Merge branch 'master' of http://repo.or.cz/r/FFMpeg-mirror
[FFMpeg-mirror/DVCPRO-HD.git] / libavcodec / fft-test.c
blob980dc8688594122d7c4317e971ed6a6ace7e8354
1 /*
2 * (c) 2002 Fabrice Bellard
4 * This file is part of FFmpeg.
6 * FFmpeg 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 2.1 of the License, or (at your option) any later version.
11 * FFmpeg 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.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 /**
22 * @file fft-test.c
23 * FFT and MDCT tests.
26 #include "dsputil.h"
27 #include <math.h>
28 #include <unistd.h>
29 #include <sys/time.h>
30 #include <stdlib.h>
31 #include <string.h>
33 #undef exit
34 #undef random
36 /* reference fft */
38 #define MUL16(a,b) ((a) * (b))
40 #define CMAC(pre, pim, are, aim, bre, bim) \
42 pre += (MUL16(are, bre) - MUL16(aim, bim));\
43 pim += (MUL16(are, bim) + MUL16(bre, aim));\
46 FFTComplex *exptab;
48 void fft_ref_init(int nbits, int inverse)
50 int n, i;
51 double c1, s1, alpha;
53 n = 1 << nbits;
54 exptab = av_malloc((n / 2) * sizeof(FFTComplex));
56 for(i=0;i<(n/2);i++) {
57 alpha = 2 * M_PI * (float)i / (float)n;
58 c1 = cos(alpha);
59 s1 = sin(alpha);
60 if (!inverse)
61 s1 = -s1;
62 exptab[i].re = c1;
63 exptab[i].im = s1;
67 void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
69 int n, i, j, k, n2;
70 double tmp_re, tmp_im, s, c;
71 FFTComplex *q;
73 n = 1 << nbits;
74 n2 = n >> 1;
75 for(i=0;i<n;i++) {
76 tmp_re = 0;
77 tmp_im = 0;
78 q = tab;
79 for(j=0;j<n;j++) {
80 k = (i * j) & (n - 1);
81 if (k >= n2) {
82 c = -exptab[k - n2].re;
83 s = -exptab[k - n2].im;
84 } else {
85 c = exptab[k].re;
86 s = exptab[k].im;
88 CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
89 q++;
91 tabr[i].re = tmp_re;
92 tabr[i].im = tmp_im;
96 void imdct_ref(float *out, float *in, int nbits)
98 int n = 1<<nbits;
99 int k, i, a;
100 double sum, f;
102 for(i=0;i<n;i++) {
103 sum = 0;
104 for(k=0;k<n/2;k++) {
105 a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
106 f = cos(M_PI * a / (double)(2 * n));
107 sum += f * in[k];
109 out[i] = -sum;
113 /* NOTE: no normalisation by 1 / N is done */
114 void mdct_ref(float *output, float *input, int nbits)
116 int n = 1<<nbits;
117 int k, i;
118 double a, s;
120 /* do it by hand */
121 for(k=0;k<n/2;k++) {
122 s = 0;
123 for(i=0;i<n;i++) {
124 a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n));
125 s += input[i] * cos(a);
127 output[k] = s;
132 float frandom(void)
134 return (float)((random() & 0xffff) - 32768) / 32768.0;
137 int64_t gettime(void)
139 struct timeval tv;
140 gettimeofday(&tv,NULL);
141 return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
144 void check_diff(float *tab1, float *tab2, int n)
146 int i;
147 double max= 0;
148 double error= 0;
150 for(i=0;i<n;i++) {
151 double e= fabsf(tab1[i] - tab2[i]);
152 if (e >= 1e-3) {
153 av_log(NULL, AV_LOG_ERROR, "ERROR %d: %f %f\n",
154 i, tab1[i], tab2[i]);
156 error+= e*e;
157 if(e>max) max= e;
159 av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error)/n);
163 void help(void)
165 av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n"
166 "-h print this help\n"
