Mercurial > libavcodec.hg
annotate dsputil.h @ 2069:a8100a516b0a libavcodec
gcc bug workaround by (Jonathan Gray <khalek at linuxgamers dot net>)
author | michael |
---|---|
date | Tue, 08 Jun 2004 17:01:17 +0000 |
parents | 9e4bebc39ade |
children | aae3b4e66560 |
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 | |
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
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); | |
45 | |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
46 void ff_fdct_mmx(DCTELEM *block); |
1565 | 47 void ff_fdct_mmx2(DCTELEM *block); |
1765
e31754bc5b65
SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents:
1739
diff
changeset
|
48 void ff_fdct_sse2(DCTELEM *block); |
0 | 49 |
34 | 50 /* encoding scans */ |
1064 | 51 extern const uint8_t ff_alternate_horizontal_scan[64]; |
52 extern const uint8_t ff_alternate_vertical_scan[64]; | |
53 extern const uint8_t ff_zigzag_direct[64]; | |
1567 | 54 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
|
55 |
0 | 56 /* pixel operations */ |
57 #define MAX_NEG_CROP 384 | |
58 | |
59 /* temporary */ | |
1064 | 60 extern uint32_t squareTbl[512]; |
61 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 62 |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
63 /* VP3 DSP functions */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
64 void vp3_dsp_init_c(void); |
1977 | 65 void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix, |
66 int coeff_count, DCTELEM *output_data); | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
67 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
68 void vp3_dsp_init_mmx(void); |
1977 | 69 void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix, |
70 int coeff_count, DCTELEM *output_data); | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
71 |
1972 | 72 void vp3_dsp_init_sse2(void); |
1977 | 73 void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix, |
74 int coeff_count, DCTELEM *output_data); | |
0 | 75 |
675 | 76 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
77 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 78 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
79 | |
80 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
81 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
82 | |
83 !future video codecs might need functions with less strict alignment | |
84 */ | |
0 | 85 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
86 /* |
1064 | 87 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
88 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
89 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
90 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 91 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
|
92 */ |
0 | 93 |
94 /* add and put pixel (decoding) */ | |
675 | 95 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1709 | 96 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4 |
1064 | 97 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
|
98 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 | 99 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
1168 | 100 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
0 | 101 |
984 | 102 #define DEF_OLD_QPEL(name)\ |
1064 | 103 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
104 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
105 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
|
106 |
984 | 107 DEF_OLD_QPEL(qpel16_mc11_old_c) |
108 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
109 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
110 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
111 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
112 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
113 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
114 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
115 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
116 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
117 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
118 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 119 |
120 #define CALL_2X_PIXELS(a, b, n)\ | |
121 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
122 b(block , pixels , line_size, h);\ | |
123 b(block+n, pixels+n, line_size, h);\ | |
124 } | |
255 | 125 |
0 | 126 /* motion estimation */ |
1709 | 127 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2 |
128 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented | |
1708 | 129 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 | 130 |
1168 | 131 |
1102 | 132 /** |
133 * DSPContext. | |
134 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
135 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
136 /* pixel ops : interface with DCT */ |
1064 | 137 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
138 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
139 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
|
140 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1064 | 141 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
1136 | 142 /** |
143 * translational global motion compensation. | |
144 */ | |
1064 | 145 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 146 /** |
147 * global motion compensation. | |
148 */ | |
1064 | 149 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
150 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
|
151 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 152 int (*pix_sum)(uint8_t * pix, int line_size); |
153 int (*pix_norm1)(uint8_t * pix, int line_size); | |
1708 | 154 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 |
155 | |
1729 | 156 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ |
157 me_cmp_func sse[5]; | |
158 me_cmp_func hadamard8_diff[5]; | |
159 me_cmp_func dct_sad[5]; | |
160 me_cmp_func quant_psnr[5]; | |
161 me_cmp_func bit[5]; | |
162 me_cmp_func rd[5]; | |
163 me_cmp_func vsad[5]; | |
164 me_cmp_func vsse[5]; | |
2065
9e4bebc39ade
noise preserving sum of squares comparission function
michael
parents:
2045
diff
changeset
|
165 me_cmp_func nsse[5]; |
936 | 166 |
1708 | 167 me_cmp_func me_pre_cmp[5]; |
168 me_cmp_func me_cmp[5]; | |
169 me_cmp_func me_sub_cmp[5]; | |
170 me_cmp_func mb_cmp[5]; | |
1729 | 171 me_cmp_func ildct_cmp[5]; //only width 16 used |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
172 |
1136 | 173 /** |
174 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
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
|
175 * this is an array[4][4] of motion compensation funcions for 4 |
