ChanServ changed the topic of #dri-devel to: <ajax> nothing involved with X should ever be unable to find a bar
<jekstrand> The FS optimizer is evil! Evil I say!
<Sachiel> throw holy water at it
nchery has quit [Ping timeout: 480 seconds]
<HdkR> Woo! Server back online
<alyssa> jekstrand: this seems... wrong
<alyssa> IBC when
ybogdano has quit [Ping timeout: 480 seconds]
<jekstrand> I think I've got it fixed. Should help TGL over-all by a bit.
<graphitemaster> Let it fight with itsellf, it should converge to the optimal non-fight state
<graphitemaster> That's how it works right?
<jekstrand> totally
<jekstrand> How does this patch delete loops?!?!?
<karolherbst> okay... I think https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17334 is finally ready :) (just missing some reviews, but I think the code is fine now)
<graphitemaster> changing how abs is lowered deletes for loops?
Danct12 has quit [Read error: Connection reset by peer]
<graphitemaster> that's impressively scary
<idr> jekstrand: Sometimes making a loop shorter by a single instruction can make it unroll.
<idr> It's also possible that the code the figures out loop counts can do it's job now that the xor is gone.
<jekstrand> idr: This is in the back-end
<idr> Oh.
Tooniis[m] has joined #dri-devel
<idr> That seems... unpossible.
co1umbarius has joined #dri-devel
<jekstrand> Right?
zzoon2627thholiday[m] has joined #dri-devel
soreau has quit [Read error: Connection reset by peer]
soreau has joined #dri-devel
columbarius has quit [Ping timeout: 480 seconds]
<jekstrand> I failed at shader-db
<jekstrand> *phew*
mbrost has joined #dri-devel
mclasen has quit [Ping timeout: 480 seconds]
mclasen has joined #dri-devel
kusma has joined #dri-devel
nchery has joined #dri-devel
mairacanal[m] has joined #dri-devel
pmoreau has joined #dri-devel
pmoreau is now known as Guest6845
emsls^ has joined #dri-devel
srslypascal is now known as Guest6847
srslypascal has joined #dri-devel
idr has quit [Quit: Leaving]
srslypascal is now known as Guest6848
srslypascal has joined #dri-devel
heat_ has quit [Ping timeout: 480 seconds]
Guest6847 has quit [Ping timeout: 480 seconds]
<graphitemaster> Suppose a GPU came out that brought back native matrix instructions like matrix mul
srslypascal has quit []
<graphitemaster> How difficult would it be to "raising" (is that the opposite of lowering) all the dot/mads produced by most compilers for matrix mul back into that native instruction
Guest6848 has quit [Ping timeout: 480 seconds]
srslypascal has joined #dri-devel
mclasen has quit [Ping timeout: 480 seconds]
doras has joined #dri-devel
cengizIO has quit [Quit: Leaving]
YuGiOhJCJ has joined #dri-devel
gagallo7[m] has joined #dri-devel
enick_747 has joined #dri-devel
saurabhg has joined #dri-devel
jenatali has joined #dri-devel
saurabhg has quit [Ping timeout: 480 seconds]
JohnnyonFlame has quit [Ping timeout: 480 seconds]
bmodem has joined #dri-devel
nielsdg has joined #dri-devel
bmodem has quit []
dcbaker has joined #dri-devel
mbrost has quit [Ping timeout: 480 seconds]
sigmaris_ has joined #dri-devel
Danct12 has joined #dri-devel
saurabhg has joined #dri-devel
Company has quit [Read error: Connection reset by peer]
aravind has joined #dri-devel
<mareko> difficult if some matrix elements are constants
emsls^ has quit [Remote host closed the connection]
pribas_ has quit [Ping timeout: 480 seconds]
ella-0[m] has joined #dri-devel
ella-0 has joined #dri-devel
r[m] has joined #dri-devel
off^ has joined #dri-devel
ella-0_ has quit [Read error: Connection reset by peer]
kts has joined #dri-devel
cwfitzgerald[m] has joined #dri-devel
aravind has quit [Ping timeout: 480 seconds]
aravind has joined #dri-devel
Mershl[m] has joined #dri-devel
ambasta[m] has joined #dri-devel
jasuarez has joined #dri-devel
egalli has joined #dri-devel
gdevi has joined #dri-devel
slattann has joined #dri-devel
pribas has joined #dri-devel
tleydxdy has joined #dri-devel
<airlied> sravn_: got time to give https://patchwork.freedesktop.org/series/105941/ another look?
pixelclu- has quit [Ping timeout: 480 seconds]
danvet has joined #dri-devel
jewins has quit [Ping timeout: 480 seconds]
aravind has quit [Ping timeout: 480 seconds]
off^ has quit [Ping timeout: 480 seconds]
off^ has joined #dri-devel
kallisti5[m] has joined #dri-devel
Danct12 has quit [Remote host closed the connection]
fab has joined #dri-devel
Danct12 has joined #dri-devel
tzimmermann has joined #dri-devel
fab has quit [Quit: fab]
feaneron has joined #dri-devel
cleverca22[m] has joined #dri-devel
JosExpsito[m] has joined #dri-devel
pixelcluster has joined #dri-devel
saurabhg has quit [Ping timeout: 480 seconds]
pixelcluster has quit [Ping timeout: 480 seconds]
<airlied> HdkR: you might have some ideas/opinions for https://gitlab.freedesktop.org/mesa/mesa/-/issues/6972
<HdkR> Some opinions you say
pixelcluster has joined #dri-devel
saurabhg has joined #dri-devel
pixelclu- has joined #dri-devel
<HdkR> airlied: Does dynamic indexing mean you need to support runtime dynamic texture format interpretation or something? Is that the reason for "texture engine"?
