perf: rework decoder interface #22
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
ianprime0509
force-pushed
the
perf/encoding
branch
from
10fa687
to
47f384f
Compare
The updated interface decodes codepoints directly from a reader rather than being implemented as a state machine. This turns out to be considerably more efficient than the previous implementation, with around 25% improvement on the `token_reader` and `reader` benchmarks: ``` Benchmark 1 (27 runs): zig-out/bin-old/token_reader Gtk-4.0.gir measurement mean ± σ min … max outliers delta wall_time 188ms ± 14.5ms 168ms … 205ms 0 ( 0%) 0% peak_rss 7.31MB ± 58.5KB 7.21MB … 7.34MB 0 ( 0%) 0% cpu_cycles 688M ± 4.20M 684M … 706M 1 ( 4%) 0% instructions 1.19G ± 29.4 1.19G … 1.19G 0 ( 0%) 0% cache_references 412K ± 763K 239K … 4.21M 2 ( 7%) 0% cache_misses 10.0K ± 7.40K 7.90K … 46.8K 2 ( 7%) 0% branch_misses 814K ± 1.37K 813K … 821K 1 ( 4%) 0% Benchmark 2 (37 runs): zig-out/bin/token_reader Gtk-4.0.gir measurement mean ± σ min … max outliers delta wall_time 136ms ± 13.8ms 115ms … 147ms 0 ( 0%) ⚡- 27.7% ± 3.8% peak_rss 7.31MB ± 54.7KB 7.21MB … 7.34MB 8 (22%) + 0.1% ± 0.4% cpu_cycles 462M ± 1.87M 459M … 466M 0 ( 0%) ⚡- 32.8% ± 0.2% instructions 1.14G ± 26.6 1.14G … 1.14G 0 ( 0%) ⚡- 4.1% ± 0.0% cache_references 236K ± 4.86K 227K … 244K 0 ( 0%) - 42.7% ± 60.7% cache_misses 9.40K ± 1.25K 7.88K … 11.5K 0 ( 0%) - 6.5% ± 24.6% branch_misses 815K ± 1.01K 813K … 817K 0 ( 0%) + 0.1% ± 0.1% ``` ``` Benchmark 1 (23 runs): zig-out/bin-old/reader Gtk-4.0.gir measurement mean ± σ min … max outliers delta wall_time 225ms ± 14.2ms 199ms … 249ms 0 ( 0%) 0% peak_rss 7.25MB ± 100KB 7.08MB … 7.34MB 0 ( 0%) 0% cpu_cycles 823M ± 12.2M 813M … 847M 0 ( 0%) 0% instructions 1.43G ± 23.0 1.43G … 1.43G 0 ( 0%) 0% cache_references 757K ± 129K 635K … 1.07M 1 ( 4%) 0% cache_misses 13.7K ± 1.18K 12.5K … 17.2K 2 ( 9%) 0% branch_misses 1.43M ± 3.35K 1.42M … 1.43M 0 ( 0%) 0% Benchmark 2 (31 runs): zig-out/bin/reader Gtk-4.0.gir measurement mean ± σ min … max outliers delta wall_time 166ms ± 13.9ms 144ms … 175ms 0 ( 0%) ⚡- 26.5% ± 3.4% peak_rss 7.27MB ± 81.8KB 7.08MB … 7.34MB 0 ( 0%) + 0.3% ± 0.7% cpu_cycles 581M ± 1.54M 579M … 584M 0 ( 0%) ⚡- 29.4% ± 0.5% instructions 1.38G ± 16.0 1.38G … 1.38G 9 (29%) ⚡- 3.8% ± 0.0% cache_references 715K ± 219K 563K … 1.71M 3 (10%) - 5.5% ± 13.6% cache_misses 13.5K ± 1.31K 11.4K … 16.5K 2 ( 6%) - 1.2% ± 5.1% branch_misses 1.07M ± 20.3K 1.05M … 1.11M 5 (16%) ⚡- 25.3% ± 0.6% ```
ianprime0509
force-pushed
the
perf/encoding
branch
from
47f384f
to
48a25fe
Compare
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.
The updated interface decodes codepoints directly from a reader rather than being implemented as a state machine. This turns out to be considerably more efficient than the previous implementation, with around 25% improvement on the
token_readerandreaderbenchmarks: