You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While on a call about things that might eventually lead to the use of this to actually help people, i figured i might collect some performance data via cargo flamegraph --bench simple_parse --features string_index -- --bench - results are interesting.
We're eating up less time than the threading libraries and benchmark suite by a good margin (expected, simple operations). Somehow you're beating me out pretty consistently on the query vs index bits :-p.
Unfortunately, GH won't lemme attach the flamegraph SVG by itself, but here's a screenshot of where we are today on an i9HK (gen9):
The pink (? sorry, kinda colorblind) parts are results for the rusthl7 search within the SVG, about 1/3 of the runtime measured.
Mostly filing this issue to plant a flag for baseline - now its here for posterity in case those numbers change for the worse.
The text was updated successfully, but these errors were encountered:
Sweet, feel free to add the raw SVG inside the codebase (in a readme in benches or something?) 👍
I have difficulty running perf on my windows box to get flamegraphs (and my linux box has dust on it I use it so rarely) so this is good to see. I started looking into adding instruction counts/cache misses to the build benchmark rather than wall time to remove VM fluctuations.
While on a call about things that might eventually lead to the use of this to actually help people, i figured i might collect some performance data via
cargo flamegraph --bench simple_parse --features string_index -- --bench
- results are interesting.We're eating up less time than the threading libraries and benchmark suite by a good margin (expected, simple operations). Somehow you're beating me out pretty consistently on the
query
vsindex
bits :-p.Unfortunately, GH won't lemme attach the flamegraph SVG by itself, but here's a screenshot of where we are today on an i9HK (gen9):
The pink (? sorry, kinda colorblind) parts are results for the
rusthl7
search within the SVG, about 1/3 of the runtime measured.Mostly filing this issue to plant a flag for baseline - now its here for posterity in case those numbers change for the worse.
The text was updated successfully, but these errors were encountered: