Mercurial > libavcodec.hg
annotate mpegvideo.c @ 3988:3041b7b5476c libavcodec
Move CFLAGS handling to common.mak.
author | diego |
---|---|
date | Wed, 11 Oct 2006 08:30:13 +0000 |
parents | c8c591fe26f8 |
children | 34fdffe98bd0 |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest mpeg encoder (well, it was the simplest!) | |
429 | 3 * Copyright (c) 2000,2001 Fabrice Bellard. |
1739
07a484280a82
copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents:
1729
diff
changeset
|
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
0 | 5 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3939
diff
changeset
|
6 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3939
diff
changeset
|
7 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3939
diff
changeset
|
8 * FFmpeg is free software; you can redistribute it and/or |
429 | 9 * modify it under the terms of the GNU Lesser General Public |
10 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3939
diff
changeset
|
11 * version 2.1 of the License, or (at your option) any later version. |
0 | 12 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3939
diff
changeset
|
13 * FFmpeg is distributed in the hope that it will be useful, |
0 | 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
429 | 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
16 * Lesser General Public License for more details. | |
0 | 17 * |
429 | 18 * You should have received a copy of the GNU Lesser General Public |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3939
diff
changeset
|
19 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3011
diff
changeset
|
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
295 | 21 * |
325 | 22 * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at> |
0 | 23 */ |
2967 | 24 |
1106 | 25 /** |
26 * @file mpegvideo.c | |
27 * The simplest mpeg encoder (well, it was the simplest!). | |
2967 | 28 */ |
29 | |
0 | 30 #include "avcodec.h" |
31 #include "dsputil.h" | |
32 #include "mpegvideo.h" | |
1557 | 33 #include "faandct.h" |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2015
diff
changeset
|
34 #include <limits.h> |
0 | 35 |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
36 #ifdef USE_FASTMEMCPY |
3589 | 37 #include "libvo/fastmemcpy.h" |
17
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
38 #endif |
b69fe46fd708
Adding fastmemcpy stuff to speedup mplayer project
nickols_k
parents:
13
diff
changeset
|
39 |
1422
efeed6e29f9b
oooooops, having 2 Eterms open and doing cvs diff in one and cvs commit in the other sucks, especially if they are in different directories (MPlayer/main/libavcodec vs. ffmpeg/libavcodec)
michaelni
parents:
1421
diff
changeset
|
40 //#undef NDEBUG |
efeed6e29f9b
oooooops, having 2 Eterms open and doing cvs diff in one and cvs commit in the other sucks, especially if they are in different directories (MPlayer/main/libavcodec vs. ffmpeg/libavcodec)
michaelni
parents:
1421
diff
changeset
|
41 //#include <assert.h> |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
42 |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
43 #ifdef CONFIG_ENCODERS |
3766 | 44 static int encode_picture(MpegEncContext *s, int picture_number); |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
45 #endif //CONFIG_ENCODERS |
2967 | 46 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s, |
1689 | 47 DCTELEM *block, int n, int qscale); |
2967 | 48 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s, |
1689 | 49 DCTELEM *block, int n, int qscale); |
50 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s, | |
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
|
51 DCTELEM *block, int n, int qscale); |
3281
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
52 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s, |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
53 DCTELEM *block, int n, int qscale); |
1689 | 54 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s, |
325 | 55 DCTELEM *block, int n, int qscale); |
2967 | 56 static void dct_unquantize_h263_intra_c(MpegEncContext *s, |
1689 | 57 DCTELEM *block, int n, int qscale); |
2967 | 58 static void dct_unquantize_h263_inter_c(MpegEncContext *s, |
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
|
59 DCTELEM *block, int n, int qscale); |
1064 | 60 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w); |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
61 #ifdef CONFIG_ENCODERS |
344 | 62 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow); |
945 | 63 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow); |
1777 | 64 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale); |
1389 | 65 static int sse_mb(MpegEncContext *s); |
1719 | 66 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block); |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
67 #endif //CONFIG_ENCODERS |
206 | 68 |
1381 | 69 #ifdef HAVE_XVMC |
70 extern int XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx); | |
71 extern void XVMC_field_end(MpegEncContext *s); | |
1580
628bf341e099
XvMC speedup by removing one memcpy and doing MB packing
iive
parents:
1579
diff
changeset
|
72 extern void XVMC_decode_mb(MpegEncContext *s); |
1381 | 73 #endif |
74 | |
1064 | 75 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w)= draw_edges_c; |
206 | 76 |
0 | 77 |
78 /* enable all paranoid tests for rounding, overflows, etc... */ | |
79 //#define PARANOID | |
80 | |
81 //#define DEBUG | |
82 | |
321 | 83 |
0 | 84 /* for jpeg fast DCT */ |
85 #define CONST_BITS 14 | |
86 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
87 static const uint16_t aanscales[64] = { |
0 | 88 /* precomputed values scaled up by 14 bits */ |
89 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
90 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
91 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
92 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
93 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
94 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
936 | 95 8867 , 12299, 11585, 10426, 8867, 6967, 4799, 2446, |
96 4520 , 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
0 | 97 }; |
98 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
99 static const uint8_t h263_chroma_roundtab[16] = { |
1013 | 100 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
0 | 101 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, |
102 }; | |
103 | |
1644 | 104 static const uint8_t ff_default_chroma_qscale_table[32]={ |
105 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 | |
106 0, 1, 2, 3, 4, 5, 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 | |
107 }; | |
108 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
109 #ifdef CONFIG_ENCODERS |
1162 | 110 static uint8_t (*default_mv_penalty)[MAX_MV*2+1]=NULL; |
1064 | 111 static uint8_t default_fcode_tab[MAX_MV*2+1]; |
277
5cb2978e701f
new motion estimation (epzs) not complete yet but allready pretty good :)
michaelni
parents:
271
diff
changeset
|
112 |
998 | 113 enum PixelFormat ff_yuv420p_list[2]= {PIX_FMT_YUV420P, -1}; |
114 | |
1579
039cb2296de2
make convert_matrix() independant of MpegEncContext
michael
parents:
1573
diff
changeset
|
115 static void convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64], |
2385
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
116 const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra) |
0 | 117 { |
344 | 118 int qscale; |
2385
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
119 int shift=0; |
0 | 120 |
709
afeff6ccb7f5
convert only needed matrixes in convert_matrix() (mjpeg calls it for every frame)
michaelni
parents:
706
diff
changeset
|
121 for(qscale=qmin; qscale<=qmax; qscale++){ |
344 | 122 int i; |
2967 | 123 if (dsp->fdct == ff_jpeg_fdct_islow |
1562
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
124 #ifdef FAAN_POSTSCALE |
1579
039cb2296de2
make convert_matrix() independant of MpegEncContext
michael
parents:
1573
diff
changeset
|
125 || dsp->fdct == ff_faandct |
1562
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
126 #endif |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
127 ) { |
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
|
128 for(i=0;i<64;i++) { |
1579
039cb2296de2
make convert_matrix() independant of MpegEncContext
michael
parents:
1573
diff
changeset
|
129 const int j= dsp->idct_permutation[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
|
130 /* 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
|
131 /* 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
|
132 /* (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
|
133 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ |
2967 | 134 |
135 qmat[qscale][i] = (int)((uint64_t_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
|
136 (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
|
137 } |
1579
039cb2296de2
make convert_matrix() independant of MpegEncContext
michael
parents:
1573
diff
changeset
|
138 } else if (dsp->fdct == fdct_ifast |
1562
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
139 #ifndef FAAN_POSTSCALE |
1579
039cb2296de2
make convert_matrix() independant of MpegEncContext
michael
parents:
1573
diff
changeset
|
140 || dsp->fdct == ff_faandct |
1562
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
141 #endif |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
142 ) { |
344 | 143 for(i=0;i<64;i++) { |
1579
039cb2296de2
make convert_matrix() independant of MpegEncContext
michael
parents:
1573
diff
changeset
|
144 const int j= dsp->idct_permutation[i]; |
344 | 145 /* 16 <= qscale * quant_matrix[i] <= 7905 */ |
146 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ | |
147 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ | |
148 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ | |
2967 | 149 |
150 qmat[qscale][i] = (int)((uint64_t_C(1) << (QMAT_SHIFT + 14)) / | |
344 | 151 (aanscales[i] * qscale * quant_matrix[j])); |
152 } | |
153 } else { | |
154 for(i=0;i<64;i++) { | |
1579
039cb2296de2
make convert_matrix() independant of MpegEncContext
michael
parents:
1573
diff
changeset
|
155 const int j= dsp->idct_permutation[i]; |
344 | 156 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 |
157 So 16 <= qscale * quant_matrix[i] <= 7905 | |
158 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 | |
159 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 | |
160 */ | |
1064 | 161 qmat[qscale][i] = (int)((uint64_t_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j])); |
945 | 162 // qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
1554
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
163 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]); |
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
164 |
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
165 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1; |
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
166 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]); |
344 | 167 } |
0 | 168 } |
2967 | 169 |
2385
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
170 for(i=intra; i<64; i++){ |
2390 | 171 int64_t max= 8191; |
172 if (dsp->fdct == fdct_ifast | |
173 #ifndef FAAN_POSTSCALE | |
174 || dsp->fdct == ff_faandct | |
175 #endif | |
176 ) { | |
177 max= (8191LL*aanscales[i]) >> 14; | |
178 } | |
2967 | 179 while(((max * qmat[qscale][i]) >> shift) > INT_MAX){ |
2385
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
180 shift++; |
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
181 } |
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
182 } |
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
183 } |
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
184 if(shift){ |
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
185 av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger then %d, overflows possible\n", QMAT_SHIFT - shift); |
0 | 186 } |
187 } | |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
188 |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
189 static inline void update_qscale(MpegEncContext *s){ |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
190 s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
191 s->qscale= clip(s->qscale, s->avctx->qmin, s->avctx->qmax); |
2967 | 192 |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
193 s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT; |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
194 } |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
195 #endif //CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
196 |
1273 | 197 void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){ |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
198 int i; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
199 int end; |
2967 | 200 |
764 | 201 st->scantable= src_scantable; |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
202 |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
203 for(i=0; i<64; i++){ |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
204 int j; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
205 j = src_scantable[i]; |
1273 | 206 st->permutated[i] = permutation[j]; |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
815
diff
changeset
|
207 #ifdef ARCH_POWERPC |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
815
diff
changeset
|
208 st->inverse[j] = i; |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
815
diff
changeset
|
209 #endif |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
210 } |
2967 | 211 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
212 end=-1; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
213 for(i=0; i<64; i++){ |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
214 int j; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
215 j = st->permutated[i]; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
216 if(j>end) end=j; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
217 st->raster_end[i]= end; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
218 } |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
219 } |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
220 |
1530
3b31998fe22f
disable encoders where appropriate (patch courtesy of BERO
melanson
parents:
1523
diff
changeset
|
221 #ifdef CONFIG_ENCODERS |
3776 | 222 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){ |
1411 | 223 int i; |
224 | |
225 if(matrix){ | |
226 put_bits(pb, 1, 1); | |
227 for(i=0;i<64;i++) { | |
228 put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]); | |
229 } | |
230 }else | |
231 put_bits(pb, 1, 0); | |
232 } | |
1530
3b31998fe22f
disable encoders where appropriate (patch courtesy of BERO
melanson
parents:
1523
diff
changeset
|
233 #endif //CONFIG_ENCODERS |
1411 | 234 |
3086 | 235 const uint8_t *ff_find_start_code(const uint8_t * restrict p, const uint8_t *end, uint32_t * restrict state){ |
236 int i; | |
237 | |
3092 | 238 assert(p<=end); |
239 if(p>=end) | |
240 return end; | |
241 | |
3086 | 242 for(i=0; i<3; i++){ |
243 uint32_t tmp= *state << 8; | |
244 *state= tmp + *(p++); | |
245 if(tmp == 0x100 || p==end) | |
246 return p; | |
247 } | |
248 | |
249 while(p<end){ | |
3087 | 250 if (p[-1] > 1 ) p+= 3; |
251 else if(p[-2] ) p+= 2; | |
252 else if(p[-3]|(p[-1]-1)) p++; | |
3086 | 253 else{ |
254 p++; | |
255 break; | |
256 } | |
257 } | |
258 | |
3087 | 259 p= FFMIN(p, end)-4; |
3086 | 260 *state= be2me_32(unaligned32(p)); |
261 | |
262 return p+4; | |
263 } | |
264 | |
726
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
265 /* init common dct for both encoder and decoder */ |
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
266 int DCT_common_init(MpegEncContext *s) |
0 | 267 { |
1689 | 268 s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c; |
269 s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c; | |
270 s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c; | |
271 s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c; | |
272 s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c; | |
3281
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
273 if(s->flags & CODEC_FLAG_BITEXACT) |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
274 s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact; |
1689 | 275 s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c; |
1092 | 276 |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
277 #ifdef CONFIG_ENCODERS |
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
|
278 s->dct_quantize= dct_quantize_c; |
1719 | 279 s->denoise_dct= denoise_dct_c; |
2616
ba6530ae8e34
remove superfluous CONFIG_ENCODERS (entire function is already wrapped)
melanson
parents:
2584
diff
changeset
|
280 #endif //CONFIG_ENCODERS |
2967 | 281 |
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
|
282 #ifdef HAVE_MMX |
1422
efeed6e29f9b
oooooops, having 2 Eterms open and doing cvs diff in one and cvs commit in the other sucks, especially if they are in different directories (MPlayer/main/libavcodec vs. ffmpeg/libavcodec)
michaelni
parents:
1421
diff
changeset
|
283 MPV_common_init_mmx(s); |
8 | 284 #endif |
514
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M«©ns Rullg«©rd
mellum
parents:
506
diff
changeset
|
285 #ifdef ARCH_ALPHA |
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M«©ns Rullg«©rd
mellum
parents:
506
diff
changeset
|
286 MPV_common_init_axp(s); |
c9f724e3a797
Update and activate dct_unquantize_h263_mvi. Thanks to M«©ns Rullg«©rd
mellum
parents:
506
diff
changeset
|
287 #endif |
628
f596db4aa871
sun solaris compilation bugfix, patch by (Martin Olschewski <olschewski at zpr dot uni-koeln dot de>)
michaelni
parents:
625
diff
changeset
|
288 #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
|
289 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
|
290 #endif |
721
71f669e9f633
ps2 optimizations update patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
717
diff
changeset
|
291 #ifdef HAVE_MMI |
71f669e9f633
ps2 optimizations update patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
717
diff
changeset
|
292 MPV_common_init_mmi(s); |
71f669e9f633
ps2 optimizations update patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
717
diff
changeset
|
293 #endif |
730 | 294 #ifdef ARCH_ARMV4L |
874 | 295 MPV_common_init_armv4l(s); |
730 | 296 #endif |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
815
diff
changeset
|
297 #ifdef ARCH_POWERPC |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
815
diff
changeset
|
298 MPV_common_init_ppc(s); |
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
815
diff
changeset
|
299 #endif |
730 | 300 |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
301 #ifdef CONFIG_ENCODERS |
1007 | 302 s->fast_dct_quantize= s->dct_quantize; |
303 | |
945 | 304 if(s->flags&CODEC_FLAG_TRELLIS_QUANT){ |
305 s->dct_quantize= dct_quantize_trellis_c; //move before MPV_common_init_* | |
306 } | |
307 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
308 #endif //CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
309 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
310 /* load & permutate scantables |
2967 | 311 note: only wmv uses different ones |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
312 */ |
1708 | 313 if(s->alternate_scan){ |
314 ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable , ff_alternate_vertical_scan); | |
315 ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable , ff_alternate_vertical_scan); | |
316 }else{ | |
317 ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable , ff_zigzag_direct); | |
318 ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable , ff_zigzag_direct); | |
319 } | |
1273 | 320 ff_init_scantable(s->dsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan); |
321 ff_init_scantable(s->dsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan); | |
591 | 322 |
726
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
323 return 0; |
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
324 } |
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
325 |
1539
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
326 static void copy_picture(Picture *dst, Picture *src){ |
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
327 *dst = *src; |
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
328 dst->type= FF_BUFFER_TYPE_COPY; |
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
329 } |
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
330 |
3777 | 331 #ifdef CONFIG_ENCODERS |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
332 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){ |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
333 int i; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
334 |
1702 | 335 dst->pict_type = src->pict_type; |
336 dst->quality = src->quality; | |
337 dst->coded_picture_number = src->coded_picture_number; | |
338 dst->display_picture_number = src->display_picture_number; | |
339 // dst->reference = src->reference; | |
340 dst->pts = src->pts; | |
341 dst->interlaced_frame = src->interlaced_frame; | |
342 dst->top_field_first = src->top_field_first; | |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
343 |
1945
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
344 if(s->avctx->me_threshold){ |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
345 if(!src->motion_val[0]) |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
346 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n"); |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
347 if(!src->mb_type) |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
348 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n"); |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
349 if(!src->ref_index[0]) |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
350 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n"); |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
351 if(src->motion_subsample_log2 != dst->motion_subsample_log2) |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
352 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n", |
1955
5dafb10e0252
reuse motion vectors/mb types/field select values of the source video, if the SSE for a macroblock which is predicted with these values is below me_threshold
michael
parents:
1948
diff
changeset
|
353 src->motion_subsample_log2, dst->motion_subsample_log2); |
1945
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
354 |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
355 memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0])); |
2967 | 356 |
1945
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
357 for(i=0; i<2; i++){ |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
358 int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
359 int height= ((16*s->mb_height)>>src->motion_subsample_log2); |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
360 |
1945
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
361 if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){ |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
362 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t)); |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
363 } |
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
364 if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){ |
1948 | 365 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t)); |
1945
32f68745c431
passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder
michael
parents:
1944
diff
changeset
|
366 } |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
367 } |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
368 } |
1702 | 369 } |
3777 | 370 #endif |
1702 | 371 |
903 | 372 /** |
924 | 373 * allocates a Picture |
374 * The pixels are allocated/set by calling get_buffer() if shared=0 | |
903 | 375 */ |
924 | 376 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared){ |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
377 const int big_mb_num= s->mb_stride*(s->mb_height+1) + 1; //the +1 is needed so memset(,,stride*height) doesnt sig11 |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
378 const int mb_array_size= s->mb_stride*s->mb_height; |
1655 | 379 const int b8_array_size= s->b8_stride*s->mb_height*2; |
380 const int b4_array_size= s->b4_stride*s->mb_height*4; | |
1168 | 381 int i; |
2967 | 382 |
924 | 383 if(shared){ |
384 assert(pic->data[0]); | |
385 assert(pic->type == 0 || pic->type == FF_BUFFER_TYPE_SHARED); | |
386 pic->type= FF_BUFFER_TYPE_SHARED; | |
387 }else{ | |
388 int r; | |
2967 | 389 |
924 | 390 assert(!pic->data[0]); |
2967 | 391 |
925 | 392 r= s->avctx->get_buffer(s->avctx, (AVFrame*)pic); |
2967 | 393 |
924 | 394 if(r<0 || !pic->age || !pic->type || !pic->data[0]){ |
2979 | 395 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]); |
924 | 396 return -1; |
397 } | |
398 | |
399 if(s->linesize && (s->linesize != pic->linesize[0] || s->uvlinesize != pic->linesize[1])){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
400 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (stride changed)\n"); |
924 | 401 return -1; |
402 } | |
403 | |
404 if(pic->linesize[1] != pic->linesize[2]){ | |
2764 | 405 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride mismatch)\n"); |
924 | 406 return -1; |
407 } | |
408 | |
409 s->linesize = pic->linesize[0]; | |
410 s->uvlinesize= pic->linesize[1]; | |
903 | 411 } |
2967 | 412 |
924 | 413 if(pic->qscale_table==NULL){ |
2967 | 414 if (s->encoding) { |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
415 CHECKED_ALLOCZ(pic->mb_var , mb_array_size * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
416 CHECKED_ALLOCZ(pic->mc_mb_var, mb_array_size * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
417 CHECKED_ALLOCZ(pic->mb_mean , mb_array_size * sizeof(int8_t)) |
924 | 418 } |
419 | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
420 CHECKED_ALLOCZ(pic->mbskip_table , mb_array_size * sizeof(uint8_t)+2) //the +2 is for the slice end check |
1422
efeed6e29f9b
oooooops, having 2 Eterms open and doing cvs diff in one and cvs commit in the other sucks, especially if they are in different directories (MPlayer/main/libavcodec vs. ffmpeg/libavcodec)
michaelni
parents:
1421
diff
changeset
|
421 CHECKED_ALLOCZ(pic->qscale_table , mb_array_size * sizeof(uint8_t)) |
1742 | 422 CHECKED_ALLOCZ(pic->mb_type_base , big_mb_num * sizeof(uint32_t)) |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
423 pic->mb_type= pic->mb_type_base + s->mb_stride+1; |
1168 | 424 if(s->out_format == FMT_H264){ |
425 for(i=0; i<2; i++){ | |
2570 | 426 CHECKED_ALLOCZ(pic->motion_val_base[i], 2 * (b4_array_size+4) * sizeof(int16_t)) |
427 pic->motion_val[i]= pic->motion_val_base[i]+4; | |
1948 | 428 CHECKED_ALLOCZ(pic->ref_index[i], b8_array_size * sizeof(uint8_t)) |
1655 | 429 } |
1669 | 430 pic->motion_subsample_log2= 2; |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
431 }else if(s->out_format == FMT_H263 || s->encoding || (s->avctx->debug&FF_DEBUG_MV) || (s->avctx->debug_mv)){ |
1655 | 432 for(i=0; i<2; i++){ |
2570 | 433 CHECKED_ALLOCZ(pic->motion_val_base[i], 2 * (b8_array_size+4) * sizeof(int16_t)) |
434 pic->motion_val[i]= pic->motion_val_base[i]+4; | |
1948 | 435 CHECKED_ALLOCZ(pic->ref_index[i], b8_array_size * sizeof(uint8_t)) |
1168 | 436 } |
1669 | 437 pic->motion_subsample_log2= 3; |
1168 | 438 } |
1833
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
439 if(s->avctx->debug&FF_DEBUG_DCT_COEFF) { |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
440 CHECKED_ALLOCZ(pic->dct_coeff, 64 * mb_array_size * sizeof(DCTELEM)*6) |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
441 } |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
442 pic->qstride= s->mb_stride; |
1546 | 443 CHECKED_ALLOCZ(pic->pan_scan , 1 * sizeof(AVPanScan)) |
924 | 444 } |
1168 | 445 |
1021
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
446 //it might be nicer if the application would keep track of these but it would require a API change |
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
447 memmove(s->prev_pict_types+1, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE-1); |
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
448 s->prev_pict_types[0]= s->pict_type; |
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
449 if(pic->age < PREV_PICT_TYPES_BUFFER_SIZE && s->prev_pict_types[pic->age] == B_TYPE) |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
450 pic->age= INT_MAX; // skipped MBs in b frames are quite rare in mpeg1/2 and its a bit tricky to skip them anyway |
2967 | 451 |
903 | 452 return 0; |
453 fail: //for the CHECKED_ALLOCZ macro | |
454 return -1; | |
455 } | |
456 | |
924 | 457 /** |
458 * deallocates a picture | |
459 */ | |
903 | 460 static void free_picture(MpegEncContext *s, Picture *pic){ |
461 int i; | |
924 | 462 |
463 if(pic->data[0] && pic->type!=FF_BUFFER_TYPE_SHARED){ | |
925 | 464 s->avctx->release_buffer(s->avctx, (AVFrame*)pic); |
924 | 465 } |
466 | |
903 | 467 av_freep(&pic->mb_var); |
468 av_freep(&pic->mc_mb_var); | |
469 av_freep(&pic->mb_mean); | |
470 av_freep(&pic->mbskip_table); | |
471 av_freep(&pic->qscale_table); | |
1168 | 472 av_freep(&pic->mb_type_base); |
1833
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
473 av_freep(&pic->dct_coeff); |
1546 | 474 av_freep(&pic->pan_scan); |
1168 | 475 pic->mb_type= NULL; |
476 for(i=0; i<2; i++){ | |
1655 | 477 av_freep(&pic->motion_val_base[i]); |
1168 | 478 av_freep(&pic->ref_index[i]); |
479 } | |
2967 | 480 |
1214 | 481 if(pic->type == FF_BUFFER_TYPE_SHARED){ |
924 | 482 for(i=0; i<4; i++){ |
483 pic->base[i]= | |
484 pic->data[i]= NULL; | |
485 } | |
2967 | 486 pic->type= 0; |
903 | 487 } |
488 } | |
489 | |
1799 | 490 static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base){ |
491 int i; | |
492 | |
2967 | 493 // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264) |
3316 | 494 CHECKED_ALLOCZ(s->allocated_edge_emu_buffer, (s->width+64)*2*21*2); //(width + edge + align)*interlaced*MBsize*tolerance |
495 s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21; | |
1799 | 496 |
497 //FIXME should be linesize instead of s->width*2 but that isnt known before get_buffer() | |
2967 | 498 CHECKED_ALLOCZ(s->me.scratchpad, (s->width+64)*4*16*2*sizeof(uint8_t)) |
1799 | 499 s->rd_scratchpad= s->me.scratchpad; |
500 s->b_scratchpad= s->me.scratchpad; | |
501 s->obmc_scratchpad= s->me.scratchpad + 16; | |
502 if (s->encoding) { | |
503 CHECKED_ALLOCZ(s->me.map , ME_MAP_SIZE*sizeof(uint32_t)) | |
504 CHECKED_ALLOCZ(s->me.score_map, ME_MAP_SIZE*sizeof(uint32_t)) | |
505 if(s->avctx->noise_reduction){ | |
506 CHECKED_ALLOCZ(s->dct_error_sum, 2 * 64 * sizeof(int)) | |
507 } | |
2967 | 508 } |
1848
c72589baee53
initial chroma_format changes,xvmc tweaks and codec_cap
iive
parents:
1846
diff
changeset
|
509 CHECKED_ALLOCZ(s->blocks, 64*12*2 * sizeof(DCTELEM)) |
1799 | 510 s->block= s->blocks[0]; |
511 | |
512 for(i=0;i<12;i++){ | |
513 s->pblocks[i] = (short *)(&s->block[i]); | |
514 } | |
515 return 0; | |
516 fail: | |
517 return -1; //free() through MPV_common_end() | |
518 } | |
519 | |
520 static void free_duplicate_context(MpegEncContext *s){ | |
521 if(s==NULL) return; | |
522 | |
523 av_freep(&s->allocated_edge_emu_buffer); s->edge_emu_buffer= NULL; | |
524 av_freep(&s->me.scratchpad); | |
2967 | 525 s->rd_scratchpad= |
526 s->b_scratchpad= | |
1799 | 527 s->obmc_scratchpad= NULL; |
2967 | 528 |
1799 | 529 av_freep(&s->dct_error_sum); |
530 av_freep(&s->me.map); | |
531 av_freep(&s->me.score_map); | |
532 av_freep(&s->blocks); | |
533 s->block= NULL; | |
534 } | |
535 | |
536 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src){ | |
537 #define COPY(a) bak->a= src->a | |
538 COPY(allocated_edge_emu_buffer); | |
539 COPY(edge_emu_buffer); | |
540 COPY(me.scratchpad); | |
541 COPY(rd_scratchpad); | |
542 COPY(b_scratchpad); | |
543 COPY(obmc_scratchpad); | |
544 COPY(me.map); | |
545 COPY(me.score_map); | |
546 COPY(blocks); | |
547 COPY(block); | |
548 COPY(start_mb_y); | |
549 COPY(end_mb_y); | |
550 COPY(me.map_generation); | |
551 COPY(pb); | |
552 COPY(dct_error_sum); | |
1810 | 553 COPY(dct_count[0]); |
554 COPY(dct_count[1]); | |
1799 | 555 #undef COPY |
556 } | |
557 | |
1827 | 558 void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src){ |
1799 | 559 MpegEncContext bak; |
1827 | 560 int i; |
1799 | 561 //FIXME copy only needed parts |
562 //START_TIMER | |
563 backup_duplicate_context(&bak, dst); | |
564 memcpy(dst, src, sizeof(MpegEncContext)); | |
565 backup_duplicate_context(dst, &bak); | |
1827 | 566 for(i=0;i<12;i++){ |
567 dst->pblocks[i] = (short *)(&dst->block[i]); | |
568 } | |
1799 | 569 //STOP_TIMER("update_duplicate_context") //about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads |
570 } | |
571 | |
3777 | 572 #ifdef CONFIG_ENCODERS |
1799 | 573 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){ |
574 #define COPY(a) dst->a= src->a | |
575 COPY(pict_type); | |
576 COPY(current_picture); | |
577 COPY(f_code); | |
578 COPY(b_code); | |
579 COPY(qscale); | |
580 COPY(lambda); | |
581 COPY(lambda2); | |
582 COPY(picture_in_gop_number); | |
583 COPY(gop_picture_number); | |
2764 | 584 COPY(frame_pred_frame_dct); //FIXME don't set in encode_header |
585 COPY(progressive_frame); //FIXME don't set in encode_header | |
586 COPY(partitioned_frame); //FIXME don't set in encode_header | |
1799 | 587 #undef COPY |
588 } | |
3777 | 589 #endif |
1799 | 590 |
1892 | 591 /** |
592 * sets the given MpegEncContext to common defaults (same for encoding and decoding). | |
593 * the changed fields will not depend upon the prior state of the MpegEncContext. | |
594 */ | |
595 static void MPV_common_defaults(MpegEncContext *s){ | |
596 s->y_dc_scale_table= | |
597 s->c_dc_scale_table= ff_mpeg1_dc_scale_table; | |
598 s->chroma_qscale_table= ff_default_chroma_qscale_table; | |
599 s->progressive_frame= 1; | |
600 s->progressive_sequence= 1; | |
601 s->picture_structure= PICT_FRAME; | |
602 | |
603 s->coded_picture_number = 0; | |
604 s->picture_number = 0; | |
605 s->input_picture_number = 0; | |
606 | |
607 s->picture_in_gop_number = 0; | |
1894 | 608 |
609 s->f_code = 1; | |
610 s->b_code = 1; | |
1892 | 611 } |
612 | |
613 /** | |
614 * sets the given MpegEncContext to defaults for decoding. | |
615 * the changed fields will not depend upon the prior state of the MpegEncContext. | |
616 */ | |
617 void MPV_decode_defaults(MpegEncContext *s){ | |
618 MPV_common_defaults(s); | |
619 } | |
620 | |
621 /** | |
622 * sets the given MpegEncContext to defaults for encoding. | |
623 * the changed fields will not depend upon the prior state of the MpegEncContext. | |
624 */ | |
1902
a5588762c4fa
Compile fix by (Chip <szarlada at freemail dot hu>)
michael
parents:
1894
diff
changeset
|
625 |
a5588762c4fa
Compile fix by (Chip <szarlada at freemail dot hu>)
michael
parents:
1894
diff
changeset
|
626 #ifdef CONFIG_ENCODERS |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2015
diff
changeset
|
627 static void MPV_encode_defaults(MpegEncContext *s){ |
1892 | 628 static int done=0; |
2967 | 629 |
1892 | 630 MPV_common_defaults(s); |
2967 | 631 |
1892 | 632 if(!done){ |
633 int i; | |
634 done=1; | |
635 | |
636 default_mv_penalty= av_mallocz( sizeof(uint8_t)*(MAX_FCODE+1)*(2*MAX_MV+1) ); | |
637 memset(default_fcode_tab , 0, sizeof(uint8_t)*(2*MAX_MV+1)); | |
638 | |
639 for(i=-16; i<16; i++){ | |
640 default_fcode_tab[i + MAX_MV]= 1; | |
641 } | |
642 } | |
643 s->me.mv_penalty= default_mv_penalty; | |
644 s->fcode_tab= default_fcode_tab; | |
645 } | |
1902
a5588762c4fa
Compile fix by (Chip <szarlada at freemail dot hu>)
michael
parents:
1894
diff
changeset
|
646 #endif //CONFIG_ENCODERS |
1892 | 647 |
2967 | 648 /** |
1892 | 649 * init common structure for both encoder and decoder. |
650 * this assumes that some variables like width/height are already set | |
651 */ | |
726
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
652 int MPV_common_init(MpegEncContext *s) |
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
653 { |
1708 | 654 int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y; |
726
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
655 |
3070 | 656 s->mb_height = (s->height + 15) / 16; |
657 | |
658 if(s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_height)){ | |
1989 | 659 av_log(s->avctx, AV_LOG_ERROR, "too many threads\n"); |
660 return -1; | |
661 } | |
662 | |
2422 | 663 if((s->width || s->height) && avcodec_check_dimensions(s->avctx, s->width, s->height)) |
664 return -1; | |
665 | |
1092 | 666 dsputil_init(&s->dsp, s->avctx); |
726
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
667 DCT_common_init(s); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
668 |
754 | 669 s->flags= s->avctx->flags; |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
670 s->flags2= s->avctx->flags2; |
726
a91203b34e71
moved dct init out from mpv_common_init to dct_common_init (for less-uglier way for dv)
al3x
parents:
721
diff
changeset
|
671 |
903 | 672 s->mb_width = (s->width + 15) / 16; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
673 s->mb_stride = s->mb_width + 1; |
1655 | 674 s->b8_stride = s->mb_width*2 + 1; |
675 s->b4_stride = s->mb_width*4 + 1; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
676 mb_array_size= s->mb_height * s->mb_stride; |
1708 | 677 mv_table_size= (s->mb_height+2) * s->mb_stride + 1; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
678 |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
679 /* set chroma shifts */ |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
680 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift), |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
681 &(s->chroma_y_shift) ); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
682 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
683 /* set default edge pos, will be overriden in decode_header if needed */ |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
684 s->h_edge_pos= s->mb_width*16; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
685 s->v_edge_pos= s->mb_height*16; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
686 |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
687 s->mb_num = s->mb_width * s->mb_height; |
2967 | 688 |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
689 s->block_wrap[0]= |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
690 s->block_wrap[1]= |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
691 s->block_wrap[2]= |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
692 s->block_wrap[3]= s->b8_stride; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
693 s->block_wrap[4]= |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
694 s->block_wrap[5]= s->mb_stride; |
2967 | 695 |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
696 y_size = s->b8_stride * (2 * s->mb_height + 1); |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
697 c_size = s->mb_stride * (s->mb_height + 1); |
756 | 698 yc_size = y_size + 2 * c_size; |
2967 | 699 |
701 | 700 /* convert fourcc to upper case */ |
2967 | 701 s->avctx->codec_tag= toupper( s->avctx->codec_tag &0xFF) |
1116 | 702 + (toupper((s->avctx->codec_tag>>8 )&0xFF)<<8 ) |
2967 | 703 + (toupper((s->avctx->codec_tag>>16)&0xFF)<<16) |
1116 | 704 + (toupper((s->avctx->codec_tag>>24)&0xFF)<<24); |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
705 |
2967 | 706 s->avctx->stream_codec_tag= toupper( s->avctx->stream_codec_tag &0xFF) |
1458
40b69d238beb
AVI stream header FCC / more reliable detection of old xvid files
michaelni
parents:
1451
diff
changeset
|
707 + (toupper((s->avctx->stream_codec_tag>>8 )&0xFF)<<8 ) |
2967 | 708 + (toupper((s->avctx->stream_codec_tag>>16)&0xFF)<<16) |
1458
40b69d238beb
AVI stream header FCC / more reliable detection of old xvid files
michaelni
parents:
1451
diff
changeset
|
709 + (toupper((s->avctx->stream_codec_tag>>24)&0xFF)<<24); |
40b69d238beb
AVI stream header FCC / more reliable detection of old xvid files
michaelni
parents:
1451
diff
changeset
|
710 |
925 | 711 s->avctx->coded_frame= (AVFrame*)&s->current_picture; |
903 | 712 |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
713 CHECKED_ALLOCZ(s->mb_index2xy, (s->mb_num+1)*sizeof(int)) //error ressilience code looks cleaner with this |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
714 for(y=0; y<s->mb_height; y++){ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
715 for(x=0; x<s->mb_width; x++){ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
716 s->mb_index2xy[ x + y*s->mb_width ] = x + y*s->mb_stride; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
717 } |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
718 } |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
719 s->mb_index2xy[ s->mb_height*s->mb_width ] = (s->mb_height-1)*s->mb_stride + s->mb_width; //FIXME really needed? |
2967 | 720 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
721 if (s->encoding) { |
324 | 722 /* Allocate MV tables */ |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
723 CHECKED_ALLOCZ(s->p_mv_table_base , mv_table_size * 2 * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
724 CHECKED_ALLOCZ(s->b_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
725 CHECKED_ALLOCZ(s->b_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
726 CHECKED_ALLOCZ(s->b_bidir_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
727 CHECKED_ALLOCZ(s->b_bidir_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
728 CHECKED_ALLOCZ(s->b_direct_mv_table_base , mv_table_size * 2 * sizeof(int16_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
729 s->p_mv_table = s->p_mv_table_base + s->mb_stride + 1; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
730 s->b_forw_mv_table = s->b_forw_mv_table_base + s->mb_stride + 1; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
731 s->b_back_mv_table = s->b_back_mv_table_base + s->mb_stride + 1; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
732 s->b_bidir_forw_mv_table= s->b_bidir_forw_mv_table_base + s->mb_stride + 1; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
733 s->b_bidir_back_mv_table= s->b_bidir_back_mv_table_base + s->mb_stride + 1; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
734 s->b_direct_mv_table = s->b_direct_mv_table_base + s->mb_stride + 1; |
324 | 735 |
650
ef4a33aad86e
reducing sizeof MpegEncContext to avoid stack overflow on crap M$ windo$
michaelni
parents:
635
diff
changeset
|
736 if(s->msmpeg4_version){ |
ef4a33aad86e
reducing sizeof MpegEncContext to avoid stack overflow on crap M$ windo$
michaelni
parents:
635
diff
changeset
|
737 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
|
738 } |
612 | 739 CHECKED_ALLOCZ(s->avctx->stats_out, 256); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
740 |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
741 /* Allocate MB type table */ |
1708 | 742 CHECKED_ALLOCZ(s->mb_type , mb_array_size * sizeof(uint16_t)) //needed for encoding |
2967 | 743 |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
744 CHECKED_ALLOCZ(s->lambda_table, mb_array_size * sizeof(int)) |
2967 | 745 |
1553
541681146f83
move q_*_matrix out of MpegEncContext (40k ->23k) dct_quantize() is even slightly faster now, dont ask my why ...
michael
parents:
1552
diff
changeset
|
746 CHECKED_ALLOCZ(s->q_intra_matrix, 64*32 * sizeof(int)) |
541681146f83
move q_*_matrix out of MpegEncContext (40k ->23k) dct_quantize() is even slightly faster now, dont ask my why ...
