff_write_pass1_stats(s); | |
456 | 856 |
324 | 857 } |
858 | |
859 s->input_picture_number++; | |
860 s->input_picture_in_gop_number++; | |
0 | 861 |
862 flush_put_bits(&s->pb); | |
268 | 863 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8; |
329 | 864 if(s->pict_type==B_TYPE) s->pb_frame_bits+= s->frame_bits; |
865 else s->pb_frame_bits= s->frame_bits; | |
866 | |
268 | 867 s->total_bits += s->frame_bits; |
286 | 868 avctx->frame_bits = s->frame_bits; |
869 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n", | |
870 //s->f_code, avctx->key_frame, s->header_bits, s->mv_bits, s->misc_bits, s->frame_bits, s->i_tex_bits, s->p_tex_bits); | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
871 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
872 if (avctx->get_psnr) { |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
873 /* At this point pict->data should have the original frame */ |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
874 /* an s->current_picture should have the coded/decoded frame */ |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
875 get_psnr(pict->data, s->current_picture, |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
876 pict->linesize, s->linesize, avctx); |
324 | 877 // printf("%f\n", avctx->psnr_y); |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
878 } |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
879 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 880 } |
881 | |
255 | 882 static inline void gmc1_motion(MpegEncContext *s, |
883 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
884 int dest_offset, | |
885 UINT8 **ref_picture, int src_offset, | |
886 int h) | |
887 { | |
888 UINT8 *ptr; | |
556 | 889 int offset, src_x, src_y, linesize, uvlinesize; |
255 | 890 int motion_x, motion_y; |
566 | 891 int emu=0; |
255 | 892 |
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
893 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n"); |
255 | 894 motion_x= s->sprite_offset[0][0]; |
895 motion_y= s->sprite_offset[0][1]; | |
896 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
897 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
898 motion_x<<=(3-s->sprite_warping_accuracy); | |
899 motion_y<<=(3-s->sprite_warping_accuracy); | |
900 src_x = clip(src_x, -16, s->width); | |
901 if (src_x == s->width) | |
902 motion_x =0; | |
903 src_y = clip(src_y, -16, s->height); | |
904 if (src_y == s->height) | |
905 motion_y =0; | |
906 | |
907 linesize = s->linesize; | |
556 | 908 uvlinesize = s->uvlinesize; |
255 | 909 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; |
910 | |
911 dest_y+=dest_offset; | |
566 | 912 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
913 if(src_x<0 || src_y<0 || src_x + (motion_x&15) + 16 > s->h_edge_pos |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
914 || src_y + (motion_y&15) + h > s->v_edge_pos){ |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
915 emulated_edge_mc(s, ptr, linesize, 17, h+1, src_x, src_y, s->h_edge_pos, s->v_edge_pos); |
566 | 916 ptr= s->edge_emu_buffer; |
917 emu=1; | |
918 } | |
919 } | |
255 | 920 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); |
921 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
922 | |
923 motion_x= s->sprite_offset[1][0]; | |
924 motion_y= s->sprite_offset[1][1]; | |
925 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
926 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
927 motion_x<<=(3-s->sprite_warping_accuracy); | |
928 motion_y<<=(3-s->sprite_warping_accuracy); | |
929 src_x = clip(src_x, -8, s->width>>1); | |
930 if (src_x == s->width>>1) | |
931 motion_x =0; | |
932 src_y = clip(src_y, -8, s->height>>1); | |
933 if (src_y == s->height>>1) | |
934 motion_y =0; | |
935 | |
556 | 936 offset = (src_y * uvlinesize) + src_x + (src_offset>>1); |
255 | 937 ptr = ref_picture[1] + offset; |
566 | 938 if(emu){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
939 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
566 | 940 ptr= s->edge_emu_buffer; |
941 } | |
556 | 942 gmc1(dest_cb + (dest_offset>>1), ptr, uvlinesize, h>>1, motion_x&15, motion_y&15, s->no_rounding); |
566 | 943 |
255 | 944 ptr = ref_picture[2] + offset; |
566 | 945 if(emu){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
946 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
566 | 947 ptr= s->edge_emu_buffer; |
948 } | |
556 | 949 gmc1(dest_cr + (dest_offset>>1), ptr, uvlinesize, h>>1, motion_x&15, motion_y&15, s->no_rounding); |
255 | 950 |
951 return; | |
952 } | |
953 | |
566 | 954 static void emulated_edge_mc(MpegEncContext *s, UINT8 *src, int linesize, int block_w, int block_h, |
553 | 955 int src_x, int src_y, int w, int h){ |
956 int x, y; | |
957 int start_y, start_x, end_y, end_x; | |
566 | 958 UINT8 *buf= s->edge_emu_buffer; |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
959 |
553 | 960 if(src_y>= h){ |
961 src+= (h-1-src_y)*linesize; | |
962 src_y=h-1; | |
554 | 963 }else if(src_y<=-block_h){ |
964 src+= (1-block_h-src_y)*linesize; | |
965 src_y=1-block_h; | |
553 | 966 } |
967 if(src_x>= w){ | |
968 src+= (w-1-src_x); | |
969 src_x=w-1; | |
554 | 970 }else if(src_x<=-block_w){ |
971 src+= (1-block_w-src_x); | |
972 src_x=1-block_w; | |
553 | 973 } |
974 | |
975 start_y= MAX(0, -src_y); | |
976 start_x= MAX(0, -src_x); | |
977 end_y= MIN(block_h, h-src_y); | |
978 end_x= MIN(block_w, w-src_x); | |
566 | 979 |
553 | 980 // copy existing part |
981 for(y=start_y; y<end_y; y++){ | |
982 for(x=start_x; x<end_x; x++){ | |
983 buf[x + y*linesize]= src[x + y*linesize]; | |
984 } | |
985 } | |
986 | |
987 //top | |
988 for(y=0; y<start_y; y++){ | |
989 for(x=start_x; x<end_x; x++){ | |
990 buf[x + y*linesize]= buf[x + start_y*linesize]; | |
991 } | |
992 } | |
993 | |
994 //bottom | |
995 for(y=end_y; y<block_h; y++){ | |
996 for(x=start_x; x<end_x; x++){ | |
997 buf[x + y*linesize]= buf[x + (end_y-1)*linesize]; | |
998 } | |
999 } | |
1000 | |
1001 for(y=0; y<block_h; y++){ | |
1002 //left | |
1003 for(x=0; x<start_x; x++){ | |
1004 buf[x + y*linesize]= buf[start_x + y*linesize]; | |
1005 } | |
1006 | |
1007 //right | |
1008 for(x=end_x; x<block_w; x++){ | |
1009 buf[x + y*linesize]= buf[end_x - 1 + y*linesize]; | |
1010 } | |
1011 } | |
1012 } | |
1013 | |
1014 | |
0 | 1015 /* apply one mpeg motion vector to the three components */ |
1016 static inline void mpeg_motion(MpegEncContext *s, | |
1017 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
1018 int dest_offset, | |
1019 UINT8 **ref_picture, int src_offset, | |
1020 int field_based, op_pixels_func *pix_op, | |
1021 int motion_x, int motion_y, int h) | |
1022 { | |
1023 UINT8 *ptr; | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1024 int dxy, offset, mx, my, src_x, src_y, height, v_edge_pos, linesize, uvlinesize; |
553 | 1025 int emu=0; |
1026 | |
255 | 1027 if(s->quarter_sample) |
1028 { | |
1029 motion_x>>=1; | |
1030 motion_y>>=1; | |
1031 } | |
0 | 1032 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
1033 src_x = s->mb_x * 16 + (motion_x >> 1); | |
1034 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
1035 | |
1036 /* WARNING: do no forget half pels */ | |
1037 height = s->height >> field_based; | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1038 v_edge_pos = s->v_edge_pos >> field_based; |
0 | 1039 src_x = clip(src_x, -16, s->width); |
1040 if (src_x == s->width) | |
1041 dxy &= ~1; | |
1042 src_y = clip(src_y, -16, height); | |
1043 if (src_y == height) | |
1044 dxy &= ~2; | |
565
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1045 linesize = s->linesize << field_based; |
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1046 uvlinesize = s->uvlinesize << field_based; |
0 | 1047 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; |
1048 dest_y += dest_offset; | |
553 | 1049 |
1050 if(s->flags&CODEC_FLAG_EMU_EDGE){ | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1051 if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 16 > s->h_edge_pos |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1052 || src_y + (motion_y&1) + h > v_edge_pos){ |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1053 emulated_edge_mc(s, ptr, linesize, 17, h+1, src_x, src_y, s->h_edge_pos, v_edge_pos); |
553 | 1054 ptr= s->edge_emu_buffer; |
1055 emu=1; | |
1056 } | |
1057 } | |
0 | 1058 pix_op[dxy](dest_y, ptr, linesize, h); |
1059 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h); | |
1060 | |
485 | 1061 if(s->flags&CODEC_FLAG_GRAY) return; |
1062 | |
0 | 1063 if (s->out_format == FMT_H263) { |
1064 dxy = 0; | |
1065 if ((motion_x & 3) != 0) | |
1066 dxy |= 1; | |
1067 if ((motion_y & 3) != 0) | |
1068 dxy |= 2; | |
1069 mx = motion_x >> 2; | |
1070 my = motion_y >> 2; | |
1071 } else { | |
1072 mx = motion_x / 2; | |
1073 my = motion_y / 2; | |
1074 dxy = ((my & 1) << 1) | (mx & 1); | |
1075 mx >>= 1; | |
1076 my >>= 1; | |
1077 } | |
1078 | |
1079 src_x = s->mb_x * 8 + mx; | |
1080 src_y = s->mb_y * (8 >> field_based) + my; | |
1081 src_x = clip(src_x, -8, s->width >> 1); | |
1082 if (src_x == (s->width >> 1)) | |
1083 dxy &= ~1; | |
1084 src_y = clip(src_y, -8, height >> 1); | |
1085 if (src_y == (height >> 1)) | |
1086 dxy &= ~2; | |
565
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1087 offset = (src_y * uvlinesize) + src_x + (src_offset >> 1); |
0 | 1088 ptr = ref_picture[1] + offset; |
553 | 1089 if(emu){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1090 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
553 | 1091 ptr= s->edge_emu_buffer; |
1092 } | |
565
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1093 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, uvlinesize, h >> 1); |
553 | 1094 |
0 | 1095 ptr = ref_picture[2] + offset; |
553 | 1096 if(emu){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1097 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
553 | 1098 ptr= s->edge_emu_buffer; |
1099 } | |
