Mercurial > libavcodec.hg
annotate dsputil.h @ 731:f011a0df269f libavcodec
set low_delay on encoding if no b frames
author | michaelni |
---|---|
date | Sun, 06 Oct 2002 17:46:22 +0000 |
parents | e65798d228ea |
children | 8e1f0939d15d |
rev | line source |
---|---|
429 | 1 /* |
2 * DSP utils | |
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard. | |
4 * | |
5 * This library is free software; you can redistribute it and/or | |
6 * modify it under the terms of the GNU Lesser General Public | |
7 * License as published by the Free Software Foundation; either | |
8 * version 2 of the License, or (at your option) any later version. | |
9 * | |
10 * This library is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 * Lesser General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU Lesser General Public | |
16 * License along with this library; if not, write to the Free Software | |
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 */ | |
0 | 19 #ifndef DSPUTIL_H |
20 #define DSPUTIL_H | |
21 | |
22 #include "common.h" | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
23 #include "avcodec.h" |
0 | 24 |
255 | 25 //#define DEBUG |
0 | 26 /* dct code */ |
27 typedef short DCTELEM; | |
28 | |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
29 void fdct_ifast (DCTELEM *data); |
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:
623
diff
changeset
|
30 void ff_jpeg_fdct_islow (DCTELEM *data); |
0 | 31 |
32 void j_rev_dct (DCTELEM *data); | |
33 | |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
34 void ff_fdct_mmx(DCTELEM *block); |
0 | 35 |
34 | 36 /* encoding scans */ |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
37 extern const UINT8 ff_alternate_horizontal_scan[64]; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
38 extern const UINT8 ff_alternate_vertical_scan[64]; |
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
39 extern const UINT8 ff_zigzag_direct[64]; |
190
9e0e56869d05
fix for non-mmx runtimedetect encoding bugs - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
174
diff
changeset
|
40 |
0 | 41 /* pixel operations */ |
42 #define MAX_NEG_CROP 384 | |
43 | |
44 /* temporary */ | |
45 extern UINT32 squareTbl[512]; | |
50 | 46 extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP]; |
0 | 47 |
48 void dsputil_init(void); | |
49 | |
675 | 50 /* minimum alignment rules ;) |
51 if u notice errors in the align stuff, need more alignment for some asm code for some cpu | |
52 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... | |
53 | |
54 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
55 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
56 | |
57 !future video codecs might need functions with less strict alignment | |
58 */ | |
0 | 59 |
675 | 60 /* pixel ops : interface with DCT */ |
61 extern void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size); | |
62 extern void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride); | |
63 extern void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size); | |
64 extern void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size); | |
65 extern void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); | |
66 extern void (*clear_blocks)(DCTELEM *blocks/*align 16*/); | |
612 | 67 extern int (*pix_sum)(UINT8 * pix, int line_size); |
68 extern int (*pix_norm1)(UINT8 * pix, int line_size); | |
69 | |
255 | 70 |
0 | 71 |
72 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size); | |
324 | 73 void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride); |
0 | 74 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size); |
75 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size); | |
296 | 76 void clear_blocks_c(DCTELEM *blocks); |
0 | 77 |
78 /* add and put pixel (decoding) */ | |
675 | 79 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
80 typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h); | |
81 typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride); | |
0 | 82 |
651 | 83 extern op_pixels_func put_pixels_tab[2][4]; |
84 extern op_pixels_func avg_pixels_tab[2][4]; | |
85 extern op_pixels_func put_no_rnd_pixels_tab[2][4]; | |
86 extern op_pixels_func avg_no_rnd_pixels_tab[2][4]; | |
87 extern qpel_mc_func put_qpel_pixels_tab[2][16]; | |
88 extern qpel_mc_func avg_qpel_pixels_tab[2][16]; | |
89 extern qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; | |
90 extern qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; | |
91 | |
92 #define CALL_2X_PIXELS(a, b, n)\ | |
93 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
94 b(block , pixels , line_size, h);\ | |
95 b(block+n, pixels+n, line_size, h);\ | |
96 } | |
255 | 97 |
0 | 98 /* motion estimation */ |
99 | |
675 | 100 typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size); |
0 | 101 |
102 extern op_pixels_abs_func pix_abs16x16; | |
103 extern op_pixels_abs_func pix_abs16x16_x2; | |
104 extern op_pixels_abs_func pix_abs16x16_y2; | |
105 extern op_pixels_abs_func pix_abs16x16_xy2; | |
294 | 106 extern op_pixels_abs_func pix_abs8x8; |
107 extern op_pixels_abs_func pix_abs8x8_x2; | |
108 extern op_pixels_abs_func pix_abs8x8_y2; | |
109 extern op_pixels_abs_func pix_abs8x8_xy2; | |
0 | 110 |
294 | 111 int pix_abs16x16_c(UINT8 *blk1, UINT8 *blk2, int lx); |
112 int pix_abs16x16_x2_c(UINT8 *blk1, UINT8 *blk2, int lx); | |
113 int pix_abs16x16_y2_c(UINT8 *blk1, UINT8 *blk2, int lx); | |
114 int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx); | |
0 | 115 |
706
e65798d228ea
idct permutation cleanup, idct can be selected per context now