michael
parents:
1552
diff
changeset
|
747 CHECKED_ALLOCZ(s->q_inter_matrix, 64*32 * sizeof(int)) |
1554
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
748 CHECKED_ALLOCZ(s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t)) |
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
749 CHECKED_ALLOCZ(s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t)) |
1556 | 750 CHECKED_ALLOCZ(s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*)) |
751 CHECKED_ALLOCZ(s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*)) | |
2967 | 752 |
1597 | 753 if(s->avctx->noise_reduction){ |
754 CHECKED_ALLOCZ(s->dct_offset, 2 * 64 * sizeof(uint16_t)) | |
755 } | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
756 } |
1552
26c6ecba99a1
dynamic alloc of picture structs instead of putting them in MpegEncContext
michael
parents:
1546
diff
changeset
|
757 CHECKED_ALLOCZ(s->picture, MAX_PICTURE_COUNT * sizeof(Picture)) |
26c6ecba99a1
dynamic alloc of picture structs instead of putting them in MpegEncContext
michael
parents:
1546
diff
changeset
|
758 |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
759 CHECKED_ALLOCZ(s->error_status_table, mb_array_size*sizeof(uint8_t)) |
2967 | 760 |
1708 | 761 if(s->codec_id==CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)){ |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
762 /* interlaced direct mode decoding tables */ |
1708 | 763 for(i=0; i<2; i++){ |
764 int j, k; | |
765 for(j=0; j<2; j++){ | |
766 for(k=0; k<2; k++){ | |
767 CHECKED_ALLOCZ(s->b_field_mv_table_base[i][j][k] , mv_table_size * 2 * sizeof(int16_t)) | |
768 s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] + s->mb_stride + 1; | |
769 } | |
770 CHECKED_ALLOCZ(s->b_field_select_table[i][j] , mb_array_size * 2 * sizeof(uint8_t)) | |
771 CHECKED_ALLOCZ(s->p_field_mv_table_base[i][j] , mv_table_size * 2 * sizeof(int16_t)) | |
772 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1; | |
773 } | |
774 CHECKED_ALLOCZ(s->p_field_select_table[i] , mb_array_size * 2 * sizeof(uint8_t)) | |
775 } | |
0 | 776 } |
767 | 777 if (s->out_format == FMT_H263) { |
0 | 778 /* ac values */ |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
779 CHECKED_ALLOCZ(s->ac_val_base, yc_size * sizeof(int16_t) * 16); |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
780 s->ac_val[0] = s->ac_val_base + s->b8_stride + 1; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
781 s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1; |
0 | 782 s->ac_val[2] = s->ac_val[1] + c_size; |
2967 | 783 |
0 | 784 /* cbp values */ |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
785 CHECKED_ALLOCZ(s->coded_block_base, y_size); |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
786 s->coded_block= s->coded_block_base + s->b8_stride + 1; |
2967 | 787 |
456 | 788 /* cbp, ac_pred, pred_dir */ |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
789 CHECKED_ALLOCZ(s->cbp_table , mb_array_size * sizeof(uint8_t)) |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
790 CHECKED_ALLOCZ(s->pred_dir_table, mb_array_size * sizeof(uint8_t)) |
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
|
791 } |
2967 | 792 |
756 | 793 if (s->h263_pred || s->h263_plus || !s->encoding) { |
794 /* dc values */ | |
795 //MN: we need these for error resilience of intra-frames | |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
796 CHECKED_ALLOCZ(s->dc_val_base, yc_size * sizeof(int16_t)); |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
797 s->dc_val[0] = s->dc_val_base + s->b8_stride + 1; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
798 s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1; |
756 | 799 s->dc_val[2] = s->dc_val[1] + c_size; |
800 for(i=0;i<yc_size;i++) | |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
801 s->dc_val_base[i] = 1024; |
756 | 802 } |
803 | |
611
3214d3f4519e
error concealment needs the mbintra_table so it should allways be allocated
michaelni
parents:
608
diff
changeset
|
804 /* which mb is a intra block */ |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
805 CHECKED_ALLOCZ(s->mbintra_table, mb_array_size); |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
806 memset(s->mbintra_table, 1, mb_array_size); |
2967 | 807 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
808 /* init macroblock skip table */ |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
809 CHECKED_ALLOCZ(s->mbskip_table, mb_array_size+2); |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
810 //Note the +1 is for a quicker mpeg4 slice_end detection |
1021
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
811 CHECKED_ALLOCZ(s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE); |
2967 | 812 |
842
e460775adb38
cleanup (breaks compatibility, requested by fabrice)
michaelni
parents:
828
diff
changeset
|
813 s->parse_context.state= -1; |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
814 if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){ |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
815 s->visualization_buffer[0] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH); |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
816 s->visualization_buffer[1] = av_malloc((s->mb_width*8 + EDGE_WIDTH) * s->mb_height*8 + EDGE_WIDTH); |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
817 s->visualization_buffer[2] = av_malloc((s->mb_width*8 + EDGE_WIDTH) * s->mb_height*8 + EDGE_WIDTH); |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
818 } |
842
e460775adb38
cleanup (breaks compatibility, requested by fabrice)
michaelni
parents:
828
diff
changeset
|
819 |
0 | 820 s->context_initialized = 1; |
1799 | 821 |
822 s->thread_context[0]= s; | |
823 for(i=1; i<s->avctx->thread_count; i++){ | |
824 s->thread_context[i]= av_malloc(sizeof(MpegEncContext)); | |
825 memcpy(s->thread_context[i], s, sizeof(MpegEncContext)); | |
826 } | |
827 | |
828 for(i=0; i<s->avctx->thread_count; i++){ | |
829 if(init_duplicate_context(s->thread_context[i], s) < 0) | |
830 goto fail; | |
831 s->thread_context[i]->start_mb_y= (s->mb_height*(i ) + s->avctx->thread_count/2) / s->avctx->thread_count; | |
832 s->thread_context[i]->end_mb_y = (s->mb_height*(i+1) + s->avctx->thread_count/2) / s->avctx->thread_count; | |
833 } | |
834 | |
0 | 835 return 0; |
836 fail: | |
244 | 837 MPV_common_end(s); |
838 return -1; | |
839 } | |
840 | |
841 /* init common structure for both encoder and decoder */ | |
842 void MPV_common_end(MpegEncContext *s) | |
843 { | |
1708 | 844 int i, j, k; |
244 | 845 |
1799 | 846 for(i=0; i<s->avctx->thread_count; i++){ |
847 free_duplicate_context(s->thread_context[i]); | |
848 } | |
849 for(i=1; i<s->avctx->thread_count; i++){ | |
850 av_freep(&s->thread_context[i]); | |
851 } | |
852 | |
1344
447b88b784ee
memleak fix by (Jake Page <jake at CS dot Stanford dot EDU>)
michaelni
parents:
1339
diff
changeset
|
853 av_freep(&s->parse_context.buffer); |
447b88b784ee
memleak fix by (Jake Page <jake at CS dot Stanford dot EDU>)
michaelni
parents:
1339
diff
changeset
|
854 s->parse_context.buffer_size=0; |
447b88b784ee
memleak fix by (Jake Page <jake at CS dot Stanford dot EDU>)
michaelni
parents:
1339
diff
changeset
|
855 |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
856 av_freep(&s->mb_type); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
857 av_freep(&s->p_mv_table_base); |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
858 av_freep(&s->b_forw_mv_table_base); |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
859 av_freep(&s->b_back_mv_table_base); |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
860 av_freep(&s->b_bidir_forw_mv_table_base); |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
861 av_freep(&s->b_bidir_back_mv_table_base); |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
862 av_freep(&s->b_direct_mv_table_base); |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
863 s->p_mv_table= NULL; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
864 s->b_forw_mv_table= NULL; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
865 s->b_back_mv_table= NULL; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
866 s->b_bidir_forw_mv_table= NULL; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
867 s->b_bidir_back_mv_table= NULL; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
868 s->b_direct_mv_table= NULL; |
1708 | 869 for(i=0; i<2; i++){ |
870 for(j=0; j<2; j++){ | |
871 for(k=0; k<2; k++){ | |
872 av_freep(&s->b_field_mv_table_base[i][j][k]); | |
873 s->b_field_mv_table[i][j][k]=NULL; | |
874 } | |
875 av_freep(&s->b_field_select_table[i][j]); | |
876 av_freep(&s->p_field_mv_table_base[i][j]); | |
877 s->p_field_mv_table[i][j]=NULL; | |
878 } | |
879 av_freep(&s->p_field_select_table[i]); | |
880 } | |
2967 | 881 |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
882 av_freep(&s->dc_val_base); |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
883 av_freep(&s->ac_val_base); |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
884 av_freep(&s->coded_block_base); |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
885 av_freep(&s->mbintra_table); |
456 | 886 av_freep(&s->cbp_table); |
887 av_freep(&s->pred_dir_table); | |
2967 | 888 |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
889 av_freep(&s->mbskip_table); |
1021
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
890 av_freep(&s->prev_pict_types); |
396
fce0a2520551
removed useless header includes - use av memory functions
glantau
parents:
376
diff
changeset
|
891 av_freep(&s->bitstream_buffer); |
2422 | 892 s->allocated_bitstream_buffer_size=0; |
893 | |
612 | 894 av_freep(&s->avctx->stats_out); |
650
ef4a33aad86e
reducing sizeof MpegEncContext to avoid stack overflow on crap M$ windo$
michaelni
parents:
635
diff
changeset
|
895 av_freep(&s->ac_stats); |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
896 av_freep(&s->error_status_table); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
897 av_freep(&s->mb_index2xy); |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
898 av_freep(&s->lambda_table); |
1553
541681146f83
move q_*_matrix out of MpegEncContext (40k ->23k) dct_quantize() is even slightly faster now, dont ask my why ...
michael
parents:
1552
diff
changeset
|
899 av_freep(&s->q_intra_matrix); |
541681146f83
move q_*_matrix out of MpegEncContext (40k ->23k) dct_quantize() is even slightly faster now, dont ask my why ...
michael
parents:
1552
diff
changeset
|
900 av_freep(&s->q_inter_matrix); |
1554
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
901 av_freep(&s->q_intra_matrix16); |
d736e24bf303
move mmx quantizer matrixes out of MpegEncContext (23k -> 7k) (no meassureable slowdown)
michael
parents:
1553
diff
changeset
|
902 av_freep(&s->q_inter_matrix16); |
1556 | 903 av_freep(&s->input_picture); |
904 av_freep(&s->reordered_input_picture); | |
1597 | 905 av_freep(&s->dct_offset); |
903 | 906 |
1573 | 907 if(s->picture){ |
908 for(i=0; i<MAX_PICTURE_COUNT; i++){ | |
909 free_picture(s, &s->picture[i]); | |
910 } | |
0 | 911 } |
1552
26c6ecba99a1
dynamic alloc of picture structs instead of putting them in MpegEncContext
michael
parents:
1546
diff
changeset
|
912 av_freep(&s->picture); |
0 | 913 s->context_initialized = 0; |
1523 | 914 s->last_picture_ptr= |
915 s->next_picture_ptr= | |
916 s->current_picture_ptr= NULL; | |
2231 | 917 s->linesize= s->uvlinesize= 0; |
1918
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
918 |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
919 for(i=0; i<3; i++) |
1918
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
920 av_freep(&s->visualization_buffer[i]); |
2231 | 921 |
922 avcodec_default_free_buffers(s->avctx); | |
0 | 923 } |
924 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
925 #ifdef CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
926 |
0 | 927 /* init video encoder */ |
928 int MPV_encode_init(AVCodecContext *avctx) | |
929 { | |
930 MpegEncContext *s = avctx->priv_data; | |
2834 | 931 int i; |
1315
6696d3bf4ff2
lossless mjpeg encoding (planar yuv & RGB) and somerelated bugfixes
michaelni
parents:
1288
diff
changeset
|
932 int chroma_h_shift, chroma_v_shift; |
2967 | 933 |
1892 | 934 MPV_encode_defaults(s); |
315 | 935 |
3309 | 936 switch (avctx->codec_id) { |
937 case CODEC_ID_MPEG2VIDEO: | |
938 if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){ | |
939 av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n"); | |
940 return -1; | |
941 } | |
942 break; | |
943 case CODEC_ID_LJPEG: | |
944 case CODEC_ID_MJPEG: | |
3929 | 945 if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && |
946 ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){ | |
2442 | 947 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n"); |
948 return -1; | |
949 } | |
3309 | 950 break; |
951 default: | |
952 if(avctx->pix_fmt != PIX_FMT_YUV420P){ | |
953 av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n"); | |
2442 | 954 return -1; |
955 } | |
956 } | |
1402
f662e9c86cf2
* fixing a regression in mpeg encoder (not setting pix_fmt),
romansh
parents:
1397
diff
changeset
|
957 |
3309 | 958 switch (avctx->pix_fmt) { |
959 case PIX_FMT_YUVJ422P: | |
960 case PIX_FMT_YUV422P: | |
961 s->chroma_format = CHROMA_422; | |
962 break; | |
963 case PIX_FMT_YUVJ420P: | |
964 case PIX_FMT_YUV420P: | |
965 default: | |
966 s->chroma_format = CHROMA_420; | |
967 break; | |
968 } | |
969 | |
0 | 970 s->bit_rate = avctx->bit_rate; |
971 s->width = avctx->width; | |
972 s->height = avctx->height; | |
3296
d185908c71cb
allow insane gop sizes if the user doesnt care about playable streams (doesnt want compliance to any standard)
michael
parents:
3286
diff
changeset
|
973 if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){ |
2979 | 974 av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n"); |
456 | 975 avctx->gop_size=600; |
976 } | |
0 | 977 s->gop_size = avctx->gop_size; |
194
27d1773552c9
mpeg4 encoder fix by Michael Niedermayer <michaelni@gmx.at>
arpi_esp
parents:
191
diff
changeset
|
978 s->avctx = avctx; |
294 | 979 s->flags= avctx->flags; |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
980 s->flags2= avctx->flags2; |
324 | 981 s->max_b_frames= avctx->max_b_frames; |
344 | 982 s->codec_id= avctx->codec->id; |
456 | 983 s->luma_elim_threshold = avctx->luma_elim_threshold; |
984 s->chroma_elim_threshold= avctx->chroma_elim_threshold; | |
985 s->strict_std_compliance= avctx->strict_std_compliance; | |
986 s->data_partitioning= avctx->flags & CODEC_FLAG_PART; | |
936 | 987 s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0; |
599 | 988 s->mpeg_quant= avctx->mpeg_quant; |
1661 | 989 s->rtp_mode= !!avctx->rtp_payload_size; |
1992 | 990 s->intra_dc_precision= avctx->intra_dc_precision; |
2243
b8bad8bbbc64
check if the user specified timestamps are strictly monotone
michael
parents:
2235
diff
changeset
|
991 s->user_specified_pts = AV_NOPTS_VALUE; |
1422
efeed6e29f9b
oooooops, having 2 Eterms open and doing cvs diff in one and cvs commit in the other sucks, especially if they are in different directories (MPlayer/main/libavcodec vs. ffmpeg/libavcodec)
michaelni
parents:
1421
diff
changeset
|
992 |
0 | 993 if (s->gop_size <= 1) { |
994 s->intra_only = 1; | |
995 s->gop_size = 12; | |
996 } else { | |
997 s->intra_only = 0; | |
998 } | |
693
b6a7ff92df57
darkness masking (lumi masking does only bright stuff now)
michaelni
parents:
690
diff
changeset
|
999 |
936 | 1000 s->me_method = avctx->me_method; |
693
b6a7ff92df57
darkness masking (lumi masking does only bright stuff now)
michaelni
parents:
690
diff
changeset
|
1001 |
320
cda7d0857baf
- ME setting moved to AVCodecContext/MpegEncContext, no longer a global.
pulento
parents:
315
diff
changeset
|
1002 /* Fixed QSCALE */ |
1708 | 1003 s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE); |
2967 | 1004 |
690
a1c69cb685b3
adaptive quantization (lumi/temporal & spatial complexity masking)
michaelni
parents:
673
diff
changeset
|
1005 s->adaptive_quant= ( s->avctx->lumi_masking |
693
b6a7ff92df57
darkness masking (lumi masking does only bright stuff now)
michaelni
parents:
690
diff
changeset
|
1006 || s->avctx->dark_masking |
2967 | 1007 || s->avctx->temporal_cplx_masking |
690
a1c69cb685b3
adaptive quantization (lumi/temporal & spatial complexity masking)
michaelni
parents:
673
diff
changeset
|
1008 || s->avctx->spatial_cplx_masking |
1616 | 1009 || s->avctx->p_masking |
2493
258120c61eea
Border processing adaptive quant patch by (Christophe Massiot |cmassiot freebox fr)
michael
parents:
2492
diff
changeset
|
1010 || s->avctx->border_masking |
1616 | 1011 || (s->flags&CODEC_FLAG_QP_RD)) |
690
a1c69cb685b3
adaptive quantization (lumi/temporal & spatial complexity masking)
michaelni
parents:
673
diff
changeset
|
1012 && !s->fixed_qscale; |
2967 | 1013 |
1708 | 1014 s->obmc= !!(s->flags & CODEC_FLAG_OBMC); |
1015 s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER); | |
1016 s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN); | |
3305 | 1017 s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC); |
1633 | 1018 |
1711 | 1019 if(avctx->rc_max_rate && !avctx->rc_buffer_size){ |
1020 av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n"); | |
1021 return -1; | |
2967 | 1022 } |
1711 | 1023 |
1024 if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){ | |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
1025 av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n"); |
1907 | 1026 } |
2967 | 1027 |
2063 | 1028 if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){ |
1029 av_log(avctx, AV_LOG_INFO, "bitrate below min bitrate\n"); | |
1030 return -1; | |
1031 } | |
2967 | 1032 |
2063 | 1033 if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){ |
1034 av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n"); | |
1035 return -1; | |
1036 } | |
2967 | 1037 |
1038 if( s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate | |
1907 | 1039 && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO) |
1040 && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){ | |
2967 | 1041 |
1907 | 1042 av_log(avctx, AV_LOG_INFO, "Warning vbv_delay will be set to 0xFFFF (=VBR) as the specified vbv buffer is too large for the given bitrate!\n"); |
1043 } | |
2967 | 1044 |
1045 if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4 | |
1790 | 1046 && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1047 av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n"); |
1188 | 1048 return -1; |
1049 } | |
2967 | 1050 |
1633 | 1051 if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){ |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
1052 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n"); |
1633 | 1053 return -1; |
1054 } | |
2967 | 1055 |
1644 | 1056 if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){ |
1057 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n"); | |
1058 return -1; | |
1059 } | |
2967 | 1060 |
1188 | 1061 if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1062 av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n"); |
1188 | 1063 return -1; |
1064 } | |
1065 | |
1066 if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1067 av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n"); |
1188 | 1068 return -1; |
1069 } | |
2967 | 1070 |
1421 | 1071 if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1072 av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n"); |
1188 | 1073 return -1; |
1074 } | |
1958 | 1075 |
2967 | 1076 if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN)) |
1958 | 1077 && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){ |
1078 av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n"); | |
1079 return -1; | |
1080 } | |
2967 | 1081 |
1188 | 1082 if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
1083 av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n"); |
1188 | 1084 return -1; |
1085 } | |
2967 | 1086 |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
1087 if((s->flags & CODEC_FLAG_CBP_RD) && !(s->flags & CODEC_FLAG_TRELLIS_QUANT)){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1088 av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n"); |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
1089 return -1; |
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
1090 } |
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
1091 |
1622 | 1092 if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){ |
1093 av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n"); | |
1094 return -1; | |
1095 } | |
2967 | 1096 |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
1097 if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){ |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
1098 av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection arent supported yet\n"); |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
1099 return -1; |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
1100 } |
2967 | 1101 |
3305 | 1102 if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){ |
1103 av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n"); | |
1104 return -1; | |
1105 } | |
1106 | |
2967 | 1107 if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4 |
1108 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO | |
1799 | 1109 && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){ |
1110 av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n"); | |
1111 return -1; | |
1112 } | |
2967 | 1113 |
1799 | 1114 if(s->avctx->thread_count > 1) |
1115 s->rtp_mode= 1; | |
1760 | 1116 |
2637 | 1117 if(!avctx->time_base.den || !avctx->time_base.num){ |
2450 | 1118 av_log(avctx, AV_LOG_ERROR, "framerate not set\n"); |
1119 return -1; | |
1120 } | |
2967 | 1121 |
2576
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1122 i= (INT_MAX/2+128)>>8; |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1123 if(avctx->me_threshold >= i){ |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1124 av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1); |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1125 return -1; |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1126 } |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1127 if(avctx->mb_threshold >= i){ |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1128 av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1); |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1129 return -1; |
e237d9bd0f8c
check mb/me_threshold range, fixes assertion failure
michael
parents:
2570
diff
changeset
|
1130 } |
2967 | 1131 |
2668 | 1132 if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){ |
3321
8869b4dd2ffe
Make b_frame_strategy not fail on pass 2, but still print a notice.
corey
parents:
3316
diff
changeset
|
1133 av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n"); |
8869b4dd2ffe
Make b_frame_strategy not fail on pass 2, but still print a notice.
corey
parents:
3316
diff
changeset
|
1134 avctx->b_frame_strategy = 0; |
2668 | 1135 } |
1136 | |
2637 | 1137 i= ff_gcd(avctx->time_base.den, avctx->time_base.num); |
1760 | 1138 if(i > 1){ |
1139 av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n"); | |
2637 | 1140 avctx->time_base.den /= i; |
1141 avctx->time_base.num /= i; | |
1760 | 1142 // return -1; |
1143 } | |
2967 | 1144 |
1150 | 1145 if(s->codec_id==CODEC_ID_MJPEG){ |
1146 s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x | |
1147 s->inter_quant_bias= 0; | |
1421 | 1148 }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){ |
1150 | 1149 s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x |
1150 s->inter_quant_bias= 0; | |
1151 }else{ | |
1152 s->intra_quant_bias=0; | |
1153 s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x | |
1154 } | |
2967 | 1155 |
1150 | 1156 if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS) |
1157 s->intra_quant_bias= avctx->intra_quant_bias; | |
1158 if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS) | |
1159 s->inter_quant_bias= avctx->inter_quant_bias; | |
2967 | 1160 |
1315
6696d3bf4ff2
lossless mjpeg encoding (planar yuv & RGB) and somerelated bugfixes
michaelni
parents:
1288
diff
changeset
|
1161 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift); |
6696d3bf4ff2
lossless mjpeg encoding (planar yuv & RGB) and somerelated bugfixes
michaelni
parents:
1288
diff
changeset
|
1162 |
2653
2515af7d3238
complain about mpeg4 limits only if codec_id == MPEG4
michael
parents:
2639
diff
changeset
|
1163 if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){ |
2637 | 1164 av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n"); |
2967 | 1165 return -1; |
2637 | 1166 } |
1167 s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1; | |
1424 | 1168 |
0 | 1169 switch(avctx->codec->id) { |
1170 case CODEC_ID_MPEG1VIDEO: | |
1171 s->out_format = FMT_MPEG1; | |
1126
77ccf7fe3bd0
per context frame_rate_base, this should finally fix frame_rate related av sync issues
michaelni
parents:
1124
diff
changeset
|
1172 s->low_delay= 0; //s->max_b_frames ? 0 : 1; |
77ccf7fe3bd0
per context frame_rate_base, this should finally fix frame_rate related av sync issues
michaelni
parents:
1124
diff
changeset
|
1173 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); |
0 | 1174 break; |
1421 | 1175 case CODEC_ID_MPEG2VIDEO: |
1176 s->out_format = FMT_MPEG1; | |
1177 s->low_delay= 0; //s->max_b_frames ? 0 : 1; | |
1178 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); | |
1661 | 1179 s->rtp_mode= 1; |
1421 | 1180 break; |
1315
6696d3bf4ff2
lossless mjpeg encoding (planar yuv & RGB) and somerelated bugfixes
michaelni
parents:
1288
diff
changeset
|
1181 case CODEC_ID_LJPEG: |
2973 | 1182 case CODEC_ID_JPEGLS: |
0 | 1183 case CODEC_ID_MJPEG: |
1184 s->out_format = FMT_MJPEG; | |
1185 s->intra_only = 1; /* force intra only for jpeg */ | |
2973 | 1186 s->mjpeg_write_tables = avctx->codec->id != CODEC_ID_JPEGLS; |
2979 | 1187 s->mjpeg_data_only_frames = 0; /* write all the needed headers */ |
3929 | 1188 s->mjpeg_vsample[0] = 2; |
1189 s->mjpeg_vsample[1] = 2>>chroma_v_shift; | |
1190 s->mjpeg_vsample[2] = 2>>chroma_v_shift; | |
1191 s->mjpeg_hsample[0] = 2; | |
1192 s->mjpeg_hsample[1] = 2>>chroma_h_shift; | |
1193 s->mjpeg_hsample[2] = 2>>chroma_h_shift; | |
0 | 1194 if (mjpeg_init(s) < 0) |
1195 return -1; | |
336 | 1196 avctx->delay=0; |
924 | 1197 s->low_delay=1; |
0 | 1198 break; |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1199 case CODEC_ID_H261: |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1200 s->out_format = FMT_H261; |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1201 avctx->delay=0; |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1202 s->low_delay=1; |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1203 break; |
0 | 1204 case CODEC_ID_H263: |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1205 if (h263_get_picture_format(s->width, s->height) == 7) { |
3011 | 1206 av_log(avctx, AV_LOG_INFO, "The specified picture size of %dx%d is not valid for the H.263 codec.\nValid sizes are 128x96, 176x144, 352x288, 704x576, and 1408x1152. Try H.263+.\n", s->width, s->height); |
0 | 1207 return -1; |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
1208 } |
0 | 1209 s->out_format = FMT_H263; |
2979 | 1210 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0; |
336 | 1211 avctx->delay=0; |
924 | 1212 s->low_delay=1; |
0 | 1213 break; |
1214 case CODEC_ID_H263P: | |
1215 s->out_format = FMT_H263; | |
1216 s->h263_plus = 1; | |
2979 | 1217 /* Fx */ |
1644 | 1218 s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0; |
2979 | 1219 s->h263_aic= (avctx->flags & CODEC_FLAG_H263P_AIC) ? 1:0; |
1220 s->modified_quant= s->h263_aic; | |
1221 s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0; | |
1222 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0; | |
1223 s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0; | |
1224 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus; | |
1661 | 1225 s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0; |
1644 | 1226 |
2979 | 1227 /* /Fx */ |
79
82e579c37bc3
Moved some H.263+ variables to MpegEncContext to be thread-safe.
pulento
parents:
78
diff
changeset
|
1228 /* These are just to be sure */ |
336 | 1229 avctx->delay=0; |
924 | 1230 s->low_delay=1; |
0 | 1231 break; |
1353
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1232 case CODEC_ID_FLV1: |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1233 s->out_format = FMT_H263; |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1234 s->h263_flv = 2; /* format = 1; 11-bit codes */ |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1235 s->unrestricted_mv = 1; |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1236 s->rtp_mode=0; /* don't allow GOB */ |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1237 avctx->delay=0; |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1238 s->low_delay=1; |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
1239 break; |
0 | 1240 case CODEC_ID_RV10: |
1241 s->out_format = FMT_H263; | |
336 | 1242 avctx->delay=0; |
924 | 1243 s->low_delay=1; |
0 | 1244 break; |
2380 | 1245 case CODEC_ID_RV20: |
1246 s->out_format = FMT_H263; | |
1247 avctx->delay=0; | |
1248 s->low_delay=1; | |
1249 s->modified_quant=1; | |
1250 s->h263_aic=1; | |
1251 s->h263_plus=1; | |
1252 s->loop_filter=1; | |
1253 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus; | |
1254 break; | |
71 | 1255 case CODEC_ID_MPEG4: |
0 | 1256 s->out_format = FMT_H263; |
1257 s->h263_pred = 1; | |
1258 s->unrestricted_mv = 1; | |
924 | 1259 s->low_delay= s->max_b_frames ? 0 : 1; |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
1260 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); |
0 | 1261 break; |
307 | 1262 case CODEC_ID_MSMPEG4V1: |
0 | 1263 s->out_format = FMT_H263; |
1264 s->h263_msmpeg4 = 1; | |
1265 s->h263_pred = 1; | |
1266 s->unrestricted_mv = 1; | |
307 | 1267 s->msmpeg4_version= 1; |
336 | 1268 avctx->delay=0; |
924 | 1269 s->low_delay=1; |
307 | 1270 break; |
1271 case CODEC_ID_MSMPEG4V2: | |
1272 s->out_format = FMT_H263; | |
1273 s->h263_msmpeg4 = 1; | |
1274 s->h263_pred = 1; | |
1275 s->unrestricted_mv = 1; | |
1276 s->msmpeg4_version= 2; | |
336 | 1277 avctx->delay=0; |
924 | 1278 s->low_delay=1; |
307 | 1279 break; |
1280 case CODEC_ID_MSMPEG4V3: | |
1281 s->out_format = FMT_H263; | |
1282 s->h263_msmpeg4 = 1; | |
1283 s->h263_pred = 1; | |
1284 s->unrestricted_mv = 1; | |
1285 s->msmpeg4_version= 3; | |
1163 | 1286 s->flipflop_rounding=1; |
336 | 1287 avctx->delay=0; |
924 | 1288 s->low_delay=1; |
0 | 1289 break; |
499 | 1290 case CODEC_ID_WMV1: |
1291 s->out_format = FMT_H263; | |
1292 s->h263_msmpeg4 = 1; | |
1293 s->h263_pred = 1; | |
1294 s->unrestricted_mv = 1; | |
1295 s->msmpeg4_version= 4; | |
1163 | 1296 s->flipflop_rounding=1; |
499 | 1297 avctx->delay=0; |
924 | 1298 s->low_delay=1; |
499 | 1299 break; |
1300 case CODEC_ID_WMV2: | |
1301 s->out_format = FMT_H263; | |
1302 s->h263_msmpeg4 = 1; | |
1303 s->h263_pred = 1; | |
1304 s->unrestricted_mv = 1; | |
1305 s->msmpeg4_version= 5; | |
1163 | 1306 s->flipflop_rounding=1; |
499 | 1307 avctx->delay=0; |
924 | 1308 s->low_delay=1; |
499 | 1309 break; |
0 | 1310 default: |
1311 return -1; | |
1312 } | |
2967 | 1313 |
2042 | 1314 avctx->has_b_frames= !s->low_delay; |
1749
7b11032507e2
removing broken framerate conversation hack in mpeg1/2
michael
parents:
1742
diff
changeset
|
1315 |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1316 s->encoding = 1; |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
1317 |
0 | 1318 /* init */ |
1319 if (MPV_common_init(s) < 0) | |
1320 return -1; | |
1678 | 1321 |
1322 if(s->modified_quant) | |
1323 s->chroma_qscale_table= ff_h263_chroma_qscale_table; | |
2967 | 1324 s->progressive_frame= |
1708 | 1325 s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME)); |
1799 | 1326 s->quant_precision=5; |
2967 | 1327 |
1729 | 1328 ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp); |
2382 | 1329 ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp); |
2967 | 1330 |
2639 | 1331 #ifdef CONFIG_H261_ENCODER |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1332 if (s->out_format == FMT_H261) |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1333 ff_h261_encode_init(s); |
2639 | 1334 #endif |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1335 if (s->out_format == FMT_H263) |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1336 h263_encode_init(s); |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1337 if(s->msmpeg4_version) |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1338 ff_msmpeg4_encode_init(s); |
1042 | 1339 if (s->out_format == FMT_MPEG1) |
1340 ff_mpeg1_encode_init(s); | |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1341 |
1892 | 1342 /* init q matrix */ |
60 | 1343 for(i=0;i<64;i++) { |
1092 | 1344 int j= s->dsp.idct_permutation[i]; |
599 | 1345 if(s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){ |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1346 s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i]; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1347 s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i]; |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
1348 }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){ |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1349 s->intra_matrix[j] = |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1350 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i]; |
1042 | 1351 }else |
1421 | 1352 { /* mpeg1/2 */ |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1353 s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i]; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
1354 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i]; |
599 | 1355 } |
1411 | 1356 if(s->avctx->intra_matrix) |
1357 s->intra_matrix[j] = s->avctx->intra_matrix[i]; | |
1358 if(s->avctx->inter_matrix) | |
1359 s->inter_matrix[j] = s->avctx->inter_matrix[i]; | |
344 | 1360 } |
1361 | |
1362 /* precompute matrix */ | |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
1363 /* for mjpeg, we do include qscale in the matrix */ |
344 | 1364 if (s->out_format != FMT_MJPEG) { |
2967 | 1365 convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16, |
2385
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
1366 s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1); |
2967 | 1367 convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16, |
2385
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
1368 s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0); |
60 | 1369 } |
1370 | |
329 | 1371 if(ff_rate_control_init(s) < 0) |
1372 return -1; | |
2967 | 1373 |
0 | 1374 return 0; |
1375 } | |
1376 | |
1377 int MPV_encode_end(AVCodecContext *avctx) | |
1378 { | |
1379 MpegEncContext *s = avctx->priv_data; | |
1380 | |
329 | 1381 ff_rate_control_uninit(s); |
1382 | |
0 | 1383 MPV_common_end(s); |
1384 if (s->out_format == FMT_MJPEG) | |
1385 mjpeg_close(s); | |
1458
40b69d238beb
AVI stream header FCC / more reliable detection of old xvid files
michaelni
parents:
1451
diff
changeset
|
1386 |
1424 | 1387 av_freep(&avctx->extradata); |
2967 | 1388 |
0 | 1389 return 0; |
1390 } | |
1391 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
1392 #endif //CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
1393 |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1394 void init_rl(RLTable *rl, int use_static) |
1042 | 1395 { |
1064 | 1396 int8_t max_level[MAX_RUN+1], max_run[MAX_LEVEL+1]; |
1397 uint8_t index_run[MAX_RUN+1]; | |
1042 | 1398 int last, run, level, start, end, i; |
1399 | |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1400 /* If table is static, we can quit if rl->max_level[0] is not NULL */ |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1401 if(use_static && rl->max_level[0]) |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1402 return; |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1403 |
1042 | 1404 /* compute max_level[], max_run[] and index_run[] */ |
1405 for(last=0;last<2;last++) { | |
1406 if (last == 0) { | |
1407 start = 0; | |
1408 end = rl->last; | |
1409 } else { | |
1410 start = rl->last; | |
1411 end = rl->n; | |
1412 } | |
1413 | |
1414 memset(max_level, 0, MAX_RUN + 1); | |
1415 memset(max_run, 0, MAX_LEVEL + 1); | |
1416 memset(index_run, rl->n, MAX_RUN + 1); | |
1417 for(i=start;i<end;i++) { | |
1418 run = rl->table_run[i]; | |
1419 level = rl->table_level[i]; | |
1420 if (index_run[run] == rl->n) | |
1421 index_run[run] = i; | |
1422 if (level > max_level[run]) | |
1423 max_level[run] = level; | |
1424 if (run > max_run[level]) | |
1425 max_run[level] = run; | |
1426 } | |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1427 if(use_static) |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1428 rl->max_level[last] = av_mallocz_static(MAX_RUN + 1); |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1429 else |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1430 rl->max_level[last] = av_malloc(MAX_RUN + 1); |
1042 | 1431 memcpy(rl->max_level[last], max_level, MAX_RUN + 1); |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1432 if(use_static) |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1433 rl->max_run[last] = av_mallocz_static(MAX_LEVEL + 1); |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1434 else |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1435 rl->max_run[last] = av_malloc(MAX_LEVEL + 1); |
1042 | 1436 memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1); |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1437 if(use_static) |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1438 rl->index_run[last] = av_mallocz_static(MAX_RUN + 1); |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1439 else |
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2364
diff
changeset
|
1440 rl->index_run[last] = av_malloc(MAX_RUN + 1); |
1042 | 1441 memcpy(rl->index_run[last], index_run, MAX_RUN + 1); |
1442 } | |
1443 } | |
1444 | |
0 | 1445 /* 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
|
1446 //FIXME check that this is ok for mpeg4 interlaced |
1064 | 1447 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w) |
0 | 1448 { |
1064 | 1449 uint8_t *ptr, *last_line; |
0 | 1450 int i; |
1451 | |
1452 last_line = buf + (height - 1) * wrap; | |
1453 for(i=0;i<w;i++) { | |
1454 /* top and bottom */ | |
1455 memcpy(buf - (i + 1) * wrap, buf, width); | |
1456 memcpy(last_line + (i + 1) * wrap, last_line, width); | |
1457 } | |
1458 /* left and right */ | |
1459 ptr = buf; | |
1460 for(i=0;i<height;i++) { | |
1461 memset(ptr - w, ptr[0], w); | |
1462 memset(ptr + width, ptr[width-1], w); | |
1463 ptr += wrap; | |
1464 } | |
1465 /* corners */ | |
1466 for(i=0;i<w;i++) { | |
1467 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */ | |
1468 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */ | |
1469 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */ | |
1470 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */ | |
1471 } | |
1472 } | |
1473 | |
1586 | 1474 int ff_find_unused_picture(MpegEncContext *s, int shared){ |
924 | 1475 int i; |
2967 | 1476 |
924 | 1477 if(shared){ |
1478 for(i=0; i<MAX_PICTURE_COUNT; i++){ | |
1586 | 1479 if(s->picture[i].data[0]==NULL && s->picture[i].type==0) return i; |
924 | 1480 } |
1481 }else{ | |
1482 for(i=0; i<MAX_PICTURE_COUNT; i++){ | |
1586 | 1483 if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) return i; //FIXME |
924 | 1484 } |
1485 for(i=0; i<MAX_PICTURE_COUNT; i++){ | |
1586 | 1486 if(s->picture[i].data[0]==NULL) return i; |
924 | 1487 } |
1488 } | |
1489 | |
1586 | 1490 assert(0); |
1491 return -1; | |
924 | 1492 } |
1493 | |
1597 | 1494 static void update_noise_reduction(MpegEncContext *s){ |
1495 int intra, i; | |
1496 | |
1497 for(intra=0; intra<2; intra++){ | |
1498 if(s->dct_count[intra] > (1<<16)){ | |
1499 for(i=0; i<64; i++){ | |
1500 s->dct_error_sum[intra][i] >>=1; | |
1501 } | |
1502 s->dct_count[intra] >>= 1; | |
1503 } | |
2967 | 1504 |
1597 | 1505 for(i=0; i<64; i++){ |
1506 s->dct_offset[intra][i]= (s->avctx->noise_reduction * s->dct_count[intra] + s->dct_error_sum[intra][i]/2) / (s->dct_error_sum[intra][i]+1); | |
1507 } | |
1508 } | |
1509 } | |
1510 | |
1586 | 1511 /** |
1512 * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded | |
1513 */ | |
771
d4cc92144266
handle direct rendering buffer allocation failure
michaelni
parents:
768
diff
changeset
|
1514 int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx) |
0 | 1515 { |
924 | 1516 int i; |
925 | 1517 AVFrame *pic; |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
1518 s->mb_skipped = 0; |
1168 | 1519 |
1234 | 1520 assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3); |
1228 | 1521 |
903 | 1522 /* mark&release old frames */ |
1944 | 1523 if (s->pict_type != B_TYPE && s->last_picture_ptr && s->last_picture_ptr != s->next_picture_ptr && s->last_picture_ptr->data[0]) { |
1138 | 1524 avctx->release_buffer(avctx, (AVFrame*)s->last_picture_ptr); |
903 | 1525 |
1526 /* release forgotten pictures */ | |
1527 /* if(mpeg124/h263) */ | |
1528 if(!s->encoding){ | |
1529 for(i=0; i<MAX_PICTURE_COUNT; i++){ | |
1138 | 1530 if(s->picture[i].data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].reference){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1531 av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n"); |
2967 | 1532 avctx->release_buffer(avctx, (AVFrame*)&s->picture[i]); |
903 | 1533 } |
1534 } | |
1535 } | |
1536 } | |
910
eb448df811be
fixing files where the first frame isn a keyframe
michaelni
parents:
909
diff
changeset
|
1537 alloc: |
eb448df811be
fixing files where the first frame isn a keyframe
michaelni
parents:
909
diff
changeset
|
1538 if(!s->encoding){ |
2764 | 1539 /* release non reference frames */ |
1228 | 1540 for(i=0; i<MAX_PICTURE_COUNT; i++){ |
1541 if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){ | |
1542 s->avctx->release_buffer(s->avctx, (AVFrame*)&s->picture[i]); | |
1543 } | |
1544 } | |
1545 | |
1586 | 1546 if(s->current_picture_ptr && s->current_picture_ptr->data[0]==NULL) |
1547 pic= (AVFrame*)s->current_picture_ptr; //we allready have a unused image (maybe it was set before reading the header) | |
1548 else{ | |
1549 i= ff_find_unused_picture(s, 0); | |
1550 pic= (AVFrame*)&s->picture[i]; | |
1551 } | |
1552 | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2534
diff
changeset
|
1553 pic->reference= (s->pict_type != B_TYPE || s->codec_id == CODEC_ID_H264) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2534
diff
changeset
|
1554 && !s->dropable ? 3 : 0; |
1138 | 1555 |
1705 | 1556 pic->coded_picture_number= s->coded_picture_number++; |
2967 | 1557 |
1384 | 1558 if( alloc_picture(s, (Picture*)pic, 0) < 0) |
1559 return -1; | |
903 | 1560 |
1586 | 1561 s->current_picture_ptr= (Picture*)pic; |
1658 | 1562 s->current_picture_ptr->top_field_first= s->top_field_first; //FIXME use only the vars from current_pic |
1659 | 1563 s->current_picture_ptr->interlaced_frame= !s->progressive_frame && !s->progressive_sequence; |
903 | 1564 } |
456 | 1565 |
1173 | 1566 s->current_picture_ptr->pict_type= s->pict_type; |
2967 | 1567 // if(s->flags && CODEC_FLAG_QSCALE) |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
1568 // s->current_picture_ptr->quality= s->new_picture_ptr->quality; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1569 s->current_picture_ptr->key_frame= s->pict_type == I_TYPE; |
1173 | 1570 |
1539
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
1571 copy_picture(&s->current_picture, s->current_picture_ptr); |
2967 | 1572 |
1234 | 1573 if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){ |
903 | 1574 if (s->pict_type != B_TYPE) { |
1138 | 1575 s->last_picture_ptr= s->next_picture_ptr; |
1944 | 1576 if(!s->dropable) |
1577 s->next_picture_ptr= s->current_picture_ptr; | |
1578 } | |
1579 /* av_log(s->avctx, AV_LOG_DEBUG, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n", s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr, | |
2967 | 1580 s->last_picture_ptr ? s->last_picture_ptr->data[0] : NULL, |
1581 s->next_picture_ptr ? s->next_picture_ptr->data[0] : NULL, | |
1944 | 1582 s->current_picture_ptr ? s->current_picture_ptr->data[0] : NULL, |
1583 s->pict_type, s->dropable);*/ | |
2967 | 1584 |
1539
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
1585 if(s->last_picture_ptr) copy_picture(&s->last_picture, s->last_picture_ptr); |
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
1586 if(s->next_picture_ptr) copy_picture(&s->next_picture, s->next_picture_ptr); |
2967 | 1587 |
1539
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
1588 if(s->pict_type != I_TYPE && (s->last_picture_ptr==NULL || s->last_picture_ptr->data[0]==NULL)){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1589 av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n"); |
2764 | 1590 assert(s->pict_type != B_TYPE); //these should have been dropped if we don't have a reference |
1393 | 1591 goto alloc; |
1592 } | |
1593 | |
1594 assert(s->pict_type == I_TYPE || (s->last_picture_ptr && s->last_picture_ptr->data[0])); | |
1595 | |
1138 | 1596 if(s->picture_structure!=PICT_FRAME){ |
1597 int i; | |
1598 for(i=0; i<4; i++){ | |
