Mercurial > libavcodec.hg
annotate dsputil.h @ 3110:b090207c3d18 libavcodec
Separate header file installation from the general install target.
author | diego |
---|---|
date | Sat, 11 Feb 2006 17:44:06 +0000 |
parents | 2d35fb3cb940 |
children | c20c181e0eca |
rev | line source |
---|---|
429 | 1 /* |
2 * DSP utils | |
3 * Copyright (c) 2000, 2001, 2002 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> |
429 | 5 * |
6 * This library is free software; you can redistribute it and/or | |
7 * modify it under the terms of the GNU Lesser General Public | |
8 * License as published by the Free Software Foundation; either | |
9 * version 2 of the License, or (at your option) any later version. | |
10 * | |
11 * This library is distributed in the hope that it will be useful, | |
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 * Lesser General Public License for more details. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
17 * License along with this library; if not, write to the Free Software | |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
429 | 19 */ |
1102 | 20 |
21 /** | |
22 * @file dsputil.h | |
1106 | 23 * DSP utils. |
1213 | 24 * note, many functions in here may use MMX which trashes the FPU state, it is |
25 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 26 */ |
27 | |
0 | 28 #ifndef DSPUTIL_H |
29 #define DSPUTIL_H | |
30 | |
31 #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
|
32 #include "avcodec.h" |
0 | 33 |
1102 | 34 |
255 | 35 //#define DEBUG |
0 | 36 /* dct code */ |
37 typedef short DCTELEM; | |
38 | |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
39 void fdct_ifast (DCTELEM *data); |
1571
aa4dc16c0f18
* adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents:
1567
diff
changeset
|
40 void fdct_ifast248 (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
|
41 void ff_jpeg_fdct_islow (DCTELEM *data); |
1567 | 42 void ff_fdct248_islow (DCTELEM *data); |
0 | 43 |
44 void j_rev_dct (DCTELEM *data); | |
2256 | 45 void j_rev_dct4 (DCTELEM *data); |
2257 | 46 void j_rev_dct2 (DCTELEM *data); |
2259 | 47 void j_rev_dct1 (DCTELEM *data); |
0 | 48 |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
49 void ff_fdct_mmx(DCTELEM *block); |
1565 | 50 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
51 void ff_fdct_sse2(DCTELEM *block); |
0 | 52 |
2755 | 53 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
54 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
55 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
56 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
57 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
58 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); |
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
59 |
34 | 60 /* encoding scans */ |
1064 | 61 extern const uint8_t ff_alternate_horizontal_scan[64]; |
62 extern const uint8_t ff_alternate_vertical_scan[64]; | |
63 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 64 extern const uint8_t ff_zigzag248_direct[64]; |
190
9e0e56869d05
fix for non-mmx runtimedetect encoding bugs - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
174
diff
changeset
|
65 |
0 | 66 /* pixel operations */ |
2090 | 67 #define MAX_NEG_CROP 1024 |
0 | 68 |
69 /* temporary */ | |
1064 | 70 extern uint32_t squareTbl[512]; |
71 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 72 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
73 /* VP3 DSP functions */ |
2693 | 74 void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
75 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
76 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); | |
0 | 77 |
675 | 78 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
79 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 80 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
81 | |
82 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
83 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
84 | |
85 !future video codecs might need functions with less strict alignment | |
86 */ | |
0 | 87 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
88 /* |
1064 | 89 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
90 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
91 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
92 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 93 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
|
94 */ |
0 | 95 |
96 /* add and put pixel (decoding) */ | |
675 | 97 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 98 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 99 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h); |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
100 typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h); |
1064 | 101 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 102 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
2415 | 103 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset); |
3029 | 104 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset); |
0 | 105 |
984 | 106 #define DEF_OLD_QPEL(name)\ |
1064 | 107 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
108 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
109 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *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
|
110 |
984 | 111 DEF_OLD_QPEL(qpel16_mc11_old_c) |
112 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
113 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
114 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
115 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
116 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
117 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
118 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
119 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
120 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
121 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
122 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 123 |
124 #define CALL_2X_PIXELS(a, b, n)\ | |
125 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
126 b(block , pixels , line_size, h);\ | |
127 b(block+n, pixels+n, line_size, h);\ | |
128 } | |
255 | 129 |
0 | 130 /* motion estimation */ |
1709 | 131 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
132 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 133 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/; |
936 | 134 |
1168 | 135 |
1102 | 136 /** |
137 * DSPContext. | |
138 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
139 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
140 /* pixel ops : interface with DCT */ |
1064 | 141 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
142 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
143 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
144 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 145 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
2763 | 146 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
147 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); | |
1136 | 148 /** |
149 * translational global motion compensation. | |
150 */ | |