<airlied> HdkR: yeah I think dynamic texture format inter is the thing
lplc has quit [Ping timeout: 480 seconds]
<airlied> but also dynamic samplers
<airlied> we have a bunch of sampler state we drag in as static
<airlied> like all the wrap modes etc
<HdkR> How do you currently do dynamic format interpretation with the current implementation? GL already requires this as an example
<HdkR> Core GL only requiring interpretation on image store I think? Where NV has an extension for load
pixelcluster has quit [Ping timeout: 480 seconds]
<airlied> we lie and cross our fingers
<HdkR> pfft
<HdkR> I love it
<airlied> we actually create a big swithc statement
<airlied> which when you have texture/sampler as separatly dynamic becomes an unmanageably big switch statement
<airlied> with GL you always have texture/sampler the same, so it's one switch entry per image in the array
<HdkR> I think that ends up being the ideal case anyway, Maybe interleaving code samples together doing some call chain to reduce how much code is generated
saurabhg has quit [Remote host closed the connection]
<airlied> I leave it to LLVM to get things "optimised" but not sure if it really does that well
saurabhg has joined #dri-devel
<HdkR> Your BTB is going to hate you regardless if this is dynamic across every pixel
<airlied> yeah I think in theory it can be
* airlied assumes shaderSampledImageArrayNonUniformIndexing; is for that
sdutt has quit [Read error: Connection reset by peer]
<HdkR> Seems likely
<airlied> I think the trick is to fast path things when you know it's uniform, which we do now for some stuff
ngcortes has quit [Ping timeout: 480 seconds]
<airlied> zmike: do we want the NonUniformIndexing ones :-)
* airlied does see a lot of dropping out of vector mode into single wide :-P
pixelclu- has quit [Ping timeout: 480 seconds]
frieder has joined #dri-devel
lplc has joined #dri-devel
pixelcluster has joined #dri-devel
kts has quit [Ping timeout: 480 seconds]
mvlad has joined #dri-devel
lemonzest has joined #dri-devel
mvlad has quit [Quit: Leaving]
mvlad has joined #dri-devel
frieder has quit [Quit: Leaving]
frieder has joined #dri-devel
sul has quit [Ping timeout: 480 seconds]
frieder has quit []
sul has joined #dri-devel
frieder has joined #dri-devel
kts has joined #dri-devel
MajorBiscuit has joined #dri-devel
frieder has quit []
frieder has joined #dri-devel
TMM_ has quit []
TMM has joined #dri-devel
Duke`` has joined #dri-devel
pochu has joined #dri-devel
eukara has quit [Remote host closed the connection]
MajorBiscuit has quit [Ping timeout: 480 seconds]
frieder has quit [Quit: Leaving]
frieder has joined #dri-devel
Danct12 has quit [Remote host closed the connection]
Danct12 has joined #dri-devel
frieder has quit [Quit: Leaving]
frieder has joined #dri-devel
frieder has quit []
frieder has joined #dri-devel
frieder has quit [Quit: Leaving]
frieder has joined #dri-devel
frieder has quit []
frieder has joined #dri-devel
frieder has quit []
eukara has joined #dri-devel
Duke`` has quit [Read error: Network is unreachable]
Venemo_ has quit []
Venemo has joined #dri-devel
kts has quit [Ping timeout: 480 seconds]
MajorBiscuit has joined #dri-devel
eukara has quit []
kts has joined #dri-devel
pixelcluster has quit [Ping timeout: 480 seconds]
pixelcluster has joined #dri-devel
YuGiOhJCJ has quit [Quit: YuGiOhJCJ]
tursulin has joined #dri-devel
Haaninjo has joined #dri-devel
rasterman has joined #dri-devel
jkrzyszt has joined #dri-devel
fahien has joined #dri-devel
pixelcluster has quit [Ping timeout: 480 seconds]
rkanwal has joined #dri-devel
fab has joined #dri-devel
Company has joined #dri-devel
_kbingham has quit []
fab has quit [Quit: fab]
fab has joined #dri-devel
bmodem has joined #dri-devel
frieder has joined #dri-devel
frieder has quit []
fab has quit [Ping timeout: 480 seconds]
zehortigoza has quit [Ping timeout: 480 seconds]
zehortigoza has joined #dri-devel
kbingham has joined #dri-devel
rkanwal has quit [Quit: rkanwal]
Guest6793 has quit []
DragoonAethis has joined #dri-devel
frankbinns has joined #dri-devel
elongbug has joined #dri-devel
mclasen has joined #dri-devel
pixelcluster has joined #dri-devel
jkrzyszt has quit [Remote host closed the connection]
sjfricke has joined #dri-devel
sjfricke has quit []
jkrzyszt has joined #dri-devel
pixelcluster has quit [Ping timeout: 480 seconds]
JohnnyonFlame has joined #dri-devel
saurabhg has quit [Ping timeout: 480 seconds]
MrCooper_ is now known as MrCooper
kts has quit [Ping timeout: 480 seconds]
Major_Biscuit has joined #dri-devel
MajorBiscuit has quit [Ping timeout: 480 seconds]
vbt has joined #dri-devel
<vbt> Hello
<vbt> The above issue exists on linux 5.19 too
Danct12 has quit [Remote host closed the connection]
heat_ has joined #dri-devel
sravn_ is now known as sravn
<sravn> airlied: I considered the "rename CMA helpers" series ready, just waiting for any final comment from danvet - he is the contact for the todo item. I will take a look later today and give feedback on the series
<MrCooper> Frogging101: once upon a time, GPU clears were expensive enough that game engines tried to avoid them; were you maybe looking at an older game engine?
