Please send requests like this to the appropriate mailing list.
Tyler Buresh <bures024(a)umn.edu> writes:
I hope you’re having a great week. If you don’t mind,
I’m eager to get your
advice on a PyCUDA issue I’m having, as it seems like you might be best
person to ask.
As part of a research project at the University of Minnesota, I have been
helping write algorithms using PyCUDA designed to be run on a GPU. I’ve run
into a problem that I have not been able to resolve.
The following error is thrown when transferring data from GPU memory back
to CPU memory when the input matrix has dimensions larger than 1,000 x
pycuda._driver.LogicError: cuMemcpyDtoH failed: an illegal memory access
PyCUDA WARNING: a clean-up operation failed (dead context maybe?)
cuMemFree failed: an illegal memory access was encountered
The code executes without errors on smaller matrices and produces correct
results. Given the relatively small matrix size, it doesn't seem like
memory should be an issue. Have you ever seen an error like this?
If it helps, the code can be found here:
I understand you are likely tremendously busy, so any advice or pointers
would be tremendously appreciated. Thank you for your time.
Room 4318 (Siebel Center), University of Illinois at Urbana-Champaign