ChanServ changed the topic of #dri-devel to: <ajax> nothing involved with X should ever be unable to find a bar
<robmur01> yeah, until Android exclusively ships bleeding-edge mainline kernels, that would require vendors to upstream drivers 2-3 years before they've even designed the SoC ;)
<robmur01> typical vendors are writing brand-new code against GKI branches based on 5.10 or 5.15 (which may eventually get forward-ported to mainline and thrown over the wall if we're lucky)
<CounterPillow> I was about to sarcastically describe why they might not bother with upstreaming but then I remembered I should not start flamewars with people who might work for companies I eventually seek employment at.
mbrost has quit [Ping timeout: 480 seconds]
<karolherbst> jenatali: uhh... I have to figure out from a spir-v if `-cl-single-precision-constant` was set... I think...
<jenatali> Oh?
<jenatali> Why?
<karolherbst> to set treat_doubles_as_floats correctly, no?
<jenatali> Nah, we can just delete that flag
<karolherbst> ohhh
<karolherbst> yeah, fair enough :)
<karolherbst> that makes everything easier :D
<jenatali> If cl-single-precision-constant was set, then you won't see doubles in the printf string unless they were really doubles
<karolherbst> right...
<karolherbst> and if it's in the spirv, the user is asking for it anyway...
<jenatali> Otherwise, clang does the C thing of promoting floats to double because double is the "native" floating point type
<karolherbst> yeah.. I'm more concerned about spirv as the input, not clc
<karolherbst> as there it's not in our control how it's compiled
<karolherbst> so I wonder if we want to keep that option anyway, but only depend on if fp64 support is enabled for the device or not?
<karolherbst> and then we can force fp32 for spir-vs
<jenatali> Ah yeah I see what you're saying, that makes sense
kzd has quit [Quit: kzd]
<karolherbst> I'll just push `-cl-single-precision-constant` to the frontend and see how that works out :)
Jeremy_Rand_Talos__ has quit [Remote host closed the connection]
Jeremy_Rand_Talos__ has joined #dri-devel
<jenatali> Yeah, that's what I've done
<karolherbst> mhh...
<karolherbst> it's not using fp64 constants by default...?
<karolherbst> was there some condition for it?
kzd has joined #dri-devel
<karolherbst> ehh wait...
<karolherbst> it's 32 bit in the spirv, but 64 bit in the LLVM...
<karolherbst> mhhhhh
<karolherbst> interesting...
<karolherbst> jenatali: looks like llvm just handles that for us now 🙃
<karolherbst> unless I pass "+cl_khr_fp64" or "+__opencl_c_fp64" to clang the constants all end up as fp32 ones in the spirv...
<karolherbst> but I'm also seeing "call spir_func i32 (i8 addrspace(2)*, ...) @printf(i8 addrspace(2)* noundef getelementptr inbounds ([8 x i8], [8 x i8] addrspace(2)* @.str, i64 0, i64 0), float noundef 0x4024B0F280000000)"...
<karolherbst> yeah.. it becomes "call spir_func i32 (i8 addrspace(2)*, ...) @printf(i8 addrspace(2)* noundef getelementptr inbounds ([8 x i8], [8 x i8] addrspace(2)* @.str, i64 0, i64 0), double noundef 1.034560e+01)" with enabled fp64
<karolherbst> ... that also explains why I have seen regressions in my spirv based runs and not the clc ones...
<karolherbst> I think that changed in llvm-16 actually...
<karolherbst> jenatali: https://reviews.llvm.org/D24235 mhhh
<karolherbst> like that's 7 years ago...
<karolherbst> ohhhhhhhhhhhhhh
<karolherbst> I know
<karolherbst> I've fixed this ....
<karolherbst> I suspect my work on forcing enabled features made us not claim fp64 support and all that
<karolherbst> so we don't run into this with llvm-14+
<jenatali> Oh cool
<karolherbst> I guess I will only handle the spir-v part then :D
<karolherbst> but yeah.. that explains it then
<jenatali> karolherbst: I added that when I was still using 12 I think
<karolherbst> yeah...
<karolherbst> that's why I was confused because that llvm check is there since llvm-8?
<karolherbst> ehh
<karolherbst> llvm-4 even
<karolherbst> could also been that `-USPIRV` and `-USPIR` part...