1064 | 151 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 152 /** |
153 * global motion compensation. | |
154 */ | |
1064 | 155 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
2979 | 156 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
157 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 158 int (*pix_sum)(uint8_t * pix, int line_size); |
159 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 160 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
2967 | 161 |
1729 | 162 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
163 me_cmp_func sse[5]; | |
164 me_cmp_func hadamard8_diff[5]; | |
165 me_cmp_func dct_sad[5]; | |
166 me_cmp_func quant_psnr[5]; | |
167 me_cmp_func bit[5]; | |
168 me_cmp_func rd[5]; | |
169 me_cmp_func vsad[5]; | |
170 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
171 me_cmp_func nsse[5]; |
2184 | 172 me_cmp_func w53[5]; |
173 me_cmp_func w97[5]; | |
2382 | 174 me_cmp_func dct_max[5]; |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
175 me_cmp_func dct264_sad[5]; |
936 | 176 |
1708 | 177 me_cmp_func me_pre_cmp[5]; |
178 me_cmp_func me_cmp[5]; | |
179 me_cmp_func me_sub_cmp[5]; | |
180 me_cmp_func mb_cmp[5]; | |
1729 | 181 me_cmp_func ildct_cmp[5]; //only width 16 used |
2382 | 182 me_cmp_func frame_skip_cmp[5]; //only width 8 used |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
183 |
1136 | 184 /** |
185 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 186 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 187 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 188 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 189 * @param block destination where the result is stored |
190 * @param pixels source | |
191 * @param line_size number of bytes in a horizontal line of block | |
192 * @param h height | |
193 */ | |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
194 op_pixels_func put_pixels_tab[4][4]; |
1136 | 195 |
196 /** | |
197 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
2967 | 198 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 199 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 200 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 201 * @param block destination into which the result is averaged (a+b+1)>>1 |
202 * @param pixels source | |
203 * @param line_size number of bytes in a horizontal line of block | |
204 * @param h height | |
205 */ | |
1319 | 206 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 207 |
208 /** | |
209 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 210 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 211 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 212 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 213 * @param block destination where the result is stored |
214 * @param pixels source | |
215 * @param line_size number of bytes in a horizontal line of block | |
216 * @param h height | |
217 */ | |
2075 | 218 op_pixels_func put_no_rnd_pixels_tab[4][4]; |
1136 | 219 |
220 /** | |
221 * Halfpel motion compensation with no rounding (a+b)>>1. | |
2967 | 222 * this is an array[2][4] of motion compensation funcions for 2 |
1225 | 223 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 224 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 225 * @param block destination into which the result is averaged (a+b)>>1 |
226 * @param pixels source | |
227 * @param line_size number of bytes in a horizontal line of block | |
228 * @param h height | |
229 */ | |
2075 | 230 op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
2967 | 231 |
1864 | 232 void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h); |
2967 | 233 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
234 /** |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
235 * Thirdpel motion compensation with rounding (a+b+1)>>1. |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
236 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br> |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
237 * *pixels_tab[ xthirdpel + 4*ythirdpel ] |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
238 * @param block destination where the result is stored |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
239 * @param pixels source |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
240 * @param line_size number of bytes in a horizontal line of block |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
241 * @param h height |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
242 */ |
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
243 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 244 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
245 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
246 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
|
247 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
|
248 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
|
249 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 250 qpel_mc_func put_mspel_pixels_tab[8]; |
2967 | 251 |
1168 | 252 /** |
253 * h264 Chram MC | |
254 */ | |
255 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
256 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
257 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
258 qpel_mc_func put_h264_qpel_pixels_tab[4][16]; |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
259 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
2967 | 260 |
2415 | 261 h264_weight_func weight_h264_pixels_tab[10]; |
262 h264_biweight_func biweight_h264_pixels_tab[10]; | |
2967 | 263 |
1708 | 264 me_cmp_func pix_abs[2][4]; |
2967 | 265 |
866 | 266 /* huffyuv specific */ |
267 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 268 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 269 /** |
270 * subtract huffyuv's variant of median prediction | |
271 * note, this might read from src1[-1], src2[-1] | |
272 */ | |
273 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 274 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
2633 | 275 |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
276 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
277 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
278 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
279 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
280 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2696
diff
changeset
|
281 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); |
2967 | 282 |
1644 | 283 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
284 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
285 | |
2045 | 286 void (*h261_loop_filter)(uint8_t *src, int stride); |
2044
b6f2add2511e
h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
michael
parents:
2024
diff
changeset
|
287 |
1092 | 288 /* (I)DCT */ |
289 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 290 void (*fdct248)(DCTELEM *block/* align 16*/); |
2967 | 291 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
292 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
293 void (*idct)(DCTELEM *block/* align 16*/); |
2967 | 294 |
1102 | 295 /** |
1104 | 296 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 297 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 298 * @param line_size size in bytes of a horizotal line of dest |
