yading@10: /*
yading@10:  * (c) 2002 Fabrice Bellard
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * FFT and MDCT tests.
yading@10:  */
yading@10: 
yading@10: #include "libavutil/cpu.h"
yading@10: #include "libavutil/mathematics.h"
yading@10: #include "libavutil/lfg.h"
yading@10: #include "libavutil/log.h"
yading@10: #include "libavutil/time.h"
yading@10: #include "fft.h"
yading@10: #if CONFIG_FFT_FLOAT
yading@10: #include "dct.h"
yading@10: #include "rdft.h"
yading@10: #endif
yading@10: #include <math.h>
yading@10: #if HAVE_UNISTD_H
yading@10: #include <unistd.h>
yading@10: #endif
yading@10: #include <stdlib.h>
yading@10: #include <string.h>
yading@10: 
yading@10: /* reference fft */
yading@10: 
yading@10: #define MUL16(a,b) ((a) * (b))
yading@10: 
yading@10: #define CMAC(pre, pim, are, aim, bre, bim) \
yading@10: {\
yading@10:    pre += (MUL16(are, bre) - MUL16(aim, bim));\
yading@10:    pim += (MUL16(are, bim) + MUL16(bre, aim));\
yading@10: }
yading@10: 
yading@10: #if CONFIG_FFT_FLOAT
yading@10: #   define RANGE 1.0
yading@10: #   define REF_SCALE(x, bits)  (x)
yading@10: #   define FMT "%10.6f"
yading@10: #else
yading@10: #   define RANGE 16384
yading@10: #   define REF_SCALE(x, bits) ((x) / (1<<(bits)))
yading@10: #   define FMT "%6d"
yading@10: #endif
yading@10: 
yading@10: struct {
yading@10:     float re, im;
yading@10: } *exptab;
yading@10: 
yading@10: static void fft_ref_init(int nbits, int inverse)
yading@10: {
yading@10:     int n, i;
yading@10:     double c1, s1, alpha;
yading@10: 
yading@10:     n = 1 << nbits;
yading@10:     exptab = av_malloc((n / 2) * sizeof(*exptab));
yading@10: 
yading@10:     for (i = 0; i < (n/2); i++) {
yading@10:         alpha = 2 * M_PI * (float)i / (float)n;
yading@10:         c1 = cos(alpha);
yading@10:         s1 = sin(alpha);
yading@10:         if (!inverse)
yading@10:             s1 = -s1;
yading@10:         exptab[i].re = c1;
yading@10:         exptab[i].im = s1;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
yading@10: {
yading@10:     int n, i, j, k, n2;
yading@10:     double tmp_re, tmp_im, s, c;
yading@10:     FFTComplex *q;
yading@10: 
yading@10:     n = 1 << nbits;
yading@10:     n2 = n >> 1;
yading@10:     for (i = 0; i < n; i++) {
yading@10:         tmp_re = 0;
yading@10:         tmp_im = 0;
yading@10:         q = tab;
yading@10:         for (j = 0; j < n; j++) {
yading@10:             k = (i * j) & (n - 1);
yading@10:             if (k >= n2) {
yading@10:                 c = -exptab[k - n2].re;
yading@10:                 s = -exptab[k - n2].im;
yading@10:             } else {
yading@10:                 c = exptab[k].re;
yading@10:                 s = exptab[k].im;
yading@10:             }
yading@10:             CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
yading@10:             q++;
yading@10:         }
yading@10:         tabr[i].re = REF_SCALE(tmp_re, nbits);
yading@10:         tabr[i].im = REF_SCALE(tmp_im, nbits);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
yading@10: {
yading@10:     int n = 1<<nbits;
yading@10:     int k, i, a;
yading@10:     double sum, f;
yading@10: 
yading@10:     for (i = 0; i < n; i++) {
yading@10:         sum = 0;
yading@10:         for (k = 0; k < n/2; k++) {
yading@10:             a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
yading@10:             f = cos(M_PI * a / (double)(2 * n));
yading@10:             sum += f * in[k];
yading@10:         }
yading@10:         out[i] = REF_SCALE(-sum, nbits - 2);
yading@10:     }
yading@10: }
yading@10: 
yading@10: /* NOTE: no normalisation by 1 / N is done */
yading@10: static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
yading@10: {
yading@10:     int n = 1<<nbits;
yading@10:     int k, i;
yading@10:     double a, s;
yading@10: 
yading@10:     /* do it by hand */
yading@10:     for (k = 0; k < n/2; k++) {
yading@10:         s = 0;
yading@10:         for (i = 0; i < n; i++) {
yading@10:             a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n));
yading@10:             s += input[i] * cos(a);
yading@10:         }
yading@10:         output[k] = REF_SCALE(s, nbits - 1);
yading@10:     }
yading@10: }
yading@10: 
yading@10: #if CONFIG_FFT_FLOAT
yading@10: static void idct_ref(FFTSample *output, FFTSample *input, int nbits)
yading@10: {
yading@10:     int n = 1<<nbits;
yading@10:     int k, i;
yading@10:     double a, s;
yading@10: 
