Mercurial > libavcodec.hg
annotate fft.h @ 12266:48d6738904a9 libavcodec
Fix SPLATB_REG mess. Used to be a if/elseif/elseif/elseif spaghetti, so this
splits it into small optimization-specific macros which are selected for each
DSP function. The advantage of this approach is that the sse4 functions now
use the ssse3 codepath also without needing an explicit sse4 codepath.
author | rbultje |
---|---|
date | Sat, 24 Jul 2010 19:33:05 +0000 |
parents | 1bf322283429 |
children | b25537518e40 |
rev | line source |
---|---|
11370 | 1 /* |
2 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard | |
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> | |
4 * | |
5 * This file is part of FFmpeg. | |
6 * | |
7 * FFmpeg is free software; you can redistribute it and/or | |
8 * modify it under the terms of the GNU Lesser General Public | |
9 * License as published by the Free Software Foundation; either | |
10 * version 2.1 of the License, or (at your option) any later version. | |
11 * | |
12 * FFmpeg is distributed in the hope that it will be useful, | |
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
18 * License along with FFmpeg; if not, write to the Free Software | |
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
20 */ | |
21 | |
22 #ifndef AVCODEC_FFT_H | |
23 #define AVCODEC_FFT_H | |
24 | |
25 #include <stdint.h> | |
26 #include "config.h" | |
27 #include "libavutil/mem.h" | |
11392 | 28 #include "avfft.h" |
11370 | 29 |
30 /* FFT computation */ | |
31 | |
11392 | 32 struct FFTContext { |
11370 | 33 int nbits; |
34 int inverse; | |
35 uint16_t *revtab; | |
36 FFTComplex *tmp_buf; | |
37 int mdct_size; /* size of MDCT (i.e. number of input data * 2) */ | |
38 int mdct_bits; /* n = 2^nbits */ | |
39 /* pre/post rotation tables */ | |
40 FFTSample *tcos; | |
41 FFTSample *tsin; | |
42 void (*fft_permute)(struct FFTContext *s, FFTComplex *z); | |
43 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
44 void (*imdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input); | |
45 void (*imdct_half)(struct FFTContext *s, FFTSample *output, const FFTSample *input); | |
46 void (*mdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input); | |
47 int permutation; | |
48 #define FF_MDCT_PERM_NONE 0 | |
49 #define FF_MDCT_PERM_INTERLEAVE 1 | |
11392 | 50 }; |
11370 | 51 |
52 #if CONFIG_HARDCODED_TABLES | |
53 #define COSTABLE_CONST const | |
54 #define SINTABLE_CONST const | |
55 #define SINETABLE_CONST const | |
56 #else | |
57 #define COSTABLE_CONST | |
58 #define SINTABLE_CONST | |
59 #define SINETABLE_CONST | |
60 #endif | |
61 | |
62 #define COSTABLE(size) \ | |
63 COSTABLE_CONST DECLARE_ALIGNED(16, FFTSample, ff_cos_##size)[size/2] | |
64 #define SINTABLE(size) \ | |
65 SINTABLE_CONST DECLARE_ALIGNED(16, FFTSample, ff_sin_##size)[size/2] | |
66 #define SINETABLE(size) \ | |
67 SINETABLE_CONST DECLARE_ALIGNED(16, float, ff_sine_##size)[size] | |
68 extern COSTABLE(16); | |
69 extern COSTABLE(32); | |
70 extern COSTABLE(64); | |
71 extern COSTABLE(128); | |
72 extern COSTABLE(256); | |
73 extern COSTABLE(512); | |
74 extern COSTABLE(1024); | |
75 extern COSTABLE(2048); | |
76 extern COSTABLE(4096); | |
77 extern COSTABLE(8192); | |
78 extern COSTABLE(16384); | |
79 extern COSTABLE(32768); | |
80 extern COSTABLE(65536); | |
81 extern COSTABLE_CONST FFTSample* const ff_cos_tabs[17]; | |
82 | |
83 /** | |
12024 | 84 * Initialize the cosine table in ff_cos_tabs[index] |
11370 | 85 * \param index index in ff_cos_tabs array of the table to initialize |