565
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1100 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, uvlinesize, h >> 1); |
0 | 1101 } |
1102 | |
255 | 1103 static inline void qpel_motion(MpegEncContext *s, |
1104 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
1105 int dest_offset, | |
1106 UINT8 **ref_picture, int src_offset, | |
1107 int field_based, op_pixels_func *pix_op, | |
1108 qpel_mc_func *qpix_op, | |
1109 int motion_x, int motion_y, int h) | |
1110 { | |
1111 UINT8 *ptr; | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1112 int dxy, offset, mx, my, src_x, src_y, height, v_edge_pos, linesize; |
554 | 1113 int emu=0; |
255 | 1114 |
1115 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
1116 src_x = s->mb_x * 16 + (motion_x >> 2); | |
1117 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
1118 | |
1119 height = s->height >> field_based; | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1120 v_edge_pos = s->v_edge_pos >> field_based; |
255 | 1121 src_x = clip(src_x, -16, s->width); |
1122 if (src_x == s->width) | |
1123 dxy &= ~3; | |
1124 src_y = clip(src_y, -16, height); | |
1125 if (src_y == height) | |
1126 dxy &= ~12; | |
1127 linesize = s->linesize << field_based; | |
1128 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; | |
1129 dest_y += dest_offset; | |
1130 //printf("%d %d %d\n", src_x, src_y, dxy); | |
554 | 1131 |
1132 if(s->flags&CODEC_FLAG_EMU_EDGE){ | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1133 if(src_x<0 || src_y<0 || src_x + (motion_x&3) + 16 > s->h_edge_pos |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1134 || src_y + (motion_y&3) + h > v_edge_pos){ |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1135 emulated_edge_mc(s, ptr, linesize, 17, h+1, src_x, src_y, s->h_edge_pos, v_edge_pos); |
554 | 1136 ptr= s->edge_emu_buffer; |
1137 emu=1; | |
1138 } | |
1139 } | |
255 | 1140 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3); |
1141 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3); | |
1142 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3); | |
1143 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3); | |
1144 | |
485 | 1145 if(s->flags&CODEC_FLAG_GRAY) return; |
1146 | |
255 | 1147 mx= (motion_x>>1) | (motion_x&1); |
1148 my= (motion_y>>1) | (motion_y&1); | |
1149 | |
1150 dxy = 0; | |
1151 if ((mx & 3) != 0) | |
1152 dxy |= 1; | |
1153 if ((my & 3) != 0) | |
1154 dxy |= 2; | |
1155 mx = mx >> 2; | |
1156 my = my >> 2; | |
1157 | |
1158 src_x = s->mb_x * 8 + mx; | |
1159 src_y = s->mb_y * (8 >> field_based) + my; | |
1160 src_x = clip(src_x, -8, s->width >> 1); | |
1161 if (src_x == (s->width >> 1)) | |
1162 dxy &= ~1; | |
1163 src_y = clip(src_y, -8, height >> 1); | |
1164 if (src_y == (height >> 1)) | |
1165 dxy &= ~2; | |
1166 | |
556 | 1167 offset = (src_y * s->uvlinesize) + src_x + (src_offset >> 1); |
255 | 1168 ptr = ref_picture[1] + offset; |
554 | 1169 if(emu){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1170 emulated_edge_mc(s, ptr, s->uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
554 | 1171 ptr= s->edge_emu_buffer; |
1172 } | |
556 | 1173 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, s->uvlinesize, h >> 1); |
554 | 1174 |
255 | 1175 ptr = ref_picture[2] + offset; |
554 | 1176 if(emu){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1177 emulated_edge_mc(s, ptr, s->uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
554 | 1178 ptr= s->edge_emu_buffer; |
1179 } | |
556 | 1180 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, s->uvlinesize, h >> 1); |
255 | 1181 } |
1182 | |
1183 | |
0 | 1184 static inline void MPV_motion(MpegEncContext *s, |
1185 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
1186 int dir, UINT8 **ref_picture, | |
255 | 1187 op_pixels_func *pix_op, qpel_mc_func *qpix_op) |
0 | 1188 { |
1189 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
1190 int mb_x, mb_y, i; | |
1191 UINT8 *ptr, *dest; | |
554 | 1192 int emu=0; |
0 | 1193 |
1194 mb_x = s->mb_x; | |
1195 mb_y = s->mb_y; | |
1196 | |
1197 switch(s->mv_type) { | |
1198 case MV_TYPE_16X16: | |
255 | 1199 if(s->mcsel){ |
1200 #if 0 | |
1201 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
1202 ref_picture, 0, | |
1203 0, pix_op, | |
1204 s->sprite_offset[0][0]>>3, | |
1205 s->sprite_offset[0][1]>>3, | |
1206 16); | |
1207 #else | |
1208 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
1209 ref_picture, 0, | |
1210 16); | |
1211 #endif | |
262 | 1212 }else if(s->quarter_sample && dir==0){ //FIXME |
255 | 1213 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, |
1214 ref_picture, 0, | |
1215 0, pix_op, qpix_op, | |
1216 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
1217 }else{ | |
1218 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
1219 ref_picture, 0, | |
1220 0, pix_op, | |
1221 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
1222 } | |
0 | 1223 break; |
1224 case MV_TYPE_8X8: | |
1225 for(i=0;i<4;i++) { | |
1226 motion_x = s->mv[dir][i][0]; | |
1227 motion_y = s->mv[dir][i][1]; | |
1228 | |
1229 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
1230 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
295 | 1231 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; |
0 | 1232 |
1233 /* WARNING: do no forget half pels */ | |
1234 src_x = clip(src_x, -16, s->width); | |
1235 if (src_x == s->width) | |
1236 dxy &= ~1; | |
1237 src_y = clip(src_y, -16, s->height); | |
1238 if (src_y == s->height) | |
1239 dxy &= ~2; | |
1240 | |
1241 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
554 | 1242 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1243 if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 8 > s->h_edge_pos |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1244 || src_y + (motion_y&1) + 8 > s->v_edge_pos){ |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1245 emulated_edge_mc(s, ptr, s->linesize, 9, 9, src_x, src_y, s->h_edge_pos, s->v_edge_pos); |
554 | 1246 ptr= s->edge_emu_buffer; |
1247 } | |
1248 } | |
0 | 1249 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; |
1250 pix_op[dxy](dest, ptr, s->linesize, 8); | |
1251 } | |
485 | 1252 |
1253 if(s->flags&CODEC_FLAG_GRAY) break; | |
0 | 1254 /* In case of 8X8, we construct a single chroma motion vector |
1255 with a special rounding */ | |
1256 mx = 0; | |
1257 my = 0; | |
1258 for(i=0;i<4;i++) { | |
1259 mx += s->mv[dir][i][0]; | |
1260 my += s->mv[dir][i][1]; | |
1261 } | |
1262 if (mx >= 0) | |
1263 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1264 else { | |
1265 mx = -mx; | |
1266 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
1267 } | |
1268 if (my >= 0) | |
1269 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1270 else { | |
1271 my = -my; | |
1272 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
1273 } | |
1274 dxy = ((my & 1) << 1) | (mx & 1); | |
1275 mx >>= 1; | |
1276 my >>= 1; | |
1277 | |
1278 src_x = mb_x * 8 + mx; | |
1279 src_y = mb_y * 8 + my; | |
1280 src_x = clip(src_x, -8, s->width/2); | |
1281 if (src_x == s->width/2) | |
1282 dxy &= ~1; | |
1283 src_y = clip(src_y, -8, s->height/2); | |
1284 if (src_y == s->height/2) | |
1285 dxy &= ~2; | |
1286 | |
556 | 1287 offset = (src_y * (s->uvlinesize)) + src_x; |
0 | 1288 ptr = ref_picture[1] + offset; |
554 | 1289 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1290 if(src_x<0 || src_y<0 || src_x + (dxy &1) + 8 > s->h_edge_pos>>1 |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1291 || src_y + (dxy>>1) + 8 > s->v_edge_pos>>1){ |
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1292 emulated_edge_mc(s, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
554 | 1293 ptr= s->edge_emu_buffer; |
1294 emu=1; | |
1295 } | |
1296 } | |
556 | 1297 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8); |
554 | 1298 |
0 | 1299 ptr = ref_picture[2] + offset; |
554 | 1300 if(emu){ |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1301 emulated_edge_mc(s, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
554 | 1302 ptr= s->edge_emu_buffer; |
1303 } | |
556 | 1304 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8); |
0 | 1305 break; |
1306 case MV_TYPE_FIELD: | |
1307 if (s->picture_structure == PICT_FRAME) { | |
1308 /* top field */ | |
1309 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
1310 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
1311 1, pix_op, | |
1312 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
1313 /* bottom field */ | |
1314 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
1315 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
1316 1, pix_op, | |
1317 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
1318 } else { | |
1319 | |
1320 | |
1321 } | |
1322 break; | |
1323 } | |
1324 } | |
1325 | |
1326 | |
1327 /* put block[] to dest[] */ | |
1328 static inline void put_dct(MpegEncContext *s, | |
1329 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
1330 { | |
1331 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
|
1332 s->dct_unquantize(s, block, i, s->qscale); |
478 | 1333 ff_idct_put (dest, line_size, block); |
0 | 1334 } |
1335 | |
1336 /* add block[] to dest[] */ | |
1337 static inline void add_dct(MpegEncContext *s, | |
1338 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
1339 { | |
1340 if (s->block_last_index[i] >= 0) { | |
481 | 1341 ff_idct_add (dest, line_size, block); |
1342 } | |
1343 } | |
1344 | |
1345 static inline void add_dequant_dct(MpegEncContext *s, | |
1346 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
1347 { | |
1348 if (s->block_last_index[i] >= 0) { | |
1349 s->dct_unquantize(s, block, i, s->qscale); | |
324 | 1350 |
478 | 1351 ff_idct_add (dest, line_size, block); |
0 | 1352 } |
1353 } | |
1354 | |
456 | 1355 /** |
1356 * cleans dc, ac, coded_block for the current non intra MB | |
1357 */ | |
1358 void ff_clean_intra_table_entries(MpegEncContext *s) | |
1359 { | |
1360 int wrap = s->block_wrap[0]; | |
1361 int xy = s->block_index[0]; | |
1362 | |
1363 s->dc_val[0][xy ] = | |
1364 s->dc_val[0][xy + 1 ] = | |
1365 s->dc_val[0][xy + wrap] = | |
1366 s->dc_val[0][xy + 1 + wrap] = 1024; | |
1367 /* ac pred */ | |
1368 memset(s->ac_val[0][xy ], 0, 32 * sizeof(INT16)); | |
1369 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(INT16)); | |
1370 if (s->msmpeg4_version>=3) { | |
1371 s->coded_block[xy ] = | |
1372 s->coded_block[xy + 1 ] = | |
1373 s->coded_block[xy + wrap] = | |
1374 s->coded_block[xy + 1 + wrap] = 0; | |
1375 } | |
1376 /* chroma */ | |
1377 wrap = s->block_wrap[4]; | |
1378 xy = s->mb_x + 1 + (s->mb_y + 1) * wrap; | |
1379 s->dc_val[1][xy] = | |
1380 s->dc_val[2][xy] = 1024; | |
1381 /* ac pred */ | |
1382 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); | |
1383 memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16)); | |
1384 | |
1385 s->mbintra_table[s->mb_x + s->mb_y*s->mb_width]= 0; | |
1386 } | |
1387 | |
0 | 1388 /* generic function called after a macroblock has been parsed by the |
1389 decoder or after it has been encoded by the encoder. | |
1390 | |
1391 Important variables used: | |
1392 s->mb_intra : true if intra macroblock | |
1393 s->mv_dir : motion vector direction | |
1394 s->mv_type : motion vector type | |
1395 s->mv : motion vector | |
1396 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
1397 */ | |
1398 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
1399 { | |
244 | 1400 int mb_x, mb_y; |
481 | 1401 const int mb_xy = s->mb_y * s->mb_width + s->mb_x; |
0 | 1402 |
1403 mb_x = s->mb_x; | |
1404 mb_y = s->mb_y; | |
1405 | |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1406 #ifdef FF_POSTPROCESS |
466
805714c0c10f
new field for communicating with external postprocessing
nickols_k
parents:
462
diff
changeset
|
1407 /* Obsolete. Exists for compatibility with mplayer only. */ |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1408 quant_store[mb_y][mb_x]=s->qscale; |
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1409 //printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale); |
466
805714c0c10f
new field for communicating with external postprocessing
nickols_k
parents:
462
diff
changeset
|
1410 #else |
575 | 1411 /* even more obsolete, exists for mplayer xp only */ |
467 | 1412 if(s->avctx->quant_store) s->avctx->quant_store[mb_y*s->avctx->qstride+mb_x] = s->qscale; |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1413 #endif |
575 | 1414 s->qscale_table[mb_xy]= s->qscale; |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1415 |
0 | 1416 /* update DC predictors for P macroblocks */ |
1417 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1418 if (s->h263_pred || s->h263_aic) { |
481 | 1419 if(s->mbintra_table[mb_xy]) |
456 | 1420 ff_clean_intra_table_entries(s); |
0 | 1421 } else { |
456 | 1422 s->last_dc[0] = |
1423 s->last_dc[1] = | |
0 | 1424 s->last_dc[2] = 128 << s->intra_dc_precision; |
1425 } | |
1426 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1427 else if (s->h263_pred || s->h263_aic) |
481 | 1428 s->mbintra_table[mb_xy]=1; |
191
883f184537e6
AC table reset (memset) optimization - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
189
diff
changeset
|
1429 |
280 | 1430 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */ |
481 | 1431 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here |
1432 int motion_x, motion_y; | |
0 | 1433 |
481 | 1434 const int wrap = s->block_wrap[0]; |
1435 const int xy = s->block_index[0]; | |
0 | 1436 if (s->mb_intra) { |
1437 motion_x = 0; | |
1438 motion_y = 0; | |
1439 goto motion_init; | |
1440 } else if (s->mv_type == MV_TYPE_16X16) { | |
1441 motion_x = s->mv[0][0][0]; | |
1442 motion_y = s->mv[0][0][1]; | |
1443 motion_init: | |
1444 /* no update if 8X8 because it has been done during parsing */ | |
244 | 1445 s->motion_val[xy][0] = motion_x; |
1446 s->motion_val[xy][1] = motion_y; | |
1447 s->motion_val[xy + 1][0] = motion_x; | |
1448 s->motion_val[xy + 1][1] = motion_y; | |
1449 s->motion_val[xy + wrap][0] = motion_x; | |
1450 s->motion_val[xy + wrap][1] = motion_y; | |
1451 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
1452 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
0 | 1453 } |
1454 } | |
1455 | |
324 | 1456 if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) { |
0 | 1457 UINT8 *dest_y, *dest_cb, *dest_cr; |
481 | 1458 int dct_linesize, dct_offset; |
1459 op_pixels_func *op_pix; | |
1460 qpel_mc_func *op_qpix; | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1461 |
324 | 1462 /* avoid copy if macroblock skipped in last frame too |
1463 dont touch it for B-frames as they need the skip info from the next p-frame */ | |
1464 if (s->pict_type != B_TYPE) { | |
481 | 1465 UINT8 *mbskip_ptr = &s->mbskip_table[mb_xy]; |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1466 if (s->mb_skiped) { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1467 s->mb_skiped = 0; |
556 | 1468 |
1469 (*mbskip_ptr) ++; /* indicate that this time we skiped it */ | |
1470 if(*mbskip_ptr >99) *mbskip_ptr= 99; | |
1471 | |
324 | 1472 /* if previous was skipped too, then nothing to do ! |
1473 skip only during decoding as we might trash the buffers during encoding a bit */ | |
556 | 1474 if (*mbskip_ptr >= s->ip_buffer_count && !s->encoding) |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1475 goto the_end; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1476 } else { |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1477 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1478 } |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1479 } |
0 | 1480 |
562
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
1481 dest_y = s->current_picture [0] + (mb_y * 16* s->linesize ) + mb_x * 16; |
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
1482 dest_cb = s->current_picture[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
1483 dest_cr = s->current_picture[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
0 | 1484 |
1485 if (s->interlaced_dct) { | |
1486 dct_linesize = s->linesize * 2; | |
1487 dct_offset = s->linesize; | |
1488 } else { | |
1489 dct_linesize = s->linesize; | |
1490 dct_offset = s->linesize * 8; | |
1491 } | |
1492 | |
1493 if (!s->mb_intra) { | |
1494 /* motion handling */ | |
456 | 1495 /* decoding or more than one mb_type (MC was allready done otherwise) */ |
481 | 1496 if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){ |
327 | 1497 if ((!s->no_rounding) || s->pict_type==B_TYPE){ |
324 | 1498 op_pix = put_pixels_tab; |
1499 op_qpix= qpel_mc_rnd_tab; | |
1500 }else{ | |
1501 op_pix = put_no_rnd_pixels_tab; | |
1502 op_qpix= qpel_mc_no_rnd_tab; | |
1503 } | |
0 | 1504 |
324 | 1505 if (s->mv_dir & MV_DIR_FORWARD) { |
1506 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
327 | 1507 if ((!s->no_rounding) || s->pict_type==B_TYPE) |
324 | 1508 op_pix = avg_pixels_tab; |
1509 else | |
1510 op_pix = avg_no_rnd_pixels_tab; | |
1511 } | |
1512 if (s->mv_dir & MV_DIR_BACKWARD) { | |
1513 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
1514 } | |
0 | 1515 } |
1516 | |
481 | 1517 /* skip dequant / idct if we are really late ;) */ |
1518 if(s->hurry_up>1) goto the_end; | |
1519 | |
0 | 1520 /* add dct residue */ |
591 | 1521 if(s->encoding || !(s->mpeg2 || s->h263_msmpeg4 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){ |
481 | 1522 add_dequant_dct(s, block[0], 0, dest_y, dct_linesize); |
1523 add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1524 add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1525 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
0 | 1526 |
485 | 1527 if(!(s->flags&CODEC_FLAG_GRAY)){ |
556 | 1528 add_dequant_dct(s, block[4], 4, dest_cb, s->uvlinesize); |
1529 add_dequant_dct(s, block[5], 5, dest_cr, s->uvlinesize); | |
485 | 1530 } |
481 | 1531 } else { |
1532 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
1533 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1534 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1535 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1536 | |
485 | 1537 if(!(s->flags&CODEC_FLAG_GRAY)){ |
556 | 1538 add_dct(s, block[4], 4, dest_cb, s->uvlinesize); |
1539 add_dct(s, block[5], 5, dest_cr, s->uvlinesize); | |
485 | 1540 } |
481 | 1541 } |
0 | 1542 } else { |
1543 /* dct only in intra block */ | |
1544 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
1545 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
1546 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
1547 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
1548 | |
485 | 1549 if(!(s->flags&CODEC_FLAG_GRAY)){ |
556 | 1550 put_dct(s, block[4], 4, dest_cb, s->uvlinesize); |
1551 put_dct(s, block[5], 5, dest_cr, s->uvlinesize); | |
485 | 1552 } |
0 | 1553 } |
1554 } | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1555 the_end: |
294 | 1556 emms_c(); //FIXME remove |
1557 } | |
1558 | |
456 | 1559 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold, int skip_dc) |
1560 { | |
1561 static const char tab[64]= | |
1562 {3,2,2,1,1,1,1,1, | |
1563 1,1,1,1,1,1,1,1, | |
1564 1,1,1,1,1,1,1,1, | |
1565 0,0,0,0,0,0,0,0, | |
1566 0,0,0,0,0,0,0,0, | |
1567 0,0,0,0,0,0,0,0, | |
1568 0,0,0,0,0,0,0,0, | |
1569 0,0,0,0,0,0,0,0}; | |
1570 int score=0; | |
1571 int run=0; | |
1572 int i; | |
1573 DCTELEM *block= s->block[n]; | |
1574 const int last_index= s->block_last_index[n]; | |
1575 | |
1576 if(skip_dc) skip_dc=1; | |
1577 | |
1578 /* are all which we could set to zero are allready zero? */ | |
1579 if(last_index<=skip_dc - 1) return; | |
1580 | |
1581 for(i=0; i<=last_index; i++){ | |
1582 const int j = zigzag_direct[i]; | |
1583 const int level = ABS(block[j]); | |
1584 if(level==1){ | |
1585 if(skip_dc && i==0) continue; | |
1586 score+= tab[run]; | |
1587 run=0; | |
1588 }else if(level>1){ | |
1589 return; | |
1590 }else{ | |
