ChanServ changed the topic of #dri-devel to: <ajax> nothing involved with X should ever be unable to find a bar
<cmarcelo>
alyssa: probably I wasn't sure whether all gallium drivers would be able to support it. the SPIR-V for GL comes with GL 4.6, so there might be assumptions that those shaders make about what's there for GL...?
<cmarcelo>
maybe I was too conservative? (-:
<cmarcelo>
the GL_ARB_gl_spirv requires 3.3, so we should be fine in that camp.
<karolherbst>
maybe we can nuke that variable pointer cap as well? not sure if that requires anything from drivers either
<cmarcelo>
maybe start enabling it for other drivers until all are happy and drop the cap
<karolherbst>
I think the problem is rather that there aren't that many tests for it, are there?
<karolherbst>
probably nobody would notice the bugs anyway
<cmarcelo>
there's some on piglit, but I remember some shenanigans to run them.. maybe was fixed now.
<karolherbst>
I think the CTS has like 8 tests? Once I enabled that on nouveau it just worked
<cmarcelo>
karolherbst: yeah, I remember it was not really much there.
rotationmatrix has quit [Remote host closed the connection]
<cmarcelo>
karolherbst: still would be nice at least do a sanity check before enabling/nuking. maybe some driver is unhappy with some NIR coming out of our spirv parser.
<cmarcelo>
sanity check of the drivers that don't enable it yet
<karolherbst>
yeah.. that's what I would be worried about
rotationmatrix has joined #dri-devel
sh-zam has joined #dri-devel
sh_zam has quit [Read error: Connection reset by peer]
<jenatali>
I'd assumed that since there was a cap I'd need to do something special to handle it
Leopold has quit [Remote host closed the connection]
Scorpi_ has joined #dri-devel
<ElementW>
Hmm, I'm having some display issues but I'm not sure where to start investigating and/or reporting a bug... Basically my display has the great idea of completely disconnecting itself when I turn it off (yes, that means its EDID is gone too), and I can consistently make amdgpu emit a kernel warning if that happens while the screen is off using DPMS. Kwin can recover if I make it release and regain DRM master thru VT switching, but Sway is in
<ElementW>
complete shambles when that happens
YuGiOhJCJ has quit [Quit: YuGiOhJCJ]
<ElementW>
The warning is `drm_WARN_ON(atomic_read(&vblank->refcount) == 0)` at drivers/gpu/drm/drm_vblank.c:1210 so I'm not sure if it's an actual kernel driver issue or if my Wayland compositors just try and use the display even when the kernel signalled it's gone
<karolherbst>
thought there are more crashes now, but maybe I misremembered
<Scorpi_>
karolherbst: Hi! I built mesa from git master to play around with rusticl. I set RUSTICL_ENABLE=radeojsi. clinfo shows that rusticl is used, but no devices are found and there are 2 messages "radeonsi: driver missing"
<karolherbst>
maybe I should try to get it applied to llvm-15
<jenatali>
karolherbst: Got another fix for you
<karolherbst>
yeah.. I think your upper patch looks fine. we shouldn't convert samplers to textures here I think :)
<Scorpi_>
karolherbst: okay, I think rebuilding llvm is something that can be done when it's dark outside...
<karolherbst>
or use llvm-git if possible
<karolherbst>
uhm.. packages I mean
<karolherbst>
but... I guess you don't want to kill your system
<Scorpi_>
karolherbst: I intend to just rebuild the Debian package with the patch applied
<karolherbst>
you'll have to fix the patch a little, also not sure if that even works before llvm-15
<Scorpi_>
Debian Sid has llvm-15
<karolherbst>
okay, still need to adjust the patch a little (it touches tests which aren't there with llvm-15)
<Scorpi_>
okay
user34234 has joined #dri-devel
bluepqnuin is now known as bluepenquin
user34234 has quit [Quit: Konversation terminated!]
rasterman has joined #dri-devel
user34234 has joined #dri-devel
<user34234>
how do I use amdgpu.virtual_display with recent kernels? Xorg will just crash on startup. Is there some secret? I can't find anything mentioned in manuals
<user34234>
i'm running it on bare haredware, no SR-IOV, IOMMU or anything like that. No display hardware attached either
heat has joined #dri-devel
kts has joined #dri-devel
<karolherbst>
jenatali: okay, seems like that fixed it :)
<karolherbst>
now let's see if anything else still works
<jenatali>
Yep
<jenatali>
Dozen is a little picky but I'd say both of these fixes produce better looking code
<karolherbst>
yeah, and it's overall more consistent anyway
<jenatali>
Yeah
<jenatali>
And this is why I'm so glad we have our drivers in CI
Haaninjo has joined #dri-devel
glennk has quit [Remote host closed the connection]
glennk has joined #dri-devel
sarnex has quit [Quit: Quit]
<karolherbst>
okay, seems like the other drivers are fine as well
<karolherbst>
jenatali: I saw some people talking about a potential MSVC compiler bug, best place to file those? (the compiler turns a conditional store into an unconditional one, still correct in terms of single threaded correctness tho