int main(int argc, char *argv[]) {
int i, j, k, ret, loops, freq, log2_N, jobs, N, mb = mbox_open();
unsigned t[2];
double tsq[2];
GPU_FFT_COMPLEX *base;
struct GPU_FFT *fft;
log2_N = argc>1? atoi(argv[1]) : 12; // 8 <= log2_N <= 21
jobs = argc>2? atoi(argv[2]) : 1; // transforms per batch
loops = argc>3? atoi(argv[3]) : 1; // test repetitions
if (argc<2 || jobs<1 || loops<1) {
printf(Usage);
return -1;
}
N = 1<<log2_N; // FFT length
ret = gpu_fft_prepare(mb, log2_N, GPU_FFT_REV, jobs, &fft); // call once
switch(ret) {
case -1: printf("Unable to enable V3D. Please check your firmware is up to date.\n"); return -1;
case -2: printf("log2_N=%d not supported. Try between 8 and 21.\n", log2_N); return -1;
case -3: printf("Out of memory. Try a smaller batch or increase GPU memory.\n"); return -1;
case -4: printf("Unable to map Videocore peripherals into ARM memory space.\n"); return -1;
}
for (k=0; k<loops; k++) {
for (j=0; j<jobs; j++) {
base = fft->in + j*fft->step; // input buffer
for (i=0; i<N; i++) base[i][0] = base[i][1] = 0;
freq = j+1;
base[freq][0] = base[N-freq][0] = 0.5;
}
usleep(1); // Yield to OS
t[0] = Microseconds();
gpu_fft_execute(fft); // call one or many times
t[1] = Microseconds();
tsq[0]=tsq[1]=0;
for (j=0; j<jobs; j++) {
base = fft->out + j*fft->step; // output buffer
freq = j+1;
for (i=0; i<N; i++) {
double re = cos(2*GPU_FFT_PI*freq*i/N);
tsq[0] += pow(re, 2);
tsq[1] += pow(re - base[i][0], 2) + pow(base[i][1], 2);
}
}
printf("rel_rms_err = %0.2g, usecs = %d, k = %d\n",
sqrt(tsq[1]/tsq[0]), (t[1]-t[0])/jobs, k);
}
gpu_fft_release(fft); // Videocore memory lost if not freed !
return 0;
}
void do_fft(double *buffer) {
unsigned int ctr = 0;
// Copy the buffer into the fft data structure
for (ctr = 0; ctr < N; ctr++) {
fft->in[ctr].re = buffer[ctr];
fft->in[ctr].im = 0.0;
}
// Push to the GPU
gpu_fft_execute(fft);
// Copy back the square magnitudes for welching
for (ctr = 0; ctr < N; ctr++) {
buffer[ctr] = sqrt( pow(fft->out[ctr].re,2) + pow(fft->out[ctr].im,2));
}
}
int main(int argc, char *argv[]) {
unsigned t[2];
struct GPU_FFT_COMPLEX *dataIn,*dataOut;
struct GPU_FFT *fftinfo;
int mb = mbox_open();
gpu_fft_prepare(mb,12,GPU_FFT_FWD,1,&fftinfo);
dataIn = fftinfo->in;
dataOut = fftinfo->out;
usleep(1); // Yield to OS
t[0] = Microseconds();
gpu_fft_execute(fftinfo); // call one or many times
t[1] = Microseconds();
printf("usecs = %d\n", (t[1]-t[0]));
gpu_fft_release(fftinfo); // Videocore memory lost if not freed !
return 0;
}