<airlied>
it's pretty rough and the api has probably changed lots since
<airlied>
I was hoping to get lavapipe on openh264 going first to have some testing
<airlied>
but if there's one thing I've learning working on video, is that I'm not really a video person :-P
<jenatali>
Yeah, we've recently hooked up software video encode/decode for our va-api stuff
<jenatali>
And same. Very same. I'm very glad we have someone who is
Simonx22 has quit [Server closed connection]
Simonx22 has joined #dri-devel
<Lynne>
if all you've been exposed to has been h264+5+av1 NAL units, no wonder no one likes video
<Lynne>
and annex-b, which every hardware decoding API seems to think is the only possible form of H264 containerization
<Lynne>
and by H264, I mean everything
<Lynne>
samsung's vp9 encoder IIRC did some bastardization of annex-b even though it was completely unnecessary and unstandardized
enunes has joined #dri-devel
<Lynne>
anyhow, I went over all the previous spec updates I had done, and nothing really stuck out as something being able to break P/B frames
<Lynne>
updated my repo again with small cleanups, hopefully today we can figure out the issue and get separate output + dpb mode running
<Lynne>
the spec is unclear with what exactly constitutes a long reference, and short and long references use different indexing systems, and the spec mixes up IDs with slots, so it could be an issue with the implementation too,
columbarius has joined #dri-devel
co1umbarius has quit [Ping timeout: 480 seconds]
ogabbay has quit [Server closed connection]
ogabbay has joined #dri-devel
CosmicPenguin has quit [Server closed connection]
CosmicPenguin has joined #dri-devel
Daanct12 has joined #dri-devel
Daanct12 has quit [Remote host closed the connection]
Akari has quit [Quit: segmentation fault (core dumped)]
<airlied>
Lynne: so one thing srcBufferRange doesn't appear aligned properly
mmenzyns has quit [Server closed connection]
<airlied>
"srcBufferRange is the size of the srcBuffer with valid encoded bitstream, starting from srcBufferOffset. It must meet the alignment requirement minBitstreamBufferSizeAlignment within VkVideoCapabilitiesKHR queried with the vkGetPhysicalDeviceVideoCapabilitiesKHR function."
mmenzyns has joined #dri-devel
<Lynne>
something else I found was that I didn't set the image usage properly, video_decode_dst wasn't even part of it
<Lynne>
I fixed both issues in my branch, nvidia is even worse now, it rejects allocating any image with a video decoding usage
isinyaaa has quit [Server closed connection]
smaeul has quit [Server closed connection]
smaeul has joined #dri-devel
ayaka_ has joined #dri-devel
isinyaaa has joined #dri-devel
caseif has quit [Server closed connection]
caseif has joined #dri-devel
<Lynne>
ugh, it wanted a profile attached to all images' pnext
robink has quit [Server closed connection]
robink has joined #dri-devel
<Lynne>
with that plumbed in, nvidia actually outputs junk rather than green for the first frame
cmarcelo has quit [Server closed connection]
cmarcelo has joined #dri-devel
<airlied>
Lynne: ah yes just found the image usage bug here
<airlied>
Lynne: it might need DECODE|DPB
<airlied>
or it might have some alginment reqs
<airlied>
Lynne: image usage at least appears to fix jellyfish here
camus1 has joined #dri-devel
* airlied
wonders am I supposed to care about a profile there
<Lynne>
pushed an updated
<Lynne>
*pushed an update, fixed the usage (I hope)
yuq825 has joined #dri-devel
<Lynne>
yeah, exactly, it's a bit of a plumbing nightmare
camus has quit [Ping timeout: 480 seconds]
<Lynne>
I'm guessing it's so that in case each decoder has its own separate pool of images
<Lynne>
P/B frames are not exactly fixed still, it's just that the image breaks down slower
<airlied>
ah indeed it does, now the question of whether driver or player is doing it wrong
* airlied
gets very lost in what the hw expects in what order here
<Lynne>
I think I've got out-of-place decoding done, I'll debug it later, wasn't that difficult, for some reason I had the idea it was going to involve copying from an output frame to a dpb pool, not the other way around
<Lynne>
I've got bigger issues though, for some reason all my clocks broke suddenly, they're all saying it's nearly 5:30 am
<Lynne>
I'll go and fix them, may take some hours
<airlied>
lols, yeah there should be no copying at all on the client side
aravind has joined #dri-devel
<airlied>
the hw is the one doing the copies, you should just be doing allocations
<airlied>
Lynne: I've pushed two hacky patches that kinda aling with a vaapi decode, but it doesn't seem to help
<airlied>
and I really don't want to include them unless we can prove they are necessary
yuq825 has quit []
rg3igalia has quit [Server closed connection]
