Fix performance regressions introduced by my recent changes #1257
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
…d value for keywords If we model `RawTokenKind` as an enum with associated values, equality comparisons call into `__derived_enum_equals`, which is less efficient than comparing the base and keyword of the `RawTokenKind` separately. I believe this is because `__derived_enum_equals` needs to handle the case where multiple cases have associated values, while we can special case for a single associated value.
If we inline the matching calls, the compiler can constant evaluate the `self.rawTokenKind.base == .keyword` condition in `matches`, allowing it to avoid comparing string contents if we are not expecting a keyword.
The ARC traffic to keep track of the state stack was significant. Since most of the time we are in the `normal` state and the stack is effectively empty, we can store any large-ish stack in a bump allocator which outlives the cursor, avoiding ARC traffic.
If you call a non-mutating method from a mutating method, for some reason that results in a retain of `self`. Marking all methods (most importantly the at` methods avoids that ARC traffic.
This avoids some more ARC traffic.
|
@swift-ci Please test |
rintaro
reviewed
Jan 24, 2023
| case .push(newState: let newState): | ||
| if let topState = topState { | ||
| if let stateStack = stateStack { | ||
| let newStateStack = stateAllocator.allocate(State.self, count: stateStack.count + 1) |
There was a problem hiding this comment.
It's a little unfortunate we don't reuse previously-allocated buffer. For example
push, push, pop, push. operation should reuse the buffer allocated at the second push, instead of allocating a whole new buffer.
There was a problem hiding this comment.
Yes, but how do you know that you have already seen this exact stack? We could try to create a dictionary for it, but given that the bump allocator doesn’t allocate more than 250 byte for any file in MovieSwiftUI, I’m not sure if there’s much to be gained here.
There was a problem hiding this comment.
I am convinced that we can't reuse the memory region since other (copied) cursor might still have a reference to it.
rintaro
approved these changes
Jan 24, 2023
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 improves parsing performance by ~2.5x. Parsing is still about 1.5x slower compared to before #1173 but given that I introduced quite a bit of necessary state and the fact that we still parse all files in MovieSwiftUI in less than 15ms I hope that’s acceptable. If we want to improve performance further, we just need to do some more investigation.
It’s probably best to review this commit by commit. All commits are standalone and have a description of what they do. I can split any commit that’s controversial to its own PR.