| 25 Sep 2025 |
 outfoxxed | trigger what? | 21:19:21 |
| outfoxxed | the vector crash or the nixpkgs eval fail | 21:19:35 |
 dramforever | the vector misalignment | 21:19:42 |
| dramforever | yeah what would be vector crash on k1 | 21:19:55 |
| outfoxxed | it looked like it happened during kernel module loading | 21:20:02 |
| outfoxxed | by systemd | 21:20:06 |
 Tristan Ross | Wtf | 21:20:09 |
| dramforever | that's not really significant | 21:20:10 |
| dramforever | it could be anywhere | 21:20:13 |
| outfoxxed | its linux 6.6.63 | 21:20:12 |
| outfoxxed | I don't really know what else I can tell you | 21:20:25 |
| dramforever | since you set v in gcc.arch it could happen to anything and anywhere | 21:20:40 |
| dramforever | kernel module loading is just one of the first thing it does on boot | 21:21:54 |
 Alex | In reply to @rosscomputerguy:matrix.org
I want to try a RISC-V + LLVM kernel NB: this will likely take longer to build because of needing to change stdenv. | 22:41:06 |
| Tristan Ross | Ik | 22:41:23 |
| Tristan Ross | I run an LLVM kernel on my desktop | 22:41:31 |
| Alex | In reply to @rosscomputerguy:matrix.org
I run an LLVM kernel on my desktop Is there any noticeable difference? | 22:42:04 |
| Tristan Ross | Kinda | 22:42:24 |
| Tristan Ross | The noise from my audio interface is less noticable | 22:42:34 |
| Tristan Ross | And pkgsLLVM.linux took 20 minutes with a GCC built kernel where with the LLVM kernel, it took 15 minutes | 22:43:02 |
| Tristan Ross | I ran the same build under the two kernels 3 times in a row on each kernel and got the same 5 minutes difference | 22:43:45 |
| Alex | In reply to @rosscomputerguy:matrix.org
The noise from my audio interface is less noticable It sounds like there is interference between your CPU and your audio interface...
I mean, it shouldn't be driving the sound card differently just because of differences in compiler optimisations and code layout, right? | 22:44:24 |
| Tristan Ross | The thing is I'm not sure | 22:44:56 |
| Tristan Ross | I shouldn't even be getting noise in the first place since it's over USB. Even on my dedicated USB card, I actually get worse audio quality. | 22:45:22 |
| Alex | In reply to @rosscomputerguy:matrix.org
I ran the same build under the two kernels 3 times in a row on each kernel and got the same 5 minutes difference This could be due to ld.bfd vs lld (assuming these are what the two stdenvs use for linking). Worth exploring... | 22:45:31 |
| Tristan Ross | I was thinking about asking some people at Ampere since I do have an Ampere CPU and the performance team has done a lot of testing as I've heard. | 22:46:13 |
| dramforever | lld should be faster than ld.bfd at processing linker relaxations... maybe | 22:48:05 |
| dramforever | i'm not sure how you would go about profiling this though | 22:48:21 |
| dramforever | oh nevermind it's not really building a riscv kernel | 22:48:42 |
| Alex | In reply to @dramforever:matrix.org
i'm not sure how you would go about profiling this though IIRC there is a function for "just change the linker" for use with replaceStdenv, so the two can be compared in isolation. | 22:49:19 |