yading@10:     /* do it by hand */
yading@10:     for (i = 0; i < n; i++) {
yading@10:         s = 0.5 * input[0];
yading@10:         for (k = 1; k < n; k++) {
yading@10:             a = M_PI*k*(i+0.5) / n;
yading@10:             s += input[k] * cos(a);
yading@10:         }
yading@10:         output[i] = 2 * s / n;
yading@10:     }
yading@10: }
yading@10: static void dct_ref(FFTSample *output, FFTSample *input, int nbits)
yading@10: {
yading@10:     int n = 1<<nbits;
yading@10:     int k, i;
yading@10:     double a, s;
yading@10: 
yading@10:     /* do it by hand */
yading@10:     for (k = 0; k < n; k++) {
yading@10:         s = 0;
yading@10:         for (i = 0; i < n; i++) {
yading@10:             a = M_PI*k*(i+0.5) / n;
yading@10:             s += input[i] * cos(a);
yading@10:         }
yading@10:         output[k] = s;
yading@10:     }
yading@10: }
yading@10: #endif
yading@10: 
yading@10: 
yading@10: static FFTSample frandom(AVLFG *prng)
yading@10: {
yading@10:     return (int16_t)av_lfg_get(prng) / 32768.0 * RANGE;
yading@10: }
yading@10: 
yading@10: static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
yading@10: {
yading@10:     int i;
yading@10:     double max= 0;
yading@10:     double error= 0;
yading@10:     int err = 0;
yading@10: 
yading@10:     for (i = 0; i < n; i++) {
yading@10:         double e = fabsf(tab1[i] - (tab2[i] / scale)) / RANGE;
yading@10:         if (e >= 1e-3) {
yading@10:             av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n",
yading@10:                    i, tab1[i], tab2[i]);
yading@10:             err = 1;
yading@10:         }
yading@10:         error+= e*e;
yading@10:         if(e>max) max= e;
yading@10:     }
yading@10:     av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error/n));
yading@10:     return err;
yading@10: }
yading@10: 
yading@10: 
yading@10: static void help(void)
yading@10: {
yading@10:     av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n"
yading@10:            "-h     print this help\n"
yading@10:            "-s     speed test\n"
yading@10:            "-m     (I)MDCT test\n"
yading@10:            "-d     (I)DCT test\n"
yading@10:            "-r     (I)RDFT test\n"
yading@10:            "-i     inverse transform test\n"
yading@10:            "-n b   set the transform size to 2^b\n"
yading@10:            "-f x   set scale factor for output data of (I)MDCT to x\n"
yading@10:            );
yading@10: }
yading@10: 
yading@10: enum tf_transform {
yading@10:     TRANSFORM_FFT,
yading@10:     TRANSFORM_MDCT,
yading@10:     TRANSFORM_RDFT,
yading@10:     TRANSFORM_DCT,
yading@10: };
yading@10: 
yading@10: #if !HAVE_GETOPT
yading@10: #include "compat/getopt.c"
yading@10: #endif
yading@10: 
yading@10: int main(int argc, char **argv)
yading@10: {
yading@10:     FFTComplex *tab, *tab1, *tab_ref;
yading@10:     FFTSample *tab2;
yading@10:     int it, i, c;
yading@10:     int cpuflags;
yading@10:     int do_speed = 0;
yading@10:     int err = 1;
yading@10:     enum tf_transform transform = TRANSFORM_FFT;
yading@10:     int do_inverse = 0;
yading@10:     FFTContext s1, *s = &s1;
yading@10:     FFTContext m1, *m = &m1;
yading@10: #if CONFIG_FFT_FLOAT
yading@10:     RDFTContext r1, *r = &r1;
yading@10:     DCTContext d1, *d = &d1;
yading@10:     int fft_size_2;
yading@10: #endif
yading@10:     int fft_nbits, fft_size;
yading@10:     double scale = 1.0;
yading@10:     AVLFG prng;
yading@10:     av_lfg_init(&prng, 1);
yading@10: 
yading@10:     fft_nbits = 9;
yading@10:     for(;;) {
yading@10:         c = getopt(argc, argv, "hsimrdn:f:c:");
yading@10:         if (c == -1)
yading@10:             break;
yading@10:         switch(c) {
yading@10:         case 'h':
yading@10:             help();
yading@10:             return 1;
yading@10:         case 's':
yading@10:             do_speed = 1;
yading@10:             break;
yading@10:         case 'i':
yading@10:             do_inverse = 1;
yading@10:             break;
yading@10:         case 'm':
yading@10:             transform = TRANSFORM_MDCT;
yading@10:             break;
yading@10:         case 'r':
yading@10:             transform = TRANSFORM_RDFT;
yading@10:             break;
yading@10:         case 'd':
yading@10:             transform = TRANSFORM_DCT;
yading@10:             break;
yading@10:         case 'n':
yading@10:             fft_nbits = atoi(optarg);
yading@10:             break;