<karolherbst> as that would otherwise enable rando extes
<karolherbst> *exts
<karolherbst> apparently not cl_khr_fp64
<karolherbst> jenatali: anyway.. the reason I ran into the 64 bit arg mess was because of `%s` args as they are 64 bit anyway ....
<jenatali> Right
<karolherbst> and then encode ids like... 0x1de
<karolherbst> maybe I really just fix the callbacks and move on...
macromorgan has quit [Quit: Leaving]
macromorgan has joined #dri-devel
<karolherbst> oh well.. I'll decide it tomorrow then
yyds has joined #dri-devel
soreau has quit [Ping timeout: 480 seconds]
columbarius has joined #dri-devel
co1umbarius has quit [Ping timeout: 480 seconds]
soreau has joined #dri-devel
yyds has quit [Remote host closed the connection]
yyds has joined #dri-devel
aravind has joined #dri-devel
benjaminl has quit [Remote host closed the connection]
benjaminl has joined #dri-devel
<gfxstrand> karolherbst: Did you test the alignment patches or just read them over?
Company has quit [Quit: Leaving]
mbrost has joined #dri-devel
yyds has quit [Remote host closed the connection]
yyds has joined #dri-devel
yyds has quit [Remote host closed the connection]
dviola has joined #dri-devel
mbrost has quit [Ping timeout: 480 seconds]
frankbinns1 has joined #dri-devel
frankbinns has quit [Ping timeout: 480 seconds]
mripard has quit [Ping timeout: 480 seconds]
aravind has quit [Ping timeout: 480 seconds]
alatiera has quit [Quit: Connection closed for inactivity]
mripard has joined #dri-devel
oneforall2 has quit [Remote host closed the connection]
oneforall2 has joined #dri-devel
glennk has joined #dri-devel
mbrost has joined #dri-devel
aravind has joined #dri-devel
Duke`` has joined #dri-devel
fab has joined #dri-devel
benjaminl has quit [Remote host closed the connection]
benjaminl has joined #dri-devel
tomeu has joined #dri-devel
<tomeu> how do people figure out what specific test might have caused a GPU hang when it only happens under high concurrency inside deqp-runner?
<tomeu> anholt_: maybe you have some tricks?
kzd has quit [Quit: kzd]
Duke`` has quit [Ping timeout: 480 seconds]
kts has joined #dri-devel
yyds has joined #dri-devel
glennk has quit [Ping timeout: 480 seconds]
fab has quit [Quit: fab]
kts has quit [Quit: Konversation terminated!]
bmodem has joined #dri-devel
kts has joined #dri-devel
kts has quit []
kts has joined #dri-devel
sima has joined #dri-devel
docelwilif^ has quit [Remote host closed the connection]
tshikaboom has joined #dri-devel
tomba has joined #dri-devel
fab_ has joined #dri-devel
fab_ is now known as Guest9387
mvlad has joined #dri-devel
frieder has joined #dri-devel
oneforall2 has quit [Remote host closed the connection]
apinheiro has joined #dri-devel
oneforall2 has joined #dri-devel
sghuge has quit [Remote host closed the connection]
sghuge has joined #dri-devel
crabbedhaloablut has joined #dri-devel
glennk has joined #dri-devel
rgallaispou has joined #dri-devel
donaldrobson has joined #dri-devel
eukara has quit [Ping timeout: 480 seconds]
tursulin has joined #dri-devel
benjaminl has quit [Remote host closed the connection]
benjaminl has joined #dri-devel
kts has quit [Ping timeout: 480 seconds]
i509vcb has quit [Quit: Connection closed for inactivity]
hansg has joined #dri-devel
rasterman has joined #dri-devel
crabbedhaloablut has quit []
lynxeye has joined #dri-devel
crabbedhaloablut has joined #dri-devel
frankbinns1 is now known as frankbinns
vliaskov has joined #dri-devel
<karolherbst> gfxstrand: I've ran them, just that they didn't do anything for the issue I was seeing as I was also hitting an LLVM bug. But I can run the CL CTS with those and see if anything break if you want
flynnjiang has joined #dri-devel
tzimmermann has joined #dri-devel
flynnjiang has quit [Quit: flynnjiang]
a-865 has quit [Ping timeout: 480 seconds]
glennk has quit [Ping timeout: 480 seconds]
anujp has joined #dri-devel
glennk has joined #dri-devel
TMM has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
TMM has joined #dri-devel
cmichael has joined #dri-devel
anujp has quit [Remote host closed the connection]
<karolherbst> jenatali: actually... for spir-v the printf double thing is also not a problem, because the only valid way of doing is to have a spir-v requiring "OpCapability Float64", which isn't valid to pass to the runtime if it doesn't support cl_khr_fp64 :D
<karolherbst> I think I'd rather fix my spirv testing here...