soreau has quit [Read error: Connection reset by peer]
MajorBiscuit has joined #dri-devel
soreau has joined #dri-devel
Major_Biscuit has quit [Ping timeout: 480 seconds]
kts has joined #dri-devel
fahien has quit [Ping timeout: 480 seconds]
saurabhg has joined #dri-devel
qyliss_ has quit []
qyliss has joined #dri-devel
jkrzyszt has quit [Ping timeout: 480 seconds]
<dv_> out of curiosity, do I see this correctly that if the periphery (hw video codec engine for example) support scatter-gather DMA, then CMA memory isn't necessary?
slattann has quit [Quit: Leaving.]
<zmike> airlied: yes
arnd__ has quit []
arnd has joined #dri-devel
Danct12 has joined #dri-devel
saurabhg has quit [Ping timeout: 480 seconds]
manmower is now known as ManMower
fahien has joined #dri-devel
<tarceri> digetx: I think everything is ready to merge, just one small nit on patch 1. otherwise looks good :)
<digetx> tarceri: cool, I'll update the patch 1 soon, thank you for the review!
danvet has quit [Ping timeout: 480 seconds]
tleydxdy has quit []
tleydxdy has joined #dri-devel
<tleydxdy> hmm, there seems to be lots of calls to drm_{,ctrc_}vblank_get in the kernel that doesn't check the return value, but drm_atomic_helper.c relies on the fact that get fails during modeset so I can't just put a warn when get failed
kts has quit [Ping timeout: 480 seconds]
eloy_ has joined #dri-devel
<LaserEyess> tleydxdy: amdgpu? I actually see that a lot, plenty of warnings about vblank/crtc ref counting errors
<LaserEyess> they don't seem to harm anything though
<tleydxdy> LaserEyess: yeah, thanks for checking out the patch btw
kts has joined #dri-devel
<LaserEyess> oh, you're very welcome
<tleydxdy> I realized that patch probably broke stuff elsewhere
<tleydxdy> so trying to find some more proper fixes
sdutt has joined #dri-devel
<LaserEyess> yeah...
<LaserEyess> Jul 30 21:37:25 mami kernel: [drm:dm_vblank_get_counter [amdgpu]] *ERROR* dc_stream_state is NULL for crtc '4'!
<LaserEyess> Jul 30 21:37:25 mami kernel: [drm:dm_crtc_get_scanoutpos [amdgpu]] *ERROR* dc_stream_state is NULL for crtc '4'!
<LaserEyess> that was the next bug I was going to report lol
<LaserEyess> but I got those before your patches as well
<LaserEyess> they also don't seem to harm anything, not that I can see at least
<tleydxdy> ye
<LaserEyess> should I try your second patch? the one you said to test separately?
<pq> isn't that some kind of power consumption thing? If you leak vblank references, the driver will never stop the vblank interrupts, which means the CPU keeps on being woken up for no good reason?
saurabhg has joined #dri-devel
alyssa has left #dri-devel [#dri-devel]
<Frogging101> MrCooper: Indeed I was
<Frogging101> Halo 3 in the Master Chief Collection
<tleydxdy> LaserEyess: I'm not sure if you are using vrr or not, if not then the patch probably doesn't work
<tleydxdy> pq: it's the other way around, get fails so when the matching put comes refcount becomes too low
<pq> ah
<MrCooper> Frogging101: hmm, I'd expect the Xbox 360 GPU to support fast clears though
alyssa has joined #dri-devel
<alyssa> Can other GPUs implement `(int) x >= y` in one instruction?
<alyssa> ('b2i32', ('uge', x, y)) -> some machine op
<alyssa> (On Mali, this is the comparison instruction in "return 0/1" mode instead of "return 0/-1" mode)
kmn has joined #dri-devel
<alyssa> (On AGX, this is a four-source CSEL "(x >= y) ? 1 : 0")
off^ has quit [Ping timeout: 480 seconds]
lkw has joined #dri-devel
bmodem has quit []
off^ has joined #dri-devel
<shadeslayer> alyssa: I think there's cmpxchng which could probably be used for that
<alyssa> shadeslayer: which ISA?
