Mercurial > libavcodec.hg
annotate mpegvideo.c @ 671:3f2e5a282f61 libavcodec
interlaced qpel mpeg4 fixed
| author | michaelni |
|---|---|
| date | Sat, 14 Sep 2002 10:11:05 +0000 |
| parents | 00a882f626bd |
| children | 242f0f7d8c15 |
| 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 |
| 39 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; | |
| 566 | 40 static void emulated_edge_mc(MpegEncContext *s, UINT8 *src, int linesize, int block_w, int block_h, |
| 41 int src_x, int src_y, int w, int h); | |
| 206 | 42 |
| 0 | 43 #define EDGE_WIDTH 16 |
| 44 | |
| 45 /* enable all paranoid tests for rounding, overflows, etc... */ | |
| 46 //#define PARANOID | |
| 47 | |
| 48 //#define DEBUG | |
| 49 | |
| 321 | 50 |
| 0 | 51 /* for jpeg fast DCT */ |
| 52 #define CONST_BITS 14 | |
| 53 | |
| 54 static const unsigned short aanscales[64] = { | |
| 55 /* precomputed values scaled up by 14 bits */ | |
| 56 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
| 57 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
| 58 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
| 59 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
| 60 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
| 61 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
| 62 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
| 63 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
| 64 }; | |
| 65 | |
| 66 static UINT8 h263_chroma_roundtab[16] = { | |
| 67 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, | |
| 68 }; | |
| 69 | |
|
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
70 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
|
71 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
|
72 |
| 321 | 73 extern UINT8 zigzag_end[64]; |
| 74 | |
| 0 | 75 /* default motion estimation */ |
| 321 | 76 int motion_estimation_method = ME_EPZS; |
| 200 | 77 |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
78 static void convert_matrix(MpegEncContext *s, int (*qmat)[64], uint16_t (*qmat16)[64], uint16_t (*qmat16_bias)[64], |
| 344 | 79 const UINT16 *quant_matrix, int bias) |
| 0 | 80 { |
| 344 | 81 int qscale; |
| 0 | 82 |
| 344 | 83 for(qscale=1; qscale<32; qscale++){ |
| 84 int i; | |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
85 if (s->fdct == ff_jpeg_fdct_islow) { |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
86 for(i=0;i<64;i++) { |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
87 const int j= block_permute_op(i); |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
88 /* 16 <= qscale * quant_matrix[i] <= 7905 */ |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
89 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
90 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
91 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
92 |
| 635 | 93 qmat[qscale][j] = (int)((UINT64_C(1) << QMAT_SHIFT) / |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
94 (qscale * quant_matrix[j])); |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
95 } |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
96 } else if (s->fdct == fdct_ifast) { |
| 344 | 97 for(i=0;i<64;i++) { |
| 98 const int j= block_permute_op(i); | |
| 99 /* 16 <= qscale * quant_matrix[i] <= 7905 */ | |
| 100 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ | |
| 101 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
| 102 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
| 103 | |
| 635 | 104 qmat[qscale][j] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) / |
| 344 | 105 (aanscales[i] * qscale * quant_matrix[j])); |
| 106 } | |
| 107 } else { | |
| 108 for(i=0;i<64;i++) { | |
| 109 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 | |
| 110 So 16 <= qscale * quant_matrix[i] <= 7905 | |
| 111 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
| 112 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
| 113 */ | |
| 114 qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); | |
| 115 qmat16[qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); | |
| 116 | |
| 117 if(qmat16[qscale][i]==0 || qmat16[qscale][i]==128*256) qmat16[qscale][i]=128*256-1; | |
| 118 qmat16_bias[qscale][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][i]); | |
| 119 } | |
| 0 | 120 } |
| 121 } | |
| 122 } | |
| 456 | 123 // move into common.c perhaps |
| 124 #define CHECKED_ALLOCZ(p, size)\ | |
| 125 {\ | |
| 126 p= av_mallocz(size);\ | |
| 127 if(p==NULL){\ | |
| 128 perror("malloc");\ | |
| 129 goto fail;\ | |
| 130 }\ | |
| 131 } | |
| 0 | 132 |
| 133 /* init common structure for both encoder and decoder */ | |
| 134 int MPV_common_init(MpegEncContext *s) | |
| 135 { | |
| 136 int c_size, i; | |
| 137 UINT8 *pict; | |
| 138 | |
| 312 | 139 s->dct_unquantize_h263 = dct_unquantize_h263_c; |
| 325 | 140 s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_c; |
| 141 s->dct_unquantize_mpeg2 = dct_unquantize_mpeg2_c; | |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
142 s->dct_quantize= dct_quantize_c; |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
143 |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
144 if(s->avctx->dct_algo==FF_DCT_FASTINT) |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
145 s->fdct = fdct_ifast; |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
146 else |
|
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
147 s->fdct = ff_jpeg_fdct_islow; |
|
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
|
148 |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
149 #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
|
150 MPV_common_init_mmx(s); |
| 8 | 151 #endif |
|
514
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M?ns Rullg?rd
mellum
parents:
506
diff
changeset
|
152 #ifdef ARCH_ALPHA |
|
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M?ns Rullg?rd
mellum
parents:
506
diff
changeset
|
153 MPV_common_init_axp(s); |
|
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M?ns Rullg?rd
mellum
parents:
506
diff
changeset
|
154 #endif |
|
628
f596db4aa871
sun solaris compilation bugfix, patch by (Martin Olschewski <olschewski at zpr dot uni-koeln dot de>)
michaelni
parents:
625
diff
changeset
|
155 #ifdef HAVE_MLIB |
|
f596db4aa871
sun solaris compilation bugfix, patch by (Martin Olschewski <olschewski at zpr dot uni-koeln dot de>)
michaelni
parents:
625
diff
changeset
|
156 MPV_common_init_mlib(s); |
|
f596db4aa871
sun solaris compilation bugfix, patch by (Martin Olschewski <olschewski at zpr dot uni-koeln dot de>)
michaelni
parents:
625
diff
changeset
|
157 #endif |
| 591 | 158 |
| 0 | 159 s->mb_width = (s->width + 15) / 16; |
| 160 s->mb_height = (s->height + 15) / 16; | |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
161 |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
162 /* 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
|
163 s->h_edge_pos= s->mb_width*16; |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
164 s->v_edge_pos= s->mb_height*16; |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
165 |
|
233
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
166 s->mb_num = s->mb_width * s->mb_height; |
| 553 | 167 if(!(s->flags&CODEC_FLAG_DR1)){ |
| 556 | 168 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH; |
| 169 s->uvlinesize = s->mb_width * 8 + EDGE_WIDTH; | |
| 0 | 170 |
| 553 | 171 for(i=0;i<3;i++) { |
| 0 | 172 int w, h, shift, pict_start; |
| 173 | |
| 174 w = s->linesize; | |
| 175 h = s->mb_height * 16 + 2 * EDGE_WIDTH; | |
| 176 shift = (i == 0) ? 0 : 1; | |
| 556 | 177 c_size = (s->linesize>>shift) * (h >> shift); |
| 178 pict_start = (s->linesize>>shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift); | |
| 0 | 179 |
| 456 | 180 CHECKED_ALLOCZ(pict, c_size) |
| 0 | 181 s->last_picture_base[i] = pict; |
| 182 s->last_picture[i] = pict + pict_start; | |
| 485 | 183 if(i>0) memset(s->last_picture_base[i], 128, c_size); |
| 0 | 184 |
| 456 | 185 CHECKED_ALLOCZ(pict, c_size) |
| 0 | 186 s->next_picture_base[i] = pict; |
| 187 s->next_picture[i] = pict + pict_start; | |
| 485 | 188 if(i>0) memset(s->next_picture_base[i], 128, c_size); |
| 324 | 189 |
|
365
fdeec2834c79
there are divx5? encoded files without a userdata section but with b-frames :(
michaelni
parents:
350
diff
changeset
|
190 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
|
191 /* 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
|
192 do low-delay encoding, so we cant allways distinguish b-frame containing streams from low_delay streams */ |
| 456 | 193 CHECKED_ALLOCZ(pict, c_size) |
| 0 | 194 s->aux_picture_base[i] = pict; |
| 195 s->aux_picture[i] = pict + pict_start; | |
| 485 | 196 if(i>0) memset(s->aux_picture_base[i], 128, c_size); |
| 0 | 197 } |
| 553 | 198 } |
| 556 | 199 s->ip_buffer_count= 2; |
| 0 | 200 } |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
201 |
|
567
5a6eaaa9fba2
larger edge emu buffer (might have been too small)
michaelni
parents:
566
diff
changeset
|
202 CHECKED_ALLOCZ(s->edge_emu_buffer, (s->width+64)*2*17*2); //(width + edge + align)*interlaced*MBsize*tolerance |
| 553 | 203 |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
204 if (s->encoding) { |
| 324 | 205 int j; |
| 206 int mv_table_size= (s->mb_width+2)*(s->mb_height+2); | |
| 239 | 207 |
| 456 | 208 CHECKED_ALLOCZ(s->mb_var , s->mb_num * sizeof(INT16)) |
| 209 CHECKED_ALLOCZ(s->mc_mb_var, s->mb_num * sizeof(INT16)) | |
| 324 | 210 |
| 211 /* Allocate MV tables */ | |
| 456 | 212 CHECKED_ALLOCZ(s->p_mv_table , mv_table_size * 2 * sizeof(INT16)) |
| 213 CHECKED_ALLOCZ(s->b_forw_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
| 214 CHECKED_ALLOCZ(s->b_back_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
| 215 CHECKED_ALLOCZ(s->b_bidir_forw_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
| 216 CHECKED_ALLOCZ(s->b_bidir_back_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
| 217 CHECKED_ALLOCZ(s->b_direct_forw_mv_table, mv_table_size * 2 * sizeof(INT16)) | |
| 218 CHECKED_ALLOCZ(s->b_direct_back_mv_table, mv_table_size * 2 * sizeof(INT16)) | |
| 219 CHECKED_ALLOCZ(s->b_direct_mv_table , mv_table_size * 2 * sizeof(INT16)) | |
| 324 | 220 |
| 456 | 221 CHECKED_ALLOCZ(s->me_scratchpad, s->linesize*16*3*sizeof(uint8_t)) |
| 222 | |
| 223 CHECKED_ALLOCZ(s->me_map , ME_MAP_SIZE*sizeof(uint32_t)) | |
| 224 CHECKED_ALLOCZ(s->me_score_map, ME_MAP_SIZE*sizeof(uint16_t)) | |
| 327 | 225 |
| 324 | 226 if(s->max_b_frames){ |
| 227 for(j=0; j<REORDER_BUFFER_SIZE; j++){ | |
| 228 int i; | |
| 229 for(i=0;i<3;i++) { | |
| 230 int w, h, shift; | |
| 231 | |
| 232 w = s->linesize; | |
| 233 h = s->mb_height * 16; | |
| 234 shift = (i == 0) ? 0 : 1; | |
| 235 c_size = (w >> shift) * (h >> shift); | |
| 236 | |
| 456 | 237 CHECKED_ALLOCZ(pict, c_size); |
| 324 | 238 s->picture_buffer[j][i] = pict; |
| 239 } | |
| 240 } | |
| 241 } | |
| 456 | 242 |
| 243 if(s->codec_id==CODEC_ID_MPEG4){ | |
| 244 CHECKED_ALLOCZ(s->tex_pb_buffer, PB_BUFFER_SIZE); | |
| 245 CHECKED_ALLOCZ( s->pb2_buffer, PB_BUFFER_SIZE); | |
| 246 } | |
| 612 | 247 |
|
650
ef4a33aad86e
reducing sizeof MpegEncContext to avoid stack overflow on crap M$ windo$
michaelni
parents:
635
diff
changeset
|
248 if(s->msmpeg4_version){ |
|
ef4a33aad86e
reducing sizeof MpegEncContext to avoid stack overflow on crap M$ windo$
michaelni
parents:
635
diff
changeset
|
249 CHECKED_ALLOCZ(s->ac_stats, 2*2*(MAX_LEVEL+1)*(MAX_RUN+1)*2*sizeof(int)); |
|
ef4a33aad86e
reducing sizeof MpegEncContext to avoid stack overflow on crap M$ windo$
michaelni
parents:
635
diff
changeset
|
250 } |
| 612 | 251 CHECKED_ALLOCZ(s->avctx->stats_out, 256); |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
252 } |
|
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
253 |
| 288 | 254 if (s->out_format == FMT_H263 || s->encoding) { |
| 0 | 255 int size; |
| 456 | 256 /* Allocate MB type table */ |
| 257 CHECKED_ALLOCZ(s->mb_type , s->mb_num * sizeof(UINT8)) | |
| 258 | |
| 0 | 259 /* MV prediction */ |
| 260 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
| 456 | 261 CHECKED_ALLOCZ(s->motion_val, size * 2 * sizeof(INT16)); |
| 664 | 262 } |
| 263 | |
| 264 if(s->codec_id==CODEC_ID_MPEG4){ | |
| 265 /* 4mv and interlaced direct mode decoding tables */ | |
| 266 CHECKED_ALLOCZ(s->co_located_type_table, s->mb_num * sizeof(UINT8)) | |
| 267 CHECKED_ALLOCZ(s->field_mv_table, s->mb_num*2*2 * sizeof(INT16)) | |
| 268 CHECKED_ALLOCZ(s->field_select_table, s->mb_num*2* sizeof(INT8)) | |
| 0 | 269 } |
| 270 | |
|
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
271 if (s->h263_pred || s->h263_plus) { |
| 0 | 272 int y_size, c_size, i, size; |
| 273 | |
| 274 /* dc values */ | |
| 275 | |
| 276 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2); | |
| 277 c_size = (s->mb_width + 2) * (s->mb_height + 2); | |
| 278 size = y_size + 2 * c_size; | |
| 456 | 279 CHECKED_ALLOCZ(s->dc_val[0], size * sizeof(INT16)); |
| 0 | 280 s->dc_val[1] = s->dc_val[0] + y_size; |
| 281 s->dc_val[2] = s->dc_val[1] + c_size; | |
| 282 for(i=0;i<size;i++) | |
| 283 s->dc_val[0][i] = 1024; | |
| 284 | |
| 285 /* ac values */ | |
| 456 | 286 CHECKED_ALLOCZ(s->ac_val[0], size * sizeof(INT16) * 16); |
| 0 | 287 s->ac_val[1] = s->ac_val[0] + y_size; |
| 288 s->ac_val[2] = s->ac_val[1] + c_size; | |
| 289 | |
| 290 /* cbp values */ | |
| 456 | 291 CHECKED_ALLOCZ(s->coded_block, y_size); |
| 333 | 292 |
| 293 /* divx501 bitstream reorder buffer */ | |
| 456 | 294 CHECKED_ALLOCZ(s->bitstream_buffer, BITSTREAM_BUFFER_SIZE); |
| 295 | |
| 296 /* cbp, ac_pred, pred_dir */ | |
| 297 CHECKED_ALLOCZ(s->cbp_table , s->mb_num * sizeof(UINT8)) | |
| 298 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
|
299 } |
| 575 | 300 CHECKED_ALLOCZ(s->qscale_table , s->mb_num * sizeof(UINT8)) |
| 301 | |
|
611
3214d3f4519e
error concealment needs the mbintra_table so it should allways be allocated
michaelni
parents:
608
diff
changeset
|
302 /* which mb is a intra block */ |
|
3214d3f4519e
error concealment needs the mbintra_table so it should allways be allocated
michaelni
parents:
608
diff
changeset
|
303 CHECKED_ALLOCZ(s->mbintra_table, s->mb_num); |
|
3214d3f4519e
error concealment needs the mbintra_table so it should allways be allocated
michaelni
parents:
608
diff
changeset
|
304 memset(s->mbintra_table, 1, s->mb_num); |
|
3214d3f4519e
error concealment needs the mbintra_table so it should allways be allocated
michaelni
parents:
608
diff
changeset
|
305 |
| 0 | 306 /* default structure is frame */ |
| 307 s->picture_structure = PICT_FRAME; | |
| 553 | 308 |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
309 /* init macroblock skip table */ |
| 456 | 310 CHECKED_ALLOCZ(s->mbskip_table, s->mb_num); |
| 294 | 311 |
| 327 | 312 s->block= s->blocks[0]; |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
313 |
| 0 | 314 s->context_initialized = 1; |
| 315 return 0; | |
| 316 fail: | |
| 244 | 317 MPV_common_end(s); |
| 318 return -1; | |
| 319 } | |
| 320 | |
| 456 | 321 |
| 322 //extern int sads; | |
| 323 | |
| 244 | 324 /* init common structure for both encoder and decoder */ |
| 325 void MPV_common_end(MpegEncContext *s) | |
| 326 { | |
| 327 int i; | |
| 328 | |
|
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
329 av_freep(&s->mb_type); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
330 av_freep(&s->mb_var); |
| 456 | 331 av_freep(&s->mc_mb_var); |
|
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
332 av_freep(&s->p_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
333 av_freep(&s->b_forw_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
334 av_freep(&s->b_back_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