<karolherbst> and we can indeed drop that option
<karolherbst> uhh.. why is `-cl-single-precision-constant` ignored on the clang CLI...
<karolherbst> needs to be -Xclang -cl-ext=-all ...
mbrost has quit [Ping timeout: 480 seconds]
<karolherbst> yeah....
<karolherbst> -cl-single-precision-constant doesn't work at all...
<karolherbst> at least now how we are using clang...
bmodem has quit [Ping timeout: 480 seconds]
<karolherbst> *sigh*
<karolherbst> not in the mood of debugging this honestly :D
kts has joined #dri-devel
heat has joined #dri-devel
kts has quit [Ping timeout: 480 seconds]
<karolherbst> oh actually.. it works in libclang.. I forgot to switch to online compilation.. oops
<karolherbst> ohh it doens't.. I forgot to enable fp64.. uhhh
<karolherbst> whatever...
<karolherbst> I spending too much time on printf...
<karolherbst> *I'm
<dj-death> karolherbst: speaking of printf
<dj-death> karolherbst: you have any opinion on those changes: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26505
aravind has quit [Ping timeout: 480 seconds]
<karolherbst> looks good
Company has joined #dri-devel
<karolherbst> I'll leave comments on the MR
<dj-death> karolherbst: thanks
<dj-death> karolherbst: I think possibly the only controversial thing is to have the printfs on all shader stages
<karolherbst> yeah
<dj-death> in the serialization
<karolherbst> that's where I'll leave a comment :D
<dj-death> I can probably workaround that for my case
<karolherbst> nah
<karolherbst> I have a better idea
<dj-death> thanks a lot
<karolherbst> done
YuGiOhJCJ has quit [Quit: YuGiOhJCJ]
<karolherbst> jenatali: treat_doubles_as_floats seems to be broken, as the printf code still parses it as 8 bytes and generates a wrong value...
<karolherbst> so we'd need to update the printf info here
<karolherbst> but anyway.. I think it's safe to remove
<karolherbst> just need to fix clang to respect -cl-single-precision-constant or figure out what we'd need to change or something...
<karolherbst> (in case we do support fp64)
glennk has quit [Ping timeout: 480 seconds]
glennk has joined #dri-devel
yyds has quit [Remote host closed the connection]
<jenatali> karolherbst: Right... I think we'd need it as a separate pass then so it can be before lowering the printf struct to explicit types
<jenatali> So yeah, just delete it I think
bmodem has joined #dri-devel
alatiera has joined #dri-devel
<karolherbst> gfxstrand: no regressions at least...
<jenatali> karolherbst: Tag me in it so I get an email so I remember to review it later, it's not even 5am here yet
<karolherbst> fair
<karolherbst> good night then :P
<karolherbst> done
tzimmermann has quit [Quit: Leaving]
a-865 has joined #dri-devel
<frankbinns> mripard: ack
glennk has quit [Ping timeout: 480 seconds]
tzimmermann has joined #dri-devel
<mripard> frankbinns: merged, thanks
<frankbinns> I had totally missed that one so thanks for the ping
tomba has quit [Quit: Leaving]
tomba1 has quit []
tomba1 has joined #dri-devel
tomba1 has quit []
tomba1 has joined #dri-devel
tomba1 is now known as tomba
tomba_ has joined #dri-devel
bmodem has quit [Ping timeout: 480 seconds]
simon-perretta-img has quit [Ping timeout: 480 seconds]
simon-perretta-img has joined #dri-devel
tomba is now known as Guest9413
tomba_ is now known as tomba
Guest9413 has quit []
enick_335 has joined #dri-devel
enick_335 has quit []
tomba is now known as Guest9415
tomba has joined #dri-devel
Guest9415 has quit []
<gfxstrand> karolherbst: Okay, cool.