<LaserEyess> tleydxdy: I use vrr on all of my monitors
<shadeslayer> alyssa: on Intel
<LaserEyess> I always assumed this was the issue because vrr is a little underused on linux, so the fact that these bugs aren't readily caught means amd probably wasn't testing vrr stuff
<shadeslayer> alyssa: BRW_AOP_CMPWR
off^ has quit [Remote host closed the connection]
<tleydxdy> I can trigger the bug when I switch from compositor to vt, since it turn off the vrr mode and that mess up the refcount
<tleydxdy> but the call stack in your traces are different so I suspect somewhere else is also doing get when it shouldn't
<alyssa> (This obviously helps Mali and would help AGX with an appropriate backend optimizer. And should probably be no worse for other GPUs.)
<LaserEyess> tleydxdy: so, I *think* this gets triggered a lot for me when I do dpms off/on
<alyssa> shadeslayer: Mmm, tasty tasty 1500 page PDF
<LaserEyess> I have never really been able to reproduce this, but sometimes spamming dpms on/off commands trigger it
<shadeslayer> alyssa: Vol 2a of 14 ;)
<shadeslayer> but that's the only one that covers instructions afaict
<alyssa> Delight
<shadeslayer> for everything else
<alyssa> doesn't look like intel's CMP can do the b2i32 fused in
<LaserEyess> tleydxdy: perhaps the dpms + vrr combination, as well as multimonitor (?) is unique in my set up and causing this; that's a lot of variables to check though...
<shadeslayer> what's the nir op for something like that?
<alyssa> shadeslayer: b2i32
<shadeslayer> ah xD
<alyssa> it does look like this should be easier to schedule regardless
<tleydxdy> so something like spamming wlr-randr --output DP-3 --off; wlr-randr --output DP-3 --off?
* alyssa builds freedreno and tries to remember how to freedreno drm-shim for fun
<tleydxdy> *--on
<LaserEyess> tleydxdy: yes, basically, though with swaymsg on sway directly
<LaserEyess> I just did on/off in a loop hundreds of times a second
<tleydxdy> lol, that did lockup my system
<tleydxdy> one try tho
<alyssa> shadeslayer: Yes. The question is whether you can do anything better than the two-instruction sequence for `b2i32(cmp())`
<LaserEyess> oh..., right yes, on 5.16+ dpms does absolutely give me a total hang
<LaserEyess> on 5.15 and below I can *sometimes* trigger the vblank ref count thing
<shadeslayer> right, I was about to say, I don't know if we can do fused in instructions
Jeremy_Rand_Talos_ has quit [Remote host closed the connection]
<LaserEyess> tleydxdy: unfortunately this week/weekend would be a really bad time for me to test stuff but I'm certainly willing to sit down the weekend after and do some real debugging on this stuff, clean configs etc.
Jeremy_Rand_Talos_ has joined #dri-devel
<LaserEyess> I really just didn't take the time since I didn't think it would ever get fixed :')
jewins has joined #dri-devel
<tleydxdy> I'll see what I can find first ig, this is also not my day job, it's just crashing my desktop all the time so gotta fix it xd
<LaserEyess> oh
<LaserEyess> I wish it was someone's day job
sdutt has quit []
<alyssa> Hum. Patch seems to hurt ir3 shaderdb. Wonder why.
<alyssa> piles more nops, I guess
<alyssa> cat0/cat1 hurt but cat3 helped.. h
<tleydxdy> hmm, the crash seem to just be my compositor bugged somehow
Daanct12 has joined #dri-devel
<tleydxdy> using sway doesn't have this so far
<LaserEyess> depending on what crash we're talking about, sway definitely has it
melissawen_ has left #dri-devel [#dri-devel]
melissawen_ has joined #dri-devel
Danct12 has quit [Ping timeout: 480 seconds]
* alyssa shrugs
<alyssa> it's easy enough to do the optimization in a backend algebraic pass, no need to touch common
<alyssa> + (('b32csel', 'b@32', ('iadd', 'a@32', 1), a), ('iadd', a, ('b2i32', b))),
sdutt has joined #dri-devel
fxkamd has joined #dri-devel
<tleydxdy> hmm I set adaptive sync to on and refresh to 144, still not crashing sway
<tleydxdy> I'm running the second patch on 5.18
gawin has joined #dri-devel
<LaserEyess> I have 0 logs about what is causing the crash but 90% of the time it's dpms coming back on
<LaserEyess> interestingly enough, sometimes if I have an active ssh session, it crashes less often
Duke`` has joined #dri-devel
<tleydxdy> how often are we talking, I had the while true loop running for a few minutes, should I see it?
<tleydxdy> just wondering if it's just not reproing and I need to try stock kernel or it's super rare
kts has quit [Ping timeout: 480 seconds]
<alyssa> jekstrand: Is the second argument to extract_* and insert_* guaranteed to be constant?
<alyssa> seemingly yes
sul has quit [Ping timeout: 480 seconds]
<bl4ckb0ne> is there a reason hw why GL_OES_EGL_image is limited to GL_TEXTURE_2D or the spec is just not written for other kind of textures
<bl4ckb0ne> im trying to construct a cubemap from several EGLImage coming from DMABUF
<bl4ckb0ne> i think the other way around is possible
<bl4ckb0ne> like creating an EGLImage from the cubemap texture face and exporting it as a dmabuf
<danylo> jekstrand: in MR for common pipeline libraries helpers you mentioned that you need to figure out how to plumb custom renderpasses. Had you thought about it after that? I'm trying to use these helpers in Turnip and we have our own renderpass struct.
