Closed joaander closed 11 months ago
I can fix this segfault using the gnu development environment:
module load PrgEnv-gnu
There is still a memory issue with HPMC. When running the HPMC pytest suite, I get:
pytest/test_clusters.py::test_valid_setattr_attached[GPU-Sphinx-8-pivot_move_probability-0.2] PASSED
*Warning*: Falling back on CPU. No GPU implementation for shape.
pytest/test_clusters.py::test_valid_setattr_attached[GPU-Sphinx-8-pivot_move_probability-0.5] PASSED
*Warning*: Falling back on CPU. No GPU implementation for shape.
pytest/test_clusters.py::test_valid_setattr_attached[GPU-Sphinx-8-pivot_move_probability-0.8] PASSED
:0:rocdevice.cpp :2614: 528277422298 us: 5289 : [tid:0x7fff2828f700] Device::callbackQueue aborting with error : HSA_STATUS_ERROR_MEMORY_FAULT: Agent attempted to access an inaccessible address. code: 0x2b
Fatal Python error: Aborted
Description
HOOMD segfaults when running the HPMC unit tests on crusher. To reproduce, build HOOMD using the crusher environment (https://github.com/glotzerlab/software/pull/263) in RelWithDebInfo mode (the segfault doesn't occur in Debug mode).
Script
Input files
No response
Output
Expected output
All tests to pass.
Platform
GPU, Linux
Installation method
Compiled from source
HOOMD-blue version
3.8.1
Python version
3.9.13