ff_fft_calc_3dn/3dn2/sse: convert intrinsics to inline asm.
[libav.git] / libavcodec / fft.c
1 /*
2 * FFT/IFFT transforms
3 * Copyright (c) 2002 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20 /**
21 * @file fft.c
22 * FFT/IFFT transforms.
23 */
24
25 #include "dsputil.h"
26
27 /**
28 * The size of the FFT is 2^nbits. If inverse is TRUE, inverse FFT is
29 * done
30 */
31 int ff_fft_init(FFTContext *s, int nbits, int inverse)
32 {
33 int i, j, m, n;
34 float alpha, c1, s1, s2;
35
36 s->nbits = nbits;
37 n = 1 << nbits;
38
39 s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
40 if (!s->exptab)
41 goto fail;
42 s->revtab = av_malloc(n * sizeof(uint16_t));
43 if (!s->revtab)
44 goto fail;
45 s->inverse = inverse;
46
47 s2 = inverse ? 1.0 : -1.0;
48
49 for(i=0;i<(n/2);i++) {
50 alpha = 2 * M_PI * (float)i / (float)n;
51 c1 = cos(alpha);
52 s1 = sin(alpha) * s2;
53 s->exptab[i].re = c1;
54 s->exptab[i].im = s1;
55 }
56 s->fft_calc = ff_fft_calc_c;
57 s->imdct_calc = ff_imdct_calc;
58 s->exptab1 = NULL;
59
60 /* compute constant table for HAVE_SSE version */
61 #if defined(HAVE_MMX) \
62 || (defined(HAVE_ALTIVEC) && !defined(ALTIVEC_USE_REFERENCE_C_CODE))
63 {
64 int has_vectors = mm_support();
65
66 if (has_vectors) {
67 #if defined(HAVE_MMX)
68 if (has_vectors & MM_3DNOWEXT)
69 s->imdct_calc = ff_imdct_calc_3dn2;
70 if (has_vectors & MM_3DNOWEXT)
71 /* 3DNowEx for Athlon(XP) */
72 s->fft_calc = ff_fft_calc_3dn2;
73 else if (has_vectors & MM_3DNOW)
74 /* 3DNow! for K6-2/3 */
75 s->fft_calc = ff_fft_calc_3dn;
76 if (has_vectors & MM_SSE2)
77 /* SSE for P4/K8 */
78 s->fft_calc = ff_fft_calc_sse;
79 else if ((has_vectors & MM_SSE) &&
80 s->fft_calc == ff_fft_calc_c)
81 /* SSE for P3 */
82 s->fft_calc = ff_fft_calc_sse;
83 #else /* HAVE_MMX */
84 if (has_vectors & MM_ALTIVEC)
85 s->fft_calc = ff_fft_calc_altivec;
86 #endif
87 }
88 if (s->fft_calc != ff_fft_calc_c) {
89 int np, nblocks, np2, l;
90 FFTComplex *q;
91
92 np = 1 << nbits;
93 nblocks = np >> 3;
94 np2 = np >> 1;
95 s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
96 if (!s->exptab1)
97 goto fail;
98 q = s->exptab1;
99 do {
100 for(l = 0; l < np2; l += 2 * nblocks) {
101 *q++ = s->exptab[l];
102 *q++ = s->exptab[l + nblocks];
103
104 q->re = -s->exptab[l].im;
105 q->im = s->exptab[l].re;
106 q++;
107 q->re = -s->exptab[l + nblocks].im;
108 q->im = s->exptab[l + nblocks].re;
109 q++;
110 }
111 nblocks = nblocks >> 1;
112 } while (nblocks != 0);
113 av_freep(&s->exptab);
114 }
115 }
116 #endif
117
118 /* compute bit reverse table */
119
120 for(i=0;i<n;i++) {
121 m=0;
122 for(j=0;j<nbits;j++) {
123 m |= ((i >> j) & 1) << (nbits-j-1);
124 }
125 s->revtab[i]=m;
126 }
127 return 0;
128 fail:
129 av_freep(&s->revtab);
130 av_freep(&s->exptab);
131 av_freep(&s->exptab1);
132 return -1;
133 }
134
135 /* butter fly op */
136 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
137 {\
138 FFTSample ax, ay, bx, by;\
139 bx=pre1;\
140 by=pim1;\
141 ax=qre1;\
142 ay=qim1;\
143 pre = (bx + ax);\
144 pim = (by + ay);\
145 qre = (bx - ax);\
146 qim = (by - ay);\
147 }
148
149 #define MUL16(a,b) ((a) * (b))
150
151 #define CMUL(pre, pim, are, aim, bre, bim) \
152 {\
153 pre = (MUL16(are, bre) - MUL16(aim, bim));\
154 pim = (MUL16(are, bim) + MUL16(bre, aim));\
155 }
156
157 /**
158 * Do a complex FFT with the parameters defined in ff_fft_init(). The
159 * input data must be permuted before with s->revtab table. No
160 * 1.0/sqrt(n) normalization is done.
