template: speedup lexer #1335
Merged
template: speedup lexer #1335
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
Taken from the yagpdb-cc repository. To avoid licensing issues, I only used programs written by me. Baseline performance: name time/op Parse/lorem_ipsum-4 2.01µs ± 7% Parse/short-4 70.2µs ± 1% Parse/medium-4 165µs ± 1% Parse/long-4 517µs ± 1% Parse/very-long-4 1.79ms ± 1% name alloc/op Parse/lorem_ipsum-4 1.23kB ± 0% Parse/short-4 5.48kB ± 0% Parse/medium-4 12.7kB ± 0% Parse/long-4 35.3kB ± 0% Parse/very-long-4 113kB ± 0% name allocs/op Parse/lorem_ipsum-4 12.0 ± 0% Parse/short-4 109 ± 0% Parse/medium-4 298 ± 0% Parse/long-4 779 ± 0% Parse/very-long-4 2.61k ± 0%
Notably, this includes https://go-review.googlesource.com/c/go/+/421883, which changes the lexer to not spawn a new goroutine. name old time/op new time/op delta Parse/lorem_ipsum-4 2.01µs ± 7% 0.73µs ± 2% -63.56% (p=0.000 n=7+8) Parse/short-4 70.2µs ± 1% 16.6µs ± 1% -76.32% (p=0.000 n=7+8) Parse/medium-4 165µs ± 1% 40µs ± 1% -75.74% (p=0.000 n=8+8) Parse/long-4 517µs ± 1% 123µs ± 1% -76.28% (p=0.000 n=8+7) Parse/very-long-4 1.79ms ± 1% 0.42ms ± 1% -76.81% (p=0.000 n=8+8) name old alloc/op new alloc/op delta Parse/lorem_ipsum-4 1.23kB ± 0% 1.16kB ± 0% -5.56% (p=0.000 n=8+8) Parse/short-4 5.48kB ± 0% 5.42kB ± 0% -1.17% (p=0.000 n=8+8) Parse/medium-4 12.7kB ± 0% 12.6kB ± 0% -0.51% (p=0.000 n=8+8) Parse/long-4 35.3kB ± 0% 35.3kB ± 0% -0.17% (p=0.000 n=8+8) Parse/very-long-4 113kB ± 0% 113kB ± 0% -0.05% (p=0.000 n=8+8) name old allocs/op new allocs/op delta Parse/lorem_ipsum-4 12.0 ± 0% 10.0 ± 0% -16.67% (p=0.000 n=8+8) Parse/short-4 109 ± 0% 107 ± 0% -1.83% (p=0.000 n=8+8) Parse/medium-4 298 ± 0% 296 ± 0% -0.67% (p=0.000 n=8+8) Parse/long-4 779 ± 0% 777 ± 0% -0.26% (p=0.000 n=8+8) Parse/very-long-4 2.61k ± 0% 2.61k ± 0% -0.08% (p=0.000 n=8+8)
lexer.next is a hot function, as demonstrated by profiling. Most programs will consist of ASCII characters only, which we can optimize for. Ideally DecodeRuneInString would be inlined here and this wouldn't be a problem at all, but that won't be the case until golang/go#31666 is resolved. name old time/op new time/op delta Parse/lorem_ipsum-4 733ns ± 2% 733ns ± 2% ~ (p=0.933 n=8+8) Parse/short-4 16.6µs ± 1% 15.5µs ± 2% -6.75% (p=0.000 n=8+8) Parse/medium-4 40.1µs ± 1% 38.7µs ± 1% -3.51% (p=0.000 n=8+8) Parse/long-4 123µs ± 1% 115µs ± 1% -5.86% (p=0.001 n=7+7) Parse/very-long-4 416µs ± 1% 396µs ± 1% -4.70% (p=0.000 n=8+8) name old alloc/op new alloc/op delta Parse/lorem_ipsum-4 1.16kB ± 0% 1.16kB ± 0% ~ (all equal) Parse/short-4 5.42kB ± 0% 5.42kB ± 0% ~ (all equal) Parse/medium-4 12.6kB ± 0% 12.6kB ± 0% ~ (all equal) Parse/long-4 35.3kB ± 0% 35.3kB ± 0% ~ (all equal) Parse/very-long-4 113kB ± 0% 113kB ± 0% ~ (all equal) name old allocs/op new allocs/op delta Parse/lorem_ipsum-4 10.0 ± 0% 10.0 ± 0% ~ (all equal) Parse/short-4 107 ± 0% 107 ± 0% ~ (all equal) Parse/medium-4 296 ± 0% 296 ± 0% ~ (all equal) Parse/long-4 777 ± 0% 777 ± 0% ~ (all equal) Parse/very-long-4 2.61k ± 0% 2.61k ± 0% ~ (all equal)
CommandNode.append and by extension runtime.growslice was showing up more than expected during profiling. Allocate enough space for four arguments up front so we don't need to reallocate as much. Although this doesn't benefit performance much, it does have a clear positive effect on memory usage. name old time/op new time/op delta Parse/lorem_ipsum-4 733ns ± 2% 728ns ± 2% ~ (p=0.315 n=8+8) Parse/short-4 15.5µs ± 2% 15.2µs ± 1% -2.12% (p=0.001 n=8+8) Parse/medium-4 38.7µs ± 1% 38.0µs ± 1% -1.99% (p=0.000 n=8+8) Parse/long-4 115µs ± 1% 117µs ± 5% ~ (p=0.281 n=7+8) Parse/very-long-4 396µs ± 1% 400µs ± 1% +1.02% (p=0.002 n=8+8) name old alloc/op new alloc/op delta Parse/lorem_ipsum-4 1.16kB ± 0% 1.16kB ± 0% ~ (all equal) Parse/short-4 5.42kB ± 0% 5.08kB ± 0% -6.20% (p=0.000 n=8+8) Parse/medium-4 12.6kB ± 0% 12.7kB ± 0% +0.51% (p=0.000 n=8+8) Parse/long-4 35.3kB ± 0% 34.4kB ± 0% -2.58% (p=0.000 n=8+8) Parse/very-long-4 113kB ± 0% 110kB ± 0% -2.37% (p=0.000 n=8+8) name old allocs/op new allocs/op delta Parse/lorem_ipsum-4 10.0 ± 0% 10.0 ± 0% ~ (all equal) Parse/short-4 107 ± 0% 93 ± 0% -13.08% (p=0.000 n=8+8) Parse/medium-4 296 ± 0% 276 ± 0% -6.76% (p=0.000 n=8+8) Parse/long-4 777 ± 0% 705 ± 0% -9.27% (p=0.000 n=8+8) Parse/very-long-4 2.61k ± 0% 2.35k ± 0% -9.77% (p=0.000 n=8+8)
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.
Speed up the template lexer by ~4x.
Best reviewed commit-by-commit; see associated messages for justification and intermediate benchmark results.
Note that although this is a large diff, a large portion of it is due to benchmark fixtures which may be ignored.