86 */ | |
87 void ff_init_ff_cos_tabs(int index); | |
88 | |
89 extern SINTABLE(16); | |
90 extern SINTABLE(32); | |
91 extern SINTABLE(64); | |
92 extern SINTABLE(128); | |
93 extern SINTABLE(256); | |
94 extern SINTABLE(512); | |
95 extern SINTABLE(1024); | |
96 extern SINTABLE(2048); | |
97 extern SINTABLE(4096); | |
98 extern SINTABLE(8192); | |
99 extern SINTABLE(16384); | |
100 extern SINTABLE(32768); | |
101 extern SINTABLE(65536); | |
102 | |
103 /** | |
12024 | 104 * Set up a complex FFT. |
11370 | 105 * @param nbits log2 of the length of the input array |
106 * @param inverse if 0 perform the forward transform, if 1 perform the inverse | |
107 */ | |
108 int ff_fft_init(FFTContext *s, int nbits, int inverse); | |
109 void ff_fft_permute_c(FFTContext *s, FFTComplex *z); | |
110 void ff_fft_calc_c(FFTContext *s, FFTComplex *z); | |
111 | |
112 void ff_fft_init_altivec(FFTContext *s); | |
113 void ff_fft_init_mmx(FFTContext *s); | |
114 void ff_fft_init_arm(FFTContext *s); | |
12099 | 115 void ff_dct_init_mmx(DCTContext *s); |
11370 | 116 |
117 /** | |
118 * Do the permutation needed BEFORE calling ff_fft_calc(). | |
119 */ | |
120 static inline void ff_fft_permute(FFTContext *s, FFTComplex *z) | |
121 { | |
122 s->fft_permute(s, z); | |
123 } | |
124 /** | |
125 * Do a complex FFT with the parameters defined in ff_fft_init(). The | |
126 * input data must be permuted before. No 1.0/sqrt(n) normalization is done. | |
127 */ | |
128 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) | |
129 { | |
130 s->fft_calc(s, z); | |
131 } | |
132 void ff_fft_end(FFTContext *s); | |
133 | |
134 /* MDCT computation */ | |
135 | |
136 static inline void ff_imdct_calc(FFTContext *s, FFTSample *output, const FFTSample *input) | |
137 { | |
138 s->imdct_calc(s, output, input); | |
139 } | |
140 static inline void ff_imdct_half(FFTContext *s, FFTSample *output, const FFTSample *input) | |
141 { | |
142 s->imdct_half(s, output, input); | |
143 } | |
144 | |
145 static inline void ff_mdct_calc(FFTContext *s, FFTSample *output, | |
146 const FFTSample *input) | |
147 { | |
148 s->mdct_calc(s, output, input); | |
149 } | |
150 | |
151 /** | |
11944
052b9c58ccc4
Remove VLA in ff_kbd_window_init, limit window size to 1024
mru
parents:
11535
diff
changeset
|
152 * Maximum window size for ff_kbd_window_init. |
052b9c58ccc4
Remove VLA in ff_kbd_window_init, limit window size to 1024
mru
parents:
11535
diff
changeset
|
153 */ |
052b9c58ccc4
Remove VLA in ff_kbd_window_init, limit window size to 1024
mru
parents:
11535
diff
changeset
|
154 #define FF_KBD_WINDOW_MAX 1024 |
052b9c58ccc4
Remove VLA in ff_kbd_window_init, limit window size to 1024
mru
parents:
11535
diff
changeset
|
155 |
052b9c58ccc4
Remove VLA in ff_kbd_window_init, limit window size to 1024
mru
parents:
11535
diff
changeset
|
156 /** |
11370 | 157 * Generate a Kaiser-Bessel Derived Window. |
158 * @param window pointer to half window | |
159 * @param alpha determines window shape | |
11944
052b9c58ccc4
Remove VLA in ff_kbd_window_init, limit window size to 1024
mru
parents:
11535
diff
changeset
|
160 * @param n size of half window, max FF_KBD_WINDOW_MAX |
11370 | 161 */ |
162 void ff_kbd_window_init(float *window, float alpha, int n); | |
163 | |
164 /** | |
165 * Generate a sine window. | |
166 * @param window pointer to half window | |
167 * @param n size of half window | |
168 */ | |
169 void ff_sine_window_init(float *window, int n); | |
170 | |
171 /** | |