1102 | 299 */ |
1092 | 300 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 301 |
1102 | 302 /** |
303 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 304 * @param line_size size in bytes of a horizotal line of dest |
1102 | 305 */ |
1092 | 306 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
2967 | 307 |
1102 | 308 /** |
1104 | 309 * idct input permutation. |
1241 | 310 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
311 * IDCT) | |
312 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
313 * with the zigzag/alternate scan<br> | |
1102 | 314 * an example to avoid confusion: |
315 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
316 * - (x -> referece dct -> reference idct -> x) | |
317 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
318 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
319 */ | |
1092 | 320 uint8_t idct_permutation[64]; |
321 int idct_permutation_type; | |
322 #define FF_NO_IDCT_PERM 1 | |
323 #define FF_LIBMPEG2_IDCT_PERM 2 | |
324 #define FF_SIMPLE_IDCT_PERM 3 | |
325 #define FF_TRANSPOSE_IDCT_PERM 4 | |
2696
9699d325049d
porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents:
2693
diff
changeset
|
326 #define FF_PARTTRANS_IDCT_PERM 5 |
1092 | 327 |
1784 | 328 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
329 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
330 #define BASIS_SHIFT 16 | |
331 #define RECON_SHIFT 6 | |
2967 | 332 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2259
diff
changeset
|
333 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); |
2755 | 334 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
335 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3089
diff
changeset
|
336 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
337 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
338 |
1201 | 339 void dsputil_static_init(void); |
1092 | 340 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 341 |
764 | 342 /** |
343 * permute block according to permuatation. | |
344 * @param last last non zero element in scantable order | |
345 */ | |
1064 | 346 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 347 |
1729 | 348 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
349 | |
2979 | 350 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
1264 | 351 |
352 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
353 { | |
354 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
355 } | |
356 | |
357 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
358 { | |
359 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
360 } | |
361 | |
2184 | 362 static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
363 switch(type&0xFF){ | |
364 default: | |
365 case FF_CMP_SAD: | |
366 return lambda>>FF_LAMBDA_SHIFT; | |
367 case FF_CMP_DCT: | |
368 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); | |
369 case FF_CMP_W53: | |
370 return (4*lambda)>>(FF_LAMBDA_SHIFT); | |
371 case FF_CMP_W97: | |
372 return (2*lambda)>>(FF_LAMBDA_SHIFT); | |
373 case FF_CMP_SATD: | |
3010
533c6386eca9
8x8 integer dct from x264 as cmp function (under CONFIG_GPL)
michael
parents:
2979
diff
changeset
|
374 case FF_CMP_DCT264: |
2184 | 375 return (2*lambda)>>FF_LAMBDA_SHIFT; |
376 case FF_CMP_RD: | |
377 case FF_CMP_PSNR: | |
378 case FF_CMP_SSE: | |
379 case FF_CMP_NSSE: | |
380 return lambda2>>FF_LAMBDA_SHIFT; | |
381 case FF_CMP_BIT: | |
382 return 1; | |
383 } | |
384 } | |
385 | |
1102 | 386 /** |
1104 | 387 * Empty mmx state. |
1102 | 388 * this must be called between any dsp function and float/double code. |
389 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
390 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
391 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
392 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
393 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
394 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
395 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
396 |
3089 | 397 #ifdef __GNUC__ |
398 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16))) | |
399 #else | |
400 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v | |
401 #endif | |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
402 |
62 | 403 #if defined(HAVE_MMX) |
0 | 404 |
862 | 405 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
406 |
0 | 407 #define MM_MMX 0x0001 /* standard MMX */ |
408 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
409 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
410 #define MM_SSE 0x0008 /* SSE functions */ | |
411 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
2388 | 412 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ |
0 | 413 |
414 extern int mm_flags; | |
415 | |
1064 | 416 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
417 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
1984
ef919e9ef73e
separate out put_signed_pixels_clamped() into its own function and
melanson
parents:
1977
diff
changeset
|
418 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 419 |
420 static inline void emms(void) | |
421 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
422 __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
|
423 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
424 |
936 | 425 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
426 #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
|
427 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
428 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
|
429 emms();\ |
0 | 430 } |
431 | |
3089 | 432 #ifdef __GNUC__ |
433 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) | |
434 #else | |
435 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v | |
436 #endif | |
437 | |
2324 | 438 #define STRIDE_ALIGN 8 |
0 | 439 |
1092 | 440 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
441 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 442 |
62 | 443 #elif defined(ARCH_ARMV4L) |
444 | |
445 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
446 line optimizations */ |
3089 | 447 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4))) |
2324 | 448 #define STRIDE_ALIGN 4 |
62 | 449 |
2776
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
450 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */ |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
451 |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
452 extern int mm_flags; |
930e56f92c57
IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern org)
michael
parents:
2763
diff
changeset
|
453 |
1092 | 454 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 455 |
88 | 456 #elif defined(HAVE_MLIB) |
457 | |
458 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
3089 | 459 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 460 #define STRIDE_ALIGN 8 |
88 | 461 |
1092 | 462 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 463 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
464 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
465 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
466 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ |
