cufft causing terminate to be called?

Accelerated Computing CUDA CUDA Programming and Performance
1

I’ve got something of a strange problem. I was writing some code that uses the cufft library and I noticed that it was segfaulting when I called cufftPlan1d. So I went ahead and updated to the latest driver (260.19.26) and cuda 3.2 and now I’m seeing that whenever I link against cufft the program terminates.

Here’s a simple example:

#include <cuda_runtime.h>

int main() {

    float *data;

    cudaMalloc(&data, 1024*sizeof(float));

}

Note that I’m not explicitly using cufft in there.

If I compile with:

It will run just fine.

However, if I compile with:

It exits with:

And if I run it in the debugger, the backtrace looks like this:

Has anyone seen this problem before? Should I downgrade cuda perhaps?

EDIT: It looks linking with libcublas causes the same problem…