Possible issue in PyOpenCL during long GPU run with PySPH #562
Unanswered
Qcellaris
asked this question in
Troubleshooting
Replies: 1 comment 1 reply
-
My read on this:
is that something inside of PySPH gets out of control and tries to allocate implausibly large amounts of memory, here in the tens of thousands of exabytes. This number is so large that it fails to fit into the 64-bit integer that OpenCL accepts as the size of the allocation (which in turn is massively larger than any available GPU memory). An unsigned integer underflow comes to mind as a possible reason. As such, I think the implausible allocation size is the issue, the |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Dear developers,
I have been working with PySPH for a while now and I started using it on a GPU. PySPH (via the Compyle packages) relies on PyOpenCL for it's GPU functionality. After running the code for a couple of hours on a GPU, the process suddenly is killed with an error message relating to PyOpenCL. I have been trying to figure out what goes wrong but I can't find the issue. I posted the original issue, together with the error log, on the GitHub of PySPH:
pypr/pysph#348
Since the error message refers to PyOpenCL, I just wanted to check if it might be a bug in the PyOpenCL code. If this is not the case, I can narrow down the scope of my investigation and try to dig deeper in the Compyle and PySPH packages that are build on the PyOpenCL package.
Kind regards,
Stephan
Beta Was this translation helpful? Give feedback.
All reactions