michaelni
parents:
689
diff
changeset
|
116 void block_permute(INT16 *block, UINT8 *permutation); |
34 | 117 |
62 | 118 #if defined(HAVE_MMX) |
0 | 119 |
120 #define MM_MMX 0x0001 /* standard MMX */ | |
121 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
122 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
123 #define MM_SSE 0x0008 /* SSE functions */ | |
124 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
125 | |
126 extern int mm_flags; | |
127 | |
128 int mm_support(void); | |
129 | |
130 static inline void emms(void) | |
131 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
132 __asm __volatile ("emms;":::"memory"); |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
133 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
134 |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
135 #define emms_c() \ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
136 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
137 if (mm_flags & MM_MMX)\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
138 emms();\ |
0 | 139 } |
140 | |
141 #define __align8 __attribute__ ((aligned (8))) | |
142 | |
143 void dsputil_init_mmx(void); | |
403
2c3e25f4c496
removed unused stuff - added dsputil_set_bit_exact() support for easier testing
glantau
parents:
324
diff
changeset
|
144 void dsputil_set_bit_exact_mmx(void); |
0 | 145 |
62 | 146 #elif defined(ARCH_ARMV4L) |
147 | |
148 #define emms_c() | |
149 | |
150 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
151 line ptimizations */ | |
152 #define __align8 __attribute__ ((aligned (4))) | |
153 | |
154 void dsputil_init_armv4l(void); | |
155 | |
88 | 156 #elif defined(HAVE_MLIB) |
157 | |
158 #define emms_c() | |
159 | |
160 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
161 #define __align8 __attribute__ ((aligned (8))) | |
162 | |
163 void dsputil_init_mlib(void); | |
164 | |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
165 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
166 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
167 #define emms_c() |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
168 #define __align8 __attribute__ ((aligned (8))) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
169 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
170 void dsputil_init_alpha(void); |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
171 |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
172 #elif defined(ARCH_POWERPC) |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
173 |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
174 #define emms_c() |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
175 #define __align8 __attribute__ ((aligned (16))) |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
176 |
638
0012f75c92bb
altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
626
diff
changeset
|
177 void dsputil_init_ppc(void); |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
178 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
179 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
180 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
181 #define emms_c() |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
182 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
183 #define __align8 __attribute__ ((aligned (16))) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
184 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
185 void dsputil_init_mmi(void); |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
186 |
0 | 187 #else |
188 | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
189 #define emms_c() |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
190 |
0 | 191 #define __align8 |
192 | |
193 #endif | |
194 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
195 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
196 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
197 struct unaligned_64 { uint64_t l; } __attribute__((packed)); |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
198 struct unaligned_32 { uint32_t l; } __attribute__((packed)); |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
199 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
200 #define LD32(a) (((const struct unaligned_32 *) (a))->l) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
201 #define LD64(a) (((const struct unaligned_64 *) (a))->l) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
202 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
203 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
204 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
205 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
206 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
207 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
208 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
209 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
210 #define ST32(a, b) *((uint32_t*)(a)) = (b) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
211 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
212 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
213 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
214 /* PSNR */ |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
215 void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3], |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
216 int orig_linesize[3], int coded_linesize, |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
217 AVCodecContext *avctx); |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
218 |
0 | 219 #endif |