1599 if(s->picture_structure == PICT_BOTTOM_FIELD){ | |
1600 s->current_picture.data[i] += s->current_picture.linesize[i]; | |
2967 | 1601 } |
1138 | 1602 s->current_picture.linesize[i] *= 2; |
1603 s->last_picture.linesize[i] *=2; | |
1604 s->next_picture.linesize[i] *=2; | |
1605 } | |
553 | 1606 } |
1168 | 1607 } |
2967 | 1608 |
910
eb448df811be
fixing files where the first frame isn a keyframe
michaelni
parents:
909
diff
changeset
|
1609 s->hurry_up= s->avctx->hurry_up; |
eb448df811be
fixing files where the first frame isn a keyframe
michaelni
parents:
909
diff
changeset
|
1610 s->error_resilience= avctx->error_resilience; |
eb448df811be
fixing files where the first frame isn a keyframe
michaelni
parents:
909
diff
changeset
|
1611 |
2764 | 1612 /* set dequantizer, we can't do it during init as it might change for mpeg4 |
1613 and we can't do it in the header decode as init isnt called for mpeg4 there yet */ | |
1689 | 1614 if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO){ |
1615 s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra; | |
1616 s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter; | |
2345 | 1617 }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){ |
1689 | 1618 s->dct_unquantize_intra = s->dct_unquantize_h263_intra; |
1619 s->dct_unquantize_inter = s->dct_unquantize_h263_inter; | |
1620 }else{ | |
1621 s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra; | |
1622 s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter; | |
1623 } | |
771
d4cc92144266
handle direct rendering buffer allocation failure
michaelni
parents:
768
diff
changeset
|
1624 |
1597 | 1625 if(s->dct_error_sum){ |
1626 assert(s->avctx->noise_reduction && s->encoding); | |
1627 | |
1628 update_noise_reduction(s); | |
1629 } | |
2967 | 1630 |
1381 | 1631 #ifdef HAVE_XVMC |
1632 if(s->avctx->xvmc_acceleration) | |
1633 return XVMC_field_start(s, avctx); | |
1634 #endif | |
771
d4cc92144266
handle direct rendering buffer allocation failure
michaelni
parents:
768
diff
changeset
|
1635 return 0; |
0 | 1636 } |
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
|
1637 |
0 | 1638 /* generic function for encode/decode called after a frame has been coded/decoded */ |
1639 void MPV_frame_end(MpegEncContext *s) | |
1640 { | |
903 | 1641 int i; |
0 | 1642 /* draw edge for correct motion prediction if outside */ |
1381 | 1643 #ifdef HAVE_XVMC |
1644 //just to make sure that all data is rendered. | |
1645 if(s->avctx->xvmc_acceleration){ | |
1646 XVMC_field_end(s); | |
1647 }else | |
1648 #endif | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2534
diff
changeset
|
1649 if(s->unrestricted_mv && s->current_picture.reference && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) { |
903 | 1650 draw_edges(s->current_picture.data[0], s->linesize , s->h_edge_pos , s->v_edge_pos , EDGE_WIDTH ); |
1651 draw_edges(s->current_picture.data[1], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2); | |
1652 draw_edges(s->current_picture.data[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2); | |
0 | 1653 } |
207 | 1654 emms_c(); |
2967 | 1655 |
612 | 1656 s->last_pict_type = s->pict_type; |
2997
188f4ef688f2
second try of improved lambda/qp guessing for motion estimation RD in first pass
michael
parents:
2990
diff
changeset
|
1657 s->last_lambda_for[s->pict_type]= s->current_picture_ptr->quality; |
329 | 1658 if(s->pict_type!=B_TYPE){ |
1659 s->last_non_b_pict_type= s->pict_type; | |
1660 } | |
1138 | 1661 #if 0 |
1662 /* copy back current_picture variables */ | |
903 | 1663 for(i=0; i<MAX_PICTURE_COUNT; i++){ |
1664 if(s->picture[i].data[0] == s->current_picture.data[0]){ | |
1665 s->picture[i]= s->current_picture; | |
1666 break; | |
2967 | 1667 } |
903 | 1668 } |
1669 assert(i<MAX_PICTURE_COUNT); | |
2967 | 1670 #endif |
903 | 1671 |
1228 | 1672 if(s->encoding){ |
2764 | 1673 /* release non-reference frames */ |
1228 | 1674 for(i=0; i<MAX_PICTURE_COUNT; i++){ |
1675 if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){ | |
1676 s->avctx->release_buffer(s->avctx, (AVFrame*)&s->picture[i]); | |
1677 } | |
1678 } | |
324 | 1679 } |
1138 | 1680 // clear copies, to avoid confusion |
1681 #if 0 | |
1682 memset(&s->last_picture, 0, sizeof(Picture)); | |
1683 memset(&s->next_picture, 0, sizeof(Picture)); | |
1684 memset(&s->current_picture, 0, sizeof(Picture)); | |
1685 #endif | |
2213 | 1686 s->avctx->coded_frame= (AVFrame*)s->current_picture_ptr; |
903 | 1687 } |
1688 | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1689 /** |
1285 | 1690 * draws an line from (ex, ey) -> (sx, sy). |
1691 * @param w width of the image | |
1692 * @param h height of the image | |
1693 * @param stride stride/linesize of the image | |
1694 * @param color color of the arrow | |
1695 */ | |
1696 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){ | |
3595 | 1697 int x, y, fr, f; |
2967 | 1698 |
1285 | 1699 sx= clip(sx, 0, w-1); |
1700 sy= clip(sy, 0, h-1); | |
1701 ex= clip(ex, 0, w-1); | |
1702 ey= clip(ey, 0, h-1); | |
2967 | 1703 |
1285 | 1704 buf[sy*stride + sx]+= color; |
2967 | 1705 |
1285 | 1706 if(ABS(ex - sx) > ABS(ey - sy)){ |
1707 if(sx > ex){ | |
3595 | 1708 SWAP(int, sx, ex); |
1709 SWAP(int, sy, ey); | |
1285 | 1710 } |
1711 buf+= sx + sy*stride; | |
1712 ex-= sx; | |
1713 f= ((ey-sy)<<16)/ex; | |
1714 for(x= 0; x <= ex; x++){ | |
1918
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1715 y = (x*f)>>16; |
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1716 fr= (x*f)&0xFFFF; |
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1717 buf[ y *stride + x]+= (color*(0x10000-fr))>>16; |
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1718 buf[(y+1)*stride + x]+= (color* fr )>>16; |
1285 | 1719 } |
1720 }else{ | |
1721 if(sy > ey){ | |
3595 | 1722 SWAP(int, sx, ex); |
1723 SWAP(int, sy, ey); | |
1285 | 1724 } |
1725 buf+= sx + sy*stride; | |
1726 ey-= sy; | |
1727 if(ey) f= ((ex-sx)<<16)/ey; | |
1728 else f= 0; | |
1729 for(y= 0; y <= ey; y++){ | |
1918
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1730 x = (y*f)>>16; |
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1731 fr= (y*f)&0xFFFF; |
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1732 buf[y*stride + x ]+= (color*(0x10000-fr))>>16;; |
ea20f38b8d57
nicer looking arrows for the motion vector vissualization
michael
parents:
1916
diff
changeset
|
1733 buf[y*stride + x+1]+= (color* fr )>>16;; |
1285 | 1734 } |
1735 } | |
1736 } | |
1737 | |
1738 /** | |
1739 * draws an arrow from (ex, ey) -> (sx, sy). | |
1740 * @param w width of the image | |
1741 * @param h height of the image | |
1742 * @param stride stride/linesize of the image | |
1743 * @param color color of the arrow | |
1744 */ | |
2967 | 1745 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){ |
1285 | 1746 int dx,dy; |
1747 | |
1748 sx= clip(sx, -100, w+100); | |
1749 sy= clip(sy, -100, h+100); | |
1750 ex= clip(ex, -100, w+100); | |
1751 ey= clip(ey, -100, h+100); | |
2967 | 1752 |
1285 | 1753 dx= ex - sx; |
1754 dy= ey - sy; | |
2967 | 1755 |
1285 | 1756 if(dx*dx + dy*dy > 3*3){ |
1757 int rx= dx + dy; | |
1758 int ry= -dx + dy; | |
1759 int length= ff_sqrt((rx*rx + ry*ry)<<8); | |
2967 | 1760 |
1285 | 1761 //FIXME subpixel accuracy |
1762 rx= ROUNDED_DIV(rx*3<<4, length); | |
1763 ry= ROUNDED_DIV(ry*3<<4, length); | |
2967 | 1764 |
1285 | 1765 draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color); |
1766 draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color); | |
1767 } | |
1768 draw_line(buf, sx, sy, ex, ey, w, h, stride, color); | |
1769 } | |
1770 | |
1771 /** | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1772 * prints debuging info for the given picture. |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1773 */ |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1774 void ff_print_debug_info(MpegEncContext *s, AVFrame *pict){ |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1775 |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1776 if(!pict || !pict->mb_type) return; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1777 |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1778 if(s->avctx->debug&(FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)){ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1779 int x,y; |
2967 | 1780 |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1781 av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: "); |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1782 switch (pict->pict_type) { |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1783 case FF_I_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"I\n"); break; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1784 case FF_P_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"P\n"); break; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1785 case FF_B_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"B\n"); break; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1786 case FF_S_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"S\n"); break; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1787 case FF_SI_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SI\n"); break; |
2967 | 1788 case FF_SP_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SP\n"); break; |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1789 } |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1790 for(y=0; y<s->mb_height; y++){ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1791 for(x=0; x<s->mb_width; x++){ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1792 if(s->avctx->debug&FF_DEBUG_SKIP){ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1793 int count= s->mbskip_table[x + y*s->mb_stride]; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1794 if(count>9) count=9; |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1795 av_log(s->avctx, AV_LOG_DEBUG, "%1d", count); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1796 } |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1797 if(s->avctx->debug&FF_DEBUG_QP){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1798 av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1799 } |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1800 if(s->avctx->debug&FF_DEBUG_MB_TYPE){ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1801 int mb_type= pict->mb_type[x + y*s->mb_stride]; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1802 //Type & MV direction |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1803 if(IS_PCM(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1804 av_log(s->avctx, AV_LOG_DEBUG, "P"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1805 else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1806 av_log(s->avctx, AV_LOG_DEBUG, "A"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1807 else if(IS_INTRA4x4(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1808 av_log(s->avctx, AV_LOG_DEBUG, "i"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1809 else if(IS_INTRA16x16(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1810 av_log(s->avctx, AV_LOG_DEBUG, "I"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1811 else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1812 av_log(s->avctx, AV_LOG_DEBUG, "d"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1813 else if(IS_DIRECT(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1814 av_log(s->avctx, AV_LOG_DEBUG, "D"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1815 else if(IS_GMC(mb_type) && IS_SKIP(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1816 av_log(s->avctx, AV_LOG_DEBUG, "g"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1817 else if(IS_GMC(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1818 av_log(s->avctx, AV_LOG_DEBUG, "G"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1819 else if(IS_SKIP(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1820 av_log(s->avctx, AV_LOG_DEBUG, "S"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1821 else if(!USES_LIST(mb_type, 1)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1822 av_log(s->avctx, AV_LOG_DEBUG, ">"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1823 else if(!USES_LIST(mb_type, 0)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1824 av_log(s->avctx, AV_LOG_DEBUG, "<"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1825 else{ |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1826 assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1)); |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1827 av_log(s->avctx, AV_LOG_DEBUG, "X"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1828 } |
2967 | 1829 |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1830 //segmentation |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1831 if(IS_8X8(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1832 av_log(s->avctx, AV_LOG_DEBUG, "+"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1833 else if(IS_16X8(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1834 av_log(s->avctx, AV_LOG_DEBUG, "-"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1835 else if(IS_8X16(mb_type)) |
2677 | 1836 av_log(s->avctx, AV_LOG_DEBUG, "|"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1837 else if(IS_INTRA(mb_type) || IS_16X16(mb_type)) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1838 av_log(s->avctx, AV_LOG_DEBUG, " "); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1839 else |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1840 av_log(s->avctx, AV_LOG_DEBUG, "?"); |
2967 | 1841 |
1842 | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1843 if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1844 av_log(s->avctx, AV_LOG_DEBUG, "="); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1845 else |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1846 av_log(s->avctx, AV_LOG_DEBUG, " "); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1847 } |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1848 // av_log(s->avctx, AV_LOG_DEBUG, " "); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1849 } |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
1850 av_log(s->avctx, AV_LOG_DEBUG, "\n"); |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1851 } |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
1852 } |
1668
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
1853 |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1854 if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){ |
1285 | 1855 const int shift= 1 + s->quarter_sample; |
1856 int mb_y; | |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1857 uint8_t *ptr; |
1707 | 1858 int i; |
1723
033d889d7c2c
non YV12 vissualization fix by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1722
diff
changeset
|
1859 int h_chroma_shift, v_chroma_shift; |
2266
514949de5d15
set AVCodecContext.width/height to the picture width/height instead of the one stored in the bitstream (that only matters if lowres!=0)
michael
parents:
2265
diff
changeset
|
1860 const int width = s->avctx->width; |
514949de5d15
set AVCodecContext.width/height to the picture width/height instead of the one stored in the bitstream (that only matters if lowres!=0)
michael
parents:
2265
diff
changeset
|
1861 const int height= s->avctx->height; |
2395 | 1862 const int mv_sample_log2= 4 - pict->motion_subsample_log2; |
3268 | 1863 const int mv_stride= (s->mb_width << mv_sample_log2) + (s->codec_id == CODEC_ID_H264 ? 0 : 1); |
1285 | 1864 s->low_delay=0; //needed to see the vectors without trashing the buffers |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1865 |
1723
033d889d7c2c
non YV12 vissualization fix by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1722
diff
changeset
|
1866 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift); |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1867 for(i=0; i<3; i++){ |
2266
514949de5d15
set AVCodecContext.width/height to the picture width/height instead of the one stored in the bitstream (that only matters if lowres!=0)
michael
parents:
2265
diff
changeset
|
1868 memcpy(s->visualization_buffer[i], pict->data[i], (i==0) ? pict->linesize[i]*height:pict->linesize[i]*height >> v_chroma_shift); |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1869 pict->data[i]= s->visualization_buffer[i]; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1870 } |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1871 pict->type= FF_BUFFER_TYPE_COPY; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1872 ptr= pict->data[0]; |
1285 | 1873 |
1874 for(mb_y=0; mb_y<s->mb_height; mb_y++){ | |
1875 int mb_x; | |
1876 for(mb_x=0; mb_x<s->mb_width; mb_x++){ | |
1877 const int mb_index= mb_x + mb_y*s->mb_stride; | |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1878 if((s->avctx->debug_mv) && pict->motion_val){ |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1879 int type; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1880 for(type=0; type<3; type++){ |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2015
diff
changeset
|
1881 int direction = 0; |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1882 switch (type) { |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1883 case 0: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_P_FOR)) || (pict->pict_type!=FF_P_TYPE)) |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1884 continue; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1885 direction = 0; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1886 break; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1887 case 1: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_FOR)) || (pict->pict_type!=FF_B_TYPE)) |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1888 continue; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1889 direction = 0; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1890 break; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1891 case 2: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_BACK)) || (pict->pict_type!=FF_B_TYPE)) |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1892 continue; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1893 direction = 1; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1894 break; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1895 } |
1712 | 1896 if(!USES_LIST(pict->mb_type[mb_index], direction)) |
1897 continue; | |
1898 | |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1899 if(IS_8X8(pict->mb_type[mb_index])){ |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1900 int i; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1901 for(i=0; i<4; i++){ |
1285 | 1902 int sx= mb_x*16 + 4 + 8*(i&1); |
1903 int sy= mb_y*16 + 4 + 8*(i>>1); | |
2522
e25782262d7d
kill warnings patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2514
diff
changeset
|
1904 int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1); |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1905 int mx= (pict->motion_val[direction][xy][0]>>shift) + sx; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1906 int my= (pict->motion_val[direction][xy][1]>>shift) + sy; |
2266
514949de5d15
set AVCodecContext.width/height to the picture width/height instead of the one stored in the bitstream (that only matters if lowres!=0)
michael
parents:
2265
diff
changeset
|
1907 draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100); |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1908 } |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1909 }else if(IS_16X8(pict->mb_type[mb_index])){ |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1910 int i; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1911 for(i=0; i<2; i++){ |
1629
74685a0ec851
16x8 MV vissualization by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1628
diff
changeset
|
1912 int sx=mb_x*16 + 8; |
74685a0ec851
16x8 MV vissualization by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1628
diff
changeset
|
1913 int sy=mb_y*16 + 4 + 8*i; |
2522
e25782262d7d
kill warnings patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2514
diff
changeset
|
1914 int xy= (mb_x*2 + (mb_y*2 + i)*mv_stride) << (mv_sample_log2-1); |
2395 | 1915 int mx=(pict->motion_val[direction][xy][0]>>shift); |
1916 int my=(pict->motion_val[direction][xy][1]>>shift); | |
2967 | 1917 |
2395 | 1918 if(IS_INTERLACED(pict->mb_type[mb_index])) |
1919 my*=2; | |
2967 | 1920 |
2395 | 1921 draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100); |
1922 } | |
1923 }else if(IS_8X16(pict->mb_type[mb_index])){ | |
1924 int i; | |
1925 for(i=0; i<2; i++){ | |
1926 int sx=mb_x*16 + 4 + 8*i; | |
1927 int sy=mb_y*16 + 8; | |
2522
e25782262d7d
kill warnings patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2514
diff
changeset
|
1928 int xy= (mb_x*2 + i + mb_y*2*mv_stride) << (mv_sample_log2-1); |
1947
b306574523ad
fix y scale of mv vissualization of interlaced MVs
michael
parents:
1945
diff
changeset
|
1929 int mx=(pict->motion_val[direction][xy][0]>>shift); |
b306574523ad
fix y scale of mv vissualization of interlaced MVs
michael
parents:
1945
diff
changeset
|
1930 int my=(pict->motion_val[direction][xy][1]>>shift); |
2967 | 1931 |
1947
b306574523ad
fix y scale of mv vissualization of interlaced MVs
michael
parents:
1945
diff
changeset
|
1932 if(IS_INTERLACED(pict->mb_type[mb_index])) |
b306574523ad
fix y scale of mv vissualization of interlaced MVs
michael
parents:
1945
diff
changeset
|
1933 my*=2; |
2967 | 1934 |
2266
514949de5d15
set AVCodecContext.width/height to the picture width/height instead of the one stored in the bitstream (that only matters if lowres!=0)
michael
parents:
2265
diff
changeset
|
1935 draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100); |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1936 } |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1937 }else{ |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1938 int sx= mb_x*16 + 8; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1939 int sy= mb_y*16 + 8; |
2395 | 1940 int xy= (mb_x + mb_y*mv_stride) << mv_sample_log2; |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1941 int mx= (pict->motion_val[direction][xy][0]>>shift) + sx; |
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1705
diff
changeset
|
1942 int my= (pict->motion_val[direction][xy][1]>>shift) + sy; |
2266
514949de5d15
set AVCodecContext.width/height to the picture width/height instead of the one stored in the bitstream (that only matters if lowres!=0)
michael
parents:
2265
diff
changeset
|
1943 draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100); |
1629
74685a0ec851
16x8 MV vissualization by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1628
diff
changeset
|
1944 } |
2967 | 1945 } |
1685 | 1946 } |
1947 if((s->avctx->debug&FF_DEBUG_VIS_QP) && pict->motion_val){ | |
1948 uint64_t c= (pict->qscale_table[mb_index]*128/31) * 0x0101010101010101ULL; | |
1949 int y; | |
1950 for(y=0; y<8; y++){ | |
1951 *(uint64_t*)(pict->data[1] + 8*mb_x + (8*mb_y + y)*pict->linesize[1])= c; | |
1952 *(uint64_t*)(pict->data[2] + 8*mb_x + (8*mb_y + y)*pict->linesize[2])= c; | |
1953 } | |
1954 } | |
1955 if((s->avctx->debug&FF_DEBUG_VIS_MB_TYPE) && pict->motion_val){ | |
1956 int mb_type= pict->mb_type[mb_index]; | |
1957 uint64_t u,v; | |
1958 int y; | |
1959 #define COLOR(theta, r)\ | |
1960 u= (int)(128 + r*cos(theta*3.141592/180));\ | |
1961 v= (int)(128 + r*sin(theta*3.141592/180)); | |
1962 | |
2967 | 1963 |
1685 | 1964 u=v=128; |
1965 if(IS_PCM(mb_type)){ | |
1966 COLOR(120,48) | |
1967 }else if((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || IS_INTRA16x16(mb_type)){ | |
1968 COLOR(30,48) | |
1969 }else if(IS_INTRA4x4(mb_type)){ | |
1970 COLOR(90,48) | |
1971 }else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)){ | |
1972 // COLOR(120,48) | |
1973 }else if(IS_DIRECT(mb_type)){ | |
1974 COLOR(150,48) | |
1975 }else if(IS_GMC(mb_type) && IS_SKIP(mb_type)){ | |
1976 COLOR(170,48) | |
1977 }else if(IS_GMC(mb_type)){ | |
1978 COLOR(190,48) | |
1979 }else if(IS_SKIP(mb_type)){ | |
1980 // COLOR(180,48) | |
1981 }else if(!USES_LIST(mb_type, 1)){ | |
1982 COLOR(240,48) | |
1983 }else if(!USES_LIST(mb_type, 0)){ | |
1984 COLOR(0,48) | |
1985 }else{ | |
1986 assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1)); | |
1987 COLOR(300,48) | |
1988 } | |
1989 | |
1990 u*= 0x0101010101010101ULL; | |
1991 v*= 0x0101010101010101ULL; | |
1992 for(y=0; y<8; y++){ | |
1993 *(uint64_t*)(pict->data[1] + 8*mb_x + (8*mb_y + y)*pict->linesize[1])= u; | |
1994 *(uint64_t*)(pict->data[2] + 8*mb_x + (8*mb_y + y)*pict->linesize[2])= v; | |
1995 } | |
1996 | |
1997 //segmentation | |
1998 if(IS_8X8(mb_type) || IS_16X8(mb_type)){ | |
1999 *(uint64_t*)(pict->data[0] + 16*mb_x + 0 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL; | |
2000 *(uint64_t*)(pict->data[0] + 16*mb_x + 8 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL; | |
2001 } | |
2002 if(IS_8X8(mb_type) || IS_8X16(mb_type)){ | |
2003 for(y=0; y<16; y++) | |
2004 pict->data[0][16*mb_x + 8 + (16*mb_y + y)*pict->linesize[0]]^= 0x80; | |
2005 } | |
2534 | 2006 if(IS_8X8(mb_type) && mv_sample_log2 >= 2){ |
2007 int dm= 1 << (mv_sample_log2-2); | |
2008 for(i=0; i<4; i++){ | |
2009 int sx= mb_x*16 + 8*(i&1); | |
2010 int sy= mb_y*16 + 8*(i>>1); | |
2011 int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1); | |
2012 //FIXME bidir | |
2013 int32_t *mv = (int32_t*)&pict->motion_val[0][xy]; | |
2014 if(mv[0] != mv[dm] || mv[dm*mv_stride] != mv[dm*(mv_stride+1)]) | |
2015 for(y=0; y<8; y++) | |
2016 pict->data[0][sx + 4 + (sy + y)*pict->linesize[0]]^= 0x80; | |
2017 if(mv[0] != mv[dm*mv_stride] || mv[dm] != mv[dm*(mv_stride+1)]) | |
2018 *(uint64_t*)(pict->data[0] + sx + (sy + 4)*pict->linesize[0])^= 0x8080808080808080ULL; | |
2019 } | |
2020 } | |
2967 | 2021 |
1685 | 2022 if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264){ |
2023 // hmm | |
2024 } | |
1285 | 2025 } |
2026 s->mbskip_table[mb_index]=0; | |
2027 } | |
2028 } | |
2029 } | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
2030 } |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
2031 |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
2032 #ifdef CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
2033 |
915 | 2034 static int get_sae(uint8_t *src, int ref, int stride){ |
2035 int x,y; | |
2036 int acc=0; | |
2967 | 2037 |
915 | 2038 for(y=0; y<16; y++){ |
2039 for(x=0; x<16; x++){ | |
2040 acc+= ABS(src[x+y*stride] - ref); | |
2041 } | |
2042 } | |
2967 | 2043 |
915 | 2044 return acc; |
2045 } | |
2046 | |
2047 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){ | |
2048 int x, y, w, h; | |
2049 int acc=0; | |
2967 | 2050 |
915 | 2051 w= s->width &~15; |
2052 h= s->height&~15; | |
2967 | 2053 |
915 | 2054 for(y=0; y<h; y+=16){ |
2055 for(x=0; x<w; x+=16){ | |
2056 int offset= x + y*stride; | |
1708 | 2057 int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16); |
915 | 2058 int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8; |
2059 int sae = get_sae(src + offset, mean, stride); | |
2967 | 2060 |
915 | 2061 acc+= sae + 500 < sad; |
2062 } | |
2063 } | |
2064 return acc; | |
2065 } | |
2066 | |
924 | 2067 |
925 | 2068 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){ |
1373 | 2069 AVFrame *pic=NULL; |
2427 | 2070 int64_t pts; |
924 | 2071 int i; |
903 | 2072 const int encoding_delay= s->max_b_frames; |
924 | 2073 int direct=1; |
2967 | 2074 |
2419 | 2075 if(pic_arg){ |
2427 | 2076 pts= pic_arg->pts; |
2077 pic_arg->display_picture_number= s->input_picture_number++; | |
2078 | |
2967 | 2079 if(pts != AV_NOPTS_VALUE){ |
2419 | 2080 if(s->user_specified_pts != AV_NOPTS_VALUE){ |
2637 | 2081 int64_t time= pts; |
2082 int64_t last= s->user_specified_pts; | |
2967 | 2083 |
2084 if(time <= last){ | |
2962 | 2085 av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts); |
2419 | 2086 return -1; |
2087 } | |
2088 } | |
2427 | 2089 s->user_specified_pts= pts; |
2419 | 2090 }else{ |
2091 if(s->user_specified_pts != AV_NOPTS_VALUE){ | |
2967 | 2092 s->user_specified_pts= |
2637 | 2093 pts= s->user_specified_pts + 1; |
2962 | 2094 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts); |
2419 | 2095 }else{ |
2637 | 2096 pts= pic_arg->display_picture_number; |
2419 | 2097 } |
2098 } | |
2099 } | |
2100 | |
1373 | 2101 if(pic_arg){ |
924 | 2102 if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0; |
2103 if(pic_arg->linesize[0] != s->linesize) direct=0; | |
2104 if(pic_arg->linesize[1] != s->uvlinesize) direct=0; | |
2105 if(pic_arg->linesize[2] != s->uvlinesize) direct=0; | |
2967 | 2106 |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
2107 // av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize); |
2967 | 2108 |
924 | 2109 if(direct){ |
1586 | 2110 i= ff_find_unused_picture(s, 1); |
924 | 2111 |
925 | 2112 pic= (AVFrame*)&s->picture[i]; |
1168 | 2113 pic->reference= 3; |
2967 | 2114 |
924 | 2115 for(i=0; i<4; i++){ |
2116 pic->data[i]= pic_arg->data[i]; | |
2117 pic->linesize[i]= pic_arg->linesize[i]; | |
2118 } | |
2119 alloc_picture(s, (Picture*)pic, 1); | |
2120 }else{ | |
1586 | 2121 i= ff_find_unused_picture(s, 0); |
924 | 2122 |
925 | 2123 pic= (AVFrame*)&s->picture[i]; |
1168 | 2124 pic->reference= 3; |
924 | 2125 |
2126 alloc_picture(s, (Picture*)pic, 0); | |
1539
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
2127 |
3068 | 2128 if( pic->data[0] + INPLACE_OFFSET == pic_arg->data[0] |
2129 && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1] | |
2130 && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){ | |
924 | 2131 // empty |
2132 }else{ | |
2133 int h_chroma_shift, v_chroma_shift; | |
2134 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift); | |
2967 | 2135 |
924 | 2136 for(i=0; i<3; i++){ |
2137 int src_stride= pic_arg->linesize[i]; | |
2138 int dst_stride= i ? s->uvlinesize : s->linesize; | |
2139 int h_shift= i ? h_chroma_shift : 0; | |
2140 int v_shift= i ? v_chroma_shift : 0; | |
2141 int w= s->width >>h_shift; | |
2142 int h= s->height>>v_shift; | |
2143 uint8_t *src= pic_arg->data[i]; | |
3939
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2144 uint8_t *dst= pic->data[i]; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2145 |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2146 if(!s->avctx->rc_buffer_size) |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2147 dst +=INPLACE_OFFSET; |
2967 | 2148 |
924 | 2149 if(src_stride==dst_stride) |
2150 memcpy(dst, src, src_stride*h); | |
2151 else{ | |
2152 while(h--){ | |
2153 memcpy(dst, src, w); | |
2154 dst += dst_stride; | |
2155 src += src_stride; | |
2156 } | |
2157 } | |
2158 } | |
2159 } | |
324 | 2160 } |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
2161 copy_picture_attributes(s, pic, pic_arg); |
2427 | 2162 pic->pts= pts; //we set this here to avoid modifiying pic_arg |
1373 | 2163 } |
2967 | 2164 |
903 | 2165 /* shift buffer entries */ |
2166 for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++) | |
2167 s->input_picture[i-1]= s->input_picture[i]; | |
2967 | 2168 |
903 | 2169 s->input_picture[encoding_delay]= (Picture*)pic; |
2170 | |
2171 return 0; | |
2172 } | |
2173 | |
2381 | 2174 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){ |
2175 int x, y, plane; | |
2176 int score=0; | |
2382 | 2177 int64_t score64=0; |
2381 | 2178 |
2179 for(plane=0; plane<3; plane++){ | |
2180 const int stride= p->linesize[plane]; | |
2181 const int bw= plane ? 1 : 2; | |
2182 for(y=0; y<s->mb_height*bw; y++){ | |
2183 for(x=0; x<s->mb_width*bw; x++){ | |
2931 | 2184 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16; |
2185 int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), stride, 8); | |
2967 | 2186 |
2382 | 2187 switch(s->avctx->frame_skip_exp){ |
2188 case 0: score= FFMAX(score, v); break; | |
2189 case 1: score+= ABS(v);break; | |
2190 case 2: score+= v*v;break; | |
2191 case 3: score64+= ABS(v*v*(int64_t)v);break; | |
2192 case 4: score64+= v*v*(int64_t)(v*v);break; | |
2193 } | |
2381 | 2194 } |
2195 } | |
2196 } | |
2967 | 2197 |
2382 | 2198 if(score) score64= score; |
2199 | |
2200 if(score64 < s->avctx->frame_skip_threshold) | |
2381 | 2201 return 1; |
2382 | 2202 if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8)) |
2381 | 2203 return 1; |
2204 return 0; | |
2205 } | |
2206 | |
2988 | 2207 static int estimate_best_b_count(MpegEncContext *s){ |
2208 AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id); | |
2209 AVCodecContext *c= avcodec_alloc_context(); | |
2210 AVFrame input[FF_MAX_B_FRAMES+2]; | |
2990
0300c2647bc3
support downscaling frames for dynamic b frame decission
michael
parents:
2988
diff
changeset
|
2211 const int scale= s->avctx->brd_scale; |
3068 | 2212 int i, j, out_size, p_lambda, b_lambda, lambda2; |
2990
0300c2647bc3
support downscaling frames for dynamic b frame decission
michael
parents:
2988
diff
changeset
|
2213 int outbuf_size= s->width * s->height; //FIXME |
2988 | 2214 uint8_t *outbuf= av_malloc(outbuf_size); |
2215 int64_t best_rd= INT64_MAX; | |
2216 int best_b_count= -1; | |
3068 | 2217 |
3245 | 2218 assert(scale>=0 && scale <=3); |
2219 | |
3068 | 2220 // emms_c(); |
2221 p_lambda= s->last_lambda_for[P_TYPE]; //s->next_picture_ptr->quality; | |
2222 b_lambda= s->last_lambda_for[B_TYPE]; //p_lambda *ABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset; | |
2223 if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else | |
2224 lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT; | |
2988 | 2225 |
2226 c->width = s->width >> scale; | |
2227 c->height= s->height>> scale; | |
2228 c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/; | |
2229 c->flags|= s->avctx->flags & CODEC_FLAG_QPEL; | |
2230 c->mb_decision= s->avctx->mb_decision; | |
2231 c->me_cmp= s->avctx->me_cmp; | |
2232 c->mb_cmp= s->avctx->mb_cmp; | |
2233 c->me_sub_cmp= s->avctx->me_sub_cmp; | |
2234 c->pix_fmt = PIX_FMT_YUV420P; | |
2235 c->time_base= s->avctx->time_base; | |
2236 c->max_b_frames= s->max_b_frames; | |
2237 | |
2238 if (avcodec_open(c, codec) < 0) | |
2239 return -1; | |
2240 | |
2241 for(i=0; i<s->max_b_frames+2; i++){ | |
2242 int ysize= c->width*c->height; | |
2243 int csize= (c->width/2)*(c->height/2); | |
3068 | 2244 Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr; |
2245 | |
2246 if(pre_input_ptr) | |
2247 pre_input= *pre_input_ptr; | |
2248 | |
2249 if(pre_input.type != FF_BUFFER_TYPE_SHARED && i){ | |
2250 pre_input.data[0]+=INPLACE_OFFSET; | |
2251 pre_input.data[1]+=INPLACE_OFFSET; | |
2252 pre_input.data[2]+=INPLACE_OFFSET; | |
2253 } | |
2988 | 2254 |
2255 avcodec_get_frame_defaults(&input[i]); | |
2256 input[i].data[0]= av_malloc(ysize + 2*csize); | |
2257 input[i].data[1]= input[i].data[0] + ysize; | |
2258 input[i].data[2]= input[i].data[1] + csize; | |
2259 input[i].linesize[0]= c->width; | |
2260 input[i].linesize[1]= | |
2261 input[i].linesize[2]= c->width/2; | |
2262 | |
3245 | 2263 if(!i || s->input_picture[i-1]){ |
2264 s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height); | |
2265 s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.data[1], pre_input.linesize[1], c->width>>1, c->height>>1); | |
2266 s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.data[2], pre_input.linesize[2], c->width>>1, c->height>>1); | |
2267 } | |
2988 | 2268 } |
2269 | |
2270 for(j=0; j<s->max_b_frames+1; j++){ | |
2271 int64_t rd=0; | |
2272 | |
2273 if(!s->input_picture[j]) | |
2274 break; | |
2275 | |
2276 c->error[0]= c->error[1]= c->error[2]= 0; | |
2277 | |
2278 input[0].pict_type= I_TYPE; | |
3068 | 2279 input[0].quality= 1 * FF_QP2LAMBDA; |
2988 | 2280 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]); |
3068 | 2281 // rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT; |
2988 | 2282 |
2283 for(i=0; i<s->max_b_frames+1; i++){ | |
2284 int is_p= i % (j+1) == j || i==s->max_b_frames; | |
2285 | |
2286 input[i+1].pict_type= is_p ? P_TYPE : B_TYPE; | |
3068 | 2287 input[i+1].quality= is_p ? p_lambda : b_lambda; |
2988 | 2288 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]); |
3068 | 2289 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3); |
2988 | 2290 } |
2291 | |
2292 /* get the delayed frames */ | |
2293 while(out_size){ | |
2294 out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL); | |
3068 | 2295 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3); |
2988 | 2296 } |
2297 | |
2298 rd += c->error[0] + c->error[1] + c->error[2]; | |
2299 | |
2300 if(rd < best_rd){ | |
2301 best_rd= rd; | |
2302 best_b_count= j; | |
2303 } | |
2304 } | |
2305 | |
2306 av_freep(&outbuf); | |
2307 avcodec_close(c); | |
2308 av_freep(&c); | |
2309 | |
2310 for(i=0; i<s->max_b_frames+2; i++){ | |
2311 av_freep(&input[i].data[0]); | |
2312 } | |
2313 | |
2314 return best_b_count; | |
2315 } | |
2316 | |
903 | 2317 static void select_input_picture(MpegEncContext *s){ |
2318 int i; | |
924 | 2319 |
903 | 2320 for(i=1; i<MAX_PICTURE_COUNT; i++) |
2321 s->reordered_input_picture[i-1]= s->reordered_input_picture[i]; | |
2322 s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL; | |
2323 | |
2764 | 2324 /* set next picture type & ordering */ |
903 | 2325 if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){ |
1138 | 2326 if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){ |
915 | 2327 s->reordered_input_picture[0]= s->input_picture[0]; |
2328 s->reordered_input_picture[0]->pict_type= I_TYPE; | |
1705 | 2329 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++; |
915 | 2330 }else{ |
2331 int b_frames; | |
2381 | 2332 |
2333 if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){ | |
2931 | 2334 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){ |
2335 //FIXME check that te gop check above is +-1 correct | |
2381 | 2336 //av_log(NULL, AV_LOG_DEBUG, "skip %p %Ld\n", s->input_picture[0]->data[0], s->input_picture[0]->pts); |
2967 | 2337 |
2381 | 2338 if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){ |
2339 for(i=0; i<4; i++) | |
2340 s->input_picture[0]->data[i]= NULL; | |
2967 | 2341 s->input_picture[0]->type= 0; |
2381 | 2342 }else{ |
2967 | 2343 assert( s->input_picture[0]->type==FF_BUFFER_TYPE_USER |
2381 | 2344 || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL); |
2967 | 2345 |
2381 | 2346 s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]); |
2347 } | |
2967 | 2348 |
2931 | 2349 emms_c(); |
2350 ff_vbv_update(s, 0); | |
2381 | 2351 |
2352 goto no_output_pic; | |
2353 } | |
2354 } | |
2355 | |
915 | 2356 if(s->flags&CODEC_FLAG_PASS2){ |
2357 for(i=0; i<s->max_b_frames+1; i++){ | |
2358 int pict_num= s->input_picture[0]->display_picture_number + i; | |
2364
f8a229dfa2e8
check num_entries too to avoid reading over the array end and dont check i>0 it must be >0 there
michael
parents:
2363
diff
changeset
|
2359 |
2967 | 2360 if(pict_num >= s->rc_context.num_entries) |
2364
f8a229dfa2e8
check num_entries too to avoid reading over the array end and dont check i>0 it must be >0 there
michael
parents:
2363
diff
changeset
|
2361 break; |
2363
e0bda0b8359a
Fixes a segfault when flushing delayed B-frames on the 2nd pass. patch by (Loren Merritt <lorenm.u@washington@edu>)
michael
parents:
2345
diff
changeset
|
2362 if(!s->input_picture[i]){ |
2364
f8a229dfa2e8
check num_entries too to avoid reading over the array end and dont check i>0 it must be >0 there
michael
parents:
2363
diff
changeset
|
2363 s->rc_context.entry[pict_num-1].new_pict_type = P_TYPE; |
2363
e0bda0b8359a
Fixes a segfault when flushing delayed B-frames on the 2nd pass. patch by (Loren Merritt <lorenm.u@washington@edu>)
michael
parents:
2345
diff
changeset
|
2364 break; |
e0bda0b8359a
Fixes a segfault when flushing delayed B-frames on the 2nd pass. patch by (Loren Merritt <lorenm.u@washington@edu>)
michael
parents:
2345
diff
changeset
|
2365 } |
2364
f8a229dfa2e8
check num_entries too to avoid reading over the array end and dont check i>0 it must be >0 there
michael
parents:
2363
diff
changeset
|
2366 |
2967 | 2367 s->input_picture[i]->pict_type= |
2364
f8a229dfa2e8
check num_entries too to avoid reading over the array end and dont check i>0 it must be >0 there
michael
parents:
2363
diff
changeset
|
2368 s->rc_context.entry[pict_num].new_pict_type; |
915 | 2369 } |
2370 } | |
924 | 2371 |
2297 | 2372 if(s->avctx->b_frame_strategy==0){ |
915 | 2373 b_frames= s->max_b_frames; |
1373 | 2374 while(b_frames && !s->input_picture[b_frames]) b_frames--; |
1687 | 2375 }else if(s->avctx->b_frame_strategy==1){ |
915 | 2376 for(i=1; i<s->max_b_frames+1; i++){ |
1373 | 2377 if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){ |
2967 | 2378 s->input_picture[i]->b_frame_score= |
2379 get_intra_count(s, s->input_picture[i ]->data[0], | |
924 | 2380 s->input_picture[i-1]->data[0], s->linesize) + 1; |
915 | 2381 } |
2382 } | |
2728 | 2383 for(i=0; i<s->max_b_frames+1; i++){ |
3335
97af1b315f59
Make B-frame reduction sensitivity by b_strategy/vb_strategy = 1
corey
parents:
3321
diff
changeset
|
2384 if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break; |
915 | 2385 } |
2967 | 2386 |
915 | 2387 b_frames= FFMAX(0, i-1); |
2967 | 2388 |
915 | 2389 /* reset scores */ |
2390 for(i=0; i<b_frames+1; i++){ | |
2391 s->input_picture[i]->b_frame_score=0; | |
2392 } | |
2988 | 2393 }else if(s->avctx->b_frame_strategy==2){ |
2394 b_frames= estimate_best_b_count(s); | |
915 | 2395 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
2396 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n"); |
915 | 2397 b_frames=0; |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
2398 } |
915 | 2399 |
2400 emms_c(); | |
2401 //static int b_count=0; | |
2402 //b_count+= b_frames; | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
2403 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count); |
2297 | 2404 |
2405 for(i= b_frames - 1; i>=0; i--){ | |
2406 int type= s->input_picture[i]->pict_type; | |
2407 if(type && type != B_TYPE) | |
2408 b_frames= i; | |
2409 } | |
2410 if(s->input_picture[b_frames]->pict_type == B_TYPE && b_frames == s->max_b_frames){ | |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
2411 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n"); |
2297 | 2412 } |
2413 | |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2414 if(s->picture_in_gop_number + b_frames >= s->gop_size){ |
2298 | 2415 if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){ |
2416 b_frames= s->gop_size - s->picture_in_gop_number - 1; | |
2417 }else{ | |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2418 if(s->flags & CODEC_FLAG_CLOSED_GOP) |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2419 b_frames=0; |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2420 s->input_picture[b_frames]->pict_type= I_TYPE; |
2298 | 2421 } |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2422 } |
2967 | 2423 |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2424 if( (s->flags & CODEC_FLAG_CLOSED_GOP) |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2425 && b_frames |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2426 && s->input_picture[b_frames]->pict_type== I_TYPE) |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2427 b_frames--; |
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2428 |
915 | 2429 s->reordered_input_picture[0]= s->input_picture[b_frames]; |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1751
diff
changeset
|
2430 if(s->reordered_input_picture[0]->pict_type != I_TYPE) |
915 | 2431 s->reordered_input_picture[0]->pict_type= P_TYPE; |
1705 | 2432 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++; |
915 | 2433 for(i=0; i<b_frames; i++){ |
2434 s->reordered_input_picture[i+1]= s->input_picture[i]; | |
2435 s->reordered_input_picture[i+1]->pict_type= B_TYPE; | |
1705 | 2436 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++; |
903 | 2437 } |
324 | 2438 } |
2439 } | |
2381 | 2440 no_output_pic: |
903 | 2441 if(s->reordered_input_picture[0]){ |
1168 | 2442 s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=B_TYPE ? 3 : 0; |
1138 | 2443 |
1539
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
2444 copy_picture(&s->new_picture, s->reordered_input_picture[0]); |
924 | 2445 |
3939
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2446 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){ |
2764 | 2447 // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable |
2967 | 2448 |
1586 | 2449 int i= ff_find_unused_picture(s, 0); |
924 | 2450 Picture *pic= &s->picture[i]; |
2451 | |
3939
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2452 pic->reference = s->reordered_input_picture[0]->reference; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2453 alloc_picture(s, pic, 0); |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2454 |
924 | 2455 /* mark us unused / free shared pic */ |
3939
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2456 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL) |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2457 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]); |
924 | 2458 for(i=0; i<4; i++) |
2459 s->reordered_input_picture[0]->data[i]= NULL; | |
2460 s->reordered_input_picture[0]->type= 0; | |
2967 | 2461 |
1955
5dafb10e0252
reuse motion vectors/mb types/field select values of the source video, if the SSE for a macroblock which is predicted with these values is below me_threshold
