| commit | 6e0481682151f9843ba4a79e42b77eae9bd28224 | [log] [tgz] |
|---|---|---|
| author | Lei Zhang <antiagainst@google.com> | Tue Aug 29 21:54:32 2023 -0400 |
| committer | GitHub <noreply@github.com> | Tue Aug 29 18:54:32 2023 -0700 |
| tree | 295c572c39ab24213ba96e97ab50d8cfdd68f37e | |
| parent | 8105859455d5111247078aae7903a424a82fae16 [diff] |
[cuda] Fix segfault caused by CUevent outliving CUdevice (#14875) We use `iree_hal_cuda2_event_t` to wrap a `CUevent` so that we can have multiple places referencing the `CUevent` with automatic resource releasing when refcount becomes zero. The `iree_hal_cuda2_event_t` objects are further placed in a pool to amortize the cost of creation via recycling. But how it is set up right now, the pool's lifetime is associated with the HAL device. So it will be destroyed together with the device, including the CUDA context and so. All events that are in the pool will be destroyed; but there is no proper handling of outstanding `iree_hal_cuda2_event_t` in this flow. If a `iree_hal_cuda2_event_t`'s refcount decreases to zero after device destruction, it will cause segfaults in CUDA driver given the CUDA context is already gone. So this commit fixes the issue by retaining the pool in the `iree_hal_cuda2_event_t` object and retaining the HAL device in the pool.
IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler and runtime that lowers Machine Learning (ML) models to a unified IR that scales up to meet the needs of the datacenter and down to satisfy the constraints and special considerations of mobile and edge deployments.
See our website for project details, user guides, and instructions on building from source.
IREE is still in its early phase. We have settled down on the overarching infrastructure and are actively improving various software components as well as project logistics. It is still quite far from ready for everyday use and is made available without any support at the moment. With that said, we welcome any kind of feedback on any communication channels!
See our website for more information.
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.