1591 run++; | |
1592 } | |
1593 } | |
1594 if(score >= threshold) return; | |
1595 for(i=skip_dc; i<=last_index; i++){ | |
1596 const int j = zigzag_direct[i]; | |
1597 block[j]=0; | |
1598 } | |
1599 if(block[0]) s->block_last_index[n]= 0; | |
1600 else s->block_last_index[n]= -1; | |
1601 } | |
1602 | |
344 | 1603 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index) |
1604 { | |
1605 int i; | |
1606 const int maxlevel= s->max_qcoeff; | |
1607 const int minlevel= s->min_qcoeff; | |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
1608 |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
1609 for(i=0;i<=last_index; i++){ |
344 | 1610 const int j = zigzag_direct[i]; |
1611 int level = block[j]; | |
1612 | |
1613 if (level>maxlevel) level=maxlevel; | |
1614 else if(level<minlevel) level=minlevel; | |
1615 block[j]= level; | |
1616 } | |
1617 } | |
324 | 1618 |
1619 static void encode_mb(MpegEncContext *s, int motion_x, int motion_y) | |
294 | 1620 { |
1621 const int mb_x= s->mb_x; | |
1622 const int mb_y= s->mb_y; | |
1623 int i; | |
456 | 1624 int skip_dct[6]; |
324 | 1625 #if 0 |
1626 if (s->interlaced_dct) { | |
1627 dct_linesize = s->linesize * 2; | |
1628 dct_offset = s->linesize; | |
1629 } else { | |
1630 dct_linesize = s->linesize; | |
1631 dct_offset = s->linesize * 8; | |
1632 } | |
1633 #endif | |
456 | 1634 for(i=0; i<6; i++) skip_dct[i]=0; |
294 | 1635 |
324 | 1636 if (s->mb_intra) { |
1637 UINT8 *ptr; | |
1638 int wrap; | |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1639 int emu=0; |
294 | 1640 |
324 | 1641 wrap = s->linesize; |
1642 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16; | |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1643 if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){ |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1644 emulated_edge_mc(s, ptr, wrap, 16, 16, mb_x*16, mb_y*16, s->width, s->height); |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1645 ptr= s->edge_emu_buffer; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1646 emu=1; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1647 } |
324 | 1648 get_pixels(s->block[0], ptr , wrap); |
1649 get_pixels(s->block[1], ptr + 8, wrap); | |
1650 get_pixels(s->block[2], ptr + 8 * wrap , wrap); | |
1651 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); | |
294 | 1652 |
487 | 1653 if(s->flags&CODEC_FLAG_GRAY){ |
1654 skip_dct[4]= 1; | |
1655 skip_dct[5]= 1; | |
1656 }else{ | |
1657 wrap >>=1; | |
1658 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8; | |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1659 if(emu){ |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1660 emulated_edge_mc(s, ptr, wrap, 8, 8, mb_x*8, mb_y*8, s->width>>1, s->height>>1); |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1661 ptr= s->edge_emu_buffer; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1662 } |
487 | 1663 get_pixels(s->block[4], ptr, wrap); |
294 | 1664 |
487 | 1665 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8; |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1666 if(emu){ |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1667 emulated_edge_mc(s, ptr, wrap, 8, 8, mb_x*8, mb_y*8, s->width>>1, s->height>>1); |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1668 ptr= s->edge_emu_buffer; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1669 } |
487 | 1670 get_pixels(s->block[5], ptr, wrap); |
1671 } | |
324 | 1672 }else{ |
1673 op_pixels_func *op_pix; | |
1674 qpel_mc_func *op_qpix; | |
1675 UINT8 *dest_y, *dest_cb, *dest_cr; | |
456 | 1676 UINT8 *ptr_y, *ptr_cb, *ptr_cr; |
1677 int wrap_y, wrap_c; | |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1678 int emu=0; |
294 | 1679 |
324 | 1680 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize ) + mb_x * 16; |
556 | 1681 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; |
1682 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; | |
456 | 1683 wrap_y = s->linesize; |
1684 wrap_c = wrap_y>>1; | |
1685 ptr_y = s->new_picture[0] + (mb_y * 16 * wrap_y) + mb_x * 16; | |
1686 ptr_cb = s->new_picture[1] + (mb_y * 8 * wrap_c) + mb_x * 8; | |
1687 ptr_cr = s->new_picture[2] + (mb_y * 8 * wrap_c) + mb_x * 8; | |
324 | 1688 |
327 | 1689 if ((!s->no_rounding) || s->pict_type==B_TYPE){ |
324 | 1690 op_pix = put_pixels_tab; |
1691 op_qpix= qpel_mc_rnd_tab; | |
295 | 1692 }else{ |
324 | 1693 op_pix = put_no_rnd_pixels_tab; |
1694 op_qpix= qpel_mc_no_rnd_tab; | |
1695 } | |
295 | 1696 |
324 | 1697 if (s->mv_dir & MV_DIR_FORWARD) { |
1698 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
327 | 1699 if ((!s->no_rounding) || s->pict_type==B_TYPE) |
324 | 1700 op_pix = avg_pixels_tab; |
1701 else | |
1702 op_pix = avg_no_rnd_pixels_tab; | |
1703 } | |
1704 if (s->mv_dir & MV_DIR_BACKWARD) { | |
1705 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
1706 } | |
295 | 1707 |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1708 if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){ |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1709 emulated_edge_mc(s, ptr_y, wrap_y, 16, 16, mb_x*16, mb_y*16, s->width, s->height); |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1710 ptr_y= s->edge_emu_buffer; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1711 emu=1; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1712 } |
456 | 1713 diff_pixels(s->block[0], ptr_y , dest_y , wrap_y); |
1714 diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y); | |
1715 diff_pixels(s->block[2], ptr_y + 8 * wrap_y , dest_y + 8 * wrap_y , wrap_y); | |
1716 diff_pixels(s->block[3], ptr_y + 8 * wrap_y + 8, dest_y + 8 * wrap_y + 8, wrap_y); | |
487 | 1717 |
1718 if(s->flags&CODEC_FLAG_GRAY){ | |
1719 skip_dct[4]= 1; | |
1720 skip_dct[5]= 1; | |
1721 }else{ | |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1722 if(emu){ |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1723 emulated_edge_mc(s, ptr_cb, wrap_c, 8, 8, mb_x*8, mb_y*8, s->width>>1, s->height>>1); |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1724 ptr_cb= s->edge_emu_buffer; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1725 } |
487 | 1726 diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c); |
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1727 if(emu){ |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1728 emulated_edge_mc(s, ptr_cr, wrap_c, 8, 8, mb_x*8, mb_y*8, s->width>>1, s->height>>1); |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1729 ptr_cr= s->edge_emu_buffer; |
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1730 } |
487 | 1731 diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c); |
1732 } | |
1733 | |
456 | 1734 /* pre quantization */ |
1735 if(s->mc_mb_var[s->mb_width*mb_y+ mb_x]<2*s->qscale*s->qscale){ | |
1736 if(pix_abs8x8(ptr_y , dest_y , wrap_y) < 20*s->qscale) skip_dct[0]= 1; | |
1737 if(pix_abs8x8(ptr_y + 8, dest_y + 8, wrap_y) < 20*s->qscale) skip_dct[1]= 1; | |
1738 if(pix_abs8x8(ptr_y + 8*wrap_y , dest_y + 8*wrap_y , wrap_y) < 20*s->qscale) skip_dct[2]= 1; | |
1739 if(pix_abs8x8(ptr_y + 8*wrap_y + 8, dest_y + 8*wrap_y + 8, wrap_y) < 20*s->qscale) skip_dct[3]= 1; | |
1740 if(pix_abs8x8(ptr_cb , dest_cb , wrap_y) < 20*s->qscale) skip_dct[4]= 1; | |
1741 if(pix_abs8x8(ptr_cr , dest_cr , wrap_y) < 20*s->qscale) skip_dct[5]= 1; | |
1742 #if 0 | |
1743 { | |
1744 static int stat[7]; | |
1745 int num=0; | |
1746 for(i=0; i<6; i++) | |
1747 if(skip_dct[i]) num++; | |
1748 stat[num]++; | |
1749 | |
1750 if(s->mb_x==0 && s->mb_y==0){ | |
1751 for(i=0; i<7; i++){ | |
1752 printf("%6d %1d\n", stat[i], i); | |
1753 } | |
1754 } | |
1755 } | |
1756 #endif | |
1757 } | |
324 | 1758 |
294 | 1759 } |
1760 | |
1761 #if 0 | |
1762 { | |
1763 float adap_parm; | |
1764 | |
1765 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / | |
1766 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); | |
1767 | |
1768 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", | |
1769 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', | |
1770 s->qscale, adap_parm, s->qscale*adap_parm, | |
1771 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); | |
1772 } | |
1773 #endif | |
1774 /* DCT & quantize */ | |
344 | 1775 if(s->out_format==FMT_MJPEG){ |
1776 for(i=0;i<6;i++) { | |
1777 int overflow; | |
1778 s->block_last_index[i] = dct_quantize(s, s->block[i], i, 8, &overflow); | |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
1779 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]); |
344 | 1780 } |
1781 }else{ | |
1782 for(i=0;i<6;i++) { | |
456 | 1783 if(!skip_dct[i]){ |
1784 int overflow; | |
1785 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale, &overflow); | |
344 | 1786 // FIXME we could decide to change to quantizer instead of clipping |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
1787 // JS: I don't think that would be a good idea it could lower quality instead |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
1788 // of improve it. Just INTRADC clipping deserves changes in quantizer |
456 | 1789 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]); |
1790 }else | |
1791 s->block_last_index[i]= -1; | |
344 | 1792 } |
456 | 1793 if(s->luma_elim_threshold && !s->mb_intra) |
1794 for(i=0; i<4; i++) | |
1795 dct_single_coeff_elimination(s, i, s->luma_elim_threshold, 0); | |
1796 if(s->chroma_elim_threshold && !s->mb_intra) | |
1797 for(i=4; i<6; i++) | |
1798 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold, 1); | |
294 | 1799 } |
1800 | |
487 | 1801 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){ |
1802 s->block_last_index[4]= | |
1803 s->block_last_index[5]= 0; | |
1804 s->block[4][0]= | |
1805 s->block[5][0]= 128; | |
1806 } | |
1807 | |
294 | 1808 /* huffman encode */ |
1809 switch(s->out_format) { | |
1810 case FMT_MPEG1: | |
1811 mpeg1_encode_mb(s, s->block, motion_x, motion_y); | |
1812 break; | |
1813 case FMT_H263: | |
1814 if (s->h263_msmpeg4) | |
1815 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1816 else if(s->h263_pred) | |
1817 mpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
1818 else | |
1819 h263_encode_mb(s, s->block, motion_x, motion_y); | |
1820 break; | |
1821 case FMT_MJPEG: | |
1822 mjpeg_encode_mb(s, s->block); | |
1823 break; | |
1824 } | |
1825 } | |
1826 | |
456 | 1827 void ff_copy_bits(PutBitContext *pb, UINT8 *src, int length) |
294 | 1828 { |
326 | 1829 int bytes= length>>4; |
1830 int bits= length&15; | |
1831 int i; | |
1832 | |
456 | 1833 if(length==0) return; |
1834 | |
326 | 1835 for(i=0; i<bytes; i++) put_bits(pb, 16, be2me_16(((uint16_t*)src)[i])); |
1836 put_bits(pb, bits, be2me_16(((uint16_t*)src)[i])>>(16-bits)); | |
0 | 1837 } |
1838 | |
456 | 1839 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){ |
326 | 1840 int i; |
1841 | |
1842 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop? | |
1843 | |
1844 /* mpeg1 */ | |
1845 d->mb_incr= s->mb_incr; | |
1846 for(i=0; i<3; i++) | |
1847 d->last_dc[i]= s->last_dc[i]; | |
1848 | |
1849 /* statistics */ | |
1850 d->mv_bits= s->mv_bits; | |
1851 d->i_tex_bits= s->i_tex_bits; | |
1852 d->p_tex_bits= s->p_tex_bits; | |
1853 d->i_count= s->i_count; | |
1854 d->p_count= s->p_count; | |
1855 d->skip_count= s->skip_count; | |
1856 d->misc_bits= s->misc_bits; | |
329 | 1857 d->last_bits= 0; |
327 | 1858 |
1859 d->mb_skiped= s->mb_skiped; | |
326 | 1860 } |
1861 | |
456 | 1862 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){ |
326 | 1863 int i; |
1864 | |
1865 memcpy(d->mv, s->mv, 2*4*2*sizeof(int)); | |
1866 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop? | |
1867 | |
1868 /* mpeg1 */ | |
1869 d->mb_incr= s->mb_incr; | |
1870 for(i=0; i<3; i++) | |
1871 d->last_dc[i]= s->last_dc[i]; | |
1872 | |
1873 /* statistics */ | |
1874 d->mv_bits= s->mv_bits; | |
1875 d->i_tex_bits= s->i_tex_bits; | |
1876 d->p_tex_bits= s->p_tex_bits; | |
1877 d->i_count= s->i_count; | |
1878 d->p_count= s->p_count; | |
1879 d->skip_count= s->skip_count; | |
1880 d->misc_bits= s->misc_bits; | |
1881 | |
1882 d->mb_intra= s->mb_intra; | |
327 | 1883 d->mb_skiped= s->mb_skiped; |
326 | 1884 d->mv_type= s->mv_type; |
1885 d->mv_dir= s->mv_dir; | |
1886 d->pb= s->pb; | |
456 | 1887 if(s->data_partitioning){ |
1888 d->pb2= s->pb2; | |
1889 d->tex_pb= s->tex_pb; | |
1890 } | |
326 | 1891 d->block= s->block; |
1892 for(i=0; i<6; i++) | |
1893 d->block_last_index[i]= s->block_last_index[i]; | |
1894 } | |
1895 | |
456 | 1896 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type, |
1897 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2], | |
1898 int *dmin, int *next_block, int motion_x, int motion_y) | |
1899 { | |
1900 int bits_count; | |
1901 | |
1902 copy_context_before_encode(s, backup, type); | |
1903 | |
1904 s->block= s->blocks[*next_block]; | |
1905 s->pb= pb[*next_block]; | |
1906 if(s->data_partitioning){ | |
1907 s->pb2 = pb2 [*next_block]; | |
1908 s->tex_pb= tex_pb[*next_block]; | |
1909 } | |
1910 | |
1911 encode_mb(s, motion_x, motion_y); | |
1912 | |
1913 bits_count= get_bit_count(&s->pb); | |
1914 if(s->data_partitioning){ | |
1915 bits_count+= get_bit_count(&s->pb2); | |
1916 bits_count+= get_bit_count(&s->tex_pb); | |
1917 } | |
1918 | |
1919 if(bits_count<*dmin){ | |
1920 *dmin= bits_count; | |
1921 *next_block^=1; | |
1922 | |
1923 copy_context_after_encode(best, s, type); | |
1924 } | |
1925 } | |
326 | 1926 |
0 | 1927 static void encode_picture(MpegEncContext *s, int picture_number) |
1928 { | |
294 | 1929 int mb_x, mb_y, last_gob, pdif = 0; |
1930 int i; | |
286 | 1931 int bits; |
326 | 1932 MpegEncContext best_s, backup_s; |
456 | 1933 UINT8 bit_buf[2][3000]; |
1934 UINT8 bit_buf2[2][3000]; | |
1935 UINT8 bit_buf_tex[2][3000]; | |
1936 PutBitContext pb[2], pb2[2], tex_pb[2]; | |
1937 | |
1938 for(i=0; i<2; i++){ | |
1939 init_put_bits(&pb [i], bit_buf [i], 3000, NULL, NULL); | |
1940 init_put_bits(&pb2 [i], bit_buf2 [i], 3000, NULL, NULL); | |
1941 init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000, NULL, NULL); | |
1942 } | |
0 | 1943 |
1944 s->picture_number = picture_number; | |
268 | 1945 |
294 | 1946 s->block_wrap[0]= |
1947 s->block_wrap[1]= | |
1948 s->block_wrap[2]= | |
1949 s->block_wrap[3]= s->mb_width*2 + 2; | |
1950 s->block_wrap[4]= | |
1951 s->block_wrap[5]= s->mb_width + 2; | |
1952 | |
268 | 1953 /* Reset the average MB variance */ |
456 | 1954 s->mb_var_sum = 0; |
1955 s->mc_mb_var_sum = 0; | |
327 | 1956 |
1957 /* we need to initialize some time vars before we can encode b-frames */ | |
1958 if (s->h263_pred && !s->h263_msmpeg4) | |
1959 ff_set_mpeg4_time(s, s->picture_number); | |
1960 | |
268 | 1961 /* Estimate motion for every MB */ |
324 | 1962 if(s->pict_type != I_TYPE){ |
294 | 1963 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
1964 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; | |
1965 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
1966 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
1967 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
1968 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
1969 s->mb_x = mb_x; | |
1970 s->mb_y = mb_y; | |
1971 s->block_index[0]+=2; | |
1972 s->block_index[1]+=2; | |
1973 s->block_index[2]+=2; | |
1974 s->block_index[3]+=2; | |
268 | 1975 |
294 | 1976 /* compute motion vector & mb_type and store in context */ |
324 | 1977 if(s->pict_type==B_TYPE) |
1978 ff_estimate_b_frame_motion(s, mb_x, mb_y); | |
1979 else | |
1980 ff_estimate_p_frame_motion(s, mb_x, mb_y); | |
294 | 1981 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER; |
268 | 1982 } |
1983 } | |
294 | 1984 emms_c(); |
456 | 1985 }else /* if(s->pict_type == I_TYPE) */{ |
294 | 1986 /* I-Frame */ |
1987 //FIXME do we need to zero them? | |
1988 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2); | |
324 | 1989 memset(s->p_mv_table , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2); |
294 | 1990 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
268 | 1991 } |
1992 | |
456 | 1993 if(s->mb_var_sum < s->mc_mb_var_sum && s->pict_type == P_TYPE){ //FIXME subtract MV bits |
271 | 1994 s->pict_type= I_TYPE; |
294 | 1995 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
329 | 1996 if(s->max_b_frames==0){ |
1997 s->input_pict_type= I_TYPE; | |
1998 s->input_picture_in_gop_number=0; | |
1999 } | |
294 | 2000 //printf("Scene change detected, encoding as I Frame\n"); |
271 | 2001 } |
324 | 2002 |
2003 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) | |
2004 s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER); | |
2005 ff_fix_long_p_mvs(s); | |
2006 if(s->pict_type==B_TYPE){ | |
2007 s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD); | |
2008 s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD); | |
327 | 2009 |
2010 ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD); | |
2011 ff_fix_long_b_mvs(s, s->b_back_mv_table, s->b_code, MB_TYPE_BACKWARD); | |
2012 ff_fix_long_b_mvs(s, s->b_bidir_forw_mv_table, s->f_code, MB_TYPE_BIDIR); | |
2013 ff_fix_long_b_mvs(s, s->b_bidir_back_mv_table, s->b_code, MB_TYPE_BIDIR); | |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
2014 } |
324 | 2015 |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
2016 //printf("f_code %d ///\n", s->f_code); |
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
2017 |
271 | 2018 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var); |
329 | 2019 if(s->flags&CODEC_FLAG_PASS2) |
2020 s->qscale = ff_rate_estimate_qscale_pass2(s); | |
2021 else if (!s->fixed_qscale) | |
2022 s->qscale = ff_rate_estimate_qscale(s); | |
2023 | |
0 | 2024 if (s->out_format == FMT_MJPEG) { |
2025 /* for mjpeg, we do include qscale in the matrix */ | |
533
3c07cf9595de
adding ff prefix to avoid global name conficts with xvid (patch by Marko Kreen <marko at l-t.ee>)
michaelni
parents:
514
diff
changeset
|
2026 s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0]; |
0 | 2027 for(i=1;i<64;i++) |
533
3c07cf9595de
adding ff prefix to avoid global name conficts with xvid (patch by Marko Kreen <marko at l-t.ee>)
michaelni
parents:
514
diff
changeset
|
2028 s->intra_matrix[i] = CLAMP_TO_8BIT((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3); |
344 | 2029 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, |
2030 s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias); | |
0 | 2031 } |
2032 | |
286 | 2033 s->last_bits= get_bit_count(&s->pb); |
0 | 2034 switch(s->out_format) { |
2035 case FMT_MJPEG: | |
2036 mjpeg_picture_header(s); | |
2037 break; | |
2038 case FMT_H263: | |
2039 if (s->h263_msmpeg4) | |
2040 msmpeg4_encode_picture_header(s, picture_number); | |
2041 else if (s->h263_pred) | |
2042 mpeg4_encode_picture_header(s, picture_number); | |
2043 else if (s->h263_rv10) | |
2044 rv10_encode_picture_header(s, picture_number); | |
2045 else | |
2046 h263_encode_picture_header(s, picture_number); | |
2047 break; | |
2048 case FMT_MPEG1: | |
2049 mpeg1_encode_picture_header(s, picture_number); | |
2050 break; | |
2051 } | |
286 | 2052 bits= get_bit_count(&s->pb); |
2053 s->header_bits= bits - s->last_bits; | |
2054 s->last_bits= bits; | |
2055 s->mv_bits=0; | |
2056 s->misc_bits=0; | |
2057 s->i_tex_bits=0; | |
2058 s->p_tex_bits=0; | |
2059 s->i_count=0; | |
2060 s->p_count=0; | |
2061 s->skip_count=0; | |
2062 | |
0 | 2063 /* init last dc values */ |
2064 /* note: quant matrix value (8) is implied here */ | |
2065 s->last_dc[0] = 128; | |
2066 s->last_dc[1] = 128; | |
2067 s->last_dc[2] = 128; | |
2068 s->mb_incr = 1; | |
2069 s->last_mv[0][0][0] = 0; | |
2070 s->last_mv[0][0][1] = 0; | |
2071 | |
162 | 2072 /* Get the GOB height based on picture height */ |
231 | 2073 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
162 | 2074 if (s->height <= 400) |
2075 s->gob_index = 1; | |
2076 else if (s->height <= 800) | |
2077 s->gob_index = 2; | |
2078 else | |
2079 s->gob_index = 4; | |