michael
parents:
1948
diff
changeset
|
2462 copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]); |
5dafb10e0252
reuse motion vectors/mb types/field select values of the source video, if the SSE for a macroblock which is predicted with these values is below me_threshold
michael
parents:
1948
diff
changeset
|
2463 |
1138 | 2464 s->current_picture_ptr= pic; |
924 | 2465 }else{ |
1138 | 2466 // input is not a shared pix -> reuse buffer for current_pix |
2467 | |
2967 | 2468 assert( s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER |
924 | 2469 || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL); |
2967 | 2470 |
1138 | 2471 s->current_picture_ptr= s->reordered_input_picture[0]; |
924 | 2472 for(i=0; i<4; i++){ |
3068 | 2473 s->new_picture.data[i]+= INPLACE_OFFSET; |
924 | 2474 } |
903 | 2475 } |
1539
14d01ccc0081
picture buffer cleanup, this fixes a assertion failure (if assertion checking was manualy enabled for utils.c) and a memleak at least
michael
parents:
1530
diff
changeset
|
2476 copy_picture(&s->current_picture, s->current_picture_ptr); |
2967 | 2477 |
903 | 2478 s->picture_number= s->new_picture.display_picture_number; |
2479 //printf("dpn:%d\n", s->picture_number); | |
2480 }else{ | |
2481 memset(&s->new_picture, 0, sizeof(Picture)); | |
324 | 2482 } |
2483 } | |
2484 | |
0 | 2485 int MPV_encode_picture(AVCodecContext *avctx, |
2486 unsigned char *buf, int buf_size, void *data) | |
2487 { | |
2488 MpegEncContext *s = avctx->priv_data; | |
925 | 2489 AVFrame *pic_arg = data; |
1684 | 2490 int i, stuffing_count; |
0 | 2491 |
1799 | 2492 for(i=0; i<avctx->thread_count; i++){ |
1838 | 2493 int start_y= s->thread_context[i]->start_mb_y; |
2494 int end_y= s->thread_context[i]-> end_mb_y; | |
1799 | 2495 int h= s->mb_height; |
2807
3137aae854c9
Fix for overflow issue in mpegvideo.c patch by (Martin Boehme: boehme, inb uni-luebeck de)
michael
parents:
2792
diff
changeset
|
2496 uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h); |
3137aae854c9
Fix for overflow issue in mpegvideo.c patch by (Martin Boehme: boehme, inb uni-luebeck de)
michael
parents:
2792
diff
changeset
|
2497 uint8_t *end = buf + (size_t)(((int64_t) buf_size)* end_y/h); |
1799 | 2498 |
2499 init_put_bits(&s->thread_context[i]->pb, start, end - start); | |
2500 } | |
0 | 2501 |
903 | 2502 s->picture_in_gop_number++; |
2503 | |
2243
b8bad8bbbc64
check if the user specified timestamps are strictly monotone
michael
parents:
2235
diff
changeset
|
2504 if(load_input_picture(s, pic_arg) < 0) |
b8bad8bbbc64
check if the user specified timestamps are strictly monotone
michael
parents:
2235
diff
changeset
|
2505 return -1; |
2967 | 2506 |
903 | 2507 select_input_picture(s); |
2967 | 2508 |
324 | 2509 /* output? */ |
903 | 2510 if(s->new_picture.data[0]){ |
2511 s->pict_type= s->new_picture.pict_type; | |
2512 //emms_c(); | |
2513 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale); | |
553 | 2514 MPV_frame_start(s, avctx); |
3939
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2515 vbv_retry: |
3766 | 2516 if (encode_picture(s, s->picture_number) < 0) |
2517 return -1; | |
2967 | 2518 |
376 | 2519 avctx->real_pict_num = s->picture_number; |
324 | 2520 avctx->header_bits = s->header_bits; |
2521 avctx->mv_bits = s->mv_bits; | |
2522 avctx->misc_bits = s->misc_bits; | |
2523 avctx->i_tex_bits = s->i_tex_bits; | |
2524 avctx->p_tex_bits = s->p_tex_bits; | |
2525 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
|
2526 avctx->p_count = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx |
324 | 2527 avctx->skip_count = s->skip_count; |
0 | 2528 |
324 | 2529 MPV_frame_end(s); |
2530 | |
2531 if (s->out_format == FMT_MJPEG) | |
2532 mjpeg_picture_trailer(s); | |
2967 | 2533 |
3939
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2534 if(avctx->rc_buffer_size){ |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2535 RateControlContext *rcc= &s->rc_context; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2536 int max_size= rcc->buffer_index/3; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2537 |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2538 if(put_bits_count(&s->pb) > max_size && s->qscale < s->avctx->qmax){ |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2539 s->next_lambda= s->lambda*(s->qscale+1) / s->qscale; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2540 s->mb_skipped = 0; //done in MPV_frame_start() |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2541 if(s->pict_type==P_TYPE){ //done in encode_picture() so we must undo it |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2542 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4) |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2543 s->no_rounding ^= 1; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2544 } |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2545 // av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda); |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2546 for(i=0; i<avctx->thread_count; i++){ |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2547 PutBitContext *pb= &s->thread_context[i]->pb; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2548 init_put_bits(pb, pb->buf, pb->buf_end - pb->buf); |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2549 } |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2550 goto vbv_retry; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2551 } |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2552 |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2553 assert(s->avctx->rc_max_rate); |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2554 } |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
2555 |
329 | 2556 if(s->flags&CODEC_FLAG_PASS1) |
2557 ff_write_pass1_stats(s); | |
1138 | 2558 |
2559 for(i=0; i<4; i++){ | |
2955 | 2560 s->current_picture_ptr->error[i]= s->current_picture.error[i]; |
1138 | 2561 avctx->error[i] += s->current_picture_ptr->error[i]; |
2562 } | |
1766 | 2563 |
2428
548c87c06dbb
assertion about bits statistic and minor bits stat fix
michael
parents:
2427
diff
changeset
|
2564 if(s->flags&CODEC_FLAG_PASS1) |
548c87c06dbb
assertion about bits statistic and minor bits stat fix
michael
parents:
2427
diff
changeset
|
2565 assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb)); |
1684 | 2566 flush_put_bits(&s->pb); |
1801
3f26dfb3eba4
replace a few pbBufPtr() by put_bits_count(), one of them was actually wrong
michael
parents:
1799
diff
changeset
|
2567 s->frame_bits = put_bits_count(&s->pb); |
1766 | 2568 |
2569 stuffing_count= ff_vbv_update(s, s->frame_bits); | |
2570 if(stuffing_count){ | |
2422 | 2571 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){ |
2572 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n"); | |
2573 return -1; | |
2574 } | |
2575 | |
1766 | 2576 switch(s->codec_id){ |
2577 case CODEC_ID_MPEG1VIDEO: | |
2578 case CODEC_ID_MPEG2VIDEO: | |
2579 while(stuffing_count--){ | |
2580 put_bits(&s->pb, 8, 0); | |
2581 } | |
2582 break; | |
2583 case CODEC_ID_MPEG4: | |
2584 put_bits(&s->pb, 16, 0); | |
2585 put_bits(&s->pb, 16, 0x1C3); | |
2586 stuffing_count -= 4; | |
2587 while(stuffing_count--){ | |
2588 put_bits(&s->pb, 8, 0xFF); | |
2589 } | |
2590 break; | |
2591 default: | |
2592 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n"); | |
2593 } | |
2594 flush_put_bits(&s->pb); | |
1801
3f26dfb3eba4
replace a few pbBufPtr() by put_bits_count(), one of them was actually wrong
michael
parents:
1799
diff
changeset
|
2595 s->frame_bits = put_bits_count(&s->pb); |
1766 | 2596 } |
2597 | |
2967 | 2598 /* update mpeg1/2 vbv_delay for CBR */ |
1907 | 2599 if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1 |
2600 && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){ | |
1766 | 2601 int vbv_delay; |
2602 | |
2603 assert(s->repeat_first_field==0); | |
2967 | 2604 |
1766 | 2605 vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate); |
2606 assert(vbv_delay < 0xFFFF); | |
2607 | |
2608 s->vbv_delay_ptr[0] &= 0xF8; | |
2609 s->vbv_delay_ptr[0] |= vbv_delay>>13; | |
2610 s->vbv_delay_ptr[1] = vbv_delay>>5; | |
2611 s->vbv_delay_ptr[2] &= 0x07; | |
2612 s->vbv_delay_ptr[2] |= vbv_delay<<3; | |
2613 } | |
2614 s->total_bits += s->frame_bits; | |
2615 avctx->frame_bits = s->frame_bits; | |
2616 }else{ | |
2617 assert((pbBufPtr(&s->pb) == s->pb.buf)); | |
2618 s->frame_bits=0; | |
1684 | 2619 } |
1766 | 2620 assert((s->frame_bits&7)==0); |
2967 | 2621 |
1684 | 2622 return s->frame_bits/8; |
0 | 2623 } |
2624 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
2625 #endif //CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
2626 |
255 | 2627 static inline void gmc1_motion(MpegEncContext *s, |
1064 | 2628 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2629 uint8_t **ref_picture) |
255 | 2630 { |
1064 | 2631 uint8_t *ptr; |
556 | 2632 int offset, src_x, src_y, linesize, uvlinesize; |
255 | 2633 int motion_x, motion_y; |
566 | 2634 int emu=0; |
255 | 2635 |
2636 motion_x= s->sprite_offset[0][0]; | |
2637 motion_y= s->sprite_offset[0][1]; | |
2638 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
2639 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
2640 motion_x<<=(3-s->sprite_warping_accuracy); | |
2641 motion_y<<=(3-s->sprite_warping_accuracy); | |
2642 src_x = clip(src_x, -16, s->width); | |
2643 if (src_x == s->width) | |
2644 motion_x =0; | |
2645 src_y = clip(src_y, -16, s->height); | |
2646 if (src_y == s->height) | |
2647 motion_y =0; | |
753 | 2648 |
255 | 2649 linesize = s->linesize; |
556 | 2650 uvlinesize = s->uvlinesize; |
2967 | 2651 |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2652 ptr = ref_picture[0] + (src_y * linesize) + src_x; |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2653 |
566 | 2654 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
1542
576861d6343a
emu_edge checks simplification and avoid redundant checks for mpeg1/2 if emu_edge is set
michael
parents:
1539
diff
changeset
|
2655 if( (unsigned)src_x >= s->h_edge_pos - 17 |
576861d6343a
emu_edge checks simplification and avoid redundant checks for mpeg1/2 if emu_edge is set
michael
parents:
1539
diff
changeset
|
2656 || (unsigned)src_y >= s->v_edge_pos - 17){ |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2657 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos); |
566 | 2658 ptr= s->edge_emu_buffer; |
2659 } | |
2660 } | |
2967 | 2661 |
753 | 2662 if((motion_x|motion_y)&7){ |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2663 s->dsp.gmc1(dest_y , ptr , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2664 s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding); |
753 | 2665 }else{ |
2666 int dxy; | |
2967 | 2667 |
753 | 2668 dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2); |
2669 if (s->no_rounding){ | |
2979 | 2670 s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16); |
753 | 2671 }else{ |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2672 s->dsp.put_pixels_tab [0][dxy](dest_y, ptr, linesize, 16); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2673 } |
753 | 2674 } |
2967 | 2675 |
753 | 2676 if(s->flags&CODEC_FLAG_GRAY) return; |
255 | 2677 |
2678 motion_x= s->sprite_offset[1][0]; | |
2679 motion_y= s->sprite_offset[1][1]; | |
2680 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1)); | |
2681 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1)); | |
2682 motion_x<<=(3-s->sprite_warping_accuracy); | |
2683 motion_y<<=(3-s->sprite_warping_accuracy); | |
2684 src_x = clip(src_x, -8, s->width>>1); | |
2685 if (src_x == s->width>>1) | |
2686 motion_x =0; | |
2687 src_y = clip(src_y, -8, s->height>>1); | |
2688 if (src_y == s->height>>1) | |
2689 motion_y =0; | |
2690 | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2691 offset = (src_y * uvlinesize) + src_x; |
255 | 2692 ptr = ref_picture[1] + offset; |
1002 | 2693 if(s->flags&CODEC_FLAG_EMU_EDGE){ |
1542
576861d6343a
emu_edge checks simplification and avoid redundant checks for mpeg1/2 if emu_edge is set
michael
parents:
1539
diff
changeset
|
2694 if( (unsigned)src_x >= (s->h_edge_pos>>1) - 9 |
576861d6343a
emu_edge checks simplification and avoid redundant checks for mpeg1/2 if emu_edge is set
michael
parents:
1539
diff
changeset
|
2695 || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){ |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2696 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
1002 | 2697 ptr= s->edge_emu_buffer; |
2698 emu=1; | |
2699 } | |
566 | 2700 } |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2701 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding); |
2967 | 2702 |
255 | 2703 ptr = ref_picture[2] + offset; |
566 | 2704 if(emu){ |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2705 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); |
566 | 2706 ptr= s->edge_emu_buffer; |
2707 } | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2708 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding); |
2967 | 2709 |
255 | 2710 return; |
2711 } | |
2712 | |
753 | 2713 static inline void gmc_motion(MpegEncContext *s, |
1064 | 2714 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2715 uint8_t **ref_picture) |
753 | 2716 { |
1064 | 2717 uint8_t *ptr; |
753 | 2718 int linesize, uvlinesize; |
2719 const int a= s->sprite_warping_accuracy; | |
2720 int ox, oy; | |
2721 | |
2722 linesize = s->linesize; | |
2723 uvlinesize = s->uvlinesize; | |
2724 | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2725 ptr = ref_picture[0]; |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2726 |
753 | 2727 ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16; |
2728 oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16; | |
2729 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2730 s->dsp.gmc(dest_y, ptr, linesize, 16, |
2967 | 2731 ox, |
2732 oy, | |
753 | 2733 s->sprite_delta[0][0], s->sprite_delta[0][1], |
2967 | 2734 s->sprite_delta[1][0], s->sprite_delta[1][1], |
753 | 2735 a+1, (1<<(2*a+1)) - s->no_rounding, |
2736 s->h_edge_pos, s->v_edge_pos); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2737 s->dsp.gmc(dest_y+8, ptr, linesize, 16, |
2967 | 2738 ox + s->sprite_delta[0][0]*8, |
2739 oy + s->sprite_delta[1][0]*8, | |
753 | 2740 s->sprite_delta[0][0], s->sprite_delta[0][1], |
2967 | 2741 s->sprite_delta[1][0], s->sprite_delta[1][1], |
753 | 2742 a+1, (1<<(2*a+1)) - s->no_rounding, |
2743 s->h_edge_pos, s->v_edge_pos); | |
2744 | |
2745 if(s->flags&CODEC_FLAG_GRAY) return; | |
2746 | |
2747 ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8; | |
2748 oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8; | |
2749 | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2750 ptr = ref_picture[1]; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2751 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8, |
2967 | 2752 ox, |
2753 oy, | |
753 | 2754 s->sprite_delta[0][0], s->sprite_delta[0][1], |
2967 | 2755 s->sprite_delta[1][0], s->sprite_delta[1][1], |
753 | 2756 a+1, (1<<(2*a+1)) - s->no_rounding, |
2757 s->h_edge_pos>>1, s->v_edge_pos>>1); | |
2967 | 2758 |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2759 ptr = ref_picture[2]; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2760 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8, |
2967 | 2761 ox, |
2762 oy, | |
753 | 2763 s->sprite_delta[0][0], s->sprite_delta[0][1], |
2967 | 2764 s->sprite_delta[1][0], s->sprite_delta[1][1], |
753 | 2765 a+1, (1<<(2*a+1)) - s->no_rounding, |
2766 s->h_edge_pos>>1, s->v_edge_pos>>1); | |
2767 } | |
2768 | |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2769 /** |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2770 * Copies a rectangular area of samples to a temporary buffer and replicates the boarder samples. |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2771 * @param buf destination buffer |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2772 * @param src source buffer |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2773 * @param linesize number of bytes between 2 vertically adjacent samples in both the source and destination buffers |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2774 * @param block_w width of block |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2775 * @param block_h height of block |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2776 * @param src_x x coordinate of the top left sample of the block in the source buffer |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2777 * @param src_y y coordinate of the top left sample of the block in the source buffer |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2778 * @param w width of the source buffer |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2779 * @param h height of the source buffer |
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
2780 */ |
2967 | 2781 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, int block_w, int block_h, |
553 | 2782 int src_x, int src_y, int w, int h){ |
2783 int x, y; | |
2784 int start_y, start_x, end_y, end_x; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
2785 |
553 | 2786 if(src_y>= h){ |
2787 src+= (h-1-src_y)*linesize; | |
2788 src_y=h-1; | |
554 | 2789 }else if(src_y<=-block_h){ |
2790 src+= (1-block_h-src_y)*linesize; | |
2791 src_y=1-block_h; | |
553 | 2792 } |
2793 if(src_x>= w){ | |
2794 src+= (w-1-src_x); | |
2795 src_x=w-1; | |
554 | 2796 }else if(src_x<=-block_w){ |
2797 src+= (1-block_w-src_x); | |
2798 src_x=1-block_w; | |
553 | 2799 } |
2800 | |
847 | 2801 start_y= FFMAX(0, -src_y); |
2802 start_x= FFMAX(0, -src_x); | |
2803 end_y= FFMIN(block_h, h-src_y); | |
2804 end_x= FFMIN(block_w, w-src_x); | |
566 | 2805 |
553 | 2806 // copy existing part |
2807 for(y=start_y; y<end_y; y++){ | |
2808 for(x=start_x; x<end_x; x++){ | |
2809 buf[x + y*linesize]= src[x + y*linesize]; | |
2810 } | |
2811 } | |
2812 | |
2813 //top | |
2814 for(y=0; y<start_y; y++){ | |
2815 for(x=start_x; x<end_x; x++){ | |
2816 buf[x + y*linesize]= buf[x + start_y*linesize]; | |
2817 } | |
2818 } | |
2819 | |
2820 //bottom | |
2821 for(y=end_y; y<block_h; y++){ | |
2822 for(x=start_x; x<end_x; x++){ | |
2823 buf[x + y*linesize]= buf[x + (end_y-1)*linesize]; | |
2824 } | |
2825 } | |
2967 | 2826 |
553 | 2827 for(y=0; y<block_h; y++){ |
2828 //left | |
2829 for(x=0; x<start_x; x++){ | |
2830 buf[x + y*linesize]= buf[start_x + y*linesize]; | |
2831 } | |
2967 | 2832 |
553 | 2833 //right |
2834 for(x=end_x; x<block_w; x++){ | |
2835 buf[x + y*linesize]= buf[end_x - 1 + y*linesize]; | |
2836 } | |
2837 } | |
2838 } | |
2839 | |
2967 | 2840 static inline int hpel_motion(MpegEncContext *s, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2841 uint8_t *dest, uint8_t *src, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2842 int field_based, int field_select, |
1633 | 2843 int src_x, int src_y, |
2844 int width, int height, int stride, | |
2845 int h_edge_pos, int v_edge_pos, | |
2846 int w, int h, op_pixels_func *pix_op, | |
2847 int motion_x, int motion_y) | |
2848 { | |
2849 int dxy; | |
2850 int emu=0; | |
2851 | |
2852 dxy = ((motion_y & 1) << 1) | (motion_x & 1); | |
2853 src_x += motion_x >> 1; | |
2854 src_y += motion_y >> 1; | |
2967 | 2855 |
1633 | 2856 /* WARNING: do no forget half pels */ |
2857 src_x = clip(src_x, -16, width); //FIXME unneeded for emu? | |
2858 if (src_x == width) | |
2859 dxy &= ~1; | |
2860 src_y = clip(src_y, -16, height); | |
2861 if (src_y == height) | |
2862 dxy &= ~2; | |
2863 src += src_y * stride + src_x; | |
2864 | |
2865 if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){ | |
2866 if( (unsigned)src_x > h_edge_pos - (motion_x&1) - w | |
2867 || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){ | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2868 ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2869 src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos); |
1633 | 2870 src= s->edge_emu_buffer; |
2871 emu=1; | |
2872 } | |
2873 } | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2874 if(field_select) |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2875 src += s->linesize; |
1633 | 2876 pix_op[dxy](dest, src, stride, h); |
2877 return emu; | |
2878 } | |
553 | 2879 |
2967 | 2880 static inline int hpel_motion_lowres(MpegEncContext *s, |
2292 | 2881 uint8_t *dest, uint8_t *src, |
2882 int field_based, int field_select, | |
2883 int src_x, int src_y, | |
2884 int width, int height, int stride, | |
2885 int h_edge_pos, int v_edge_pos, | |
2886 int w, int h, h264_chroma_mc_func *pix_op, | |
2887 int motion_x, int motion_y) | |
2888 { | |
2889 const int lowres= s->avctx->lowres; | |
2890 const int s_mask= (2<<lowres)-1; | |
2891 int emu=0; | |
2892 int sx, sy; | |
2893 | |
2894 if(s->quarter_sample){ | |
2895 motion_x/=2; | |
2896 motion_y/=2; | |
2897 } | |
2898 | |
2899 sx= motion_x & s_mask; | |
2900 sy= motion_y & s_mask; | |
2901 src_x += motion_x >> (lowres+1); | |
2902 src_y += motion_y >> (lowres+1); | |
2967 | 2903 |
2292 | 2904 src += src_y * stride + src_x; |
2905 | |
2906 if( (unsigned)src_x > h_edge_pos - (!!sx) - w | |
2907 || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){ | |
2908 ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based, | |
2909 src_x, src_y<<field_based, h_edge_pos, v_edge_pos); | |
2910 src= s->edge_emu_buffer; | |
2911 emu=1; | |
2912 } | |
2913 | |
2914 sx <<= 2 - lowres; | |
2915 sy <<= 2 - lowres; | |
2916 if(field_select) | |
2917 src += s->linesize; | |
2918 pix_op[lowres](dest, src, stride, h, sx, sy); | |
2919 return emu; | |
2920 } | |
2921 | |
0 | 2922 /* apply one mpeg motion vector to the three components */ |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2923 static always_inline void mpeg_motion(MpegEncContext *s, |
1064 | 2924 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2925 int field_based, int bottom_field, int field_select, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
2926 uint8_t **ref_picture, op_pixels_func (*pix_op)[4], |
0 | 2927 int motion_x, int motion_y, int h) |
2928 { | |
1846 | 2929 uint8_t *ptr_y, *ptr_cb, *ptr_cr; |
2930 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize; | |
2967 | 2931 |
2932 #if 0 | |
255 | 2933 if(s->quarter_sample) |
2934 { | |
2935 motion_x>>=1; | |
2936 motion_y>>=1; | |
2937 } | |
651 | 2938 #endif |
1633 | 2939 |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
2940 v_edge_pos = s->v_edge_pos >> field_based; |
1846 | 2941 linesize = s->current_picture.linesize[0] << field_based; |
1138 | 2942 uvlinesize = s->current_picture.linesize[1] << field_based; |
1633 | 2943 |
1846 | 2944 dxy = ((motion_y & 1) << 1) | (motion_x & 1); |
2945 src_x = s->mb_x* 16 + (motion_x >> 1); | |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2946 src_y =(s->mb_y<<(4-field_based)) + (motion_y >> 1); |
485 | 2947 |
0 | 2948 if (s->out_format == FMT_H263) { |
1916 | 2949 if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){ |
2950 mx = (motion_x>>1)|(motion_x&1); | |
2951 my = motion_y >>1; | |
2952 uvdxy = ((my & 1) << 1) | (mx & 1); | |
2953 uvsrc_x = s->mb_x* 8 + (mx >> 1); | |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2954 uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1); |
1916 | 2955 }else{ |
2956 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1); | |
2957 uvsrc_x = src_x>>1; | |
2958 uvsrc_y = src_y>>1; | |
2959 } | |
2044
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2042
diff
changeset
|
2960 }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261 |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2042
diff
changeset
|
2961 mx = motion_x / 4; |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2042
diff
changeset
|
2962 my = motion_y / 4; |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2042
diff
changeset
|
2963 uvdxy = 0; |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2042
diff
changeset
|
2964 uvsrc_x = s->mb_x*8 + mx; |
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2042
diff
changeset
|
2965 uvsrc_y = s->mb_y*8 + my; |
0 | 2966 } else { |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2967 if(s->chroma_y_shift){ |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2968 mx = motion_x / 2; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2969 my = motion_y / 2; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2970 uvdxy = ((my & 1) << 1) | (mx & 1); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2971 uvsrc_x = s->mb_x* 8 + (mx >> 1); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2972 uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2973 } else { |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2974 if(s->chroma_x_shift){ |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2975 //Chroma422 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2976 mx = motion_x / 2; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2977 uvdxy = ((motion_y & 1) << 1) | (mx & 1); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2978 uvsrc_x = s->mb_x* 8 + (mx >> 1); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2979 uvsrc_y = src_y; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2980 } else { |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2981 //Chroma444 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2982 uvdxy = dxy; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2983 uvsrc_x = src_x; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2984 uvsrc_y = src_y; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2985 } |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2986 } |
1846 | 2987 } |
2988 | |
2989 ptr_y = ref_picture[0] + src_y * linesize + src_x; | |
2990 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x; | |
2991 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x; | |
2992 | |
2993 if( (unsigned)src_x > s->h_edge_pos - (motion_x&1) - 16 | |
2994 || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){ | |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2995 if(s->codec_id == CODEC_ID_MPEG2VIDEO || |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2996 s->codec_id == CODEC_ID_MPEG1VIDEO){ |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2997 av_log(s->avctx,AV_LOG_DEBUG,"MPEG motion vector out of boundary\n"); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2998 return ; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
2999 } |
1846 | 3000 ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based, |
3001 src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos); | |
3002 ptr_y = s->edge_emu_buffer; | |
3003 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
1874 | 3004 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize; |
2967 | 3005 ff_emulated_edge_mc(uvbuf , ptr_cb, s->uvlinesize, 9, 9+field_based, |
1846 | 3006 uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1); |
2967 | 3007 ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based, |
1846 | 3008 uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1); |
3009 ptr_cb= uvbuf; | |
3010 ptr_cr= uvbuf+16; | |
3011 } | |
3012 } | |
3013 | |
3014 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data | |
3015 dest_y += s->linesize; | |
3016 dest_cb+= s->uvlinesize; | |
3017 dest_cr+= s->uvlinesize; | |
3018 } | |
3019 | |
3020 if(field_select){ | |
3021 ptr_y += s->linesize; | |
3022 ptr_cb+= s->uvlinesize; | |
3023 ptr_cr+= s->uvlinesize; | |
3024 } | |
3025 | |
3026 pix_op[0][dxy](dest_y, ptr_y, linesize, h); | |
2967 | 3027 |
1846 | 3028 if(!(s->flags&CODEC_FLAG_GRAY)){ |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
3029 pix_op[s->chroma_x_shift][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
3030 pix_op[s->chroma_x_shift][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift); |
1846 | 3031 } |
2639 | 3032 #if defined(CONFIG_H261_ENCODER) || defined(CONFIG_H261_DECODER) |
2291 | 3033 if(s->out_format == FMT_H261){ |
3034 ff_h261_loop_filter(s); | |
3035 } | |
2639 | 3036 #endif |
0 | 3037 } |
2256 | 3038 |
3039 /* apply one mpeg motion vector to the three components */ | |
3040 static always_inline void mpeg_motion_lowres(MpegEncContext *s, | |
3041 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
3042 int field_based, int bottom_field, int field_select, | |
3043 uint8_t **ref_picture, h264_chroma_mc_func *pix_op, | |
3044 int motion_x, int motion_y, int h) | |
3045 { | |
3046 uint8_t *ptr_y, *ptr_cb, *ptr_cr; | |
3047 int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, uvlinesize, linesize, sx, sy, uvsx, uvsy; | |
3048 const int lowres= s->avctx->lowres; | |
3049 const int block_s= 8>>lowres; | |
3050 const int s_mask= (2<<lowres)-1; | |
3051 const int h_edge_pos = s->h_edge_pos >> lowres; | |
3052 const int v_edge_pos = s->v_edge_pos >> lowres; | |
3053 linesize = s->current_picture.linesize[0] << field_based; | |
3054 uvlinesize = s->current_picture.linesize[1] << field_based; | |
3055 | |
2260 | 3056 if(s->quarter_sample){ //FIXME obviously not perfect but qpel wont work in lowres anyway |
3057 motion_x/=2; | |
3058 motion_y/=2; | |
3059 } | |
2967 | 3060 |
2260 | 3061 if(field_based){ |
3062 motion_y += (bottom_field - field_select)*((1<<lowres)-1); | |
3063 } | |
3064 | |
2256 | 3065 sx= motion_x & s_mask; |
3066 sy= motion_y & s_mask; | |
3067 src_x = s->mb_x*2*block_s + (motion_x >> (lowres+1)); | |
2260 | 3068 src_y =(s->mb_y*2*block_s>>field_based) + (motion_y >> (lowres+1)); |
2967 | 3069 |
2256 | 3070 if (s->out_format == FMT_H263) { |
2265 | 3071 uvsx = ((motion_x>>1) & s_mask) | (sx&1); |
3072 uvsy = ((motion_y>>1) & s_mask) | (sy&1); | |
2256 | 3073 uvsrc_x = src_x>>1; |
3074 uvsrc_y = src_y>>1; | |
3075 }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261 | |
3076 mx = motion_x / 4; | |
3077 my = motion_y / 4; | |
3078 uvsx = (2*mx) & s_mask; | |
3079 uvsy = (2*my) & s_mask; | |
3080 uvsrc_x = s->mb_x*block_s + (mx >> lowres); | |
3081 uvsrc_y = s->mb_y*block_s + (my >> lowres); | |
3082 } else { | |
3083 mx = motion_x / 2; | |
3084 my = motion_y / 2; | |
3085 uvsx = mx & s_mask; | |
3086 uvsy = my & s_mask; | |
3087 uvsrc_x = s->mb_x*block_s + (mx >> (lowres+1)); | |
2260 | 3088 uvsrc_y =(s->mb_y*block_s>>field_based) + (my >> (lowres+1)); |
2256 | 3089 } |
3090 | |
3091 ptr_y = ref_picture[0] + src_y * linesize + src_x; | |
3092 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x; | |
3093 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x; | |
3094 | |
3095 if( (unsigned)src_x > h_edge_pos - (!!sx) - 2*block_s | |
3096 || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){ | |
3097 ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based, | |
3098 src_x, src_y<<field_based, h_edge_pos, v_edge_pos); | |
3099 ptr_y = s->edge_emu_buffer; | |
3100 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
3101 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize; | |
2967 | 3102 ff_emulated_edge_mc(uvbuf , ptr_cb, s->uvlinesize, 9, 9+field_based, |
2256 | 3103 uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1); |
2967 | 3104 ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based, |
2256 | 3105 uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1); |
3106 ptr_cb= uvbuf; | |
3107 ptr_cr= uvbuf+16; | |
3108 } | |
3109 } | |
3110 | |
2260 | 3111 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data |
3112 dest_y += s->linesize; | |
3113 dest_cb+= s->uvlinesize; | |
3114 dest_cr+= s->uvlinesize; | |
3115 } | |
3116 | |
3117 if(field_select){ | |
3118 ptr_y += s->linesize; | |
3119 ptr_cb+= s->uvlinesize; | |
3120 ptr_cr+= s->uvlinesize; | |
3121 } | |
3122 | |
2256 | 3123 sx <<= 2 - lowres; |
3124 sy <<= 2 - lowres; | |
3125 pix_op[lowres-1](dest_y, ptr_y, linesize, h, sx, sy); | |
2967 | 3126 |
2256 | 3127 if(!(s->flags&CODEC_FLAG_GRAY)){ |
3128 uvsx <<= 2 - lowres; | |
3129 uvsy <<= 2 - lowres; | |
3130 pix_op[lowres](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy); | |
3131 pix_op[lowres](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy); | |
3132 } | |
2291 | 3133 //FIXME h261 lowres loop filter |
2256 | 3134 } |
3135 | |
1633 | 3136 //FIXME move to dsputil, avg variant, 16x16 version |
3137 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){ | |
3138 int x; | |
3139 uint8_t * const top = src[1]; | |
3140 uint8_t * const left = src[2]; | |
3141 uint8_t * const mid = src[0]; | |
3142 uint8_t * const right = src[3]; | |
3143 uint8_t * const bottom= src[4]; | |
3144 #define OBMC_FILTER(x, t, l, m, r, b)\ | |
3145 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3 | |
3146 #define OBMC_FILTER4(x, t, l, m, r, b)\ | |
3147 OBMC_FILTER(x , t, l, m, r, b);\ | |
3148 OBMC_FILTER(x+1 , t, l, m, r, b);\ | |
3149 OBMC_FILTER(x +stride, t, l, m, r, b);\ | |
3150 OBMC_FILTER(x+1+stride, t, l, m, r, b); | |
2967 | 3151 |
1633 | 3152 x=0; |
3153 OBMC_FILTER (x , 2, 2, 4, 0, 0); | |
3154 OBMC_FILTER (x+1, 2, 1, 5, 0, 0); | |
3155 OBMC_FILTER4(x+2, 2, 1, 5, 0, 0); | |
3156 OBMC_FILTER4(x+4, 2, 0, 5, 1, 0); | |
3157 OBMC_FILTER (x+6, 2, 0, 5, 1, 0); | |
3158 OBMC_FILTER (x+7, 2, 0, 4, 2, 0); | |
3159 x+= stride; | |
3160 OBMC_FILTER (x , 1, 2, 5, 0, 0); | |
3161 OBMC_FILTER (x+1, 1, 2, 5, 0, 0); | |
3162 OBMC_FILTER (x+6, 1, 0, 5, 2, 0); | |
3163 OBMC_FILTER (x+7, 1, 0, 5, 2, 0); | |
3164 x+= stride; | |
3165 OBMC_FILTER4(x , 1, 2, 5, 0, 0); | |
3166 OBMC_FILTER4(x+2, 1, 1, 6, 0, 0); | |
3167 OBMC_FILTER4(x+4, 1, 0, 6, 1, 0); | |
3168 OBMC_FILTER4(x+6, 1, 0, 5, 2, 0); | |
3169 x+= 2*stride; | |
3170 OBMC_FILTER4(x , 0, 2, 5, 0, 1); | |
3171 OBMC_FILTER4(x+2, 0, 1, 6, 0, 1); | |
3172 OBMC_FILTER4(x+4, 0, 0, 6, 1, 1); | |
3173 OBMC_FILTER4(x+6, 0, 0, 5, 2, 1); | |
3174 x+= 2*stride; | |
3175 OBMC_FILTER (x , 0, 2, 5, 0, 1); | |
3176 OBMC_FILTER (x+1, 0, 2, 5, 0, 1); | |
3177 OBMC_FILTER4(x+2, 0, 1, 5, 0, 2); | |
3178 OBMC_FILTER4(x+4, 0, 0, 5, 1, 2); | |
3179 OBMC_FILTER (x+6, 0, 0, 5, 2, 1); | |
3180 OBMC_FILTER (x+7, 0, 0, 5, 2, 1); | |
3181 x+= stride; | |
3182 OBMC_FILTER (x , 0, 2, 4, 0, 2); | |
3183 OBMC_FILTER (x+1, 0, 1, 5, 0, 2); | |
3184 OBMC_FILTER (x+6, 0, 0, 5, 1, 2); | |
3185 OBMC_FILTER (x+7, 0, 0, 4, 2, 2); | |
3186 } | |
3187 | |
3188 /* obmc for 1 8x8 luma block */ | |
3189 static inline void obmc_motion(MpegEncContext *s, | |
3190 uint8_t *dest, uint8_t *src, | |
3191 int src_x, int src_y, | |
3192 op_pixels_func *pix_op, | |
3193 int16_t mv[5][2]/* mid top left right bottom*/) | |
3194 #define MID 0 | |
3195 { | |
3196 int i; | |
3197 uint8_t *ptr[5]; | |
2967 | 3198 |
1633 | 3199 assert(s->quarter_sample==0); |
2967 | 3200 |
1633 | 3201 for(i=0; i<5; i++){ |
3202 if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){ | |
3203 ptr[i]= ptr[MID]; | |
3204 }else{ | |
1799 | 3205 ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1); |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3206 hpel_motion(s, ptr[i], src, 0, 0, |
1633 | 3207 src_x, src_y, |
3208 s->width, s->height, s->linesize, | |
3209 s->h_edge_pos, s->v_edge_pos, | |
3210 8, 8, pix_op, | |
3211 mv[i][0], mv[i][1]); | |
3212 } | |
3213 } | |
3214 | |
2967 | 3215 put_obmc(dest, ptr, s->linesize); |
1633 | 3216 } |
0 | 3217 |
255 | 3218 static inline void qpel_motion(MpegEncContext *s, |
1064 | 3219 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3220 int field_based, int bottom_field, int field_select, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3221 uint8_t **ref_picture, op_pixels_func (*pix_op)[4], |
651 | 3222 qpel_mc_func (*qpix_op)[16], |
255 | 3223 int motion_x, int motion_y, int h) |
3224 { | |
1846 | 3225 uint8_t *ptr_y, *ptr_cb, *ptr_cr; |
3226 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize; | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3227 |
255 | 3228 dxy = ((motion_y & 3) << 2) | (motion_x & 3); |
1846 | 3229 src_x = s->mb_x * 16 + (motion_x >> 2); |
255 | 3230 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2); |
3231 | |
582
5132a4ee50cd
different edge positions fixed with edge emu / dr1
michaelni
parents:
575
diff
changeset
|
3232 v_edge_pos = s->v_edge_pos >> field_based; |
255 | 3233 linesize = s->linesize << field_based; |
671 | 3234 uvlinesize = s->uvlinesize << field_based; |
2967 | 3235 |
671 | 3236 if(field_based){ |
3237 mx= motion_x/2; | |
3238 my= motion_y>>1; | |
1048 | 3239 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){ |
3240 static const int rtab[8]= {0,0,1,1,0,0,0,1}; | |
3241 mx= (motion_x>>1) + rtab[motion_x&7]; | |
3242 my= (motion_y>>1) + rtab[motion_y&7]; | |
760 | 3243 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){ |
671 | 3244 mx= (motion_x>>1)|(motion_x&1); |
3245 my= (motion_y>>1)|(motion_y&1); | |
3246 }else{ | |
3247 mx= motion_x/2; | |
3248 my= motion_y/2; | |
3249 } | |
3250 mx= (mx>>1)|(mx&1); | |
3251 my= (my>>1)|(my&1); | |
1048 | 3252 |
1846 | 3253 uvdxy= (mx&1) | ((my&1)<<1); |
671 | 3254 mx>>=1; |
3255 my>>=1; | |
255 | 3256 |
1846 | 3257 uvsrc_x = s->mb_x * 8 + mx; |
3258 uvsrc_y = s->mb_y * (8 >> field_based) + my; | |
3259 | |
3260 ptr_y = ref_picture[0] + src_y * linesize + src_x; | |
3261 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x; | |
3262 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x; | |
3263 | |
2967 | 3264 if( (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16 |
1846 | 3265 || (unsigned)src_y > v_edge_pos - (motion_y&3) - h ){ |
2967 | 3266 ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based, |
1846 | 3267 src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos); |
3268 ptr_y= s->edge_emu_buffer; | |
3269 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
1874 | 3270 uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize; |
2967 | 3271 ff_emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize, 9, 9 + field_based, |
1846 | 3272 uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1); |
2967 | 3273 ff_emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize, 9, 9 + field_based, |
1846 | 3274 uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1); |
3275 ptr_cb= uvbuf; | |
3276 ptr_cr= uvbuf + 16; | |
3277 } | |
3278 } | |
3279 | |
3280 if(!field_based) | |
3281 qpix_op[0][dxy](dest_y, ptr_y, linesize); | |
3282 else{ | |
3283 if(bottom_field){ | |
3284 dest_y += s->linesize; | |
3285 dest_cb+= s->uvlinesize; | |
3286 dest_cr+= s->uvlinesize; | |
3287 } | |
3288 | |
3289 if(field_select){ | |
3290 ptr_y += s->linesize; | |
3291 ptr_cb += s->uvlinesize; | |
3292 ptr_cr += s->uvlinesize; | |
3293 } | |
3294 //damn interlaced mode | |
3295 //FIXME boundary mirroring is not exactly correct here | |
3296 qpix_op[1][dxy](dest_y , ptr_y , linesize); | |
3297 qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize); | |
3298 } | |
3299 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
3300 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1); | |
3301 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1); | |
3302 } | |
255 | 3303 } |
3304 | |
1013 | 3305 inline int ff_h263_round_chroma(int x){ |
3306 if (x >= 0) | |
3307 return (h263_chroma_roundtab[x & 0xf] + ((x >> 3) & ~1)); | |
3308 else { | |
3309 x = -x; | |
3310 return -(h263_chroma_roundtab[x & 0xf] + ((x >> 3) & ~1)); | |
3311 } | |
3312 } | |
255 | 3313 |
1225 | 3314 /** |
1633 | 3315 * h263 chorma 4mv motion compensation. |
3316 */ | |
3317 static inline void chroma_4mv_motion(MpegEncContext *s, | |
3318 uint8_t *dest_cb, uint8_t *dest_cr, | |
3319 uint8_t **ref_picture, | |
3320 op_pixels_func *pix_op, | |
3321 int mx, int my){ | |
3322 int dxy, emu=0, src_x, src_y, offset; | |
3323 uint8_t *ptr; | |
2967 | 3324 |
1633 | 3325 /* In case of 8X8, we construct a single chroma motion vector |
3326 with a special rounding */ | |
3327 mx= ff_h263_round_chroma(mx); | |
3328 my= ff_h263_round_chroma(my); | |
2967 | 3329 |
1633 | 3330 dxy = ((my & 1) << 1) | (mx & 1); |
3331 mx >>= 1; | |
3332 my >>= 1; | |
3333 | |
3334 src_x = s->mb_x * 8 + mx; | |
3335 src_y = s->mb_y * 8 + my; | |
3336 src_x = clip(src_x, -8, s->width/2); | |
3337 if (src_x == s->width/2) | |
3338 dxy &= ~1; | |
3339 src_y = clip(src_y, -8, s->height/2); | |
3340 if (src_y == s->height/2) | |
3341 dxy &= ~2; | |
2967 | 3342 |
1633 | 3343 offset = (src_y * (s->uvlinesize)) + src_x; |
3344 ptr = ref_picture[1] + offset; | |
3345 if(s->flags&CODEC_FLAG_EMU_EDGE){ | |
3346 if( (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8 | |
3347 || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){ | |
3348 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); | |
3349 ptr= s->edge_emu_buffer; | |
3350 emu=1; | |
3351 } | |
3352 } | |
3353 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8); | |
3354 | |
3355 ptr = ref_picture[2] + offset; | |
3356 if(emu){ | |
3357 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); | |
3358 ptr= s->edge_emu_buffer; | |
3359 } | |
3360 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8); | |
3361 } | |
3362 | |
2292 | 3363 static inline void chroma_4mv_motion_lowres(MpegEncContext *s, |
3364 uint8_t *dest_cb, uint8_t *dest_cr, | |
3365 uint8_t **ref_picture, | |
3366 h264_chroma_mc_func *pix_op, | |
3367 int mx, int my){ | |
3368 const int lowres= s->avctx->lowres; | |
3369 const int block_s= 8>>lowres; | |
3370 const int s_mask= (2<<lowres)-1; | |
3371 const int h_edge_pos = s->h_edge_pos >> (lowres+1); | |
3372 const int v_edge_pos = s->v_edge_pos >> (lowres+1); | |
3373 int emu=0, src_x, src_y, offset, sx, sy; | |
3374 uint8_t *ptr; | |
2967 | 3375 |
2292 | 3376 if(s->quarter_sample){ |
3377 mx/=2; | |
3378 my/=2; | |
3379 } | |
3380 | |
3381 /* In case of 8X8, we construct a single chroma motion vector | |
3382 with a special rounding */ | |
3383 mx= ff_h263_round_chroma(mx); | |
3384 my= ff_h263_round_chroma(my); | |
2967 | 3385 |
2292 | 3386 sx= mx & s_mask; |
3387 sy= my & s_mask; | |
3388 src_x = s->mb_x*block_s + (mx >> (lowres+1)); | |
3389 src_y = s->mb_y*block_s + (my >> (lowres+1)); | |
2967 | 3390 |
2292 | 3391 offset = src_y * s->uvlinesize + src_x; |
3392 ptr = ref_picture[1] + offset; | |
3393 if(s->flags&CODEC_FLAG_EMU_EDGE){ | |
3394 if( (unsigned)src_x > h_edge_pos - (!!sx) - block_s | |
3395 || (unsigned)src_y > v_edge_pos - (!!sy) - block_s){ | |
3396 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos); | |
3397 ptr= s->edge_emu_buffer; | |
3398 emu=1; | |
3399 } | |
2967 | 3400 } |
2292 | 3401 sx <<= 2 - lowres; |
3402 sy <<= 2 - lowres; | |
3403 pix_op[lowres](dest_cb, ptr, s->uvlinesize, block_s, sx, sy); | |
2967 | 3404 |
2292 | 3405 ptr = ref_picture[2] + offset; |
3406 if(emu){ | |
3407 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos); | |
3408 ptr= s->edge_emu_buffer; | |
3409 } | |
3410 pix_op[lowres](dest_cr, ptr, s->uvlinesize, block_s, sx, sy); | |
3411 } | |
3412 | |
3216
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3413 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){ |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3414 /* fetch pixels for estimated mv 4 macroblocks ahead |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3415 * optimized for 64byte cache lines */ |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3416 const int shift = s->quarter_sample ? 2 : 1; |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3417 const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8; |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3418 const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y; |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3419 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64; |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3420 s->dsp.prefetch(pix[0]+off, s->linesize, 4); |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3421 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64; |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3422 s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2); |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3423 } |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3424 |
1633 | 3425 /** |
2764 | 3426 * motion compensation of a single macroblock |
1225 | 3427 * @param s context |
3428 * @param dest_y luma destination pointer | |
3429 * @param dest_cb chroma cb/u destination pointer | |
3430 * @param dest_cr chroma cr/v destination pointer | |
3431 * @param dir direction (0->forward, 1->backward) | |
3432 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture | |
3433 * @param pic_op halfpel motion compensation function (average or put normally) | |
3434 * @param pic_op qpel motion compensation function (average or put normally) | |
3435 * the motion vectors are taken from s->mv and the MV type from s->mv_type | |
3436 */ | |
2967 | 3437 static inline void MPV_motion(MpegEncContext *s, |
1064 | 3438 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
2967 | 3439 int dir, uint8_t **ref_picture, |
651 | 3440 op_pixels_func (*pix_op)[4], qpel_mc_func (*qpix_op)[16]) |
0 | 3441 { |
1633 | 3442 int dxy, mx, my, src_x, src_y, motion_x, motion_y; |
0 | 3443 int mb_x, mb_y, i; |
1064 | 3444 uint8_t *ptr, *dest; |
0 | 3445 |
3446 mb_x = s->mb_x; | |
3447 mb_y = s->mb_y; | |
3448 | |
3216
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3449 prefetch_motion(s, ref_picture, dir); |
4186eb8d29a1
prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).