rg3igalia has joined #dri-devel
mhenning has quit [Quit: mhenning]
bmodem has joined #dri-devel
<airlied>
and I really don't want to include them unless we can prove they are necessary
<airlied>
oops
swivel has quit [Server closed connection]
swivel has joined #dri-devel
eric_engestrom has quit [Server closed connection]
eric_engestrom has joined #dri-devel
hwentlan_ has quit [Server closed connection]
hwentlan_ has joined #dri-devel
Company has quit [Quit: Leaving]
jimjams has quit [Server closed connection]
jimjams has joined #dri-devel
ddavenport_ has quit [Server closed connection]
ddavenport_ has joined #dri-devel
benettig has quit [Server closed connection]
benettig has joined #dri-devel
zx2c4 has quit [Server closed connection]
zx2c4 has joined #dri-devel
gildekel has quit [Server closed connection]
gildekel has joined #dri-devel
dolphin has quit [Server closed connection]
dolphin has joined #dri-devel
bmodem1 has joined #dri-devel
bmodem has quit [Ping timeout: 480 seconds]
bmodem has joined #dri-devel
rodrigovivi has quit [Server closed connection]
rodrigovivi has joined #dri-devel
mattst88 has quit [Server closed connection]
mattst88 has joined #dri-devel
bmodem has quit [Remote host closed the connection]
bmodem1 has quit [Read error: Connection reset by peer]
tarceri has joined #dri-devel
Duke`` has joined #dri-devel
yuq825 has joined #dri-devel
<tomeu>
karolherbst: ok, I see now that it is indeed an aarch64 thing, my local builds succeeded because I was using the x86 bindgen when cross-compiling
aravind has quit [Ping timeout: 480 seconds]
itoral has joined #dri-devel
Duke`` has quit [Ping timeout: 480 seconds]
gio has quit [Server closed connection]
gio has joined #dri-devel
ccaione has quit [Server closed connection]
ccaione has joined #dri-devel
fab has joined #dri-devel
bgs has quit [Remote host closed the connection]
tzimmermann has joined #dri-devel
lemonzest has joined #dri-devel
Sachiel has quit [Server closed connection]
Sachiel has joined #dri-devel
sarnex has quit [Server closed connection]
sarnex has joined #dri-devel
kts has joined #dri-devel
aravind has joined #dri-devel
flto has quit [Server closed connection]
flto has joined #dri-devel
zaratustra has quit [Server closed connection]
zaratustra has joined #dri-devel
aravind has quit [Ping timeout: 480 seconds]
aravind has joined #dri-devel
robher has quit [Server closed connection]
robher has joined #dri-devel
ahajda has joined #dri-devel
alatiera has quit [Server closed connection]
alatiera has joined #dri-devel
pH5 has quit [Server closed connection]
pH5 has joined #dri-devel
danvet has joined #dri-devel
mvlad has joined #dri-devel
pa- has quit [Server closed connection]
angular_mike______ has quit [Server closed connection]
angular_mike______ has joined #dri-devel
<tomeu>
karolherbst: I'm going to look at skip them, because this makes me suspect it may be quite of a rathole: https://reviews.llvm.org/D116774
pa has joined #dri-devel
jhugo___ has quit [Server closed connection]
jhugo___ has joined #dri-devel
fab has quit [Quit: fab]
_alice has quit [Server closed connection]
_alice has joined #dri-devel
alanc has quit [Remote host closed the connection]
<ayaka_>
Anything that I need to do to make it merge
mdnavare has quit [Server closed connection]
mdnavare has joined #dri-devel
dianders has quit [Server closed connection]
gouchi has joined #dri-devel
<mripard>
zackr: drm-misc-fixes is pretty much always open, and we're likely to have a -rc8
dianders has joined #dri-devel
<emersion>
ayaka_: address the comments and resend?
gouchi has quit [Remote host closed the connection]
<mripard>
zackr: if it's not urgent, you can use drm-misc-next-fixes too
<ayaka_>
emersion, you see the last comment is made by me, I think the only thing I didn't do in the document where is in mesa
<ayaka_>
but the problem is nobody review it in mesa
<mripard>
zackr: my general rule (but I don't know if everyone agrees on that) is that if it's a fix for something affecting someone (so you have a report, or you found it yourself), it goes into drm-misc-fixes, if it's a "theoretical" bug (something that is indeed a bug, but no-one has been/is affected by it), it can go into drm-misc-next/drm-misc-next-fixes
camus has joined #dri-devel
camus1 has quit [Ping timeout: 480 seconds]
pzanoni has quit [Server closed connection]
pzanoni has joined #dri-devel
tanty has quit [Remote host closed the connection]
frieder has joined #dri-devel
mcoffin has joined #dri-devel
tanty has joined #dri-devel
<mcoffin>
After 1.3.235 of `vulkan-icd-loader`, placing 32-bit versions of layers into `VK_INSTANCE_LAYERS` segfaults for me (whereas it was previously a silent soft-fail). Does anyone know if 1. this was intentional and 2. a workaround?
