Sharing mData between device and staging tensors #130
Closed
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Outdated suggestions cannot be applied.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.
Currently, when creating a tensor the data is copied several times:
Tensor
constructorOpTensorCreate
when creating a staged tensorOptionally for input/output tensors:
OpTensorSyncDevice
/OpTensorSyncLocal
when creating another staged tensorThis is 5x the required memory. Additionally, those staging tensors are not destroyed because sequences are never deleted (#36) which creates a memory leak. This is unacceptable when working with large amounts of data. It has already happened that I've run out of host memory several times.
To mitigate this, I've converted
Tensor::mData
fromstd::vector
to astd::shared_ptr<std::vector>
to enable sharing the data between device tensors and staging tensors to reduce memory footprint a little.In the long term larger refactoring is needed (esp. #36 and #14)