Reduce allocations during graph execution #243
Merged
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
Given that: - Node IDs are incrementally assigned (because they are indices into an array) - A significant proportion of the valid node IDs are value nodes and will be assigned a refcount at some point during the run - The maximum refcount for each value is typically small Then it works out to be more efficient to use an array of refcounts rather than a hash map.
Refactor internals to reduce the number of allocations required in the likely case that the number of inputs is small.
- Change `InputList` to store inputs as a `Cow` - Change `Graph::run_plan` to use a `SmallVec` to store the input list for operators In the process I ran into a borrow-checking issue in the `Slice` op that I didn't get to the bottom of, but found a workaround for.
`<SmallVec as Clone>::clone` internally uses `SmallVec::from`. We want to use the more efficient `SmallVec::from_slice` instead since we know we're dealing with `Copy` items.
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Outdated suggestions cannot be applied.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.
This is a collection of micro-optimizations to reduce from overhead when executing a graph. Mostly this involves using
SmallVecinstead ofVecin cases where we expect the number of items will almost always be small.Also I found that the overhead of creating tensor views could be reduced by implementing
Clonemanually forDynLayout, in order to use aSmallVecfast path forCopy-able items.In the test scenario described in #239 this decreases time spent doing non-compute work on the "main" thread (the one where
Graph::run_planruns) by ~3%.