10:22:37 <sech1> hyc https://github.com/tevador/RandomX/pull/273 - it looks like that we didn't save FP rounding mode in Windows binaries. I added CI tests and discovered it.
14:46:51 <hyc> I dunno how to review the yaml stuff but the other two look ok to me
14:48:32 <sech1> you can check syntax at https://www.yamllint.com/
14:48:50 <sech1> It's just a sequence of commands for Github actions
14:51:08 <hyc> anyway, isn't fpu state per-thread already? failing to save/restore it isn't going to affect most users because they're only running a miner anyway
14:51:15 <sech1> it's per thread
14:51:42 <sech1> all incoming Monero blocks are also verified, so RandomX hash is calculated somewhere (not sure which thread)
14:52:05 <hyc> yeah but monerod doesn't use floating point for anything else
14:52:16 <hyc> or nothing consensus related anyway
14:52:20 <sech1> yes
14:52:38 <sech1> but someone who uses RandomX lib can get unexpected results
14:53:42 <hyc> right. just seems unlikely they're doing anything besides mining.
14:58:35 <sech1> This PR also runs tests on s390x architecture (big endian), among other things. Good to know it's not broken :)
14:59:56 <hyc> nice... it's so annoying that so much software is written with implicit little-endian dependency these days. like authors aren't even aware
15:02:38 <sech1> I tried it on p2pool and it's broken everywhere on s390x :D