01:28 <thefossguy> Gotcha :)

01:40 <thefossguy> The sun doesn't seem too bright today (metaphorically speaking)

06:19 <davidlt[m]> rwmjones: want more work? :)

08:49 <rwmjones_> davidlt[m]: morning, yes, but not right now

08:50 <rwmjones_> I want to try to concentrate on making my build system a bit faster first

08:50 <rwmjones_> I was going to try to install a qemu VM and benchmark it against the real hardware

09:05 <davidlt[m]> rwmjones_: Note, you need to benchmarks. One that is exclusive single-thread bound and another one that can go very parallel.

09:05 <davidlt[m]> For example, compiling kernel most likely is faster on QEMU if you can go -j32.

09:06 <davidlt[m]> Well, okay, things might stop scaling well after -j8 :)

09:06 <davidlt[m]> If it's a stupidly long configure script that executing tests one by one, the board is faster.

09:08 <davidlt[m]> Of course this might not apply if you have something like 13900KS running at extremely high freq or something :)

09:09 <davidlt[m]> I tried to find some benchmarks for QEMU (freq. vs cache size), but nothing interesting on the Internet.

09:09 <davidlt[m]> Anything even if QEMU wins today, probably it already looses to TH1520 and others.

10:27 <davidlt[m]> Oh, another patch for TH1520 showed up on the mailing list.

11:25 <rwmjones> davidlt[m]: we're still limited to 8 vCPUs on qemu I guess?

11:25 <rwmjones> IIRC it was some problem with the interrupt mechanism

11:27 <davidlt[m]> No, it's up to 32 vCPUs.

11:27 <davidlt[m]> Well, the scaling might not be that great. I am running 32-core VM, but never spend too much time testings 8 vs 16 vs 32 vCPUs.

11:28 <davidlt[m]> My biggest VM is 32 vCPU 80GB RAM.

11:29 <davidlt[m]> Note, you need the latest QEMU and must restrict it to Sv39.

11:29 <davidlt[m]> Well, I need to update instructions for the latest QEMU :)

13:06 <davidlt[m]> From: https://wiki.haskell.org/GHC/FAQ#When_I_try_to_start_ghci_.28probably_one_I_compiled_myself.29_it_says_ghc-5.02:_not_built_for_interactive_use

13:06 <davidlt[m]> rwmjones_:... (full message at <https://libera.ems.host/_matrix/media/v3/download/libera.chat/3f1e1599a37b70306530ab109cbfb62766d520c0>)

13:08 <davidlt[m]> But yeah, it was built with the same version.

13:08 <davidlt[m]> It also used to work in the past. Looking at ghcid build logs.

13:16 <thefossguy> <davidlt[m]> "Note, you need the latest QEMU..." <- *restrict* it to sv39? OoO

13:17 <davidlt[m]> Yes

13:18 <thefossguy> How do I check if the hw booted with what amount of addr space support?

13:18 <thefossguy> I saw this in the firmware log but can't see it in dmesg

13:19 <davidlt[m]> Check DT?

13:19 <thefossguy> good idea :D

13:21 <davidlt[m]> See: https://github.com/qemu/qemu/commit/6f23aaeb9be13d4a26b7f863533c316105481640

13:22 <davidlt[m]> I get (for virt machine):... (full message at <https://libera.ems.host/_matrix/media/v3/download/libera.chat/40deb08089adb023fa01dc5ee9eb2f726ecdbcbe>)

13:23 <thefossguy> Just realized that sv39 is what current hardware supports :sweat_smile:

13:23 <thefossguy> I thought sv48 was the "current" one

13:23 <thefossguy> My bad

13:25 <davidlt[m]> I get:... (full message at <https://libera.ems.host/_matrix/media/v3/download/libera.chat/b18b61622b63fa3656a5f772ccff2021a3c49c01>)

13:25 <davidlt[m]> I think TH1520 might support Sv48.

13:26 <davidlt[m]> Ah, their DT has it at Sv39.

13:27 <davidlt[m]> The same is with C920 cores.

13:27 <davidlt[m]> It's Ventana Veyron 1 that has Sv48.

13:38 <rwmjones> cool, I will build a 32 core VM in that case :-/

13:42 <thefossguy> > As per the specification, sv64 must support sv57, which in turn must support sv48...etc. So we can restrict machine support by simply setting the "highest" supported mode and the bare mode is always supported.

13:42 <thefossguy> TIL :)

15:38 <davidlt[m]> Oh, golang 1.20 has a fix for this already: https://github.com/golang/go/commit/1e3c19f3fee12e5e2b7802a54908a4d4d03960da

15:39 <davidlt[m]> Not sure about others.