int main(int argc, char **argv) { FFTComplex *tab, *tab1, *tab_ref; FFTSample *tab2; int it, i, c; int do_speed = 0; int err = 1; enum tf_transform transform = TRANSFORM_FFT; int do_inverse = 0; FFTContext s1, *s = &s1; FFTContext m1, *m = &m1; RDFTContext r1, *r = &r1; DCTContext d1, *d = &d1; int fft_nbits, fft_size, fft_size_2; double scale = 1.0; AVLFG prng; av_lfg_init(&prng, 1); fft_nbits = 9; for(;;) { c = getopt(argc, argv, "hsimrdn:f:"); if (c == -1) break; switch(c) { case 'h': help(); break; case 's': do_speed = 1; break; case 'i': do_inverse = 1; break; case 'm': transform = TRANSFORM_MDCT; break; case 'r': transform = TRANSFORM_RDFT; break; case 'd': transform = TRANSFORM_DCT; break; case 'n': fft_nbits = atoi(optarg); break; case 'f': scale = atof(optarg); break; } } fft_size = 1 << fft_nbits; fft_size_2 = fft_size >> 1; tab = av_malloc(fft_size * sizeof(FFTComplex)); tab1 = av_malloc(fft_size * sizeof(FFTComplex)); tab_ref = av_malloc(fft_size * sizeof(FFTComplex)); tab2 = av_malloc(fft_size * sizeof(FFTSample)); switch (transform) { case TRANSFORM_MDCT: av_log(NULL, AV_LOG_INFO,"Scale factor is set to %f\n", scale); if (do_inverse) av_log(NULL, AV_LOG_INFO,"IMDCT"); else av_log(NULL, AV_LOG_INFO,"MDCT"); ff_mdct_init(m, fft_nbits, do_inverse, scale); break; case TRANSFORM_FFT: if (do_inverse) av_log(NULL, AV_LOG_INFO,"IFFT"); else av_log(NULL, AV_LOG_INFO,"FFT"); ff_fft_init(s, fft_nbits, do_inverse); fft_ref_init(fft_nbits, do_inverse); break; case TRANSFORM_RDFT: if (do_inverse) av_log(NULL, AV_LOG_INFO,"IDFT_C2R"); else av_log(NULL, AV_LOG_INFO,"DFT_R2C"); ff_rdft_init(r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C); fft_ref_init(fft_nbits, do_inverse); break; case TRANSFORM_DCT: if (do_inverse) av_log(NULL, AV_LOG_INFO,"DCT_III"); else av_log(NULL, AV_LOG_INFO,"DCT_II"); ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II); break; } av_log(NULL, AV_LOG_INFO," %d test\n", fft_size); /* generate random data */ for (i = 0; i < fft_size; i++) { tab1[i].re = frandom(&prng); tab1[i].im = frandom(&prng); } /* checking result */ av_log(NULL, AV_LOG_INFO,"Checking...\n"); switch (transform) { case TRANSFORM_MDCT: if (do_inverse) { imdct_ref((float *)tab_ref, (float *)tab1, fft_nbits); ff_imdct_calc(m, tab2, (float *)tab1); err = check_diff((float *)tab_ref, tab2, fft_size, scale); } else { mdct_ref((float *)tab_ref, (float *)tab1, fft_nbits); ff_mdct_calc(m, tab2, (float *)tab1); err = check_diff((float *)tab_ref, tab2, fft_size / 2, scale); } break; case TRANSFORM_FFT: memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); ff_fft_permute(s, tab); ff_fft_calc(s, tab); fft_ref(tab_ref, tab1, fft_nbits); err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 1.0); break; case TRANSFORM_RDFT: if (do_inverse) { tab1[ 0].im = 0; tab1[fft_size_2].im = 0; for (i = 1; i < fft_size_2; i++) { tab1[fft_size_2+i].re = tab1[fft_size_2-i].re; tab1[fft_size_2+i].im = -tab1[fft_size_2-i].im; } memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); tab2[1] = tab1[fft_size_2].re; ff_rdft_calc(r, tab2); fft_ref(tab_ref, tab1, fft_nbits); for (i = 0; i < fft_size; i++) { tab[i].re = tab2[i]; tab[i].im = 0; } err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 0.5); } else { for (i = 0; i < fft_size; i++) { tab2[i] = tab1[i].re; tab1[i].im = 0; } ff_rdft_calc(r, tab2); fft_ref(tab_ref, tab1, fft_nbits); tab_ref[0].im = tab_ref[fft_size_2].re; err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0); } break; case TRANSFORM_DCT: memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); ff_dct_calc(d, tab); if (do_inverse) { idct_ref(tab_ref, tab1, fft_nbits); } else { dct_ref(tab_ref, tab1, fft_nbits); } err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0); break; } /* do a speed test */ if (do_speed) { int64_t time_start, duration; int nb_its; av_log(NULL, AV_LOG_INFO,"Speed test...\n"); /* we measure during about 1 seconds */ nb_its = 1; for(;;) { time_start = gettime(); for (it = 0; it < nb_its; it++) { switch (transform) { case TRANSFORM_MDCT: if (do_inverse) { ff_imdct_calc(m, (float *)tab, (float *)tab1); } else { ff_mdct_calc(m, (float *)tab, (float *)tab1); } break; case TRANSFORM_FFT: memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); ff_fft_calc(s, tab); break; case TRANSFORM_RDFT: memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); ff_rdft_calc(r, tab2); break; case TRANSFORM_DCT: memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); ff_dct_calc(d, tab2); break; } } duration = gettime() - time_start; if (duration >= 1000000) break; nb_its *= 2; } av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n", (double)duration / nb_its, (double)duration / 1000000.0, nb_its); } switch (transform) { case TRANSFORM_MDCT: ff_mdct_end(m); break; case TRANSFORM_FFT: ff_fft_end(s); break; case TRANSFORM_RDFT: ff_rdft_end(r); break; case TRANSFORM_DCT: ff_dct_end(d); break; } av_free(tab); av_free(tab1); av_free(tab2); av_free(tab_ref); av_free(exptab); return err; }
int main(int argc, char **argv) { FFTComplex *tab, *tab1, *tab_ref; FFTSample *tab2; int it, i, c; int do_speed = 0; int do_mdct = 0; int do_inverse = 0; FFTContext s1, *s = &s1; MDCTContext m1, *m = &m1; int fft_nbits, fft_size; fft_nbits = 9; for(;;) { c = getopt(argc, argv, "hsimn:"); if (c == -1) break; switch(c) { case 'h': help(); break; case 's': do_speed = 1; break; case 'i': do_inverse = 1; break; case 'm': do_mdct = 1; break; case 'n': fft_nbits = atoi(optarg); break; } } fft_size = 1 << fft_nbits; tab = av_malloc(fft_size * sizeof(FFTComplex)); tab1 = av_malloc(fft_size * sizeof(FFTComplex)); tab_ref = av_malloc(fft_size * sizeof(FFTComplex)); tab2 = av_malloc(fft_size * sizeof(FFTSample)); if (do_mdct) { if (do_inverse) av_log(NULL, AV_LOG_INFO,"IMDCT"); else av_log(NULL, AV_LOG_INFO,"MDCT"); ff_mdct_init(m, fft_nbits, do_inverse); } else { if (do_inverse) av_log(NULL, AV_LOG_INFO,"IFFT"); else av_log(NULL, AV_LOG_INFO,"FFT"); ff_fft_init(s, fft_nbits, do_inverse); fft_ref_init(fft_nbits, do_inverse); } av_log(NULL, AV_LOG_INFO," %d test\n", fft_size); /* generate random data */ for(i=0;i<fft_size;i++) { tab1[i].re = frandom(); tab1[i].im = frandom(); } /* checking result */ av_log(NULL, AV_LOG_INFO,"Checking...