<gfxstrand> karolherbst: Did you verify it fixed the kernel you were looking at?
<karolherbst> sadly it doesn't as it only has one shared variable (so the location is 0 anyway) and the align_mul=1 comes from LLVM being silly :'(
<karolherbst> I just noticed that we don't handle the Alignment thing on variables while looking into it
simon-perretta-img has quit [Ping timeout: 480 seconds]
simon-perretta-img has joined #dri-devel
<karolherbst> but I could probably come up with a test to verify
glennk has joined #dri-devel
<karolherbst> wondering if you have a better idea :)
<karolherbst> (and yes, it should loop over the mov as it could be a mov chain)
<karolherbst> this is needed with opaque pointers and we ending up with some movs along the way
yyds has joined #dri-devel
<jenatali> karolherbst: Run copy prop before whatever pass was blowing up?
<karolherbst> that's in vtn
<karolherbst> `vtn_add_printf_string`
<karolherbst> so.. no :)
<jenatali> Ohh
<gfxstrand> karolherbst: Ugh... I don't want to think about printf. :sob:
<karolherbst> gfxstrand: it's your lucky day, it's not a printf specific issue :P
<karolherbst> just if you have movs in your deref chain
<karolherbst> and you are still in vtn
<gfxstrand> karolherbst, eric_engestrom: Is there a reason why the gitlab CI's rustfmt is only running on lib.rs or app.rs?
<karolherbst> gfxstrand: this pulls in the entire project
<karolherbst> otherwise you'd run it on all files multiple times
<gfxstrand> karolherbst: Only if you have a lib.rs. :P
<karolherbst> what's your file called then?
<gfxstrand> nak.rs
<karolherbst> mhh.. could rename to lib.rs 🙃
<karolherbst> but I guess...
<gfxstrand> Yeah, I probably should.
<karolherbst> you kinda want to have that to be able to structurize your code, but yeah...
<karolherbst> worst case we just add custom file names
<gfxstrand> Yeah, I can rename to lib.rs
<karolherbst> `lib.rs` and `app.rs` is just what most rust projects use and it's a cargo default
<karolherbst> ehh
<karolherbst> lib.rs and main.rs
<karolherbst> I should fix the app.rs one...
<eric_engestrom> yeah it's trivial to add nak.rs
<eric_engestrom> app.rs should be gone, unless I missed something somewhere?
<karolherbst> yeah.. it should be main.rs instead
<karolherbst> but we can also remove it
<karolherbst> and we could also add nak.rs
<gfxstrand> eric_engestrom: I think I'm going to just restructure nak a bit
<eric_engestrom> ack
<karolherbst> I have a handy git alias btw: rustfmt = ! git ls-files */{lib,main}.rs | xargs rustfmt
<karolherbst> (which I would have to update for nak... 🙃)
<eric_engestrom> gfxstrand: I'll drop this MR then, but now you know which line to edit when you have your new filename :)
Guest9387 has quit []
<eric_engestrom> karolherbst: if you have `shopt -s globstar` in your shell you can simply do `rustfmt **/lib.rs` :)
<karolherbst> but I just want to do that on git tracked files tho :D
<karolherbst> otherwise it might pull in generated files and such
<karolherbst> but yeah...
<eric_engestrom> well, in ci it's run on `src/**/lib.rs`, if you do the same locally that will not include generated files unless you put your builddir in src/ :P
yyds has quit [Remote host closed the connection]
<gfxstrand> I think that should do the trick
<gfxstrand> I would actually like an ACK or two on that MR. Just to make sure I didn't do anything exceptionally stupid.
<gfxstrand> Maybe look at the meson a bit.
fab has joined #dri-devel
<karolherbst> you don't necessarily have to split it up on a crate level
<karolherbst> that reminds me.. I wanted to move to rust.proc_macro at some point :)
<karolherbst> but yeah.. I think they are fine
<gfxstrand> karolherbst: Yeah, I know it's not necessary. But, if my crates are going to be lib.rs..
<MrCooper> hmm, does MR Label Maker not know about zink yet?
<gfxstrand> Is MR Label Maker working?