1713 | 176 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 177 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 178 * @param block destination where the result is stored |
179 * @param pixels source | |
180 * @param line_size number of bytes in a horizontal line of block | |
181 * @param h height | |
182 */ | |
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
|
183 op_pixels_func put_pixels_tab[4][4]; |
1136 | 184 |
185 /** | |
186 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
1320 | 187 * This is an array[4][4] of motion compensation functions for 4 |
1713 | 188 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> |
1213 | 189 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 190 * @param block destination into which the result is averaged (a+b+1)>>1 |
191 * @param pixels source | |
192 * @param line_size number of bytes in a horizontal line of block | |
193 * @param h height | |
194 */ | |
1319 | 195 op_pixels_func avg_pixels_tab[4][4]; |
1136 | 196 |
197 /** | |
198 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 199 * this is an array[2][4] of motion compensation funcions for 2 |
200 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 201 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 202 * @param block destination where the result is stored |
203 * @param pixels source | |
204 * @param line_size number of bytes in a horizontal line of block | |
205 * @param h height | |
206 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
207 op_pixels_func put_no_rnd_pixels_tab[2][4]; |
1136 | 208 |
209 /** | |
210 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1225 | 211 * this is an array[2][4] of motion compensation funcions for 2 |
212 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> | |
1213 | 213 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 214 * @param block destination into which the result is averaged (a+b)>>1 |
215 * @param pixels source | |
216 * @param line_size number of bytes in a horizontal line of block | |
217 * @param h height | |
218 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
219 op_pixels_func avg_no_rnd_pixels_tab[2][4]; |
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
|
220 |
1864 | 221 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); |
222 | |
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
|
223 /** |
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
|
224 * 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
|
225 * 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
|
226 * *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
|
227 * @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
|
228 * @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
|
229 * @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
|
230 * @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
|
231 */ |
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
|
232 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
1319 | 233 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
234 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
235 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
|
236 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
|
237 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
|
238 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 239 qpel_mc_func put_mspel_pixels_tab[8]; |
1168 | 240 |
241 /** | |
242 * h264 Chram MC | |
243 */ | |
244 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
245 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
|
246 |
1168 | 247 qpel_mc_func put_h264_qpel_pixels_tab[3][16]; |
248 qpel_mc_func avg_h264_qpel_pixels_tab[3][16]; | |
249 | |
1708 | 250 me_cmp_func pix_abs[2][4]; |
866 | 251 |
252 /* huffyuv specific */ | |
253 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 254 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1527 | 255 /** |
256 * subtract huffyuv's variant of median prediction | |
257 * note, this might read from src1[-1], src2[-1] | |
258 */ | |
259 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); | |
1273 | 260 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); |
1092 | 261 |
1644 | 262 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
263 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); | |
264 | |
2045 | 265 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
|
266 |
1092 | 267 /* (I)DCT */ |
268 void (*fdct)(DCTELEM *block/* align 16*/); | |
1567 | 269 void (*fdct248)(DCTELEM *block/* align 16*/); |
1102 | 270 |
1324
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
271 /* IDCT really*/ |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
272 void (*idct)(DCTELEM *block/* align 16*/); |
7d328fd9d8a5
the return of the idct with 16bit output by ("Ivan Kalvachev" <ivan at cacad dot com>)
michaelni
parents:
1320
diff
changeset
|
273 |
1102 | 274 /** |
1104 | 275 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 276 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 277 * @param line_size size in bytes of a horizotal line of dest |
1102 | 278 */ |
1092 | 279 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 280 |
281 /** | |
282 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 283 * @param line_size size in bytes of a horizotal line of dest |
1102 | 284 */ |
1092 | 285 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 286 |
287 /** | |
1104 | 288 * idct input permutation. |
1241 | 289 * several optimized IDCTs need a permutated input (relative to the normal order of the reference |
290 * IDCT) | |
291 * this permutation must be performed before the idct_put/add, note, normally this can be merged | |
292 * with the zigzag/alternate scan<br> | |
1102 | 293 * an example to avoid confusion: |
294 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
295 * - (x -> referece dct -> reference idct -> x) | |
296 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
297 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
298 */ | |
1092 | 299 uint8_t idct_permutation[64]; |
300 int idct_permutation_type; | |
301 #define FF_NO_IDCT_PERM 1 | |
302 #define FF_LIBMPEG2_IDCT_PERM 2 | |
303 #define FF_SIMPLE_IDCT_PERM 3 | |
304 #define FF_TRANSPOSE_IDCT_PERM 4 | |
305 | |
1784 | 306 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
307 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
308 #define BASIS_SHIFT 16 | |
309 #define RECON_SHIFT 6 | |
310 | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
311 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
312 * This function handles any initialization for the VP3 DSP functions. |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
313 */ |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
314 void (*vp3_dsp_init)(void); |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
315 |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
316 /** |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
317 * This function is responsible for taking a block of zigzag'd, |
1977 | 318 * quantized DCT coefficients and reconstructing the original block of |
319 * samples. | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
320 * @param input_data 64 zigzag'd, quantized DCT coefficients |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
321 * @param dequant_matrix 64 zigzag'd quantizer coefficients |
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
322 * @param coeff_count index of the last coefficient |
1977 | 323 * @param output_samples space for 64 DCTELEMs where the transformed |
324 * samples will be stored | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
325 */ |
1977 | 326 void (*vp3_idct)(int16_t *input_data, int16_t *dequant_matrix, |
327 int coeff_count, DCTELEM *output_samples); | |
1866
1755f959ab7f
seperated out the C-based VP3 DSP functions into a different file; also
melanson
parents:
1864
diff
changeset
|
328 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
329 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
330 |
1201 | 331 void dsputil_static_init(void); |
1092 | 332 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 333 |
764 | 334 /** |
335 * permute block according to permuatation. | |
336 * @param last last non zero element in scantable order | |
337 */ | |
1064 | 338 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 339 |
1729 | 340 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
341 | |
1264 | 342 #define BYTE_VEC32(c) ((c)*0x01010101UL) |
343 | |
344 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) | |
345 { | |
346 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
347 } | |
348 | |
349 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) | |
350 { | |
351 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); | |
352 } | |
353 | |
1102 | 354 /** |
1104 | 355 * Empty mmx state. |
1102 | 356 * this must be called between any dsp function and float/double code. |
357 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
358 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
359 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
360 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
361 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
362 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
363 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
364 |
1974
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
365 #define __align16 __attribute__ ((aligned (16))) |
8c5489b2cf3e
move __align16 some place where non-MMX machines can see it
melanson
parents:
1972
diff
changeset
|
366 |
62 | 367 #if defined(HAVE_MMX) |
0 | 368 |
862 | 369 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
370 |
0 | 371 #define MM_MMX 0x0001 /* standard MMX */ |
372 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
373 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
374 #define MM_SSE 0x0008 /* SSE functions */ | |
375 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
376 | |
377 extern int mm_flags; | |
378 | |
1064 | 379 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
380 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
|
381 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
0 | 382 |
383 static inline void emms(void) | |
384 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
385 __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
|
386 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
387 |
936 | 388 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
389 #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
|
390 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
391 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
|
392 emms();\ |
0 | 393 } |
394 | |
395 #define __align8 __attribute__ ((aligned (8))) | |
396 | |
1092 | 397 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
398 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 399 |
62 | 400 #elif defined(ARCH_ARMV4L) |
401 | |
402 /* 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
|
403 line optimizations */ |
62 | 404 #define __align8 __attribute__ ((aligned (4))) |
405 | |
1092 | 406 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 407 |
88 | 408 #elif defined(HAVE_MLIB) |
409 | |
410 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
411 #define __align8 __attribute__ ((aligned (8))) | |
412 | |
1092 | 413 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 414 |
1959
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
415 #elif defined(ARCH_SPARC) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
416 |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
417 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
418 #define __align8 __attribute__ ((aligned (8))) |
55b7435c59b8
VIS optimized motion compensation code. by (David S. Miller <davem at redhat dot com>)
michael
parents:
1879
diff
changeset
|
419 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
|
420 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
421 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
422 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
423 #define __align8 __attribute__ ((aligned (8))) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
424 |
1092 | 425 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
|
426 |
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
|
427 #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
|
428 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
429 #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
|
430 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
431 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
|
432 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
433 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
1653 | 434 #define pixel altivec_pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
435 #include <altivec.h> |
1653 | 436 #undef pixel |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
437 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
438 |
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
|
439 #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
|
440 |
