--- Log opened Thu May 11 00:00:58 2023
03:35 -!- jonatack [~jonatack@user/jonatack] has quit [Ping timeout: 264 seconds]
05:34 -!- jonatack [~jonatack@user/jonatack] has joined #secp256k1
05:49 -!- preimage [~halosghos@user/halosghost] has joined #secp256k1
05:56 -!- jonatack [~jonatack@user/jonatack] has quit [Quit: WeeChat 3.8]
06:06 -!- jonatack [~jonatack@user/jonatack] has joined #secp256k1
09:14 < hebasto> is there a plan to tag a new release after getting #1129 merged?
10:04 < andytoshi> hehe, my ack on #1305 (removing the scratch space) came from my new nix-based script that builds a giant matrix of config params, including precomp tables of a few sizes including 24. doing multiple of these in parallel i wound up using 450gb of ram at one point. total runtime was 32 minutes for a one-commit PR
10:09 < sipa> How much RAM do you have?
10:18 < andytoshi> i just bought 1tb
10:18 < andytoshi> specifically because i wanted to be able to do this :P
10:18 < fanquake> how many sticks is that
10:18 < andytoshi> before i had 256gb, which i was straining against already, but it was manageable
10:18 < andytoshi> fanquake: 16
10:20 < andytoshi> fanquake: i have this motherboard https://www.newegg.com/asus-krpa-u16-amd-epyc-7002-generation/p/N82E16813119360
10:20 < fanquake> still got room for another TB
10:20 < sipa> @andytoshi It's absolutely ludicrous that you have half a million times more RAM than my first PC.
10:20 < fanquake> nice
10:21 < andytoshi> lol! to both of you
10:22 < sipa> @andytoshi Want to run your script on master? We merged a bunch of things.
10:22 < andytoshi> interestingly clang cannot handle a precomp table of size 24. at least with msan
10:23 < andytoshi> sipa: sure, one sec
10:23 < andytoshi> just started. it may be like an hour, there's something else running already
10:24 < andytoshi> on fb3a806365743261ca0f6158e154d338a851bbf9
10:26 < sipa> Sounds like you need that other Tb.
10:26 < andytoshi> :P
10:26 < andytoshi> so, i currently have 32 cores. there is a new proc on the market that has 96
10:27 < andytoshi> and if i had that, i think i could make use of 2tb
10:28 < sipa> Ok, sounds like you need more machines.
10:28 < andytoshi> annoyingly though the bottleneck on my current development pattern is usually my upload bandwidth. nix sends all the source tree, and often other build requiremnets (such as new nightly rust compilers), from my laptop to the build server. and i have only 20mbps up :(
10:28 < andytoshi> i am a little tempted to buy a local machine
10:29 < andytoshi> with the 96-core proc and stuff
10:29 < andytoshi> but right now there are still lots of "soft" speedups i can get ... e.g. by using cron to actively watch for new PRs overnight, at least ones from people i trust, and start building those ahead of time rather than waiting for me to wake up and start them
10:30 < andytoshi> and i could switch back to working directly on the build machine with mosh+tmux, though even with mosh it's still much nicer to work locally
10:52 -!- jonatack [~jonatack@user/jonatack] has quit [Ping timeout: 265 seconds]
10:52 -!- jonatack [~jonatack@user/jonatack] has joined #secp256k1
10:55 < andytoshi> sipa: done, all checks passed. it runs the full test suite, with and without msan, and runs ./tests 1 and exhaustive tests and ctime tests in valgrind (in non-msan mode). with and without recovery. with and without x86 asm. and with a bunch of different values for the ecmult window and ecmult gen precision
10:55 < andytoshi> ..oh lol, it's not trying both 32 and 64-bit scalar/field choices. i should fix that
10:56 < sipa> @andytoshi I think we may have merged more since :p
10:56 < sipa> Probably there will be a bunch more tomorrow.
10:57 < andytoshi> lol :P well, lemme add the widemul setting and i'll restart on current master. at some point i'll setup a cronjob as well, since i don't actively check the libsecp repo too often
10:58 < andytoshi> btw ./configure --help does not show widemul as an option
10:58 < andytoshi> i guess it's --with-test-override-wide-multiply .. which is not supposed to be twiddled with by normal users?
10:58 < sipa> Indeed.
11:00 < sipa> We believe our int size autodetection code for the widemul is good enough it never needs to be touched, except for testing.
11:00 < andytoshi> very nice
11:01 < andytoshi> ok, added the 3 possible values to my test matrix, re-fetched origin (only one PR merged, 1306) and restarted. curious how long it'll take now. in theory 90 minutes
11:19 -!- jonatack [~jonatack@user/jonatack] has quit [Quit: WeeChat 3.8]
11:53 < andytoshi> sipa: done. 52 minutes, not bad
11:53 < andytoshi> and i wrote a script that will go through all the recent-ish open PRs in a bunch of repos (including libsecp) and test them. will run that tonight (and maybe every night) and see how it works
11:54 < sipa> We were just talking about to set up a nightly CI task that can run a bigger set of combinations (so that we can reduce the per-PR ones).
11:55 < sipa> Sounds like we may just invoke you?
11:55 < andytoshi> at ~1hr/commit i may well need to reduce the number of combinations i do :P
11:55 < andytoshi> at least if development stays at the pace it's been this week..
11:56 < andytoshi> sipa: i'm not quite ready to commit to doing that (even a soft commit) .. but i think if this works out and turns out that it requires very little maintenance and "just works" for more than a week or so, i'd be happy to
13:28 -!- jonatack [~jonatack@user/jonatack] has joined #secp256k1
16:34 < andytoshi> kek my nix-daemon was oom-killed
17:32 -!- jonatack [~jonatack@user/jonatack] has quit [Quit: WeeChat 3.8]
17:36 -!- jonatack [~jonatack@user/jonatack] has joined #secp256k1
18:11 -!- preimage [~halosghos@user/halosghost] has quit [Quit: WeeChat 3.8]
18:40 -!- tromp [~textual@92-110-219-57.cable.dynamic.v4.ziggo.nl] has quit [Ping timeout: 240 seconds]
--- Log closed Fri May 12 00:00:59 2023