<mcoffin>
I know the loader isn't mesa-specific, but thought there might be some folks in here that would know, so lmk if I'm off-base on that
<mcoffin>
Downgrade to 1.3.226 works as "expected". Layers utilizing this pattern of "_{32,64}" suffixing include VK_LAYER_VALVE_steam_overlay_{32,64} and the same for steam_fossilize, so it would be a pattern in real-world use that I wouldn't *think* the loader would break, but I could be wrong
<mcoffin>
Any confirmation that this happens for you as well would also help me rule out some part of my setup that I haven't been considering. Thanks again, and cheers
kem has quit [Server closed connection]
jkrzyszt has joined #dri-devel
kem has joined #dri-devel
MajorBiscuit has joined #dri-devel
fab has joined #dri-devel
fxkamd has quit []
<mcoffin>
And if nobody knows, then where would loader issues be tracked so I can try there? Sorry again, I've just never had issues with this component so don't really know where to go xD
tursulin has joined #dri-devel
anarsoul has quit [Server closed connection]
anarsoul has joined #dri-devel
Akari has joined #dri-devel
heat_ has joined #dri-devel
camus has quit [Ping timeout: 480 seconds]
dcz_ has joined #dri-devel
rasterman has joined #dri-devel
mcoffin has quit [Quit: leaving]
arnd has quit [Server closed connection]
arnd has joined #dri-devel
FLHerne has quit [Server closed connection]
FLHerne has joined #dri-devel
aswar002 has quit [Server closed connection]
aswar002 has joined #dri-devel
zf has quit [Server closed connection]
zf has joined #dri-devel
everfree has quit [Server closed connection]
everfree has joined #dri-devel
anholt has quit [Server closed connection]
anholt has joined #dri-devel
Akari has quit [Quit: segmentation fault (core dumped)]
genpaku has quit [Server closed connection]
genpaku has joined #dri-devel
lynxeye has joined #dri-devel
vliaskov has joined #dri-devel
simon-perretta-img has quit [Server closed connection]
simon-perretta-img has joined #dri-devel
aravind has joined #dri-devel
eukara has quit [Server closed connection]
ceyusa has quit [Server closed connection]
ceyusa has joined #dri-devel
Daanct12 has joined #dri-devel
Daanct12 has quit [Remote host closed the connection]
rib has quit [Server closed connection]
rib has joined #dri-devel
Daanct12 has joined #dri-devel
pcercuei has joined #dri-devel
apinheiro has joined #dri-devel
kts has quit [Quit: Leaving]
mceier has quit [Server closed connection]
mceier has joined #dri-devel
Daaanct12 has joined #dri-devel
kts has joined #dri-devel
yoslin_ has quit []
yoslin has joined #dri-devel
pcercuei has quit [Quit: brb]
amber-luci has joined #dri-devel
jkrzyszt has quit [Remote host closed the connection]
Duke`` has joined #dri-devel
GeorgesStavracasfeaneron[m] has quit [Server closed connection]
GeorgesStavracasfeaneron[m] has joined #dri-devel
Daaanct12 has quit [Remote host closed the connection]
yuq825 has quit []
jkrzyszt has joined #dri-devel
aravind has quit [Ping timeout: 480 seconds]
devilhorns has joined #dri-devel
aravind has joined #dri-devel
aravind has quit [Remote host closed the connection]
junaid has joined #dri-devel
aravind has joined #dri-devel
bbrezill1 has quit []
bbrezillon has joined #dri-devel
mkoncek has joined #dri-devel
jkrzyszt has quit [Remote host closed the connection]
<Lynne>
airlied: no copying by the client? how are clients supposed to get the output then?