456 | 2080 }else if(s->codec_id==CODEC_ID_MPEG4){ |
2081 s->gob_index = 1; | |
162 | 2082 } |
456 | 2083 |
2084 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE) | |
2085 ff_mpeg4_init_partitions(s); | |
2086 | |
2087 s->resync_mb_x=0; | |
2088 s->resync_mb_y=0; | |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2089 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
456 | 2090 /* Put GOB header based on RTP MTU for formats which support it per line (H263*)*/ |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2091 /* TODO: Put all this stuff in a separate generic function */ |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2092 if (s->rtp_mode) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2093 if (!mb_y) { |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2094 s->ptr_lastgob = s->pb.buf; |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2095 s->ptr_last_mb_line = s->pb.buf; |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2096 } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) { |
456 | 2097 // MN: we could move the space check from h263 -> here, as its not h263 specific |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2098 last_gob = h263_encode_gob_header(s, mb_y); |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2099 if (last_gob) { |
456 | 2100 s->first_slice_line = 1; |
2101 }else{ | |
2102 /*MN: we reset it here instead at the end of each line cuz mpeg4 can have | |
2103 slice lines starting & ending in the middle*/ | |
2104 s->first_slice_line = 0; | |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2105 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2106 } |
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2107 } |
499 | 2108 |
2109 s->y_dc_scale= s->y_dc_scale_table[ s->qscale ]; | |
2110 s->c_dc_scale= s->c_dc_scale_table[ s->qscale ]; | |
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2111 |
266 | 2112 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; |
2113 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
2114 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
2115 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
2116 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
2117 s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
2118 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
327 | 2119 const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x]; |
2120 const int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
456 | 2121 // int d; |
294 | 2122 int dmin=10000000; |
0 | 2123 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
2124 s->mb_x = mb_x; |
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
2125 s->mb_y = mb_y; |
266 | 2126 s->block_index[0]+=2; |
2127 s->block_index[1]+=2; | |
2128 s->block_index[2]+=2; | |
2129 s->block_index[3]+=2; | |
2130 s->block_index[4]++; | |
2131 s->block_index[5]++; | |
456 | 2132 |
2133 /* write gob / video packet header for formats which support it at any MB (MPEG4) */ | |
2134 if(s->rtp_mode && s->mb_y>0 && s->codec_id==CODEC_ID_MPEG4){ | |
2135 int pdif= pbBufPtr(&s->pb) - s->ptr_lastgob; | |
2136 | |
2137 //the *2 is there so we stay below the requested size | |
2138 if(pdif + s->mb_line_avgsize/s->mb_width >= s->rtp_payload_size){ | |
2139 if(s->codec_id==CODEC_ID_MPEG4){ | |
2140 if(s->data_partitioning && s->pict_type!=B_TYPE){ | |
2141 ff_mpeg4_merge_partitions(s); | |
2142 ff_mpeg4_init_partitions(s); | |
2143 } | |
2144 ff_mpeg4_encode_video_packet_header(s); | |
2145 | |
2146 if(s->flags&CODEC_FLAG_PASS1){ | |
2147 int bits= get_bit_count(&s->pb); | |
2148 s->misc_bits+= bits - s->last_bits; | |
2149 s->last_bits= bits; | |
2150 } | |
2151 ff_mpeg4_clean_buffers(s); | |
2152 } | |
2153 s->ptr_lastgob = pbBufPtr(&s->pb); | |
2154 s->first_slice_line=1; | |
2155 s->resync_mb_x=mb_x; | |
2156 s->resync_mb_y=mb_y; | |
2157 } | |
2158 | |
2159 if( (s->resync_mb_x == s->mb_x) | |
2160 && s->resync_mb_y+1 == s->mb_y){ | |
2161 s->first_slice_line=0; | |
2162 } | |
2163 } | |
2164 | |
294 | 2165 if(mb_type & (mb_type-1)){ // more than 1 MB type possible |
327 | 2166 int next_block=0; |
456 | 2167 int pb_bits_count, pb2_bits_count, tex_pb_bits_count; |
326 | 2168 |
2169 copy_context_before_encode(&backup_s, s, -1); | |
456 | 2170 backup_s.pb= s->pb; |
2171 best_s.data_partitioning= s->data_partitioning; | |
2172 if(s->data_partitioning){ | |
2173 backup_s.pb2= s->pb2; | |
2174 backup_s.tex_pb= s->tex_pb; | |
2175 } | |
326 | 2176 |
294 | 2177 if(mb_type&MB_TYPE_INTER){ |
327 | 2178 s->mv_dir = MV_DIR_FORWARD; |
295 | 2179 s->mv_type = MV_TYPE_16X16; |
294 | 2180 s->mb_intra= 0; |
324 | 2181 s->mv[0][0][0] = s->p_mv_table[xy][0]; |
2182 s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
456 | 2183 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER, pb, pb2, tex_pb, |
2184 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]); | |
0 | 2185 } |
326 | 2186 if(mb_type&MB_TYPE_INTER4V){ |
327 | 2187 s->mv_dir = MV_DIR_FORWARD; |
295 | 2188 s->mv_type = MV_TYPE_8X8; |
2189 s->mb_intra= 0; | |
2190 for(i=0; i<4; i++){ | |
2191 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
2192 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
2193 } | |
456 | 2194 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER4V, pb, pb2, tex_pb, |
2195 &dmin, &next_block, 0, 0); | |
327 | 2196 } |
2197 if(mb_type&MB_TYPE_FORWARD){ | |
2198 s->mv_dir = MV_DIR_FORWARD; | |
2199 s->mv_type = MV_TYPE_16X16; | |
2200 s->mb_intra= 0; | |
2201 s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
2202 s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
456 | 2203 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_FORWARD, pb, pb2, tex_pb, |
2204 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]); | |
327 | 2205 } |
2206 if(mb_type&MB_TYPE_BACKWARD){ | |
2207 s->mv_dir = MV_DIR_BACKWARD; | |
2208 s->mv_type = MV_TYPE_16X16; | |
2209 s->mb_intra= 0; | |
2210 s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
2211 s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
456 | 2212 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BACKWARD, pb, pb2, tex_pb, |
2213 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]); | |
327 | 2214 } |
2215 if(mb_type&MB_TYPE_BIDIR){ | |
2216 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; | |
2217 s->mv_type = MV_TYPE_16X16; | |
2218 s->mb_intra= 0; | |
2219 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
2220 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
2221 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
2222 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
456 | 2223 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BIDIR, pb, pb2, tex_pb, |
2224 &dmin, &next_block, 0, 0); | |
327 | 2225 } |
2226 if(mb_type&MB_TYPE_DIRECT){ | |
2227 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; | |
2228 s->mv_type = MV_TYPE_16X16; //FIXME | |
2229 s->mb_intra= 0; | |
2230 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0]; | |
2231 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1]; | |
2232 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0]; | |
2233 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1]; | |
456 | 2234 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_DIRECT, pb, pb2, tex_pb, |
2235 &dmin, &next_block, s->b_direct_mv_table[xy][0], s->b_direct_mv_table[xy][1]); | |
295 | 2236 } |
294 | 2237 if(mb_type&MB_TYPE_INTRA){ |
327 | 2238 s->mv_dir = MV_DIR_FORWARD; |
295 | 2239 s->mv_type = MV_TYPE_16X16; |
294 | 2240 s->mb_intra= 1; |
2241 s->mv[0][0][0] = 0; | |
2242 s->mv[0][0][1] = 0; | |
456 | 2243 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTRA, pb, pb2, tex_pb, |
2244 &dmin, &next_block, 0, 0); | |
326 | 2245 /* force cleaning of ac/dc pred stuff if needed ... */ |
2246 if(s->h263_pred || s->h263_aic) | |
2247 s->mbintra_table[mb_x + mb_y*s->mb_width]=1; | |
295 | 2248 } |
326 | 2249 copy_context_after_encode(s, &best_s, -1); |
456 | 2250 |
2251 pb_bits_count= get_bit_count(&s->pb); | |
2252 flush_put_bits(&s->pb); | |
2253 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count); | |
2254 s->pb= backup_s.pb; | |
2255 | |
2256 if(s->data_partitioning){ | |
2257 pb2_bits_count= get_bit_count(&s->pb2); | |
2258 flush_put_bits(&s->pb2); | |
2259 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count); | |
2260 s->pb2= backup_s.pb2; | |
2261 | |
2262 tex_pb_bits_count= get_bit_count(&s->tex_pb); | |
2263 flush_put_bits(&s->tex_pb); | |
2264 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count); | |
2265 s->tex_pb= backup_s.tex_pb; | |
2266 } | |
329 | 2267 s->last_bits= get_bit_count(&s->pb); |
294 | 2268 } else { |
324 | 2269 int motion_x, motion_y; |
2270 s->mv_type=MV_TYPE_16X16; | |
294 | 2271 // only one MB-Type possible |
327 | 2272 switch(mb_type){ |
2273 case MB_TYPE_INTRA: | |
324 | 2274 s->mv_dir = MV_DIR_FORWARD; |
294 | 2275 s->mb_intra= 1; |
324 | 2276 motion_x= s->mv[0][0][0] = 0; |
2277 motion_y= s->mv[0][0][1] = 0; | |
327 | 2278 break; |
2279 case MB_TYPE_INTER: | |
324 | 2280 s->mv_dir = MV_DIR_FORWARD; |
2281 s->mb_intra= 0; | |
2282 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0]; | |
2283 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
327 | 2284 break; |
456 | 2285 case MB_TYPE_INTER4V: |
2286 s->mv_dir = MV_DIR_FORWARD; | |
2287 s->mv_type = MV_TYPE_8X8; | |
2288 s->mb_intra= 0; | |
2289 for(i=0; i<4; i++){ | |
2290 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
2291 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
2292 } | |
2293 motion_x= motion_y= 0; | |
2294 break; | |
327 | 2295 case MB_TYPE_DIRECT: |
324 | 2296 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; |
2297 s->mb_intra= 0; | |
327 | 2298 motion_x=s->b_direct_mv_table[xy][0]; |
2299 motion_y=s->b_direct_mv_table[xy][1]; | |
2300 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0]; | |
2301 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1]; | |
2302 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0]; | |
2303 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1]; | |
2304 break; | |
2305 case MB_TYPE_BIDIR: | |
324 | 2306 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; |
294 | 2307 s->mb_intra= 0; |
324 | 2308 motion_x=0; |
2309 motion_y=0; | |
2310 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
2311 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
2312 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
2313 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
327 | 2314 break; |
2315 case MB_TYPE_BACKWARD: | |
324 | 2316 s->mv_dir = MV_DIR_BACKWARD; |
2317 s->mb_intra= 0; | |
2318 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
2319 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
327 | 2320 break; |
2321 case MB_TYPE_FORWARD: | |
324 | 2322 s->mv_dir = MV_DIR_FORWARD; |
2323 s->mb_intra= 0; | |
2324 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
2325 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
2326 // printf(" %d %d ", motion_x, motion_y); | |
327 | 2327 break; |
2328 default: | |
324 | 2329 motion_x=motion_y=0; //gcc warning fix |
2330 printf("illegal MB type\n"); | |
294 | 2331 } |
324 | 2332 encode_mb(s, motion_x, motion_y); |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2333 } |
327 | 2334 /* clean the MV table in IPS frames for direct mode in B frames */ |
2335 if(s->mb_intra /* && I,P,S_TYPE */){ | |
2336 s->p_mv_table[xy][0]=0; | |
2337 s->p_mv_table[xy][1]=0; | |
2338 } | |
0 | 2339 |
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
|
2340 MPV_decode_mb(s, s->block); |
456 | 2341 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_width, get_bit_count(&s->pb)); |
0 | 2342 } |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
2343 |
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
2344 |
231 | 2345 /* Obtain average GOB size for RTP */ |
2346 if (s->rtp_mode) { | |
2347 if (!mb_y) | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
2348 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
231 | 2349 else if (!(mb_y % s->gob_index)) { |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
2350 s->mb_line_avgsize = (s->mb_line_avgsize + pbBufPtr(&s->pb) - s->ptr_last_mb_line) >> 1; |
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
2351 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
231 | 2352 } |
2353 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
2354 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
456 | 2355 if(s->codec_id!=CODEC_ID_MPEG4) s->first_slice_line = 0; //FIXME clean |
231 | 2356 } |
0 | 2357 } |
294 | 2358 emms_c(); |
286 | 2359 |
456 | 2360 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE) |
2361 ff_mpeg4_merge_partitions(s); | |
2362 | |
2363 if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE) | |
208 | 2364 msmpeg4_encode_ext_header(s); |
2365 | |
456 | 2366 if(s->codec_id==CODEC_ID_MPEG4) |
2367 ff_mpeg4_stuffing(&s->pb); | |
2368 | |
162 | 2369 //if (s->gob_number) |
2370 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
231 | 2371 |
2372 /* Send the last GOB if RTP */ | |
2373 if (s->rtp_mode) { | |
2374 flush_put_bits(&s->pb); | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
2375 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 2376 /* Call the RTP callback to send the last GOB */ |
2377 if (s->rtp_callback) | |
2378 s->rtp_callback(s->ptr_lastgob, pdif, s->gob_number); | |
234
5fc0c3af3fe4
alternative bitstream writer (disabled by default, uncomment #define ALT_BISTREAM_WRITER in common.h if u want to try it)
michaelni
parents:
233
diff
changeset
|
2379 s->ptr_lastgob = pbBufPtr(&s->pb); |
231 | 2380 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
2381 } | |
0 | 2382 } |
2383 | |
220 | 2384 static int dct_quantize_c(MpegEncContext *s, |
0 | 2385 DCTELEM *block, int n, |
344 | 2386 int qscale, int *overflow) |
0 | 2387 { |
2388 int i, j, level, last_non_zero, q; | |
2389 const int *qmat; | |
344 | 2390 int bias; |
2391 int max=0; | |
2392 unsigned int threshold1, threshold2; | |
456 | 2393 |
0 | 2394 av_fdct (block); |
2395 | |
64 | 2396 /* we need this permutation so that we correct the IDCT |
2397 permutation. will be moved into DCT code */ | |
2398 block_permute(block); | |
2399 | |
0 | 2400 if (s->mb_intra) { |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2401 if (!s->h263_aic) { |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2402 if (n < 4) |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2403 q = s->y_dc_scale; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2404 else |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2405 q = s->c_dc_scale; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2406 q = q << 3; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2407 } else |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2408 /* For AIC we skip quant/dequant of INTRADC */ |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2409 q = 1 << 3; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2410 |
0 | 2411 /* note: block[0] is assumed to be positive */ |
2412 block[0] = (block[0] + (q >> 1)) / q; | |
2413 i = 1; | |
2414 last_non_zero = 0; | |
344 | 2415 qmat = s->q_intra_matrix[qscale]; |
2416 bias= s->intra_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT); | |
0 | 2417 } else { |
2418 i = 0; | |
2419 last_non_zero = -1; | |
344 | 2420 qmat = s->q_inter_matrix[qscale]; |
2421 bias= s->inter_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT); | |
0 | 2422 } |
344 | 2423 threshold1= (1<<(QMAT_SHIFT - 3)) - bias - 1; |
2424 threshold2= threshold1<<1; | |
0 | 2425 |
2426 for(;i<64;i++) { | |
2427 j = zigzag_direct[i]; | |
2428 level = block[j]; | |
2429 level = level * qmat[j]; | |
344 | 2430 |
2431 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
2432 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
2433 if(((unsigned)(level+threshold1))>threshold2){ | |
2434 if(level>0){ | |
2435 level= (bias + level)>>(QMAT_SHIFT - 3); | |
2436 block[j]= level; | |
2437 }else{ | |
2438 level= (bias - level)>>(QMAT_SHIFT - 3); | |
2439 block[j]= -level; | |
0 | 2440 } |
344 | 2441 max |=level; |
0 | 2442 last_non_zero = i; |
344 | 2443 }else{ |
2444 block[j]=0; | |
0 | 2445 } |
2446 } | |
344 | 2447 *overflow= s->max_qcoeff < max; //overflow might have happend |
2448 | |
0 | 2449 return last_non_zero; |
2450 } | |
2451 | |
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
|
2452 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
|
2453 DCTELEM *block, int n, int qscale) |
0 | 2454 { |
200 | 2455 int i, level, nCoeffs; |
0 | 2456 const UINT16 *quant_matrix; |
2457 | |
200 | 2458 if(s->alternate_scan) nCoeffs= 64; |
2459 else nCoeffs= s->block_last_index[n]+1; | |
2460 | |
0 | 2461 if (s->mb_intra) { |
2462 if (n < 4) | |
2463 block[0] = block[0] * s->y_dc_scale; | |
2464 else | |
2465 block[0] = block[0] * s->c_dc_scale; | |
2466 /* XXX: only mpeg1 */ | |
2467 quant_matrix = s->intra_matrix; | |
200 | 2468 for(i=1;i<nCoeffs;i++) { |
2469 int j= zigzag_direct[i]; | |
2470 level = block[j]; | |
0 | 2471 if (level) { |
2472 if (level < 0) { | |
2473 level = -level; | |
200 | 2474 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 2475 level = (level - 1) | 1; |
2476 level = -level; | |
2477 } else { | |
200 | 2478 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
0 | 2479 level = (level - 1) | 1; |
2480 } | |
2481 #ifdef PARANOID | |
2482 if (level < -2048 || level > 2047) | |
2483 fprintf(stderr, "unquant error %d %d\n", i, level); | |
2484 #endif | |
200 | 2485 block[j] = level; |
0 | 2486 } |
2487 } | |
2488 } else { | |
2489 i = 0; | |
344 | 2490 quant_matrix = s->inter_matrix; |
217 | 2491 for(;i<nCoeffs;i++) { |
200 | 2492 int j= zigzag_direct[i]; |
2493 level = block[j]; | |
0 | 2494 if (level) { |
2495 if (level < 0) { | |
2496 level = -level; | |
2497 level = (((level << 1) + 1) * qscale * | |
200 | 2498 ((int) (quant_matrix[j]))) >> 4; |
0 | 2499 level = (level - 1) | 1; |
2500 level = -level; | |
2501 } else { | |
2502 level = (((level << 1) + 1) * qscale * | |
200 | 2503 ((int) (quant_matrix[j]))) >> 4; |
0 | 2504 level = (level - 1) | 1; |
2505 } | |
2506 #ifdef PARANOID | |
2507 if (level < -2048 || level > 2047) | |
2508 fprintf(stderr, "unquant error %d %d\n", i, level); | |
2509 #endif | |
200 | 2510 block[j] = level; |
0 | 2511 } |
2512 } | |
2513 } | |
2514 } | |
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
|
2515 |
325 | 2516 static void dct_unquantize_mpeg2_c(MpegEncContext *s, |
2517 DCTELEM *block, int n, int qscale) | |
2518 { | |
2519 int i, level, nCoeffs; | |
2520 const UINT16 *quant_matrix; | |
2521 | |
2522 if(s->alternate_scan) nCoeffs= 64; | |
2523 else nCoeffs= s->block_last_index[n]+1; | |
2524 | |
2525 if (s->mb_intra) { | |
2526 if (n < 4) | |
2527 block[0] = block[0] * s->y_dc_scale; | |
2528 else | |
2529 block[0] = block[0] * s->c_dc_scale; | |
2530 quant_matrix = s->intra_matrix; | |
2531 for(i=1;i<nCoeffs;i++) { | |
2532 int j= zigzag_direct[i]; | |
2533 level = block[j]; | |
2534 if (level) { | |
2535 if (level < 0) { | |
2536 level = -level; | |
2537 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
2538 level = -level; | |
2539 } else { | |
2540 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
2541 } | |
2542 #ifdef PARANOID | |
2543 if (level < -2048 || level > 2047) | |
2544 fprintf(stderr, "unquant error %d %d\n", i, level); | |
2545 #endif | |
2546 block[j] = level; | |
2547 } | |
2548 } | |
2549 } else { | |
2550 int sum=-1; | |
2551 i = 0; | |
344 | 2552 quant_matrix = s->inter_matrix; |
325 | 2553 for(;i<nCoeffs;i++) { |
2554 int j= zigzag_direct[i]; | |
2555 level = block[j]; | |
2556 if (level) { | |
2557 if (level < 0) { | |
2558 level = -level; | |
2559 level = (((level << 1) + 1) * qscale * | |
2560 ((int) (quant_matrix[j]))) >> 4; | |
