jax
c761d902 - [ROCm] Fix GPU handle pool singleton aliasing using opaque handle type separation

Commit

150 days ago

[ROCm] Fix GPU handle pool singleton aliasing using opaque handle type separation Problem: - hipBLAS, hipSOLVER, and other GPU libraries were unintentionally sharing the same singleton instance of `HandlePool` - This occurred because all pools used the same template type signature (e.g. `HandlePool<hipblasHandle_t, Stream>`), leading to a single static instantiation - As a result, using FFI to call hipSOLVER followed by hipBLAS could lead to overlapping internal state - This caused potential memory corruption and unpredictable behavior, especially in mixed or concurrent workloads

References

upstream_fix_handle_pool

#30715 - [ROCm] Fix GPU handle pool singleton aliasing using opaque handle typ…

#31381 - Remove old ROCm build code

#31720 - Fix ann_test.py numerical bug in target reshape

#31768 - [ROCm] Support lowering through PJRT_Triton_Extension

#32115 - Relax version requirements for ROCm Jax Plugin wheels

#33157 - Resolve undefined behavior in bitshift unit test

#33186 - Make nvidia version data optional for ROCm builds

#579 - Create rocm-test-requirements.txt

#580 - GESVDJ support for ROCm GPUs in JAX

#581 - Fix/pallas tests shared memory