static void ff_dct_calc_I_c(DCTContext *ctx, FFTSample *data)
{
int n = 1 << ctx->nbits;
int i;
float next = -0.5f * (data[0] - data[n]);
for(i = 0; i < n/2; i++) {
float tmp1 = data[i ];
float tmp2 = data[n - i];
float s = SIN(ctx, n, 2*i);
float c = COS(ctx, n, 2*i);
c *= tmp1 - tmp2;
s *= tmp1 - tmp2;
next += c;
tmp1 = (tmp1 + tmp2) * 0.5f;
data[i ] = tmp1 - s;
data[n - i] = tmp1 + s;
}
ff_rdft_calc(&ctx->rdft, data);
data[n] = data[1];
data[1] = next;
for(i = 3; i <= n; i += 2)
data[i] = data[i - 2] - data[i];
}
static void ff_dst_calc_I_c(DCTContext *ctx, FFTSample *data)
{
int n = 1 << ctx->nbits;
int i;
data[0] = 0;
for(i = 1; i < n/2; i++) {
float tmp1 = data[i ];
float tmp2 = data[n - i];
float s = SIN(ctx, n, 2*i);
s *= tmp1 + tmp2;
tmp1 = (tmp1 - tmp2) * 0.5f;
data[i ] = s + tmp1;
data[n - i] = s - tmp1;
}
data[n/2] *= 2;
ff_rdft_calc(&ctx->rdft, data);
data[0] *= 0.5f;
for(i = 1; i < n-2; i += 2) {
data[i + 1] += data[i - 1];
data[i ] = -data[i + 2];
}
data[n-1] = 0;
}
static void ff_dct_calc_III_c(DCTContext *ctx, FFTSample *data)
{
int n = 1 << ctx->nbits;
int i;
float next = data[n - 1];
float inv_n = 1.0f / n;
for (i = n - 2; i >= 2; i -= 2) {
float val1 = data[i ];
float val2 = data[i - 1] - data[i + 1];
float c = COS(ctx, n, i);
float s = SIN(ctx, n, i);
data[i ] = c * val1 + s * val2;
data[i + 1] = s * val1 - c * val2;
}
data[1] = 2 * next;
ff_rdft_calc(&ctx->rdft, data);
for (i = 0; i < n / 2; i++) {
float tmp1 = data[i ] * inv_n;
float tmp2 = data[n - i - 1] * inv_n;
float csc = ctx->csc2[i] * (tmp1 - tmp2);
tmp1 += tmp2;
data[i ] = tmp1 + csc;
data[n - i - 1] = tmp1 - csc;
}
}
static void ff_dct_calc_II_c(DCTContext *ctx, FFTSample *data)
{
int n = 1 << ctx->nbits;
int i;
float next;
for (i=0; i < n/2; i++) {
float tmp1 = data[i ];
float tmp2 = data[n - i - 1];
float s = SIN(ctx, n, 2*i + 1);
s *= tmp1 - tmp2;
tmp1 = (tmp1 + tmp2) * 0.5f;
data[i ] = tmp1 + s;
data[n-i-1] = tmp1 - s;
}
ff_rdft_calc(&ctx->rdft, data);
next = data[1] * 0.5;
data[1] *= -1;
for (i = n - 2; i >= 0; i -= 2) {
float inr = data[i ];
float ini = data[i + 1];
float c = COS(ctx, n, i);
float s = SIN(ctx, n, i);
data[i ] = c * inr + s * ini;
data[i+1] = next;
next += s * inr - c * ini;
}
}
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;
}
void av_rdft_calc(RDFTContext *s, FFTSample *data)
{
ff_rdft_calc(s, data);
}
