<karolherbst>
pepp: well.. we have emulation for most cards. There is a small range of nv4x cards which can do it, but yeah.. generally we have to emulate this feature
<karolherbst>
so this being enabled for all draws does degrade perf for us as we end up uploading to GPU and then mapping on the CPU side again and reading it out
<karolherbst>
this also fixes some old bugs we never got fixed :O
<karolherbst>
so maybe it was a good thing it broke so generally
<bbrezillon>
danvet: I skimmed through your "drm/sched" series at all panfrost changes look good to me, but you already had Steven's R-b, so I'm not sure mine is super useful
flacks has quit [Quit: Quitter]
flacks has joined #dri-devel
Hi-Angel has joined #dri-devel
Surkow|laptop has joined #dri-devel
MrCooper has quit [Remote host closed the connection]
MrCooper has joined #dri-devel
jrayhawk has quit []
Company has joined #dri-devel
jrayhawk has joined #dri-devel
FLHerne has joined #dri-devel
camus has quit []
agx has quit [Read error: Connection reset by peer]
agx has joined #dri-devel
mlankhorst has quit [Ping timeout: 480 seconds]
mlankhorst has joined #dri-devel
tzimmermann has joined #dri-devel
glennk has quit [Read error: Connection reset by peer]
glennk has joined #dri-devel
glennk has quit []
glennk has joined #dri-devel
agx has quit [Read error: Connection reset by peer]
agx has joined #dri-devel
itoral has quit []
<glennk>
anyone else bumped into libclc.pc having incorrect include/libexecdir paths?
agx has quit [Read error: Connection reset by peer]
agx has joined #dri-devel
<pepp>
karolherbst: is there a CAP describing this? (the display list code could be adapted to avoid uploading the IB to the GPU in this case)
<karolherbst>
pepp: that was removed
<karolherbst>
I think...
<karolherbst>
let me see
<karolherbst>
pepp: anyway, maybe it makes sense to add a CAP for that now or maybe there never was.. not sure
jewins has joined #dri-devel
sarnex has quit [Quit: Quit]
sarnex has joined #dri-devel
jernej_ has quit []
<karolherbst>
ehh.. how can I launch csgo from the command line? I need to debug it and gdbserver is so slow it's no fun
jernej has joined #dri-devel
<HdkR>
karolherbst: There's a couple of ways you can launch it with gdb
<HdkR>
Easiest way is likely just changing the launch arguments in steam to `DEBUGGER=gdb %command%` and launch through steam. It'll then be captured in gdb inside your terminal
<alyssa>
cmutt
<karolherbst>
HdkR: huh? mhh
<karolherbst>
interesting idea
<HdkR>
This only works in this regard because the csgo launch script understands that environment variable
<karolherbst>
yeah, I know
<HdkR>
Wrapping other games now can be a bit of a pain since Steam now sticks a "reaper" process between steam and the game launch. So you can't do game args as just `gdb --args %command%` anymore. It'll just capture the reaper program instead
<HdkR>
Not sure how to get around that one
<karolherbst>
HdkR: most games have shell wrappers anyway
<karolherbst>
HdkR: edit those shell wrappers? :p
<HdkR>
For those ones yes. I have quite a few games that just launch eg `dowser` which is an elf, and if you rename it to insert a script then it gets angry
<HdkR>
Hm. Maybe, I'll have to try something like that
<HdkR>
No wait, that would still capture reaper
<HdkR>
...I think
<karolherbst>
reaper should be on the outer side
<karolherbst>
what is reaper doing anyway?
<HdkR>
It's a lightweight application that is doing `prctl(PR_SET_CHILD_SUBREAPER)` to capture things trying to unparent themselves
<karolherbst>
ahh
<HdkR>
Fairly new thing. Was surprised when I noticed it just working under FEX :D
<HdkR>
"What is this new process stealing 128TB of VA?"
<karolherbst>
I guess game launchers to the unparent stuff like origin and so on?
<karolherbst>
*do
<karolherbst>
but that's usually wine domain
<HdkR>
I think the main thing that happens is an application's parent thread crashes, leaving some child processes that were reparenting to init and Steam couldn't track them anymore
<karolherbst>
mhh
<karolherbst>
maybe that as well
<HdkR>
Which I've seen happen all the time with steamwebhelper crashing. Maybe it needs to be wrapped in the reaper as well :P
<karolherbst>
:D
<karolherbst>
okay nice.. I can even launch games thoruhg ssh :)
<HdkR>
With x forwarding or something?