335 av_freep(&s->b_bidir_forw_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
336 av_freep(&s->b_bidir_back_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
337 av_freep(&s->b_direct_forw_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
338 av_freep(&s->b_direct_back_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
339 av_freep(&s->b_direct_mv_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
340 av_freep(&s->motion_val); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
341 av_freep(&s->dc_val[0]); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
342 av_freep(&s->ac_val[0]); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
343 av_freep(&s->coded_block); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
344 av_freep(&s->mbintra_table); |
| 456 | 345 av_freep(&s->cbp_table); |
| 346 av_freep(&s->pred_dir_table); | |
| 347 av_freep(&s->qscale_table); | |
|
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
348 av_freep(&s->me_scratchpad); |
| 456 | 349 av_freep(&s->me_map); |
| 350 av_freep(&s->me_score_map); | |
| 351 | |
|
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
352 av_freep(&s->mbskip_table); |
|
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
353 av_freep(&s->bitstream_buffer); |
| 456 | 354 av_freep(&s->tex_pb_buffer); |
| 355 av_freep(&s->pb2_buffer); | |
| 553 | 356 av_freep(&s->edge_emu_buffer); |
| 664 | 357 av_freep(&s->co_located_type_table); |
| 358 av_freep(&s->field_mv_table); | |
| 359 av_freep(&s->field_select_table); | |
| 612 | 360 av_freep(&s->avctx->stats_out); |
|
650
ef4a33aad86e
reducing sizeof MpegEncContext to avoid stack overflow on crap M$ windo$
michaelni
parents:
635
diff
changeset
|
361 av_freep(&s->ac_stats); |
| 612 | 362 |
| 0 | 363 for(i=0;i<3;i++) { |
| 324 | 364 int j; |
| 553 | 365 if(!(s->flags&CODEC_FLAG_DR1)){ |
| 366 av_freep(&s->last_picture_base[i]); | |
| 367 av_freep(&s->next_picture_base[i]); | |
| 368 av_freep(&s->aux_picture_base[i]); | |
| 369 } | |
| 370 s->last_picture_base[i]= | |
| 371 s->next_picture_base[i]= | |
| 372 s->aux_picture_base [i] = NULL; | |
| 373 s->last_picture[i]= | |
| 374 s->next_picture[i]= | |
| 375 s->aux_picture [i] = NULL; | |
| 376 | |
| 324 | 377 for(j=0; j<REORDER_BUFFER_SIZE; j++){ |
|
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
378 av_freep(&s->picture_buffer[j][i]); |
| 324 | 379 } |
| 0 | 380 } |
| 381 s->context_initialized = 0; | |
| 382 } | |
| 383 | |
| 384 /* init video encoder */ | |
| 385 int MPV_encode_init(AVCodecContext *avctx) | |
| 386 { | |
| 387 MpegEncContext *s = avctx->priv_data; | |
| 60 | 388 int i; |
| 0 | 389 |
| 315 | 390 avctx->pix_fmt = PIX_FMT_YUV420P; |
| 391 | |
| 0 | 392 s->bit_rate = avctx->bit_rate; |
| 268 | 393 s->bit_rate_tolerance = avctx->bit_rate_tolerance; |
| 0 | 394 s->frame_rate = avctx->frame_rate; |
| 395 s->width = avctx->width; | |
| 396 s->height = avctx->height; | |
| 456 | 397 if(avctx->gop_size > 600){ |
|
462
3fdd0627f6bd
typo (found by Bohdan Horst <nexus at hoth.amu.edu.pl>)
michaelni
parents:
461
diff
changeset
|
398 fprintf(stderr, "Warning keyframe interval too large! reducing it ...\n"); |
| 456 | 399 avctx->gop_size=600; |
| 400 } | |
| 0 | 401 s->gop_size = avctx->gop_size; |
| 162 | 402 s->rtp_mode = avctx->rtp_mode; |
| 403 s->rtp_payload_size = avctx->rtp_payload_size; | |
| 231 | 404 if (avctx->rtp_callback) |
| 405 s->rtp_callback = avctx->rtp_callback; | |
| 268 | 406 s->qmin= avctx->qmin; |
| 407 s->qmax= avctx->qmax; | |
| 408 s->max_qdiff= avctx->max_qdiff; | |
| 409 s->qcompress= avctx->qcompress; | |
| 410 s->qblur= avctx->qblur; | |
|
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
411 s->avctx = avctx; |
|
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
412 s->aspect_ratio_info= avctx->aspect_ratio_info; |
| 618 | 413 if (avctx->aspect_ratio_info == FF_ASPECT_EXTENDED) |
| 414 { | |
| 415 s->aspected_width = avctx->aspected_width; | |
| 416 s->aspected_height = avctx->aspected_height; | |
| 417 } | |
| 294 | 418 s->flags= avctx->flags; |
| 324 | 419 s->max_b_frames= avctx->max_b_frames; |
| 329 | 420 s->b_frame_strategy= avctx->b_frame_strategy; |
| 344 | 421 s->codec_id= avctx->codec->id; |
| 456 | 422 s->luma_elim_threshold = avctx->luma_elim_threshold; |
| 423 s->chroma_elim_threshold= avctx->chroma_elim_threshold; | |
| 424 s->strict_std_compliance= avctx->strict_std_compliance; | |
| 425 s->data_partitioning= avctx->flags & CODEC_FLAG_PART; | |
| 599 | 426 s->mpeg_quant= avctx->mpeg_quant; |
| 344 | 427 |
| 0 | 428 if (s->gop_size <= 1) { |
| 429 s->intra_only = 1; | |
| 430 s->gop_size = 12; | |
| 431 } else { | |
| 432 s->intra_only = 0; | |
| 433 } | |
|
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
434 |
|
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
435 /* ME algorithm */ |
| 321 | 436 if (avctx->me_method == 0) |
| 437 /* For compatibility */ | |
| 438 s->me_method = motion_estimation_method; | |
| 439 else | |
| 440 s->me_method = avctx->me_method; | |
| 441 | |
|
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
442 /* Fixed QSCALE */ |
| 0 | 443 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
|
444 |
| 0 | 445 switch(avctx->codec->id) { |
| 446 case CODEC_ID_MPEG1VIDEO: | |
| 447 s->out_format = FMT_MPEG1; | |
| 336 | 448 avctx->delay=0; //FIXME not sure, should check the spec |
| 0 | 449 break; |
| 450 case CODEC_ID_MJPEG: | |
| 451 s->out_format = FMT_MJPEG; | |
| 452 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
|
453 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
|
454 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
|
455 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
|
456 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
|
457 s->mjpeg_vsample[2] = 1; |
|
370
0eca28d16cbd
clamp intra matrix to 8bit for mjpeg (workaround for qscale>=25)
al3x
parents:
365
diff
changeset
|
458 s->mjpeg_hsample[0] = 2; |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
459 s->mjpeg_hsample[1] = 1; |
|
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
460 s->mjpeg_hsample[2] = 1; |
| 0 | 461 if (mjpeg_init(s) < 0) |
| 462 return -1; | |
| 336 | 463 avctx->delay=0; |
| 0 | 464 break; |
| 465 case CODEC_ID_H263: | |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
466 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
|
467 printf("Input picture size isn't suitable for h263 codec! try h263+\n"); |
| 0 | 468 return -1; |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
469 } |
| 0 | 470 s->out_format = FMT_H263; |
| 336 | 471 avctx->delay=0; |
| 0 | 472 break; |
| 473 case CODEC_ID_H263P: | |
| 474 s->out_format = FMT_H263; | |
| 162 | 475 s->rtp_mode = 1; |
| 476 s->rtp_payload_size = 1200; | |
| 0 | 477 s->h263_plus = 1; |
| 78 | 478 s->unrestricted_mv = 1; |
|
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
479 s->h263_aic = 1; |
|
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
480 |
|
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
481 /* These are just to be sure */ |
|
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
482 s->umvplus = 0; |
|
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
483 s->umvplus_dec = 0; |
| 336 | 484 avctx->delay=0; |
| 0 | 485 break; |
| 486 case CODEC_ID_RV10: | |
| 487 s->out_format = FMT_H263; | |
| 488 s->h263_rv10 = 1; | |
| 336 | 489 avctx->delay=0; |
| 0 | 490 break; |
| 71 | 491 case CODEC_ID_MPEG4: |
| 0 | 492 s->out_format = FMT_H263; |
| 493 s->h263_pred = 1; | |
| 494 s->unrestricted_mv = 1; | |
| 324 | 495 s->has_b_frames= s->max_b_frames ? 1 : 0; |
| 336 | 496 s->low_delay=0; |
| 497 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); | |
| 0 | 498 break; |
| 307 | 499 case CODEC_ID_MSMPEG4V1: |
| 0 | 500 s->out_format = FMT_H263; |
| 501 s->h263_msmpeg4 = 1; | |
| 502 s->h263_pred = 1; | |
| 503 s->unrestricted_mv = 1; | |
| 307 | 504 s->msmpeg4_version= 1; |
| 336 | 505 avctx->delay=0; |
| 307 | 506 break; |
| 507 case CODEC_ID_MSMPEG4V2: | |
| 508 s->out_format = FMT_H263; | |
| 509 s->h263_msmpeg4 = 1; | |
| 510 s->h263_pred = 1; | |
| 511 s->unrestricted_mv = 1; | |
| 512 s->msmpeg4_version= 2; | |
| 336 | 513 avctx->delay=0; |
| 307 | 514 break; |
| 515 case CODEC_ID_MSMPEG4V3: | |
| 516 s->out_format = FMT_H263; | |
| 517 s->h263_msmpeg4 = 1; | |
| 518 s->h263_pred = 1; | |
| 519 s->unrestricted_mv = 1; | |
| 520 s->msmpeg4_version= 3; | |
| 336 | 521 avctx->delay=0; |
| 0 | 522 break; |
| 499 | 523 case CODEC_ID_WMV1: |
| 524 s->out_format = FMT_H263; | |
| 525 s->h263_msmpeg4 = 1; | |
| 526 s->h263_pred = 1; | |
| 527 s->unrestricted_mv = 1; | |
| 528 s->msmpeg4_version= 4; | |
| 529 avctx->delay=0; | |
| 530 break; | |
| 531 case CODEC_ID_WMV2: | |
| 532 s->out_format = FMT_H263; | |
| 533 s->h263_msmpeg4 = 1; | |
| 534 s->h263_pred = 1; | |
| 535 s->unrestricted_mv = 1; | |
| 536 s->msmpeg4_version= 5; | |
| 537 avctx->delay=0; | |
| 538 break; | |
| 0 | 539 default: |
| 540 return -1; | |
| 541 } | |
| 295 | 542 |
|
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
543 { /* 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
|
544 static int done=0; |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
545 if(!done){ |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
546 int i; |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
547 done=1; |
|
292
73a9ce3d9715
fcode_tables where too small, found by Klaas-Pieter Vlieg <vlieg@eurescom.de>
michaelni
parents:
288
diff
changeset
|
548 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
|
549 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
|
550 |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
551 for(i=-16; i<16; i++){ |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
552 default_fcode_tab[i + MAX_MV]= 1; |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
553 } |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
554 } |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
555 } |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
556 s->mv_penalty= default_mv_penalty; |
|
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
557 s->fcode_tab= default_fcode_tab; |
|
506
b9ed0ae72d51
init dc_scale tables to defaults (fixes mjpeg sig11)
michaelni
parents:
499
diff
changeset
|
558 s->y_dc_scale_table= |
|
b9ed0ae72d51
init dc_scale tables to defaults (fixes mjpeg sig11)
michaelni
parents:
499
diff
changeset
|
559 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
|
560 |
| 0 | 561 if (s->out_format == FMT_H263) |
|
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
562 h263_encode_init(s); |
|
281
1fc96b02142e
mpeg4 aspect_ratio_info in AVCodecContext (requested by alex)
michaelni
parents:
280
diff
changeset
|
563 else if (s->out_format == FMT_MPEG1) |
| 499 | 564 ff_mpeg1_encode_init(s); |
| 565 if(s->msmpeg4_version) | |
| 566 ff_msmpeg4_encode_init(s); | |
| 0 | 567 |
| 287 | 568 /* dont use mv_penalty table for crap MV as it would be confused */ |
| 324 | 569 if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty; |
| 287 | 570 |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
571 s->encoding = 1; |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
572 |
| 0 | 573 /* init */ |
| 574 if (MPV_common_init(s) < 0) | |
| 575 return -1; | |
| 576 | |
| 60 | 577 /* init default q matrix */ |
| 578 for(i=0;i<64;i++) { | |
| 599 | 579 if(s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){ |
| 580 s->intra_matrix[i] = ff_mpeg4_default_intra_matrix[i]; | |
| 581 s->inter_matrix[i] = ff_mpeg4_default_non_intra_matrix[i]; | |
| 582 }else if(s->out_format == FMT_H263){ | |
| 583 s->intra_matrix[i] = | |
| 584 s->inter_matrix[i] = ff_mpeg1_default_non_intra_matrix[i]; | |
| 585 }else{ /* mpeg1 */ | |
|
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
|
586 s->intra_matrix[i] = ff_mpeg1_default_intra_matrix[i]; |
| 599 | 587 s->inter_matrix[i] = ff_mpeg1_default_non_intra_matrix[i]; |
| 588 } | |
| 344 | 589 } |
| 590 | |
| 591 /* precompute matrix */ | |
|
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
592 /* for mjpeg, we do include qscale in the matrix */ |
| 344 | 593 if (s->out_format != FMT_MJPEG) { |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
594 convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias, |
| 344 | 595 s->intra_matrix, s->intra_quant_bias); |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
596 convert_matrix(s, s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias, |
| 344 | 597 s->inter_matrix, s->inter_quant_bias); |
| 60 | 598 } |
| 599 | |
| 329 | 600 if(ff_rate_control_init(s) < 0) |
| 601 return -1; | |
| 0 | 602 |
| 603 s->picture_number = 0; | |
|
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
604 s->picture_in_gop_number = 0; |
| 0 | 605 s->fake_picture_number = 0; |
| 606 /* motion detector init */ | |
| 607 s->f_code = 1; | |
| 324 | 608 s->b_code = 1; |
| 0 | 609 |
| 610 return 0; | |
| 611 } | |
| 612 | |
| 613 int MPV_encode_end(AVCodecContext *avctx) | |
| 614 { | |
| 615 MpegEncContext *s = avctx->priv_data; | |
| 616 | |
| 617 #ifdef STATS | |
| 618 print_stats(); | |
| 619 #endif | |
| 329 | 620 |
| 621 ff_rate_control_uninit(s); | |
| 622 | |
| 0 | 623 MPV_common_end(s); |
| 624 if (s->out_format == FMT_MJPEG) | |
| 625 mjpeg_close(s); | |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
626 |
| 0 | 627 return 0; |
| 628 } | |
| 629 | |
| 630 /* draw the edges of width 'w' of an image of size width, height */ | |
|
619
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
631 //FIXME check that this is ok for mpeg4 interlaced |
| 206 | 632 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w) |
| 0 | 633 { |
| 634 UINT8 *ptr, *last_line; | |
| 635 int i; | |
| 636 | |
| 637 last_line = buf + (height - 1) * wrap; | |
| 638 for(i=0;i<w;i++) { | |
| 639 /* top and bottom */ | |
| 640 memcpy(buf - (i + 1) * wrap, buf, width); | |
| 641 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
| 642 } | |
| 643 /* left and right */ | |
| 644 ptr = buf; | |
| 645 for(i=0;i<height;i++) { | |
| 646 memset(ptr - w, ptr[0], w); | |
| 647 memset(ptr + width, ptr[width-1], w); | |
| 648 ptr += wrap; | |
| 649 } | |
| 650 /* corners */ | |
| 651 for(i=0;i<w;i++) { | |
| 652 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
| 653 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
| 654 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
| 655 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
| 656 } | |
| 657 } | |
| 658 | |
| 659 /* generic function for encode/decode called before a frame is coded/decoded */ | |
| 553 | 660 void MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx) |
| 0 | 661 { |
| 662 int i; | |
| 663 UINT8 *tmp; | |
| 664 | |
|
46
931417475f5b
fixed mpeg1 first block bug (pb with black picture optimisation for B frames)
glantau
parents:
40
diff
changeset
|
665 s->mb_skiped = 0; |
| 456 | 666 s->decoding_error=0; |
|
562
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
667 avctx->mbskip_table= s->mbskip_table; |
| 456 | 668 |
| 553 | 669 if(avctx->flags&CODEC_FLAG_DR1){ |
| 670 avctx->get_buffer_callback(avctx, s->width, s->height, s->pict_type); | |
| 671 | |
| 556 | 672 s->linesize = avctx->dr_stride; |
| 673 s->uvlinesize= avctx->dr_uvstride; | |
| 674 s->ip_buffer_count= avctx->dr_ip_buffer_count; | |
| 553 | 675 } |
|
562
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
676 avctx->dr_ip_buffer_count= s->ip_buffer_count; |
| 553 | 677 |
| 0 | 678 if (s->pict_type == B_TYPE) { |
| 679 for(i=0;i<3;i++) { | |
| 553 | 680 if(avctx->flags&CODEC_FLAG_DR1) |
| 681 s->aux_picture[i]= avctx->dr_buffer[i]; | |
| 569 | 682 |
