Difficulty with CUFFT lib. HELP! cufft Inverse is not giving the supposed results

Accelerated Computing CUDA CUDA Programming and Performance
1

I am trying to use the cufft library. I am creating a 2D signal, applying a 2D fft on the 2D signal and then applying fft inverse to get back the original signal. The problem is that the output is nowhere near the original signal. Can someone guide me and tell me if I am doing something wrong please.

Here is the code:

int
main( int argc, char** argv)
{
int NX = 17, NY = 17;
cufftHandle plan;
cufftComplex idata, odata, tdata;
cudaMalloc((void)&idata, sizeof(cufftComplex)NXNY);
cudaMalloc((void*)&odata, sizeof(cufftComplex)NXNY);
cudaMalloc((void**)&tdata, sizeof(cufftComplex)NXNY);

cufftComplex* Input = (cufftComplex*)malloc(NY * NX * sizeof(cufftComplex));

for(int y = 0; y < NY; y++)
for (int x = 0; x < NX; x++)
{
Input[(y * NX) + x].x = 255;
Input[(y * NX) + x].y = 0;
}

cudaMemcpy(idata, Input, (NY * NX * sizeof(cufftComplex)), cudaMemcpyHostToDevice);
/* Create a 2D FFT plan. /
cufftPlan2d(&plan, NX, NY, CUFFT_C2C);
/ Use the CUFFT plan to transform the signal out of place. */
cufftExecC2C(plan, idata, tdata, CUFFT_FORWARD);

cufftExecC2C(plan, tdata, odata, CUFFT_INVERSE);

cufftComplex* InputImage = (cufftComplex*)malloc(NY * NX * sizeof(cufftComplex));
cufftComplex* OutputImageFFT = (cufftComplex*)malloc(NY * NX * sizeof(cufftComplex));
cudaMemcpy(InputImage, idata, (NY * NX * sizeof(cufftComplex)), cudaMemcpyDeviceToHost);
cudaMemcpy(OutputImageFFT, odata, (NY * NX * sizeof(cufftComplex)), cudaMemcpyDeviceToHost);
for (int y = 0; y < NY; y++)
for (int x = 0; x < NX; x++)
printf(“X:%d, Y:%d, A:%f = B:%f %fi\n”, x, y, InputImage[((y * NX) + x)].x, OutputImageFFT[((y * NX) + x)].x, OutputImageFFT[((y * NX) + x)].y);

getchar();
}

Output:
X:0, Y:0, A:255.000000 0.000000i = B:73694.984375 0.085018i
X:1, Y:0, A:255.000000 0.000000i = B:73695.039063 0.059780i
X:2, Y:0, A:255.000000 0.000000i = B:73695.156250 0.002077i
X:3, Y:0, A:255.000000 0.000000i = B:73695.054688 -0.106803i
X:4, Y:0, A:255.000000 0.000000i = B:73694.976563 -0.119947i
X:5, Y:0, A:255.000000 0.000000i = B:73694.882813 -0.017992i
…

The output should be:
X:0, Y:0, A:255.000000 0.000000i = B:255.000000 0.000000i
X:1, Y:0, A:255.000000 0.000000i = B:255.000000 0.000000i
X:2, Y:0, A:255.000000 0.000000i = B:255.000000 0.000000i
X:3, Y:0, A:255.000000 0.000000i = B:255.000000 0.000000i
X:4, Y:0, A:255.000000 0.000000i = B:255.000000 0.000000i
X:5, Y:0, A:255.000000 0.000000i = B:255.000000 0.000000i
…

2

You need to scale the results.

With CUFFT:
IFFT(FFT(A))=len(A)*A

3

Thank you so much!!!