161 */
162 void ff_fft_calc_c(FFTContext *s, FFTComplex *z)
163 {
164 int ln = s->nbits;
165 int j, np, np2;
166 int nblocks, nloops;
167 register FFTComplex *p, *q;
168 FFTComplex *exptab = s->exptab;
169 int l;
170 FFTSample tmp_re, tmp_im;
171
172 np = 1 << ln;
173
174 /* pass 0 */
175
176 p=&z[0];
177 j=(np >> 1);
178 do {
179 BF(p[0].re, p[0].im, p[1].re, p[1].im,
180 p[0].re, p[0].im, p[1].re, p[1].im);
181 p+=2;
182 } while (--j != 0);
183
184 /* pass 1 */
185
186
187 p=&z[0];
188 j=np >> 2;
189 if (s->inverse) {
190 do {
191 BF(p[0].re, p[0].im, p[2].re, p[2].im,
192 p[0].re, p[0].im, p[2].re, p[2].im);
193 BF(p[1].re, p[1].im, p[3].re, p[3].im,
194 p[1].re, p[1].im, -p[3].im, p[3].re);
195 p+=4;
196 } while (--j != 0);
197 } else {
198 do {
199 BF(p[0].re, p[0].im, p[2].re, p[2].im,
200 p[0].re, p[0].im, p[2].re, p[2].im);
201 BF(p[1].re, p[1].im, p[3].re, p[3].im,
202 p[1].re, p[1].im, p[3].im, -p[3].re);
203 p+=4;
204 } while (--j != 0);
205 }
206 /* pass 2 .. ln-1 */
207
208 nblocks = np >> 3;
209 nloops = 1 << 2;
210 np2 = np >> 1;
211 do {
212 p = z;
213 q = z + nloops;
214 for (j = 0; j < nblocks; ++j) {
215 BF(p->re, p->im, q->re, q->im,
216 p->re, p->im, q->re, q->im);
217
218 p++;
219 q++;
220 for(l = nblocks; l < np2; l += nblocks) {
221 CMUL(tmp_re, tmp_im, exptab[l].re, exptab[l].im, q->re, q->im);
222 BF(p->re, p->im, q->re, q->im,
223 p->re, p->im, tmp_re, tmp_im);
224 p++;
225 q++;
226 }
227
228 p += nloops;
229 q += nloops;
230 }
231 nblocks = nblocks >> 1;
232 nloops = nloops << 1;
233 } while (nblocks != 0);
234 }
235
236 /**
237 * Do the permutation needed BEFORE calling ff_fft_calc()
238 */
239 void ff_fft_permute(FFTContext *s, FFTComplex *z)
240 {
241 int j, k, np;
242 FFTComplex tmp;
243 const uint16_t *revtab = s->revtab;
244
245 /* reverse */
246 np = 1 << s->nbits;
247 for(j=0;j<np;j++) {
248 k = revtab[j];
249 if (k < j) {
250 tmp = z[k];
251 z[k] = z[j];
252 z[j] = tmp;
253 }
254 }
255 }
256
257 void ff_fft_end(FFTContext *s)
258 {
259 av_freep(&s->revtab);
260 av_freep(&s->exptab);
261 av_freep(&s->exptab1);
262 }
263