167 "-s speed test\n"
168 "-m (I)MDCT test\n"
169 "-i inverse transform test\n"
170 "-n b set the transform size to 2^b\n"
172 exit(1);
177 int main(int argc, char **argv)
179 FFTComplex *tab, *tab1, *tab_ref;
180 FFTSample *tabtmp, *tab2;
181 int it, i, c;
182 int do_speed = 0;
183 int do_mdct = 0;
184 int do_inverse = 0;
185 FFTContext s1, *s = &s1;
186 MDCTContext m1, *m = &m1;
187 int fft_nbits, fft_size;
189 fft_nbits = 9;
190 for(;;) {
191 c = getopt(argc, argv, "hsimn:");
192 if (c == -1)
193 break;
194 switch(c) {
195 case 'h':
196 help();
197 break;
198 case 's':
199 do_speed = 1;
200 break;
201 case 'i':
202 do_inverse = 1;
203 break;
204 case 'm':
205 do_mdct = 1;
206 break;
207 case 'n':
208 fft_nbits = atoi(optarg);
209 break;
213 fft_size = 1 << fft_nbits;
214 tab = av_malloc(fft_size * sizeof(FFTComplex));
215 tab1 = av_malloc(fft_size * sizeof(FFTComplex));
216 tab_ref = av_malloc(fft_size * sizeof(FFTComplex));
217 tabtmp = av_malloc(fft_size / 2 * sizeof(FFTSample));
218 tab2 = av_malloc(fft_size * sizeof(FFTSample));
220 if (do_mdct) {
221 if (do_inverse)
222 av_log(NULL, AV_LOG_INFO,"IMDCT");
223 else
224 av_log(NULL, AV_LOG_INFO,"MDCT");
225 ff_mdct_init(m, fft_nbits, do_inverse);
226 } else {
227 if (do_inverse)
228 av_log(NULL, AV_LOG_INFO,"IFFT");
229 else
230 av_log(NULL, AV_LOG_INFO,"FFT");
231 ff_fft_init(s, fft_nbits, do_inverse);
232 fft_ref_init(fft_nbits, do_inverse);
234 av_log(NULL, AV_LOG_INFO," %d test\n", fft_size);
236 /* generate random data */
238 for(i=0;i<fft_size;i++) {
239 tab1[i].re = frandom();
240 tab1[i].im = frandom();
243 /* checking result */
244 av_log(NULL, AV_LOG_INFO,"Checking...\n");
246 if (do_mdct) {
247 if (do_inverse) {
248 imdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
249 ff_imdct_calc(m, tab2, (float *)tab1, tabtmp);
250 check_diff((float *)tab_ref, tab2, fft_size);
251 } else {
252 mdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
254 ff_mdct_calc(m, tab2, (float *)tab1, tabtmp);
256 check_diff((float *)tab_ref, tab2, fft_size / 2);
258 } else {
259 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
260 ff_fft_permute(s, tab);
261 ff_fft_calc(s, tab);
263 fft_ref(tab_ref, tab1, fft_nbits);
264 check_diff((float *)tab_ref, (float *)tab, fft_size * 2);
267 /* do a speed test */
269 if (do_speed) {
270 int64_t time_start, duration;
271 int nb_its;
273 av_log(NULL, AV_LOG_INFO,"Speed test...\n");
274 /* we measure during about 1 seconds */
275 nb_its = 1;
276 for(;;) {
277 time_start = gettime();
278 for(it=0;it<nb_its;it++) {
279 if (do_mdct) {
280 if (do_inverse) {
281 ff_imdct_calc(m, (float *)tab, (float *)tab1, tabtmp);
282 } else {
283 ff_mdct_calc(m, (float *)tab, (float *)tab1, tabtmp);
285 } else {
286 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
287 ff_fft_calc(s, tab);
290 duration = gettime() - time_start;
291 if (duration >= 1000000)
292 break;
293 nb_its *= 2;
295 av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
296 (double)duration / nb_its,
297 (double)duration / 1000000.0,
298 nb_its);
301 if (do_mdct) {
302 ff_mdct_end(m);
303 } else {
304 ff_fft_end(s);
306 return 0;