JoniSt has joined #dri-devel
<jekstrand> alyssa: yes
<jekstrand> danylo: I did think about that!
<jekstrand> danylo: If you do custom render passes, you need to look at the render pass and fill out a vk_subpass_info and pass that to `vk_graphics_pipeline_state_fill()`
<ajax> bl4ckb0ne: i think the spec's just not that ambitious
<danylo> jekstrand: Ah, thanks! Shame on me for not properly reading the functions I use =)
<ajax> sixteen years old, that extension
<bl4ckb0ne> could it be due for a round of upgrade
<bl4ckb0ne> after some search i dont think the other way around is possible
Duke`` has quit [Read error: No route to host]
<ajax> if you want to define how it should work for cubemaps, hack mesa until it works and then write an extension spec for how it works ;)
<bl4ckb0ne> i dont think its going to be very hard
<bl4ckb0ne> cubemap faces are 2d textures with a fancy name right?
<ajax> 2d array textures with six levels and a fixed-function binding, yeah
<bl4ckb0ne> there's a check in egl_image_target_texture that puts any texture that are not GL_TEXTURE_2D/EXTERNAL_OES as invalid
<ajax> layers, excuse me. mips have levels.
Haaninjo has quit [Quit: Ex-Chat]
<bl4ckb0ne> i think i have found my starting point
melissawen_ has left #dri-devel [#dri-devel]
TMM has quit [Ping timeout: 480 seconds]
melissawen has joined #dri-devel
<jekstrand> danylo: No worries. It's sort of hidden in there.
<bl4ckb0ne> ajax: but each layer is a 2d tex
idr has joined #dri-devel
<alyssa> jekstrand: :+1:
<zmike> done
<gawin> thank you
TMM has joined #dri-devel
<bl4ckb0ne> gah, immutable textures
pcercuei has joined #dri-devel
<jekstrand> bl4ckb0ne: What you're trying to do can't be done.
<jekstrand> bl4ckb0ne: You can't pull multiple dma-bufs into a single texture.
<jekstrand> Not without serious sparse resource black magic, and maybe not even then.
<bl4ckb0ne> can I pass 6 distinct textures and stitch them as a cubemap using a framebuffer
<jekstrand> nope
<bl4ckb0ne> what's im trying to achieve is sending a cubemap from one process to the other
<jekstrand> You need to use 6 distinct textures in your shader. That or do a blit to combine them.
<jekstrand> Ok, well that's a bit different
<LaserEyess> tleydxdy: we again we're talking about two different things here, if you mean #1247 that doesn't crash anything for me it just gives those vblank ref count warnings in my logs
<JoniSt> If you want to "just" get a cubemap from one process to another, why don't you use a single dma-buf for the entire cubemap texture?
<bl4ckb0ne> i tried but i dont think its possible
<jekstrand> Yeah, sending an entire cube map is more tractable but still dicey
<bl4ckb0ne> i thought about the other way around
<LaserEyess> tleydxdy: if you're talking about #1869 that is a hard hang that is triggered roughly 1/3 times on waking up my monitors from DPMS; I have no idea what causes that and the hang is so severe that it makes logs useless
<JoniSt> bl4ckb0ne: No, splitting one texture over multiple memory allocations is just not a thing
<bl4ckb0ne> oh its all done underneath?
<jekstrand> bl4ckb0ne: No, cubes are all one allocation
Duke`` has joined #dri-devel
<bl4ckb0ne> im using EGLImages to transform the texture as a dmabuf, is there another way?
<JoniSt> Yeah, it's just one contiguous piece of memory
<LaserEyess> I have no idea what triggers #1247, it happens seemingly randomly but I know it is related to dpms somehow; for #1869 100% of the time it happens is waking up my monitors
<bl4ckb0ne> https://registry.khronos.org/EGL/sdk/docs/man/html/eglCreateImage.xhtml i cant use a cubemap as a target, only faces
<tleydxdy> LaserEyess: I see, so #1247 has no way to repro reliably right now
<LaserEyess> nope, absolutely none
<LaserEyess> I *suspect* it is vrr + dpms related, I only suspect this because I assume this is not a setup people at AMD test
<LaserEyess> I can sometimes trigger the warning by spamming dpms events, but not always
<LaserEyess> I honestly think you're on the right track with some ref count not being checked with vrr, but I have no clue how amdgpu works and since it doesn't really affect my desktop I have not been willing to debug it further, especially with no one at AMD who seems to care or be able to reproduce it
alanc has quit [Remote host closed the connection]
<LaserEyess> I mean, to be fair, I can't even reproduce it myself
alanc has joined #dri-devel
<tleydxdy> in that case the second patch should only solve the vrr refcount issue
<tleydxdy> so if the other issue is somewhere else it won't get masked
<LaserEyess> yeah, I don't really knwo if they're related
<LaserEyess> #1247 is the only one I suspect is vrr related, I think I turned off VRR for a while and it went away or something
<tleydxdy> the more I look the more I realize the first patch is no good
<LaserEyess> #1869, I have no idea...