172 * initialize the specified entry of ff_sine_windows | |
173 */ | |
174 void ff_init_ff_sine_windows(int index); | |
175 extern SINETABLE( 32); | |
176 extern SINETABLE( 64); | |
177 extern SINETABLE( 128); | |
178 extern SINETABLE( 256); | |
179 extern SINETABLE( 512); | |
180 extern SINETABLE(1024); | |
181 extern SINETABLE(2048); | |
182 extern SINETABLE(4096); | |
183 extern SINETABLE_CONST float * const ff_sine_windows[13]; | |
184 | |
185 int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale); | |
186 void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input); | |
187 void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input); | |
188 void ff_mdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input); | |
189 void ff_mdct_end(FFTContext *s); | |
190 | |
191 /* Real Discrete Fourier Transform */ | |
192 | |
11392 | 193 struct RDFTContext { |
11370 | 194 int nbits; |
195 int inverse; | |
196 int sign_convention; | |
197 | |
198 /* pre/post rotation tables */ | |
199 const FFTSample *tcos; | |
200 SINTABLE_CONST FFTSample *tsin; | |
201 FFTContext fft; | |
11512 | 202 void (*rdft_calc)(struct RDFTContext *s, FFTSample *z); |
11392 | 203 }; |
11370 | 204 |
205 /** | |
12024 | 206 * Set up a real FFT. |
11370 | 207 * @param nbits log2 of the length of the input array |
208 * @param trans the type of transform | |
209 */ | |
210 int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans); | |
211 void ff_rdft_end(RDFTContext *s); | |
212 | |
11532 | 213 void ff_rdft_init_arm(RDFTContext *s); |
214 | |
11512 | 215 static av_always_inline void ff_rdft_calc(RDFTContext *s, FFTSample *data) |
216 { | |
217 s->rdft_calc(s, data); | |
218 } | |
219 | |
11370 | 220 /* Discrete Cosine Transform */ |
221 | |
11392 | 222 struct DCTContext { |
11370 | 223 int nbits; |
224 int inverse; | |
225 RDFTContext rdft; | |
226 const float *costab; | |
227 FFTSample *csc2; | |
11518
c4d18d452f82
Call DCT by function pointer. Needed for any future ASM implementation and
vitor
parents:
11516
diff
changeset
|
228 void (*dct_calc)(struct DCTContext *s, FFTSample *data); |
12026
3f3d08bb5cf8
More mp{1,2,3} 32-point DCT transform to our common DCT framework.
vitor
parents:
12024
diff
changeset
|
229 void (*dct32)(FFTSample *out, const FFTSample *in); |
11392 | 230 }; |
11370 | 231 |
232 /** | |
12024 | 233 * Set up DCT. |
11535
f468aac92300
Implement the discrete sine/cosine transforms DCT-I and DST-I
vitor
parents:
11532
diff
changeset
|
234 * @param nbits size of the input array: |
f468aac92300
Implement the discrete sine/cosine transforms DCT-I and DST-I
vitor
parents:
11532
diff
changeset
|
235 * (1 << nbits) for DCT-II, DCT-III and DST-I |
f468aac92300
Implement the discrete sine/cosine transforms DCT-I and DST-I
vitor
parents:
11532
diff
changeset
|
236 * (1 << nbits) + 1 for DCT-I |
f468aac92300
Implement the discrete sine/cosine transforms DCT-I and DST-I
vitor
parents:
11532
diff
changeset
|
237 * |
f468aac92300
Implement the discrete sine/cosine transforms DCT-I and DST-I
vitor
parents:
11532
diff
changeset
|
238 * @note the first element of the input of DST-I is ignored |
11370 | 239 */ |
11535
f468aac92300
Implement the discrete sine/cosine transforms DCT-I and DST-I
vitor
parents:
11532
diff
changeset
|
240 int ff_dct_init(DCTContext *s, int nbits, enum DCTTransformType type); |
11370 | 241 void ff_dct_calc(DCTContext *s, FFTSample *data); |
242 void ff_dct_end (DCTContext *s); | |
243 | |
244 #endif /* AVCODEC_FFT_H */ |