<karolherbst> it should be fixed now
<karolherbst> but it was broken for a while
<gfxstrand> Okay
<gfxstrand> I was trying to get it to label my dot product MR
<gfxstrand> But whaver. I think I found them all manually
mbrost has joined #dri-devel
macslayer has joined #dri-devel
<karolherbst> yeah... we've debugged the server a few hours ago
<gfxstrand> Ah
<MrCooper> thanks
immibis_ is now known as immibis
immibis is now known as immibis_
immibis_ has left #dri-devel [#dri-devel]
immibis has joined #dri-devel
aravind has joined #dri-devel
Jeremy_Rand_Talos_ has joined #dri-devel
Jeremy_Rand_Talos__ has quit [Remote host closed the connection]
kzd has joined #dri-devel
Daanct12 is now known as Danct12
tshikaboom has quit []
mvlad has quit [Remote host closed the connection]
jsa has joined #dri-devel
tshikaboom has joined #dri-devel
aravind has quit [Remote host closed the connection]
tzimmermann has quit [Quit: Leaving]
Duke`` has joined #dri-devel
Haaninjo has joined #dri-devel
tshikaboom has quit []
rgallaispou has quit [Quit: Leaving.]
asriel has quit [Quit: Don't drink the water. They put something in it to make you forget.]
Danct12 has quit [Quit: ZNC 1.8.2 - https://znc.in]
Danct12 has joined #dri-devel
asriel has joined #dri-devel
kasper93_ has joined #dri-devel
kasper93 has quit [Ping timeout: 480 seconds]
cmichael has quit [Quit: Leaving]
apinheiro has quit [Ping timeout: 480 seconds]
apinheiro has joined #dri-devel
oneforall2 has quit [Remote host closed the connection]
oneforall2 has joined #dri-devel
mvlad has joined #dri-devel
TMM has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
TMM has joined #dri-devel
pcercuei_ has joined #dri-devel
pcercuei has quit [Ping timeout: 480 seconds]
mbrost has quit [Ping timeout: 480 seconds]
pcercuei_ has quit [Ping timeout: 480 seconds]
alanc has quit [Remote host closed the connection]
alanc has joined #dri-devel
lynxeye has quit [Quit: Leaving.]
benjaminl has quit [Ping timeout: 480 seconds]
donaldrobson has quit [Ping timeout: 480 seconds]
frieder has quit [Remote host closed the connection]
hansg has quit [Quit: Leaving]
ced117 has quit [Ping timeout: 480 seconds]
JohnnyonFlame has joined #dri-devel
ced117 has joined #dri-devel
gouchi has joined #dri-devel
benjaminl has joined #dri-devel
rasterman has quit [Quit: Gettin' stinky!]
ced117 has quit [Ping timeout: 480 seconds]
sarnex has quit [Read error: No route to host]
glennk has quit [Ping timeout: 480 seconds]
sarnex has joined #dri-devel
pcercuei has joined #dri-devel
fab has quit [Quit: fab]
fab has joined #dri-devel
fab has quit []
fab has joined #dri-devel
mandelstays has joined #dri-devel
<DemiMarie> mattst88: sorry about that.
<DemiMarie> robmur01: I had not considered that Android doesn’t ship bleeding-edge kernels. That explains so much!
<DemiMarie> I’m a bit salty about Android using proprietary drivers because it means no virtio-GPU native context support, which means no decently secure GPU acceleration in pKVM VMs.
tursulin has quit [Ping timeout: 480 seconds]
YuGiOhJCJ has joined #dri-devel
fab has quit [Quit: fab]
pac85 has joined #dri-devel
mbrost has joined #dri-devel
i509vcb has joined #dri-devel
alyssa has quit [Quit: alyssa]
Duke`` has quit [Ping timeout: 480 seconds]
gouchi has quit [Remote host closed the connection]
sima has quit [Ping timeout: 480 seconds]
Haaninjo has quit [Quit: Ex-Chat]
illwieckz has quit [Ping timeout: 480 seconds]
crabbedhaloablut has quit []
thelounge14738 has quit []
thelounge14738 has joined #dri-devel
apinheiro has quit [Quit: Leaving]
glennk has joined #dri-devel
mbrost has quit [Ping timeout: 480 seconds]
mbrost has joined #dri-devel
jsa has quit []
pcercuei has quit [Quit: dodo]
<mareko> DemiMarie: Android has virtio
<mareko> also native context