/**
* @brief Mixed radix-2/4 IFFT (complex to real) of float(32-bit) data.
* @param[out] *fout point to the output buffer
* @param[in] *fin point to the input buffer
* @param[in] cfg point to the config struct
* @return none.
* The function implements a mixed radix-2/4 FFT (complex to real). The length of 2^N(N is 2, 3, 4, 5, 6 ....etc) is supported.
* Otherwise, we alloc a temp buffer(the size is same as input buffer) for storing intermedia.
* For the usage of this function, please check test/test_suite_fft_float32.c
*/
void ne10_fft_c2r_1d_float32_neon (ne10_float32_t *fout,
ne10_fft_cpx_float32_t *fin,
ne10_fft_r2c_cfg_float32_t cfg)
{
ne10_fft_cpx_float32_t * tmpbuf1 = cfg->buffer;
ne10_fft_cpx_float32_t * tmpbuf2 = cfg->buffer + cfg->ncfft;
ne10_fft_state_float32_t c2c_state;
c2c_state.nfft = cfg->ncfft;
c2c_state.factors = cfg->factors;
c2c_state.twiddles = cfg->twiddles;
c2c_state.buffer = tmpbuf2;
ne10_fft_split_c2r_1d_float32_neon (tmpbuf1, fin, cfg->super_twiddles, cfg->ncfft);
ne10_fft_c2c_1d_float32_neon ( (ne10_fft_cpx_float32_t*) fout, tmpbuf1, &c2c_state, 1);
}
void TimeOneNE10FFT(int count, int fft_log_size, float signal_value,
int signal_type) {
struct AlignedPtr* x_aligned;
struct AlignedPtr* y_aligned;
struct AlignedPtr* z_aligned;
struct ComplexFloat* x;
struct ComplexFloat* y;
OMX_FC32* z;
struct ComplexFloat* y_true;
int n;
ne10_result_t status;
ne10_fft_cfg_float32_t fft_fwd_spec;
int fft_size;
struct timeval start_time;
struct timeval end_time;
double elapsed_time;
struct SnrResult snr_forward;
struct SnrResult snr_inverse;
fft_size = 1 << fft_log_size;
x_aligned = AllocAlignedPointer(32, sizeof(*x) * fft_size);
y_aligned = AllocAlignedPointer(32, sizeof(*y) * 2 * fft_size);
z_aligned = AllocAlignedPointer(32, sizeof(*z) * 2 * fft_size);
y_true = (struct ComplexFloat*) malloc(sizeof(*y_true) * fft_size);
x = x_aligned->aligned_pointer_;
y = y_aligned->aligned_pointer_;
z = z_aligned->aligned_pointer_;
GenerateTestSignalAndFFT(x, y_true, fft_size, signal_type, signal_value, 0);
fft_fwd_spec = ne10_fft_alloc_c2c_float32(fft_size);
if (!fft_fwd_spec) {
fprintf(stderr, "NE10 FFT: Cannot initialize FFT structure for order %d\n", fft_log_size);
return;
}
if (do_forward_test) {
GetUserTime(&start_time);
for (n = 0; n < count; ++n) {
ne10_fft_c2c_1d_float32_neon((ne10_fft_cpx_float32_t *) y,
(ne10_fft_cpx_float32_t *) x,
fft_fwd_spec,
0);
}
GetUserTime(&end_time);
elapsed_time = TimeDifference(&start_time, &end_time);
CompareComplexFloat(&snr_forward, (OMX_FC32*) y, (OMX_FC32*) y_true, fft_size);
PrintResult("Forward NE10 FFT", fft_log_size, elapsed_time, count, snr_forward.complex_snr_);
if (verbose >= 255) {
printf("Input data:\n");
DumpArrayComplexFloat("x", fft_size, (OMX_FC32*) x);
printf("FFT Actual:\n");
DumpArrayComplexFloat("y", fft_size, (OMX_FC32*) y);
printf("FFT Expected:\n");
DumpArrayComplexFloat("true", fft_size, (OMX_FC32*) y_true);
}
}
if (do_inverse_test) {
GetUserTime(&start_time);
for (n = 0; n < count; ++n) {
ne10_fft_c2c_1d_float32_neon((ne10_fft_cpx_float32_t *) z,
(ne10_fft_cpx_float32_t *) y_true,
fft_fwd_spec,
1);
}
GetUserTime(&end_time);
elapsed_time = TimeDifference(&start_time, &end_time);
CompareComplexFloat(&snr_inverse, (OMX_FC32*) z, (OMX_FC32*) x, fft_size);
PrintResult("Inverse NE10 FFT", fft_log_size, elapsed_time, count, snr_inverse.complex_snr_);
if (verbose >= 255) {
printf("Input data:\n");
DumpArrayComplexFloat("y", fft_size, (OMX_FC32*) y_true);
printf("IFFT Actual:\n");
DumpArrayComplexFloat("z", fft_size, z);
printf("IFFT Expected:\n");
DumpArrayComplexFloat("x", fft_size, (OMX_FC32*) x);
}
}
ne10_fft_destroy_c2c_float32(fft_fwd_spec);
FreeAlignedPointer(x_aligned);
FreeAlignedPointer(y_aligned);
FreeAlignedPointer(z_aligned);
free(y_true);
}