lorenm
parents:
3145
diff
changeset
|
3450 |
1655 | 3451 if(s->obmc && s->pict_type != B_TYPE){ |
1633 | 3452 int16_t mv_cache[4][4][2]; |
3453 const int xy= s->mb_x + s->mb_y*s->mb_stride; | |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
3454 const int mot_stride= s->b8_stride; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
3455 const int mot_xy= mb_x*2 + mb_y*2*mot_stride; |
1633 | 3456 |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
3457 assert(!s->mb_skipped); |
2967 | 3458 |
1668
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3459 memcpy(mv_cache[1][1], s->current_picture.motion_val[0][mot_xy ], sizeof(int16_t)*4); |
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3460 memcpy(mv_cache[2][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4); |
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3461 memcpy(mv_cache[3][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4); |
1633 | 3462 |
3463 if(mb_y==0 || IS_INTRA(s->current_picture.mb_type[xy-s->mb_stride])){ | |
3464 memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4); | |
3465 }else{ | |
1668
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3466 memcpy(mv_cache[0][1], s->current_picture.motion_val[0][mot_xy-mot_stride], sizeof(int16_t)*4); |
1633 | 3467 } |
3468 | |
3469 if(mb_x==0 || IS_INTRA(s->current_picture.mb_type[xy-1])){ | |
3470 *(int32_t*)mv_cache[1][0]= *(int32_t*)mv_cache[1][1]; | |
3471 *(int32_t*)mv_cache[2][0]= *(int32_t*)mv_cache[2][1]; | |
3472 }else{ | |
1668
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3473 *(int32_t*)mv_cache[1][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1]; |
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3474 *(int32_t*)mv_cache[2][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1+mot_stride]; |
1633 | 3475 } |
3476 | |
3477 if(mb_x+1>=s->mb_width || IS_INTRA(s->current_picture.mb_type[xy+1])){ | |
3478 *(int32_t*)mv_cache[1][3]= *(int32_t*)mv_cache[1][2]; | |
3479 *(int32_t*)mv_cache[2][3]= *(int32_t*)mv_cache[2][2]; | |
3480 }else{ | |
1668
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3481 *(int32_t*)mv_cache[1][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2]; |
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
3482 *(int32_t*)mv_cache[2][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2+mot_stride]; |
1633 | 3483 } |
2967 | 3484 |
1633 | 3485 mx = 0; |
3486 my = 0; | |
3487 for(i=0;i<4;i++) { | |
3488 const int x= (i&1)+1; | |
3489 const int y= (i>>1)+1; | |
3490 int16_t mv[5][2]= { | |
3491 {mv_cache[y][x ][0], mv_cache[y][x ][1]}, | |
3492 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]}, | |
3493 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]}, | |
3494 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]}, | |
3495 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}}; | |
3496 //FIXME cleanup | |
3497 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize, | |
3498 ref_picture[0], | |
3499 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8, | |
3500 pix_op[1], | |
3501 mv); | |
3502 | |
3503 mx += mv[0][0]; | |
3504 my += mv[0][1]; | |
3505 } | |
3506 if(!(s->flags&CODEC_FLAG_GRAY)) | |
3507 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my); | |
3508 | |
3509 return; | |
3510 } | |
2967 | 3511 |
0 | 3512 switch(s->mv_type) { |
3513 case MV_TYPE_16X16: | |
255 | 3514 if(s->mcsel){ |
753 | 3515 if(s->real_sprite_warping_points==1){ |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3516 gmc1_motion(s, dest_y, dest_cb, dest_cr, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3517 ref_picture); |
753 | 3518 }else{ |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3519 gmc_motion(s, dest_y, dest_cb, dest_cr, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3520 ref_picture); |
753 | 3521 } |
651 | 3522 }else if(s->quarter_sample){ |
2967 | 3523 qpel_motion(s, dest_y, dest_cb, dest_cr, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3524 0, 0, 0, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3525 ref_picture, pix_op, qpix_op, |
255 | 3526 s->mv[dir][0][0], s->mv[dir][0][1], 16); |
936 | 3527 }else if(s->mspel){ |
3528 ff_mspel_motion(s, dest_y, dest_cb, dest_cr, | |
3529 ref_picture, pix_op, | |
3530 s->mv[dir][0][0], s->mv[dir][0][1], 16); | |
1042 | 3531 }else |
3532 { | |
2967 | 3533 mpeg_motion(s, dest_y, dest_cb, dest_cr, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3534 0, 0, 0, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3535 ref_picture, pix_op, |
255 | 3536 s->mv[dir][0][0], s->mv[dir][0][1], 16); |
2967 | 3537 } |
0 | 3538 break; |
3539 case MV_TYPE_8X8: | |
673 | 3540 mx = 0; |
3541 my = 0; | |
3542 if(s->quarter_sample){ | |
3543 for(i=0;i<4;i++) { | |
3544 motion_x = s->mv[dir][i][0]; | |
3545 motion_y = s->mv[dir][i][1]; | |
0 | 3546 |
673 | 3547 dxy = ((motion_y & 3) << 2) | (motion_x & 3); |
3548 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8; | |
3549 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8; | |
2967 | 3550 |
673 | 3551 /* WARNING: do no forget half pels */ |
3552 src_x = clip(src_x, -16, s->width); | |
3553 if (src_x == s->width) | |
3554 dxy &= ~3; | |
3555 src_y = clip(src_y, -16, s->height); | |
3556 if (src_y == s->height) | |
3557 dxy &= ~12; | |
2967 | 3558 |
673 | 3559 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x); |
3560 if(s->flags&CODEC_FLAG_EMU_EDGE){ | |
2967 | 3561 if( (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 8 |
1542
576861d6343a
emu_edge checks simplification and avoid redundant checks for mpeg1/2 if emu_edge is set
michael
parents:
1539
diff
changeset
|
3562 || (unsigned)src_y > s->v_edge_pos - (motion_y&3) - 8 ){ |
1317
26c44d2433c1
make ff_emulated_edge_mc() independant of MpegEncContext
michaelni
parents:
1315
diff
changeset
|
3563 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize, 9, 9, src_x, src_y, s->h_edge_pos, s->v_edge_pos); |
673 | 3564 ptr= s->edge_emu_buffer; |
3565 } | |
3566 } | |
3567 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize; | |
3568 qpix_op[1][dxy](dest, ptr, s->linesize); | |
3569 | |
3570 mx += s->mv[dir][i][0]/2; | |
3571 my += s->mv[dir][i][1]/2; | |
3572 } | |
3573 }else{ | |
3574 for(i=0;i<4;i++) { | |
1633 | 3575 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize, |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3576 ref_picture[0], 0, 0, |
1633 | 3577 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8, |
3578 s->width, s->height, s->linesize, | |
3579 s->h_edge_pos, s->v_edge_pos, | |
3580 8, 8, pix_op[1], | |
3581 s->mv[dir][i][0], s->mv[dir][i][1]); | |
673 | 3582 |
3583 mx += s->mv[dir][i][0]; | |
3584 my += s->mv[dir][i][1]; | |
554 | 3585 } |
0 | 3586 } |
673 | 3587 |
1633 | 3588 if(!(s->flags&CODEC_FLAG_GRAY)) |
3589 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my); | |
0 | 3590 break; |
3591 case MV_TYPE_FIELD: | |
3592 if (s->picture_structure == PICT_FRAME) { | |
671 | 3593 if(s->quarter_sample){ |
1843 | 3594 for(i=0; i<2; i++){ |
3595 qpel_motion(s, dest_y, dest_cb, dest_cr, | |
3596 1, i, s->field_select[dir][i], | |
3597 ref_picture, pix_op, qpix_op, | |
3598 s->mv[dir][i][0], s->mv[dir][i][1], 8); | |
3599 } | |
671 | 3600 }else{ |
2967 | 3601 /* top field */ |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3602 mpeg_motion(s, dest_y, dest_cb, dest_cr, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3603 1, 0, s->field_select[dir][0], |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3604 ref_picture, pix_op, |
671 | 3605 s->mv[dir][0][0], s->mv[dir][0][1], 8); |
3606 /* bottom field */ | |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3607 mpeg_motion(s, dest_y, dest_cb, dest_cr, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3608 1, 1, s->field_select[dir][1], |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3609 ref_picture, pix_op, |
671 | 3610 s->mv[dir][1][0], s->mv[dir][1][1], 8); |
3611 } | |
0 | 3612 } else { |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3613 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != B_TYPE && !s->first_field){ |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3614 ref_picture= s->current_picture_ptr->data; |
2967 | 3615 } |
1096
5e6e505d8997
field picture decoding support (16x16 MC blocks only as i dont have any samples which use other modes ...)
michaelni
parents:
1095
diff
changeset
|
3616 |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3617 mpeg_motion(s, dest_y, dest_cb, dest_cr, |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3618 0, 0, s->field_select[dir][0], |
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3619 ref_picture, pix_op, |
1096
5e6e505d8997
field picture decoding support (16x16 MC blocks only as i dont have any samples which use other modes ...)
michaelni
parents:
1095
diff
changeset
|
3620 s->mv[dir][0][0], s->mv[dir][0][1], 16); |
0 | 3621 } |
3622 break; | |
1843 | 3623 case MV_TYPE_16X8: |
3624 for(i=0; i<2; i++){ | |
3625 uint8_t ** ref2picture; | |
3626 | |
3627 if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == B_TYPE || s->first_field){ | |
1339
338a2f6e6402
Mpeg2 16x8 Patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1328
diff
changeset
|
3628 ref2picture= ref_picture; |
338a2f6e6402
Mpeg2 16x8 Patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1328
diff
changeset
|
3629 }else{ |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3630 ref2picture= s->current_picture_ptr->data; |
2967 | 3631 } |
3632 | |
3633 mpeg_motion(s, dest_y, dest_cb, dest_cr, | |
1843 | 3634 0, 0, s->field_select[dir][i], |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3635 ref2picture, pix_op, |
1843 | 3636 s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8); |
2967 | 3637 |
1843 | 3638 dest_y += 16*s->linesize; |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
3639 dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
3640 dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize; |
2967 | 3641 } |
1339
338a2f6e6402
Mpeg2 16x8 Patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1328
diff
changeset
|
3642 break; |
1326
6cdd3b8f4fd3
DMV support patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1317
diff
changeset
|
3643 case MV_TYPE_DMV: |
6cdd3b8f4fd3
DMV support patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1317
diff
changeset
|
3644 if(s->picture_structure == PICT_FRAME){ |
1843 | 3645 for(i=0; i<2; i++){ |
3646 int j; | |
3647 for(j=0; j<2; j++){ | |
3648 mpeg_motion(s, dest_y, dest_cb, dest_cr, | |
3649 1, j, j^i, | |
3650 ref_picture, pix_op, | |
3651 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8); | |
3652 } | |
2967 | 3653 pix_op = s->dsp.avg_pixels_tab; |
1843 | 3654 } |
1326
6cdd3b8f4fd3
DMV support patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1317
diff
changeset
|
3655 }else{ |
1843 | 3656 for(i=0; i<2; i++){ |
2967 | 3657 mpeg_motion(s, dest_y, dest_cb, dest_cr, |
1843 | 3658 0, 0, s->picture_structure != i+1, |
3659 ref_picture, pix_op, | |
3660 s->mv[dir][2*i][0],s->mv[dir][2*i][1],16); | |
3661 | |
3662 // after put we make avg of the same block | |
2967 | 3663 pix_op=s->dsp.avg_pixels_tab; |
1843 | 3664 |
3665 //opposite parity is always in the same frame if this is second field | |
3666 if(!s->first_field){ | |
2967 | 3667 ref_picture = s->current_picture_ptr->data; |
1843 | 3668 } |
1842
0d82db458a3c
interlaced mpeg4 + edge emu fix (fixes decoding of mermaid.avi)
michael
parents:
1838
diff
changeset
|
3669 } |
1326
6cdd3b8f4fd3
DMV support patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1317
diff
changeset
|
3670 } |
6cdd3b8f4fd3
DMV support patch by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1317
diff
changeset
|
3671 break; |
1633 | 3672 default: assert(0); |
0 | 3673 } |
3674 } | |
3675 | |
2256 | 3676 /** |
2764 | 3677 * motion compensation of a single macroblock |
2256 | 3678 * @param s context |
3679 * @param dest_y luma destination pointer | |
3680 * @param dest_cb chroma cb/u destination pointer | |
3681 * @param dest_cr chroma cr/v destination pointer | |
3682 * @param dir direction (0->forward, 1->backward) | |
3683 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture | |
3684 * @param pic_op halfpel motion compensation function (average or put normally) | |
3685 * the motion vectors are taken from s->mv and the MV type from s->mv_type | |
3686 */ | |
2967 | 3687 static inline void MPV_motion_lowres(MpegEncContext *s, |
2256 | 3688 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
2967 | 3689 int dir, uint8_t **ref_picture, |
2256 | 3690 h264_chroma_mc_func *pix_op) |
3691 { | |
2292 | 3692 int mx, my; |
2260 | 3693 int mb_x, mb_y, i; |
3694 const int lowres= s->avctx->lowres; | |
2967 | 3695 const int block_s= 8>>lowres; |
2260 | 3696 |
3697 mb_x = s->mb_x; | |
3698 mb_y = s->mb_y; | |
3699 | |
3700 switch(s->mv_type) { | |
3701 case MV_TYPE_16X16: | |
2967 | 3702 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
2260 | 3703 0, 0, 0, |
3704 ref_picture, pix_op, | |
3705 s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s); | |
3706 break; | |
2292 | 3707 case MV_TYPE_8X8: |
2260 | 3708 mx = 0; |
3709 my = 0; | |
3710 for(i=0;i<4;i++) { | |
2292 | 3711 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) * s->linesize)*block_s, |
2260 | 3712 ref_picture[0], 0, 0, |
2292 | 3713 (2*mb_x + (i & 1))*block_s, (2*mb_y + (i >>1))*block_s, |
2260 | 3714 s->width, s->height, s->linesize, |
2292 | 3715 s->h_edge_pos >> lowres, s->v_edge_pos >> lowres, |
3716 block_s, block_s, pix_op, | |
2260 | 3717 s->mv[dir][i][0], s->mv[dir][i][1]); |
3718 | |
3719 mx += s->mv[dir][i][0]; | |
3720 my += s->mv[dir][i][1]; | |
3721 } | |
3722 | |
3723 if(!(s->flags&CODEC_FLAG_GRAY)) | |
2292 | 3724 chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture, pix_op, mx, my); |
3725 break; | |
2260 | 3726 case MV_TYPE_FIELD: |
3727 if (s->picture_structure == PICT_FRAME) { | |
2967 | 3728 /* top field */ |
2260 | 3729 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
3730 1, 0, s->field_select[dir][0], | |
3731 ref_picture, pix_op, | |
3732 s->mv[dir][0][0], s->mv[dir][0][1], block_s); | |
3733 /* bottom field */ | |
3734 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, | |
3735 1, 1, s->field_select[dir][1], | |
3736 ref_picture, pix_op, | |
3737 s->mv[dir][1][0], s->mv[dir][1][1], block_s); | |
3738 } else { | |
3739 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != B_TYPE && !s->first_field){ | |
3740 ref_picture= s->current_picture_ptr->data; | |
2967 | 3741 } |
2260 | 3742 |
3743 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, | |
3744 0, 0, s->field_select[dir][0], | |
3745 ref_picture, pix_op, | |
3746 s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s); | |
3747 } | |
3748 break; | |
3749 case MV_TYPE_16X8: | |
3750 for(i=0; i<2; i++){ | |
3751 uint8_t ** ref2picture; | |
3752 | |
3753 if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == B_TYPE || s->first_field){ | |
3754 ref2picture= ref_picture; | |
3755 }else{ | |
3756 ref2picture= s->current_picture_ptr->data; | |
2967 | 3757 } |
3758 | |
3759 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, | |
2260 | 3760 0, 0, s->field_select[dir][i], |
3761 ref2picture, pix_op, | |
3762 s->mv[dir][i][0], s->mv[dir][i][1] + 2*block_s*i, block_s); | |
2967 | 3763 |
2260 | 3764 dest_y += 2*block_s*s->linesize; |
3765 dest_cb+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize; | |
3766 dest_cr+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize; | |
2967 | 3767 } |
2260 | 3768 break; |
3769 case MV_TYPE_DMV: | |
3770 if(s->picture_structure == PICT_FRAME){ | |
3771 for(i=0; i<2; i++){ | |
3772 int j; | |
3773 for(j=0; j<2; j++){ | |
3774 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, | |
3775 1, j, j^i, | |
3776 ref_picture, pix_op, | |
3777 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], block_s); | |
3778 } | |
3779 pix_op = s->dsp.avg_h264_chroma_pixels_tab; | |
3780 } | |
3781 }else{ | |
3782 for(i=0; i<2; i++){ | |
2967 | 3783 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
2260 | 3784 0, 0, s->picture_structure != i+1, |
3785 ref_picture, pix_op, | |
3786 s->mv[dir][2*i][0],s->mv[dir][2*i][1],2*block_s); | |
3787 | |
3788 // after put we make avg of the same block | |
3789 pix_op = s->dsp.avg_h264_chroma_pixels_tab; | |
3790 | |
3791 //opposite parity is always in the same frame if this is second field | |
3792 if(!s->first_field){ | |
2967 | 3793 ref_picture = s->current_picture_ptr->data; |
2260 | 3794 } |
3795 } | |
3796 } | |
3797 break; | |
3798 default: assert(0); | |
3799 } | |
2256 | 3800 } |
0 | 3801 |
3802 /* put block[] to dest[] */ | |
2967 | 3803 static inline void put_dct(MpegEncContext *s, |
1651 | 3804 DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale) |
0 | 3805 { |
1689 | 3806 s->dct_unquantize_intra(s, block, i, qscale); |
1092 | 3807 s->dsp.idct_put (dest, line_size, block); |
0 | 3808 } |
3809 | |
3810 /* add block[] to dest[] */ | |
2967 | 3811 static inline void add_dct(MpegEncContext *s, |
1064 | 3812 DCTELEM *block, int i, uint8_t *dest, int line_size) |
0 | 3813 { |
3814 if (s->block_last_index[i] >= 0) { | |
1092 | 3815 s->dsp.idct_add (dest, line_size, block); |
481 | 3816 } |
3817 } | |
3818 | |
2967 | 3819 static inline void add_dequant_dct(MpegEncContext *s, |
1644 | 3820 DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale) |
481 | 3821 { |
3822 if (s->block_last_index[i] >= 0) { | |
1689 | 3823 s->dct_unquantize_inter(s, block, i, qscale); |
324 | 3824 |
1092 | 3825 s->dsp.idct_add (dest, line_size, block); |
0 | 3826 } |
3827 } | |
3828 | |
456 | 3829 /** |
3830 * cleans dc, ac, coded_block for the current non intra MB | |
3831 */ | |
3832 void ff_clean_intra_table_entries(MpegEncContext *s) | |
3833 { | |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
3834 int wrap = s->b8_stride; |
456 | 3835 int xy = s->block_index[0]; |
2967 | 3836 |
3837 s->dc_val[0][xy ] = | |
3838 s->dc_val[0][xy + 1 ] = | |
456 | 3839 s->dc_val[0][xy + wrap] = |
3840 s->dc_val[0][xy + 1 + wrap] = 1024; | |
3841 /* ac pred */ | |
1064 | 3842 memset(s->ac_val[0][xy ], 0, 32 * sizeof(int16_t)); |
3843 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t)); | |
456 | 3844 if (s->msmpeg4_version>=3) { |
3845 s->coded_block[xy ] = | |
3846 s->coded_block[xy + 1 ] = | |
3847 s->coded_block[xy + wrap] = | |
3848 s->coded_block[xy + 1 + wrap] = 0; | |
3849 } | |
3850 /* chroma */ | |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
3851 wrap = s->mb_stride; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
3852 xy = s->mb_x + s->mb_y * wrap; |
456 | 3853 s->dc_val[1][xy] = |
3854 s->dc_val[2][xy] = 1024; | |
3855 /* ac pred */ | |
1064 | 3856 memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t)); |
3857 memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t)); | |
2967 | 3858 |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
3859 s->mbintra_table[xy]= 0; |
456 | 3860 } |
3861 | |
0 | 3862 /* generic function called after a macroblock has been parsed by the |
3863 decoder or after it has been encoded by the encoder. | |
3864 | |
3865 Important variables used: | |
3866 s->mb_intra : true if intra macroblock | |
3867 s->mv_dir : motion vector direction | |
3868 s->mv_type : motion vector type | |
3869 s->mv : motion vector | |
3870 s->interlaced_dct : true if interlaced dct used (mpeg2) | |
3871 */ | |
2256 | 3872 static always_inline void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64], int lowres_flag) |
0 | 3873 { |
244 | 3874 int mb_x, mb_y; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
3875 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x; |
1381 | 3876 #ifdef HAVE_XVMC |
3877 if(s->avctx->xvmc_acceleration){ | |
1580
628bf341e099
XvMC speedup by removing one memcpy and doing MB packing
iive
parents:
1579
diff
changeset
|
3878 XVMC_decode_mb(s);//xvmc uses pblocks |
1381 | 3879 return; |
3880 } | |
3881 #endif | |
0 | 3882 |
3883 mb_x = s->mb_x; | |
3884 mb_y = s->mb_y; | |
3885 | |
1833
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3886 if(s->avctx->debug&FF_DEBUG_DCT_COEFF) { |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3887 /* save DCT coefficients */ |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3888 int i,j; |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3889 DCTELEM *dct = &s->current_picture.dct_coeff[mb_xy*64*6]; |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3890 for(i=0; i<6; i++) |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3891 for(j=0; j<64; j++) |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3892 *dct++ = block[i][s->dsp.idct_permutation[j]]; |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3893 } |
ba27ce284d07
export DCT coefficients patch by (Wolfgang Hesseler <wolfgang.hesseler at imk dot fraunhofer dot de>)
michael
parents:
1831
diff
changeset
|
3894 |
903 | 3895 s->current_picture.qscale_table[mb_xy]= s->qscale; |
108
1e4a4af694d1
exporting qscale data for postprocessing (for MPlayer)
arpi_esp
parents:
79
diff
changeset
|
3896 |
0 | 3897 /* update DC predictors for P macroblocks */ |
3898 if (!s->mb_intra) { | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
3899 if (s->h263_pred || s->h263_aic) { |
481 | 3900 if(s->mbintra_table[mb_xy]) |
456 | 3901 ff_clean_intra_table_entries(s); |
0 | 3902 } else { |
456 | 3903 s->last_dc[0] = |
3904 s->last_dc[1] = | |
0 | 3905 s->last_dc[2] = 128 << s->intra_dc_precision; |
3906 } | |
3907 } | |
249
42a0b7b16738
- Bug fixes in H.263+ Advanced INTRA Coding decoder.
pulento
parents:
248
diff
changeset
|
3908 else if (s->h263_pred || s->h263_aic) |
481 | 3909 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
|
3910 |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
3911 if ((s->flags&CODEC_FLAG_PSNR) || !(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) { //FIXME precalc |
1064 | 3912 uint8_t *dest_y, *dest_cb, *dest_cr; |
481 | 3913 int dct_linesize, dct_offset; |
651 | 3914 op_pixels_func (*op_pix)[4]; |
3915 qpel_mc_func (*op_qpix)[16]; | |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
3916 const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics |
1138 | 3917 const int uvlinesize= s->current_picture.linesize[1]; |
2256 | 3918 const int readable= s->pict_type != B_TYPE || s->encoding || s->avctx->draw_horiz_band || lowres_flag; |
3919 const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8; | |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
3920 |
903 | 3921 /* avoid copy if macroblock skipped in last frame too */ |
3922 /* skip only during decoding as we might trash the buffers during encoding a bit */ | |
3923 if(!s->encoding){ | |
1064 | 3924 uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy]; |
903 | 3925 const int age= s->current_picture.age; |
3926 | |
3927 assert(age); | |
3928 | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
3929 if (s->mb_skipped) { |
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
3930 s->mb_skipped= 0; |
903 | 3931 assert(s->pict_type!=I_TYPE); |
2967 | 3932 |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
3933 (*mbskip_ptr) ++; /* indicate that this time we skipped it */ |
556 | 3934 if(*mbskip_ptr >99) *mbskip_ptr= 99; |
3935 | |
903 | 3936 /* if previous was skipped too, then nothing to do ! */ |
1021
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
3937 if (*mbskip_ptr >= age && s->current_picture.reference){ |
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
3938 return; |
903 | 3939 } |
1021
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
3940 } else if(!s->current_picture.reference){ |
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
3941 (*mbskip_ptr) ++; /* increase counter so the age can be compared cleanly */ |
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
3942 if(*mbskip_ptr >99) *mbskip_ptr= 99; |
2d7c9f5738de
trying to fix mb skip bug in mpeg1/2 if slices are not used
michaelni
parents:
1014
diff
changeset
|
3943 } else{ |
7
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
3944 *mbskip_ptr = 0; /* not skipped */ |
1d3ac9654178
added skip macroblock optimization (big perf win on black regions for example)
glantau
parents:
0
diff
changeset
|
3945 } |
717 | 3946 } |
2967 | 3947 |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
3948 dct_linesize = linesize << s->interlaced_dct; |
2256 | 3949 dct_offset =(s->interlaced_dct)? linesize : linesize*block_size; |
2967 | 3950 |
1632 | 3951 if(readable){ |
3952 dest_y= s->dest[0]; | |
3953 dest_cb= s->dest[1]; | |
3954 dest_cr= s->dest[2]; | |
3955 }else{ | |
1799 | 3956 dest_y = s->b_scratchpad; |
1834 | 3957 dest_cb= s->b_scratchpad+16*linesize; |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
3958 dest_cr= s->b_scratchpad+32*linesize; |
1632 | 3959 } |
2256 | 3960 |
0 | 3961 if (!s->mb_intra) { |
3962 /* motion handling */ | |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
3963 /* decoding or more than one mb_type (MC was already done otherwise) */ |
1389 | 3964 if(!s->encoding){ |
2256 | 3965 if(lowres_flag){ |
3966 h264_chroma_mc_func *op_pix = s->dsp.put_h264_chroma_pixels_tab; | |
3967 | |
3968 if (s->mv_dir & MV_DIR_FORWARD) { | |
3969 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix); | |
3970 op_pix = s->dsp.avg_h264_chroma_pixels_tab; | |
3971 } | |
3972 if (s->mv_dir & MV_DIR_BACKWARD) { | |
3973 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix); | |
3974 } | |
324 | 3975 }else{ |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3780
diff
changeset
|
3976 op_qpix= s->me.qpel_put; |
2967 | 3977 if ((!s->no_rounding) || s->pict_type==B_TYPE){ |
2256 | 3978 op_pix = s->dsp.put_pixels_tab; |
3979 }else{ | |
3980 op_pix = s->dsp.put_no_rnd_pixels_tab; | |
3981 } | |
3982 if (s->mv_dir & MV_DIR_FORWARD) { | |
3983 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix); | |
3984 op_pix = s->dsp.avg_pixels_tab; | |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3780
diff
changeset
|
3985 op_qpix= s->me.qpel_avg; |
2256 | 3986 } |
3987 if (s->mv_dir & MV_DIR_BACKWARD) { | |
3988 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix); | |
3989 } | |
324 | 3990 } |
0 | 3991 } |
3992 | |
481 | 3993 /* skip dequant / idct if we are really late ;) */ |
2792 | 3994 if(s->hurry_up>1) goto skip_idct; |
3995 if(s->avctx->skip_idct){ | |
3996 if( (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == B_TYPE) | |
3997 ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != I_TYPE) | |
3998 || s->avctx->skip_idct >= AVDISCARD_ALL) | |
3999 goto skip_idct; | |
4000 } | |
481 | 4001 |
0 | 4002 /* add dct residue */ |
1421 | 4003 if(s->encoding || !( s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO |
711 | 4004 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){ |
2256 | 4005 add_dequant_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale); |
4006 add_dequant_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale); | |
4007 add_dequant_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale); | |
4008 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale); | |
0 | 4009 |
485 | 4010 if(!(s->flags&CODEC_FLAG_GRAY)){ |
3309 | 4011 if (s->chroma_y_shift){ |
4012 add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale); | |
4013 add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale); | |
4014 }else{ | |
4015 dct_linesize >>= 1; | |
4016 dct_offset >>=1; | |
4017 add_dequant_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale); | |
4018 add_dequant_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale); | |
4019 add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale); | |
4020 add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale); | |
4021 } | |
485 | 4022 } |
936 | 4023 } else if(s->codec_id != CODEC_ID_WMV2){ |
2256 | 4024 add_dct(s, block[0], 0, dest_y , dct_linesize); |
4025 add_dct(s, block[1], 1, dest_y + block_size, dct_linesize); | |
4026 add_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize); | |
4027 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize); | |
481 | 4028 |
485 | 4029 if(!(s->flags&CODEC_FLAG_GRAY)){ |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4030 if(s->chroma_y_shift){//Chroma420 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4031 add_dct(s, block[4], 4, dest_cb, uvlinesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4032 add_dct(s, block[5], 5, dest_cr, uvlinesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4033 }else{ |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4034 //chroma422 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4035 dct_linesize = uvlinesize << s->interlaced_dct; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4036 dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4037 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4038 add_dct(s, block[4], 4, dest_cb, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4039 add_dct(s, block[5], 5, dest_cr, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4040 add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4041 add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4042 if(!s->chroma_x_shift){//Chroma444 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4043 add_dct(s, block[8], 8, dest_cb+8, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4044 add_dct(s, block[9], 9, dest_cr+8, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4045 add_dct(s, block[10], 10, dest_cb+8+dct_offset, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4046 add_dct(s, block[11], 11, dest_cr+8+dct_offset, dct_linesize); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4047 } |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4048 } |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4049 }//fi gray |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4050 } |
1042 | 4051 else{ |
936 | 4052 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr); |
481 | 4053 } |
0 | 4054 } else { |
4055 /* dct only in intra block */ | |
1421 | 4056 if(s->encoding || !(s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO)){ |
2256 | 4057 put_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale); |
4058 put_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale); | |
4059 put_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale); | |
4060 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale); | |
0 | 4061 |
711 | 4062 if(!(s->flags&CODEC_FLAG_GRAY)){ |
3309 | 4063 if(s->chroma_y_shift){ |
4064 put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale); | |
4065 put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale); | |
4066 }else{ | |
4067 dct_offset >>=1; | |
4068 dct_linesize >>=1; | |
4069 put_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale); | |
4070 put_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale); | |
4071 put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale); | |
4072 put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale); | |
4073 } | |
711 | 4074 } |
4075 }else{ | |
2256 | 4076 s->dsp.idct_put(dest_y , dct_linesize, block[0]); |
4077 s->dsp.idct_put(dest_y + block_size, dct_linesize, block[1]); | |
4078 s->dsp.idct_put(dest_y + dct_offset , dct_linesize, block[2]); | |
4079 s->dsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]); | |
711 | 4080 |
4081 if(!(s->flags&CODEC_FLAG_GRAY)){ | |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4082 if(s->chroma_y_shift){ |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4083 s->dsp.idct_put(dest_cb, uvlinesize, block[4]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4084 s->dsp.idct_put(dest_cr, uvlinesize, block[5]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4085 }else{ |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4086 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4087 dct_linesize = uvlinesize << s->interlaced_dct; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4088 dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8; |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4089 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4090 s->dsp.idct_put(dest_cb, dct_linesize, block[4]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4091 s->dsp.idct_put(dest_cr, dct_linesize, block[5]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4092 s->dsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4093 s->dsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4094 if(!s->chroma_x_shift){//Chroma444 |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4095 s->dsp.idct_put(dest_cb + 8, dct_linesize, block[8]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4096 s->dsp.idct_put(dest_cr + 8, dct_linesize, block[9]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4097 s->dsp.idct_put(dest_cb + 8 + dct_offset, dct_linesize, block[10]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4098 s->dsp.idct_put(dest_cr + 8 + dct_offset, dct_linesize, block[11]); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4099 } |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4100 } |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4101 }//gray |
485 | 4102 } |
0 | 4103 } |
2792 | 4104 skip_idct: |
1632 | 4105 if(!readable){ |
4106 s->dsp.put_pixels_tab[0][0](s->dest[0], dest_y , linesize,16); | |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4107 s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift); |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4108 s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift); |
1632 | 4109 } |
0 | 4110 } |
294 | 4111 } |
4112 | |
2256 | 4113 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[12][64]){ |
4114 if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1); | |
4115 else MPV_decode_mb_internal(s, block, 0); | |
4116 } | |
4117 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4118 #ifdef CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4119 |
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
4120 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold) |
456 | 4121 { |
4122 static const char tab[64]= | |
4123 {3,2,2,1,1,1,1,1, | |
4124 1,1,1,1,1,1,1,1, | |
4125 1,1,1,1,1,1,1,1, | |
4126 0,0,0,0,0,0,0,0, | |
4127 0,0,0,0,0,0,0,0, | |
4128 0,0,0,0,0,0,0,0, | |
4129 0,0,0,0,0,0,0,0, | |
4130 0,0,0,0,0,0,0,0}; | |
4131 int score=0; | |
4132 int run=0; | |
4133 int i; | |
4134 DCTELEM *block= s->block[n]; | |
4135 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
|
4136 int skip_dc; |
456 | 4137 |
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
|
4138 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
|
4139 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
|
4140 threshold= -threshold; |
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
4141 }else |
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
4142 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
|
4143 |
456 | 4144 /* are all which we could set to zero are allready zero? */ |
4145 if(last_index<=skip_dc - 1) return; | |
4146 | |
4147 for(i=0; i<=last_index; i++){ | |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4148 const int j = s->intra_scantable.permutated[i]; |
456 | 4149 const int level = ABS(block[j]); |
4150 if(level==1){ | |
4151 if(skip_dc && i==0) continue; | |
4152 score+= tab[run]; | |
4153 run=0; | |
4154 }else if(level>1){ | |
4155 return; | |
4156 }else{ | |
4157 run++; | |
4158 } | |
4159 } | |
4160 if(score >= threshold) return; | |
4161 for(i=skip_dc; i<=last_index; i++){ | |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4162 const int j = s->intra_scantable.permutated[i]; |
456 | 4163 block[j]=0; |
4164 } | |
4165 if(block[0]) s->block_last_index[n]= 0; | |
4166 else s->block_last_index[n]= -1; | |
4167 } | |
4168 | |
344 | 4169 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index) |
4170 { | |
4171 int i; | |
4172 const int maxlevel= s->max_qcoeff; | |
4173 const int minlevel= s->min_qcoeff; | |
1722
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4174 int overflow=0; |
2967 | 4175 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4176 if(s->mb_intra){ |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4177 i=1; //skip clipping of intra dc |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4178 }else |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4179 i=0; |
2967 | 4180 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4181 for(;i<=last_index; i++){ |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
4182 const int j= s->intra_scantable.permutated[i]; |
344 | 4183 int level = block[j]; |
2967 | 4184 |
1722
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4185 if (level>maxlevel){ |
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4186 level=maxlevel; |
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4187 overflow++; |
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4188 }else if(level<minlevel){ |
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4189 level=minlevel; |
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4190 overflow++; |
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4191 } |
2967 | 4192 |
344 | 4193 block[j]= level; |
4194 } | |
2967 | 4195 |
1722
7e040c2a86e4
warn the user if we had to clip some dct coefficient due to a crappy format which doenst support the whole needed range (msmpeg4/wmv mostly but mpeg1 too to a lesser extend)
michael
parents:
1719
diff
changeset
|
4196 if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE) |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
4197 av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel); |
344 | 4198 } |
324 | 4199 |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4200 #endif //CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4201 |
1098
b7f267d168b7
mpeg2 field pictures + sliced mode (doesnt work with mplayer though, dunno why)
michaelni
parents:
1096
diff
changeset
|
4202 /** |
b7f267d168b7
mpeg2 field pictures + sliced mode (doesnt work with mplayer though, dunno why)
michaelni
parents:
1096
diff
changeset
|
4203 * |
b7f267d168b7
mpeg2 field pictures + sliced mode (doesnt work with mplayer though, dunno why)
michaelni
parents:
1096
diff
changeset
|
4204 * @param h is the normal height, this will be reduced automatically if needed for the last row |
b7f267d168b7
mpeg2 field pictures + sliced mode (doesnt work with mplayer though, dunno why)
michaelni
parents:
1096
diff
changeset
|
4205 */ |
b7f267d168b7
mpeg2 field pictures + sliced mode (doesnt work with mplayer though, dunno why)
michaelni
parents:
1096
diff
changeset
|
4206 void ff_draw_horiz_band(MpegEncContext *s, int y, int h){ |
1368 | 4207 if (s->avctx->draw_horiz_band) { |
1370 | 4208 AVFrame *src; |
1368 | 4209 int offset[4]; |
2967 | 4210 |
1370 | 4211 if(s->picture_structure != PICT_FRAME){ |
4212 h <<= 1; | |
4213 y <<= 1; | |
4214 if(s->first_field && !(s->avctx->slice_flags&SLICE_FLAG_ALLOW_FIELD)) return; | |
4215 } | |
4216 | |
2266
514949de5d15
set AVCodecContext.width/height to the picture width/height instead of the one stored in the bitstream (that only matters if lowres!=0)
michael
parents:
2265
diff
changeset
|
4217 h= FFMIN(h, s->avctx->height - y); |
1098
b7f267d168b7
mpeg2 field pictures + sliced mode (doesnt work with mplayer though, dunno why)
michaelni
parents:
1096
diff
changeset
|
4218 |
2967 | 4219 if(s->pict_type==B_TYPE || s->low_delay || (s->avctx->slice_flags&SLICE_FLAG_CODED_ORDER)) |
1370 | 4220 src= (AVFrame*)s->current_picture_ptr; |
4221 else if(s->last_picture_ptr) | |
4222 src= (AVFrame*)s->last_picture_ptr; | |
4223 else | |
4224 return; | |
2967 | 4225 |
1369 | 4226 if(s->pict_type==B_TYPE && s->picture_structure == PICT_FRAME && s->out_format != FMT_H264){ |
1368 | 4227 offset[0]= |
4228 offset[1]= | |
4229 offset[2]= | |
4230 offset[3]= 0; | |
4231 }else{ | |
4232 offset[0]= y * s->linesize;; | |
2967 | 4233 offset[1]= |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4234 offset[2]= (y >> s->chroma_y_shift) * s->uvlinesize; |
1368 | 4235 offset[3]= 0; |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
4236 } |
1368 | 4237 |
813 | 4238 emms_c(); |
4239 | |
1370 | 4240 s->avctx->draw_horiz_band(s->avctx, src, offset, |
4241 y, s->picture_structure, h); | |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
4242 } |
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
4243 } |
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
4244 |
1389 | 4245 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename |
2764 | 4246 const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics |
1389 | 4247 const int uvlinesize= s->current_picture.linesize[1]; |
2256 | 4248 const int mb_size= 4 - s->avctx->lowres; |
2967 | 4249 |
1938
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
4250 s->block_index[0]= s->b8_stride*(s->mb_y*2 ) - 2 + s->mb_x*2; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
4251 s->block_index[1]= s->b8_stride*(s->mb_y*2 ) - 1 + s->mb_x*2; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
4252 s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
4253 s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
4254 s->block_index[4]= s->mb_stride*(s->mb_y + 1) + s->b8_stride*s->mb_height*2 + s->mb_x - 1; |
e2501e6e7ff7
unify table indexing (motion_val,dc_val,ac_val,coded_block changed)
michael
parents:
1918
diff
changeset
|
4255 s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1; |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4256 //block_index is not used by mpeg2, so it is not affected by chroma_format |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4257 |
2256 | 4258 s->dest[0] = s->current_picture.data[0] + ((s->mb_x - 1) << mb_size); |
4259 s->dest[1] = s->current_picture.data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift)); | |
4260 s->dest[2] = s->current_picture.data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift)); | |
2168 | 4261 |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4262 if(!(s->pict_type==B_TYPE && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME)) |
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4263 { |
2256 | 4264 s->dest[0] += s->mb_y * linesize << mb_size; |
4265 s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift); | |
4266 s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift); | |
2074
2faafe7a3db6
mpeg2 chroma422/444 support, may be slower, may be faster for other codecs
iive
parents:
2066
diff
changeset
|
4267 } |
1389 | 4268 } |
4269 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4270 #ifdef CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4271 |
1777 | 4272 static void get_vissual_weight(int16_t *weight, uint8_t *ptr, int stride){ |
4273 int x, y; | |
4274 //FIXME optimize | |
4275 for(y=0; y<8; y++){ | |
4276 for(x=0; x<8; x++){ | |
4277 int x2, y2; | |
4278 int sum=0; | |
4279 int sqr=0; | |
4280 int count=0; | |
4281 | |
4282 for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){ | |
4283 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){ | |
4284 int v= ptr[x2 + y2*stride]; | |
4285 sum += v; | |
4286 sqr += v*v; | |
4287 count++; | |
4288 } | |
4289 } | |
4290 weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count; | |
4291 } | |
4292 } | |
4293 } | |
4294 | |
3309 | 4295 static always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count) |
294 | 4296 { |
3309 | 4297 int16_t weight[8][64]; |
4298 DCTELEM orig[8][64]; | |
294 | 4299 const int mb_x= s->mb_x; |
4300 const int mb_y= s->mb_y; | |
4301 int i; | |
3309 | 4302 int skip_dct[8]; |
697 | 4303 int dct_offset = s->linesize*8; //default for progressive frames |
1778 | 4304 uint8_t *ptr_y, *ptr_cb, *ptr_cr; |
4305 int wrap_y, wrap_c; | |
2967 | 4306 |
3309 | 4307 for(i=0; i<mb_block_count; i++) skip_dct[i]=0; |
2967 | 4308 |
690
a1c69cb685b3
adaptive quantization (lumi/temporal & spatial complexity masking)
michaelni
parents:
673
diff
changeset
|
4309 if(s->adaptive_quant){ |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
4310 const int last_qp= s->qscale; |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
4311 const int mb_xy= mb_x + mb_y*s->mb_stride; |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
4312 |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
4313 s->lambda= s->lambda_table[mb_xy]; |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
4314 update_qscale(s); |
2967 | 4315 |
1616 | 4316 if(!(s->flags&CODEC_FLAG_QP_RD)){ |
4317 s->dquant= s->qscale - last_qp; | |
4318 | |
1805
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4319 if(s->out_format==FMT_H263){ |
1616 | 4320 s->dquant= clip(s->dquant, -2, 2); //FIXME RD |
2967 | 4321 |
4322 if(s->codec_id==CODEC_ID_MPEG4){ | |
1805
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4323 if(!s->mb_intra){ |
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4324 if(s->pict_type == B_TYPE){ |
2967 | 4325 if(s->dquant&1) |
1805
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4326 s->dquant= (s->dquant/2)*2; |
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4327 if(s->mv_dir&MV_DIRECT) |
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4328 s->dquant= 0; |
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4329 } |
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4330 if(s->mv_type==MV_TYPE_8X8) |
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4331 s->dquant=0; |
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4332 } |
1616 | 4333 } |
690
a1c69cb685b3
adaptive quantization (lumi/temporal & spatial complexity masking)
michaelni
parents:
673
diff
changeset
|
4334 } |
a1c69cb685b3
adaptive quantization (lumi/temporal & spatial complexity masking)
michaelni
parents:
673
diff
changeset
|
4335 } |
1652 | 4336 ff_set_qscale(s, last_qp + s->dquant); |
2054 | 4337 }else if(s->flags&CODEC_FLAG_QP_RD) |
4338 ff_set_qscale(s, s->qscale + s->dquant); | |
294 | 4339 |
1778 | 4340 wrap_y = s->linesize; |
4341 wrap_c = s->uvlinesize; | |
4342 ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16; | |
3309 | 4343 ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8; |
4344 ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8; | |
1778 | 4345 |
4346 if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){ | |
2165
5ca07515a309
10l (dont store 2 different things at the same place ...)