<karolherbst>
nope
<karolherbst>
just "DISPLAY=:0 steam 'steam://run/730'"
<HdkR>
oh right
<karolherbst>
and in the second shell I launched steam and get my gdb console :)
<karolherbst>
I only have one mouse/keyboard pair, but with multiple connection profiles. so I just wanted to remove the step of having to use the mouse on the other machine and stuff
<HdkR>
Not sure if there is an easy way to set launch options without the GUI
<karolherbst>
well.. changing the options isn't the issue, so much I can do without it getting annoying :D
<karolherbst>
relaunching games is just annoying
rsalvaterra_ has joined #dri-devel
<alyssa>
Incidentally I had a nightmare last night involving the Utgard renderer state word from lima
<alyssa>
anarsoul: ^^
<HdkR>
That's a horror
dviola has quit []
rsalvaterra has quit [Ping timeout: 480 seconds]
rsalvaterra_ has quit []
rsalvaterra has joined #dri-devel
iive has joined #dri-devel
dviola has joined #dri-devel
<karolherbst>
HdkR: whelp... it messes up stdin :/
<HdkR>
oh, hah.
<karolherbst>
but it seems like the first launch is _okay_
<karolherbst>
so worst case I just retart steam
imre has joined #dri-devel
imre has quit []
imre has joined #dri-devel
Duke`` has joined #dri-devel
cedric is now known as bluebugs
uis_ has quit [Read error: Connection reset by peer]
mattrope has joined #dri-devel
rsalvaterra_ has joined #dri-devel
rsalvaterra has quit [Ping timeout: 480 seconds]
Sumera[m] is now known as Sumera
nchery has joined #dri-devel
<karolherbst>
HdkR: turns out.. my high CPU usage under gdbserver might be my fault afterall :D
<karolherbst>
but this way I can actually restart the app from within gdb
<HdkR>
oops?
<karolherbst>
well...
<karolherbst>
gdbserver makes it worse
<karolherbst>
but I also use very annoying breakpoints
<karolherbst>
I get like 5 fps, with gdbserver I got like 5 spf
<imirkin_>
admittedly it's a GeForce 6200, so...
<karolherbst>
ehh no
<karolherbst>
it's on a gm206 now :)
<karolherbst>
I have a conditional breakpoint in a function which should get called like millions of times per frame or so...
<karolherbst>
maybe there is a shortcut I pressed by accident or so
<jekstrand>
my .gdbinit appears to be python
<karolherbst>
it is
<karolherbst>
gdb.execute('set print pretty on')
<jekstrand>
Ah
<karolherbst>
I have asan and tsan helpers :)
<karolherbst>
there might be a more python like way of setting such options, but in doubt use gdb.execute :D
<HdkR>
inside gdbinit you don't need to do gdb.execute for set
<jekstrand>
One of these days, I'm going to write a NIR extension for gdb
<karolherbst>
well, if it's python you have to
<karolherbst>
jekstrand: :D
<HdkR>
I just have two set lines which seems to work for me?
<karolherbst>
HdkR: sure, but I and I guess jekstrand as well, have python scripts for .gdbinit
<jekstrand>
HdkR: Mine starts with "python" so maybe it checks for that and decides whether to handle it as an ini file or a python script?
<karolherbst>
yeah
<jenatali>
Does GDB do NatVis or is that only VS/WinDbg?
<HdkR>
ah
<jekstrand>
jenatali: GDB has python-based extensions which can give you better pretty-print of stuff.
<jenatali>
Ah looks like someone wrote an extension for parsing natvis, that's cool
<jekstrand>
jenatali: Qt has support for it so all their base container types print nicely.
<jekstrand>
UE4 also has extensions, aparently.
<jenatali>
Yeah, on the Windows side of things, WinDbg can do C++ and Javascript extensions, and VS has its own extension mechanism, but they both now also support NatVis, which is an XML pretty-print description
<jenatali>
The nice thing about NatVis is that you can embed it in your code's debug symbols and the debugger will automatically load it
<jekstrand>
Oh, that's neat
<jenatali>
So if I was being selfish I'd say we should do NIR debugger pretty print stuff in that so we can share it cross-plat :P
<jenatali>
But if it's not well-supported by gdb it's probably not the right thing to do
<jekstrand>
Yeah, I doubt natvis would get enough use that people would notice when it breaks.
