[VM] [Hexagon] Add buffers to dma_wait builtin
#16706
Merged
Conversation
This file contains hidden or 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
While introducing dma operations at graph level, relax KillAfterLastUse pass introduces kill_tensor operation after dma_copy. This leads to memory being deallocated when asynchronous copy operation is in progress. Hence, moving the input/output buffers to dma_wait to ensure kill_tensor is introduced after dma_wait at the graph level. Also, the logic for size calculation is updated to use GetDataSize function. The test case is updated to use offsets instead of allocating different storage in VTCM.
Contributor
Author
quic-sanirudh
approved these changes
Mar 14, 2024
Contributor
quic-sanirudh
left a comment
quic-sanirudh
left a comment
There was a problem hiding this comment.
LGTM, apart from the minor comment for unused argument. Thanks.
Comment on lines
55
to
56
| .set_body_typed([](TVMArgValue vm_ptr, int queue_id, int inflight_dma, NDArray src_arr, | ||
| NDArray dst_arr) { |
Contributor
There was a problem hiding this comment.
For src_arr and dst_arr, perhaps we should add the [[maybe_unused]].
Contributor
Author
There was a problem hiding this comment.
Thank you @quic-sanirudh ! I have fixed this in the latest patch.
thaisacs
pushed a commit
to thaisacs/tvm
that referenced
this pull request
Apr 3, 2024
* [VM] [Hexagon] Add buffers to dma_wait builtin While introducing dma operations at graph level, relax KillAfterLastUse pass introduces kill_tensor operation after dma_copy. This leads to memory being deallocated when asynchronous copy operation is in progress. Hence, moving the input/output buffers to dma_wait to ensure kill_tensor is introduced after dma_wait at the graph level. Also, the logic for size calculation is updated to use GetDataSize function. The test case is updated to use offsets instead of allocating different storage in VTCM. * Fix review comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
KillAfterLastUsepass introduceskill_tensoroperation afterdma_copy. This leads to memory being deallocated when asynchronous copy operation is in progress. Hence, moving the input/output buffers todma_waitto ensurekill_tensoris introduced afterdma_waitat the graph level.GetDataSizefunction.