\n"); if (do_mdct) { if (do_inverse) { imdct_ref((float *)tab_ref, (float *)tab1, fft_nbits); ff_imdct_calc(m, tab2, (float *)tab1); check_diff((float *)tab_ref, tab2, fft_size); } else { mdct_ref((float *)tab_ref, (float *)tab1, fft_nbits); ff_mdct_calc(m, tab2, (float *)tab1); check_diff((float *)tab_ref, tab2, fft_size / 2); } } else { memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); ff_fft_permute(s, tab); ff_fft_calc(s, tab); fft_ref(tab_ref, tab1, fft_nbits); check_diff((float *)tab_ref, (float *)tab, fft_size * 2); } /* do a speed test */ if (do_speed) { int64_t time_start, duration; int nb_its; av_log(NULL, AV_LOG_INFO,"Speed test...\n"); /* we measure during about 1 seconds */ nb_its = 1; for(;;) { time_start = gettime(); for(it=0;it<nb_its;it++) { if (do_mdct) { if (do_inverse) { ff_imdct_calc(m, (float *)tab, (float *)tab1); } else { ff_mdct_calc(m, (float *)tab, (float *)tab1); } } else { memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); ff_fft_calc(s, tab); } } duration = gettime() - time_start; if (duration >= 1000000) break; nb_its *= 2; } av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n", (double)duration / nb_its, (double)duration / 1000000.0, nb_its); } if (do_mdct) { ff_mdct_end(m); } else { ff_fft_end(s); } return 0; }
int main(int argc, char **argv) { FFTComplex *tab, *tab1, *tab_ref; FFTSample *tab2; enum tf_transform transform = TRANSFORM_FFT; FFTContext m, s; #if FFT_FLOAT RDFTContext r; DCTContext d; #if CONFIG_LIBFFTW3 FFTWContext fftw; FFTWComplex *tab_fftw, *tab_fftw_copy; #endif /* CONFIG_LIBFFTW3 */ #endif /* FFT_FLOAT */ int it, i, err = 1; int do_speed = 0, do_inverse = 0; int fft_nbits = 9, fft_size; double scale = 1.0; AVLFG prng; av_lfg_init(&prng, 1); for (;;) { int c = getopt(argc, argv, "hsitmrdn:f:c:"); if (c == -1) break; switch (c) { case 'h': help(); return 1; case 's': do_speed = 1; break; case 'i': do_inverse = 1; break; #if CONFIG_LIBFFTW3 case 't': transform = TRANSFORM_FFTW; break; #endif /* CONFIG_LIBFFTW3 */ case 'm': transform = TRANSFORM_MDCT; break; case 'r': transform = TRANSFORM_RDFT; break; case 'd': transform = TRANSFORM_DCT; break; case 'n': fft_nbits = atoi(optarg); break; case 'f': scale = atof(optarg); break; case 'c': { unsigned cpuflags = av_get_cpu_flags(); if (av_parse_cpu_caps(&cpuflags, optarg) < 0) return 1; av_force_cpu_flags(cpuflags); break; } } } fft_size = 1 << fft_nbits; tab = av_malloc_array(fft_size, sizeof(FFTComplex)); tab1 = av_malloc_array(fft_size, sizeof(FFTComplex)); tab_ref = av_malloc_array(fft_size, sizeof(FFTComplex)); tab2 = av_malloc_array(fft_size, sizeof(FFTSample)); #if CONFIG_LIBFFTW3 && FFT_FLOAT tab_fftw = av_malloc_array(fft_size, sizeof(*tab_fftw)); tab_fftw_copy = av_malloc_array(fft_size, sizeof(*tab_fftw_copy)); if (!(tab_fftw && tab_fftw_copy)) goto cleanup_fftw; #endif /* CONFIG_LIBFFTW3 */ if (!