| 683 //FIXME the following should never be needed, the decoder should drop b frames if no reference is available | |
| 684 if(s->next_picture[i]==NULL) | |
| 685 s->next_picture[i]= s->aux_picture[i]; | |
| 686 if(s->last_picture[i]==NULL) | |
| 687 s->last_picture[i]= s->next_picture[i]; | |
| 553 | 688 |
| 0 | 689 s->current_picture[i] = s->aux_picture[i]; |
| 690 } | |
| 691 } else { | |
| 692 for(i=0;i<3;i++) { | |
| 693 /* swap next and last */ | |
| 553 | 694 if(avctx->flags&CODEC_FLAG_DR1) |
| 695 tmp= avctx->dr_buffer[i]; | |
| 696 else | |
| 697 tmp = s->last_picture[i]; | |
| 698 | |
| 0 | 699 s->last_picture[i] = s->next_picture[i]; |
| 700 s->next_picture[i] = tmp; | |
| 701 s->current_picture[i] = tmp; | |
| 553 | 702 |
|
592
909e50c67d0e
last_picture should be never == NULL (it was with dr1) this might fix a segfault with error concealment
michaelni
parents:
591
diff
changeset
|
703 if(s->last_picture[i]==NULL) |
|
909e50c67d0e
last_picture should be never == NULL (it was with dr1) this might fix a segfault with error concealment
michaelni
parents:
591
diff
changeset
|
704 s->last_picture[i]= s->next_picture[i]; |
|
909e50c67d0e
last_picture should be never == NULL (it was with dr1) this might fix a segfault with error concealment
michaelni
parents:
591
diff
changeset
|
705 |
| 553 | 706 s->last_dr_opaque= s->next_dr_opaque; |
| 707 s->next_dr_opaque= avctx->dr_opaque_frame; | |
| 708 | |
| 557 | 709 if(s->has_b_frames && s->last_dr_opaque && s->codec_id!=CODEC_ID_SVQ1) |
| 553 | 710 avctx->dr_opaque_frame= s->last_dr_opaque; |
| 711 else | |
| 712 avctx->dr_opaque_frame= s->next_dr_opaque; | |
| 0 | 713 } |
| 714 } | |
| 591 | 715 /* set dequantizer, we cant do it during init as it might change for mpeg4 |
| 716 and we cant do it in the header decode as init isnt called for mpeg4 there yet */ | |
| 717 if(s->out_format == FMT_H263){ | |
| 718 if(s->mpeg_quant) | |
| 719 s->dct_unquantize = s->dct_unquantize_mpeg2; | |
| 720 else | |
| 721 s->dct_unquantize = s->dct_unquantize_h263; | |
| 722 }else | |
| 723 s->dct_unquantize = s->dct_unquantize_mpeg1; | |
| 0 | 724 } |
|
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
|
725 |
| 0 | 726 /* generic function for encode/decode called after a frame has been coded/decoded */ |
| 727 void MPV_frame_end(MpegEncContext *s) | |
| 728 { | |
| 652 | 729 s->avctx->key_frame = (s->pict_type == I_TYPE); |
| 730 s->avctx->pict_type = s->pict_type; | |
| 456 | 731 |
| 0 | 732 /* draw edge for correct motion prediction if outside */ |
| 553 | 733 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
|
734 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
|
735 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
|
736 draw_edges(s->current_picture[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2); |
| 0 | 737 } |
| 207 | 738 emms_c(); |
| 329 | 739 |
| 612 | 740 s->last_pict_type = s->pict_type; |
| 329 | 741 if(s->pict_type!=B_TYPE){ |
| 742 s->last_non_b_pict_type= s->pict_type; | |
| 341 | 743 s->num_available_buffers++; |
| 744 if(s->num_available_buffers>2) s->num_available_buffers= 2; | |
| 329 | 745 } |
| 0 | 746 } |
| 747 | |
| 324 | 748 /* reorder input for encoding */ |
| 749 void reorder_input(MpegEncContext *s, AVPicture *pict) | |
| 750 { | |
| 751 int i, j, index; | |
| 752 | |
| 753 if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES; | |
| 754 | |
| 755 // delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff) | |
| 756 | |
| 757 for(j=0; j<REORDER_BUFFER_SIZE-1; j++){ | |
| 758 s->coded_order[j]= s->coded_order[j+1]; | |
| 759 } | |
| 760 s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers | |
| 339 | 761 s->coded_order[j].pict_type=0; |
| 324 | 762 |
| 763 switch(s->input_pict_type){ | |
| 764 default: | |
| 765 case I_TYPE: | |
| 766 case S_TYPE: | |
| 767 case P_TYPE: | |
| 768 index= s->max_b_frames - s->b_frames_since_non_b; | |
| 769 s->b_frames_since_non_b=0; | |
| 770 break; | |
| 771 case B_TYPE: | |
| 772 index= s->max_b_frames + 1; | |
| 773 s->b_frames_since_non_b++; | |
| 774 break; | |
| 775 } | |
| 776 //printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize); | |
| 777 if( (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED)) | |
| 778 && pict->linesize[0] == s->linesize | |
| 556 | 779 && pict->linesize[1] == s->uvlinesize |
| 780 && pict->linesize[2] == s->uvlinesize){ | |
| 324 | 781 //printf("ptr\n"); |
| 782 for(i=0; i<3; i++){ | |
| 783 s->coded_order[index].picture[i]= pict->data[i]; | |
| 784 } | |
| 785 }else{ | |
| 786 //printf("copy\n"); | |
| 787 for(i=0; i<3; i++){ | |
| 788 uint8_t *src = pict->data[i]; | |
| 789 uint8_t *dest; | |
| 790 int src_wrap = pict->linesize[i]; | |
| 791 int dest_wrap = s->linesize; | |
| 792 int w = s->width; | |
| 793 int h = s->height; | |
| 794 | |
| 795 if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering | |
| 796 else dest= s->picture_buffer[s->picture_buffer_index][i]; | |
| 797 | |
| 798 if (i >= 1) { | |
| 799 dest_wrap >>= 1; | |
| 800 w >>= 1; | |
| 801 h >>= 1; | |
| 802 } | |
| 803 | |
| 804 s->coded_order[index].picture[i]= dest; | |
| 805 for(j=0;j<h;j++) { | |
| 806 memcpy(dest, src, w); | |
| 807 dest += dest_wrap; | |
| 808 src += src_wrap; | |
| 809 } | |
| 810 } | |
| 811 if(index!=0){ | |
| 812 s->picture_buffer_index++; | |
| 588 | 813 if(s->picture_buffer_index >= REORDER_BUFFER_SIZE) s->picture_buffer_index=0; |
| 324 | 814 } |
| 815 } | |
| 816 s->coded_order[index].pict_type = s->input_pict_type; | |
| 817 s->coded_order[index].qscale = s->input_qscale; | |
| 818 s->coded_order[index].force_type= s->force_input_type; | |
| 819 s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number; | |
| 820 s->coded_order[index].picture_number= s->input_picture_number; | |
| 821 | |
| 822 for(i=0; i<3; i++){ | |
| 823 s->new_picture[i]= s->coded_order[0].picture[i]; | |
| 824 } | |
| 825 } | |
| 826 | |
| 0 | 827 int MPV_encode_picture(AVCodecContext *avctx, |
| 828 unsigned char *buf, int buf_size, void *data) | |
| 829 { | |
| 830 MpegEncContext *s = avctx->priv_data; | |
| 831 AVPicture *pict = data; | |
| 832 | |
| 324 | 833 s->input_qscale = avctx->quality; |
| 0 | 834 |
| 835 init_put_bits(&s->pb, buf, buf_size, NULL, NULL); | |
| 836 | |
| 329 | 837 if(avctx->flags&CODEC_FLAG_TYPE){ |
| 838 s->input_pict_type= | |
| 839 s->force_input_type= avctx->key_frame ? I_TYPE : P_TYPE; | |
| 840 }else if(s->flags&CODEC_FLAG_PASS2){ | |
| 841 s->input_pict_type= | |
| 842 s->force_input_type= s->rc_context.entry[s->input_picture_number].new_pict_type; | |
| 843 }else{ | |
| 844 s->force_input_type=0; | |
| 845 if (!s->intra_only) { | |
| 846 /* first picture of GOP is intra */ | |
| 847 if (s->input_picture_in_gop_number % s->gop_size==0){ | |
| 848 s->input_pict_type = I_TYPE; | |
| 849 }else if(s->max_b_frames==0){ | |
| 850 s->input_pict_type = P_TYPE; | |
| 851 }else{ | |
| 852 if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ | |
| 853 s->input_pict_type = B_TYPE; | |
| 854 else | |
| 855 s->input_pict_type = P_TYPE; | |
| 856 } | |
| 857 } else { | |
| 324 | 858 s->input_pict_type = I_TYPE; |
| 859 } | |
| 0 | 860 } |
| 861 | |
| 329 | 862 if(s->input_pict_type==I_TYPE) |
| 863 s->input_picture_in_gop_number=0; | |
| 864 | |
| 324 | 865 reorder_input(s, pict); |
| 866 | |
| 867 /* output? */ | |
| 868 if(s->coded_order[0].picture[0]){ | |
| 869 | |
| 870 s->pict_type= s->coded_order[0].pict_type; | |
| 871 if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */ | |
| 872 s->qscale= s->coded_order[0].qscale; | |
| 873 s->force_type= s->coded_order[0].force_type; | |
| 874 s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number; | |
| 875 s->picture_number= s->coded_order[0].picture_number; | |
| 876 | |
| 553 | 877 MPV_frame_start(s, avctx); |
| 286 | 878 |
| 324 | 879 encode_picture(s, s->picture_number); |
| 652 | 880 |
| 376 | 881 avctx->real_pict_num = s->picture_number; |
| 324 | 882 avctx->header_bits = s->header_bits; |
| 883 avctx->mv_bits = s->mv_bits; | |
| 884 avctx->misc_bits = s->misc_bits; | |
| 885 avctx->i_tex_bits = s->i_tex_bits; | |
| 886 avctx->p_tex_bits = s->p_tex_bits; | |
| 887 avctx->i_count = s->i_count; | |
|
656
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
888 avctx->p_count = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx |
| 324 | 889 avctx->skip_count = s->skip_count; |
| 0 | 890 |
| 324 | 891 MPV_frame_end(s); |
| 892 | |
| 893 if (s->out_format == FMT_MJPEG) | |
| 894 mjpeg_picture_trailer(s); | |
| 895 | |
| 896 avctx->quality = s->qscale; | |
| 329 | 897 |
| 898 if(s->flags&CODEC_FLAG_PASS1) | |
| 899 ff_write_pass1_stats(s); | |
| 456 | 900 |
| 324 | 901 } |
| 902 | |
| 903 s->input_picture_number++; | |
| 904 s->input_picture_in_gop_number++; | |
| 0 | 905 |
| 906 flush_put_bits(&s->pb); | |
| 268 | 907 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8; |
| 612 | 908 |
| 268 | 909 s->total_bits += s->frame_bits; |
| 286 | 910 avctx->frame_bits = s->frame_bits; |
| 911 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n", | |
| 912 //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); | |
| 612 | 913 #if 0 //dump some stats to stats.txt for testing/debuging |
| 914 if(s->max_b_frames==0) | |
| 915 { | |
| 916 static FILE *f=NULL; | |
| 917 if(!f) f= fopen("stats.txt", "wb"); | |
| 918 get_psnr(pict->data, s->current_picture, | |
| 919 pict->linesize, s->linesize, avctx); | |
| 920 fprintf(f, "%7d, %7d, %2.4f\n", pbBufPtr(&s->pb) - s->pb.buf, s->qscale, avctx->psnr_y); | |
| 921 } | |
| 922 #endif | |
|
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
|
923 |
|
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
251
diff
changeset
|
924 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
|
925 /* 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
|
926 /* 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
|
927 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
|
928 pict->linesize, s->linesize, avctx); |
| 324 | 929 // 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
|
930 } |
|
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
|
931 return pbBufPtr(&s->pb) - s->pb.buf; |
| 0 | 932 } |
| 933 | |
| 255 | 934 static inline void gmc1_motion(MpegEncContext *s, |
| 935 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
| 936 int dest_offset, | |
| 937 UINT8 **ref_picture, int src_offset, | |
| 938 int h) | |
| 939 { | |
| 940 UINT8 *ptr; | |
| 556 | 941 int offset, src_x, src_y, linesize, uvlinesize; |
| 255 | 942 int motion_x, motion_y; |
| 566 | 943 int emu=0; |
| 255 | 944 |
|
265
4e9e728021d8
use ac prediction in mpeg4 encoding (5% smaller intra-blocks/keyframes)
michaelni
parents:
262
diff
changeset
|
945 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n"); |
| 255 | 946 motion_x= s->sprite_offset[0][0]; |
| 947 motion_y= s->sprite_offset[0][1]; | |
| 948 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
| 949 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
| 950 motion_x<<=(3-s->sprite_warping_accuracy); | |
| 951 motion_y<<=(3-s->sprite_warping_accuracy); | |
| 952 src_x = clip(src_x, -16, s->width); | |
| 953 if (src_x == s->width) | |
| 954 motion_x =0; | |
| 955 src_y = clip(src_y, -16, s->height); | |
| 956 if (src_y == s->height) | |
| 957 motion_y =0; | |
| 958 | |
| 959 linesize = s->linesize; | |
| 556 | 960 uvlinesize = s->uvlinesize; |
| 255 | 961 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; |
| 962 | |
| 963 dest_y+=dest_offset; | |
| 566 | 964 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
965 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
|
966 || src_y + (motion_y&15) + h > s->v_edge_pos){ |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
967 emulated_edge_mc(s, ptr, linesize, 17, h+1, src_x, src_y, s->h_edge_pos, s->v_edge_pos); |
| 566 | 968 ptr= s->edge_emu_buffer; |
| 969 emu=1; | |
| 970 } | |
| 971 } | |
| 255 | 972 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding); |
| 973 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding); | |
| 974 | |
| 975 motion_x= s->sprite_offset[1][0]; | |
| 976 motion_y= s->sprite_offset[1][1]; | |
| 977 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
| 978 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
| 979 motion_x<<=(3-s->sprite_warping_accuracy); | |
| 980 motion_y<<=(3-s->sprite_warping_accuracy); | |
| 981 src_x = clip(src_x, -8, s->width>>1); | |
| 982 if (src_x == s->width>>1) | |
| 983 motion_x =0; | |
| 984 src_y = clip(src_y, -8, s->height>>1); | |
| 985 if (src_y == s->height>>1) | |
| 986 motion_y =0; | |
| 987 | |
| 556 | 988 offset = (src_y * uvlinesize) + src_x + (src_offset>>1); |
| 255 | 989 ptr = ref_picture[1] + offset; |
| 566 | 990 if(emu){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
991 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 | 992 ptr= s->edge_emu_buffer; |
| 993 } | |
| 556 | 994 gmc1(dest_cb + (dest_offset>>1), ptr, uvlinesize, h>>1, motion_x&15, motion_y&15, s->no_rounding); |
| 566 | 995 |
| 255 | 996 ptr = ref_picture[2] + offset; |
| 566 | 997 if(emu){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
998 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 | 999 ptr= s->edge_emu_buffer; |
| 1000 } | |
| 556 | 1001 gmc1(dest_cr + (dest_offset>>1), ptr, uvlinesize, h>>1, motion_x&15, motion_y&15, s->no_rounding); |
| 255 | 1002 |
| 1003 return; | |
| 1004 } | |
| 1005 | |
| 566 | 1006 static void emulated_edge_mc(MpegEncContext *s, UINT8 *src, int linesize, int block_w, int block_h, |
| 553 | 1007 int src_x, int src_y, int w, int h){ |
| 1008 int x, y; | |
| 1009 int start_y, start_x, end_y, end_x; | |
| 566 | 1010 UINT8 *buf= s->edge_emu_buffer; |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1011 |
| 553 | 1012 if(src_y>= h){ |
| 1013 src+= (h-1-src_y)*linesize; | |
| 1014 src_y=h-1; | |
| 554 | 1015 }else if(src_y<=-block_h){ |
| 1016 src+= (1-block_h-src_y)*linesize; | |
| 1017 src_y=1-block_h; | |
| 553 | 1018 } |
| 1019 if(src_x>= w){ | |
| 1020 src+= (w-1-src_x); | |
| 1021 src_x=w-1; | |
| 554 | 1022 }else if(src_x<=-block_w){ |
| 1023 src+= (1-block_w-src_x); | |
| 1024 src_x=1-block_w; | |
| 553 | 1025 } |
| 1026 | |
| 1027 start_y= MAX(0, -src_y); | |
| 1028 start_x= MAX(0, -src_x); | |
| 1029 end_y= MIN(block_h, h-src_y); | |
| 1030 end_x= MIN(block_w, w-src_x); | |
| 566 | 1031 |
| 553 | 1032 // copy existing part |
| 1033 for(y=start_y; y<end_y; y++){ | |
| 1034 for(x=start_x; x<end_x; x++){ | |
| 1035 buf[x + y*linesize]= src[x + y*linesize]; | |
| 1036 } | |
| 1037 } | |
| 1038 | |
| 1039 //top | |
| 1040 for(y=0; y<start_y; y++){ | |
| 1041 for(x=start_x; x<end_x; x++){ | |
| 1042 buf[x + y*linesize]= buf[x + start_y*linesize]; | |
| 1043 } | |
| 1044 } | |
| 1045 | |
| 1046 //bottom | |
| 1047 for(y=end_y; y<block_h; y++){ | |
| 1048 for(x=start_x; x<end_x; x++){ | |
| 1049 buf[x + y*linesize]= buf[x + (end_y-1)*linesize]; | |
| 1050 } | |
| 1051 } | |
| 1052 | |
| 1053 for(y=0; y<block_h; y++){ | |
| 1054 //left | |
| 1055 for(x=0; x<start_x; x++){ | |
| 1056 buf[x + y*linesize]= buf[start_x + y*linesize]; | |
| 1057 } | |
| 1058 | |
| 1059 //right | |
| 1060 for(x=end_x; x<block_w; x++){ | |
| 1061 buf[x + y*linesize]= buf[end_x - 1 + y*linesize]; | |
| 1062 } | |
| 1063 } | |
| 1064 } | |
| 1065 | |
| 1066 | |
| 0 | 1067 /* apply one mpeg motion vector to the three components */ |
| 1068 static inline void mpeg_motion(MpegEncContext *s, | |
| 1069 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
| 1070 int dest_offset, | |
| 1071 UINT8 **ref_picture, int src_offset, | |
| 651 | 1072 int field_based, op_pixels_func (*pix_op)[4], |
| 0 | 1073 int motion_x, int motion_y, int h) |
| 1074 { | |
