09:11nomore201: hello! I have a problem with discrete graphics and suspend which I reported here: https://bbs.archlinux.org/viewtopic.php?id=240354
09:11nomore201: do you have an idea about what is happening and how I can debug this?
09:59RSpliet: nomore201: we recommend against using bumblebee on a system with nouveau, as you probably read on the Arch wiki. I don't know whether the daemon might interfere with Prime's ability to put the discrete GPU to sleep when it's idle after suspend.
10:00RSpliet: But either way, standard procedure apply: get the latest kernel, obtain us more info (like a full unfiltered dmesg, potentially with some debugging messages turned on if the base log doesn't reveal anything)
10:29nomore201: RSpliet: ofc I'm not using nouveau with bumblebee, I evaluated both scenarios: 1) plain nouveau 2) bumblebee+bbswitch with the results described in the post
10:31RSpliet: nomore201: Thanks for that clarification
10:33RSpliet: I suspect Lyude is the person with the most insight on these kind of issues. Be aware that the X.org Developers Conference is about to kick off, so devs could be slow to respond and in a different timezone than usual
10:34RSpliet: The asynchronous nature of the bugtracker might be helpful to manage those delays ;-) Take a peek on https://nouveau.freedesktop.org/wiki/Bugs/ for hints on how to gather data for a bug report that'll minimise the risk of bouncing back and forth.
10:35nomore201: I'll provide some logs, but usually dmesg shows nothing more than warnings like "proc_thermal 0000:00:04.0: Unsupported event [0x84]"
10:36nomore201: Yes, probably filing a bug it's better
10:36RSpliet: That's okay. Absence of messages is information too
10:36RSpliet: Thanks in advance!
10:37nomore201: Which debug levels should I use in this particular case? https://nouveau.freedesktop.org/wiki/KernelModuleParameters/
10:37nomore201: (I'm using archlinux with latest kernel 4.18.9)
10:37RSpliet: That's... actually a very good question!
10:38RSpliet: For now I'd assume a default debugging level, they might ask you to retry with debug level "debug" for certain engines turned on. I'm unable to tell you which engines might be relevant, sorry
10:44nomore201: I guess the default debugging is "nouveau.debug=debug" right?
16:55mslusarz_: RSpliet: what kind of problems? (WRT OpenCL benchmarks and mmt)
17:01pmoreau: mslusarz: Kernels that end up using double operations because they forgot the 'f' suffix on a floating-point literal. (They might be more than that, but that’s the issue he was talking about.)
17:02pmoreau: That wouldn’t really show up when looking at the code itself, but if you look at the disassembly of the binary, it becomes way easier to spot f64 operations.
17:43RSpliet: msluarz_: ah sorry, no I didn't mean problems with the mmt tools (although I have trouble tracing cuda kernels... but for me that now has a negative prio to figure out), but thanks for asking!
17:43RSpliet: pmoreau: well in this case it was using the wrong constant (/macro) :-D