Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

HISTORGRAM-OMPTarget triggers a memory error during checksum computation #493

Open
ThorBl opened this issue Nov 12, 2024 · 0 comments
Open

Comments

@ThorBl
Copy link
Contributor

ThorBl commented Nov 12, 2024

Describe the bug

Memory access fault during execution of updateChecksum().

OFFLOAD ERROR: Memory access fault by GPU 8 (agent 0xa11a80) at virtual address 0xf15000. Reasons: Unknown (0)
Device pointer 0xf15000 does not point into any (current or prior) host-issued allocation (might be a global, stack, or shared location).

To Reproduce
Run HISTOGRAM-OMPTarget. I was also able to reproduce the same issue with MULTI_REDUCE-OMPTarget.

Expected behavior

There is no memory access fault.

Compilers & Libraries (please complete the following information):

  • Compiler & version: [e.g. GCC 4.9.3]: Clang 20
  • ROCm version (if applicable): 6.2.1

Additional context

The error occurs during the checksum computation in the calcChecksum method (DataUtils.hpp:387). The problem seems to be that 'dataSpace' (rajaperf::DataSpace::OmpTarget) and check_dataSpace (rajaperf::DataSpace::Host) mismatch. If the data spaces mismatch, it is falsely assumed that ptr (m_counts_final.data()) points to GPU memory. The following call of copyData() (DataUtils.hpp:397) triggers the memory error.

ThorBl added a commit to ROCm/aomp that referenced this issue Nov 13, 2024
Updated script to use the latest version of RAJAPerf. It contains a bug fix for the build error we currently observe. Another bug ticket is still open (LLNL/RAJAPerf#493).

!!! Requires cmake version 3.23  !!!!
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant