00:09 <HdkR> alyssa_: If cwabbott takes over the bifrost shader stuff then maybe I'll spend more spare cycles on Vulkan stuff :P

00:10 <alyssa_> HdkR: If Connor takes bf shaders, you can take up bf cmdstream? :P

00:10 <HdkR> That too

00:11 <HdkR> Who says they can't come up at the same time?

00:11 <alyssa_> All known laws of Newtonian mechanics

00:13 <HdkR> I'm pretty excited about it still :P

00:13 <HdkR> With the new Vulkan mesa driver there is even more material to pull data from

00:14 <Lyude> I haven't had the energy to continue working on the ra stuff since flying home from vacation ;_;

00:14 <Lyude> really hoping I can start getting to it this weekend

00:15 <HdkR> It hasn't really moved on my end since I've been busy

00:15 <HdkR> Which should change in....2 weeks?

00:15 <HdkR> 3 weeks?

00:16 <alyssa_> You silly people with real life

00:20 <alyssa_> I'd like to wack the ctx->fragment abstraction entirely

00:20 <alyssa_> And also to cleanly separate MFBD and SFBD, even at the risk of a bit of code repitition

00:33 <alyssa_> tomeu: Just pushed your DEBUG patches, thanks! :)

04:22 <alyssa_> tomeu: Lot more to go but I've been taking a stab at properly refactoring the fragment stuff (v2 of the other half of the series I guess) in a personal branch (fragment-yoi I think, last letters were keyboard mashing) if you're curious

04:22 <alyssa_> Feeling good about it this time

05:42 <tomeu> alyssa_: nice, will give it a look later today

05:43 <tomeu> alyssa_: btw, any suggestions for a test scenario in which a gpu reset is needed?

14:51 <alyssa_> tomeu: I would think any time we fault?

14:51 <tomeu> but for some reason, without any resets, things keep working afterwards

14:52 <tomeu> so I'm guessing there's some faults more catastrofics than others

14:52 <alyssa_> Interesting

14:52 <tomeu> in any case, I'm just doing what kbase does

14:52 <alyssa_> tomeu: In that case, if you really want to go wild, I found a super bad bit which actually hangs the GPU, no fault

14:52 <tomeu> I'm not sure of what would be a good way of testing though

14:52 <tomeu> ah, that sounds interesting

14:53 <alyssa_> No clue what the intended purpose is but it should do what you need

14:53 <tomeu> best would be to have tests in igt, but I don't know how to do that without bringing in all the cmdstream thing

14:53 <alyssa_> tomeu: TBH, don't bother?

16:07 <Lyude> tomeu: ask the i915 guys what they do maybe?

16:07 <Lyude> I know i915 has a bunch of tests for testing GPU resets

16:08 <tomeu> good idea

16:20 <tomeu> robher: do you remember where this 8 came from?

16:20 <tomeu> if (job_read(pfdev, JS_STATUS(js)) == 8) {

16:20 <tomeu> in panfrost_job_timedout

16:22 <robher> tomeu: I think that's 'active' which is what it was stuck at before we got things working.

16:25 <robher> tomeu: I plan to make gpu va 0 fault. So just passing that as a jc address would be good test.

16:25 <tomeu> would that put the GPU in a non-working state?

16:27 <robher> tomeu: I think so. I couldn't get the job to stop otherwise.

16:27 <tomeu> ah, cool

16:30 <robher> tomeu: change drm_mm_init call to: drm_mm_init(&pfdev->mm, SZ_2M, SZ_4G);

16:30 <robher> SZ_2M for large page alignment.

16:31 <robher> tomeu: I think I have an IGT test, but it will be with the old ABI.

16:34 <robher> tomeu: turns out I already pushed it. https://gitlab.freedesktop.org/robh/igt-gpu-tools/tree/panfrost