cusolverDnCreate segmentation fault

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everybody,

I had implemented an eigensolver following page 217 of https://docs.nvidia.com/cuda/pdf/CUSOLVER_Library.pdf which was working perfectly as of today. Nevertheless, it now throws a segmentation fault error that I have narrowed down to the line

cusolver_status = cusolverDnCreate(&cusolverH);

I honestly have no idea how to debug this error. Google has been of little help. How may I print/assert the handle cusolverH so that I know that it has been correctly initialized? Otherwise, how can I print cusolver_status itself?

Let’s say that the initialization of the handle is correct, then which is the usual cause of a segmentation fault in these cases?

Thank you very much for your time.

2

Hi again everybody,

there has been a “development”. The segmentation fault is related to being unable to access [the handle i]cusolverH[/i], which is a host side problem. Nevertheless, when the handle is accessed, the assertion following the line of code mentioned on my previous post fails (see the two lines of code below):

cusolver_status = cusolverDnCreate(&cusolverH);
assert(CUSOLVER_STATUS_SUCCESS == cusolver_status);

Any idea on why cusolver does not succeed? What is the usual reason of cusolver’s failure in these cases?

3

This sounds to me like an example of stack corruption in your program. That isn’t related to CUDA.