Mercurial > libavcodec.hg
annotate dsputil.h @ 987:295ec9bebf93 libavcodec
new image formats - suppressed ill defined formats (endianness, unless specified, is always the one of the CPU) - added avcodec_get_pix_fmt_name()
author | bellard |
---|---|
date | Sat, 11 Jan 2003 00:04:36 +0000 |
parents | e162c09efbe7 |
children | edc10966b081 |
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 | |
675 | 49 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
50 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 51 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
52 | |
53 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
54 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
55 | |
56 !future video codecs might need functions with less strict alignment | |
57 */ | |
0 | 58 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
59 /* |
0 | 60 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size); |
324 | 61 void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride); |
0 | 62 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size); |
63 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size); | |
296 | 64 void clear_blocks_c(DCTELEM *blocks); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
65 */ |
0 | 66 |
67 /* add and put pixel (decoding) */ | |
675 | 68 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
69 typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h); | |
70 typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride); | |
0 | 71 |
984 | 72 #define DEF_OLD_QPEL(name)\ |
73 void ff_put_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\ | |
74 void ff_put_no_rnd_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\ | |
75 void ff_avg_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
76 |
984 | 77 DEF_OLD_QPEL(qpel16_mc11_old_c) |
78 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
79 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
80 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
81 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
82 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
83 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
84 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
85 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
86 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
87 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
88 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 89 |
90 #define CALL_2X_PIXELS(a, b, n)\ | |
91 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
92 b(block , pixels , line_size, h);\ | |
93 b(block+n, pixels+n, line_size, h);\ | |
94 } | |
255 | 95 |
0 | 96 /* motion estimation */ |
97 | |
936 | 98 typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/; |
99 | |
100 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/; | |
101 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
102 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
103 /* pixel ops : interface with DCT */ |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
104 void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
105 void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
106 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
107 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
108 void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
109 void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy, |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
110 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
111 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
112 int (*pix_sum)(UINT8 * pix, int line_size); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
113 int (*pix_norm1)(UINT8 * pix, int line_size); |
936 | 114 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */ |
115 me_cmp_func sse[2]; | |
116 me_cmp_func hadamard8_diff[2]; | |
117 me_cmp_func dct_sad[2]; | |
118 me_cmp_func quant_psnr[2]; | |
119 int (*hadamard8_abs )(uint8_t *src, int stride, int mean); | |
120 | |
954 | 121 me_cmp_func me_pre_cmp[11]; |
936 | 122 me_cmp_func me_cmp[11]; |
123 me_cmp_func me_sub_cmp[11]; | |
124 me_cmp_func mb_cmp[11]; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
125 |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
126 /* maybe create an array for 16/8 functions */ |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
127 op_pixels_func put_pixels_tab[2][4]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
128 op_pixels_func avg_pixels_tab[2][4]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
129 op_pixels_func put_no_rnd_pixels_tab[2][4]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
130 op_pixels_func avg_no_rnd_pixels_tab[2][4]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
131 qpel_mc_func put_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
132 qpel_mc_func avg_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
133 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
134 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 135 qpel_mc_func put_mspel_pixels_tab[8]; |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
136 |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
137 op_pixels_abs_func pix_abs16x16; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
138 op_pixels_abs_func pix_abs16x16_x2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
139 op_pixels_abs_func pix_abs16x16_y2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
140 op_pixels_abs_func pix_abs16x16_xy2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
141 op_pixels_abs_func pix_abs8x8; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
142 op_pixels_abs_func pix_abs8x8_x2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
143 op_pixels_abs_func pix_abs8x8_y2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
144 op_pixels_abs_func pix_abs8x8_xy2; |
866 | 145 |
146 /* huffyuv specific */ | |
147 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 148 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
149 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
150 |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
151 void dsputil_init(DSPContext* p, unsigned mask); |
0 | 152 |
764 | 153 /** |
154 * permute block according to permuatation. | |
155 * @param last last non zero element in scantable order | |
156 */ | |
157 void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last); | |
34 | 158 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
159 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
160 |
62 | 161 #if defined(HAVE_MMX) |
0 | 162 |
862 | 163 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
164 |
0 | 165 #define MM_MMX 0x0001 /* standard MMX */ |
166 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
167 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
168 #define MM_SSE 0x0008 /* SSE functions */ | |
169 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
170 | |
171 extern int mm_flags; | |
172 | |
173 int mm_support(void); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
174 void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
175 void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size); |
0 | 176 |
177 static inline void emms(void) | |
178 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
179 __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
|