<LaserEyess> I need to debug with netconsole because there is no other way to get feedback on what in the fuck is happening
<LaserEyess> just been too lazy when 5.15 still works though
<tleydxdy> ye, if it works it works ;)
cengiz_io has joined #dri-devel
<LaserEyess> eventually there's going to be a new LTS kernel though so I'm pretty screwed if I don't debug/bisect before that
<bl4ckb0ne> i dont understand why EGL_GL_TEXTURE_CUBE_MAP_POSITIVE_X exists if cubemaps are one allocation
<tleydxdy> if only there's some way to repro it :(
Daaanct12 has joined #dri-devel
<ajax> bl4ckb0ne: https://registry.khronos.org/EGL/extensions/KHR/EGL_KHR_gl_image.txt says it's the <target> for eglCreateImageKHR. which kinda suggests that an EGLImage really is just a single 2d image
<bl4ckb0ne> so i should be able to make a cubemap out of it
<bl4ckb0ne> i dont see how i can get a dmabuf from a cubemap without using eglimage
<bl4ckb0ne> hm wait, would giving the cubemap as EGLClientBuffer to eglCreateImage give one EGLImage per cube face?
Daanct12 has quit [Ping timeout: 480 seconds]
bmodem has joined #dri-devel
bmodem has quit []
<bl4ckb0ne> after reading the ext, it is
lkw has quit [Ping timeout: 480 seconds]
saurabhg has quit [Ping timeout: 480 seconds]
lkw has joined #dri-devel
<alyssa> anholt: so, for deleting the broken idiv lowering
<alyssa> current users: panfrost, asahi, vc4, v3d, etnaviv, freedreno
<alyssa> I have MRs up for both panfrost compilers and will take care of asahi next
<alyssa> you up for any of vc4/v3d/freedreno?
<alyssa> I have half a mind to just bulk delete from the tree but letting shaderdb runs happen is probably good
gawin has quit [Ping timeout: 480 seconds]
TMM has quit [Ping timeout: 480 seconds]
TMM has joined #dri-devel
rsripada_ has quit []
rsripada_ has joined #dri-devel
rsripada_ has quit []
* alyssa wrote the patch for agx
<alyssa> not tested though, booted into the wrong OS...
ybogdano has joined #dri-devel
danvet has joined #dri-devel
fab has joined #dri-devel
lkw has quit [Quit: leaving]
dviola has quit [Ping timeout: 480 seconds]
gawin has joined #dri-devel
sul has joined #dri-devel
saurabhg has joined #dri-devel
TMM has quit [Ping timeout: 480 seconds]
Daanct12 has joined #dri-devel
ngcortes has joined #dri-devel
slattann has joined #dri-devel
eukara has joined #dri-devel
TMM has joined #dri-devel
Daaanct12 has quit [Ping timeout: 480 seconds]
fab has quit [Quit: fab]
dviola has joined #dri-devel
srslypascal has quit [Quit: Leaving]
srslypascal has joined #dri-devel
gawin has quit [Ping timeout: 480 seconds]
fahien has quit [Quit: fahien]
kts has joined #dri-devel
<sravn> danvet: OK that I push the "rename CMA helpers to DMA helpers" series? I will give it a second look and build test first ofc
pcercuei has quit [Quit: brb]
vbt has left #dri-devel [#dri-devel]
pcercuei has joined #dri-devel
eukara has quit []
eukara has joined #dri-devel
kem has quit [Ping timeout: 480 seconds]
nchery has quit [Remote host closed the connection]
pcercuei has quit []
pcercuei has joined #dri-devel
kem has joined #dri-devel
nchery has joined #dri-devel
Haaninjo has joined #dri-devel
gawin has joined #dri-devel
ybogdano has quit [Ping timeout: 480 seconds]
JoniSt has quit [Quit: KVIrc 5.0.0 Aria http://www.kvirc.net/]
Duke`` has quit [Read error: No route to host]
dviola has left #dri-devel [#dri-devel]
dviola has joined #dri-devel
sul has quit [Ping timeout: 480 seconds]
sul has joined #dri-devel
<alyssa> sdeafsbg
<alyssa> sdfb
<alyssa> gfsdb
<alyssa> gfsbgffgbx
alyssa has quit [Quit: leaving]
MajorBiscuit has quit [Ping timeout: 480 seconds]
tzimmermann has quit [Quit: Leaving]
<karolherbst> ahhh... why is "this" becoming NULL randomly...
<airlied> g
<karolherbst> throwing libasan into the mix.. how absurd is this.. I moved my delete[] call up and now it doesn't happen...
<sravn> airlied: For "rename CMA helpers to DMA helpers", the series can IMO go in. I have pinged danvet, and unless I hear otherwise then I will apply tomorrow. There is already enough a-b, r-b
Jeremy_Rand_Talos_ has quit [Remote host closed the connection]
Jeremy_Rand_Talos_ has joined #dri-devel
frankbinns has quit [Remote host closed the connection]
gawin has quit [Ping timeout: 480 seconds]
Danct12 has joined #dri-devel
Daanct12 has quit [Ping timeout: 480 seconds]
ybogdano has joined #dri-devel
tursulin has quit [Ping timeout: 480 seconds]
oneforall2 has quit [Remote host closed the connection]
gawin has joined #dri-devel
gouchi has joined #dri-devel
alyssa has joined #dri-devel
<airlied> sravn: thx
lemonzest has quit [Quit: WeeChat 3.5]
danvet has quit [Ping timeout: 480 seconds]
ybogdano has quit [Read error: Connection reset by peer]
ybogdano has joined #dri-devel
kts has quit [Quit: Konversation terminated!]
oneforall2 has joined #dri-devel
<airlied> jekstrand: so the nir lowering for descriptor indexing just lowers to a bunch of loops?
