<marex>
uh, if I run MESA_SHADER_CACHE_DISABLE=true ./bin/tex-miplevel-selection *GradARB 2D -png -auto , what exactly does the output PNG show me ?
<LaserEyess>
does amd not implement dmabuf modifiers? I'm trying to figure out why I can't get direct scanout on wayland working with the latest radv/radeonsi
<LaserEyess>
it works with a similar software stack on intel for both anv and iris
<LaserEyess>
GPU is a 5700XT
FireBurn has quit [Quit: Konversation terminated!]
<Lynne>
airlied: are the headers stable enough to finally work on vulkan video again?
<Lynne>
they were an utter mess a month ago, and the lack of justification for the changes made me give up
dj-death has quit [Ping timeout: 480 seconds]
iive has joined #dri-devel
<danvet>
hwentlan____, agd5f just scrolled through the psr-su stuff, might be good to extract the "calculate total crtc damage" code from i915 in intel_psr2_sel_fetch_update, stuff that into damage helpers and reuse for amdgpu
<danvet>
and then dc just has the single su rect on the dc_pipe as the interface to the hw code
heat has joined #dri-devel
<danvet>
there's probably more drivers somewhere which do the crtc damage calc
lumag_ has quit [Ping timeout: 480 seconds]
<danvet>
we do have drm_atomic_helper_damage_merged but that's just the plane damage for upload to the buffer, not the crtc damage for psr su or similar things
<danvet>
robclark, pinchartl do we really not have any other driver that does crtc damage rects?
<robclark>
IIRC msm probably could for dsi cmd mode.. not sure if that is a thing drm-hwc supports?
<robclark>
it's pretty much only phones that have cmd mode panels
<danvet>
robclark, we have damage rects since a while
<robclark>
we do use the damage helper to trigger full screen flush on cmd mode phones.. but other than that I've not really looked much at it.. pretty much all the hw I have is video mode or something other than dsi
<danvet>
hwentlan____, agd5f https://lists.freedesktop.org/archives/amd-gfx/2022-May/078768.html <- commit message on this is a bit funny, since if you use the helpers right you always get damage information, just when it's from userspace that doesn't set explicit damage it's just always the entire plane
<danvet>
so maybe update that to reflect reality (and then also less hand rolling of stuff)
<danvet>
robclark, yeah a few quick git grep didn't show up anything interesting
<danvet>
there's a bunch of self-refresh helper use, but that's not damage tracking
<danvet>
just a helper to shut down/restart the entire display hw as needed
<robclark>
it's possible that qcom has something downstream? If so abhinav__ might know about it
<danvet>
yeah I'm expected all android disp drivers to have a massive thing downstream
<danvet>
robclark, also might be that desktop compositors are better at this, since it helps for virtual hw and stuff due to copy operations needed
<danvet>
those are the currently overwhelming cases of driver support for damage rects
<robclark>
qcom has been working incrementally to get downstream things (like writeback) upstream.. but I guess damage rect tracking might not be the highest on the priority list? It isn't something we have a strong upstream need for, other than just reducing downstream delta
<danvet>
robclark, I'm a bit surprised tbh, are the cmd mode panels simply more expensive and not worth it wrt battery time?
<danvet>
for arm cros devices I mean
<robclark>
I think the savings are negligible on a larger panel on tablets/detachables, where backlight is larger part of the power draw (but you also have a larger battery).. and all the clamshells are eDP because mipi can't do the longer signal length
DanaG has quit [Remote host closed the connection]
DanaG has joined #dri-devel
heat has quit [Remote host closed the connection]
dllud has quit []
dllud has joined #dri-devel
<karolherbst>
DanaG: Well, looks like a driver bug to me
<DanaG>
Do you happen to know if this card is supposed to have fixed PCIe reset? I think some Navi cards had it fixed.
<karolherbst>
DanaG: dunno, but amdgpu shouldn't leave sysfs around after unloading, so this has to be fixed in the driver regardless
<karolherbst>
anyway, non of this is officially supported, so you are kind of on your own
<karolherbst>
and what do you mean by reloading anyway? Normally you'd rmmod and modprobe and not using the pci reset thing (which nobody should ever use as long as a driver is bound)
Haaninjo has quit [Quit: Ex-Chat]
<DanaG>
I mean, my other GPU is using amdgpu too, so I couldn't just unload the module.
<karolherbst>
well, you can unbind a device from the driver
<karolherbst>
there are bind/unbind methods where the PCI address gets written into