X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;ds=inline;f=libavcodec%2Ffft-test.c;h=83b5546f0826e98594a82776898889fb70fc79cd;hb=7ca3e5203f133eb41a0b5c3a1d753a7427ba72e7;hp=0313154ecf4a6098077dfde0c136ba653bc544ee;hpb=2912e87a6c9264d556734e2bf94a99c64cf9b102;p=ffmpeg

diff --git a/libavcodec/fft-test.c b/libavcodec/fft-test.c
index 0313154ecf4..83b5546f082 100644
--- a/libavcodec/fft-test.c
+++ b/libavcodec/fft-test.c
@@ -23,63 +23,83 @@
  * FFT and MDCT tests.
  */
 
-#include "libavutil/mathematics.h"
-#include "libavutil/lfg.h"
-#include "libavutil/log.h"
-#include "fft.h"
+#include "config.h"
+
 #include <math.h>
+#if HAVE_UNISTD_H
 #include <unistd.h>
-#include <sys/time.h>
+#endif
+#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
-#undef exit
+#include "libavutil/cpu.h"
+#include "libavutil/lfg.h"
+#include "libavutil/log.h"
+#include "libavutil/mathematics.h"
+#include "libavutil/time.h"
 
-/* reference fft */
+#include "fft.h"
+#if FFT_FLOAT
+#include "dct.h"
+#include "rdft.h"
+#endif
 
-#define MUL16(a,b) ((a) * (b))
+/* reference fft */
 
-#define CMAC(pre, pim, are, aim, bre, bim) \
-{\
-   pre += (MUL16(are, bre) - MUL16(aim, bim));\
-   pim += (MUL16(are, bim) + MUL16(bre, aim));\
-}
+#define MUL16(a, b) ((a) * (b))
 
-FFTComplex *exptab;
+#define CMAC(pre, pim, are, aim, bre, bim)          \
+    {                                               \
+        pre += (MUL16(are, bre) - MUL16(aim, bim)); \
+        pim += (MUL16(are, bim) + MUL16(bre, aim)); \
+    }
 
-static void fft_ref_init(int nbits, int inverse)
+#if FFT_FLOAT
+#define RANGE 1.0
+#define REF_SCALE(x, bits)  (x)
+#define FMT "%10.6f"
+#else
+#define RANGE 16384
+#define REF_SCALE(x, bits) ((x) / (1 << (bits)))
+#define FMT "%6d"
+#endif
+
+static struct {
+    float re, im;
+} *exptab;
+
+static int fft_ref_init(int nbits, int inverse)
 {
-    int n, i;
-    double c1, s1, alpha;
+    int i, n = 1 << nbits;
 
-    n = 1 << nbits;
-    exptab = av_malloc((n / 2) * sizeof(FFTComplex));
+    exptab = av_malloc((n / 2) * sizeof(*exptab));
+    if (!exptab)
+        return AVERROR(ENOMEM);
 
-    for (i = 0; i < (n/2); i++) {
-        alpha = 2 * M_PI * (float)i / (float)n;
-        c1 = cos(alpha);
-        s1 = sin(alpha);
+    for (i = 0; i < (n / 2); i++) {
+        double alpha = 2 * M_PI * (float) i / (float) n;
+        double c1 = cos(alpha), s1 = sin(alpha);
         if (!inverse)
             s1 = -s1;
         exptab[i].re = c1;
         exptab[i].im = s1;
     }
+    return 0;
 }
 
 static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
 {
-    int n, i, j, k, n2;
-    double tmp_re, tmp_im, s, c;
-    FFTComplex *q;
+    int i, j;
+    int n  = 1 << nbits;
+    int n2 = n >> 1;
 
-    n = 1 << nbits;
-    n2 = n >> 1;
     for (i = 0; i < n; i++) {
-        tmp_re = 0;
-        tmp_im = 0;
-        q = tab;
+        double tmp_re = 0, tmp_im = 0;
+        FFTComplex *q = tab;
         for (j = 0; j < n; j++) {
-            k = (i * j) & (n - 1);
+            double s, c;
+            int k = (i * j) & (n - 1);
             if (k >= n2) {
                 c = -exptab[k - n2].re;
                 s = -exptab[k - n2].im;
@@ -90,117 +110,107 @@ static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
             CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
             q++;
         }
-        tabr[i].re = tmp_re;
-        tabr[i].im = tmp_im;
+        tabr[i].re = REF_SCALE(tmp_re, nbits);
+        tabr[i].im = REF_SCALE(tmp_im, nbits);
     }
 }
 
-static void imdct_ref(float *out, float *in, int nbits)
+#if CONFIG_MDCT
+static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
 {
-    int n = 1<<nbits;
-    int k, i, a;
-    double sum, f;
+    int i, k, n = 1 << nbits;
 
     for (i = 0; i < n; i++) {
-        sum = 0;
-        for (k = 0; k < n/2; k++) {
-            a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
-            f = cos(M_PI * a / (double)(2 * n));
+        double sum = 0;
+        for (k = 0; k < n / 2; k++) {
+            int a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
+            double f = cos(M_PI * a / (double) (2 * n));
             sum += f * in[k];
         }
-        out[i] = -sum;
+        out[i] = REF_SCALE(-sum, nbits - 2);
     }
 }
 