| 1075 UINT8 *ptr; | |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1076 int dxy, offset, mx, my, src_x, src_y, height, v_edge_pos, linesize, uvlinesize; |
| 553 | 1077 int emu=0; |
| 651 | 1078 #if 0 |
| 255 | 1079 if(s->quarter_sample) |
| 1080 { | |
| 1081 motion_x>>=1; | |
| 1082 motion_y>>=1; | |
| 1083 } | |
| 651 | 1084 #endif |
| 0 | 1085 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
| 1086 src_x = s->mb_x * 16 + (motion_x >> 1); | |
| 1087 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1); | |
| 1088 | |
| 1089 /* WARNING: do no forget half pels */ | |
| 1090 height = s->height >> field_based; | |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1091 v_edge_pos = s->v_edge_pos >> field_based; |
| 0 | 1092 src_x = clip(src_x, -16, s->width); |
| 1093 if (src_x == s->width) | |
| 1094 dxy &= ~1; | |
| 1095 src_y = clip(src_y, -16, height); | |
| 1096 if (src_y == height) | |
| 1097 dxy &= ~2; | |
|
565
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1098 linesize = s->linesize << field_based; |
|
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1099 uvlinesize = s->uvlinesize << field_based; |
| 0 | 1100 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset; |
| 1101 dest_y += dest_offset; | |
| 553 | 1102 |
| 1103 if(s->flags&CODEC_FLAG_EMU_EDGE){ | |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1104 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
|
1105 || src_y + (motion_y&1) + h > v_edge_pos){ |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1106 emulated_edge_mc(s, ptr, linesize, 17, h+1, src_x, src_y, s->h_edge_pos, v_edge_pos); |
| 553 | 1107 ptr= s->edge_emu_buffer; |
| 1108 emu=1; | |
| 1109 } | |
| 1110 } | |
| 651 | 1111 pix_op[0][dxy](dest_y, ptr, linesize, h); |
| 0 | 1112 |
| 485 | 1113 if(s->flags&CODEC_FLAG_GRAY) return; |
| 1114 | |
| 0 | 1115 if (s->out_format == FMT_H263) { |
| 1116 dxy = 0; | |
| 1117 if ((motion_x & 3) != 0) | |
| 1118 dxy |= 1; | |
| 1119 if ((motion_y & 3) != 0) | |
| 1120 dxy |= 2; | |
| 1121 mx = motion_x >> 2; | |
| 1122 my = motion_y >> 2; | |
| 1123 } else { | |
| 1124 mx = motion_x / 2; | |
| 1125 my = motion_y / 2; | |
| 1126 dxy = ((my & 1) << 1) | (mx & 1); | |
| 1127 mx >>= 1; | |
| 1128 my >>= 1; | |
| 1129 } | |
| 1130 | |
| 1131 src_x = s->mb_x * 8 + mx; | |
| 1132 src_y = s->mb_y * (8 >> field_based) + my; | |
| 1133 src_x = clip(src_x, -8, s->width >> 1); | |
| 1134 if (src_x == (s->width >> 1)) | |
| 1135 dxy &= ~1; | |
| 1136 src_y = clip(src_y, -8, height >> 1); | |
| 1137 if (src_y == (height >> 1)) | |
| 1138 dxy &= ~2; | |
|
565
44d744901ded
interlaced mpeg2 fix ... replacing linesize>>1 by uvlinesize brainlessly wasnt a good idea
michaelni
parents:
562
diff
changeset
|
1139 offset = (src_y * uvlinesize) + src_x + (src_offset >> 1); |
| 0 | 1140 ptr = ref_picture[1] + offset; |
| 553 | 1141 if(emu){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1142 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
| 553 | 1143 ptr= s->edge_emu_buffer; |
| 1144 } | |
| 651 | 1145 pix_op[1][dxy](dest_cb + (dest_offset >> 1), ptr, uvlinesize, h >> 1); |
| 553 | 1146 |
| 0 | 1147 ptr = ref_picture[2] + offset; |
| 553 | 1148 if(emu){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1149 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
| 553 | 1150 ptr= s->edge_emu_buffer; |
| 1151 } | |
| 651 | 1152 pix_op[1][dxy](dest_cr + (dest_offset >> 1), ptr, uvlinesize, h >> 1); |
| 0 | 1153 } |
| 1154 | |
| 255 | 1155 static inline void qpel_motion(MpegEncContext *s, |
| 1156 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
| 1157 int dest_offset, | |
| 1158 UINT8 **ref_picture, int src_offset, | |
| 651 | 1159 int field_based, op_pixels_func (*pix_op)[4], |
| 1160 qpel_mc_func (*qpix_op)[16], | |
| 255 | 1161 int motion_x, int motion_y, int h) |
| 1162 { | |
| 1163 UINT8 *ptr; | |
| 671 | 1164 int dxy, offset, mx, my, src_x, src_y, height, v_edge_pos, linesize, uvlinesize; |
| 554 | 1165 int emu=0; |
| 255 | 1166 |
| 1167 dxy = ((motion_y & 3) << 2) | (motion_x & 3); | |
| 1168 src_x = s->mb_x * 16 + (motion_x >> 2); | |
| 1169 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); | |
| 1170 | |
| 1171 height = s->height >> field_based; | |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1172 v_edge_pos = s->v_edge_pos >> field_based; |
| 255 | 1173 src_x = clip(src_x, -16, s->width); |
| 1174 if (src_x == s->width) | |
| 1175 dxy &= ~3; | |
| 1176 src_y = clip(src_y, -16, height); | |
| 1177 if (src_y == height) | |
| 1178 dxy &= ~12; | |
| 1179 linesize = s->linesize << field_based; | |
| 671 | 1180 uvlinesize = s->uvlinesize << field_based; |
| 255 | 1181 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset; |
| 1182 dest_y += dest_offset; | |
| 1183 //printf("%d %d %d\n", src_x, src_y, dxy); | |
| 554 | 1184 |
| 1185 if(s->flags&CODEC_FLAG_EMU_EDGE){ | |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1186 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
|
1187 || src_y + (motion_y&3) + h > v_edge_pos){ |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1188 emulated_edge_mc(s, ptr, linesize, 17, h+1, src_x, src_y, s->h_edge_pos, v_edge_pos); |
| 554 | 1189 ptr= s->edge_emu_buffer; |
| 1190 emu=1; | |
| 1191 } | |
| 1192 } | |
| 671 | 1193 if(!field_based) |
| 1194 qpix_op[0][dxy](dest_y, ptr, linesize); | |
| 1195 else{ | |
| 1196 //damn interlaced mode | |
| 1197 //FIXME boundary mirroring is not exactly correct here | |
| 1198 qpix_op[1][dxy](dest_y , ptr , linesize); | |
| 1199 qpix_op[1][dxy](dest_y+8, ptr+8, linesize); | |
| 1200 } | |
| 651 | 1201 |
| 485 | 1202 if(s->flags&CODEC_FLAG_GRAY) return; |
| 1203 | |
| 671 | 1204 if(field_based){ |
| 1205 mx= motion_x/2; | |
| 1206 my= motion_y>>1; | |
| 1207 }else if(s->divx_version){ | |
| 1208 mx= (motion_x>>1)|(motion_x&1); | |
| 1209 my= (motion_y>>1)|(motion_y&1); | |
| 1210 }else{ | |
| 1211 mx= motion_x/2; | |
| 1212 my= motion_y/2; | |
| 1213 } | |
| 1214 mx= (mx>>1)|(mx&1); | |
| 1215 my= (my>>1)|(my&1); | |
| 1216 dxy= (mx&1) | ((my&1)<<1); | |
| 1217 mx>>=1; | |
| 1218 my>>=1; | |
| 255 | 1219 |
| 1220 src_x = s->mb_x * 8 + mx; | |
| 1221 src_y = s->mb_y * (8 >> field_based) + my; | |
| 1222 src_x = clip(src_x, -8, s->width >> 1); | |
| 1223 if (src_x == (s->width >> 1)) | |
| 1224 dxy &= ~1; | |
| 1225 src_y = clip(src_y, -8, height >> 1); | |
| 1226 if (src_y == (height >> 1)) | |
| 1227 dxy &= ~2; | |
| 1228 | |
| 671 | 1229 offset = (src_y * uvlinesize) + src_x + (src_offset >> 1); |
| 255 | 1230 ptr = ref_picture[1] + offset; |
| 554 | 1231 if(emu){ |
| 671 | 1232 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
| 554 | 1233 ptr= s->edge_emu_buffer; |
| 1234 } | |
| 671 | 1235 pix_op[1][dxy](dest_cb + (dest_offset >> 1), ptr, uvlinesize, h >> 1); |
| 554 | 1236 |
| 255 | 1237 ptr = ref_picture[2] + offset; |
| 554 | 1238 if(emu){ |
| 671 | 1239 emulated_edge_mc(s, ptr, uvlinesize, 9, (h>>1)+1, src_x, src_y, s->h_edge_pos>>1, v_edge_pos>>1); |
| 554 | 1240 ptr= s->edge_emu_buffer; |
| 1241 } | |
| 671 | 1242 pix_op[1][dxy](dest_cr + (dest_offset >> 1), ptr, uvlinesize, h >> 1); |
| 255 | 1243 } |
| 1244 | |
| 1245 | |
| 0 | 1246 static inline void MPV_motion(MpegEncContext *s, |
| 1247 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr, | |
| 1248 int dir, UINT8 **ref_picture, | |
| 651 | 1249 op_pixels_func (*pix_op)[4], qpel_mc_func (*qpix_op)[16]) |
| 0 | 1250 { |
| 1251 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y; | |
| 1252 int mb_x, mb_y, i; | |
| 1253 UINT8 *ptr, *dest; | |
| 554 | 1254 int emu=0; |
| 0 | 1255 |
| 1256 mb_x = s->mb_x; | |
| 1257 mb_y = s->mb_y; | |
| 1258 | |
| 1259 switch(s->mv_type) { | |
| 1260 case MV_TYPE_16X16: | |
| 255 | 1261 if(s->mcsel){ |
| 1262 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 1263 ref_picture, 0, | |
| 1264 16); | |
| 651 | 1265 }else if(s->quarter_sample){ |
| 255 | 1266 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, |
| 1267 ref_picture, 0, | |
| 1268 0, pix_op, qpix_op, | |
| 1269 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
| 1270 }else{ | |
| 1271 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 1272 ref_picture, 0, | |
| 1273 0, pix_op, | |
| 1274 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
| 1275 } | |
| 0 | 1276 break; |
| 1277 case MV_TYPE_8X8: | |
| 671 | 1278 if(s->quarter_sample){ fprintf(stderr, "qpel4mv"); fflush(stderr);} |
| 0 | 1279 for(i=0;i<4;i++) { |
| 1280 motion_x = s->mv[dir][i][0]; | |
| 1281 motion_y = s->mv[dir][i][1]; | |
| 1282 | |
| 1283 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
| 1284 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8; | |
| 295 | 1285 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8; |
| 0 | 1286 |
| 1287 /* WARNING: do no forget half pels */ | |
| 1288 src_x = clip(src_x, -16, s->width); | |
| 1289 if (src_x == s->width) | |
| 1290 dxy &= ~1; | |
| 1291 src_y = clip(src_y, -16, s->height); | |
| 1292 if (src_y == s->height) | |
| 1293 dxy &= ~2; | |
| 1294 | |
| 1295 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); | |
| 554 | 1296 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1297 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
|
1298 || src_y + (motion_y&1) + 8 > s->v_edge_pos){ |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1299 emulated_edge_mc(s, ptr, s->linesize, 9, 9, src_x, src_y, s->h_edge_pos, s->v_edge_pos); |
| 554 | 1300 ptr= s->edge_emu_buffer; |
| 1301 } | |
| 1302 } | |
| 0 | 1303 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; |
| 651 | 1304 pix_op[1][dxy](dest, ptr, s->linesize, 8); |
| 0 | 1305 } |
| 485 | 1306 |
| 1307 if(s->flags&CODEC_FLAG_GRAY) break; | |
| 0 | 1308 /* In case of 8X8, we construct a single chroma motion vector |
| 1309 with a special rounding */ | |
| 1310 mx = 0; | |
| 1311 my = 0; | |
| 1312 for(i=0;i<4;i++) { | |
| 1313 mx += s->mv[dir][i][0]; | |
| 1314 my += s->mv[dir][i][1]; | |
| 1315 } | |
| 1316 if (mx >= 0) | |
| 1317 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
| 1318 else { | |
| 1319 mx = -mx; | |
| 1320 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1)); | |
| 1321 } | |
| 1322 if (my >= 0) | |
| 1323 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
| 1324 else { | |
| 1325 my = -my; | |
| 1326 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1)); | |
| 1327 } | |
| 1328 dxy = ((my & 1) << 1) | (mx & 1); | |
| 1329 mx >>= 1; | |
| 1330 my >>= 1; | |
| 1331 | |
| 1332 src_x = mb_x * 8 + mx; | |
| 1333 src_y = mb_y * 8 + my; | |
| 1334 src_x = clip(src_x, -8, s->width/2); | |
| 1335 if (src_x == s->width/2) | |
| 1336 dxy &= ~1; | |
| 1337 src_y = clip(src_y, -8, s->height/2); | |
| 1338 if (src_y == s->height/2) | |
| 1339 dxy &= ~2; | |
| 1340 | |
| 556 | 1341 offset = (src_y * (s->uvlinesize)) + src_x; |
| 0 | 1342 ptr = ref_picture[1] + offset; |
| 554 | 1343 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1344 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
|
1345 || src_y + (dxy>>1) + 8 > s->v_edge_pos>>1){ |
|
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1346 emulated_edge_mc(s, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
| 554 | 1347 ptr= s->edge_emu_buffer; |
| 1348 emu=1; | |
| 1349 } | |
| 1350 } | |
| 651 | 1351 pix_op[1][dxy](dest_cb, ptr, s->uvlinesize, 8); |
| 554 | 1352 |
| 0 | 1353 ptr = ref_picture[2] + offset; |
| 554 | 1354 if(emu){ |
|
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
1355 emulated_edge_mc(s, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
| 554 | 1356 ptr= s->edge_emu_buffer; |
| 1357 } | |
| 651 | 1358 pix_op[1][dxy](dest_cr, ptr, s->uvlinesize, 8); |
| 0 | 1359 break; |
| 1360 case MV_TYPE_FIELD: | |
| 1361 if (s->picture_structure == PICT_FRAME) { | |
| 671 | 1362 if(s->quarter_sample){ |
| 1363 /* top field */ | |
| 1364 qpel_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 1365 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
| 1366 1, pix_op, qpix_op, | |
| 1367 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
| 1368 /* bottom field */ | |
| 1369 qpel_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
| 1370 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
| 1371 1, pix_op, qpix_op, | |
| 1372 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
| 1373 }else{ | |
| 1374 /* top field */ | |
| 1375 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 1376 ref_picture, s->field_select[dir][0] ? s->linesize : 0, | |
| 1377 1, pix_op, | |
| 1378 s->mv[dir][0][0], s->mv[dir][0][1], 8); | |
| 1379 /* bottom field */ | |
| 1380 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize, | |
| 1381 ref_picture, s->field_select[dir][1] ? s->linesize : 0, | |
| 1382 1, pix_op, | |
| 1383 s->mv[dir][1][0], s->mv[dir][1][1], 8); | |
| 1384 } | |
| 0 | 1385 } else { |
| 1386 | |
| 1387 | |
| 1388 } | |
| 1389 break; | |
| 1390 } | |
| 1391 } | |
| 1392 | |
| 1393 | |
| 1394 /* put block[] to dest[] */ | |
| 1395 static inline void put_dct(MpegEncContext *s, | |
| 1396 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
| 1397 { | |
| 1398 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
|
1399 s->dct_unquantize(s, block, i, s->qscale); |
| 478 | 1400 ff_idct_put (dest, line_size, block); |
| 0 | 1401 } |
| 1402 | |
| 1403 /* add block[] to dest[] */ | |
| 1404 static inline void add_dct(MpegEncContext *s, | |
| 1405 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
| 1406 { | |
| 1407 if (s->block_last_index[i] >= 0) { | |
| 481 | 1408 ff_idct_add (dest, line_size, block); |
| 1409 } | |
| 1410 } | |
| 1411 | |
| 1412 static inline void add_dequant_dct(MpegEncContext *s, | |
| 1413 DCTELEM *block, int i, UINT8 *dest, int line_size) | |
| 1414 { | |
| 1415 if (s->block_last_index[i] >= 0) { | |
| 1416 s->dct_unquantize(s, block, i, s->qscale); | |
| 324 | 1417 |
| 478 | 1418 ff_idct_add (dest, line_size, block); |
| 0 | 1419 } |
| 1420 } | |
| 1421 | |
| 456 | 1422 /** |
| 1423 * cleans dc, ac, coded_block for the current non intra MB | |
| 1424 */ | |
| 1425 void ff_clean_intra_table_entries(MpegEncContext *s) | |
| 1426 { | |
| 1427 int wrap = s->block_wrap[0]; | |
| 1428 int xy = s->block_index[0]; | |
| 1429 | |
| 1430 s->dc_val[0][xy ] = | |
| 1431 s->dc_val[0][xy + 1 ] = | |
| 1432 s->dc_val[0][xy + wrap] = | |
| 1433 s->dc_val[0][xy + 1 + wrap] = 1024; | |
| 1434 /* ac pred */ | |
| 1435 memset(s->ac_val[0][xy ], 0, 32 * sizeof(INT16)); | |
| 1436 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(INT16)); | |
| 1437 if (s->msmpeg4_version>=3) { | |
| 1438 s->coded_block[xy ] = | |
| 1439 s->coded_block[xy + 1 ] = | |
| 1440 s->coded_block[xy + wrap] = | |
| 1441 s->coded_block[xy + 1 + wrap] = 0; | |
| 1442 } | |
| 1443 /* chroma */ | |
| 1444 wrap = s->block_wrap[4]; | |
| 1445 xy = s->mb_x + 1 + (s->mb_y + 1) * wrap; | |
| 1446 s->dc_val[1][xy] = | |
| 1447 s->dc_val[2][xy] = 1024; | |
| 1448 /* ac pred */ | |
| 1449 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16)); | |
| 1450 memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16)); | |
| 1451 | |
| 1452 s->mbintra_table[s->mb_x + s->mb_y*s->mb_width]= 0; | |
| 1453 } | |
| 1454 | |
| 0 | 1455 /* generic function called after a macroblock has been parsed by the |
| 1456 decoder or after it has been encoded by the encoder. | |
| 1457 | |
| 1458 Important variables used: | |
| 1459 s->mb_intra : true if intra macroblock | |
| 1460 s->mv_dir : motion vector direction | |
| 1461 s->mv_type : motion vector type | |
| 1462 s->mv : motion vector | |
| 1463 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
| 1464 */ | |
| 1465 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) | |
| 1466 { | |
