[Relax] Share storage allocs among functions after cuda graph rewriting#16830
Merged
vinx13 merged 1 commit intoapache:mainfrom Apr 2, 2024
Merged
[Relax] Share storage allocs among functions after cuda graph rewriting#16830vinx13 merged 1 commit intoapache:mainfrom
vinx13 merged 1 commit intoapache:mainfrom
Conversation
vinx13
force-pushed
the
feat/cuda-graph-merge-1
branch
3 times, most recently
from
55c8b4e to
97dbcf6
Compare
vinx13
force-pushed
the
feat/cuda-graph-merge-1
branch
from
97dbcf6 to
045342f
Compare
tqchen
approved these changes
Apr 1, 2024
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR makes storage among different functions shared after cuda graph rewriting. Because CUDA graph cache storage, storage objects are not freed after function execution, this will increase memory usage if there are multiple functions. Making storage objects shared eliminate such overhead.
It also updates rewriting to prevent capturing storages and bindings used as function output. Previous we relies on the fact output tensors are allocated with
R.builtin.alloc_tensor, however, this behavior changed after we enable storage planning for output tensor, which may also useR.memory.alloc_memorycc @tqchen