<jenatali>
Too bad
vivek has quit [Ping timeout: 480 seconds]
ngcortes has joined #dri-devel
rsalvaterra_ has quit []
rsalvaterra has joined #dri-devel
<anarsoul>
alyssa: hehe, is it that different from later malis? :)
mbrost has joined #dri-devel
<alyssa>
anarsoul: the rsw/rsd gets redesigned every gen
<alyssa>
valhall finally threw it out
<anarsoul>
so what kind of descriptor is it in valhall? :)
<alyssa>
split up into like 6 descriptors
<alyssa>
shader program desc, depth stencil desc, ...
<alyssa>
each desc locally looks like bifrost/midgard/utgard though :-p
<anarsoul>
I guess few more gens and they will reintroduce rsw
<anarsoul>
and the circle of gpu designs gets closed :)
mbrost_ has joined #dri-devel
frieder has quit [Remote host closed the connection]
sdutt has joined #dri-devel
vivek has joined #dri-devel
mlankhorst has quit [Ping timeout: 480 seconds]
mlankhorst has joined #dri-devel
gpoo has joined #dri-devel
xantoz has quit []
xantoz has joined #dri-devel
jkrzyszt has quit [Remote host closed the connection]
mbrost has quit [Ping timeout: 480 seconds]
mbrost_ has quit [Ping timeout: 480 seconds]
dviola has quit [Quit: WeeChat 3.2]
<alyssa>
dschuermann: Does NIR have a pass to shrink the load in "vec4 foo; .... foo.y" to load 32-bits (&foo + 4) instead of 64-bits &foo and then swizzle the y?
<alyssa>
opt_shrink_vector in the other direction, I guess
<jekstrand>
alyssa: We have an intel-specific pass: brw_nir_lower_mem_access_bit_sizes().
<jekstrand>
alyssa: I don't think there's a super-generic one.
<jekstrand>
The brw_nir_* pass does have some cleverness that could probably be turned into something generic, though.
<alyssa>
jekstrand: Gotcha.. I see `load_offset` in there, am I missing out on NIR for that too?
cphealy has quit [Ping timeout: 480 seconds]
<alyssa>
I see both global_constant and global_consant_offset so I guess we don't have global_offset
<alyssa>
(both valhall and agx let you add a constant offset with global load/store ... I could scalar_chase through 64-bit iadds at codegen time but that seems uncomfortable)
K`den has joined #dri-devel
Kayden has quit [Read error: Connection reset by peer]
K`den is now known as Kayden
<jekstrand>
alyssa: There's a load_global_constant_offset() which uses a 64+32 address format.
<jekstrand>
We get some small benefit from using that
yoslin has quit [Quit: WeeChat 3.2]
yoslin has joined #dri-devel
dcbaker has joined #dri-devel
<dcbaker>
alyssa: I had to pick over your "pan/mdg: Garbage collect silly quirk" to get jekstrand's "nir/lower_tex: Rework invalid implicit LOD lowering" to compile on 21.2, is that okay?
<alyssa>
dcbaker: Those two commits should be squashed together, we realized it was cherrypicked incorrectly when marge was about to merge it
<alyssa>
and didnt' think of implications for stable
<dcbaker>
no worries, it was easy enough to figure out, I just wanted to make sure there wasn't some bigger rework I was missing
tzimmermann has quit [Quit: Leaving]
<alyssa>
no, pan/mdg was just my fix for jekstrand's patch
rasterman has quit [Quit: Gettin' stinky!]
dviola has joined #dri-devel
lemonzest has quit [Quit: Quitting]
zackr has joined #dri-devel
danvet has quit [Ping timeout: 480 seconds]
<zackr>
i had to cherry-pick a fix to drm-misc-next-fixes and during dim push it apparently didn't apply cleanly to drm-tip and broke it. it's resolved and it was broken for only about a minute, but just in case any CI's catch it my apologies
<alyssa>
jekstrand: load_global_constant_offset() seems to be a flavour of load_constant