| 244 | 1467 int mb_x, mb_y; |
| 481 | 1468 const int mb_xy = s->mb_y * s->mb_width + s->mb_x; |
| 0 | 1469 |
| 1470 mb_x = s->mb_x; | |
| 1471 mb_y = s->mb_y; | |
| 1472 | |
|
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1473 #ifdef FF_POSTPROCESS |
|
466
805714c0c10f
new field for communicating with external postprocessing
nickols_k
parents:
462
diff
changeset
|
1474 /* Obsolete. Exists for compatibility with mplayer only. */ |
|
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1475 quant_store[mb_y][mb_x]=s->qscale; |
|
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1476 //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
|
1477 #else |
| 575 | 1478 /* even more obsolete, exists for mplayer xp only */ |
| 467 | 1479 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
|
1480 #endif |
| 575 | 1481 s->qscale_table[mb_xy]= s->qscale; |
|
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
1482 |
| 0 | 1483 /* update DC predictors for P macroblocks */ |
| 1484 if (!s->mb_intra) { | |
|
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1485 if (s->h263_pred || s->h263_aic) { |
| 481 | 1486 if(s->mbintra_table[mb_xy]) |
| 456 | 1487 ff_clean_intra_table_entries(s); |
| 0 | 1488 } else { |
| 456 | 1489 s->last_dc[0] = |
| 1490 s->last_dc[1] = | |
| 0 | 1491 s->last_dc[2] = 128 << s->intra_dc_precision; |
| 1492 } | |
| 1493 } | |
|
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
1494 else if (s->h263_pred || s->h263_aic) |
| 481 | 1495 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
|
1496 |
| 280 | 1497 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */ |
| 481 | 1498 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here |
| 0 | 1499 |
| 481 | 1500 const int wrap = s->block_wrap[0]; |
| 1501 const int xy = s->block_index[0]; | |
| 664 | 1502 const int mb_index= s->mb_x + s->mb_y*s->mb_width; |
|
619
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1503 if(s->mv_type == MV_TYPE_8X8){ |
| 664 | 1504 s->co_located_type_table[mb_index]= CO_LOCATED_TYPE_4MV; |
|
619
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1505 } else { |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1506 int motion_x, motion_y; |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1507 if (s->mb_intra) { |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1508 motion_x = 0; |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1509 motion_y = 0; |
| 664 | 1510 if(s->co_located_type_table) |
| 1511 s->co_located_type_table[mb_index]= 0; | |
|
619
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1512 } else if (s->mv_type == MV_TYPE_16X16) { |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1513 motion_x = s->mv[0][0][0]; |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1514 motion_y = s->mv[0][0][1]; |
| 664 | 1515 if(s->co_located_type_table) |
| 1516 s->co_located_type_table[mb_index]= 0; | |
|
619
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1517 } else /*if (s->mv_type == MV_TYPE_FIELD)*/ { |
| 664 | 1518 int i; |
|
619
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1519 motion_x = s->mv[0][0][0] + s->mv[0][1][0]; |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1520 motion_y = s->mv[0][0][1] + s->mv[0][1][1]; |
|
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1521 motion_x = (motion_x>>1) | (motion_x&1); |
| 664 | 1522 for(i=0; i<2; i++){ |
| 1523 s->field_mv_table[mb_index][i][0]= s->mv[0][i][0]; | |
| 1524 s->field_mv_table[mb_index][i][1]= s->mv[0][i][1]; | |
| 1525 s->field_select_table[mb_index][i]= s->field_select[0][i]; | |
| 1526 } | |
| 1527 s->co_located_type_table[mb_index]= CO_LOCATED_TYPE_FIELDMV; | |
|
619
2be2cc8fd0a1
mpeg4 interlaced decoding support (not completly implemented/tested due to lack of samples)
michaelni
parents:
618
diff
changeset
|
1528 } |
| 0 | 1529 /* no update if 8X8 because it has been done during parsing */ |
| 244 | 1530 s->motion_val[xy][0] = motion_x; |
| 1531 s->motion_val[xy][1] = motion_y; | |
| 1532 s->motion_val[xy + 1][0] = motion_x; | |
| 1533 s->motion_val[xy + 1][1] = motion_y; | |
| 1534 s->motion_val[xy + wrap][0] = motion_x; | |
| 1535 s->motion_val[xy + wrap][1] = motion_y; | |
| 1536 s->motion_val[xy + 1 + wrap][0] = motion_x; | |
| 1537 s->motion_val[xy + 1 + wrap][1] = motion_y; | |
| 0 | 1538 } |
| 1539 } | |
| 1540 | |
| 324 | 1541 if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) { |
| 0 | 1542 UINT8 *dest_y, *dest_cb, *dest_cr; |
| 481 | 1543 int dct_linesize, dct_offset; |
| 651 | 1544 op_pixels_func (*op_pix)[4]; |
| 1545 qpel_mc_func (*op_qpix)[16]; | |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1546 |
| 324 | 1547 /* avoid copy if macroblock skipped in last frame too |
| 1548 dont touch it for B-frames as they need the skip info from the next p-frame */ | |
| 1549 if (s->pict_type != B_TYPE) { | |
| 481 | 1550 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
|
1551 if (s->mb_skiped) { |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1552 s->mb_skiped = 0; |
| 556 | 1553 |
| 1554 (*mbskip_ptr) ++; /* indicate that this time we skiped it */ | |
| 1555 if(*mbskip_ptr >99) *mbskip_ptr= 99; | |
| 1556 | |
| 324 | 1557 /* if previous was skipped too, then nothing to do ! |
| 1558 skip only during decoding as we might trash the buffers during encoding a bit */ | |
| 556 | 1559 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
|
1560 goto the_end; |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1561 } else { |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1562 *mbskip_ptr = 0; /* not skipped */ |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1563 } |
|
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1564 } |
| 0 | 1565 |
|
562
23e58889a108
exporting mbskip_table after it has been allocated
michaelni
parents:
557
diff
changeset
|
1566 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
|
1567 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
|
1568 dest_cr = s->current_picture[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
| 0 | 1569 |
| 1570 if (s->interlaced_dct) { | |
| 1571 dct_linesize = s->linesize * 2; | |
| 1572 dct_offset = s->linesize; | |
| 1573 } else { | |
| 1574 dct_linesize = s->linesize; | |
| 1575 dct_offset = s->linesize * 8; | |
| 1576 } | |
| 1577 | |
| 1578 if (!s->mb_intra) { | |
| 1579 /* motion handling */ | |
| 456 | 1580 /* decoding or more than one mb_type (MC was allready done otherwise) */ |
| 481 | 1581 if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){ |
| 327 | 1582 if ((!s->no_rounding) || s->pict_type==B_TYPE){ |
| 324 | 1583 op_pix = put_pixels_tab; |
| 651 | 1584 op_qpix= put_qpel_pixels_tab; |
| 324 | 1585 }else{ |
| 1586 op_pix = put_no_rnd_pixels_tab; | |
| 651 | 1587 op_qpix= put_no_rnd_qpel_pixels_tab; |
| 324 | 1588 } |
| 0 | 1589 |
| 324 | 1590 if (s->mv_dir & MV_DIR_FORWARD) { |
| 1591 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
| 651 | 1592 op_pix = avg_pixels_tab; |
| 1593 op_qpix= avg_qpel_pixels_tab; | |
| 324 | 1594 } |
| 1595 if (s->mv_dir & MV_DIR_BACKWARD) { | |
| 1596 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
| 1597 } | |
| 0 | 1598 } |
| 1599 | |
| 481 | 1600 /* skip dequant / idct if we are really late ;) */ |
| 1601 if(s->hurry_up>1) goto the_end; | |
| 1602 | |
| 0 | 1603 /* add dct residue */ |
| 591 | 1604 if(s->encoding || !(s->mpeg2 || s->h263_msmpeg4 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){ |
| 481 | 1605 add_dequant_dct(s, block[0], 0, dest_y, dct_linesize); |
| 1606 add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
| 1607 add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
| 1608 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
| 0 | 1609 |
| 485 | 1610 if(!(s->flags&CODEC_FLAG_GRAY)){ |
| 556 | 1611 add_dequant_dct(s, block[4], 4, dest_cb, s->uvlinesize); |
| 1612 add_dequant_dct(s, block[5], 5, dest_cr, s->uvlinesize); | |
| 485 | 1613 } |
| 481 | 1614 } else { |
| 1615 add_dct(s, block[0], 0, dest_y, dct_linesize); | |
| 1616 add_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
| 1617 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
| 1618 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
| 1619 | |
| 485 | 1620 if(!(s->flags&CODEC_FLAG_GRAY)){ |
| 556 | 1621 add_dct(s, block[4], 4, dest_cb, s->uvlinesize); |
| 1622 add_dct(s, block[5], 5, dest_cr, s->uvlinesize); | |
| 485 | 1623 } |
| 481 | 1624 } |
| 0 | 1625 } else { |
| 1626 /* dct only in intra block */ | |
| 1627 put_dct(s, block[0], 0, dest_y, dct_linesize); | |
| 1628 put_dct(s, block[1], 1, dest_y + 8, dct_linesize); | |
| 1629 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize); | |
| 1630 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize); | |
| 1631 | |
| 485 | 1632 if(!(s->flags&CODEC_FLAG_GRAY)){ |
| 556 | 1633 put_dct(s, block[4], 4, dest_cb, s->uvlinesize); |
| 1634 put_dct(s, block[5], 5, dest_cr, s->uvlinesize); | |
| 485 | 1635 } |
| 0 | 1636 } |
| 1637 } | |
|
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1638 the_end: |
| 294 | 1639 emms_c(); //FIXME remove |
| 1640 } | |
| 1641 | |
|
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
1642 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold) |
| 456 | 1643 { |
| 1644 static const char tab[64]= | |
| 1645 {3,2,2,1,1,1,1,1, | |
| 1646 1,1,1,1,1,1,1,1, | |
| 1647 1,1,1,1,1,1,1,1, | |
| 1648 0,0,0,0,0,0,0,0, | |
| 1649 0,0,0,0,0,0,0,0, | |
| 1650 0,0,0,0,0,0,0,0, | |
| 1651 0,0,0,0,0,0,0,0, | |
| 1652 0,0,0,0,0,0,0,0}; | |
| 1653 int score=0; | |
| 1654 int run=0; | |
| 1655 int i; | |
| 1656 DCTELEM *block= s->block[n]; | |
| 1657 const int last_index= s->block_last_index[n]; | |
|
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
1658 int skip_dc; |
| 456 | 1659 |
|
604
4344cc8033bd
skip blocks with small chroma dc too (if the user wants it) needed to avoid IDCT(input coeffs !=0) == 0 problems which cause catastrophic error accumulation at qp=1
michaelni
parents:
599
diff
changeset
|
1660 if(threshold<0){ |
|
4344cc8033bd
skip blocks with small chroma dc too (if the user wants it) needed to avoid IDCT(input coeffs !=0) == 0 problems which cause catastrophic error accumulation at qp=1
michaelni
parents:
599
diff
changeset
|
1661 skip_dc=0; |
|
4344cc8033bd
skip blocks with small chroma dc too (if the user wants it) needed to avoid IDCT(input coeffs !=0) == 0 problems which cause catastrophic error accumulation at qp=1
michaelni
parents:
599
diff
changeset
|
1662 threshold= -threshold; |
|
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
1663 }else |
|
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
1664 skip_dc=1; |
|
604
4344cc8033bd
skip blocks with small chroma dc too (if the user wants it) needed to avoid IDCT(input coeffs !=0) == 0 problems which cause catastrophic error accumulation at qp=1
michaelni
parents:
599
diff
changeset
|
1665 |
| 456 | 1666 /* are all which we could set to zero are allready zero? */ |
| 1667 if(last_index<=skip_dc - 1) return; | |
| 1668 | |
| 1669 for(i=0; i<=last_index; i++){ | |
| 1670 const int j = zigzag_direct[i]; | |
| 1671 const int level = ABS(block[j]); | |
| 1672 if(level==1){ | |
| 1673 if(skip_dc && i==0) continue; | |
| 1674 score+= tab[run]; | |
| 1675 run=0; | |
| 1676 }else if(level>1){ | |
| 1677 return; | |
| 1678 }else{ | |
| 1679 run++; | |
| 1680 } | |
| 1681 } | |
| 1682 if(score >= threshold) return; | |
| 1683 for(i=skip_dc; i<=last_index; i++){ | |
| 1684 const int j = zigzag_direct[i]; | |
| 1685 block[j]=0; | |
| 1686 } | |
| 1687 if(block[0]) s->block_last_index[n]= 0; | |
| 1688 else s->block_last_index[n]= -1; | |
| 1689 } | |
| 1690 | |
| 344 | 1691 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index) |
| 1692 { | |
| 1693 int i; | |
| 1694 const int maxlevel= s->max_qcoeff; | |
| 1695 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
|
1696 |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
1697 for(i=0;i<=last_index; i++){ |
| 344 | 1698 const int j = zigzag_direct[i]; |
| 1699 int level = block[j]; | |
| 1700 | |
| 1701 if (level>maxlevel) level=maxlevel; | |
| 1702 else if(level<minlevel) level=minlevel; | |
| 1703 block[j]= level; | |
| 1704 } | |
| 1705 } | |
| 324 | 1706 |
| 1707 static void encode_mb(MpegEncContext *s, int motion_x, int motion_y) | |
| 294 | 1708 { |
| 1709 const int mb_x= s->mb_x; | |
| 1710 const int mb_y= s->mb_y; | |
| 1711 int i; | |
| 456 | 1712 int skip_dct[6]; |
| 324 | 1713 #if 0 |
| 1714 if (s->interlaced_dct) { | |
| 1715 dct_linesize = s->linesize * 2; | |
| 1716 dct_offset = s->linesize; | |
| 1717 } else { | |
| 1718 dct_linesize = s->linesize; | |
| 1719 dct_offset = s->linesize * 8; | |
| 1720 } | |
| 1721 #endif | |
| 456 | 1722 for(i=0; i<6; i++) skip_dct[i]=0; |
| 294 | 1723 |
| 324 | 1724 if (s->mb_intra) { |
| 1725 UINT8 *ptr; | |
| 1726 int wrap; | |
|
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1727 int emu=0; |
| 294 | 1728 |
| 324 | 1729 wrap = s->linesize; |
| 1730 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
|
1731 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
|
1732 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
|
1733 ptr= s->edge_emu_buffer; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1734 emu=1; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1735 } |
| 324 | 1736 get_pixels(s->block[0], ptr , wrap); |
| 1737 get_pixels(s->block[1], ptr + 8, wrap); | |
| 1738 get_pixels(s->block[2], ptr + 8 * wrap , wrap); | |
| 1739 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap); | |
| 294 | 1740 |
| 487 | 1741 if(s->flags&CODEC_FLAG_GRAY){ |
| 1742 skip_dct[4]= 1; | |
| 1743 skip_dct[5]= 1; | |
| 1744 }else{ | |
| 1745 wrap >>=1; | |
| 1746 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
|
1747 if(emu){ |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1748 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
|
1749 ptr= s->edge_emu_buffer; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1750 } |
| 487 | 1751 get_pixels(s->block[4], ptr, wrap); |
| 294 | 1752 |
| 487 | 1753 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
|
1754 if(emu){ |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1755 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
|
1756 ptr= s->edge_emu_buffer; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1757 } |
| 487 | 1758 get_pixels(s->block[5], ptr, wrap); |
| 1759 } | |
| 324 | 1760 }else{ |
| 651 | 1761 op_pixels_func (*op_pix)[4]; |
| 1762 qpel_mc_func (*op_qpix)[16]; | |
| 324 | 1763 UINT8 *dest_y, *dest_cb, *dest_cr; |
| 456 | 1764 UINT8 *ptr_y, *ptr_cb, *ptr_cr; |
| 1765 int wrap_y, wrap_c; | |
|
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1766 int emu=0; |
| 294 | 1767 |
| 324 | 1768 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize ) + mb_x * 16; |
| 556 | 1769 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; |
| 1770 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; | |
| 456 | 1771 wrap_y = s->linesize; |
| 1772 wrap_c = wrap_y>>1; | |
| 1773 ptr_y = s->new_picture[0] + (mb_y * 16 * wrap_y) + mb_x * 16; | |
| 1774 ptr_cb = s->new_picture[1] + (mb_y * 8 * wrap_c) + mb_x * 8; | |
| 1775 ptr_cr = s->new_picture[2] + (mb_y * 8 * wrap_c) + mb_x * 8; | |
| 324 | 1776 |
| 327 | 1777 if ((!s->no_rounding) || s->pict_type==B_TYPE){ |
| 324 | 1778 op_pix = put_pixels_tab; |
| 651 | 1779 op_qpix= put_qpel_pixels_tab; |
| 295 | 1780 }else{ |
| 324 | 1781 op_pix = put_no_rnd_pixels_tab; |
| 651 | 1782 op_qpix= put_no_rnd_qpel_pixels_tab; |
| 324 | 1783 } |
| 295 | 1784 |
| 324 | 1785 if (s->mv_dir & MV_DIR_FORWARD) { |
| 1786 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix); | |
| 651 | 1787 op_pix = avg_pixels_tab; |
