Cursed project end of day 1: clinfo runs

[karolherbst@chaos.social]

oh shoo.. I'm like 10% slower than Nvidia's implementation 🙃

[karolherbst@chaos.social]

you all don't want to know the most cursed part about this 🙃

[oblomov@sociale.network]

@karolherbst we do actually, and you know we do 8-D

[karolherbst@chaos.social]

Pass 3769 Fails 112 Crashes 4 Timeouts 0

Something something atomics... and a few other random things, should be like 3 or 4 bugs in total...

[karolherbst@chaos.social]

okay... I found one of the atomic bugs.. it's when two kernels are launched back to back and apparently they can interact weirdly with each other. When I force a flush+wait between them those fails go away... curious

[karolherbst@chaos.social]

or maybe it's an ordering issue? mhh

[oblomov@sociale.network]

@karolherbst how would two kernels affect each other, unless they stomp on each other's memory? Did the hardware support concurrent kennel execution?

[karolherbst@chaos.social]

@oblomov either that or just executed in a different order

[karolherbst@chaos.social]

Pass 3871 Fails 10 Crashes 4 Timeouts 0

I think that's good enough for an initial MR 🙃

[karolherbst@chaos.social]

Here it is: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37831

Hopefully that's not too much of a shock.

I don't have a blog, so my MR is the blog post I guess?

[oblomov@sociale.network]

@karolherbst there's hardware whose queues are OOO?

[oblomov@sociale.network]

@karolherbst interesting! Pocl does something similar too too the same effect, BTW.

And if you can fix typos, there's a cl_gl_shring that needs an a ;-)