(tab && tab1 && tab_ref && tab2)) goto cleanup; switch (transform) { #if CONFIG_MDCT case TRANSFORM_MDCT: av_log(NULL, AV_LOG_INFO, "Scale factor is set to %f\n", scale); if (do_inverse) av_log(NULL, AV_LOG_INFO, "IMDCT"); else av_log(NULL, AV_LOG_INFO, "MDCT"); ff_mdct_init(&m, fft_nbits, do_inverse, scale); break; #endif /* CONFIG_MDCT */ case TRANSFORM_FFT: if (do_inverse) av_log(NULL, AV_LOG_INFO, "IFFT"); else av_log(NULL, AV_LOG_INFO, "FFT"); ff_fft_init(&s, fft_nbits, do_inverse); if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0) goto cleanup; break; #if CONFIG_LIBFFTW3 && FFT_FLOAT case TRANSFORM_FFTW: if (do_inverse) av_log(NULL, AV_LOG_INFO, "IFFTW"); else av_log(NULL, AV_LOG_INFO, "FFTW"); ff_fftw_init(&fftw, fft_size, do_inverse); if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0) goto cleanup; break; #endif /* CONFIG_LIBFFTW3 */ #if FFT_FLOAT # if CONFIG_RDFT case TRANSFORM_RDFT: if (do_inverse) av_log(NULL, AV_LOG_INFO, "IDFT_C2R"); else av_log(NULL, AV_LOG_INFO, "DFT_R2C"); ff_rdft_init(&r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C); if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0) goto cleanup; break; # endif /* CONFIG_RDFT */ # if CONFIG_DCT case TRANSFORM_DCT: if (do_inverse) av_log(NULL, AV_LOG_INFO, "DCT_III"); else av_log(NULL, AV_LOG_INFO, "DCT_II"); ff_dct_init(&d, fft_nbits, do_inverse ? DCT_III : DCT_II); break; # endif /* CONFIG_DCT */ #endif /* FFT_FLOAT */ default: av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n"); goto cleanup; } av_log(NULL, AV_LOG_INFO, " %d test\n", fft_size); /* generate random data */ for (i = 0; i < fft_size; i++) { tab1[i].re = frandom(&prng); tab1[i].im = frandom(&prng); #if CONFIG_LIBFFTW3 && FFT_FLOAT tab_fftw[i][0] = tab1[i].re; tab_fftw[i][1] = tab1[i].im; #endif /* CONFIG_LIBFFTW3 */ } #if CONFIG_LIBFFTW3 && FFT_FLOAT memcpy(tab_fftw_copy, tab_fftw, fft_size * sizeof(*tab_fftw)); #endif /* checking result */ av_log(NULL, AV_LOG_INFO, "Checking...\n"); switch (transform) { #if CONFIG_MDCT case TRANSFORM_MDCT: if (do_inverse) { imdct_ref(&tab_ref->re, &tab1->re, fft_nbits); m.imdct_calc(&m, tab2, &tab1->re); err = check_diff(&tab_ref->re, tab2, fft_size, scale); } else { mdct_ref(&tab_ref->re, &tab1->re, fft_nbits); m.mdct_calc(&m, tab2, &tab1->re); err = check_diff(&tab_ref->re, tab2, fft_size / 2, scale); } break; #endif /* CONFIG_MDCT */ case TRANSFORM_FFT: memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); s.fft_permute(&s, tab); s.fft_calc(&s, tab); fft_ref(tab_ref, tab1, fft_nbits); err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 1.0); break; #if CONFIG_LIBFFTW3 && FFT_FLOAT case TRANSFORM_FFTW: fftw.fft_calc(&fftw, tab_fftw); fft_ref(tab_ref, tab1, fft_nbits); for (i = 0; i < fft_size; i++) { tab[i].re = tab_fftw[i][0]; tab[i].im = tab_fftw[i][1]; } err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 1.0); break; #endif /* CONFIG_LIBFFTW3 */ #if FFT_FLOAT #if CONFIG_RDFT case TRANSFORM_RDFT: { int fft_size_2 = fft_size >> 1; if (do_inverse) { tab1[0].im = 0; tab1[fft_size_2].im = 0; for (i = 1; i < fft_size_2; i++) { tab1[fft_size_2 + i].re = tab1[fft_size_2 - i].re; tab1[fft_size_2 + i].im = -tab1[fft_size_2 - i].im; } memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); tab2[1] = tab1[fft_size_2].re; r.rdft_calc(&r, tab2); fft_ref(tab_ref, tab1, fft_nbits); for (i = 0; i < fft_size; i++) { tab[i].re = tab2[i]; tab[i].im = 0; } err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 0.5); } else { for (i = 0; i < fft_size; i++) { tab2[i] = tab1[i].re; tab1[i].im = 0; } r.rdft_calc(&r, tab2); fft_ref(tab_ref, tab1, fft_nbits); tab_ref[0].im = tab_ref[fft_size_2].re; err = check_diff(&tab_ref->re, tab2, fft_size, 1.0); } break; } #endif /* CONFIG_RDFT */ #if CONFIG_DCT case TRANSFORM_DCT: memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); d.dct_calc(&d, &tab->re); if (do_inverse) idct_ref(&tab_ref->re, &tab1->re, fft_nbits); else dct_ref(&tab_ref->re, &tab1->re, fft_nbits); err = check_diff(&tab_ref->re, &tab->re, fft_size, 1.0); break; #endif /* CONFIG_DCT */ #endif /* FFT_FLOAT */ } /* do a speed test */ if (do_speed) { int64_t time_start, duration; int nb_its; av_log(NULL, AV_LOG_INFO, "Speed test...\n"); /* we measure during about 1 seconds */ nb_its = 1; for (;;) { time_start = av_gettime_relative(); for (it = 0; it < nb_its; it++) { switch (transform) { case TRANSFORM_MDCT: if (do_inverse) m.imdct_calc(&m, &tab->re, &tab1->re); else m.mdct_calc(&m, &tab->re, &tab1->re); break; case TRANSFORM_FFT: memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); s.fft_permute(&s, tab); s.fft_calc(&s, tab); break; #if CONFIG_LIBFFTW3 && FFT_FLOAT case TRANSFORM_FFTW: memcpy(tab_fftw, tab_fftw_copy, fft_size * sizeof(*tab_fftw)); fftw.fft_calc(&fftw, tab_fftw); break; #endif /* CONFIG_LIBFFTW3 */ #if FFT_FLOAT case TRANSFORM_RDFT: memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); r.rdft_calc(&r, tab2); break; case TRANSFORM_DCT: memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); d.dct_calc(&d, tab2); break; #endif /* FFT_FLOAT */ } } duration = av_gettime_relative() - time_start; if (duration >= 1000000) break; nb_its *= 2; } av_log(NULL, AV_LOG_INFO, "time: %0.2f us/transform [total time=%0.2f s its=%d]\n", (double) duration / nb_its, (double) duration / 1000000.0, nb_its); } switch (transform) { #if CONFIG_MDCT case TRANSFORM_MDCT: ff_mdct_end(&m); break; #endif /* CONFIG_MDCT */ case TRANSFORM_FFT: ff_fft_end(&s); break; #if CONFIG_LIBFFTW3 && FFT_FLOAT case TRANSFORM_FFTW: ff_fftw_deinit(&fftw); break; #endif /* CONFIG_LIBFFTW3 */ #if FFT_FLOAT # if CONFIG_RDFT case TRANSFORM_RDFT: ff_rdft_end(&r); break; # endif /* CONFIG_RDFT */ # if CONFIG_DCT case TRANSFORM_DCT: ff_dct_end(&d); break; # endif /* CONFIG_DCT */ #endif /* FFT_FLOAT */ } #if CONFIG_LIBFFTW3 && FFT_FLOAT cleanup_fftw: av_freep(&tab_fftw); #endif /* CONFIG_LIBFFTW3 */ cleanup: av_freep(&tab); av_freep(&tab1); av_freep(&tab2); av_freep(&tab_ref); av_freep(&exptab); if (err) printf("Error: %d.\n", err); return !!err; }