--- Log opened Sun May 28 00:00:14 2023 06:58 < hebasto> I've benchmarked the recent master branch on arn32 pltaform -- https://gist.github.com/hebasto/4ef870213543ce74133c82ce8ef2d150 06:59 < hebasto> and observed a slight regression in the `hash_sha256` internal benchmark when configured `--with-asm=arm32` 07:00 < hebasto> is it expected or known already? 07:00 < sipa> Slight regression compared to what? 07:00 < sipa> That code has not changed in a long time. 07:01 < hebasto> compared to `--with-asm=no` 07:02 < sipa> That shouldn't be related. 07:02 < sipa> Because there is no sha256 asm code in libsecp256k1 07:03 < sipa> If you do notice a systemic regression, it's probably due to random cache page alignment changes in the binary. 07:04 < hebasto> yes, regression is a systematic one; at least, benchmarks results are stable 07:06 < hebasto> does cache page alignment depends on a particular CPU? 07:15 < sipa> no, on the compiler 07:16 < sipa> if you change anything about the binary, code will be arranged differently 07:16 < sipa> just which instructions end up in which place in the binary 07:16 < sipa> and that can affect caching effects on CPUs 07:25 -!- tromp [~textual@92-110-219-57.cable.dynamic.v4.ziggo.nl] has quit [Read error: Connection reset by peer] 07:29 < sipa> oh yes, it depends on the CPU too of course 07:30 < hebasto> thanks for clarification 23:00 -!- Netsplit *.net <-> *.split quits: jonasschnelli, user_ 23:00 -!- Netsplit over, joins: user_ 23:01 -!- Netsplit over, joins: jonasschnelli --- Log closed Mon May 29 00:00:15 2023