Sam[m]5 has quit [Remote host closed the connection]
cvtsi2sd has joined #dri-devel
cvtsi2sd has quit [Remote host closed the connection]
reductum has quit [Quit: WeeChat 2.8]
Lightkey has quit [Ping timeout: 480 seconds]
boistordu_old has joined #dri-devel
boistordu has quit [Ping timeout: 480 seconds]
Lightkey has joined #dri-devel
sdutt has quit [Remote host closed the connection]
fly_hawk[m|gr] has joined #dri-devel
fly_hawk[m|gr] has quit [Remote host closed the connection]
gpoo has joined #dri-devel
ppascher has joined #dri-devel
mhy[m] has joined #dri-devel
mhy[m] has quit [Remote host closed the connection]
mbrost has quit [Remote host closed the connection]
sdutt has joined #dri-devel
sdutt has quit []
sdutt has joined #dri-devel
ppascher has quit [Ping timeout: 480 seconds]
camus has joined #dri-devel
Duke`` has joined #dri-devel
gpoo has quit [Ping timeout: 480 seconds]
boistordu_old has quit []
boistordu_old has joined #dri-devel
boistordu_old has quit []
lemonzest has joined #dri-devel
boistordu_ex has joined #dri-devel
sdutt has quit [Ping timeout: 480 seconds]
pnowack has joined #dri-devel
pnowack has quit []
alanc has quit [Remote host closed the connection]
alanc has joined #dri-devel
JoostKopano[m] has joined #dri-devel
JoostKopano[m] has quit [Remote host closed the connection]
pnowack has joined #dri-devel
ppascher has joined #dri-devel
thellstrom1 has joined #dri-devel
thellstrom has quit [Remote host closed the connection]
i-garrison has joined #dri-devel
danvet has joined #dri-devel
mlankhorst has joined #dri-devel
pekkari has joined #dri-devel
thellstrom has joined #dri-devel
danvet has quit [Ping timeout: 480 seconds]
thellstrom1 has quit [Ping timeout: 480 seconds]
pnowack has quit [Quit: pnowack]
CornFly has joined #dri-devel
CornFly has quit [Remote host closed the connection]
pcercuei has joined #dri-devel
yoslin_ has joined #dri-devel
yoslin_ has left #dri-devel [#dri-devel]
yoslin has quit [Ping timeout: 480 seconds]
yoslin has joined #dri-devel
Enock4seth9 has joined #dri-devel
Enock4seth9 has quit [Remote host closed the connection]
gouchi has joined #dri-devel
pmonson has joined #dri-devel
RobertC has joined #dri-devel
pmonson has quit [Remote host closed the connection]
yk has quit [Remote host closed the connection]
RobertC has quit [Ping timeout: 480 seconds]
andrey-konovalov has joined #dri-devel
Frigolit has joined #dri-devel
Frigolit has quit [Remote host closed the connection]
lord4163 has joined #dri-devel
lord4163 has quit [Remote host closed the connection]
whosgonna has joined #dri-devel
whosgonna has quit [autokilled: Suspected spammer. Mail support@oftc.net with questions (2021-06-12 11:12:46)]
luzipher_ has joined #dri-devel
tobiasjakobi has joined #dri-devel
luzipher__ has quit [Ping timeout: 480 seconds]
luzipher__ has joined #dri-devel
luzipher_ has quit [Ping timeout: 480 seconds]
blue__penquin has joined #dri-devel
couchkitten16 has joined #dri-devel
couchkitten16 has quit [Remote host closed the connection]
karolherbst has quit [Ping timeout: 480 seconds]
yoslin has quit [Quit: WeeChat 3.1]
Nuc1eoN2 has joined #dri-devel
Nuc1eoN2 has quit [autokilled: Suspected spammer. Mail support@oftc.net with questions (2021-06-12 12:35:26)]
yoslin has joined #dri-devel
Bad_K4rMa has joined #dri-devel
Bad_K4rMa has quit [Remote host closed the connection]
yoslin has quit [Quit: WeeChat 3.1]
blue__penquin has quit []
Creak[m] has joined #dri-devel
danvet has joined #dri-devel
yoslin has joined #dri-devel
yoslin has quit []
yoslin has joined #dri-devel
Creak[m] has left #dri-devel [#dri-devel]
fasxmut[m] has joined #dri-devel
MatrixTravelerbot[m] has joined #dri-devel
ricardoamaro has joined #dri-devel
ricardoamaro has quit [Remote host closed the connection]
karolherbst has joined #dri-devel
honigkuchen has joined #dri-devel
honigkuchen has quit [Remote host closed the connection]
moobar has joined #dri-devel
moobar has quit [Remote host closed the connection]
<mareko>
I'm thinking about how difficult it would be to replace DXVK with a DX11 Gallium frontend. If we had that, we wouldn't need shader pre-caching at all.
<mareko>
we wouldn't even need a shader cache
wolpho[m|gr] has joined #dri-devel
wolpho[m|gr] has quit [Remote host closed the connection]
<alyssa>
didn't vmware post patches for that?
<alyssa>
but also I question the long term maintainability of adding more gallium frontends
luzipher_ has joined #dri-devel
<mareko>
there would have to be a business case of course, maintainability will come with that
<jekstrand>
It's probably quite tractable. It's just a matter of someone actually doing the work to write it, get it up to perf, and get the games debugged.
<jekstrand>
But given how good DXVK is and how much work it took to get it there, I'm sort of in the camp of "why bother?"
<mareko>
DXVK is unusable with a cold shader cache, and if a customer picked up DXVK, the complaint that it stutters would eventually end up on my desk, and we would have to do something
luzipher__ has quit [Ping timeout: 480 seconds]
<mareko>
I'm also thinking about how difficult it would be to develop a solution that precompiles shaders for DXVK as an alternative to the gallium frontend. If the size of the shader cache were ~1GB per game, it might not be palatable to some...
<jekstrand>
Steam already does that as part of the game install
<mareko>
does it also do for each driver update?
<jekstrand>
I don't know. It may detect a driver change at launch and re-compile. I don't remember.
<jekstrand>
Plagman knows those details
Daanct12 has joined #dri-devel
<kisak>
just like gallium-nine, it's going to be hard to win a debate to get gallium-eleven to be used over DXVK based on calling cold cache stuttering unusable. I think you'd get more traction if the approach ended up getting 32 bit virtual address exhaustion under control resulting in otherwise unreliable games from OOM stabilizing.
<kisak>
mareko: Steam shader pre-caching should detect driver changes
<kisak>
if not, that's a bug
Danct12 has quit [Ping timeout: 480 seconds]
Corvus` has joined #dri-devel
Corvus` has quit [autokilled: Suspected spammer. Mail support@oftc.net with questions (2021-06-12 15:06:27)]
<mareko>
I think that's good enough for now
rasterman has joined #dri-devel
<HdkR>
steam shader pre-caching was definitely working as of yesterday. It noticed a driver change and spammed my threadripper for an hour :)
<mareko>
an hour?!
<HdkR>
I have something like 500 games installed ¯\_(ツ)_/¯
<mareko>
makes sense
<HdkR>
Shader cache is only about 15GB as well
<swick>
it also makes gaming without steam basically unusable
<swick>
relying on a proprietary cloud solution to be able to play games is not a good situation at all
pq has quit [Ping timeout: 480 seconds]
<orbea>
most of my GOG games work and those that don't aren't neccesarily better on steam...
pq has joined #dri-devel
Estintax_ has joined #dri-devel
Estintax_ has quit [Remote host closed the connection]
iive has joined #dri-devel
<vsyrjala>
sometimes i wonder if the precaching is doing anything sensible for me because i use a custom mesa build for a bunch of games, and i tend to get the shader precaching thing quite a lot even when i haven't rebuilt mesa
Lucretia has quit []
Lucretia has joined #dri-devel
iive has quit []
<alyssa>
mareko: I'm out-of-the-loop -- why is shader caching so hard for radv but not necessary with gallium-eleven+radeonsi?
<alyssa>
Aren't you still doing the same number of compilers in each case, just at slightly different times?
<alyssa>
Or is it just that radeonsi does prologs/epilogs but radv does not?
<FLHerne>
DX11 Gallium would be a step toward replacing Windows blob drivers with Mesa :p
<FLHerne>
Did that guy who was making radeonsi/radv compile for Windows ever explain what the point was?
gpoo has joined #dri-devel
aaro has joined #dri-devel
aaro is now known as Guest1850
Guest1850 has quit [autokilled: Possible spambot. Mail support@oftc.net if you think this is in error. (2021-06-12 16:43:00)]
<hch12907>
FLHerne: I dont think he had ever explained why
macromorgan has quit [Quit: Leaving]
waterkip27 has joined #dri-devel
waterkip27 has quit [autokilled: spambot. Dont mail support@oftc.net with questions (2021-06-12 16:58:52)]
mlankhorst has quit [Ping timeout: 480 seconds]
drawat has quit [Remote host closed the connection]
camus has quit []
andrey-konovalov has quit [Ping timeout: 480 seconds]
jwpapi- has joined #dri-devel
jwpapi- has quit [Remote host closed the connection]
sdutt has joined #dri-devel
drawat has joined #dri-devel
pekkari has quit []
Daaanct12 has joined #dri-devel
Daanct12 has quit [Ping timeout: 480 seconds]
LeLutin12 has joined #dri-devel
LeLutin12 has quit [autokilled: Suspected spammer. Mail support@oftc.net with questions (2021-06-12 18:18:04)]
Sxem has joined #dri-devel
Sxem has quit [Remote host closed the connection]
camus has joined #dri-devel
camus1 has joined #dri-devel
camus has quit [Ping timeout: 480 seconds]
camus1 has quit [Remote host closed the connection]
camus has joined #dri-devel
flto has quit [Remote host closed the connection]
flto has joined #dri-devel
jrklein has joined #dri-devel
jrklein has quit [Remote host closed the connection]
camus has quit [Remote host closed the connection]
camus has joined #dri-devel
Patater7 has joined #dri-devel
Patater7 has quit [Remote host closed the connection]
andrey-konovalov has joined #dri-devel
sai[m|gr] has joined #dri-devel
sai[m|gr] has quit [Remote host closed the connection]
<mareko>
alyssa: basically radeonsi doesn't compile almost anything on demand
zadig has joined #dri-devel
zadig has quit [Remote host closed the connection]
imirkin has quit [Quit: Leaving]
<airlied>
mareko: just make a vulkan ext that lets dxvk work like radeonsi
<airlied>
zink could also use it
<kusma>
VK_EXT_no_performance_problems
<airlied>
compile faster exts
<HdkR>
Just use the Intel blackhole extension to optimize. It's the user's problem if they don't like the visuals :P
NinjaTrappeur has joined #dri-devel
NinjaTrappeur has quit [Remote host closed the connection]
<kusma>
Pixels or perf, pick one
<airlied>
making radv build shaders like radeonsi is lot more tractable than dx11 st
<kisak>
VK_EXT_lag_free
<alyssa>
kisak: why is that an extension everyone knows vulkan is automatically lag-free
<alyssa>
:-p
gouchi has quit [Remote host closed the connection]
<kisak>
because we get features in as extensions before promoting to the core spec
<bnieuwenhuizen>
so next vulkan version will be lag free?
<airlied>
you just front load the lag to loading screen
<bnieuwenhuizen>
how long will we wait in the loading screen?
<alyssa>
yes
<airlied>
f1 2017 a week or so?
camus has quit [Remote host closed the connection]
tobiasjakobi has quit [Remote host closed the connection]
<robclark>
alyssa: vk is lag free when used properly.. gl on top of vk is not holding it correctly :-P
<HdkR>
How soon until Vulkan on top of GL for legacy support? :)
<bnieuwenhuizen>
HdkR: isn't that lavapipe already?
danvet has quit [Ping timeout: 480 seconds]
<bnieuwenhuizen>
er nvm, that is vulkan on top of a GL driver, not on top of GL ...
luzipher__ has joined #dri-devel
<robclark>
HdkR: I think when all new games that are pushing limits of hw, and only things that need gl are already fast enough given improvements in hw, we'll see a switch to vk-only hw support
<robclark>
so maybe a generation or two?
<robclark>
maybe there is some room for vk extensions to speed that up.. idk
<robclark>
*all new games are vk
<HdkR>
I was thinking of those poor Apple users that have hardware that is new enough for GL 3.3 or 4.1, yet not new enough to gain their Metal driver :D
<robclark>
I don't think we see gl going away for existing hw.. for apple users, meh, not my problem if they bought the wrong hw ;-)
<kusma>
bnieuwenhuizen:: if we get Lavapipe running on top of virgl, maybe? :-P
luzipher_ has quit [Ping timeout: 480 seconds]
<mareko>
GL on VK probably makes sense for non-gaming
<HdkR>
Oh right. A whole suite of applications that will never move to Vulkan
<mareko>
like X
<HdkR>
Is line stippling fixed with Vulkan yet for that use case?
<mareko>
hm, GS could generate stippled lines
<HdkR>
Yea. I don't know if anyone has done that yet, or just went with an extension to hide the details somewhere :D
<airlied>
i did start glamor on vk
<airlied>
using zink spirv code
<airlied>
should see if i can make it work
<mareko>
actually no, GS can't generate stippled lines for strips and loops
<bnieuwenhuizen>
compute shaders can do everything :)
<mareko>
they can't either
<mareko>
not quickly though
<bnieuwenhuizen>
is there anybody that needs quick line stippling?
<mareko>
me :)
<bnieuwenhuizen>
for what?
<mareko>
I wanna beat the hw!
<bnieuwenhuizen>
so not that familiar but is this something we can do in the FS?
<mareko>
nope
tobiasjakobi has joined #dri-devel
<bnieuwenhuizen>
like we can get both endpoints, I presume we can build some discard formula?
chrisedsall[m] has joined #dri-devel
chrisedsall[m] has quit [Remote host closed the connection]
<mareko>
see, the line stipple pattern isn't reset between primitives, it just continues
<alyssa>
robclark: "for apple users, meh, not my problem if they bought the wrong hw ;-)"
<alyssa>
I really need new hobbies, huh? :_p
<bnieuwenhuizen>
uhh not resetting between primitives is very weird
<mareko>
only GDS Ordered Append on AMD can emulate line stipling quickly
<mareko>
it can also be emulated with memory at 1% performance
<bnieuwenhuizen>
is GDS reliable on amdgpu these days?
<mareko>
no
<bnieuwenhuizen>
:(
<mareko>
GDS as a storage is, GDS OA isn't
<bnieuwenhuizen>
any idea what is missing/wrong?
<mareko>
I think power management corrupts the OA state, so ds_ordered_count never returns
<bnieuwenhuizen>
:(
<robclark>
alyssa: I suppose there is an exemption to that rule for people who buy the hw planning to write their own driver ;-)
<bnieuwenhuizen>
power management corrupting timestamps is currently a bigger priority I guess :(
<mareko>
if you want to use GDS for super fast atomics, you can
<bnieuwenhuizen>
sounds nice for prefix sort
<bnieuwenhuizen>
er, radix sort*
<mareko>
actually I think GDS worked OK with streamout on the gfx queue
<mareko>
I mean GDS OA
<mareko>
it's just the compute queues where it hangs
<bnieuwenhuizen>
with async compute is there a way to partition GDS between gfx and compute?
<bnieuwenhuizen>
or assuming amdgpu does that is there a way to aget the address/offset
<mareko>
ašw
<mareko>
yes
<mareko>
you allocate GDS resources like BOs and it's fully virtualized, but you can have only one per queue, and your GDS allocation starts at offset 0, the OA counter starts as 0, etc.
<bnieuwenhuizen>
a hok
Duke`` has quit [Ping timeout: 480 seconds]
sdutt has quit [Remote host closed the connection]
luzipher_ has joined #dri-devel
luzipher__ has quit [Ping timeout: 480 seconds]
Luck has joined #dri-devel
Luck has quit [Read error: Connection reset by peer]
luzipher__ has joined #dri-devel
luzipher_ has quit [Ping timeout: 480 seconds]
<Venemo>
mareko: the problem with a gallium frontend for D3D11 is that is only works on HW that has a Gallium driver
<Venemo>
I think there was one a few years ago but it was removed because the code was abandoned :(
<dcbaker>
There was also some legal concerns IIRC. The
<dcbaker>
Author may have had access to the original headers or something like that
<zmike>
now we can just have jenatali write one to uh... verify the d3d12 driver...
<jenatali>
The VMware thing is using the D3D DDI though, so it still requires a D3D runtime, as opposed to DXVK which implements the API directly
ngcortes has joined #dri-devel
<airlied>
mareko: just come to the darkside, work on radv and fix it all up :-P
andrey-konovalov has quit [Ping timeout: 480 seconds]
ngcortes has quit [Remote host closed the connection]
pcercuei has quit [Quit: dodo]
<mareko>
I guess ACO supporting radeonsi would be a milestone towards that