<jekstrand> airlied: Yup
<jekstrand> airlied: Not a bunch of loops, one per tex/image instruction
<airlied> I suppose that is as optimal as anything I could do on the backend
<alyssa> airlied: also see src/panfrost/bifrost/bi_lower_divergent_indirects.c for an alternate approach
<alyssa> that's probably worse for CPUs
oneforall2 has quit [Remote host closed the connection]
ybogdano has quit [Ping timeout: 480 seconds]
MajorBiscuit has joined #dri-devel
oneforall2 has joined #dri-devel
<karolherbst> jekstrand: writing those printf tests is a bit painful :D
ybogdano has joined #dri-devel
<jekstrand> karolherbst: That's not as bad as it could be, at least
<karolherbst> yeah...
<karolherbst> but figuring out what needs to be inside the buffer is a bit annoying :)
<jekstrand> karolherbst: Suggestion: Make your test case class a builder.
<jekstrand> You should have a new instance of it for each test
<karolherbst> that's already happening
<karolherbst> or what do you mean?
<alyssa> karolherbst: Seems nice :) though I'm known for writing questionable GTest
<karolherbst> gtest creates a new fixture instance for each test by default
<jekstrand> So add an add_arg() helper which you can call multiple times to add arguments.
<jekstrand> Should be a lot easier than constructing the buffer manually
ngcortes has quit [Ping timeout: 480 seconds]
<karolherbst> mhhh
<jekstrand> You can even make it templated to handle different sizes
<karolherbst> I guess that makes sense
MajorBiscuit has quit [Ping timeout: 480 seconds]
JohnnyonF has joined #dri-devel
<Frogging101> Where can I find the definitions of stuff used in NIR shader building? Like this one: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/amd/vulkan/radv_meta_bufimage.c#L208
<Frogging101> I can't find the definition of nir_load_push_constant for example
<alyssa> Frogging101: they're generated at build time
<Frogging101> ah
<alyssa> if you really want:
<alyssa> mesa/build/src/compiler/nir/nir_builder_opcodes.h
<Frogging101> I basically just want to know what this shader is doing and it's not very readable
<alyssa> (reading the generated builder prototypes isn't too weird ... I do it all the time for bifrost ...)
<alyssa> ultimately generated from mesa/src/compiler/nir/nir_opcodes.py
<alyssa> and mesa/src/compiler/nir/nir_intrincs.py
<alyssa> intrinsics
<Frogging101> I'll find the generated ones
JohnnyonFlame has quit [Ping timeout: 480 seconds]
<Frogging101> Can I get this in SPIR-V somehow?
<Frogging101> I tried RADV_DEBUG=metashaders but I didn't get any output
<Frogging101> actually spir-v is higher level than NIR, isn't it
<alyssa> correct
<alyssa> zmike: don't.
<alyssa> ;p
* zmike doesn't
TMM has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
TMM has joined #dri-devel
mclasen has quit [Ping timeout: 480 seconds]
gouchi has quit [Quit: Quitte]
<graphitemaster> AMDs new GL driver in 22.7.1 is actually ruining my day
MajorBiscuit has joined #dri-devel
<graphitemaster> Like the max SSBO binding index is now 32 when it used to be 64 in their previous driver so I worked around that. But now I'm getting shader link errors with no info log at all.
akselmo_ has left #dri-devel [#dri-devel]
alyssa has quit [Quit: leaving]
<graphitemaster> And strange compilation errors like "cannot convert from const float to temp uint" in code that has no uints
Jeremy_Rand_Talos_ has quit [Remote host closed the connection]
Jeremy_Rand_Talos_ has joined #dri-devel
eukara has quit []
rsripada has quit [Remote host closed the connection]
rsripada has joined #dri-devel
eukara has joined #dri-devel
ngcortes has joined #dri-devel
mvlad has quit [Remote host closed the connection]
Company has quit [Read error: Connection reset by peer]
<mareko> 22.7.1, what's that?
