Closed fasterthanlime closed 3 months ago
Comparing back-to-stable
(68995c1) with main
(9df2294)
โ 2 (๐ 2)
regressions
โ
4
untouched benchmarks
๐ 4
new benchmarks
โ๏ธ 2 (๐ 2)
dropped benchmarks
Benchmark | main |
back-to-stable |
Change | |
---|---|---|---|---|
๐ | format_content_length/itoa/buffet |
203.6 ยตs | 285.5 ยตs | -28.69% |
๐ | format_content_length/itoa/heap |
196.7 ยตs | 865.7 ยตs | -77.28% |
๐ | format_content_length/itoa/pool |
N/A | 169.4 ยตs | N/A |
๐ | format_content_length/itoa/stack |
34.7 ยตs | N/A | N/A |
๐ | format_content_length/lut |
N/A | 153 ยตs | N/A |
๐ | format_content_length/std_fmt |
350.8 ยตs | N/A | N/A |
๐ | format_content_length/std_fmt/buffet |
N/A | 504.6 ยตs | N/A |
๐ | format_content_length/std_fmt/heap |
N/A | 1 ms | N/A |
That's super interesting โ on macOS I see a 20% regression. I guess I don't care for absolute macOS performance that much, but it could be worth adding a nightly
feature that uses #[thread_local]
instead of thread_local!
It would be nice to be on stable. We'll see if the perf is acceptable.