08:42:46 <rbrunner> I am trying to do initial blockchain sync, with self-compiled latest master, on OpenBSD 7.4
08:43:49 <rbrunner> There is a problem that syncing gets slower and slower, until over the course of about 1 hour you can hardly move the X cursor anymore, and echo of single characters on the monerod console takes a minute or so
08:45:03 <rbrunner> It looks to me like the first suspicion of a memory leak is not the answer because "top" shows an almost constant memory use of about half a gig for monerod
08:46:31 <rbrunner> If I still manage to exit monerod (still possible after about 1 hour, after 2 hours the systems seems to be frozen completely) everything returns to normal
08:47:37 <rbrunner> and then after a restart syncing resumes and is fast again
08:47:47 <rbrunner> for about 1 hour. Rinse and repeat.
08:47:53 <rbrunner> Any idea?
10:47:40 <m-relay> <s​yntheticbird:monero.social> 01GXQ7PK2DK0KQX35H8D6MZ5ES.gif
10:47:43 <m-relay> <s​yntheticbird:monero.social> cry about it
10:49:03 <m-relay> <s​yntheticbird:monero.social> This would require flamegraph I think
11:08:32 <selsta> rbrunner: https://www.reddit.com/r/openbsd/comments/txdt4o/comment/i3orkh0/ related?
11:08:38 <selsta> https://github.com/monero-project/monero/issues/7027
11:49:56 <SyntheticBird> Related to BP+ optimization discussion brought by ArticMine. Is the actual implementation making use of SIMD ? If yes up to which one, SSE2 ? SSE4x ? AVX2 ?
11:49:59 <rbrunner> selsta: Thanks. Not sure, but it gave me ideas what to try next.
11:50:50 <SyntheticBird> Checking crypto/crypto-ops.c, I would like to believe the compiler vectorize the work. But I'm not entirely sure
11:53:19 <moneromooo> src/ringct/bullet*. Building blocks are crypto-ops.c, but I would touch src/ringct first.
11:53:58 <moneromooo> The multiexp is a large part of the processing. Multiprocessing would be a better performance/work ratio to go for first.
11:55:06 <moneromooo> There's a version of the stuff from crypto-ops for the wallet which uses asm (can't recall whether SIMD, but likely), it did not replace those ops for the node due to fork concerns.
11:55:27 <selsta> https://github.com/monero-project/supercop/tree/monero
11:57:34 <SyntheticBird> alr. thx for the link