2561 level = -level; | |
2562 } else { | |
2563 level = (((level << 1) + 1) * qscale * | |
2564 ((int) (quant_matrix[j]))) >> 4; | |
2565 } | |
2566 #ifdef PARANOID | |
2567 if (level < -2048 || level > 2047) | |
2568 fprintf(stderr, "unquant error %d %d\n", i, level); | |
2569 #endif | |
2570 block[j] = level; | |
2571 sum+=level; | |
2572 } | |
2573 } | |
2574 block[63]^=sum&1; | |
2575 } | |
2576 } | |
2577 | |
2578 | |
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
|
2579 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
|
2580 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
|
2581 { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2582 int i, level, qmul, qadd; |
200 | 2583 int nCoeffs; |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2584 |
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
|
2585 if (s->mb_intra) { |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2586 if (!s->h263_aic) { |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2587 if (n < 4) |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2588 block[0] = block[0] * s->y_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2589 else |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2590 block[0] = block[0] * s->c_dc_scale; |
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2591 } |
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
|
2592 i = 1; |
200 | 2593 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
|
2594 } 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
|
2595 i = 0; |
200 | 2596 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
|
2597 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2598 |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2599 qmul = s->qscale << 1; |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2600 if (s->h263_aic && s->mb_intra) |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2601 qadd = 0; |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2602 else |
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2603 qadd = (s->qscale - 1) | 1; |
13
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2604 |
200 | 2605 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
|
2606 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
|
2607 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
|
2608 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
|
2609 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
|
2610 } 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
|
2611 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
|
2612 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2613 #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
|
2614 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
|
2615 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
|
2616 #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
|
2617 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
|
2618 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2619 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2620 } |
0 | 2621 |
456 | 2622 static void remove_ac(MpegEncContext *s, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, int mb_x, int mb_y) |
2623 { | |
2624 int dc, dcb, dcr, y, i; | |
2625 for(i=0; i<4; i++){ | |
2626 dc= s->dc_val[0][mb_x*2+1 + (i&1) + (mb_y*2+1 + (i>>1))*(s->mb_width*2+2)]; | |
2627 for(y=0; y<8; y++){ | |
2628 int x; | |
2629 for(x=0; x<8; x++){ | |
2630 dest_y[x + (i&1)*8 + (y + (i>>1)*8)*s->linesize]= dc/8; | |
2631 } | |
2632 } | |
2633 } | |
2634 dcb = s->dc_val[1][mb_x+1 + (mb_y+1)*(s->mb_width+2)]; | |
2635 dcr= s->dc_val[2][mb_x+1 + (mb_y+1)*(s->mb_width+2)]; | |
2636 for(y=0; y<8; y++){ | |
2637 int x; | |
2638 for(x=0; x<8; x++){ | |
556 | 2639 dest_cb[x + y*(s->uvlinesize)]= dcb/8; |
2640 dest_cr[x + y*(s->uvlinesize)]= dcr/8; | |
456 | 2641 } |
2642 } | |
2643 } | |
2644 | |
2645 /** | |
2646 * will conceal past errors, and allso drop b frames if needed | |
2647 * | |
2648 */ | |
2649 void ff_conceal_past_errors(MpegEncContext *s, int unknown_pos) | |
2650 { | |
2651 int mb_x= s->mb_x; | |
2652 int mb_y= s->mb_y; | |
2653 int mb_dist=0; | |
2654 int i, intra_count=0, inter_count=0; | |
2655 int intra_conceal= s->msmpeg4_version ? 50 : 50; //FIXME finetune | |
2656 int inter_conceal= s->msmpeg4_version ? 50 : 50; | |
2657 | |
2658 // for last block | |
2659 if(mb_x>=s->mb_width) mb_x= s->mb_width -1; | |
2660 if(mb_y>=s->mb_height) mb_y= s->mb_height-1; | |
2661 | |
2662 if(s->decoding_error==0 && unknown_pos){ | |
2663 if(s->data_partitioning && s->pict_type!=B_TYPE) | |
2664 s->decoding_error= DECODING_AC_LOST; | |
2665 else | |
2666 s->decoding_error= DECODING_DESYNC; | |
2667 } | |
2668 | |
2669 if(s->decoding_error==DECODING_DESYNC && s->pict_type!=B_TYPE) s->next_p_frame_damaged=1; | |
2670 | |
2671 for(i=mb_x + mb_y*s->mb_width; i>=0; i--){ | |
2672 if(s->mbintra_table[i]) intra_count++; | |
2673 else inter_count++; | |
2674 } | |
2675 | |
2676 if(s->decoding_error==DECODING_AC_LOST){ | |
2677 intra_conceal*=2; | |
2678 inter_conceal*=2; | |
2679 }else if(s->decoding_error==DECODING_ACDC_LOST){ | |
2680 intra_conceal*=2; | |
2681 inter_conceal*=2; | |
2682 } | |
2683 | |
2684 if(unknown_pos && (intra_count<inter_count)){ | |
2685 intra_conceal= inter_conceal= s->mb_num; | |
2686 // printf("%d %d\n",intra_count, inter_count); | |
2687 } | |
2688 | |
2689 fprintf(stderr, "concealing errors\n"); | |
2690 | |
2691 /* for all MBs from the current one back until the last resync marker */ | |
2692 for(; mb_y>=0 && mb_y>=s->resync_mb_y; mb_y--){ | |
2693 for(; mb_x>=0; mb_x--){ | |
2694 uint8_t *dest_y = s->current_picture[0] + (mb_y * 16* s->linesize ) + mb_x * 16; | |
556 | 2695 uint8_t *dest_cb = s->current_picture[1] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; |
2696 uint8_t *dest_cr = s->current_picture[2] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; | |
456 | 2697 int mb_x_backup= s->mb_x; //FIXME pass xy to mpeg_motion |
2698 int mb_y_backup= s->mb_y; | |
2699 s->mb_x=mb_x; | |
2700 s->mb_y=mb_y; | |
2701 if(s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<intra_conceal){ | |
2702 if(s->decoding_error==DECODING_AC_LOST){ | |
2703 remove_ac(s, dest_y, dest_cb, dest_cr, mb_x, mb_y); | |
2704 // printf("remove ac to %d %d\n", mb_x, mb_y); | |
2705 }else{ | |
2706 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
2707 s->last_picture, 0, 0, put_pixels_tab, | |
2708 0/*mx*/, 0/*my*/, 16); | |
2709 } | |
2710 } | |
2711 else if(!s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<inter_conceal){ | |
2712 int mx=0; | |
2713 int my=0; | |
2714 | |
2715 if(s->decoding_error!=DECODING_DESYNC){ | |
2716 int xy= mb_x*2+1 + (mb_y*2+1)*(s->mb_width*2+2); | |
2717 mx= s->motion_val[ xy ][0]; | |
2718 my= s->motion_val[ xy ][1]; | |
2719 } | |
2720 | |
2721 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
2722 s->last_picture, 0, 0, put_pixels_tab, | |
2723 mx, my, 16); | |
2724 } | |
2725 s->mb_x= mb_x_backup; | |
2726 s->mb_y= mb_y_backup; | |
2727 | |
2728 if(mb_x== s->resync_mb_x && mb_y== s->resync_mb_y) return; | |
2729 if(!s->mbskip_table[mb_x + mb_y*s->mb_width]) mb_dist++; | |
2730 } | |
2731 mb_x=s->mb_width-1; | |
2732 } | |
2733 } | |
2734 | |
0 | 2735 AVCodec mpeg1video_encoder = { |
2736 "mpeg1video", | |
2737 CODEC_TYPE_VIDEO, | |
2738 CODEC_ID_MPEG1VIDEO, | |
2739 sizeof(MpegEncContext), | |
2740 MPV_encode_init, | |
2741 MPV_encode_picture, | |
2742 MPV_encode_end, | |
2743 }; | |
2744 | |
2745 AVCodec h263_encoder = { | |
2746 "h263", | |
2747 CODEC_TYPE_VIDEO, | |
2748 CODEC_ID_H263, | |
2749 sizeof(MpegEncContext), | |
2750 MPV_encode_init, | |
2751 MPV_encode_picture, | |
2752 MPV_encode_end, | |
2753 }; | |
2754 | |
2755 AVCodec h263p_encoder = { | |
2756 "h263p", | |
2757 CODEC_TYPE_VIDEO, | |
2758 CODEC_ID_H263P, | |
2759 sizeof(MpegEncContext), | |
2760 MPV_encode_init, | |
2761 MPV_encode_picture, | |
2762 MPV_encode_end, | |
2763 }; | |
2764 | |
2765 AVCodec rv10_encoder = { | |
2766 "rv10", | |
2767 CODEC_TYPE_VIDEO, | |
2768 CODEC_ID_RV10, | |
2769 sizeof(MpegEncContext), | |
2770 MPV_encode_init, | |
2771 MPV_encode_picture, | |
2772 MPV_encode_end, | |
2773 }; | |
2774 | |
2775 AVCodec mjpeg_encoder = { | |
2776 "mjpeg", | |
2777 CODEC_TYPE_VIDEO, | |
2778 CODEC_ID_MJPEG, | |
2779 sizeof(MpegEncContext), | |
2780 MPV_encode_init, | |
2781 MPV_encode_picture, | |
2782 MPV_encode_end, | |
2783 }; | |
2784 | |
71 | 2785 AVCodec mpeg4_encoder = { |
2786 "mpeg4", | |
0 | 2787 CODEC_TYPE_VIDEO, |
71 | 2788 CODEC_ID_MPEG4, |
0 | 2789 sizeof(MpegEncContext), |
2790 MPV_encode_init, | |
2791 MPV_encode_picture, | |
2792 MPV_encode_end, | |
2793 }; | |
2794 | |
307 | 2795 AVCodec msmpeg4v1_encoder = { |
2796 "msmpeg4v1", | |
0 | 2797 CODEC_TYPE_VIDEO, |
307 | 2798 CODEC_ID_MSMPEG4V1, |
0 | 2799 sizeof(MpegEncContext), |
2800 MPV_encode_init, | |
2801 MPV_encode_picture, | |
2802 MPV_encode_end, | |
2803 }; | |
307 | 2804 |
2805 AVCodec msmpeg4v2_encoder = { | |
2806 "msmpeg4v2", | |
2807 CODEC_TYPE_VIDEO, | |
2808 CODEC_ID_MSMPEG4V2, | |
2809 sizeof(MpegEncContext), | |
2810 MPV_encode_init, | |
2811 MPV_encode_picture, | |
2812 MPV_encode_end, | |
2813 }; | |
2814 | |
2815 AVCodec msmpeg4v3_encoder = { | |
2816 "msmpeg4", | |
2817 CODEC_TYPE_VIDEO, | |
2818 CODEC_ID_MSMPEG4V3, | |
2819 sizeof(MpegEncContext), | |
2820 MPV_encode_init, | |
2821 MPV_encode_picture, | |
2822 MPV_encode_end, | |
2823 }; | |
499 | 2824 |
2825 AVCodec wmv1_encoder = { | |
2826 "wmv1", | |
2827 CODEC_TYPE_VIDEO, | |
2828 CODEC_ID_WMV1, | |
2829 sizeof(MpegEncContext), | |
2830 MPV_encode_init, | |
2831 MPV_encode_picture, | |
2832 MPV_encode_end, | |
2833 }; | |
2834 | |
2835 AVCodec wmv2_encoder = { | |
2836 "wmv2", | |
2837 CODEC_TYPE_VIDEO, | |
2838 CODEC_ID_WMV2, | |
2839 sizeof(MpegEncContext), | |
2840 MPV_encode_init, | |
2841 MPV_encode_picture, | |
2842 MPV_encode_end, | |
2843 }; |