3089 | 467 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 468 #define STRIDE_ALIGN 8 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
469 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
470 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
471 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
472 |
3089 | 473 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 474 #define STRIDE_ALIGN 8 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
475 |
1092 | 476 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
477 |
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
|
478 #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
|
479 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
480 #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
|
481 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
482 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
|
483 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
484 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 485 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
486 #include <altivec.h> |
1653 | 487 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
488 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
489 |
3089 | 490 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 491 #define STRIDE_ALIGN 16 |
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
|
492 |
1092 | 493 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); |
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
|
494 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
495 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
496 |
3089 | 497 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16))) |
2324 | 498 #define STRIDE_ALIGN 16 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
499 |
1092 | 500 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
501 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
502 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
503 |
3089 | 504 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 505 #define STRIDE_ALIGN 8 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
506 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
507 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
508 |
0 | 509 #else |
510 | |
3089 | 511 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8))) |
2324 | 512 #define STRIDE_ALIGN 8 |
0 | 513 |
514 #endif | |
515 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
516 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
517 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
518 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
|
519 struct unaligned_32 { uint32_t l; } __attribute__((packed)); |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
520 struct unaligned_16 { uint16_t l; } __attribute__((packed)); |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
521 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
522 #define LD16(a) (((const struct unaligned_16 *) (a))->l) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
523 #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
|
524 #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
|
525 |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3010
diff
changeset
|
526 #define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
527 #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
|
528 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
529 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
530 |
1267
85b71f9f7450
moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)
michaelni
parents:
1264
diff
changeset
|
531 #define LD16(a) (*((uint16_t*)(a))) |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
532 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
533 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
534 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
535 #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
|
536 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
537 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
538 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
539 /* PSNR */ |
1064 | 540 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3], |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
541 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
|
542 AVCodecContext *avctx); |
781 | 543 |
544 /* FFT computation */ | |
545 | |
546 /* NOTE: soon integer code will be added, so you must use the | |
547 FFTSample type */ | |
548 typedef float FFTSample; | |
549 | |
550 typedef struct FFTComplex { | |
551 FFTSample re, im; | |
552 } FFTComplex; | |
553 | |
554 typedef struct FFTContext { | |
555 int nbits; | |
556 int inverse; | |
557 uint16_t *revtab; | |
558 FFTComplex *exptab; | |
559 FFTComplex *exptab1; /* only used by SSE code */ | |
560 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
561 } FFTContext; | |
562 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
563 int ff_fft_init(FFTContext *s, int nbits, int inverse); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
564 void ff_fft_permute(FFTContext *s, FFTComplex *z); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
565 void ff_fft_calc_c(FFTContext *s, FFTComplex *z); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
566 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z); |
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
567 void ff_fft_calc_altivec(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
|
568 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
569 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 570 { |
571 s->fft_calc(s, z); | |
572 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
573 void ff_fft_end(FFTContext *s); |
781 | 574 |
575 /* MDCT computation */ | |
576 | |
577 typedef struct MDCTContext { | |
578 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
579 int nbits; /* n = 2^nbits */ | |
580 /* pre/post rotation tables */ | |
581 FFTSample *tcos; | |
582 FFTSample *tsin; | |
583 FFTContext fft; | |
584 } MDCTContext; | |
585 | |
794 | 586 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
|
587 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 588 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
|
589 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 590 const FFTSample *input, FFTSample *tmp); |
794 | 591 void ff_mdct_end(MDCTContext *s); |
781 | 592 |
1708 | 593 #define WARPER8_16(name8, name16)\ |
594 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
595 return name8(s, dst , src , stride, h)\ | |
596 +name8(s, dst+8 , src+8 , stride, h);\ | |
597 } | |
598 | |
599 #define WARPER8_16_SQ(name8, name16)\ | |
600 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
601 int score=0;\ | |
602 score +=name8(s, dst , src , stride, 8);\ | |
603 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
604 if(h==16){\ | |
605 dst += 8*stride;\ | |
606 src += 8*stride;\ | |
607 score +=name8(s, dst , src , stride, 8);\ | |
608 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
609 }\ | |
610 return score;\ | |
936 | 611 } |
612 | |
1040
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
613 #endif |