Crash fix cudaMallocAsync usage of TF_CUDA_MALLOC_ASYNC_SUPPORTED_PREALLOC. #50962
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
sanjoy
approved these changes
Jul 27, 2021
|
Please make the PR description a bit more descriptive (what was the bug?). CC @hanbinyoon |
|
I updated the description. |
1cb58ca
to
17d3b28
Compare
sanjoy
suggested changes
Aug 4, 2021
tensorflow/core/common_runtime/gpu/gpu_cudamallocasync_allocator.cc
Outdated
Show resolved
Hide resolved
17d3b28
to
e41fc39
Compare
e41fc39
to
f95e862
Compare
|
I rebased this PR since the dependent PR is merged. |
|
Any update? |
sanjoy
approved these changes
Sep 8, 2021
|
The CI seems to have failed: Is this related to this PR? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Some of the refactoring caused crash when using TF_CUDA_MALLOC_ASYNC_SUPPORTED_PREALLOC.
This PR fix them and add a test to be sure it continue to work.
This PR is on top of #50961 as otherwise there would be code conflict.
Now, GpuCudaMallocAsyncAllocator doesn't create its own compute stream. But it use one stream set after construction when GpuCudaMallocAsyncAllocator::SetStream is called. So it is now impossible to preallocate all memory during the object construction.
So this PR postpone the memory preallocation to when the stream is available, in SetStream.
This PR also check that the stream isn't set when SetStream is called.
@sanjoy