Closed seliopou closed 6 years ago
just me being curious, but is there any benchmark difference before and after (i understand that in theory there shouldn't be any)
@hannesm FWIW I ran some internal benchmarks before/after this PR and saw no significant change in those tests.
This was motivated by httpaf benchmarks. In particular I was running perf on wrk_async_benchmark.exe
during a load test and saw a bunch of C calls and caml_ba_set_1
's, which were coming from the Faraday Bigstring
module, which I hadn't been paying as much attention to as Angstrom's.
So to be clear, there shouldn't be a difference in Angstom, but there should be a measurable difference in Faraday.
FWIW - I had my pins setup incorrectly in my previous tests. There is a nice speedup with this change.
Looks good to me! Nice to have this factored out and usable externally - thanks!