| 1788 op_qpix= avg_qpel_pixels_tab; | |
| 324 | 1789 } |
| 1790 if (s->mv_dir & MV_DIR_BACKWARD) { | |
| 1791 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix); | |
| 1792 } | |
| 295 | 1793 |
|
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1794 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
|
1795 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
|
1796 ptr_y= s->edge_emu_buffer; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1797 emu=1; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1798 } |
| 456 | 1799 diff_pixels(s->block[0], ptr_y , dest_y , wrap_y); |
| 1800 diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y); | |
| 1801 diff_pixels(s->block[2], ptr_y + 8 * wrap_y , dest_y + 8 * wrap_y , wrap_y); | |
| 1802 diff_pixels(s->block[3], ptr_y + 8 * wrap_y + 8, dest_y + 8 * wrap_y + 8, wrap_y); | |
| 487 | 1803 |
| 1804 if(s->flags&CODEC_FLAG_GRAY){ | |
| 1805 skip_dct[4]= 1; | |
| 1806 skip_dct[5]= 1; | |
| 1807 }else{ | |
|
570
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1808 if(emu){ |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1809 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
|
1810 ptr_cb= s->edge_emu_buffer; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1811 } |
| 487 | 1812 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
|
1813 if(emu){ |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1814 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
|
1815 ptr_cr= s->edge_emu_buffer; |
|
274d9c5a75ee
use edge emu for encoding of width or height %16!=0 files
michaelni
parents:
569
diff
changeset
|
1816 } |
| 487 | 1817 diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c); |
| 1818 } | |
| 1819 | |
| 456 | 1820 /* pre quantization */ |
| 1821 if(s->mc_mb_var[s->mb_width*mb_y+ mb_x]<2*s->qscale*s->qscale){ | |
| 1822 if(pix_abs8x8(ptr_y , dest_y , wrap_y) < 20*s->qscale) skip_dct[0]= 1; | |
| 1823 if(pix_abs8x8(ptr_y + 8, dest_y + 8, wrap_y) < 20*s->qscale) skip_dct[1]= 1; | |
| 1824 if(pix_abs8x8(ptr_y + 8*wrap_y , dest_y + 8*wrap_y , wrap_y) < 20*s->qscale) skip_dct[2]= 1; | |
| 1825 if(pix_abs8x8(ptr_y + 8*wrap_y + 8, dest_y + 8*wrap_y + 8, wrap_y) < 20*s->qscale) skip_dct[3]= 1; | |
| 1826 if(pix_abs8x8(ptr_cb , dest_cb , wrap_y) < 20*s->qscale) skip_dct[4]= 1; | |
| 1827 if(pix_abs8x8(ptr_cr , dest_cr , wrap_y) < 20*s->qscale) skip_dct[5]= 1; | |
| 1828 #if 0 | |
| 1829 { | |
| 1830 static int stat[7]; | |
| 1831 int num=0; | |
| 1832 for(i=0; i<6; i++) | |
| 1833 if(skip_dct[i]) num++; | |
| 1834 stat[num]++; | |
| 1835 | |
| 1836 if(s->mb_x==0 && s->mb_y==0){ | |
| 1837 for(i=0; i<7; i++){ | |
| 1838 printf("%6d %1d\n", stat[i], i); | |
| 1839 } | |
| 1840 } | |
| 1841 } | |
| 1842 #endif | |
| 1843 } | |
| 324 | 1844 |
| 294 | 1845 } |
| 1846 | |
| 1847 #if 0 | |
| 1848 { | |
| 1849 float adap_parm; | |
| 1850 | |
| 1851 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) / | |
| 1852 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0); | |
| 1853 | |
| 1854 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", | |
| 1855 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P', | |
| 1856 s->qscale, adap_parm, s->qscale*adap_parm, | |
| 1857 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var); | |
| 1858 } | |
| 1859 #endif | |
| 1860 /* DCT & quantize */ | |
| 344 | 1861 if(s->out_format==FMT_MJPEG){ |
| 1862 for(i=0;i<6;i++) { | |
| 1863 int overflow; | |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
1864 s->block_last_index[i] = s->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
|
1865 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]); |
| 344 | 1866 } |
| 1867 }else{ | |
| 1868 for(i=0;i<6;i++) { | |
| 456 | 1869 if(!skip_dct[i]){ |
| 1870 int overflow; | |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
1871 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow); |
| 344 | 1872 // 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
|
1873 // 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
|
1874 // of improve it. Just INTRADC clipping deserves changes in quantizer |
| 456 | 1875 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]); |
| 1876 }else | |
| 1877 s->block_last_index[i]= -1; | |
| 344 | 1878 } |
| 456 | 1879 if(s->luma_elim_threshold && !s->mb_intra) |
| 1880 for(i=0; i<4; i++) | |
|
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
1881 dct_single_coeff_elimination(s, i, s->luma_elim_threshold); |
| 456 | 1882 if(s->chroma_elim_threshold && !s->mb_intra) |
| 1883 for(i=4; i<6; i++) | |
|
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
1884 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold); |
| 294 | 1885 } |
| 1886 | |
| 487 | 1887 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){ |
| 1888 s->block_last_index[4]= | |
| 1889 s->block_last_index[5]= 0; | |
| 1890 s->block[4][0]= | |
| 1891 s->block[5][0]= 128; | |
| 1892 } | |
| 1893 | |
| 294 | 1894 /* huffman encode */ |
| 1895 switch(s->out_format) { | |
| 1896 case FMT_MPEG1: | |
| 1897 mpeg1_encode_mb(s, s->block, motion_x, motion_y); | |
| 1898 break; | |
| 1899 case FMT_H263: | |
| 1900 if (s->h263_msmpeg4) | |
| 1901 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
| 1902 else if(s->h263_pred) | |
| 1903 mpeg4_encode_mb(s, s->block, motion_x, motion_y); | |
| 1904 else | |
| 1905 h263_encode_mb(s, s->block, motion_x, motion_y); | |
| 1906 break; | |
| 1907 case FMT_MJPEG: | |
| 1908 mjpeg_encode_mb(s, s->block); | |
| 1909 break; | |
| 1910 } | |
| 1911 } | |
| 1912 | |
| 456 | 1913 void ff_copy_bits(PutBitContext *pb, UINT8 *src, int length) |
| 294 | 1914 { |
| 326 | 1915 int bytes= length>>4; |
| 1916 int bits= length&15; | |
| 1917 int i; | |
| 1918 | |
| 456 | 1919 if(length==0) return; |
| 1920 | |
| 326 | 1921 for(i=0; i<bytes; i++) put_bits(pb, 16, be2me_16(((uint16_t*)src)[i])); |
| 1922 put_bits(pb, bits, be2me_16(((uint16_t*)src)[i])>>(16-bits)); | |
| 0 | 1923 } |
| 1924 | |
| 456 | 1925 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){ |
| 326 | 1926 int i; |
| 1927 | |
| 1928 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop? | |
| 1929 | |
| 1930 /* mpeg1 */ | |
| 1931 d->mb_incr= s->mb_incr; | |
| 1932 for(i=0; i<3; i++) | |
| 1933 d->last_dc[i]= s->last_dc[i]; | |
| 1934 | |
| 1935 /* statistics */ | |
| 1936 d->mv_bits= s->mv_bits; | |
| 1937 d->i_tex_bits= s->i_tex_bits; | |
| 1938 d->p_tex_bits= s->p_tex_bits; | |
| 1939 d->i_count= s->i_count; | |
|
656
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
1940 d->f_count= s->f_count; |
|
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
1941 d->b_count= s->b_count; |
| 326 | 1942 d->skip_count= s->skip_count; |
| 1943 d->misc_bits= s->misc_bits; | |
| 329 | 1944 d->last_bits= 0; |
| 327 | 1945 |
| 1946 d->mb_skiped= s->mb_skiped; | |
| 326 | 1947 } |
| 1948 | |
| 456 | 1949 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){ |
| 326 | 1950 int i; |
| 1951 | |
| 1952 memcpy(d->mv, s->mv, 2*4*2*sizeof(int)); | |
| 1953 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop? | |
| 1954 | |
| 1955 /* mpeg1 */ | |
| 1956 d->mb_incr= s->mb_incr; | |
| 1957 for(i=0; i<3; i++) | |
| 1958 d->last_dc[i]= s->last_dc[i]; | |
| 1959 | |
| 1960 /* statistics */ | |
| 1961 d->mv_bits= s->mv_bits; | |
| 1962 d->i_tex_bits= s->i_tex_bits; | |
| 1963 d->p_tex_bits= s->p_tex_bits; | |
| 1964 d->i_count= s->i_count; | |
|
656
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
1965 d->f_count= s->f_count; |
|
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
1966 d->b_count= s->b_count; |
| 326 | 1967 d->skip_count= s->skip_count; |
| 1968 d->misc_bits= s->misc_bits; | |
| 1969 | |
| 1970 d->mb_intra= s->mb_intra; | |
| 327 | 1971 d->mb_skiped= s->mb_skiped; |
| 326 | 1972 d->mv_type= s->mv_type; |
| 1973 d->mv_dir= s->mv_dir; | |
| 1974 d->pb= s->pb; | |
| 456 | 1975 if(s->data_partitioning){ |
| 1976 d->pb2= s->pb2; | |
| 1977 d->tex_pb= s->tex_pb; | |
| 1978 } | |
| 326 | 1979 d->block= s->block; |
| 1980 for(i=0; i<6; i++) | |
| 1981 d->block_last_index[i]= s->block_last_index[i]; | |
| 1982 } | |
| 1983 | |
| 456 | 1984 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type, |
| 1985 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2], | |
| 1986 int *dmin, int *next_block, int motion_x, int motion_y) | |
| 1987 { | |
| 1988 int bits_count; | |
| 1989 | |
| 1990 copy_context_before_encode(s, backup, type); | |
| 1991 | |
| 1992 s->block= s->blocks[*next_block]; | |
| 1993 s->pb= pb[*next_block]; | |
| 1994 if(s->data_partitioning){ | |
| 1995 s->pb2 = pb2 [*next_block]; | |
| 1996 s->tex_pb= tex_pb[*next_block]; | |
| 1997 } | |
| 1998 | |
| 1999 encode_mb(s, motion_x, motion_y); | |
| 2000 | |
| 2001 bits_count= get_bit_count(&s->pb); | |
| 2002 if(s->data_partitioning){ | |
| 2003 bits_count+= get_bit_count(&s->pb2); | |
| 2004 bits_count+= get_bit_count(&s->tex_pb); | |
| 2005 } | |
| 2006 | |
| 2007 if(bits_count<*dmin){ | |
| 2008 *dmin= bits_count; | |
| 2009 *next_block^=1; | |
| 2010 | |
| 2011 copy_context_after_encode(best, s, type); | |
| 2012 } | |
| 2013 } | |
| 326 | 2014 |
| 0 | 2015 static void encode_picture(MpegEncContext *s, int picture_number) |
| 2016 { | |
| 294 | 2017 int mb_x, mb_y, last_gob, pdif = 0; |
| 2018 int i; | |
| 286 | 2019 int bits; |
| 326 | 2020 MpegEncContext best_s, backup_s; |
| 456 | 2021 UINT8 bit_buf[2][3000]; |
| 2022 UINT8 bit_buf2[2][3000]; | |
| 2023 UINT8 bit_buf_tex[2][3000]; | |
| 2024 PutBitContext pb[2], pb2[2], tex_pb[2]; | |
| 2025 | |
| 2026 for(i=0; i<2; i++){ | |
| 2027 init_put_bits(&pb [i], bit_buf [i], 3000, NULL, NULL); | |
| 2028 init_put_bits(&pb2 [i], bit_buf2 [i], 3000, NULL, NULL); | |
| 2029 init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000, NULL, NULL); | |
| 2030 } | |
| 0 | 2031 |
| 2032 s->picture_number = picture_number; | |
| 268 | 2033 |
| 294 | 2034 s->block_wrap[0]= |
| 2035 s->block_wrap[1]= | |
| 2036 s->block_wrap[2]= | |
| 2037 s->block_wrap[3]= s->mb_width*2 + 2; | |
| 2038 s->block_wrap[4]= | |
| 2039 s->block_wrap[5]= s->mb_width + 2; | |
| 2040 | |
| 268 | 2041 /* Reset the average MB variance */ |
| 456 | 2042 s->mb_var_sum = 0; |
| 2043 s->mc_mb_var_sum = 0; | |
| 327 | 2044 |
| 2045 /* we need to initialize some time vars before we can encode b-frames */ | |
| 2046 if (s->h263_pred && !s->h263_msmpeg4) | |
| 2047 ff_set_mpeg4_time(s, s->picture_number); | |
| 2048 | |
| 608 | 2049 s->scene_change_score=0; |
| 2050 | |
| 268 | 2051 /* Estimate motion for every MB */ |
| 324 | 2052 if(s->pict_type != I_TYPE){ |
| 294 | 2053 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
| 2054 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; | |
| 2055 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
| 2056 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
| 2057 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
| 2058 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
| 2059 s->mb_x = mb_x; | |
| 2060 s->mb_y = mb_y; | |
| 2061 s->block_index[0]+=2; | |
| 2062 s->block_index[1]+=2; | |
| 2063 s->block_index[2]+=2; | |
| 2064 s->block_index[3]+=2; | |
| 268 | 2065 |
| 294 | 2066 /* compute motion vector & mb_type and store in context */ |
| 324 | 2067 if(s->pict_type==B_TYPE) |
| 2068 ff_estimate_b_frame_motion(s, mb_x, mb_y); | |
| 2069 else | |
| 2070 ff_estimate_p_frame_motion(s, mb_x, mb_y); | |
| 294 | 2071 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER; |
| 268 | 2072 } |
| 2073 } | |
| 294 | 2074 emms_c(); |
| 456 | 2075 }else /* if(s->pict_type == I_TYPE) */{ |
| 294 | 2076 /* I-Frame */ |
| 2077 //FIXME do we need to zero them? | |
| 2078 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2); | |
| 324 | 2079 memset(s->p_mv_table , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2); |
| 294 | 2080 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
| 612 | 2081 |
| 2082 if(!s->fixed_qscale){ | |
| 2083 /* finding spatial complexity for I-frame rate control */ | |
| 2084 for(mb_y=0; mb_y < s->mb_height; mb_y++) { | |
| 2085 for(mb_x=0; mb_x < s->mb_width; mb_x++) { | |
| 2086 int xx = mb_x * 16; | |
| 2087 int yy = mb_y * 16; | |
| 2088 uint8_t *pix = s->new_picture[0] + (yy * s->linesize) + xx; | |
| 2089 int varc; | |
| 2090 int sum = pix_sum(pix, s->linesize); | |
| 2091 | |
| 2092 sum= (sum+8)>>4; | |
| 2093 varc = (pix_norm1(pix, s->linesize) - sum*sum + 500 + 128)>>8; | |
| 2094 | |
| 2095 s->mb_var[s->mb_width * mb_y + mb_x] = varc; | |
| 2096 s->mb_var_sum += varc; | |
| 2097 } | |
| 2098 } | |
| 2099 } | |
| 268 | 2100 } |
| 608 | 2101 if(s->scene_change_score > 0 && s->pict_type == P_TYPE){ |
| 271 | 2102 s->pict_type= I_TYPE; |
| 294 | 2103 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height); |
| 329 | 2104 if(s->max_b_frames==0){ |
| 2105 s->input_pict_type= I_TYPE; | |
| 2106 s->input_picture_in_gop_number=0; | |
| 2107 } | |
| 612 | 2108 //printf("Scene change detected, encoding as I Frame %d %d\n", s->mb_var_sum, s->mc_mb_var_sum); |
| 271 | 2109 } |
| 324 | 2110 |
| 2111 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) | |
| 2112 s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER); | |
| 2113 ff_fix_long_p_mvs(s); | |
| 2114 if(s->pict_type==B_TYPE){ | |
| 2115 s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD); | |
| 2116 s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD); | |
| 327 | 2117 |
| 2118 ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD); | |
| 2119 ff_fix_long_b_mvs(s, s->b_back_mv_table, s->b_code, MB_TYPE_BACKWARD); | |
| 2120 ff_fix_long_b_mvs(s, s->b_bidir_forw_mv_table, s->f_code, MB_TYPE_BIDIR); | |
| 2121 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
|
2122 } |
| 324 | 2123 |
|
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
2124 //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
|
2125 |
| 271 | 2126 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var); |
| 612 | 2127 if (!s->fixed_qscale) |
| 329 | 2128 s->qscale = ff_rate_estimate_qscale(s); |
| 2129 | |
| 0 | 2130 if (s->out_format == FMT_MJPEG) { |
| 2131 /* 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
|
2132 s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0]; |
| 0 | 2133 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
|
2134 s->intra_matrix[i] = CLAMP_TO_8BIT((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3); |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
2135 convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16, |
| 344 | 2136 s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias); |
| 0 | 2137 } |
| 2138 | |
| 286 | 2139 s->last_bits= get_bit_count(&s->pb); |
| 0 | 2140 switch(s->out_format) { |
| 2141 case FMT_MJPEG: | |
| 2142 mjpeg_picture_header(s); | |
| 2143 break; | |
| 2144 case FMT_H263: | |
| 2145 if (s->h263_msmpeg4) | |
| 2146 msmpeg4_encode_picture_header(s, picture_number); | |
| 2147 else if (s->h263_pred) | |
| 2148 mpeg4_encode_picture_header(s, picture_number); | |
| 2149 else if (s->h263_rv10) | |
| 2150 rv10_encode_picture_header(s, picture_number); | |
| 2151 else | |
| 2152 h263_encode_picture_header(s, picture_number); | |
| 2153 break; | |
| 2154 case FMT_MPEG1: | |
| 2155 mpeg1_encode_picture_header(s, picture_number); | |
| 2156 break; | |
| 2157 } | |
| 286 | 2158 bits= get_bit_count(&s->pb); |
| 2159 s->header_bits= bits - s->last_bits; | |
| 2160 s->last_bits= bits; | |
| 2161 s->mv_bits=0; | |
| 2162 s->misc_bits=0; | |
| 2163 s->i_tex_bits=0; | |
| 2164 s->p_tex_bits=0; | |
| 2165 s->i_count=0; | |
|
656
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
2166 s->f_count=0; |
|
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
2167 s->b_count=0; |
| 286 | 2168 s->skip_count=0; |
| 2169 | |
| 0 | 2170 /* init last dc values */ |
| 2171 /* note: quant matrix value (8) is implied here */ | |
| 2172 s->last_dc[0] = 128; | |
| 2173 s->last_dc[1] = 128; | |
| 2174 s->last_dc[2] = 128; | |
| 2175 s->mb_incr = 1; | |
| 2176 s->last_mv[0][0][0] = 0; | |
| 2177 s->last_mv[0][0][1] = 0; | |
| 2178 | |
| 162 | 2179 /* Get the GOB height based on picture height */ |
| 231 | 2180 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) { |
| 162 | 2181 if (s->height <= 400) |
| 2182 s->gob_index = 1; | |
| 2183 else if (s->height <= 800) | |
| 2184 s->gob_index = 2; | |
| 2185 else | |
| 2186 s->gob_index = 4; | |
| 456 | 2187 }else if(s->codec_id==CODEC_ID_MPEG4){ |
| 2188 s->gob_index = 1; | |
| 162 | 2189 } |
| 456 | 2190 |
| 2191 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE) | |
| 2192 ff_mpeg4_init_partitions(s); | |
| 2193 | |
| 2194 s->resync_mb_x=0; | |
| 2195 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
|
2196 for(mb_y=0; mb_y < s->mb_height; mb_y++) { |
| 456 | 2197 /* 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
|
2198 /* 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
|
2199 if (s->rtp_mode) { |
|
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2200 if (!mb_y) { |
|
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2201 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
|
2202 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
|
2203 } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) { |
| 456 | 2204 // 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
|
2205 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
|
2206 if (last_gob) { |
| 456 | 2207 s->first_slice_line = 1; |
| 2208 }else{ | |
| 2209 /*MN: we reset it here instead at the end of each line cuz mpeg4 can have | |
| 2210 slice lines starting & ending in the middle*/ | |
| 2211 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
|
2212 } |
|
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2213 } |
|
3f5b72726118
- More work on preliminary bit rate control, just to be able to get an
pulento
parents:
232
diff
changeset
|
2214 } |
| 499 | 2215 |
| 2216 s->y_dc_scale= s->y_dc_scale_table[ s->qscale ]; | |
| 2217 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
|
2218 |
| 266 | 2219 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1; |
| 2220 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1); | |
| 2221 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1; | |
| 2222 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2); | |
| 2223 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2); | |
| 2224 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
|
2225 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
| 327 | 2226 const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x]; |
| 2227 const int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1; | |
| 456 | 2228 // int d; |
| 294 | 2229 int dmin=10000000; |
| 0 | 2230 |
|
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
2231 s->mb_x = mb_x; |
|
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
2232 s->mb_y = mb_y; |
| 266 | 2233 s->block_index[0]+=2; |
| 2234 s->block_index[1]+=2; | |
| 2235 s->block_index[2]+=2; | |
| 2236 s->block_index[3]+=2; | |
| 2237 s->block_index[4]++; | |
| 2238 s->block_index[5]++; | |
| 456 | 2239 |
| 2240 /* write gob / video packet header for formats which support it at any MB (MPEG4) */ | |
| 2241 if(s->rtp_mode && s->mb_y>0 && s->codec_id==CODEC_ID_MPEG4){ | |
| 2242 int pdif= pbBufPtr(&s->pb) - s->ptr_lastgob; | |
| 2243 | |
| 2244 //the *2 is there so we stay below the requested size | |
| 2245 if(pdif + s->mb_line_avgsize/s->mb_width >= s->rtp_payload_size){ | |
| 2246 if(s->codec_id==CODEC_ID_MPEG4){ | |
| 2247 if(s->data_partitioning && s->pict_type!=B_TYPE){ | |
| 2248 ff_mpeg4_merge_partitions(s); | |
| 2249 ff_mpeg4_init_partitions(s); | |
| 2250 } | |
| 2251 ff_mpeg4_encode_video_packet_header(s); | |
| 2252 | |
| 2253 if(s->flags&CODEC_FLAG_PASS1){ | |
| 2254 int bits= get_bit_count(&s->pb); | |
| 2255 s->misc_bits+= bits - s->last_bits; | |
| 2256 s->last_bits= bits; | |
| 2257 } | |
| 2258 ff_mpeg4_clean_buffers(s); | |
| 2259 } | |
| 2260 s->ptr_lastgob = pbBufPtr(&s->pb); | |
| 2261 s->first_slice_line=1; | |
| 2262 s->resync_mb_x=mb_x; | |
| 2263 s->resync_mb_y=mb_y; | |
| 2264 } | |
| 2265 | |
| 2266 if( (s->resync_mb_x == s->mb_x) | |
| 2267 && s->resync_mb_y+1 == s->mb_y){ | |
| 2268 s->first_slice_line=0; | |
| 2269 } | |
| 2270 } | |
| 2271 | |
| 294 | 2272 if(mb_type & (mb_type-1)){ // more than 1 MB type possible |
| 327 | 2273 int next_block=0; |
| 456 | 2274 int pb_bits_count, pb2_bits_count, tex_pb_bits_count; |
| 326 | 2275 |
| 2276 copy_context_before_encode(&backup_s, s, -1); | |
| 456 | 2277 backup_s.pb= s->pb; |
| 2278 best_s.data_partitioning= s->data_partitioning; | |
| 2279 if(s->data_partitioning){ | |
| 2280 backup_s.pb2= s->pb2; | |
| 2281 backup_s.tex_pb= s->tex_pb; | |
| 2282 } | |
| 326 | 2283 |
| 294 | 2284 if(mb_type&MB_TYPE_INTER){ |
| 327 | 2285 s->mv_dir = MV_DIR_FORWARD; |
| 295 | 2286 s->mv_type = MV_TYPE_16X16; |
| 294 | 2287 s->mb_intra= 0; |
| 324 | 2288 s->mv[0][0][0] = s->p_mv_table[xy][0]; |
| 2289 s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
| 456 | 2290 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER, pb, pb2, tex_pb, |
| 2291 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]); | |
| 0 | 2292 } |
| 326 | 2293 if(mb_type&MB_TYPE_INTER4V){ |
| 327 | 2294 s->mv_dir = MV_DIR_FORWARD; |
| 295 | 2295 s->mv_type = MV_TYPE_8X8; |
| 2296 s->mb_intra= 0; | |
| 2297 for(i=0; i<4; i++){ | |
| 2298 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
| 2299 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
| 2300 } | |
| 456 | 2301 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER4V, pb, pb2, tex_pb, |
| 2302 &dmin, &next_block, 0, 0); | |
| 327 | 2303 } |
| 2304 if(mb_type&MB_TYPE_FORWARD){ | |
| 2305 s->mv_dir = MV_DIR_FORWARD; | |
| 2306 s->mv_type = MV_TYPE_16X16; | |
| 2307 s->mb_intra= 0; | |
| 2308 s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
| 2309 s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
| 456 | 2310 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_FORWARD, pb, pb2, tex_pb, |
| 2311 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]); | |
| 327 | 2312 } |
| 2313 if(mb_type&MB_TYPE_BACKWARD){ | |
| 2314 s->mv_dir = MV_DIR_BACKWARD; | |
| 2315 s->mv_type = MV_TYPE_16X16; | |
| 2316 s->mb_intra= 0; | |
| 2317 s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
| 2318 s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
| 456 | 2319 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BACKWARD, pb, pb2, tex_pb, |
| 2320 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]); | |
| 327 | 2321 } |
| 2322 if(mb_type&MB_TYPE_BIDIR){ | |
| 2323 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; | |
| 2324 s->mv_type = MV_TYPE_16X16; | |
| 2325 s->mb_intra= 0; | |
| 2326 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
| 2327 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
| 2328 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
| 2329 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
| 456 | 2330 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BIDIR, pb, pb2, tex_pb, |
| 2331 &dmin, &next_block, 0, 0); | |
| 327 | 2332 } |
| 2333 if(mb_type&MB_TYPE_DIRECT){ | |
| 2334 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; | |
| 2335 s->mv_type = MV_TYPE_16X16; //FIXME | |
| 2336 s->mb_intra= 0; | |
| 2337 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0]; | |
| 2338 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1]; | |
| 2339 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0]; | |
| 2340 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1]; | |
| 456 | 2341 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_DIRECT, pb, pb2, tex_pb, |
| 2342 &dmin, &next_block, s->b_direct_mv_table[xy][0], s->b_direct_mv_table[xy][1]); | |
| 295 | 2343 } |
| 294 | 2344 if(mb_type&MB_TYPE_INTRA){ |
| 327 | 2345 s->mv_dir = MV_DIR_FORWARD; |
| 295 | 2346 s->mv_type = MV_TYPE_16X16; |
| 294 | 2347 s->mb_intra= 1; |
| 2348 s->mv[0][0][0] = 0; | |
| 2349 s->mv[0][0][1] = 0; | |
| 456 | 2350 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTRA, pb, pb2, tex_pb, |
| 2351 &dmin, &next_block, 0, 0); | |
| 326 | 2352 /* force cleaning of ac/dc pred stuff if needed ... */ |
| 2353 if(s->h263_pred || s->h263_aic) | |
| 2354 s->mbintra_table[mb_x + mb_y*s->mb_width]=1; | |
| 295 | 2355 } |
| 326 | 2356 copy_context_after_encode(s, &best_s, -1); |
| 456 | 2357 |
| 2358 pb_bits_count= get_bit_count(&s->pb); | |
| 2359 flush_put_bits(&s->pb); | |
| 2360 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count); | |
| 2361 s->pb= backup_s.pb; | |
| 2362 | |
| 2363 if(s->data_partitioning){ | |
| 2364 pb2_bits_count= get_bit_count(&s->pb2); | |
| 2365 flush_put_bits(&s->pb2); | |
| 2366 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count); | |
| 2367 s->pb2= backup_s.pb2; | |
| 2368 | |
| 2369 tex_pb_bits_count= get_bit_count(&s->tex_pb); | |
| 2370 flush_put_bits(&s->tex_pb); | |
| 2371 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count); | |
| 2372 s->tex_pb= backup_s.tex_pb; | |
| 2373 } | |
| 329 | 2374 s->last_bits= get_bit_count(&s->pb); |
| 294 | 2375 } else { |
| 324 | 2376 int motion_x, motion_y; |
| 2377 s->mv_type=MV_TYPE_16X16; | |
| 294 | 2378 // only one MB-Type possible |
| 327 | 2379 switch(mb_type){ |
| 2380 case MB_TYPE_INTRA: | |
| 324 | 2381 s->mv_dir = MV_DIR_FORWARD; |
| 294 | 2382 s->mb_intra= 1; |
| 324 | 2383 motion_x= s->mv[0][0][0] = 0; |
| 2384 motion_y= s->mv[0][0][1] = 0; | |
| 327 | 2385 break; |
| 2386 case MB_TYPE_INTER: | |
| 324 | 2387 s->mv_dir = MV_DIR_FORWARD; |
| 2388 s->mb_intra= 0; | |
| 2389 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0]; | |
| 2390 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
| 327 | 2391 break; |
| 456 | 2392 case MB_TYPE_INTER4V: |
| 2393 s->mv_dir = MV_DIR_FORWARD; | |
| 2394 s->mv_type = MV_TYPE_8X8; | |
| 2395 s->mb_intra= 0; | |
| 2396 for(i=0; i<4; i++){ | |
| 2397 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0]; | |
| 2398 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1]; | |
| 2399 } | |
| 2400 motion_x= motion_y= 0; | |
| 2401 break; | |
| 327 | 2402 case MB_TYPE_DIRECT: |
| 324 | 2403 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; |
| 2404 s->mb_intra= 0; | |
| 327 | 2405 motion_x=s->b_direct_mv_table[xy][0]; |
| 2406 motion_y=s->b_direct_mv_table[xy][1]; | |
| 2407 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0]; | |
| 2408 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1]; | |
| 2409 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0]; | |
| 2410 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1]; | |
| 2411 break; | |
| 2412 case MB_TYPE_BIDIR: | |
| 324 | 2413 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; |
| 294 | 2414 s->mb_intra= 0; |
| 324 | 2415 motion_x=0; |
| 2416 motion_y=0; | |
| 2417 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
| 2418 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
| 2419 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
| 2420 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
| 327 | 2421 break; |
| 2422 case MB_TYPE_BACKWARD: | |
| 324 | 2423 s->mv_dir = MV_DIR_BACKWARD; |
| 2424 s->mb_intra= 0; | |
| 2425 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
| 2426 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
| 327 | 2427 break; |
| 2428 case MB_TYPE_FORWARD: | |
| 324 | 2429 s->mv_dir = MV_DIR_FORWARD; |
| 2430 s->mb_intra= 0; | |
| 2431 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
| 2432 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
| 2433 // printf(" %d %d ", motion_x, motion_y); | |
| 327 | 2434 break; |
| 2435 default: | |
| 324 | 2436 motion_x=motion_y=0; //gcc warning fix |
| 2437 printf("illegal MB type\n"); | |
| 294 | 2438 } |
| 324 | 2439 encode_mb(s, motion_x, motion_y); |
|
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2440 } |
| 327 | 2441 /* clean the MV table in IPS frames for direct mode in B frames */ |
| 2442 if(s->mb_intra /* && I,P,S_TYPE */){ | |
| 2443 s->p_mv_table[xy][0]=0; | |
| 2444 s->p_mv_table[xy][1]=0; | |
| 2445 } | |
| 0 | 2446 |
|
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
|
2447 MPV_decode_mb(s, s->block); |
| 456 | 2448 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_width, get_bit_count(&s->pb)); |
| 0 | 2449 } |
|
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
|
2450 |
|
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
|
2451 |
| 231 | 2452 /* Obtain average GOB size for RTP */ |
| 2453 if (s->rtp_mode) { | |
| 2454 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
|
2455 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line; |
| 231 | 2456 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
|
2457 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
|
2458 s->ptr_last_mb_line = pbBufPtr(&s->pb); |
| 231 | 2459 } |
| 2460 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y, | |
| 2461 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize); | |
| 456 | 2462 if(s->codec_id!=CODEC_ID_MPEG4) s->first_slice_line = 0; //FIXME clean |
| 231 | 2463 } |
| 0 | 2464 } |
| 294 | 2465 emms_c(); |
| 286 | 2466 |
| 456 | 2467 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE) |
| 2468 ff_mpeg4_merge_partitions(s); | |
| 2469 | |
| 2470 if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE) | |
| 208 | 2471 msmpeg4_encode_ext_header(s); |
| 2472 | |
| 456 | 2473 if(s->codec_id==CODEC_ID_MPEG4) |
| 2474 ff_mpeg4_stuffing(&s->pb); | |
| 2475 | |
| 162 | 2476 //if (s->gob_number) |
| 2477 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number); | |
| 231 | 2478 |
| 2479 /* Send the last GOB if RTP */ | |
| 2480 if (s->rtp_mode) { | |
| 2481 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
|
2482 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
| 231 | 2483 /* Call the RTP callback to send the last GOB */ |
| 2484 if (s->rtp_callback) | |
| 2485 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
|
2486 s->ptr_lastgob = pbBufPtr(&s->pb); |
| 231 | 2487 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif); |
| 2488 } | |
| 0 | 2489 } |
| 2490 | |
| 220 | 2491 static int dct_quantize_c(MpegEncContext *s, |
| 0 | 2492 DCTELEM *block, int n, |
| 344 | 2493 int qscale, int *overflow) |
| 0 | 2494 { |
| 2495 int i, j, level, last_non_zero, q; | |
| 2496 const int *qmat; | |
| 344 | 2497 int bias; |
| 2498 int max=0; | |
| 2499 unsigned int threshold1, threshold2; | |
| 456 | 2500 |
|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
619
diff
changeset
|
2501 s->fdct (block); |
| 0 | 2502 |
| 64 | 2503 /* we need this permutation so that we correct the IDCT |
| 2504 permutation. will be moved into DCT code */ | |
| 2505 block_permute(block); | |
| 2506 | |
| 0 | 2507 if (s->mb_intra) { |
|
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2508 if (!s->h263_aic) { |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2509 if (n < 4) |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2510 q = s->y_dc_scale; |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2511 else |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2512 q = s->c_dc_scale; |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2513 q = q << 3; |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2514 } else |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2515 /* 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
