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

CUDA Misaligned Data #1233

Open
aescobed opened this issue Feb 10, 2025 · 3 comments
Open

CUDA Misaligned Data #1233

aescobed opened this issue Feb 10, 2025 · 3 comments

Comments

@aescobed
Copy link

Hello, I am having an issue with GPU enabled HYPRE. In some of my test cases when I attempt to solve the pressure equations I get a CUDA misaligned data error. Any ideas as to why this is happening? It happens at the following line:
HYPRE_ParCSRGMRESSolve
(
peqn->hypreSlvr,
peqn->hypreParA,
peqn->hypreParRhs,
peqn->hypreParP
);

@victorapm
Copy link
Contributor

Hi @aescobed, could you provide more details about your setup? Specifically:

  1. Which version of HYPRE are you using?
  2. What GPU architecture are you running on?
  3. How was HYPRE configured and built (e.g., any specific CMake options)?
  4. What are the matrix and right-hand side sizes in the failing test cases?
  5. Does the error occur with all cases you tested or only specific ones?
  6. Is your code open source
  7. Did you get full stacktrace of the error?

With this information, we can help diagnose the issue more effectively.

Best,
Victor

@aescobed
Copy link
Author

Hi @victorapm, here are the answers to your questions, thanks for the response:

  1. 2.32.0

  2. I am using CUDA, sometimes using the thrust api with -arch=sm_75

  3. I used this when configuring petsc
    --prefix=/home/andres/Software/petsc/petsc-3.22.3-cuda-streams2/petsc/install --with-fc=0 --download-f2cblaslapack --download-hdf5 --with-mpi-dir=/home/andres/Software/openmpi/openmpi-5.0.3/openmpi-5.0.3/install --download-hypre --with-64-bit-indices=1 --with-debugging=0 --with-cuda --with-cuda-dir=/home/andres/Software/cuda/cuda_toolkit_12.4/install --with-precision=double --with-device-memory-pool=0 --CUDAFLAGS="-default-stream legacy"

  4. matrix (hypreA): 0 170568 0 170568, rhs: 0 170568

  5. It only happens with certain test cases, but other test cases which also incorporate a moving IBM seem to work fine so I'm not sure why it doesn't work with these test cases

  6. The CPU version is open source: https://github.com/sebastipa/TOSCA
    The GPU version's repo is not public but I can ask my supervisor if I can share it with you.

  7. I have been having trouble getting the full stacktrace, I can keep working on that if that would be of benefit to you.

Thank you!

@victorapm
Copy link
Contributor

Thank you for the info! Seeing the full stacktrace would help.

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

2 participants