michael
parents:
2132
diff
changeset
|
4347 uint8_t *ebuf= s->edge_emu_buffer + 32; |
5ca07515a309
10l (dont store 2 different things at the same place ...)
michael
parents:
2132
diff
changeset
|
4348 ff_emulated_edge_mc(ebuf , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width , s->height); |
5ca07515a309
10l (dont store 2 different things at the same place ...)
michael
parents:
2132
diff
changeset
|
4349 ptr_y= ebuf; |
3309 | 4350 ff_emulated_edge_mc(ebuf+18*wrap_y , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1); |
2165
5ca07515a309
10l (dont store 2 different things at the same place ...)
michael
parents:
2132
diff
changeset
|
4351 ptr_cb= ebuf+18*wrap_y; |
3309 | 4352 ff_emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1); |
2165
5ca07515a309
10l (dont store 2 different things at the same place ...)
michael
parents:
2132
diff
changeset
|
4353 ptr_cr= ebuf+18*wrap_y+8; |
1778 | 4354 } |
4355 | |
324 | 4356 if (s->mb_intra) { |
697 | 4357 if(s->flags&CODEC_FLAG_INTERLACED_DCT){ |
4358 int progressive_score, interlaced_score; | |
1708 | 4359 |
1729 | 4360 s->interlaced_dct=0; |
2967 | 4361 progressive_score= s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y, 8) |
1777 | 4362 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400; |
1729 | 4363 |
4364 if(progressive_score > 0){ | |
2967 | 4365 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y*2, 8) |
1777 | 4366 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y , NULL, wrap_y*2, 8); |
1729 | 4367 if(progressive_score > interlaced_score){ |
4368 s->interlaced_dct=1; | |
2967 | 4369 |
1729 | 4370 dct_offset= wrap_y; |
4371 wrap_y<<=1; | |
3309 | 4372 if (s->chroma_format == CHROMA_422) |
4373 wrap_c<<=1; | |
1729 | 4374 } |
4375 } | |
697 | 4376 } |
2967 | 4377 |
2979 | 4378 s->dsp.get_pixels(s->block[0], ptr_y , wrap_y); |
1777 | 4379 s->dsp.get_pixels(s->block[1], ptr_y + 8, wrap_y); |
4380 s->dsp.get_pixels(s->block[2], ptr_y + dct_offset , wrap_y); | |
4381 s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y); | |
294 | 4382 |
487 | 4383 if(s->flags&CODEC_FLAG_GRAY){ |
4384 skip_dct[4]= 1; | |
4385 skip_dct[5]= 1; | |
4386 }else{ | |
2979 | 4387 s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c); |
1777 | 4388 s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c); |
3309 | 4389 if(!s->chroma_y_shift){ /* 422 */ |
4390 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c); | |
4391 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c); | |
4392 } | |
1777 | 4393 } |
324 | 4394 }else{ |
651 | 4395 op_pixels_func (*op_pix)[4]; |
4396 qpel_mc_func (*op_qpix)[16]; | |
1064 | 4397 uint8_t *dest_y, *dest_cb, *dest_cr; |
294 | 4398 |
1389 | 4399 dest_y = s->dest[0]; |
4400 dest_cb = s->dest[1]; | |
4401 dest_cr = s->dest[2]; | |
324 | 4402 |
327 | 4403 if ((!s->no_rounding) || s->pict_type==B_TYPE){ |
2979 | 4404 op_pix = s->dsp.put_pixels_tab; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4405 op_qpix= s->dsp.put_qpel_pixels_tab; |
295 | 4406 }else{ |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4407 op_pix = s->dsp.put_no_rnd_pixels_tab; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4408 op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab; |
324 | 4409 } |
295 | 4410 |
324 | 4411 if (s->mv_dir & MV_DIR_FORWARD) { |
903 | 4412 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4413 op_pix = s->dsp.avg_pixels_tab; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4414 op_qpix= s->dsp.avg_qpel_pixels_tab; |
324 | 4415 } |
4416 if (s->mv_dir & MV_DIR_BACKWARD) { | |
903 | 4417 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix); |
324 | 4418 } |
295 | 4419 |
697 | 4420 if(s->flags&CODEC_FLAG_INTERLACED_DCT){ |
4421 int progressive_score, interlaced_score; | |
1729 | 4422 |
4423 s->interlaced_dct=0; | |
2967 | 4424 progressive_score= s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y, 8) |
1729 | 4425 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400; |
2967 | 4426 |
1729 | 4427 if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400; |
4428 | |
4429 if(progressive_score>0){ | |
2967 | 4430 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y*2, 8) |
1729 | 4431 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y , ptr_y + wrap_y , wrap_y*2, 8); |
2967 | 4432 |
1729 | 4433 if(progressive_score > interlaced_score){ |
4434 s->interlaced_dct=1; | |
2967 | 4435 |
1729 | 4436 dct_offset= wrap_y; |
4437 wrap_y<<=1; | |
3309 | 4438 if (s->chroma_format == CHROMA_422) |
4439 wrap_c<<=1; | |
1729 | 4440 } |
4441 } | |
697 | 4442 } |
2967 | 4443 |
2979 | 4444 s->dsp.diff_pixels(s->block[0], ptr_y , dest_y , wrap_y); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4445 s->dsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4446 s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset , dest_y + dct_offset , wrap_y); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4447 s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y); |
2967 | 4448 |
487 | 4449 if(s->flags&CODEC_FLAG_GRAY){ |
4450 skip_dct[4]= 1; | |
4451 skip_dct[5]= 1; | |
4452 }else{ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4453 s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
847
diff
changeset
|
4454 s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c); |
3309 | 4455 if(!s->chroma_y_shift){ /* 422 */ |
4456 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c); | |
4457 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c); | |
4458 } | |
487 | 4459 } |
2967 | 4460 /* pre quantization */ |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
4461 if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){ |
697 | 4462 //FIXME optimize |
2979 | 4463 if(s->dsp.sad[1](NULL, ptr_y , dest_y , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1; |
1708 | 4464 if(s->dsp.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1; |
4465 if(s->dsp.sad[1](NULL, ptr_y +dct_offset , dest_y +dct_offset , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1; | |
4466 if(s->dsp.sad[1](NULL, ptr_y +dct_offset+ 8, dest_y +dct_offset+ 8, wrap_y, 8) < 20*s->qscale) skip_dct[3]= 1; | |
4467 if(s->dsp.sad[1](NULL, ptr_cb , dest_cb , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1; | |
4468 if(s->dsp.sad[1](NULL, ptr_cr , dest_cr , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1; | |
3309 | 4469 if(!s->chroma_y_shift){ /* 422 */ |
4470 if(s->dsp.sad[1](NULL, ptr_cb +(dct_offset>>1), dest_cb +(dct_offset>>1), wrap_c, 8) < 20*s->qscale) skip_dct[6]= 1; | |
4471 if(s->dsp.sad[1](NULL, ptr_cr +(dct_offset>>1), dest_cr +(dct_offset>>1), wrap_c, 8) < 20*s->qscale) skip_dct[7]= 1; | |
4472 } | |
456 | 4473 } |
1778 | 4474 } |
4475 | |
4476 if(s->avctx->quantizer_noise_shaping){ | |
4477 if(!skip_dct[0]) get_vissual_weight(weight[0], ptr_y , wrap_y); | |
4478 if(!skip_dct[1]) get_vissual_weight(weight[1], ptr_y + 8, wrap_y); | |
4479 if(!skip_dct[2]) get_vissual_weight(weight[2], ptr_y + dct_offset , wrap_y); | |
4480 if(!skip_dct[3]) get_vissual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y); | |
4481 if(!skip_dct[4]) get_vissual_weight(weight[4], ptr_cb , wrap_c); | |
4482 if(!skip_dct[5]) get_vissual_weight(weight[5], ptr_cr , wrap_c); | |
3309 | 4483 if(!s->chroma_y_shift){ /* 422 */ |
4484 if(!skip_dct[6]) get_vissual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c); | |
4485 if(!skip_dct[7]) get_vissual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c); | |
4486 } | |
4487 memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count); | |
294 | 4488 } |
2967 | 4489 |
294 | 4490 /* DCT & quantize */ |
1778 | 4491 assert(s->out_format!=FMT_MJPEG || s->qscale==8); |
4492 { | |
3309 | 4493 for(i=0;i<mb_block_count;i++) { |
456 | 4494 if(!skip_dct[i]){ |
4495 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
|
4496 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow); |
344 | 4497 // 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
|
4498 // 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
|
4499 // of improve it. Just INTRADC clipping deserves changes in quantizer |
456 | 4500 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]); |
4501 }else | |
4502 s->block_last_index[i]= -1; | |
344 | 4503 } |
1777 | 4504 if(s->avctx->quantizer_noise_shaping){ |
3309 | 4505 for(i=0;i<mb_block_count;i++) { |
1777 | 4506 if(!skip_dct[i]){ |
4507 s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale); | |
4508 } | |
4509 } | |
4510 } | |
2967 | 4511 |
456 | 4512 if(s->luma_elim_threshold && !s->mb_intra) |
4513 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
|
4514 dct_single_coeff_elimination(s, i, s->luma_elim_threshold); |
456 | 4515 if(s->chroma_elim_threshold && !s->mb_intra) |
3309 | 4516 for(i=4; i<mb_block_count; i++) |
605
40874804a5af
same dc skip behavior for chroma & luma elimination, its confusing otherwise imho
michaelni
parents:
604
diff
changeset
|
4517 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold); |
1497 | 4518 |
4519 if(s->flags & CODEC_FLAG_CBP_RD){ | |
3309 | 4520 for(i=0;i<mb_block_count;i++) { |
1497 | 4521 if(s->block_last_index[i] == -1) |
4522 s->coded_score[i]= INT_MAX/256; | |
4523 } | |
4524 } | |
294 | 4525 } |
4526 | |
487 | 4527 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){ |
4528 s->block_last_index[4]= | |
4529 s->block_last_index[5]= 0; | |
4530 s->block[4][0]= | |
1011 | 4531 s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale; |
487 | 4532 } |
4533 | |
1708 | 4534 //non c quantize code returns incorrect block_last_index FIXME |
4535 if(s->alternate_scan && s->dct_quantize != dct_quantize_c){ | |
3309 | 4536 for(i=0; i<mb_block_count; i++){ |
1708 | 4537 int j; |
4538 if(s->block_last_index[i]>0){ | |
4539 for(j=63; j>0; j--){ | |
4540 if(s->block[i][ s->intra_scantable.permutated[j] ]) break; | |
4541 } | |
4542 s->block_last_index[i]= j; | |
4543 } | |
4544 } | |
4545 } | |
4546 | |
294 | 4547 /* huffman encode */ |
936 | 4548 switch(s->codec_id){ //FIXME funct ptr could be slightly faster |
4549 case CODEC_ID_MPEG1VIDEO: | |
1421 | 4550 case CODEC_ID_MPEG2VIDEO: |
936 | 4551 mpeg1_encode_mb(s, s->block, motion_x, motion_y); break; |
4552 case CODEC_ID_MPEG4: | |
4553 mpeg4_encode_mb(s, s->block, motion_x, motion_y); break; | |
4554 case CODEC_ID_MSMPEG4V2: | |
4555 case CODEC_ID_MSMPEG4V3: | |
4556 case CODEC_ID_WMV1: | |
4557 msmpeg4_encode_mb(s, s->block, motion_x, motion_y); break; | |
4558 case CODEC_ID_WMV2: | |
4559 ff_wmv2_encode_mb(s, s->block, motion_x, motion_y); break; | |
2639 | 4560 #ifdef CONFIG_H261_ENCODER |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4561 case CODEC_ID_H261: |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4562 ff_h261_encode_mb(s, s->block, motion_x, motion_y); break; |
2639 | 4563 #endif |
936 | 4564 case CODEC_ID_H263: |
4565 case CODEC_ID_H263P: | |
1353
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4566 case CODEC_ID_FLV1: |
936 | 4567 case CODEC_ID_RV10: |
2380 | 4568 case CODEC_ID_RV20: |
936 | 4569 h263_encode_mb(s, s->block, motion_x, motion_y); break; |
1042 | 4570 case CODEC_ID_MJPEG: |
4571 mjpeg_encode_mb(s, s->block); break; | |
936 | 4572 default: |
4573 assert(0); | |
294 | 4574 } |
4575 } | |
4576 | |
3309 | 4577 static always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y) |
4578 { | |
4579 if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y, 8, 6); | |
4580 else encode_mb_internal(s, motion_x, motion_y, 16, 8); | |
4581 } | |
4582 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4583 #endif //CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4584 |
1368 | 4585 void ff_mpeg_flush(AVCodecContext *avctx){ |
4586 int i; | |
4587 MpegEncContext *s = avctx->priv_data; | |
2967 | 4588 |
4589 if(s==NULL || s->picture==NULL) | |
1751 | 4590 return; |
2967 | 4591 |
1368 | 4592 for(i=0; i<MAX_PICTURE_COUNT; i++){ |
4593 if(s->picture[i].data[0] && ( s->picture[i].type == FF_BUFFER_TYPE_INTERNAL | |
4594 || s->picture[i].type == FF_BUFFER_TYPE_USER)) | |
4595 avctx->release_buffer(avctx, (AVFrame*)&s->picture[i]); | |
4596 } | |
1601 | 4597 s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL; |
2967 | 4598 |
2380 | 4599 s->mb_x= s->mb_y= 0; |
2967 | 4600 |
1395 | 4601 s->parse_context.state= -1; |
4602 s->parse_context.frame_start_found= 0; | |
4603 s->parse_context.overread= 0; | |
4604 s->parse_context.overread_index= 0; | |
4605 s->parse_context.index= 0; | |
4606 s->parse_context.last_index= 0; | |
1851 | 4607 s->bitstream_buffer_size=0; |
1368 | 4608 } |
4609 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
4610 #ifdef CONFIG_ENCODERS |
1064 | 4611 void ff_copy_bits(PutBitContext *pb, uint8_t *src, int length) |
294 | 4612 { |
1810 | 4613 const uint16_t *srcw= (uint16_t*)src; |
1799 | 4614 int words= length>>4; |
326 | 4615 int bits= length&15; |
4616 int i; | |
4617 | |
456 | 4618 if(length==0) return; |
2967 | 4619 |
1803 | 4620 if(words < 16){ |
4621 for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i])); | |
4622 }else if(put_bits_count(pb)&7){ | |
4623 for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i])); | |
1802
e91dbb5b1a8d
activate the memcpy special case for the byte aligned case in ff_copy_bits(), wasnt working before because of a incorrectly used pbBufPtr()
michael
parents:
1801
diff
changeset
|
4624 }else{ |
1803 | 4625 for(i=0; put_bits_count(pb)&31; i++) |
4626 put_bits(pb, 8, src[i]); | |
1799 | 4627 flush_put_bits(pb); |
1803 | 4628 memcpy(pbBufPtr(pb), src+i, 2*words-i); |
4629 skip_put_bytes(pb, 2*words-i); | |
1802
e91dbb5b1a8d
activate the memcpy special case for the byte aligned case in ff_copy_bits(), wasnt working before because of a incorrectly used pbBufPtr()
michael
parents:
1801
diff
changeset
|
4630 } |
2967 | 4631 |
1803 | 4632 put_bits(pb, bits, be2me_16(srcw[words])>>(16-bits)); |
0 | 4633 } |
4634 | |
456 | 4635 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){ |
326 | 4636 int i; |
4637 | |
4638 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop? | |
4639 | |
4640 /* mpeg1 */ | |
1160 | 4641 d->mb_skip_run= s->mb_skip_run; |
326 | 4642 for(i=0; i<3; i++) |
4643 d->last_dc[i]= s->last_dc[i]; | |
2967 | 4644 |
326 | 4645 /* statistics */ |
4646 d->mv_bits= s->mv_bits; | |
4647 d->i_tex_bits= s->i_tex_bits; | |
4648 d->p_tex_bits= s->p_tex_bits; | |
4649 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
|
4650 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
|
4651 d->b_count= s->b_count; |
326 | 4652 d->skip_count= s->skip_count; |
4653 d->misc_bits= s->misc_bits; | |
329 | 4654 d->last_bits= 0; |
327 | 4655 |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
4656 d->mb_skipped= 0; |
912 | 4657 d->qscale= s->qscale; |
1616 | 4658 d->dquant= s->dquant; |
326 | 4659 } |
4660 | |
456 | 4661 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){ |
326 | 4662 int i; |
4663 | |
2967 | 4664 memcpy(d->mv, s->mv, 2*4*2*sizeof(int)); |
326 | 4665 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop? |
2967 | 4666 |
326 | 4667 /* mpeg1 */ |
1160 | 4668 d->mb_skip_run= s->mb_skip_run; |
326 | 4669 for(i=0; i<3; i++) |
4670 d->last_dc[i]= s->last_dc[i]; | |
2967 | 4671 |
326 | 4672 /* statistics */ |
4673 d->mv_bits= s->mv_bits; | |
4674 d->i_tex_bits= s->i_tex_bits; | |
4675 d->p_tex_bits= s->p_tex_bits; | |
4676 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
|
4677 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
|
4678 d->b_count= s->b_count; |
326 | 4679 d->skip_count= s->skip_count; |
4680 d->misc_bits= s->misc_bits; | |
4681 | |
4682 d->mb_intra= s->mb_intra; | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
4683 d->mb_skipped= s->mb_skipped; |
326 | 4684 d->mv_type= s->mv_type; |
4685 d->mv_dir= s->mv_dir; | |
4686 d->pb= s->pb; | |
456 | 4687 if(s->data_partitioning){ |
4688 d->pb2= s->pb2; | |
4689 d->tex_pb= s->tex_pb; | |
4690 } | |
326 | 4691 d->block= s->block; |
3309 | 4692 for(i=0; i<8; i++) |
326 | 4693 d->block_last_index[i]= s->block_last_index[i]; |
755 | 4694 d->interlaced_dct= s->interlaced_dct; |
912 | 4695 d->qscale= s->qscale; |
326 | 4696 } |
4697 | |
2967 | 4698 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type, |
456 | 4699 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2], |
4700 int *dmin, int *next_block, int motion_x, int motion_y) | |
4701 { | |
1389 | 4702 int score; |
4703 uint8_t *dest_backup[3]; | |
2967 | 4704 |
456 | 4705 copy_context_before_encode(s, backup, type); |
4706 | |
4707 s->block= s->blocks[*next_block]; | |
4708 s->pb= pb[*next_block]; | |
4709 if(s->data_partitioning){ | |
4710 s->pb2 = pb2 [*next_block]; | |
4711 s->tex_pb= tex_pb[*next_block]; | |
4712 } | |
2967 | 4713 |
1389 | 4714 if(*next_block){ |
4715 memcpy(dest_backup, s->dest, sizeof(s->dest)); | |
1799 | 4716 s->dest[0] = s->rd_scratchpad; |
4717 s->dest[1] = s->rd_scratchpad + 16*s->linesize; | |
4718 s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8; | |
4719 assert(s->linesize >= 32); //FIXME | |
1389 | 4720 } |
456 | 4721 |
4722 encode_mb(s, motion_x, motion_y); | |
2967 | 4723 |
1786 | 4724 score= put_bits_count(&s->pb); |
456 | 4725 if(s->data_partitioning){ |
1786 | 4726 score+= put_bits_count(&s->pb2); |
4727 score+= put_bits_count(&s->tex_pb); | |
456 | 4728 } |
2967 | 4729 |
1389 | 4730 if(s->avctx->mb_decision == FF_MB_DECISION_RD){ |
4731 MPV_decode_mb(s, s->block); | |
4732 | |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
4733 score *= s->lambda2; |
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
4734 score += sse_mb(s) << FF_LAMBDA_SHIFT; |
1389 | 4735 } |
2967 | 4736 |
1389 | 4737 if(*next_block){ |
4738 memcpy(s->dest, dest_backup, sizeof(s->dest)); | |
4739 } | |
4740 | |
4741 if(score<*dmin){ | |
4742 *dmin= score; | |
456 | 4743 *next_block^=1; |
4744 | |
4745 copy_context_after_encode(best, s, type); | |
4746 } | |
4747 } | |
2967 | 4748 |
1389 | 4749 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){ |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4750 uint32_t *sq = squareTbl + 256; |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4751 int acc=0; |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4752 int x,y; |
2967 | 4753 |
4754 if(w==16 && h==16) | |
1708 | 4755 return s->dsp.sse[0](NULL, src1, src2, stride, 16); |
936 | 4756 else if(w==8 && h==8) |
1708 | 4757 return s->dsp.sse[1](NULL, src1, src2, stride, 8); |
2967 | 4758 |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4759 for(y=0; y<h; y++){ |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4760 for(x=0; x<w; x++){ |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4761 acc+= sq[src1[x + y*stride] - src2[x + y*stride]]; |
2967 | 4762 } |
4763 } | |
4764 | |
936 | 4765 assert(acc>=0); |
2967 | 4766 |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4767 return acc; |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4768 } |
326 | 4769 |
1389 | 4770 static int sse_mb(MpegEncContext *s){ |
4771 int w= 16; | |
4772 int h= 16; | |
4773 | |
4774 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16; | |
4775 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16; | |
4776 | |
4777 if(w==16 && h==16) | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2063
diff
changeset
|
4778 if(s->avctx->mb_cmp == FF_CMP_NSSE){ |
2066 | 4779 return s->dsp.nsse[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16) |
4780 +s->dsp.nsse[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8) | |
4781 +s->dsp.nsse[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8); | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2063
diff
changeset
|
4782 }else{ |
1708 | 4783 return s->dsp.sse[0](NULL, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16) |
4784 +s->dsp.sse[1](NULL, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8) | |
4785 +s->dsp.sse[1](NULL, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8); | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2063
diff
changeset
|
4786 } |
1389 | 4787 else |
4788 return sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize) | |
4789 +sse(s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize) | |
4790 +sse(s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize); | |
4791 } | |
4792 | |
1799 | 4793 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){ |
4794 MpegEncContext *s= arg; | |
4795 | |
2967 | 4796 |
1799 | 4797 s->me.pre_pass=1; |
4798 s->me.dia_size= s->avctx->pre_dia_size; | |
4799 s->first_slice_line=1; | |
4800 for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) { | |
4801 for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) { | |
4802 ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y); | |
4803 } | |
4804 s->first_slice_line=0; | |
4805 } | |
2967 | 4806 |
1799 | 4807 s->me.pre_pass=0; |
2967 | 4808 |
1799 | 4809 return 0; |
4810 } | |
4811 | |
4812 static int estimate_motion_thread(AVCodecContext *c, void *arg){ | |
4813 MpegEncContext *s= arg; | |
4814 | |
4815 s->me.dia_size= s->avctx->dia_size; | |
4816 s->first_slice_line=1; | |
4817 for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) { | |
4818 s->mb_x=0; //for block init below | |
4819 ff_init_block_index(s); | |
4820 for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) { | |
4821 s->block_index[0]+=2; | |
4822 s->block_index[1]+=2; | |
4823 s->block_index[2]+=2; | |
4824 s->block_index[3]+=2; | |
2967 | 4825 |
1799 | 4826 /* compute motion vector & mb_type and store in context */ |
4827 if(s->pict_type==B_TYPE) | |
4828 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y); | |
4829 else | |
4830 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y); | |
4831 } | |
4832 s->first_slice_line=0; | |
4833 } | |
4834 return 0; | |
4835 } | |
4836 | |
1804
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4837 static int mb_var_thread(AVCodecContext *c, void *arg){ |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4838 MpegEncContext *s= arg; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4839 int mb_x, mb_y; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4840 |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4841 for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) { |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4842 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4843 int xx = mb_x * 16; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4844 int yy = mb_y * 16; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4845 uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4846 int varc; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4847 int sum = s->dsp.pix_sum(pix, s->linesize); |
2967 | 4848 |
1804
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4849 varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4850 |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4851 s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc; |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4852 s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8; |
2015
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
4853 s->me.mb_var_sum_temp += varc; |
1804
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4854 } |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4855 } |
1805
5dd5b2052f5b
mpeg4 b frame + adaptive quantization assertion failure fix
michael
parents:
1804
diff
changeset
|
4856 return 0; |
1804
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4857 } |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
4858 |
1799 | 4859 static void write_slice_end(MpegEncContext *s){ |
4860 if(s->codec_id==CODEC_ID_MPEG4){ | |
4861 if(s->partitioned_frame){ | |
4862 ff_mpeg4_merge_partitions(s); | |
4863 } | |
2967 | 4864 |
1799 | 4865 ff_mpeg4_stuffing(&s->pb); |
4866 }else if(s->out_format == FMT_MJPEG){ | |
4867 ff_mjpeg_stuffing(&s->pb); | |
4868 } | |
4869 | |
4870 align_put_bits(&s->pb); | |
4871 flush_put_bits(&s->pb); | |
2967 | 4872 |
2428
548c87c06dbb
assertion about bits statistic and minor bits stat fix
michael
parents:
2427
diff
changeset
|
4873 if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame) |
548c87c06dbb
assertion about bits statistic and minor bits stat fix
michael
parents:
2427
diff
changeset
|
4874 s->misc_bits+= get_bits_diff(s); |
1799 | 4875 } |
4876 | |
4877 static int encode_thread(AVCodecContext *c, void *arg){ | |
4878 MpegEncContext *s= arg; | |
766 | 4879 int mb_x, mb_y, pdif = 0; |
1708 | 4880 int i, j; |
326 | 4881 MpegEncContext best_s, backup_s; |
2422 | 4882 uint8_t bit_buf[2][MAX_MB_BYTES]; |
4883 uint8_t bit_buf2[2][MAX_MB_BYTES]; | |
4884 uint8_t bit_buf_tex[2][MAX_MB_BYTES]; | |
456 | 4885 PutBitContext pb[2], pb2[2], tex_pb[2]; |
1799 | 4886 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y); |
456 | 4887 |
4888 for(i=0; i<2; i++){ | |
2422 | 4889 init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES); |
4890 init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES); | |
4891 init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES); | |
456 | 4892 } |
0 | 4893 |
1786 | 4894 s->last_bits= put_bits_count(&s->pb); |
286 | 4895 s->mv_bits=0; |
4896 s->misc_bits=0; | |
4897 s->i_tex_bits=0; | |
4898 s->p_tex_bits=0; | |
4899 s->i_count=0; | |
656
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
4900 s->f_count=0; |
e47fa3e3f2d5
statistics for forw & back p-MBs instead of just one counter for both
michaelni
parents:
652
diff
changeset
|
4901 s->b_count=0; |
286 | 4902 s->skip_count=0; |
4903 | |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4904 for(i=0; i<3; i++){ |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4905 /* init last dc values */ |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4906 /* note: quant matrix value (8) is implied here */ |
1992 | 4907 s->last_dc[i] = 128 << s->intra_dc_precision; |
2967 | 4908 |
2955 | 4909 s->current_picture.error[i] = 0; |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
4910 } |
1160 | 4911 s->mb_skip_run = 0; |
1708 | 4912 memset(s->last_mv, 0, sizeof(s->last_mv)); |
2967 | 4913 |
1051
e5a9dbf597d4
mpeg1 bframe encoding patch by (Rapha«³l LEGRAND) with some modifications by me
michaelni
parents:
1048
diff
changeset
|
4914 s->last_mv_dir = 0; |
0 | 4915 |
1353
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4916 switch(s->codec_id){ |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4917 case CODEC_ID_H263: |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4918 case CODEC_ID_H263P: |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4919 case CODEC_ID_FLV1: |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
4920 s->gob_index = ff_h263_get_gob_height(s); |
1353
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4921 break; |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4922 case CODEC_ID_MPEG4: |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4923 if(s->partitioned_frame) |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4924 ff_mpeg4_init_partitions(s); |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4925 break; |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
4926 } |
456 | 4927 |
4928 s->resync_mb_x=0; | |
2967 | 4929 s->resync_mb_y=0; |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
4930 s->first_slice_line = 1; |
766 | 4931 s->ptr_lastgob = s->pb.buf; |
1799 | 4932 for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) { |
4933 // printf("row %d at %X\n", s->mb_y, (int)s); | |
1389 | 4934 s->mb_x=0; |
4935 s->mb_y= mb_y; | |
4936 | |
1652 | 4937 ff_set_qscale(s, s->qscale); |
1389 | 4938 ff_init_block_index(s); |
2967 | 4939 |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
4940 for(mb_x=0; mb_x < s->mb_width; mb_x++) { |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4941 int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1173
diff
changeset
|
4942 int mb_type= s->mb_type[xy]; |
456 | 4943 // int d; |
1389 | 4944 int dmin= INT_MAX; |
1708 | 4945 int dir; |
0 | 4946 |
2422 | 4947 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){ |
4948 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); | |
4949 return -1; | |
4950 } | |
4951 if(s->data_partitioning){ | |
4952 if( s->pb2 .buf_end - s->pb2 .buf - (put_bits_count(&s-> pb2)>>3) < MAX_MB_BYTES | |
4953 || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){ | |
4954 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); | |
4955 return -1; | |
4956 } | |
4957 } | |
4958 | |
232
b640ec5948b0
- Now the ME is done for the entire picture when enconding, the
pulento
parents:
231
diff
changeset
|
4959 s->mb_x = mb_x; |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4960 s->mb_y = mb_y; // moved into loop, can get changed by H.261 |
1389 | 4961 ff_update_block_index(s); |
456 | 4962 |
2639 | 4963 #ifdef CONFIG_H261_ENCODER |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4964 if(s->codec_id == CODEC_ID_H261){ |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4965 ff_h261_reorder_mb_index(s); |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4966 xy= s->mb_y*s->mb_stride + s->mb_x; |
2345 | 4967 mb_type= s->mb_type[xy]; |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4968 } |
2639 | 4969 #endif |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
4970 |
766 | 4971 /* write gob / video packet header */ |
1661 | 4972 if(s->rtp_mode){ |
766 | 4973 int current_packet_size, is_gob_start; |
2967 | 4974 |
1801
3f26dfb3eba4
replace a few pbBufPtr() by put_bits_count(), one of them was actually wrong
michael
parents:
1799
diff
changeset
|
4975 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf); |
2967 | 4976 |
4977 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0; | |
4978 | |
1799 | 4979 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1; |
2967 | 4980 |
1661 | 4981 switch(s->codec_id){ |
4982 case CODEC_ID_H263: | |
4983 case CODEC_ID_H263P: | |
4984 if(!s->h263_slice_structured) | |
4985 if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0; | |
4986 break; | |
4987 case CODEC_ID_MPEG2VIDEO: | |
4988 if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1; | |
4989 case CODEC_ID_MPEG1VIDEO: | |
4990 if(s->mb_skip_run) is_gob_start=0; | |
4991 break; | |
4992 } | |
1799 | 4993 |
1661 | 4994 if(is_gob_start){ |
1799 | 4995 if(s->start_mb_y != mb_y || mb_x!=0){ |
4996 write_slice_end(s); | |
4997 | |
4998 if(s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){ | |
4999 ff_mpeg4_init_partitions(s); | |
5000 } | |
1661 | 5001 } |
2967 | 5002 |
1786 | 5003 assert((put_bits_count(&s->pb)&7) == 0); |
1661 | 5004 current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob; |
2967 | 5005 |
1688 | 5006 if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){ |
2371 | 5007 int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y; |
1688 | 5008 int d= 100 / s->avctx->error_rate; |
5009 if(r % d == 0){ | |
5010 current_packet_size=0; | |
5011 #ifndef ALT_BITSTREAM_WRITER | |
5012 s->pb.buf_ptr= s->ptr_lastgob; | |
5013 #endif | |
5014 assert(pbBufPtr(&s->pb) == s->ptr_lastgob); | |
5015 } | |
5016 } | |
2492
57ba17bd3924
rtp_callback: send number of mb patch by (Johan Bilien {jobi via.ecp fr)
michael
parents:
2450
diff
changeset
|
5017 |
57ba17bd3924
rtp_callback: send number of mb patch by (Johan Bilien {jobi via.ecp fr)
michael
parents:
2450
diff
changeset
|
5018 if (s->avctx->rtp_callback){ |
57ba17bd3924
rtp_callback: send number of mb patch by (Johan Bilien {jobi via.ecp fr)
michael
parents:
2450
diff
changeset
|
5019 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x; |
57ba17bd3924
rtp_callback: send number of mb patch by (Johan Bilien {jobi via.ecp fr)
michael
parents:
2450
diff
changeset
|
5020 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb); |
57ba17bd3924
rtp_callback: send number of mb patch by (Johan Bilien {jobi via.ecp fr)
michael
parents:
2450
diff
changeset
|
5021 } |
2967 | 5022 |
1661 | 5023 switch(s->codec_id){ |
5024 case CODEC_ID_MPEG4: | |
456 | 5025 ff_mpeg4_encode_video_packet_header(s); |
5026 ff_mpeg4_clean_buffers(s); | |
1661 | 5027 break; |
5028 case CODEC_ID_MPEG1VIDEO: | |
5029 case CODEC_ID_MPEG2VIDEO: | |
1421 | 5030 ff_mpeg1_encode_slice_header(s); |
5031 ff_mpeg1_clean_buffers(s); | |
1661 | 5032 break; |
5033 case CODEC_ID_H263: | |
5034 case CODEC_ID_H263P: | |
2967 | 5035 h263_encode_gob_header(s, mb_y); |
1661 | 5036 break; |
1160 | 5037 } |
1661 | 5038 |
5039 if(s->flags&CODEC_FLAG_PASS1){ | |
1786 | 5040 int bits= put_bits_count(&s->pb); |
1661 | 5041 s->misc_bits+= bits - s->last_bits; |
5042 s->last_bits= bits; | |
766 | 5043 } |
2967 | 5044 |
1662 | 5045 s->ptr_lastgob += current_packet_size; |
456 | 5046 s->first_slice_line=1; |
5047 s->resync_mb_x=mb_x; | |
5048 s->resync_mb_y=mb_y; | |
5049 } | |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
5050 } |
456 | 5051 |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
5052 if( (s->resync_mb_x == s->mb_x) |
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
5053 && s->resync_mb_y+1 == s->mb_y){ |
2967 | 5054 s->first_slice_line=0; |
456 | 5055 } |
5056 | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
5057 s->mb_skipped=0; |
1616 | 5058 s->dquant=0; //only for QP_RD |
5059 | |
2230 | 5060 if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD |
327 | 5061 int next_block=0; |
456 | 5062 int pb_bits_count, pb2_bits_count, tex_pb_bits_count; |
326 | 5063 |
5064 copy_context_before_encode(&backup_s, s, -1); | |
456 | 5065 backup_s.pb= s->pb; |
5066 best_s.data_partitioning= s->data_partitioning; | |
745
25d7fb7c89be
better/cleaner error resilience (done in a 2nd pass after decoding)
michaelni
parents:
731
diff
changeset
|
5067 best_s.partitioned_frame= s->partitioned_frame; |
456 | 5068 if(s->data_partitioning){ |
5069 backup_s.pb2= s->pb2; | |
5070 backup_s.tex_pb= s->tex_pb; | |
5071 } | |
326 | 5072 |
1708 | 5073 if(mb_type&CANDIDATE_MB_TYPE_INTER){ |
327 | 5074 s->mv_dir = MV_DIR_FORWARD; |
295 | 5075 s->mv_type = MV_TYPE_16X16; |
294 | 5076 s->mb_intra= 0; |
324 | 5077 s->mv[0][0][0] = s->p_mv_table[xy][0]; |
5078 s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
2967 | 5079 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb, |
456 | 5080 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]); |
0 | 5081 } |
2967 | 5082 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){ |
1708 | 5083 s->mv_dir = MV_DIR_FORWARD; |
5084 s->mv_type = MV_TYPE_FIELD; | |
5085 s->mb_intra= 0; | |
5086 for(i=0; i<2; i++){ | |
5087 j= s->field_select[0][i] = s->p_field_select_table[i][xy]; | |
5088 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0]; | |
5089 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1]; | |
5090 } | |
2967 | 5091 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb, |
1708 | 5092 &dmin, &next_block, 0, 0); |
5093 } | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2627
diff
changeset
|
5094 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){ |
1494
3ee63c12ea30
optionally try to encode each MB with MV=<0,0> and choose the one with better RD
michaelni
parents:
1490
diff
changeset
|
5095 s->mv_dir = MV_DIR_FORWARD; |
3ee63c12ea30
optionally try to encode each MB with MV=<0,0> and choose the one with better RD
michaelni
parents:
1490
diff
changeset
|
5096 s->mv_type = MV_TYPE_16X16; |
3ee63c12ea30
optionally try to encode each MB with MV=<0,0> and choose the one with better RD
michaelni
parents:
1490
diff
changeset
|
5097 s->mb_intra= 0; |
3ee63c12ea30
optionally try to encode each MB with MV=<0,0> and choose the one with better RD
michaelni
parents:
1490
diff
changeset
|
5098 s->mv[0][0][0] = 0; |
3ee63c12ea30
optionally try to encode each MB with MV=<0,0> and choose the one with better RD
michaelni
parents:
1490
diff
changeset
|
5099 s->mv[0][0][1] = 0; |
2967 | 5100 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb, |
1494
3ee63c12ea30
optionally try to encode each MB with MV=<0,0> and choose the one with better RD
michaelni
parents:
1490
diff
changeset
|
5101 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]); |
3ee63c12ea30
optionally try to encode each MB with MV=<0,0> and choose the one with better RD
michaelni
parents:
1490
diff
changeset
|
5102 } |
2967 | 5103 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){ |
327 | 5104 s->mv_dir = MV_DIR_FORWARD; |
295 | 5105 s->mv_type = MV_TYPE_8X8; |
5106 s->mb_intra= 0; | |
5107 for(i=0; i<4; i++){ | |
1668
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
5108 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0]; |
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
5109 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1]; |