 /* NOTE: no normalisation by 1 / N is done */
-static void mdct_ref(float *output, float *input, int nbits)
+static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
 {
-    int n = 1<<nbits;
-    int k, i;
-    double a, s;
+    int i, k, n = 1 << nbits;
 
     /* do it by hand */
-    for (k = 0; k < n/2; k++) {
-        s = 0;
+    for (k = 0; k < n / 2; k++) {
+        double s = 0;
         for (i = 0; i < n; i++) {
-            a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n));
+            double a = (2 * M_PI * (2 * i + 1 + n / 2) * (2 * k + 1) / (4 * n));
             s += input[i] * cos(a);
         }
-        output[k] = s;
+        output[k] = REF_SCALE(s, nbits - 1);
     }
 }
+#endif /* CONFIG_MDCT */
 
+#if FFT_FLOAT
+#if CONFIG_DCT
 static void idct_ref(float *output, float *input, int nbits)
 {
-    int n = 1<<nbits;
-    int k, i;
-    double a, s;
+    int i, k, n = 1 << nbits;
 
     /* do it by hand */
     for (i = 0; i < n; i++) {
-        s = 0.5 * input[0];
+        double s = 0.5 * input[0];
         for (k = 1; k < n; k++) {
-            a = M_PI*k*(i+0.5) / n;
+            double a = M_PI * k * (i + 0.5) / n;
             s += input[k] * cos(a);
         }
         output[i] = 2 * s / n;
     }
 }
+
 static void dct_ref(float *output, float *input, int nbits)
 {
-    int n = 1<<nbits;
-    int k, i;
-    double a, s;
+    int i, k, n = 1 << nbits;
 
     /* do it by hand */
     for (k = 0; k < n; k++) {
-        s = 0;
+        double s = 0;
         for (i = 0; i < n; i++) {
-            a = M_PI*k*(i+0.5) / n;
+            double a = M_PI * k * (i + 0.5) / n;
             s += input[i] * cos(a);
         }
         output[k] = s;
     }
 }
+#endif /* CONFIG_DCT */
+#endif /* FFT_FLOAT */
 
-
-static float frandom(AVLFG *prng)
-{
-    return (int16_t)av_lfg_get(prng) / 32768.0;
-}
-
-static int64_t gettime(void)
+static FFTSample frandom(AVLFG *prng)
 {
-    struct timeval tv;
-    gettimeofday(&tv,NULL);
-    return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
+    return (int16_t) av_lfg_get(prng) / 32768.0 * RANGE;
 }
 
-static int check_diff(float *tab1, float *tab2, int n, double scale)
+static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
 {
-    int i;
-    double max= 0;
-    double error= 0;
-    int err = 0;
+    int i, err = 0;
+    double error = 0, max = 0;
 
     for (i = 0; i < n; i++) {
-        double e= fabsf(tab1[i] - (tab2[i] / scale));
+        double e = fabs(tab1[i] - (tab2[i] / scale)) / RANGE;
         if (e >= 1e-3) {
-            av_log(NULL, AV_LOG_ERROR, "ERROR %5d: %10.6f %10.6f\n",
+            av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n",
                    i, tab1[i], tab2[i]);
             err = 1;
         }
-        error+= e*e;
-        if(e>max) max= e;
+        error += e * e;
+        if (e > max)
+            max = e;
     }
-    av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error)/n);
+    av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error) / n);
     return err;
 }
 
-
 static void help(void)
 {
-    av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n"
+    av_log(NULL, AV_LOG_INFO,
+           "usage: fft-test [-h] [-s] [-i] [-n b]\n"
            "-h     print this help\n"
            "-s     speed test\n"
            "-m     (I)MDCT test\n"
@@ -208,9 +218,7 @@ static void help(void)
            "-r     (I)RDFT test\n"
            "-i     inverse transform test\n"
            "-n b   set the transform size to 2^b\n"
-           "-f x   set scale factor for output data of (I)MDCT to x\n"
-           );
-    exit(1);
+           "-f x   set scale factor for output data of (I)MDCT to x\n");
 }
 
 enum tf_transform {
@@ -220,33 +228,36 @@ enum tf_transform {
     TRANSFORM_DCT,
 };
 
+#if !HAVE_GETOPT
+#include "compat/getopt.c"
+#endif
+
 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;
+    FFTContext m, s;
+#if FFT_FLOAT
+    RDFTContext r;
+    DCTContext d;
+#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);
 