180 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
181 |
936 | 182 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
183 #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
|
184 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
185 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
|
186 emms();\ |
0 | 187 } |
188 | |
189 #define __align8 __attribute__ ((aligned (8))) | |
190 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
191 void dsputil_init_mmx(DSPContext* c, unsigned mask); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
192 void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask); |
0 | 193 |
62 | 194 #elif defined(ARCH_ARMV4L) |
195 | |
196 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
197 line ptimizations */ | |
198 #define __align8 __attribute__ ((aligned (4))) | |
199 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
200 void dsputil_init_armv4l(DSPContext* c, unsigned mask); |
62 | 201 |
88 | 202 #elif defined(HAVE_MLIB) |
203 | |
204 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
205 #define __align8 __attribute__ ((aligned (8))) | |
206 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
207 void dsputil_init_mlib(DSPContext* c, unsigned mask); |
88 | 208 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
209 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
210 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
211 #define __align8 __attribute__ ((aligned (8))) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
212 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
213 void dsputil_init_alpha(DSPContext* c, unsigned mask); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
214 |
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
|
215 #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
|
216 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
217 #define MM_ALTIVEC 0x0001 /* standard AltiVec */ |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
218 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
219 extern int mm_flags; |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
220 |
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
|
221 #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
|
222 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
223 void dsputil_init_ppc(DSPContext* c, unsigned mask); |
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
|
224 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
225 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
226 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
227 #define __align8 __attribute__ ((aligned (16))) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
228 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
229 void dsputil_init_mmi(DSPContext* c, unsigned mask); |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
230 |
0 | 231 #else |
232 | |
233 #define __align8 | |
234 | |
235 #endif | |
236 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
237 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
238 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
239 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
|
240 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
|
241 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
242 #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
|
243 #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
|
244 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
245 #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
|
246 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
247 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
248 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
249 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
250 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
251 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
252 #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
|
253 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
254 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
255 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
256 /* PSNR */ |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
257 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
|
258 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
|
259 AVCodecContext *avctx); |
781 | 260 |
261 /* FFT computation */ | |
262 | |
263 /* NOTE: soon integer code will be added, so you must use the | |
264 FFTSample type */ | |
265 typedef float FFTSample; | |
266 | |
267 typedef struct FFTComplex { | |
268 FFTSample re, im; | |
269 } FFTComplex; | |
270 | |
271 typedef struct FFTContext { | |
272 int nbits; | |
273 int inverse; | |
274 uint16_t *revtab; | |
275 FFTComplex *exptab; | |
276 FFTComplex *exptab1; /* only used by SSE code */ | |
277 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
278 } FFTContext; | |
279 | |
280 int fft_init(FFTContext *s, int nbits, int inverse); | |
281 void fft_permute(FFTContext *s, FFTComplex *z); | |
282 void fft_calc_c(FFTContext *s, FFTComplex *z); | |
283 void fft_calc_sse(FFTContext *s, FFTComplex *z); | |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
954
diff
changeset
|
284 void fft_calc_altivec(FFTContext *s, FFTComplex *z); |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
954
diff
changeset
|
285 |
781 | 286 static inline void fft_calc(FFTContext *s, FFTComplex *z) |
287 { | |
288 s->fft_calc(s, z); | |
289 } | |
290 void fft_end(FFTContext *s); | |
291 | |
292 /* MDCT computation */ | |
293 | |
294 typedef struct MDCTContext { | |
295 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
296 int nbits; /* n = 2^nbits */ | |
297 /* pre/post rotation tables */ | |
298 FFTSample *tcos; | |
299 FFTSample *tsin; | |
300 FFTContext fft; | |
301 } MDCTContext; | |
302 | |
794 | 303 int ff_mdct_init(MDCTContext *s, int nbits, int inverse); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
304 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 305 const FFTSample *input, FFTSample *tmp); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
306 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 307 const FFTSample *input, FFTSample *tmp); |
794 | 308 void ff_mdct_end(MDCTContext *s); |
781 | 309 |
936 | 310 #define WARPER88_1616(name8, name16)\ |
311 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\ | |
312 return name8(s, dst , src , stride)\ | |
313 +name8(s, dst+8 , src+8 , stride)\ | |
314 +name8(s, dst +8*stride, src +8*stride, stride)\ | |
315 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\ | |
316 } | |
317 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
318 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
319 /* XXX: add ISOC specific test to avoid specific BSD testing. */ |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
320 /* better than nothing implementation. */ |
802 | 321 /* btw, rintf() is existing on fbsd too -- alex */ |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
322 static inline long int lrintf(float x) |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
323 { |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
324 return (int)(rint(x)); |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
325 } |
0 | 326 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
327 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
328 #endif |