295 | 5110 } |
2967 | 5111 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb, |
456 | 5112 &dmin, &next_block, 0, 0); |
327 | 5113 } |
1708 | 5114 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){ |
327 | 5115 s->mv_dir = MV_DIR_FORWARD; |
5116 s->mv_type = MV_TYPE_16X16; | |
5117 s->mb_intra= 0; | |
5118 s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
5119 s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
2967 | 5120 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb, |
456 | 5121 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]); |
327 | 5122 } |
1708 | 5123 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){ |
327 | 5124 s->mv_dir = MV_DIR_BACKWARD; |
5125 s->mv_type = MV_TYPE_16X16; | |
5126 s->mb_intra= 0; | |
5127 s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
5128 s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
2967 | 5129 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb, |
456 | 5130 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]); |
327 | 5131 } |
1708 | 5132 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){ |
327 | 5133 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; |
5134 s->mv_type = MV_TYPE_16X16; | |
5135 s->mb_intra= 0; | |
5136 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
5137 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
5138 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
5139 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
2967 | 5140 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb, |
456 | 5141 &dmin, &next_block, 0, 0); |
327 | 5142 } |
1708 | 5143 if(mb_type&CANDIDATE_MB_TYPE_DIRECT){ |
936 | 5144 int mx= s->b_direct_mv_table[xy][0]; |
5145 int my= s->b_direct_mv_table[xy][1]; | |
2967 | 5146 |
327 | 5147 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; |
5148 s->mb_intra= 0; | |
936 | 5149 ff_mpeg4_set_direct_mv(s, mx, my); |
2967 | 5150 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb, |
936 | 5151 &dmin, &next_block, mx, my); |
295 | 5152 } |
2967 | 5153 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){ |
1708 | 5154 s->mv_dir = MV_DIR_FORWARD; |
5155 s->mv_type = MV_TYPE_FIELD; | |
5156 s->mb_intra= 0; | |
5157 for(i=0; i<2; i++){ | |
5158 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy]; | |
5159 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0]; | |
5160 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1]; | |
5161 } | |
2967 | 5162 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb, |
1708 | 5163 &dmin, &next_block, 0, 0); |
5164 } | |
2967 | 5165 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){ |
1708 | 5166 s->mv_dir = MV_DIR_BACKWARD; |
5167 s->mv_type = MV_TYPE_FIELD; | |
5168 s->mb_intra= 0; | |
5169 for(i=0; i<2; i++){ | |
5170 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy]; | |
5171 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0]; | |
5172 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1]; | |
5173 } | |
2967 | 5174 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb, |
1708 | 5175 &dmin, &next_block, 0, 0); |
5176 } | |
2967 | 5177 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){ |
1708 | 5178 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; |
5179 s->mv_type = MV_TYPE_FIELD; | |
5180 s->mb_intra= 0; | |
5181 for(dir=0; dir<2; dir++){ | |
5182 for(i=0; i<2; i++){ | |
5183 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy]; | |
5184 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0]; | |
5185 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1]; | |
5186 } | |
5187 } | |
2967 | 5188 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb, |
1708 | 5189 &dmin, &next_block, 0, 0); |
5190 } | |
5191 if(mb_type&CANDIDATE_MB_TYPE_INTRA){ | |
1051
e5a9dbf597d4
mpeg1 bframe encoding patch by (Rapha«³l LEGRAND) with some modifications by me
michaelni
parents:
1048
diff
changeset
|
5192 s->mv_dir = 0; |
295 | 5193 s->mv_type = MV_TYPE_16X16; |
294 | 5194 s->mb_intra= 1; |
5195 s->mv[0][0][0] = 0; | |
5196 s->mv[0][0][1] = 0; | |
2967 | 5197 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb, |
456 | 5198 &dmin, &next_block, 0, 0); |
1389 | 5199 if(s->h263_pred || s->h263_aic){ |
5200 if(best_s.mb_intra) | |
5201 s->mbintra_table[mb_x + mb_y*s->mb_stride]=1; | |
5202 else | |
5203 ff_clean_intra_table_entries(s); //old mode? | |
5204 } | |
295 | 5205 } |
1616 | 5206 |
5207 if(s->flags & CODEC_FLAG_QP_RD){ | |
5208 if(best_s.mv_type==MV_TYPE_16X16 && !(best_s.mv_dir&MV_DIRECT)){ | |
5209 const int last_qp= backup_s.qscale; | |
5210 int dquant, dir, qp, dc[6]; | |
1617 | 5211 DCTELEM ac[6][16]; |
1623 | 5212 const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0; |
2967 | 5213 |
1616 | 5214 assert(backup_s.dquant == 0); |
5215 | |
5216 //FIXME intra | |
5217 s->mv_dir= best_s.mv_dir; | |
5218 s->mv_type = MV_TYPE_16X16; | |
5219 s->mb_intra= best_s.mb_intra; | |
5220 s->mv[0][0][0] = best_s.mv[0][0][0]; | |
5221 s->mv[0][0][1] = best_s.mv[0][0][1]; | |
5222 s->mv[1][0][0] = best_s.mv[1][0][0]; | |
5223 s->mv[1][0][1] = best_s.mv[1][0][1]; | |
2967 | 5224 |
1616 | 5225 dir= s->pict_type == B_TYPE ? 2 : 1; |
1623 | 5226 if(last_qp + dir > s->avctx->qmax) dir= -dir; |
1616 | 5227 for(dquant= dir; dquant<=2 && dquant>=-2; dquant += dir){ |
5228 qp= last_qp + dquant; | |
5229 if(qp < s->avctx->qmin || qp > s->avctx->qmax) | |
5230 break; | |
5231 backup_s.dquant= dquant; | |
2229 | 5232 if(s->mb_intra && s->dc_val[0]){ |
1617 | 5233 for(i=0; i<6; i++){ |
5234 dc[i]= s->dc_val[0][ s->block_index[i] ]; | |
5235 memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16); | |
5236 } | |
1616 | 5237 } |
1617 | 5238 |
2967 | 5239 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb, |
1623 | 5240 &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]); |
1616 | 5241 if(best_s.qscale != qp){ |
2229 | 5242 if(s->mb_intra && s->dc_val[0]){ |
1617 | 5243 for(i=0; i<6; i++){ |
5244 s->dc_val[0][ s->block_index[i] ]= dc[i]; | |
5245 memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16); | |
5246 } | |
1616 | 5247 } |
5248 if(dir > 0 && dquant==dir){ | |
5249 dquant= 0; | |
5250 dir= -dir; | |
5251 }else | |
5252 break; | |
5253 } | |
5254 } | |
5255 qp= best_s.qscale; | |
5256 s->current_picture.qscale_table[xy]= qp; | |
5257 } | |
5258 } | |
5259 | |
326 | 5260 copy_context_after_encode(s, &best_s, -1); |
2967 | 5261 |
1786 | 5262 pb_bits_count= put_bits_count(&s->pb); |
456 | 5263 flush_put_bits(&s->pb); |
5264 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count); | |
5265 s->pb= backup_s.pb; | |
2967 | 5266 |
456 | 5267 if(s->data_partitioning){ |
1786 | 5268 pb2_bits_count= put_bits_count(&s->pb2); |
456 | 5269 flush_put_bits(&s->pb2); |
5270 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count); | |
5271 s->pb2= backup_s.pb2; | |
2967 | 5272 |
1786 | 5273 tex_pb_bits_count= put_bits_count(&s->tex_pb); |
456 | 5274 flush_put_bits(&s->tex_pb); |
5275 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count); | |
5276 s->tex_pb= backup_s.tex_pb; | |
5277 } | |
1786 | 5278 s->last_bits= put_bits_count(&s->pb); |
2967 | 5279 |
1389 | 5280 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) |
5281 ff_h263_update_motion_val(s); | |
2967 | 5282 |
1799 | 5283 if(next_block==0){ //FIXME 16 vs linesize16 |
5284 s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad , s->linesize ,16); | |
5285 s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize , s->uvlinesize, 8); | |
5286 s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8); | |
1389 | 5287 } |
5288 | |
5289 if(s->avctx->mb_decision == FF_MB_DECISION_BITS) | |
5290 MPV_decode_mb(s, s->block); | |
294 | 5291 } else { |
324 | 5292 int motion_x, motion_y; |
5293 s->mv_type=MV_TYPE_16X16; | |
294 | 5294 // only one MB-Type possible |
2967 | 5295 |
327 | 5296 switch(mb_type){ |
1708 | 5297 case CANDIDATE_MB_TYPE_INTRA: |
1051
e5a9dbf597d4
mpeg1 bframe encoding patch by (Rapha«³l LEGRAND) with some modifications by me
michaelni
parents:
1048
diff
changeset
|
5298 s->mv_dir = 0; |
294 | 5299 s->mb_intra= 1; |
324 | 5300 motion_x= s->mv[0][0][0] = 0; |
5301 motion_y= s->mv[0][0][1] = 0; | |
327 | 5302 break; |
1708 | 5303 case CANDIDATE_MB_TYPE_INTER: |
324 | 5304 s->mv_dir = MV_DIR_FORWARD; |
5305 s->mb_intra= 0; | |
5306 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0]; | |
5307 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1]; | |
327 | 5308 break; |
1708 | 5309 case CANDIDATE_MB_TYPE_INTER_I: |
5310 s->mv_dir = MV_DIR_FORWARD; | |
5311 s->mv_type = MV_TYPE_FIELD; | |
5312 s->mb_intra= 0; | |
5313 for(i=0; i<2; i++){ | |
5314 j= s->field_select[0][i] = s->p_field_select_table[i][xy]; | |
5315 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0]; | |
5316 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1]; | |
5317 } | |
5318 motion_x = motion_y = 0; | |
5319 break; | |
5320 case CANDIDATE_MB_TYPE_INTER4V: | |
456 | 5321 s->mv_dir = MV_DIR_FORWARD; |
5322 s->mv_type = MV_TYPE_8X8; | |
5323 s->mb_intra= 0; | |
5324 for(i=0; i<4; i++){ | |
1668
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
5325 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0]; |
30746f429df6
move motion_val & mb_type to AVFrame patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1662
diff
changeset
|
5326 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1]; |
456 | 5327 } |
5328 motion_x= motion_y= 0; | |
5329 break; | |
1708 | 5330 case CANDIDATE_MB_TYPE_DIRECT: |
324 | 5331 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT; |
5332 s->mb_intra= 0; | |
327 | 5333 motion_x=s->b_direct_mv_table[xy][0]; |
5334 motion_y=s->b_direct_mv_table[xy][1]; | |
936 | 5335 ff_mpeg4_set_direct_mv(s, motion_x, motion_y); |
327 | 5336 break; |
1708 | 5337 case CANDIDATE_MB_TYPE_BIDIR: |
324 | 5338 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; |
294 | 5339 s->mb_intra= 0; |
324 | 5340 motion_x=0; |
5341 motion_y=0; | |
5342 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0]; | |
5343 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1]; | |
5344 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0]; | |
5345 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1]; | |
327 | 5346 break; |
1708 | 5347 case CANDIDATE_MB_TYPE_BACKWARD: |
324 | 5348 s->mv_dir = MV_DIR_BACKWARD; |
5349 s->mb_intra= 0; | |
5350 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0]; | |
5351 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1]; | |
327 | 5352 break; |
1708 | 5353 case CANDIDATE_MB_TYPE_FORWARD: |
324 | 5354 s->mv_dir = MV_DIR_FORWARD; |
5355 s->mb_intra= 0; | |
5356 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0]; | |
5357 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1]; | |
5358 // printf(" %d %d ", motion_x, motion_y); | |
327 | 5359 break; |
1708 | 5360 case CANDIDATE_MB_TYPE_FORWARD_I: |
5361 s->mv_dir = MV_DIR_FORWARD; | |
5362 s->mv_type = MV_TYPE_FIELD; | |
5363 s->mb_intra= 0; | |
5364 for(i=0; i<2; i++){ | |
5365 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy]; | |
5366 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0]; | |
5367 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1]; | |
5368 } | |
5369 motion_x=motion_y=0; | |
5370 break; | |
5371 case CANDIDATE_MB_TYPE_BACKWARD_I: | |
5372 s->mv_dir = MV_DIR_BACKWARD; | |
5373 s->mv_type = MV_TYPE_FIELD; | |
5374 s->mb_intra= 0; | |
5375 for(i=0; i<2; i++){ | |
5376 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy]; | |
5377 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0]; | |
5378 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1]; | |
5379 } | |
5380 motion_x=motion_y=0; | |
5381 break; | |
5382 case CANDIDATE_MB_TYPE_BIDIR_I: | |
5383 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD; | |
5384 s->mv_type = MV_TYPE_FIELD; | |
5385 s->mb_intra= 0; | |
5386 for(dir=0; dir<2; dir++){ | |
5387 for(i=0; i<2; i++){ | |
5388 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy]; | |
5389 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0]; | |
5390 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1]; | |
5391 } | |
5392 } | |
5393 motion_x=motion_y=0; | |
5394 break; | |
327 | 5395 default: |
324 | 5396 motion_x=motion_y=0; //gcc warning fix |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1597
diff
changeset
|
5397 av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n"); |
294 | 5398 } |
1051
e5a9dbf597d4
mpeg1 bframe encoding patch by (Rapha«³l LEGRAND) with some modifications by me
michaelni
parents:
1048
diff
changeset
|
5399 |
324 | 5400 encode_mb(s, motion_x, motion_y); |
1051
e5a9dbf597d4
mpeg1 bframe encoding patch by (Rapha«³l LEGRAND) with some modifications by me
michaelni
parents:
1048
diff
changeset
|
5401 |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
5402 // RAL: Update last macroblock type |
1051
e5a9dbf597d4
mpeg1 bframe encoding patch by (Rapha«³l LEGRAND) with some modifications by me
michaelni
parents:
1048
diff
changeset
|
5403 s->last_mv_dir = s->mv_dir; |
2967 | 5404 |
1389 | 5405 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) |
5406 ff_h263_update_motion_val(s); | |
2967 | 5407 |
1389 | 5408 MPV_decode_mb(s, s->block); |
248
56ee684c48bb
- H.263+ decoder support for Advanded INTRA Coding (buggy)
pulento
parents:
244
diff
changeset
|
5409 } |
1051
e5a9dbf597d4
mpeg1 bframe encoding patch by (Rapha«³l LEGRAND) with some modifications by me
michaelni
parents:
1048
diff
changeset
|
5410 |
327 | 5411 /* clean the MV table in IPS frames for direct mode in B frames */ |
5412 if(s->mb_intra /* && I,P,S_TYPE */){ | |
5413 s->p_mv_table[xy][0]=0; | |
5414 s->p_mv_table[xy][1]=0; | |
5415 } | |
2967 | 5416 |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
5417 if(s->flags&CODEC_FLAG_PSNR){ |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
5418 int w= 16; |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
5419 int h= 16; |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
5420 |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
5421 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16; |
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
5422 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16; |
936 | 5423 |
2955 | 5424 s->current_picture.error[0] += sse( |
1389 | 5425 s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, |
5426 s->dest[0], w, h, s->linesize); | |
2955 | 5427 s->current_picture.error[1] += sse( |
1389 | 5428 s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8, |
5429 s->dest[1], w>>1, h>>1, s->uvlinesize); | |
2955 | 5430 s->current_picture.error[2] += sse( |
1389 | 5431 s, s->new_picture .data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8, |
5432 s->dest[2], w>>1, h>>1, s->uvlinesize); | |
909
8ae1e4c24e91
new PSNR code (now works with chroma, b frames, ...)
michaelni
parents:
903
diff
changeset
|
5433 } |
2345 | 5434 if(s->loop_filter){ |
5435 if(s->out_format == FMT_H263) | |
5436 ff_h263_loop_filter(s); | |
5437 } | |
1786 | 5438 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb)); |
0 | 5439 } |
5440 } | |
286 | 5441 |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
5442 //not beautiful here but we must write it before flushing so it has to be here |
456 | 5443 if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE) |
208 | 5444 msmpeg4_encode_ext_header(s); |
456 | 5445 |
1799 | 5446 write_slice_end(s); |
5447 | |
2967 | 5448 /* Send the last GOB if RTP */ |
1661 | 5449 if (s->avctx->rtp_callback) { |
2492
57ba17bd3924
rtp_callback: send number of mb patch by (Johan Bilien {jobi via.ecp fr)
michael
parents:
2450
diff
changeset
|
5450 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x; |
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
|
5451 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob; |
231 | 5452 /* Call the RTP callback to send the last GOB */ |
1799 | 5453 emms_c(); |
2492
57ba17bd3924
rtp_callback: send number of mb patch by (Johan Bilien {jobi via.ecp fr)
michael
parents:
2450
diff
changeset
|
5454 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb); |
231 | 5455 } |
1799 | 5456 |
5457 return 0; | |
5458 } | |
5459 | |
5460 #define MERGE(field) dst->field += src->field; src->field=0 | |
5461 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){ | |
2015
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5462 MERGE(me.scene_change_score); |
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5463 MERGE(me.mc_mb_var_sum_temp); |
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5464 MERGE(me.mb_var_sum_temp); |
1799 | 5465 } |
5466 | |
5467 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){ | |
5468 int i; | |
5469 | |
5470 MERGE(dct_count[0]); //note, the other dct vars are not part of the context | |
5471 MERGE(dct_count[1]); | |
5472 MERGE(mv_bits); | |
5473 MERGE(i_tex_bits); | |
5474 MERGE(p_tex_bits); | |
5475 MERGE(i_count); | |
5476 MERGE(f_count); | |
5477 MERGE(b_count); | |
5478 MERGE(skip_count); | |
5479 MERGE(misc_bits); | |
5480 MERGE(error_count); | |
5481 MERGE(padding_bug_score); | |
2955 | 5482 MERGE(current_picture.error[0]); |
5483 MERGE(current_picture.error[1]); | |
5484 MERGE(current_picture.error[2]); | |
1799 | 5485 |
5486 if(dst->avctx->noise_reduction){ | |
5487 for(i=0; i<64; i++){ | |
5488 MERGE(dct_error_sum[0][i]); | |
5489 MERGE(dct_error_sum[1][i]); | |
5490 } | |
5491 } | |
2967 | 5492 |
1799 | 5493 assert(put_bits_count(&src->pb) % 8 ==0); |
5494 assert(put_bits_count(&dst->pb) % 8 ==0); | |
5495 ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb)); | |
5496 flush_put_bits(&dst->pb); | |
5497 } | |
5498 | |
3766 | 5499 static int estimate_qp(MpegEncContext *s, int dry_run){ |
3939
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
5500 if (s->next_lambda){ |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
5501 s->current_picture_ptr->quality= |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
5502 s->current_picture.quality = s->next_lambda; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
5503 if(!dry_run) s->next_lambda= 0; |
16ace038f452
fix buffer underflows by reencoding the current frame with a higher QP
michael
parents:
3929
diff
changeset
|
5504 } else if (!s->fixed_qscale) { |
2974 | 5505 s->current_picture_ptr->quality= |
5506 s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run); | |
3766 | 5507 if (s->current_picture.quality < 0) |
5508 return -1; | |
5509 } | |
2974 | 5510 |
5511 if(s->adaptive_quant){ | |
5512 switch(s->codec_id){ | |
5513 case CODEC_ID_MPEG4: | |
5514 ff_clean_mpeg4_qscales(s); | |
5515 break; | |
5516 case CODEC_ID_H263: | |
5517 case CODEC_ID_H263P: | |
5518 case CODEC_ID_FLV1: | |
5519 ff_clean_h263_qscales(s); | |
5520 break; | |
5521 } | |
5522 | |
5523 s->lambda= s->lambda_table[0]; | |
5524 //FIXME broken | |
5525 }else | |
5526 s->lambda= s->current_picture.quality; | |
5527 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality); | |
5528 update_qscale(s); | |
3780
6741c0e88310
return 0 from encode_picture() and estimate_qp() on success
mru
parents:
3777
diff
changeset
|
5529 return 0; |
2974 | 5530 } |
5531 | |
3766 | 5532 static int encode_picture(MpegEncContext *s, int picture_number) |
1799 | 5533 { |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2015
diff
changeset
|
5534 int i; |
1799 | 5535 int bits; |
5536 | |
5537 s->picture_number = picture_number; | |
2967 | 5538 |
1799 | 5539 /* Reset the average MB variance */ |
2015
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5540 s->me.mb_var_sum_temp = |
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5541 s->me.mc_mb_var_sum_temp = 0; |
1799 | 5542 |
5543 /* we need to initialize some time vars before we can encode b-frames */ | |
5544 // RAL: Condition added for MPEG1VIDEO | |
5545 if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4)) | |
1834 | 5546 ff_set_mpeg4_time(s, s->picture_number); //FIXME rename and use has_b_frames or similar |
2967 | 5547 |
2015
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5548 s->me.scene_change_score=0; |
2967 | 5549 |
2053 | 5550 // s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME ratedistoration |
2967 | 5551 |
1799 | 5552 if(s->pict_type==I_TYPE){ |
5553 if(s->msmpeg4_version >= 3) s->no_rounding=1; | |
5554 else s->no_rounding=0; | |
5555 }else if(s->pict_type!=B_TYPE){ | |
5556 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4) | |
2967 | 5557 s->no_rounding ^= 1; |
5558 } | |
5559 | |
2981 | 5560 if(s->flags & CODEC_FLAG_PASS2){ |
3766 | 5561 if (estimate_qp(s,1) < 0) |
5562 return -1; | |
2981 | 5563 ff_get_2pass_fcode(s); |
5564 }else if(!(s->flags & CODEC_FLAG_QSCALE)){ | |
2980
529163f99474
improved lambda/qp guessing for motion estimation RD in first pass
michael
parents:
2979
diff
changeset
|
5565 if(s->pict_type==B_TYPE) |
2997
188f4ef688f2
second try of improved lambda/qp guessing for motion estimation RD in first pass
michael
parents:
2990
diff
changeset
|
5566 s->lambda= s->last_lambda_for[s->pict_type]; |
2980
529163f99474
improved lambda/qp guessing for motion estimation RD in first pass
michael
parents:
2979
diff
changeset
|
5567 else |
2997
188f4ef688f2
second try of improved lambda/qp guessing for motion estimation RD in first pass
michael
parents:
2990
diff
changeset
|
5568 s->lambda= s->last_lambda_for[s->last_non_b_pict_type]; |
2980
529163f99474
improved lambda/qp guessing for motion estimation RD in first pass
michael
parents:
2979
diff
changeset
|
5569 update_qscale(s); |
529163f99474
improved lambda/qp guessing for motion estimation RD in first pass
michael
parents:
2979
diff
changeset
|
5570 } |
2974 | 5571 |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
5572 s->mb_intra=0; //for the rate distortion & bit compare functions |
1799 | 5573 for(i=1; i<s->avctx->thread_count; i++){ |
1827 | 5574 ff_update_duplicate_context(s->thread_context[i], s); |
1799 | 5575 } |
1962 | 5576 |
5577 ff_init_me(s); | |
5578 | |
1799 | 5579 /* Estimate motion for every MB */ |
5580 if(s->pict_type != I_TYPE){ | |
2584 | 5581 s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8; |
5582 s->lambda2= (s->lambda2* s->avctx->me_penalty_compensation + 128)>>8; | |
1955
5dafb10e0252
reuse motion vectors/mb types/field select values of the source video, if the SSE for a macroblock which is predicted with these values is below me_threshold
michael
parents:
1948
diff
changeset
|
5583 if(s->pict_type != B_TYPE && s->avctx->me_threshold==0){ |
1799 | 5584 if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){ |
5585 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count); | |
5586 } | |
5587 } | |
5588 | |
5589 s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count); | |
5590 }else /* if(s->pict_type == I_TYPE) */{ | |
5591 /* I-Frame */ | |
5592 for(i=0; i<s->mb_stride*s->mb_height; i++) | |
5593 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA; | |
2967 | 5594 |
1799 | 5595 if(!s->fixed_qscale){ |
5596 /* finding spatial complexity for I-frame rate control */ | |
1804
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
5597 s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count); |
1799 | 5598 } |
5599 } | |
1804
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
5600 for(i=1; i<s->avctx->thread_count; i++){ |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
5601 merge_context_after_me(s, s->thread_context[i]); |
cc0a15943a87
multithreaded spatial complexity calculation for i frames
michael
parents:
1803
diff
changeset
|
5602 } |
2015
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5603 s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp; |
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5604 s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp; |
1799 | 5605 emms_c(); |
5606 | |
2015
3ab8f3e2ae6a
moving motion estimation specific variables from MpegEncContext -> MotionEstContext
michael
parents:
1994
diff
changeset
|
5607 if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == P_TYPE){ |
1799 | 5608 s->pict_type= I_TYPE; |
5609 for(i=0; i<s->mb_stride*s->mb_height; i++) | |
5610 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA; | |
5611 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum); | |
5612 } | |
5613 | |
5614 if(!s->umvplus){ | |
5615 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) { | |
5616 s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER); | |
5617 | |
5618 if(s->flags & CODEC_FLAG_INTERLACED_ME){ | |
5619 int a,b; | |
5620 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select | |
5621 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I); | |
5622 s->f_code= FFMAX(s->f_code, FFMAX(a,b)); | |
5623 } | |
2967 | 5624 |
1799 | 5625 ff_fix_long_p_mvs(s); |
5626 ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0); | |
5627 if(s->flags & CODEC_FLAG_INTERLACED_ME){ | |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2015
diff
changeset
|
5628 int j; |
1799 | 5629 for(i=0; i<2; i++){ |
5630 for(j=0; j<2; j++) | |
2967 | 5631 ff_fix_long_mvs(s, s->p_field_select_table[i], j, |
1799 | 5632 s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0); |
5633 } | |
5634 } | |
5635 } | |
5636 | |
5637 if(s->pict_type==B_TYPE){ | |
5638 int a, b; | |
5639 | |
5640 a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD); | |
5641 b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR); | |
5642 s->f_code = FFMAX(a, b); | |
5643 | |
5644 a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD); | |
5645 b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR); | |
5646 s->b_code = FFMAX(a, b); | |
5647 | |
5648 ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1); | |
5649 ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1); | |
5650 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1); | |
5651 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1); | |
5652 if(s->flags & CODEC_FLAG_INTERLACED_ME){ | |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2015
diff
changeset
|
5653 int dir, j; |
1799 | 5654 for(dir=0; dir<2; dir++){ |
5655 for(i=0; i<2; i++){ | |
5656 for(j=0; j<2; j++){ | |
2967 | 5657 int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I) |
1799 | 5658 : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I); |
2967 | 5659 ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j, |
1799 | 5660 s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1); |
5661 } | |
5662 } | |
5663 } | |
5664 } | |
5665 } | |
5666 } | |
5667 | |
3766 | 5668 if (estimate_qp(s, 0) < 0) |
5669 return -1; | |
2967 | 5670 |
5671 if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==I_TYPE && !(s->flags & CODEC_FLAG_QSCALE)) | |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
5672 s->qscale= 3; //reduce clipping problems |
2967 | 5673 |
1799 | 5674 if (s->out_format == FMT_MJPEG) { |
5675 /* for mjpeg, we do include qscale in the matrix */ | |
5676 s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0]; | |
5677 for(i=1;i<64;i++){ | |
5678 int j= s->dsp.idct_permutation[i]; | |
5679 | |
3286
30d412952107
clip_uint8 now return an uint8_t so no need to &0xFF
aurel
parents:
3281
diff
changeset
|
5680 s->intra_matrix[j] = clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3); |
1799 | 5681 } |
2967 | 5682 convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16, |
2385
4605a8b188c3
warn the user if the quantization code could overflow with the selected custom matrix and qp limits
michael
parents:
2382
diff
changeset
|
5683 s->intra_matrix, s->intra_quant_bias, 8, 8, 1); |
1799 | 5684 s->qscale= 8; |
5685 } | |
2967 | 5686 |
1799 | 5687 //FIXME var duplication |
2235
0c04463f79d8
10l (scene change pict_type was wrong after coded_frame fix) patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2231
diff
changeset
|
5688 s->current_picture_ptr->key_frame= |
1799 | 5689 s->current_picture.key_frame= s->pict_type == I_TYPE; //FIXME pic_ptr |
2235
0c04463f79d8
10l (scene change pict_type was wrong after coded_frame fix) patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2231
diff
changeset
|
5690 s->current_picture_ptr->pict_type= |
1799 | 5691 s->current_picture.pict_type= s->pict_type; |
5692 | |
5693 if(s->current_picture.key_frame) | |
5694 s->picture_in_gop_number=0; | |
5695 | |
5696 s->last_bits= put_bits_count(&s->pb); | |
5697 switch(s->out_format) { | |
5698 case FMT_MJPEG: | |
5699 mjpeg_picture_header(s); | |
5700 break; | |
2639 | 5701 #ifdef CONFIG_H261_ENCODER |
2327
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
5702 case FMT_H261: |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
5703 ff_h261_encode_picture_header(s, picture_number); |
5e5cf598a48b
H.261 encoder by (Maarten Daniels <maarten dot daniels at luc dot ac dot be>)
michael
parents:
2298
diff
changeset
|
5704 break; |
2639 | 5705 #endif |
1799 | 5706 case FMT_H263: |
2967 | 5707 if (s->codec_id == CODEC_ID_WMV2) |
1799 | 5708 ff_wmv2_encode_picture_header(s, picture_number); |
2967 | 5709 else if (s->h263_msmpeg4) |
1799 | 5710 msmpeg4_encode_picture_header(s, picture_number); |
5711 else if (s->h263_pred) | |
5712 mpeg4_encode_picture_header(s, picture_number); | |
2639 | 5713 #ifdef CONFIG_RV10_ENCODER |
2967 | 5714 else if (s->codec_id == CODEC_ID_RV10) |
1799 | 5715 rv10_encode_picture_header(s, picture_number); |
2639 | 5716 #endif |
5717 #ifdef CONFIG_RV20_ENCODER | |
2967 | 5718 else if (s->codec_id == CODEC_ID_RV20) |
2380 | 5719 rv20_encode_picture_header(s, picture_number); |
2639 | 5720 #endif |
1799 | 5721 else if (s->codec_id == CODEC_ID_FLV1) |
5722 ff_flv_encode_picture_header(s, picture_number); | |
5723 else | |
5724 h263_encode_picture_header(s, picture_number); | |
5725 break; | |
5726 case FMT_MPEG1: | |
5727 mpeg1_encode_picture_header(s, picture_number); | |
5728 break; | |
5729 case FMT_H264: | |
5730 break; | |
5731 default: | |
5732 assert(0); | |
5733 } | |
5734 bits= put_bits_count(&s->pb); | |
5735 s->header_bits= bits - s->last_bits; | |
2967 | 5736 |
1799 | 5737 for(i=1; i<s->avctx->thread_count; i++){ |
5738 update_duplicate_context_after_me(s->thread_context[i], s); | |
5739 } | |
5740 s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count); | |
5741 for(i=1; i<s->avctx->thread_count; i++){ | |
5742 merge_context_after_encode(s, s->thread_context[i]); | |
5743 } | |
5744 emms_c(); | |
3780
6741c0e88310
return 0 from encode_picture() and estimate_qp() on success
mru
parents:
3777
diff
changeset
|
5745 return 0; |
0 | 5746 } |
5747 | |
1719 | 5748 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){ |
1597 | 5749 const int intra= s->mb_intra; |
5750 int i; | |
5751 | |
1599 | 5752 s->dct_count[intra]++; |
5753 | |
1597 | 5754 for(i=0; i<64; i++){ |
5755 int level= block[i]; | |
5756 | |
5757 if(level){ | |
5758 if(level>0){ | |
5759 s->dct_error_sum[intra][i] += level; | |
5760 level -= s->dct_offset[intra][i]; | |
5761 if(level<0) level=0; | |
5762 }else{ | |
5763 s->dct_error_sum[intra][i] -= level; | |
5764 level += s->dct_offset[intra][i]; | |
5765 if(level>0) level=0; | |
5766 } | |
5767 block[i]= level; | |
5768 } | |
5769 } | |
5770 } | |
5771 | |
2967 | 5772 static int dct_quantize_trellis_c(MpegEncContext *s, |
945 | 5773 DCTELEM *block, int n, |
5774 int qscale, int *overflow){ | |
5775 const int *qmat; | |
1064 | 5776 const uint8_t *scantable= s->intra_scantable.scantable; |
1718 | 5777 const uint8_t *perm_scantable= s->intra_scantable.permutated; |
945 | 5778 int max=0; |
5779 unsigned int threshold1, threshold2; | |
5780 int bias=0; | |
5781 int run_tab[65]; | |
5782 int level_tab[65]; | |
5783 int score_tab[65]; | |
1718 | 5784 int survivor[65]; |
5785 int survivor_count; | |
946 | 5786 int last_run=0; |
5787 int last_level=0; | |
5788 int last_score= 0; | |
1718 | 5789 int last_i; |
1715 | 5790 int coeff[2][64]; |
945 | 5791 int coeff_count[64]; |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
5792 int qmul, qadd, start_i, last_non_zero, i, dc; |
945 | 5793 const int esc_length= s->ac_esc_length; |
5794 uint8_t * length; | |
5795 uint8_t * last_length; | |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
5796 const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6); |
2967 | 5797 |
1092 | 5798 s->dsp.fdct (block); |
2967 | 5799 |
1597 | 5800 if(s->dct_error_sum) |
1719 | 5801 s->denoise_dct(s, block); |
945 | 5802 qmul= qscale*16; |
5803 qadd= ((qscale-1)|1)*8; | |
946 | 5804 |
945 | 5805 if (s->mb_intra) { |
5806 int q; | |
5807 if (!s->h263_aic) { | |
5808 if (n < 4) | |
5809 q = s->y_dc_scale; | |
5810 else | |
5811 q = s->c_dc_scale; | |
5812 q = q << 3; | |
5813 } else{ | |
5814 /* For AIC we skip quant/dequant of INTRADC */ | |
5815 q = 1 << 3; | |
5816 qadd=0; | |
5817 } | |
2967 | 5818 |
945 | 5819 /* note: block[0] is assumed to be positive */ |
5820 block[0] = (block[0] + (q >> 1)) / q; | |
5821 start_i = 1; | |
5822 last_non_zero = 0; | |
5823 qmat = s->q_intra_matrix[qscale]; | |
1422
efeed6e29f9b
oooooops, having 2 Eterms open and doing cvs diff in one and cvs commit in the other sucks, especially if they are in different directories (MPlayer/main/libavcodec vs. ffmpeg/libavcodec)
michaelni
parents:
1421
diff
changeset
|
5824 if(s->mpeg_quant || s->out_format == FMT_MPEG1) |
945 | 5825 bias= 1<<(QMAT_SHIFT-1); |
5826 length = s->intra_ac_vlc_length; | |
5827 last_length= s->intra_ac_vlc_last_length; | |
5828 } else { | |
5829 start_i = 0; | |
5830 last_non_zero = -1; | |
5831 qmat = s->q_inter_matrix[qscale]; | |
5832 length = s->inter_ac_vlc_length; | |
5833 last_length= s->inter_ac_vlc_last_length; | |
5834 } | |
1718 | 5835 last_i= start_i; |
945 | 5836 |
5837 threshold1= (1<<QMAT_SHIFT) - bias - 1; | |
5838 threshold2= (threshold1<<1); | |
946 | 5839 |
1715 | 5840 for(i=63; i>=start_i; i--) { |
5841 const int j = scantable[i]; | |
5842 int level = block[j] * qmat[j]; | |
5843 | |
5844 if(((unsigned)(level+threshold1))>threshold2){ | |
5845 last_non_zero = i; | |
5846 break; | |
5847 } | |
5848 } | |
5849 | |
5850 for(i=start_i; i<=last_non_zero; i++) { | |
945 | 5851 const int j = scantable[i]; |
1715 | 5852 int level = block[j] * qmat[j]; |
945 | 5853 |
5854 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
5855 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
5856 if(((unsigned)(level+threshold1))>threshold2){ | |
5857 if(level>0){ | |
5858 level= (bias + level)>>QMAT_SHIFT; | |
1718 | 5859 coeff[0][i]= level; |
5860 coeff[1][i]= level-1; | |
946 | 5861 // coeff[2][k]= level-2; |
945 | 5862 }else{ |
5863 level= (bias - level)>>QMAT_SHIFT; | |
1718 | 5864 coeff[0][i]= -level; |
5865 coeff[1][i]= -level+1; | |
946 | 5866 // coeff[2][k]= -level+2; |
945 | 5867 } |
1718 | 5868 coeff_count[i]= FFMIN(level, 2); |
5869 assert(coeff_count[i]); | |
945 | 5870 max |=level; |
5871 }else{ | |
1718 | 5872 coeff[0][i]= (level>>31)|1; |
5873 coeff_count[i]= 1; | |
945 | 5874 } |
5875 } | |
2967 | 5876 |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
5877 *overflow= s->max_qcoeff < max; //overflow might have happened |
2967 | 5878 |
945 | 5879 if(last_non_zero < start_i){ |
5880 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM)); | |
5881 return last_non_zero; | |
5882 } | |
5883 | |
1718 | 5884 score_tab[start_i]= 0; |
5885 survivor[0]= start_i; | |
5886 survivor_count= 1; | |
2967 | 5887 |
1718 | 5888 for(i=start_i; i<=last_non_zero; i++){ |
5889 int level_index, j; | |
5890 const int dct_coeff= ABS(block[ scantable[i] ]); | |
945 | 5891 const int zero_distoration= dct_coeff*dct_coeff; |
946 | 5892 int best_score=256*256*256*120; |
945 | 5893 for(level_index=0; level_index < coeff_count[i]; level_index++){ |
5894 int distoration; | |
5895 int level= coeff[level_index][i]; | |
1716 | 5896 const int alevel= ABS(level); |
945 | 5897 int unquant_coeff; |
2967 | 5898 |
945 | 5899 assert(level); |
5900 | |
5901 if(s->out_format == FMT_H263){ | |
1716 | 5902 unquant_coeff= alevel*qmul + qadd; |
947 | 5903 }else{ //MPEG1 |
1718 | 5904 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize |
947 | 5905 if(s->mb_intra){ |
1716 | 5906 unquant_coeff = (int)( alevel * qscale * s->intra_matrix[j]) >> 3; |
947 | 5907 unquant_coeff = (unquant_coeff - 1) | 1; |
5908 }else{ | |
1716 | 5909 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4; |
947 | 5910 unquant_coeff = (unquant_coeff - 1) | 1; |
5911 } | |
5912 unquant_coeff<<= 3; | |
5913 } | |
946 | 5914 |
1715 | 5915 distoration= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distoration; |
945 | 5916 level+=64; |
5917 if((level&(~127)) == 0){ | |
1718 | 5918 for(j=survivor_count-1; j>=0; j--){ |