-    fft_nbits = 9;
-    for(;;) {
-        c = getopt(argc, argv, "hsimrdn:f:");
+    for (;;) {
+        int c = getopt(argc, argv, "hsimrdn:f:c:");
         if (c == -1)
             break;
-        switch(c) {
+        switch (c) {
         case 'h':
             help();
-            break;
+            return 1;
         case 's':
             do_speed = 1;
             break;
@@ -268,50 +279,73 @@ int main(int argc, char **argv)
         case 'f':
             scale = atof(optarg);
             break;
+        case 'c':
+        {
+            int cpuflags = av_parse_cpu_flags(optarg);
+            if (cpuflags < 0)
+                return 1;
+            av_set_cpu_flags_mask(cpuflags);
+            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));
+    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 (!(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);
+        av_log(NULL, AV_LOG_INFO, "Scale factor is set to %f\n", scale);
         if (do_inverse)
-            av_log(NULL, AV_LOG_INFO,"IMDCT");
+            av_log(NULL, AV_LOG_INFO, "IMDCT");
         else
-            av_log(NULL, AV_LOG_INFO,"MDCT");
-        ff_mdct_init(m, fft_nbits, do_inverse, scale);
+            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");
+            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, "FFT");
+        ff_fft_init(&s, fft_nbits, do_inverse);
+        if (err = fft_ref_init(fft_nbits, do_inverse) < 0)
+            goto cleanup;
         break;
+#if FFT_FLOAT
+#if CONFIG_RDFT
     case TRANSFORM_RDFT:
         if (do_inverse)
-            av_log(NULL, AV_LOG_INFO,"IDFT_C2R");
+            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);
+            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");
+            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);
+            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);
+    av_log(NULL, AV_LOG_INFO, " %d test\n", fft_size);
 
     /* generate random data */
 
@@ -321,70 +355,78 @@ int main(int argc, char **argv)
     }
 
     /* checking result */
-    av_log(NULL, AV_LOG_INFO,"Checking...\n");
+    av_log(NULL, AV_LOG_INFO, "Checking...\n");
 
     switch (transform) {
+#if CONFIG_MDCT
     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);
+            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((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);
+            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));
-        ff_fft_permute(s, tab);
-        ff_fft_calc(s, tab);
+        s.fft_permute(&s, tab);
+        s.fft_calc(&s, tab);
 
         fft_ref(tab_ref, tab1, fft_nbits);
-        err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 1.0);
+        err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 1.0);
         break;
+#if FFT_FLOAT
+#if CONFIG_RDFT
     case TRANSFORM_RDFT:
+    {
+        int fft_size_2 = fft_size >> 1;
         if (do_inverse) {
-            tab1[         0].im = 0;
+            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;
+                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);
+            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((float *)tab_ref, (float *)tab, fft_size * 2, 0.5);
+            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;
             }
-            ff_rdft_calc(r, tab2);
+            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((float *)tab_ref, (float *)tab2, fft_size, 1.0);
+            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));
-        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);
+        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 */
@@ -393,65 +435,79 @@ int main(int argc, char **argv)
         int64_t time_start, duration;
         int nb_its;
 
-        av_log(NULL, AV_LOG_INFO,"Speed test...\n");
+        av_log(NULL, AV_LOG_INFO, "Speed test...\n");
         /* we measure during about 1 seconds */
         nb_its = 1;
-        for(;;) {
-            time_start = gettime();
+        for (;;) {
+            time_start = av_gettime_relative();
             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);
-                    }
+                    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));
-                    ff_fft_calc(s, tab);
+                    s.fft_calc(&s, tab);
                     break;
+#if FFT_FLOAT
                 case TRANSFORM_RDFT:
                     memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
-                    ff_rdft_calc(r, tab2);
+                    r.rdft_calc(&r, tab2);
                     break;
                 case TRANSFORM_DCT:
                     memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
-                    ff_dct_calc(d, tab2);
+                    d.dct_calc(&d, tab2);
                     break;
+#endif /* FFT_FLOAT */
                 }
             }
-            duration = gettime() - time_start;
+            duration = av_gettime_relative() - 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,
+        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) {
+#if CONFIG_MDCT
     case TRANSFORM_MDCT:
-        ff_mdct_end(m);
+        ff_mdct_end(&m);
         break;
+#endif /* CONFIG_MDCT */
     case TRANSFORM_FFT:
-        ff_fft_end(s);
+        ff_fft_end(&s);
         break;
+#if FFT_FLOAT
+#if CONFIG_RDFT
     case TRANSFORM_RDFT:
-        ff_rdft_end(r);
+        ff_rdft_end(&r);
         break;
+#endif /* CONFIG_RDFT */
+#if CONFIG_DCT
     case TRANSFORM_DCT:
-        ff_dct_end(d);
+        ff_dct_end(&d);
         break;
+#endif /* CONFIG_DCT */
+#endif /* FFT_FLOAT */
     }
 
+cleanup:
     av_free(tab);
     av_free(tab1);
     av_free(tab2);
     av_free(tab_ref);
     av_free(exptab);
 
-    return err;
+    if (err)
+        printf("Error: %d.\n", err);
+
+    return !!err;
 }