Company has joined #dri-devel
ella-0_ has joined #dri-devel
qyliss_ has joined #dri-devel
srslypascal_ has joined #dri-devel
sigmaris has joined #dri-devel
pochu_ has joined #dri-devel
TMM_ has joined #dri-devel
Dragoon has joined #dri-devel
TMM has quit [charon.oftc.net coulomb.oftc.net]
gawin has quit [charon.oftc.net coulomb.oftc.net]
Haaninjo has quit [charon.oftc.net coulomb.oftc.net]
pcercuei has quit [charon.oftc.net coulomb.oftc.net]
srslypascal has quit [charon.oftc.net coulomb.oftc.net]
tleydxdy has quit [charon.oftc.net coulomb.oftc.net]
MajorBiscuit has quit [charon.oftc.net coulomb.oftc.net]
elongbug has quit [charon.oftc.net coulomb.oftc.net]
DragoonAethis has quit [charon.oftc.net coulomb.oftc.net]
kbingham has quit [charon.oftc.net coulomb.oftc.net]
pochu has quit [charon.oftc.net coulomb.oftc.net]
qyliss has quit [charon.oftc.net coulomb.oftc.net]
cleverca22[m] has quit [charon.oftc.net coulomb.oftc.net]
feaneron has quit [charon.oftc.net coulomb.oftc.net]
gdevi has quit [charon.oftc.net coulomb.oftc.net]
egalli has quit [charon.oftc.net coulomb.oftc.net]
r[m] has quit [charon.oftc.net coulomb.oftc.net]
ella-0 has quit [charon.oftc.net coulomb.oftc.net]
ella-0[m] has quit [charon.oftc.net coulomb.oftc.net]
sigmaris_ has quit [charon.oftc.net coulomb.oftc.net]
enick_747 has quit [charon.oftc.net coulomb.oftc.net]
mairacanal[m] has quit [charon.oftc.net coulomb.oftc.net]
Tooniis[m] has quit [charon.oftc.net coulomb.oftc.net]
javierm has quit [charon.oftc.net coulomb.oftc.net]
Dragoon is now known as Guest6943
_kbingham has joined #dri-devel
HdkR has quit [Ping timeout: 481 seconds]
ds` has quit [Ping timeout: 480 seconds]
pcercuei has joined #dri-devel
<daniels> graphitemaster: this channel is not for reporting bugs and disappointments
<graphitemaster> i'm just venting okay
<daniels> I can see that
<zmike> perhaps it would be more productive to vent into a bug report where those interested might see it
Haaninjo has joined #dri-devel
<graphitemaster> oh i'm going to do that, once i finish venting
<graphitemaster> if i write the bug report while i'm frustrated it's just going to come off passive-aggressive and hurtful
<graphitemaster> which is not nice
slattann has quit []
<airlied> venting is what twitter is for :-P
<daniels> graphitemaster: that is how it came across here, yeah
oneforall2 has quit [Remote host closed the connection]
<karolherbst> jekstrand: https://gitlab.freedesktop.org/karolherbst/mesa/-/commit/97da4a61e2bece11f34fec4d44ea441c90e58df7 ... I have to see how that looks like with multiple formats/args, but I think it's not too bad like this
<jekstrand> karolherbst: What's up with use_format()?
ella-0[m] has joined #dri-devel
<karolherbst> jekstrand: that selects the format which is used, if you got multiple ones
<jekstrand> karolherbst: Why isn't that part of add_format()?
<karolherbst> you can have e.g. two formats, but 5 printf invocations
<jekstrand> Ah
<karolherbst> like... in a loop or something
<jekstrand> So use_format goes with add_arg()?
<karolherbst> yes
<jekstrand> Ok, makes sense.
<karolherbst> I think add_arg doesn't really need the index except for string literal args
<karolherbst> so that part is a little confusing
gdevi has joined #dri-devel
pcercuei has quit [Quit: dodo]
<karolherbst> mhh...
<karolherbst> maybe I store the value passed to use_format, so add_arg won't need it...
tleydxdy has joined #dri-devel
<karolherbst> the only purpose is really just so that the literal string gets added to the format string
<jekstrand> karolherbst: I think what you've got is fine.
<jekstrand> I just got a bit confused.
<karolherbst> putting any other number is wrong though :)
<karolherbst> and once I add ints/floats, that index wouldn't be part of the overloaded function...
<karolherbst> yeah... I think I just store it, so it's easier to use/read
saurabhg has quit [Ping timeout: 480 seconds]
javierm has joined #dri-devel
ds` has joined #dri-devel
HdkR has joined #dri-devel
JohnnyonF has quit [Read error: Connection reset by peer]
JohnnyonFlame has joined #dri-devel
cengiz_io_ has joined #dri-devel
ds- has joined #dri-devel
cengiz_io has quit [Ping timeout: 480 seconds]
ds` has quit [Read error: Connection reset by peer]
ds- is now known as ds`
cengiz_io has joined #dri-devel
<karolherbst> uhh...
<karolherbst> vectors are broken
<karolherbst> well.. kind of
<karolherbst> as long as they are at the begining it is fine
enick_747 has joined #dri-devel
<karolherbst> tomorrow me will deal with that :)
cengiz_io_ has quit [Ping timeout: 480 seconds]
oneforall2 has joined #dri-devel
dreda has quit [Remote host closed the connection]
dreda has joined #dri-devel
dreda is now known as Guest6954
jkrzyszt has joined #dri-devel
rasterman has quit [Quit: Gettin' stinky!]
mclasen has joined #dri-devel
oneforall2 has quit [Remote host closed the connection]
Tooniis[m] has joined #dri-devel
oneforall2 has joined #dri-devel
egalli has joined #dri-devel
simon-perretta-img has quit [Quit: Leaving]
r[m] has joined #dri-devel
feaneron has joined #dri-devel
heat_ has quit [Ping timeout: 480 seconds]