5919 int run= i - survivor[j]; | |
947 | 5920 int score= distoration + length[UNI_AC_ENC_INDEX(run, level)]*lambda; |
945 | 5921 score += score_tab[i-run]; |
2967 | 5922 |
945 | 5923 if(score < best_score){ |
1718 | 5924 best_score= score; |
945 | 5925 run_tab[i+1]= run; |
5926 level_tab[i+1]= level-64; | |
5927 } | |
5928 } | |
5929 | |
5930 if(s->out_format == FMT_H263){ | |
1718 | 5931 for(j=survivor_count-1; j>=0; j--){ |
5932 int run= i - survivor[j]; | |
947 | 5933 int score= distoration + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda; |
945 | 5934 score += score_tab[i-run]; |
946 | 5935 if(score < last_score){ |
5936 last_score= score; | |
5937 last_run= run; | |
5938 last_level= level-64; | |
5939 last_i= i+1; | |
945 | 5940 } |
5941 } | |
5942 } | |
5943 }else{ | |
5944 distoration += esc_length*lambda; | |
1718 | 5945 for(j=survivor_count-1; j>=0; j--){ |
5946 int run= i - survivor[j]; | |
945 | 5947 int score= distoration + score_tab[i-run]; |
2967 | 5948 |
945 | 5949 if(score < best_score){ |
1718 | 5950 best_score= score; |
945 | 5951 run_tab[i+1]= run; |
5952 level_tab[i+1]= level-64; | |
5953 } | |
5954 } | |
5955 | |
5956 if(s->out_format == FMT_H263){ | |
1718 | 5957 for(j=survivor_count-1; j>=0; j--){ |
5958 int run= i - survivor[j]; | |
945 | 5959 int score= distoration + score_tab[i-run]; |
946 | 5960 if(score < last_score){ |
5961 last_score= score; | |
5962 last_run= run; | |
5963 last_level= level-64; | |
5964 last_i= i+1; | |
945 | 5965 } |
5966 } | |
5967 } | |
5968 } | |
5969 } | |
2967 | 5970 |
1718 | 5971 score_tab[i+1]= best_score; |
5972 | |
946 | 5973 //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level |
1718 | 5974 if(last_non_zero <= 27){ |
5975 for(; survivor_count; survivor_count--){ | |
5976 if(score_tab[ survivor[survivor_count-1] ] <= best_score) | |
5977 break; | |
5978 } | |
5979 }else{ | |
5980 for(; survivor_count; survivor_count--){ | |
5981 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda) | |
5982 break; | |
5983 } | |
1505
010f76d07a27
use lagrange multipler instead of qp for ratecontrol, this may break some things, tell me ASAP if u notice anything broken
michaelni
parents:
1497
diff
changeset
|
5984 } |
1718 | 5985 |
5986 survivor[ survivor_count++ ]= i+1; | |
945 | 5987 } |
946 | 5988 |
945 | 5989 if(s->out_format != FMT_H263){ |
946 | 5990 last_score= 256*256*256*120; |
1718 | 5991 for(i= survivor[0]; i<=last_non_zero + 1; i++){ |
946 | 5992 int score= score_tab[i]; |
947 | 5993 if(i) score += lambda*2; //FIXME exacter? |
5994 | |
946 | 5995 if(score < last_score){ |
5996 last_score= score; | |
5997 last_i= i; | |
5998 last_level= level_tab[i]; | |
5999 last_run= run_tab[i]; | |
6000 } | |
945 | 6001 } |
6002 } | |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6003 |
1715 | 6004 s->coded_score[n] = last_score; |
2967 | 6005 |
1716 | 6006 dc= ABS(block[0]); |
1718 | 6007 last_non_zero= last_i - 1; |
945 | 6008 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM)); |
2967 | 6009 |
945 | 6010 if(last_non_zero < start_i) |
6011 return last_non_zero; | |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6012 |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6013 if(last_non_zero == 0 && start_i == 0){ |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6014 int best_level= 0; |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6015 int best_score= dc * dc; |
2967 | 6016 |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6017 for(i=0; i<coeff_count[0]; i++){ |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6018 int level= coeff[i][0]; |
1716 | 6019 int alevel= ABS(level); |
6020 int unquant_coeff, score, distortion; | |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6021 |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6022 if(s->out_format == FMT_H263){ |
1716 | 6023 unquant_coeff= (alevel*qmul + qadd)>>3; |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6024 }else{ //MPEG1 |
1716 | 6025 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4; |
6026 unquant_coeff = (unquant_coeff - 1) | 1; | |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6027 } |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6028 unquant_coeff = (unquant_coeff + 4) >> 3; |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6029 unquant_coeff<<= 3 + 3; |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6030 |
1716 | 6031 distortion= (unquant_coeff - dc) * (unquant_coeff - dc); |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6032 level+=64; |
1716 | 6033 if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda; |
6034 else score= distortion + esc_length*lambda; | |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6035 |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6036 if(score < best_score){ |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6037 best_score= score; |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6038 best_level= level - 64; |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6039 } |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6040 } |
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6041 block[0]= best_level; |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6042 s->coded_score[n] = best_score - dc*dc; |
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6043 if(best_level == 0) return -1; |
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6044 else return last_non_zero; |
1480
8657d6800d15
slight quality improvement ( < +0.04 PSNR) for high bitrate videos with trellis quantization
michaelni
parents:
1471
diff
changeset
|
6045 } |
1490
0355f2b3519a
rate distortion optimal cbp support (h263/mpeg4 non intra only)
michaelni
parents:
1480
diff
changeset
|
6046 |
946 | 6047 i= last_i; |
6048 assert(last_level); | |
1718 | 6049 |
6050 block[ perm_scantable[last_non_zero] ]= last_level; | |
946 | 6051 i -= last_run + 1; |
2967 | 6052 |
1718 | 6053 for(; i>start_i; i -= run_tab[i] + 1){ |
6054 block[ perm_scantable[i-1] ]= level_tab[i]; | |
945 | 6055 } |
6056 | |
6057 return last_non_zero; | |
6058 } | |
6059 | |
1777 | 6060 //#define REFINE_STATS 1 |
6061 static int16_t basis[64][64]; | |
6062 | |
6063 static void build_basis(uint8_t *perm){ | |
6064 int i, j, x, y; | |
6065 emms_c(); | |
6066 for(i=0; i<8; i++){ | |
6067 for(j=0; j<8; j++){ | |
6068 for(y=0; y<8; y++){ | |
6069 for(x=0; x<8; x++){ | |
6070 double s= 0.25*(1<<BASIS_SHIFT); | |
6071 int index= 8*i + j; | |
6072 int perm_index= perm[index]; | |
6073 if(i==0) s*= sqrt(0.5); | |
6074 if(j==0) s*= sqrt(0.5); | |
6075 basis[perm_index][8*x + y]= lrintf(s * cos((M_PI/8.0)*i*(x+0.5)) * cos((M_PI/8.0)*j*(y+0.5))); | |
6076 } | |
6077 } | |
6078 } | |
6079 } | |
6080 } | |
6081 | |
6082 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise? | |
6083 DCTELEM *block, int16_t *weight, DCTELEM *orig, | |
6084 int n, int qscale){ | |
6085 int16_t rem[64]; | |
3089 | 6086 DECLARE_ALIGNED_16(DCTELEM, d1[64]); |
1777 | 6087 const int *qmat; |
6088 const uint8_t *scantable= s->intra_scantable.scantable; | |
6089 const uint8_t *perm_scantable= s->intra_scantable.permutated; | |
6090 // unsigned int threshold1, threshold2; | |
6091 // int bias=0; | |
6092 int run_tab[65]; | |
6093 int prev_run=0; | |
6094 int prev_level=0; | |
6095 int qmul, qadd, start_i, last_non_zero, i, dc; | |
6096 uint8_t * length; | |
6097 uint8_t * last_length; | |
6098 int lambda; | |
6099 int rle_index, run, q, sum; | |
6100 #ifdef REFINE_STATS | |
6101 static int count=0; | |
6102 static int after_last=0; | |
6103 static int to_zero=0; | |
6104 static int from_zero=0; | |
6105 static int raise=0; | |
6106 static int lower=0; | |
6107 static int messed_sign=0; | |
6108 #endif | |
6109 | |
6110 if(basis[0][0] == 0) | |
6111 build_basis(s->dsp.idct_permutation); | |
2967 | 6112 |
1777 | 6113 qmul= qscale*2; |
6114 qadd= (qscale-1)|1; | |
6115 if (s->mb_intra) { | |
6116 if (!s->h263_aic) { | |
6117 if (n < 4) | |
6118 q = s->y_dc_scale; | |
6119 else | |
6120 q = s->c_dc_scale; | |
6121 } else{ | |
6122 /* For AIC we skip quant/dequant of INTRADC */ | |
6123 q = 1; | |
6124 qadd=0; | |
6125 } | |
6126 q <<= RECON_SHIFT-3; | |
6127 /* note: block[0] is assumed to be positive */ | |
6128 dc= block[0]*q; | |
6129 // block[0] = (block[0] + (q >> 1)) / q; | |
6130 start_i = 1; | |
6131 qmat = s->q_intra_matrix[qscale]; | |
6132 // if(s->mpeg_quant || s->out_format == FMT_MPEG1) | |
6133 // bias= 1<<(QMAT_SHIFT-1); | |
6134 length = s->intra_ac_vlc_length; | |
6135 last_length= s->intra_ac_vlc_last_length; | |
6136 } else { | |
6137 dc= 0; | |
6138 start_i = 0; | |
6139 qmat = s->q_inter_matrix[qscale]; | |
6140 length = s->inter_ac_vlc_length; | |
6141 last_length= s->inter_ac_vlc_last_length; | |
6142 } | |
6143 last_non_zero = s->block_last_index[n]; | |
6144 | |
6145 #ifdef REFINE_STATS | |
6146 {START_TIMER | |
6147 #endif | |
1784 | 6148 dc += (1<<(RECON_SHIFT-1)); |
6149 for(i=0; i<64; i++){ | |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
6150 rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME use orig dirrectly instead of copying to rem[] |
1777 | 6151 } |
6152 #ifdef REFINE_STATS | |
6153 STOP_TIMER("memset rem[]")} | |
6154 #endif | |
6155 sum=0; | |
6156 for(i=0; i<64; i++){ | |
6157 int one= 36; | |
6158 int qns=4; | |
6159 int w; | |
6160 | |
6161 w= ABS(weight[i]) + qns*one; | |
6162 w= 15 + (48*qns*one + w/2)/w; // 16 .. 63 | |
6163 | |
6164 weight[i] = w; | |
6165 // w=weight[i] = (63*qns + (w/2)) / w; | |
2967 | 6166 |
1777 | 6167 assert(w>0); |
6168 assert(w<(1<<6)); | |
6169 sum += w*w; | |
6170 } | |
6171 lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6); | |
6172 #ifdef REFINE_STATS | |
6173 {START_TIMER | |
6174 #endif | |
6175 run=0; | |
6176 rle_index=0; | |
6177 for(i=start_i; i<=last_non_zero; i++){ | |
6178 int j= perm_scantable[i]; | |
6179 const int level= block[j]; | |
6180 int coeff; | |
2967 | 6181 |
1777 | 6182 if(level){ |
6183 if(level<0) coeff= qmul*level - qadd; | |
6184 else coeff= qmul*level + qadd; | |
6185 run_tab[rle_index++]=run; | |
6186 run=0; | |
6187 | |
1784 | 6188 s->dsp.add_8x8basis(rem, basis[j], coeff); |
1777 | 6189 }else{ |
6190 run++; | |
6191 } | |
6192 } | |
6193 #ifdef REFINE_STATS | |
6194 if(last_non_zero>0){ | |
6195 STOP_TIMER("init rem[]") | |
6196 } | |
6197 } | |
6198 | |
6199 {START_TIMER | |
6200 #endif | |
6201 for(;;){ | |
1784 | 6202 int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0); |
1777 | 6203 int best_coeff=0; |
6204 int best_change=0; | |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2015
diff
changeset
|
6205 int run2, best_unquant_change=0, analyze_gradient; |
1785
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6206 #ifdef REFINE_STATS |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6207 {START_TIMER |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6208 #endif |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6209 analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3; |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6210 |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6211 if(analyze_gradient){ |
1777 | 6212 #ifdef REFINE_STATS |
6213 {START_TIMER | |
6214 #endif | |
1785
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6215 for(i=0; i<64; i++){ |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6216 int w= weight[i]; |
2967 | 6217 |
1785
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6218 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12); |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6219 } |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6220 #ifdef REFINE_STATS |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6221 STOP_TIMER("rem*w*w")} |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6222 {START_TIMER |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6223 #endif |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6224 s->dsp.fdct(d1); |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6225 #ifdef REFINE_STATS |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6226 STOP_TIMER("dct")} |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6227 #endif |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6228 } |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6229 |
1777 | 6230 if(start_i){ |
6231 const int level= block[0]; | |
6232 int change, old_coeff; | |
6233 | |
6234 assert(s->mb_intra); | |
2967 | 6235 |
1777 | 6236 old_coeff= q*level; |
2967 | 6237 |
1777 | 6238 for(change=-1; change<=1; change+=2){ |
6239 int new_level= level + change; | |
6240 int score, new_coeff; | |
2967 | 6241 |
1777 | 6242 new_coeff= q*new_level; |
6243 if(new_coeff >= 2048 || new_coeff < 0) | |
6244 continue; | |
6245 | |
1784 | 6246 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff); |
1777 | 6247 if(score<best_score){ |
6248 best_score= score; | |
6249 best_coeff= 0; | |
6250 best_change= change; | |
6251 best_unquant_change= new_coeff - old_coeff; | |
6252 } | |
6253 } | |
6254 } | |
2967 | 6255 |
1777 | 6256 run=0; |
6257 rle_index=0; | |
6258 run2= run_tab[rle_index++]; | |
6259 prev_level=0; | |
6260 prev_run=0; | |
6261 | |
6262 for(i=start_i; i<64; i++){ | |
6263 int j= perm_scantable[i]; | |
6264 const int level= block[j]; | |
6265 int change, old_coeff; | |
6266 | |
6267 if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1) | |
6268 break; | |
6269 | |
6270 if(level){ | |
6271 if(level<0) old_coeff= qmul*level - qadd; | |
6272 else old_coeff= qmul*level + qadd; | |
6273 run2= run_tab[rle_index++]; //FIXME ! maybe after last | |
6274 }else{ | |
6275 old_coeff=0; | |
6276 run2--; | |
6277 assert(run2>=0 || i >= last_non_zero ); | |
6278 } | |
2967 | 6279 |
1777 | 6280 for(change=-1; change<=1; change+=2){ |
6281 int new_level= level + change; | |
6282 int score, new_coeff, unquant_change; | |
2967 | 6283 |
1777 | 6284 score=0; |
6285 if(s->avctx->quantizer_noise_shaping < 2 && ABS(new_level) > ABS(level)) | |
6286 continue; | |
6287 | |
6288 if(new_level){ | |
6289 if(new_level<0) new_coeff= qmul*new_level - qadd; | |
6290 else new_coeff= qmul*new_level + qadd; | |
6291 if(new_coeff >= 2048 || new_coeff <= -2048) | |
6292 continue; | |
6293 //FIXME check for overflow | |
2967 | 6294 |
1777 | 6295 if(level){ |
6296 if(level < 63 && level > -63){ | |
6297 if(i < last_non_zero) | |
6298 score += length[UNI_AC_ENC_INDEX(run, new_level+64)] | |
6299 - length[UNI_AC_ENC_INDEX(run, level+64)]; | |
6300 else | |
6301 score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)] | |
6302 - last_length[UNI_AC_ENC_INDEX(run, level+64)]; | |
6303 } | |
6304 }else{ | |
6305 assert(ABS(new_level)==1); | |
2967 | 6306 |
1785
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6307 if(analyze_gradient){ |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6308 int g= d1[ scantable[i] ]; |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6309 if(g && (g^new_level) >= 0) |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6310 continue; |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6311 } |
14d918d0ef42
analyze the gradient of the error surface to avoid a few checks in the quantizing noise shaper
michael
parents:
1784
diff
changeset
|
6312 |
1777 | 6313 if(i < last_non_zero){ |
6314 int next_i= i + run2 + 1; | |
6315 int next_level= block[ perm_scantable[next_i] ] + 64; | |
2967 | 6316 |
1777 | 6317 if(next_level&(~127)) |
6318 next_level= 0; | |
6319 | |
6320 if(next_i < last_non_zero) | |
6321 score += length[UNI_AC_ENC_INDEX(run, 65)] | |
6322 + length[UNI_AC_ENC_INDEX(run2, next_level)] | |
6323 - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]; | |
6324 else | |
6325 score += length[UNI_AC_ENC_INDEX(run, 65)] | |
6326 + last_length[UNI_AC_ENC_INDEX(run2, next_level)] | |
6327 - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]; | |
6328 }else{ | |
6329 score += last_length[UNI_AC_ENC_INDEX(run, 65)]; | |
6330 if(prev_level){ | |
6331 score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)] | |
6332 - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]; | |
6333 } | |
6334 } | |
6335 } | |
6336 }else{ | |
6337 new_coeff=0; | |
6338 assert(ABS(level)==1); | |
6339 | |
6340 if(i < last_non_zero){ | |
6341 int next_i= i + run2 + 1; | |
6342 int next_level= block[ perm_scantable[next_i] ] + 64; | |
2967 | 6343 |
1777 | 6344 if(next_level&(~127)) |
6345 next_level= 0; | |
6346 | |
6347 if(next_i < last_non_zero) | |
6348 score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)] | |
6349 - length[UNI_AC_ENC_INDEX(run2, next_level)] | |
6350 - length[UNI_AC_ENC_INDEX(run, 65)]; | |
6351 else | |
6352 score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)] | |
6353 - last_length[UNI_AC_ENC_INDEX(run2, next_level)] | |
6354 - length[UNI_AC_ENC_INDEX(run, 65)]; | |
6355 }else{ | |
6356 score += -last_length[UNI_AC_ENC_INDEX(run, 65)]; | |
6357 if(prev_level){ | |
6358 score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)] | |
6359 - length[UNI_AC_ENC_INDEX(prev_run, prev_level)]; | |
6360 } | |
6361 } | |
6362 } | |
2967 | 6363 |
1777 | 6364 score *= lambda; |
6365 | |
6366 unquant_change= new_coeff - old_coeff; | |
6367 assert((score < 100*lambda && score > -100*lambda) || lambda==0); | |
2967 | 6368 |
1784 | 6369 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change); |
1777 | 6370 if(score<best_score){ |
6371 best_score= score; | |
6372 best_coeff= i; | |
6373 best_change= change; | |
6374 best_unquant_change= unquant_change; | |
6375 } | |
6376 } | |
6377 if(level){ | |
6378 prev_level= level + 64; | |
6379 if(prev_level&(~127)) | |
6380 prev_level= 0; | |
6381 prev_run= run; | |
6382 run=0; | |
6383 }else{ | |
6384 run++; | |
6385 } | |
6386 } | |
6387 #ifdef REFINE_STATS | |
6388 STOP_TIMER("iterative step")} | |
6389 #endif | |
6390 | |
6391 if(best_change){ | |
6392 int j= perm_scantable[ best_coeff ]; | |
2967 | 6393 |
1777 | 6394 block[j] += best_change; |
2967 | 6395 |
1777 | 6396 if(best_coeff > last_non_zero){ |
6397 last_non_zero= best_coeff; | |
6398 assert(block[j]); | |
6399 #ifdef REFINE_STATS | |
6400 after_last++; | |
6401 #endif | |
6402 }else{ | |
6403 #ifdef REFINE_STATS | |
6404 if(block[j]){ | |
6405 if(block[j] - best_change){ | |
6406 if(ABS(block[j]) > ABS(block[j] - best_change)){ | |
6407 raise++; | |
6408 }else{ | |
6409 lower++; | |
6410 } | |
6411 }else{ | |
6412 from_zero++; | |
6413 } | |
6414 }else{ | |
6415 to_zero++; | |
6416 } | |
6417 #endif | |
6418 for(; last_non_zero>=start_i; last_non_zero--){ | |
6419 if(block[perm_scantable[last_non_zero]]) | |
6420 break; | |
6421 } | |
6422 } | |
6423 #ifdef REFINE_STATS | |
6424 count++; | |
6425 if(256*256*256*64 % count == 0){ | |
6426 printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number); | |
6427 } | |
6428 #endif | |
6429 run=0; | |
6430 rle_index=0; | |
6431 for(i=start_i; i<=last_non_zero; i++){ | |
6432 int j= perm_scantable[i]; | |
6433 const int level= block[j]; | |
2967 | 6434 |
1777 | 6435 if(level){ |
6436 run_tab[rle_index++]=run; | |
6437 run=0; | |
6438 }else{ | |
6439 run++; | |
6440 } | |
6441 } | |
2967 | 6442 |
1784 | 6443 s->dsp.add_8x8basis(rem, basis[j], best_unquant_change); |
1777 | 6444 }else{ |
6445 break; | |
6446 } | |
6447 } | |
6448 #ifdef REFINE_STATS | |
6449 if(last_non_zero>0){ | |
6450 STOP_TIMER("iterative search") | |
6451 } | |
6452 } | |
6453 #endif | |
6454 | |
6455 return last_non_zero; | |
6456 } | |
6457 | |
2967 | 6458 static int dct_quantize_c(MpegEncContext *s, |
0 | 6459 DCTELEM *block, int n, |
344 | 6460 int qscale, int *overflow) |
0 | 6461 { |
1715 | 6462 int i, j, level, last_non_zero, q, start_i; |
0 | 6463 const int *qmat; |
1064 | 6464 const uint8_t *scantable= s->intra_scantable.scantable; |
344 | 6465 int bias; |
6466 int max=0; | |
6467 unsigned int threshold1, threshold2; | |
828
ace3ccd18dd2
Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents:
815
diff
changeset
|
6468 |
1092 | 6469 s->dsp.fdct (block); |
0 | 6470 |
1597 | 6471 if(s->dct_error_sum) |
1719 | 6472 s->denoise_dct(s, block); |
1597 | 6473 |
0 | 6474 if (s->mb_intra) { |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6475 if (!s->h263_aic) { |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6476 if (n < 4) |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6477 q = s->y_dc_scale; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6478 else |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6479 q = s->c_dc_scale; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6480 q = q << 3; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6481 } else |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
347
diff
changeset
|
6482 /* 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
|
6483 q = 1 << 3; |
2967 | 6484 |
0 | 6485 /* note: block[0] is assumed to be positive */ |
6486 block[0] = (block[0] + (q >> 1)) / q; | |
1715 | 6487 start_i = 1; |
0 | 6488 last_non_zero = 0; |
344 | 6489 qmat = s->q_intra_matrix[qscale]; |
635 | 6490 bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT); |
0 | 6491 } else { |
1715 | 6492 start_i = 0; |
0 | 6493 last_non_zero = -1; |
344 | 6494 qmat = s->q_inter_matrix[qscale]; |
635 | 6495 bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT); |
0 | 6496 } |
635 | 6497 threshold1= (1<<QMAT_SHIFT) - bias - 1; |
6498 threshold2= (threshold1<<1); | |
1715 | 6499 for(i=63;i>=start_i;i--) { |
764 | 6500 j = scantable[i]; |
1715 | 6501 level = block[j] * qmat[j]; |
6502 | |
6503 if(((unsigned)(level+threshold1))>threshold2){ | |
6504 last_non_zero = i; | |
6505 break; | |
6506 }else{ | |
6507 block[j]=0; | |
6508 } | |
6509 } | |
6510 for(i=start_i; i<=last_non_zero; i++) { | |
6511 j = scantable[i]; | |
6512 level = block[j] * qmat[j]; | |
344 | 6513 |
1215 | 6514 // if( bias+level >= (1<<QMAT_SHIFT) |
6515 // || bias-level >= (1<<QMAT_SHIFT)){ | |
344 | 6516 if(((unsigned)(level+threshold1))>threshold2){ |
6517 if(level>0){ | |
635 | 6518 level= (bias + level)>>QMAT_SHIFT; |
344 | 6519 block[j]= level; |
6520 }else{ | |
635 | 6521 level= (bias - level)>>QMAT_SHIFT; |
344 | 6522 block[j]= -level; |
0 | 6523 } |
344 | 6524 max |=level; |
6525 }else{ | |
6526 block[j]=0; | |
0 | 6527 } |
6528 } | |
2627
bf158d23bbcc
Spelling errors patch by (Kevin Baragona | kevinmb500 gawab com)
michael
parents:
2620
diff
changeset
|
6529 *overflow= s->max_qcoeff < max; //overflow might have happened |
2967 | 6530 |
764 | 6531 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */ |
1092 | 6532 if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM) |
2979 | 6533 ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero); |
764 | 6534 |
0 | 6535 return last_non_zero; |
6536 } | |
6537 | |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
6538 #endif //CONFIG_ENCODERS |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
6539 |
2967 | 6540 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s, |
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
|
6541 DCTELEM *block, int n, int qscale) |
0 | 6542 { |
200 | 6543 int i, level, nCoeffs; |
1064 | 6544 const uint16_t *quant_matrix; |
0 | 6545 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
6546 nCoeffs= s->block_last_index[n]; |
2967 | 6547 |
6548 if (n < 4) | |
1689 | 6549 block[0] = block[0] * s->y_dc_scale; |
6550 else | |
6551 block[0] = block[0] * s->c_dc_scale; | |
6552 /* XXX: only mpeg1 */ | |
6553 quant_matrix = s->intra_matrix; | |
6554 for(i=1;i<=nCoeffs;i++) { | |
6555 int j= s->intra_scantable.permutated[i]; | |
6556 level = block[j]; | |
6557 if (level) { | |
6558 if (level < 0) { | |
6559 level = -level; | |
6560 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
6561 level = (level - 1) | 1; | |
6562 level = -level; | |
6563 } else { | |
6564 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
6565 level = (level - 1) | 1; | |
0 | 6566 } |
1689 | 6567 block[j] = level; |
0 | 6568 } |
6569 } | |
6570 } | |
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
|
6571 |
2967 | 6572 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s, |
1689 | 6573 DCTELEM *block, int n, int qscale) |
6574 { | |
6575 int i, level, nCoeffs; | |
6576 const uint16_t *quant_matrix; | |
6577 | |
6578 nCoeffs= s->block_last_index[n]; | |
2967 | 6579 |
1689 | 6580 quant_matrix = s->inter_matrix; |
6581 for(i=0; i<=nCoeffs; i++) { | |
6582 int j= s->intra_scantable.permutated[i]; | |
6583 level = block[j]; | |
6584 if (level) { | |
6585 if (level < 0) { | |
6586 level = -level; | |
6587 level = (((level << 1) + 1) * qscale * | |
6588 ((int) (quant_matrix[j]))) >> 4; | |
6589 level = (level - 1) | 1; | |
6590 level = -level; | |
6591 } else { | |
6592 level = (((level << 1) + 1) * qscale * | |
6593 ((int) (quant_matrix[j]))) >> 4; | |
6594 level = (level - 1) | 1; | |
6595 } | |
6596 block[j] = level; | |
6597 } | |
6598 } | |
6599 } | |
6600 | |
2967 | 6601 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s, |
325 | 6602 DCTELEM *block, int n, int qscale) |
6603 { | |
6604 int i, level, nCoeffs; | |
1064 | 6605 const uint16_t *quant_matrix; |
325 | 6606 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
6607 if(s->alternate_scan) nCoeffs= 63; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
6608 else nCoeffs= s->block_last_index[n]; |
2967 | 6609 |
6610 if (n < 4) | |
1689 | 6611 block[0] = block[0] * s->y_dc_scale; |
6612 else | |
6613 block[0] = block[0] * s->c_dc_scale; | |
6614 quant_matrix = s->intra_matrix; | |
6615 for(i=1;i<=nCoeffs;i++) { | |
6616 int j= s->intra_scantable.permutated[i]; | |
6617 level = block[j]; | |
6618 if (level) { | |
6619 if (level < 0) { | |
6620 level = -level; | |
6621 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
6622 level = -level; | |
6623 } else { | |
6624 level = (int)(level * qscale * quant_matrix[j]) >> 3; | |
6625 } | |
6626 block[j] = level; | |
6627 } | |
6628 } | |
6629 } | |
6630 | |
3281
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6631 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s, |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6632 DCTELEM *block, int n, int qscale) |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6633 { |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6634 int i, level, nCoeffs; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6635 const uint16_t *quant_matrix; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6636 int sum=-1; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6637 |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6638 if(s->alternate_scan) nCoeffs= 63; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6639 else nCoeffs= s->block_last_index[n]; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6640 |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6641 if (n < 4) |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6642 block[0] = block[0] * s->y_dc_scale; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6643 else |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6644 block[0] = block[0] * s->c_dc_scale; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6645 quant_matrix = s->intra_matrix; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6646 for(i=1;i<=nCoeffs;i++) { |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6647 int j= s->intra_scantable.permutated[i]; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6648 level = block[j]; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6649 if (level) { |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6650 if (level < 0) { |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6651 level = -level; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6652 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6653 level = -level; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6654 } else { |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6655 level = (int)(level * qscale * quant_matrix[j]) >> 3; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6656 } |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6657 block[j] = level; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6658 sum+=level; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6659 } |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6660 } |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6661 block[63]^=sum&1; |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6662 } |
7fac25904a8b
missmatch control for mpeg2 intra dequantization if bitexact=1
michael
parents:
3268
diff
changeset
|
6663 |
2967 | 6664 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s, |
1689 | 6665 DCTELEM *block, int n, int qscale) |
6666 { | |
6667 int i, level, nCoeffs; | |
6668 const uint16_t *quant_matrix; | |
6669 int sum=-1; | |
6670 | |
6671 if(s->alternate_scan) nCoeffs= 63; | |
6672 else nCoeffs= s->block_last_index[n]; | |
2967 | 6673 |
1689 | 6674 quant_matrix = s->inter_matrix; |
6675 for(i=0; i<=nCoeffs; i++) { | |
6676 int j= s->intra_scantable.permutated[i]; | |
6677 level = block[j]; | |
6678 if (level) { | |
6679 if (level < 0) { | |
6680 level = -level; | |
6681 level = (((level << 1) + 1) * qscale * | |
6682 ((int) (quant_matrix[j]))) >> 4; | |
6683 level = -level; | |
6684 } else { | |
6685 level = (((level << 1) + 1) * qscale * | |
6686 ((int) (quant_matrix[j]))) >> 4; | |
6687 } | |
6688 block[j] = level; | |
6689 sum+=level; | |
6690 } | |
6691 } | |
6692 block[63]^=sum&1; | |
6693 } | |
6694 | |
2967 | 6695 static void dct_unquantize_h263_intra_c(MpegEncContext *s, |
1689 | 6696 DCTELEM *block, int n, int qscale) |
6697 { | |
6698 int i, level, qmul, qadd; | |
6699 int nCoeffs; | |
2967 | 6700 |
1689 | 6701 assert(s->block_last_index[n]>=0); |
2967 | 6702 |
1689 | 6703 qmul = qscale << 1; |
2967 | 6704 |
1689 | 6705 if (!s->h263_aic) { |
2967 | 6706 if (n < 4) |
325 | 6707 block[0] = block[0] * s->y_dc_scale; |
6708 else | |
6709 block[0] = block[0] * s->c_dc_scale; | |
1689 | 6710 qadd = (qscale - 1) | 1; |
6711 }else{ | |
6712 qadd = 0; | |
6713 } | |
6714 if(s->ac_pred) | |
6715 nCoeffs=63; | |
6716 else | |
6717 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ]; | |
6718 | |
6719 for(i=1; i<=nCoeffs; i++) { | |
6720 level = block[i]; | |
6721 if (level) { | |
6722 if (level < 0) { | |
6723 level = level * qmul - qadd; | |
6724 } else { | |
6725 level = level * qmul + qadd; | |
325 | 6726 } |
1689 | 6727 block[i] = level; |
325 | 6728 } |
6729 } | |
6730 } | |
6731 | |
2967 | 6732 static void dct_unquantize_h263_inter_c(MpegEncContext *s, |
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
|
6733 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
|
6734 { |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
6735 int i, level, qmul, qadd; |
200 | 6736 int nCoeffs; |
2967 | 6737 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
6738 assert(s->block_last_index[n]>=0); |
2967 | 6739 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
6740 qadd = (qscale - 1) | 1; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
704
diff
changeset
|
6741 qmul = qscale << 1; |
2967 | 6742 |
1689 | 6743 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ]; |
6744 | |
6745 for(i=0; 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
|
6746 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
|
6747 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
|
6748 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
|
6749 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
|
6750 } 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
|
6751 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
|
6752 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
6753 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
|
6754 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
6755 } |
174ef88f619a
use block[] in structure to have it aligned on 8 bytes for mmx optimizations - dct_unquantize is always a function pointer - added specialized dct_unquantize_h263
glantau
parents:
8
diff
changeset
|
6756 } |
0 | 6757 |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
6758 #ifdef CONFIG_ENCODERS |
0 | 6759 AVCodec h263_encoder = { |
6760 "h263", | |
6761 CODEC_TYPE_VIDEO, | |
6762 CODEC_ID_H263, | |
6763 sizeof(MpegEncContext), | |
6764 MPV_encode_init, | |
6765 MPV_encode_picture, | |
6766 MPV_encode_end, | |
2524 | 6767 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
0 | 6768 }; |
6769 | |
6770 AVCodec h263p_encoder = { | |
6771 "h263p", | |
6772 CODEC_TYPE_VIDEO, | |
6773 CODEC_ID_H263P, | |
6774 sizeof(MpegEncContext), | |
6775 MPV_encode_init, | |
6776 MPV_encode_picture, | |
6777 MPV_encode_end, | |
2524 | 6778 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
0 | 6779 }; |
6780 | |
1353
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6781 AVCodec flv_encoder = { |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6782 "flv", |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6783 CODEC_TYPE_VIDEO, |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6784 CODEC_ID_FLV1, |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6785 sizeof(MpegEncContext), |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6786 MPV_encode_init, |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6787 MPV_encode_picture, |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6788 MPV_encode_end, |
2524 | 6789 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
1353
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6790 }; |
cfc80b3a4ada
flash video (flv) support patch by (Garrick Meeker <gmeeker at theoryllc dot com>)
michaelni
parents:
1344
diff
changeset
|
6791 |
0 | 6792 AVCodec rv10_encoder = { |
6793 "rv10", | |
6794 CODEC_TYPE_VIDEO, | |
6795 CODEC_ID_RV10, | |
6796 sizeof(MpegEncContext), | |
6797 MPV_encode_init, | |
6798 MPV_encode_picture, | |
6799 MPV_encode_end, | |
2524 | 6800 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
0 | 6801 }; |
6802 | |
2380 | 6803 AVCodec rv20_encoder = { |
6804 "rv20", | |
6805 CODEC_TYPE_VIDEO, | |
6806 CODEC_ID_RV20, | |
6807 sizeof(MpegEncContext), | |
6808 MPV_encode_init, | |
6809 MPV_encode_picture, | |
6810 MPV_encode_end, | |
2524 | 6811 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
2380 | 6812 }; |
6813 | |
71 | 6814 AVCodec mpeg4_encoder = { |
6815 "mpeg4", | |
0 | 6816 CODEC_TYPE_VIDEO, |
71 | 6817 CODEC_ID_MPEG4, |
0 | 6818 sizeof(MpegEncContext), |
6819 MPV_encode_init, | |
6820 MPV_encode_picture, | |
6821 MPV_encode_end, | |
2524 | 6822 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
2111 | 6823 .capabilities= CODEC_CAP_DELAY, |
0 | 6824 }; |
6825 | |
307 | 6826 AVCodec msmpeg4v1_encoder = { |
6827 "msmpeg4v1", | |
0 | 6828 CODEC_TYPE_VIDEO, |
307 | 6829 CODEC_ID_MSMPEG4V1, |
0 | 6830 sizeof(MpegEncContext), |
6831 MPV_encode_init, | |
6832 MPV_encode_picture, | |
6833 MPV_encode_end, | |
2524 | 6834 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
0 | 6835 }; |
307 | 6836 |
6837 AVCodec msmpeg4v2_encoder = { | |
6838 "msmpeg4v2", | |
6839 CODEC_TYPE_VIDEO, | |
6840 CODEC_ID_MSMPEG4V2, | |
6841 sizeof(MpegEncContext), | |
6842 MPV_encode_init, | |
6843 MPV_encode_picture, | |
6844 MPV_encode_end, | |
2524 | 6845 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
307 | 6846 }; |
6847 | |
6848 AVCodec msmpeg4v3_encoder = { | |
6849 "msmpeg4", | |
6850 CODEC_TYPE_VIDEO, | |
6851 CODEC_ID_MSMPEG4V3, | |
6852 sizeof(MpegEncContext), | |
6853 MPV_encode_init, | |
6854 MPV_encode_picture, | |
6855 MPV_encode_end, | |
2524 | 6856 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
307 | 6857 }; |
499 | 6858 |
6859 AVCodec wmv1_encoder = { | |
6860 "wmv1", | |
6861 CODEC_TYPE_VIDEO, | |
6862 CODEC_ID_WMV1, | |
6863 sizeof(MpegEncContext), | |
6864 MPV_encode_init, | |
6865 MPV_encode_picture, | |
6866 MPV_encode_end, | |
2524 | 6867 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1}, |
499 | 6868 }; |
6869 | |
1042 | 6870 AVCodec mjpeg_encoder = { |
6871 "mjpeg", | |
6872 CODEC_TYPE_VIDEO, | |
6873 CODEC_ID_MJPEG, | |
6874 sizeof(MpegEncContext), | |
6875 MPV_encode_init, | |
6876 MPV_encode_picture, | |
6877 MPV_encode_end, | |
3929 | 6878 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUVJ420P, PIX_FMT_YUVJ422P, -1}, |
1042 | 6879 }; |
1070
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
6880 |
6da5ae9ee199
more #ifdef CONFIG_ENCODERS patch by (Wolfgang Hesseler <qv at multimediaware dot com>) with modifications by me (s/WOLFGANG/CONFIG_ENCODERS/ and some other fixes)
michaelni
parents:
1064
diff
changeset
|
6881 #endif //CONFIG_ENCODERS |