|
2516 q = 1 << 3; |
|
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
2517 |
| 0 | 2518 /* note: block[0] is assumed to be positive */ |
| 2519 block[0] = (block[0] + (q >> 1)) / q; | |
| 2520 i = 1; | |
| 2521 last_non_zero = 0; | |
| 344 | 2522 qmat = s->q_intra_matrix[qscale]; |
| 635 | 2523 bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT); |
| 0 | 2524 } else { |
| 2525 i = 0; | |
| 2526 last_non_zero = -1; | |
| 344 | 2527 qmat = s->q_inter_matrix[qscale]; |
| 635 | 2528 bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT); |
| 0 | 2529 } |
| 635 | 2530 threshold1= (1<<QMAT_SHIFT) - bias - 1; |
| 2531 threshold2= (threshold1<<1); | |
| 0 | 2532 |
| 2533 for(;i<64;i++) { | |
| 2534 j = zigzag_direct[i]; | |
| 2535 level = block[j]; | |
| 2536 level = level * qmat[j]; | |
| 344 | 2537 |
| 2538 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
| 2539 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
| 2540 if(((unsigned)(level+threshold1))>threshold2){ | |
| 2541 if(level>0){ | |
| 635 | 2542 level= (bias + level)>>QMAT_SHIFT; |
| 344 | 2543 block[j]= level; |
| 2544 }else{ | |
| 635 | 2545 level= (bias - level)>>QMAT_SHIFT; |
| 344 | 2546 block[j]= -level; |
| 0 | 2547 } |
| 344 | 2548 max |=level; |
| 0 | 2549 last_non_zero = i; |
| 344 | 2550 }else{ |
| 2551 block[j]=0; | |
| 0 | 2552 } |
| 2553 } | |
| 344 | 2554 *overflow= s->max_qcoeff < max; //overflow might have happend |
| 2555 | |
| 0 | 2556 return last_non_zero; |
| 2557 } | |
| 2558 | |
|
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
|
2559 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
|
2560 DCTELEM *block, int n, int qscale) |
| 0 | 2561 { |
| 200 | 2562 int i, level, nCoeffs; |
| 0 | 2563 const UINT16 *quant_matrix; |
| 2564 | |
| 200 | 2565 if(s->alternate_scan) nCoeffs= 64; |
| 2566 else nCoeffs= s->block_last_index[n]+1; | |
| 2567 | |
| 0 | 2568 if (s->mb_intra) { |
| 2569 if (n < 4) | |
| 2570 block[0] = block[0] * s->y_dc_scale; | |
| 2571 else | |
| 2572 block[0] = block[0] * s->c_dc_scale; | |
| 2573 /* XXX: only mpeg1 */ | |
| 2574 quant_matrix = s->intra_matrix; | |
| 200 | 2575 for(i=1;i<nCoeffs;i++) { |
| 2576 int j= zigzag_direct[i]; | |
| 2577 level = block[j]; | |
| 0 | 2578 if (level) { |
| 2579 if (level < 0) { | |
| 2580 level = -level; | |
| 200 | 2581 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 0 | 2582 level = (level - 1) | 1; |
| 2583 level = -level; | |
| 2584 } else { | |
| 200 | 2585 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 0 | 2586 level = (level - 1) | 1; |
| 2587 } | |
| 2588 #ifdef PARANOID | |
| 2589 if (level < -2048 || level > 2047) | |
| 2590 fprintf(stderr, "unquant error %d %d\n", i, level); | |
| 2591 #endif | |
| 200 | 2592 block[j] = level; |
| 0 | 2593 } |
| 2594 } | |
| 2595 } else { | |
| 2596 i = 0; | |
| 344 | 2597 quant_matrix = s->inter_matrix; |
| 217 | 2598 for(;i<nCoeffs;i++) { |
| 200 | 2599 int j= zigzag_direct[i]; |
| 2600 level = block[j]; | |
| 0 | 2601 if (level) { |
| 2602 if (level < 0) { | |
| 2603 level = -level; | |
| 2604 level = (((level << 1) + 1) * qscale * | |
| 200 | 2605 ((int) (quant_matrix[j]))) >> 4; |
| 0 | 2606 level = (level - 1) | 1; |
| 2607 level = -level; | |
| 2608 } else { | |
| 2609 level = (((level << 1) + 1) * qscale * | |
| 200 | 2610 ((int) (quant_matrix[j]))) >> 4; |
| 0 | 2611 level = (level - 1) | 1; |
| 2612 } | |
| 2613 #ifdef PARANOID | |
| 2614 if (level < -2048 || level > 2047) | |
| 2615 fprintf(stderr, "unquant error %d %d\n", i, level); | |
| 2616 #endif | |
| 200 | 2617 block[j] = level; |
| 0 | 2618 } |
| 2619 } | |
| 2620 } | |
| 2621 } | |
|
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
|
2622 |
| 325 | 2623 static void dct_unquantize_mpeg2_c(MpegEncContext *s, |
| 2624 DCTELEM *block, int n, int qscale) | |
| 2625 { | |
| 2626 int i, level, nCoeffs; | |
| 2627 const UINT16 *quant_matrix; | |
| 2628 | |
| 2629 if(s->alternate_scan) nCoeffs= 64; | |
| 2630 else nCoeffs= s->block_last_index[n]+1; | |
| 2631 | |
| 2632 if (s->mb_intra) { | |
| 2633 if (n < 4) | |
| 2634 block[0] = block[0] * s->y_dc_scale; | |
| 2635 else | |
| 2636 block[0] = block[0] * s->c_dc_scale; | |
| 2637 quant_matrix = s->intra_matrix; | |
| 2638 for(i=1;i<nCoeffs;i++) { | |
| 2639 int j= zigzag_direct[i]; | |
| 2640 level = block[j]; | |
| 2641 if (level) { | |
| 2642 if (level < 0) { | |
| 2643 level = -level; | |
| 2644 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
| 2645 level = -level; | |
| 2646 } else { | |
| 2647 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
| 2648 } | |
| 2649 #ifdef PARANOID | |
| 2650 if (level < -2048 || level > 2047) | |
| 2651 fprintf(stderr, "unquant error %d %d\n", i, level); | |
| 2652 #endif | |
| 2653 block[j] = level; | |
| 2654 } | |
| 2655 } | |
| 2656 } else { | |
| 2657 int sum=-1; | |
| 2658 i = 0; | |
| 344 | 2659 quant_matrix = s->inter_matrix; |
| 325 | 2660 for(;i<nCoeffs;i++) { |
| 2661 int j= zigzag_direct[i]; | |
| 2662 level = block[j]; | |
| 2663 if (level) { | |
| 2664 if (level < 0) { | |
| 2665 level = -level; | |
| 2666 level = (((level << 1) + 1) * qscale * | |
| 2667 ((int) (quant_matrix[j]))) >> 4; | |
| 2668 level = -level; | |
| 2669 } else { | |
| 2670 level = (((level << 1) + 1) * qscale * | |
| 2671 ((int) (quant_matrix[j]))) >> 4; | |
| 2672 } | |
| 2673 #ifdef PARANOID | |
| 2674 if (level < -2048 || level > 2047) | |
| 2675 fprintf(stderr, "unquant error %d %d\n", i, level); | |
| 2676 #endif | |
| 2677 block[j] = level; | |
| 2678 sum+=level; | |
| 2679 } | |
| 2680 } | |
| 2681 block[63]^=sum&1; | |
| 2682 } | |
| 2683 } | |
| 2684 | |
| 2685 | |
|
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
|
2686 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
|
2687 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
|
2688 { |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2689 int i, level, qmul, qadd; |
| 200 | 2690 int nCoeffs; |
|
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2691 |
|
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
|
2692 if (s->mb_intra) { |
|
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2693 if (!s->h263_aic) { |
|
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2694 if (n < 4) |
|
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2695 block[0] = block[0] * s->y_dc_scale; |
|
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2696 else |
|
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2697 block[0] = block[0] * s->c_dc_scale; |
|
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
2698 } |
|
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
|
2699 i = 1; |
| 200 | 2700 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
|
2701 } 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
|
2702 i = 0; |
| 200 | 2703 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
|
2704 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2705 |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2706 qmul = s->qscale << 1; |
|
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2707 if (s->h263_aic && s->mb_intra) |
|
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2708 qadd = 0; |
|
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2709 else |
|
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
2710 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
|
2711 |
| 200 | 2712 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
|
2713 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
|
2714 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
|
2715 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
|
2716 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
|
2717 } 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
|
2718 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
|
2719 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2720 #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
|
2721 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
|
2722 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
|
2723 #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
|
2724 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
|
2725 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2726 } |
|
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
2727 } |
| 0 | 2728 |
| 456 | 2729 static void remove_ac(MpegEncContext *s, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, int mb_x, int mb_y) |
| 2730 { | |
| 2731 int dc, dcb, dcr, y, i; | |
| 2732 for(i=0; i<4; i++){ | |
| 2733 dc= s->dc_val[0][mb_x*2+1 + (i&1) + (mb_y*2+1 + (i>>1))*(s->mb_width*2+2)]; | |
| 2734 for(y=0; y<8; y++){ | |
| 2735 int x; | |
| 2736 for(x=0; x<8; x++){ | |
| 2737 dest_y[x + (i&1)*8 + (y + (i>>1)*8)*s->linesize]= dc/8; | |
| 2738 } | |
| 2739 } | |
| 2740 } | |
| 2741 dcb = s->dc_val[1][mb_x+1 + (mb_y+1)*(s->mb_width+2)]; | |
| 2742 dcr= s->dc_val[2][mb_x+1 + (mb_y+1)*(s->mb_width+2)]; | |
| 2743 for(y=0; y<8; y++){ | |
| 2744 int x; | |
| 2745 for(x=0; x<8; x++){ | |
| 556 | 2746 dest_cb[x + y*(s->uvlinesize)]= dcb/8; |
| 2747 dest_cr[x + y*(s->uvlinesize)]= dcr/8; | |
| 456 | 2748 } |
| 2749 } | |
| 2750 } | |
| 2751 | |
| 2752 /** | |
| 2753 * will conceal past errors, and allso drop b frames if needed | |
| 2754 * | |
| 2755 */ | |
| 2756 void ff_conceal_past_errors(MpegEncContext *s, int unknown_pos) | |
| 2757 { | |
| 2758 int mb_x= s->mb_x; | |
| 2759 int mb_y= s->mb_y; | |
| 2760 int mb_dist=0; | |
| 2761 int i, intra_count=0, inter_count=0; | |
| 2762 int intra_conceal= s->msmpeg4_version ? 50 : 50; //FIXME finetune | |
| 2763 int inter_conceal= s->msmpeg4_version ? 50 : 50; | |
| 612 | 2764 |
| 456 | 2765 // for last block |
| 2766 if(mb_x>=s->mb_width) mb_x= s->mb_width -1; | |
| 2767 if(mb_y>=s->mb_height) mb_y= s->mb_height-1; | |
| 2768 | |
| 2769 if(s->decoding_error==0 && unknown_pos){ | |
| 2770 if(s->data_partitioning && s->pict_type!=B_TYPE) | |
| 2771 s->decoding_error= DECODING_AC_LOST; | |
| 2772 else | |
| 2773 s->decoding_error= DECODING_DESYNC; | |
| 2774 } | |
| 2775 | |
| 2776 if(s->decoding_error==DECODING_DESYNC && s->pict_type!=B_TYPE) s->next_p_frame_damaged=1; | |
| 2777 | |
| 2778 for(i=mb_x + mb_y*s->mb_width; i>=0; i--){ | |
| 2779 if(s->mbintra_table[i]) intra_count++; | |
| 2780 else inter_count++; | |
| 2781 } | |
| 2782 | |
| 2783 if(s->decoding_error==DECODING_AC_LOST){ | |
| 2784 intra_conceal*=2; | |
| 2785 inter_conceal*=2; | |
| 2786 }else if(s->decoding_error==DECODING_ACDC_LOST){ | |
| 2787 intra_conceal*=2; | |
| 2788 inter_conceal*=2; | |
| 2789 } | |
| 2790 | |
| 2791 if(unknown_pos && (intra_count<inter_count)){ | |
| 2792 intra_conceal= inter_conceal= s->mb_num; | |
| 2793 // printf("%d %d\n",intra_count, inter_count); | |
| 2794 } | |
| 2795 | |
| 2796 fprintf(stderr, "concealing errors\n"); | |
| 2797 | |
| 2798 /* for all MBs from the current one back until the last resync marker */ | |
| 2799 for(; mb_y>=0 && mb_y>=s->resync_mb_y; mb_y--){ | |
| 2800 for(; mb_x>=0; mb_x--){ | |
| 2801 uint8_t *dest_y = s->current_picture[0] + (mb_y * 16* s->linesize ) + mb_x * 16; | |
| 556 | 2802 uint8_t *dest_cb = s->current_picture[1] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; |
| 2803 uint8_t *dest_cr = s->current_picture[2] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8; | |
| 456 | 2804 int mb_x_backup= s->mb_x; //FIXME pass xy to mpeg_motion |
| 2805 int mb_y_backup= s->mb_y; | |
| 2806 s->mb_x=mb_x; | |
| 2807 s->mb_y=mb_y; | |
| 2808 if(s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<intra_conceal){ | |
| 2809 if(s->decoding_error==DECODING_AC_LOST){ | |
| 2810 remove_ac(s, dest_y, dest_cb, dest_cr, mb_x, mb_y); | |
| 2811 // printf("remove ac to %d %d\n", mb_x, mb_y); | |
| 2812 }else{ | |
| 2813 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 2814 s->last_picture, 0, 0, put_pixels_tab, | |
| 2815 0/*mx*/, 0/*my*/, 16); | |
| 2816 } | |
| 2817 } | |
| 2818 else if(!s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<inter_conceal){ | |
| 2819 int mx=0; | |
| 2820 int my=0; | |
| 2821 | |
| 2822 if(s->decoding_error!=DECODING_DESYNC){ | |
| 2823 int xy= mb_x*2+1 + (mb_y*2+1)*(s->mb_width*2+2); | |
| 2824 mx= s->motion_val[ xy ][0]; | |
| 2825 my= s->motion_val[ xy ][1]; | |
| 2826 } | |
| 2827 | |
| 2828 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0, | |
| 2829 s->last_picture, 0, 0, put_pixels_tab, | |
| 2830 mx, my, 16); | |
| 2831 } | |
| 2832 s->mb_x= mb_x_backup; | |
| 2833 s->mb_y= mb_y_backup; | |
| 2834 | |
| 2835 if(mb_x== s->resync_mb_x && mb_y== s->resync_mb_y) return; | |
| 2836 if(!s->mbskip_table[mb_x + mb_y*s->mb_width]) mb_dist++; | |
| 2837 } | |
| 2838 mb_x=s->mb_width-1; | |
| 2839 } | |
| 2840 } | |
| 2841 | |
| 0 | 2842 AVCodec mpeg1video_encoder = { |
| 2843 "mpeg1video", | |
| 2844 CODEC_TYPE_VIDEO, | |
| 2845 CODEC_ID_MPEG1VIDEO, | |
| 2846 sizeof(MpegEncContext), | |
| 2847 MPV_encode_init, | |
| 2848 MPV_encode_picture, | |
| 2849 MPV_encode_end, | |
| 2850 }; | |
| 2851 | |
| 2852 AVCodec h263_encoder = { | |
| 2853 "h263", | |
| 2854 CODEC_TYPE_VIDEO, | |
| 2855 CODEC_ID_H263, | |
| 2856 sizeof(MpegEncContext), | |
| 2857 MPV_encode_init, | |
| 2858 MPV_encode_picture, | |
| 2859 MPV_encode_end, | |
| 2860 }; | |
| 2861 | |
| 2862 AVCodec h263p_encoder = { | |
| 2863 "h263p", | |
| 2864 CODEC_TYPE_VIDEO, | |
| 2865 CODEC_ID_H263P, | |
| 2866 sizeof(MpegEncContext), | |
| 2867 MPV_encode_init, | |
| 2868 MPV_encode_picture, | |
| 2869 MPV_encode_end, | |
| 2870 }; | |
| 2871 | |
| 2872 AVCodec rv10_encoder = { | |
| 2873 "rv10", | |
| 2874 CODEC_TYPE_VIDEO, | |
| 2875 CODEC_ID_RV10, | |
| 2876 sizeof(MpegEncContext), | |
| 2877 MPV_encode_init, | |
| 2878 MPV_encode_picture, | |
| 2879 MPV_encode_end, | |
| 2880 }; | |
| 2881 | |
| 2882 AVCodec mjpeg_encoder = { | |
| 2883 "mjpeg", | |
| 2884 CODEC_TYPE_VIDEO, | |
| 2885 CODEC_ID_MJPEG, | |
| 2886 sizeof(MpegEncContext), | |
| 2887 MPV_encode_init, | |
| 2888 MPV_encode_picture, | |
| 2889 MPV_encode_end, | |
| 2890 }; | |
| 2891 | |
| 71 | 2892 AVCodec mpeg4_encoder = { |
| 2893 "mpeg4", | |
| 0 | 2894 CODEC_TYPE_VIDEO, |
| 71 | 2895 CODEC_ID_MPEG4, |
| 0 | 2896 sizeof(MpegEncContext), |
| 2897 MPV_encode_init, | |
| 2898 MPV_encode_picture, | |
| 2899 MPV_encode_end, | |
| 2900 }; | |
| 2901 | |
| 307 | 2902 AVCodec msmpeg4v1_encoder = { |
| 2903 "msmpeg4v1", | |
| 0 | 2904 CODEC_TYPE_VIDEO, |
| 307 | 2905 CODEC_ID_MSMPEG4V1, |
| 0 | 2906 sizeof(MpegEncContext), |
| 2907 MPV_encode_init, | |
| 2908 MPV_encode_picture, | |
| 2909 MPV_encode_end, | |
| 2910 }; | |
| 307 | 2911 |
| 2912 AVCodec msmpeg4v2_encoder = { | |
| 2913 "msmpeg4v2", | |
| 2914 CODEC_TYPE_VIDEO, | |
| 2915 CODEC_ID_MSMPEG4V2, | |
| 2916 sizeof(MpegEncContext), | |
| 2917 MPV_encode_init, | |
| 2918 MPV_encode_picture, | |
| 2919 MPV_encode_end, | |
| 2920 }; | |
| 2921 | |
| 2922 AVCodec msmpeg4v3_encoder = { | |
| 2923 "msmpeg4", | |
| 2924 CODEC_TYPE_VIDEO, | |
| 2925 CODEC_ID_MSMPEG4V3, | |
| 2926 sizeof(MpegEncContext), | |
| 2927 MPV_encode_init, | |
| 2928 MPV_encode_picture, | |
| 2929 MPV_encode_end, | |
| 2930 }; | |
| 499 | 2931 |
| 2932 AVCodec wmv1_encoder = { | |
| 2933 "wmv1", | |
| 2934 CODEC_TYPE_VIDEO, | |
| 2935 CODEC_ID_WMV1, | |
| 2936 sizeof(MpegEncContext), | |
| 2937 MPV_encode_init, | |
| 2938 MPV_encode_picture, | |
| 2939 MPV_encode_end, | |
| 2940 }; | |
| 2941 | |
| 2942 AVCodec wmv2_encoder = { | |
| 2943 "wmv2", | |
| 2944 CODEC_TYPE_VIDEO, | |
| 2945 CODEC_ID_WMV2, | |
| 2946 sizeof(MpegEncContext), | |
| 2947 MPV_encode_init, | |
| 2948 MPV_encode_picture, | |
| 2949 MPV_encode_end, | |
| 2950 }; |