yading@10:         case 'f':
yading@10:             scale = atof(optarg);
yading@10:             break;
yading@10:         case 'c':
yading@10:             cpuflags = av_get_cpu_flags();
yading@10: 
yading@10:             if (av_parse_cpu_caps(&cpuflags, optarg) < 0)
yading@10:                 return 1;
yading@10: 
yading@10:             av_force_cpu_flags(cpuflags);
yading@10:             break;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     fft_size = 1 << fft_nbits;
yading@10:     tab = av_malloc(fft_size * sizeof(FFTComplex));
yading@10:     tab1 = av_malloc(fft_size * sizeof(FFTComplex));
yading@10:     tab_ref = av_malloc(fft_size * sizeof(FFTComplex));
yading@10:     tab2 = av_malloc(fft_size * sizeof(FFTSample));
yading@10: 
yading@10:     switch (transform) {
yading@10:     case TRANSFORM_MDCT:
yading@10:         av_log(NULL, AV_LOG_INFO,"Scale factor is set to %f\n", scale);
yading@10:         if (do_inverse)
yading@10:             av_log(NULL, AV_LOG_INFO,"IMDCT");
yading@10:         else
yading@10:             av_log(NULL, AV_LOG_INFO,"MDCT");
yading@10:         ff_mdct_init(m, fft_nbits, do_inverse, scale);
yading@10:         break;
yading@10:     case TRANSFORM_FFT:
yading@10:         if (do_inverse)
yading@10:             av_log(NULL, AV_LOG_INFO,"IFFT");
yading@10:         else
yading@10:             av_log(NULL, AV_LOG_INFO,"FFT");
yading@10:         ff_fft_init(s, fft_nbits, do_inverse);
yading@10:         fft_ref_init(fft_nbits, do_inverse);
yading@10:         break;
yading@10: #if CONFIG_FFT_FLOAT
yading@10:     case TRANSFORM_RDFT:
yading@10:         if (do_inverse)
yading@10:             av_log(NULL, AV_LOG_INFO,"IDFT_C2R");
yading@10:         else
yading@10:             av_log(NULL, AV_LOG_INFO,"DFT_R2C");
yading@10:         ff_rdft_init(r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C);
yading@10:         fft_ref_init(fft_nbits, do_inverse);
yading@10:         break;
yading@10:     case TRANSFORM_DCT:
yading@10:         if (do_inverse)
yading@10:             av_log(NULL, AV_LOG_INFO,"DCT_III");
yading@10:         else
yading@10:             av_log(NULL, AV_LOG_INFO,"DCT_II");
yading@10:         ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II);
yading@10:         break;
yading@10: #endif
yading@10:     default:
yading@10:         av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n");
yading@10:         return 1;
yading@10:     }
yading@10:     av_log(NULL, AV_LOG_INFO," %d test\n", fft_size);
yading@10: 
yading@10:     /* generate random data */
yading@10: 
yading@10:     for (i = 0; i < fft_size; i++) {
yading@10:         tab1[i].re = frandom(&prng);
yading@10:         tab1[i].im = frandom(&prng);
yading@10:     }
yading@10: 
yading@10:     /* checking result */
yading@10:     av_log(NULL, AV_LOG_INFO,"Checking...\n");
yading@10: 
yading@10:     switch (transform) {
yading@10:     case TRANSFORM_MDCT:
yading@10:         if (do_inverse) {
yading@10:             imdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
yading@10:             m->imdct_calc(m, tab2, (FFTSample *)tab1);
yading@10:             err = check_diff((FFTSample *)tab_ref, tab2, fft_size, scale);
yading@10:         } else {
yading@10:             mdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
yading@10: 
yading@10:             m->mdct_calc(m, tab2, (FFTSample *)tab1);
yading@10: 
yading@10:             err = check_diff((FFTSample *)tab_ref, tab2, fft_size / 2, scale);
yading@10:         }
yading@10:         break;
yading@10:     case TRANSFORM_FFT:
yading@10:         memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
yading@10:         s->fft_permute(s, tab);
yading@10:         s->fft_calc(s, tab);
yading@10: 
yading@10:         fft_ref(tab_ref, tab1, fft_nbits);
yading@10:         err = check_diff((FFTSample *)tab_ref, (FFTSample *)tab, fft_size * 2, 1.0);
yading@10:         break;
yading@10: #if CONFIG_FFT_FLOAT
yading@10:     case TRANSFORM_RDFT:
yading@10:         fft_size_2 = fft_size >> 1;
yading@10:         if (do_inverse) {
yading@10:             tab1[         0].im = 0;
yading@10:             tab1[fft_size_2].im = 0;
yading@10:             for (i = 1; i < fft_size_2; i++) {
yading@10:                 tab1[fft_size_2+i].re =  tab1[fft_size_2-i].re;
yading@10:                 tab1[fft_size_2+i].im = -tab1[fft_size_2-i].im;
yading@10:             }
yading@10: 
