CUFFT CUFFT_EXEC_FAILED x86_64 RHL 5.3 CUFFT_EXEC_FAILED while using x86_64 on RHL EE 5.3

Accelerated Computing CUDA CUDA Programming and Performance
1

Greetings,

CUFFT_EXEC_FAILED message reported when using libcufft.so.2.2 as follows:

  cufftHandle plan;
  cufftPlan1d(&plan, 1024, CUFFT_R2C, 1);
  cufftExecR2C(plan, dataInput, dataOutput);

  //destroy the plan
  cufftDestroy(plan);

Here is the reported error:
“cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.2/cufft/src/execute.cu, line 1070
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.2/cufft/src/cufft.cu, line 147
cufft: ERROR: CUFFT_EXEC_FAILED”

For some reason this error only reports once. Even when the same sequence (listed above) is called 100,000 times the error posted above only appears once. The output values are correct for both 1 run or 100,000 runs so cufft seems to be successful despite this indication.

Anyone run across this symptom using RHL 5.3 x86_64 with cufft 2.2?

Thank you for your thoughts. ;-)

2

Problem solved!

This error was caused by a prior runtime error (that was not being checked).

The error was discovered by adding the following code to check if there were any existing errors prior to calling the FFT:

[u]cudaError_t cudaError;

cudaError = cudaGetLastError();

if( cudaError != cudaSuccess )

{

fprintf(stderr, “CUDA Runtime API Error reported : %s\n”, cudaGetErrorString(cudaError));

exit(EXIT_FAILURE);

}[/u]

Hope this helps anyone else in debugging their use of the Cuda Runtime API.

3
4

multiple calls of cudaSetDevice(SelectedDevice); in a short period of time also cause this errors in cufft. line 147 and execute.cu line 1070.

(int SelectedDevice = 1;)