8000 Buffer usage causes a tensor pack test to fail on gfx942 · Issue #20294 · iree-org/iree · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

Buffer usage causes a tensor pack test to fail on gfx942 #20294

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

Open
krzysz00 opened this issue Mar 18, 2025 · 0 comments
Open

Buffer usage causes a tensor pack test to fail on gfx942 #20294

krzysz00 opened this issue Mar 18, 2025 · 0 comments
Assignees
Labels
bug 🐞 Something isn't working hal/amdgpu Runtime AMDGPU HAL backend

Comments

@krzysz00
Copy link
Contributor

tests/e2e/tensor_ops/pack.mlir has a static_pack_pad_transpose_inner_and_outer_dims_large test in it that fails on gfx942 due to llvm/llvm-project#131386 causing one of its dispatches to produce incorrect values.

This ticket is to track the fact that we've excluded all of pack.mlir from the test suite on all GPUs to work around this so we can land #19918 . We expect this could be resolved by

  1. Fixes to the backend
  2. Fixes to pack kernel generation
  3. Changes that cause masked loads from buffers to generate better code - see [MLIR][AMDGPU] Adding Vector transfer_read to load rewrite pattern llvm/llvm-project#131803 - which could maybe un-break this incidentally.
@krzysz00 krzysz00 self-assigned this Mar 18, 2025
@krzysz00 krzysz00 added bug 🐞 Something isn't working hal/amdgpu Runtime AMDGPU HAL backend labels Mar 18, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug 🐞 Something isn't working hal/amdgpu Runtime AMDGPU HAL backend
Projects
None yet
Development

No branches or pull requests

1 participant
0