1092 | 441 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
|
442 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
443 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
444 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
445 #define __align8 __attribute__ ((aligned (16))) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
446 |
1092 | 447 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
|
448 |
1259
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
449 #elif defined(ARCH_SH4) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
450 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
451 #define __align8 __attribute__ ((aligned (8))) |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
452 |
e8c3884f2c7e
sh4 optimized idct & bswap patch by (BERO <bero at geocities dot co dot jp>)
michaelni
parents:
1241
diff
changeset
|
453 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
|
454 |
0 | 455 #else |
456 | |
457 #define __align8 | |
458 | |
459 #endif | |
460 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
461 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
462 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
463 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
|
464 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
|
465 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
|
466 |
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
|
467 #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
|
468 #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
|
469 #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
|
470 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
471 #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
|
472 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
473 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
474 |
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
|
475 #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
|
476 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
477 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
478 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
479 #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
|
480 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
481 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
482 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
483 /* PSNR */ |
1064 | 484 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
|
485 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
|
486 AVCodecContext *avctx); |
781 | 487 |
488 /* FFT computation */ | |
489 | |
490 /* NOTE: soon integer code will be added, so you must use the | |
491 FFTSample type */ | |
492 typedef float FFTSample; | |
493 | |
494 typedef struct FFTComplex { | |
495 FFTSample re, im; | |
496 } FFTComplex; | |
497 | |
498 typedef struct FFTContext { | |
499 int nbits; | |
500 int inverse; | |
501 uint16_t *revtab; | |
502 FFTComplex *exptab; | |
503 FFTComplex *exptab1; /* only used by SSE code */ | |
504 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
505 } FFTContext; | |
506 | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
507 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
|
508 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
|
509 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
|
510 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
|
511 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
|
512 |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
513 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) |
781 | 514 { |
515 s->fft_calc(s, z); | |
516 } | |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1866
diff
changeset
|
517 void ff_fft_end(FFTContext *s); |
781 | 518 |
519 /* MDCT computation */ | |
520 | |
521 typedef struct MDCTContext { | |
522 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
523 int nbits; /* n = 2^nbits */ | |
524 /* pre/post rotation tables */ | |
525 FFTSample *tcos; | |
526 FFTSample *tsin; | |
527 FFTContext fft; | |
528 } MDCTContext; | |
529 | |
794 | 530 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
|
531 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 532 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
|
533 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 534 const FFTSample *input, FFTSample *tmp); |
794 | 535 void ff_mdct_end(MDCTContext *s); |
781 | 536 |
1708 | 537 #define WARPER8_16(name8, name16)\ |
538 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
539 return name8(s, dst , src , stride, h)\ | |
540 +name8(s, dst+8 , src+8 , stride, h);\ | |
541 } | |
542 | |
543 #define WARPER8_16_SQ(name8, name16)\ | |
544 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ | |
545 int score=0;\ | |
546 score +=name8(s, dst , src , stride, 8);\ | |
547 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
548 if(h==16){\ | |
549 dst += 8*stride;\ | |
550 src += 8*stride;\ | |
551 score +=name8(s, dst , src , stride, 8);\ | |
552 score +=name8(s, dst+8 , src+8 , stride, 8);\ | |
553 }\ | |
554 return score;\ | |
936 | 555 } |
556 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
557 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
558 /* 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
|
559 /* better than nothing implementation. */ |
802 | 560 /* 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
|
561 static inline long int lrintf(float x) |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
562 { |
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
|
563 #ifdef CONFIG_WIN32 |
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
|
564 /* XXX: incorrect, but make it compile */ |
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
|
565 return (int)(x); |
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
|
566 #else |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
567 return (int)(rint(x)); |
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
|
568 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
569 } |
2024
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
570 #else |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
571 #ifndef _ISOC9X_SOURCE |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
572 #define _ISOC9X_SOURCE |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
573 #endif |
f65d87bfdd5a
some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
1984
diff
changeset
|
574 #include <math.h> |
0 | 575 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
576 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
577 #endif |