yading@10:             memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
yading@10:             tab2[1] = tab1[fft_size_2].re;
yading@10: 
yading@10:             r->rdft_calc(r, tab2);
yading@10:             fft_ref(tab_ref, tab1, fft_nbits);
yading@10:             for (i = 0; i < fft_size; i++) {
yading@10:                 tab[i].re = tab2[i];
yading@10:                 tab[i].im = 0;
yading@10:             }
yading@10:             err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 0.5);
yading@10:         } else {
yading@10:             for (i = 0; i < fft_size; i++) {
yading@10:                 tab2[i]    = tab1[i].re;
yading@10:                 tab1[i].im = 0;
yading@10:             }
yading@10:             r->rdft_calc(r, tab2);
yading@10:             fft_ref(tab_ref, tab1, fft_nbits);
yading@10:             tab_ref[0].im = tab_ref[fft_size_2].re;
yading@10:             err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0);
yading@10:         }
yading@10:         break;
yading@10:     case TRANSFORM_DCT:
yading@10:         memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
yading@10:         d->dct_calc(d, (FFTSample *)tab);
yading@10:         if (do_inverse) {
yading@10:             idct_ref((FFTSample*)tab_ref, (FFTSample *)tab1, fft_nbits);
yading@10:         } else {
yading@10:             dct_ref((FFTSample*)tab_ref, (FFTSample *)tab1, fft_nbits);
yading@10:         }
yading@10:         err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0);
yading@10:         break;
yading@10: #endif
yading@10:     }
yading@10: 
yading@10:     /* do a speed test */
yading@10: 
yading@10:     if (do_speed) {
yading@10:         int64_t time_start, duration;
yading@10:         int nb_its;
yading@10: 
yading@10:         av_log(NULL, AV_LOG_INFO,"Speed test...\n");
yading@10:         /* we measure during about 1 seconds */
yading@10:         nb_its = 1;
yading@10:         for(;;) {
yading@10:             time_start = av_gettime();
yading@10:             for (it = 0; it < nb_its; it++) {
yading@10:                 switch (transform) {
yading@10:                 case TRANSFORM_MDCT:
yading@10:                     if (do_inverse) {
yading@10:                         m->imdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
yading@10:                     } else {
yading@10:                         m->mdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
yading@10:                     }
yading@10:                     break;
yading@10:                 case TRANSFORM_FFT:
yading@10:                     memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
yading@10:                     s->fft_calc(s, tab);
yading@10:                     break;
yading@10: #if CONFIG_FFT_FLOAT
yading@10:                 case TRANSFORM_RDFT:
yading@10:                     memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
yading@10:                     r->rdft_calc(r, tab2);
yading@10:                     break;
yading@10:                 case TRANSFORM_DCT:
yading@10:                     memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
yading@10:                     d->dct_calc(d, tab2);
yading@10:                     break;
yading@10: #endif
yading@10:                 }
yading@10:             }
yading@10:             duration = av_gettime() - time_start;
yading@10:             if (duration >= 1000000)
yading@10:                 break;
yading@10:             nb_its *= 2;
yading@10:         }
yading@10:         av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
yading@10:                (double)duration / nb_its,
yading@10:                (double)duration / 1000000.0,
yading@10:                nb_its);
yading@10:     }
yading@10: 
yading@10:     switch (transform) {
yading@10:     case TRANSFORM_MDCT:
yading@10:         ff_mdct_end(m);
yading@10:         break;
yading@10:     case TRANSFORM_FFT:
yading@10:         ff_fft_end(s);
yading@10:         break;
yading@10: #if CONFIG_FFT_FLOAT
yading@10:     case TRANSFORM_RDFT:
yading@10:         ff_rdft_end(r);
yading@10:         break;
yading@10:     case TRANSFORM_DCT:
yading@10:         ff_dct_end(d);
yading@10:         break;
yading@10: #endif
yading@10:     }
yading@10: 
yading@10:     av_free(tab);
yading@10:     av_free(tab1);
yading@10:     av_free(tab2);
yading@10:     av_free(tab_ref);
yading@10:     av_free(exptab);
yading@10: 
yading@10:     return err;
yading@10: }