<Lynne>
is the output just another standard dpb frame? what slotindex should it have?
illwieckz has quit [Remote host closed the connection]
lynxeye has quit [Quit: Leaving.]
illwieckz has joined #dri-devel
agd5f has quit [Server closed connection]
agd5f has joined #dri-devel
yuq825 has joined #dri-devel
pcercuei has joined #dri-devel
RSpliet has quit [Server closed connection]
RSpliet has joined #dri-devel
junaid has quit [Ping timeout: 480 seconds]
xroumegue has quit [Ping timeout: 480 seconds]
jkrzyszt has joined #dri-devel
xroumegue has joined #dri-devel
<jenatali>
mcoffin: there's a GitHub project for issues
Leopold_ has joined #dri-devel
<karolherbst>
tomeu: yeah.. I wouldn't mind to hide the APIs using varargs for now anyway
<tomeu>
cool, that worked
Sumera[m] has quit [Server closed connection]
Sumera[m] has joined #dri-devel
Leopold_ has quit []
Leopold has joined #dri-devel
Leopold has quit []
Leopold has joined #dri-devel
aravind has quit [Ping timeout: 480 seconds]
Leopold has quit [Remote host closed the connection]
illwieckz has quit [Remote host closed the connection]
Leopold has joined #dri-devel
illwieckz has joined #dri-devel
Daanct12 has quit [Quit: Quitting]
Lucretia has quit [Remote host closed the connection]
MajorBiscuit has quit [Ping timeout: 480 seconds]
Lucretia has joined #dri-devel
haasn has quit [Server closed connection]
haasn has joined #dri-devel
jfalempe__ has joined #dri-devel
jfalempe_ has quit [Read error: Connection reset by peer]
lynxeye has joined #dri-devel
mkoncek has quit [Ping timeout: 480 seconds]
devilhorns has quit [Ping timeout: 480 seconds]
kts has quit [Quit: Leaving]
frieder has quit [Quit: Leaving]
frieder has joined #dri-devel
frieder has quit [Remote host closed the connection]
devilhorns has joined #dri-devel
<DavidHeidelberg[m]>
robclark: Hello! Could I patch the Mesa Kernel with your GPU hang patches from mailing list or still too soon?
<robclark>
DavidHeidelberg[m]: go for it.. the patches are in msm-next
<DavidHeidelberg[m]>
robclark: nice! marked for todo :)
itoral has quit [Remote host closed the connection]
<kusma>
Is it just me, or does the radeonsi-stoney-gl:amd64 tests run everything... twice?
<kusma>
I see a job that finished, then "LAVA Job finished with status: hung", and then... reboot and running again...
MajorBiscuit has joined #dri-devel
<kusma>
https://gitlab.freedesktop.org/mesa/mesa/-/jobs/32649703#L2811 -- the "command deqp-runner suite --suite //install/deqp-radeonsi-stoney.toml --output //results --skips //install/all-skips.txt //install/radeonsi-stoney-skips.txt -" is twice in that log...
aravind has joined #dri-devel
Akari has joined #dri-devel
MajorBiscuit has quit [Read error: Connection reset by peer]
MajorBiscuit has joined #dri-devel
<jenatali>
kusma: If it hangs it auto-restarts, yeah
<kusma>
Doesn't really look like a real hang, though...
<kusma>
I also feel like I see this all the time
<DavidHeidelberg[m]>
I can confirm I also seen it few times, not 100% sure it was stoney, but I recall radeonsi and runtime > 20 minutes
<jenatali>
Yes I've seen at least 2 or 3 restarts in a single run before
<kusma>
From some quick checking, it doesn't seem to happen as often as I thought I'd been seeing it
<dcz_>
okay, this sounds stupid, but I'm out of ideas. Can the execution of a shader depend on where the source buffer (texture) comes from?
<dcz_>
all works well when I allocate a GBM buffer and fill it up myself, but when I try a buffer from a camera, the destination buffer isn't even changed
<jenatali>
What does that mean? "The execution of a shader"?
yuq825 has left #dri-devel [#dri-devel]
<dcz_>
seems the fragment shader isn't even executed in one case
<dcz_>
seems it's not the fault of the buffer, as nothing happens even if I don't pass it to the shader
zehortigoza has quit [Remote host closed the connection]
mriesch has quit [Server closed connection]
mriesch has joined #dri-devel
DemiMarie has quit [Server closed connection]
DemiMarie has joined #dri-devel
jekstrand[m] has quit [Server closed connection]
jekstrand[m] has joined #dri-devel
DemiMarie is now known as Guest376
apinheiro has quit [Ping timeout: 480 seconds]
Andy[m]1 has quit [Server closed connection]
Andy[m]1 has joined #dri-devel
gnustomp[m] has quit [Server closed connection]
gnustomp[m] has joined #dri-devel
fab has quit [Quit: fab]
Jeremy_Rand_Talos has joined #dri-devel
junaid has joined #dri-devel
jkrzyszt has quit [Remote host closed the connection]
devilhorns has quit [Ping timeout: 480 seconds]
Haaninjo has joined #dri-devel
aravind has quit [Ping timeout: 480 seconds]
jkrzyszt has joined #dri-devel
fxkamd has joined #dri-devel
fab has joined #dri-devel
Company has joined #dri-devel
fab has quit [Quit: fab]
Leopold has quit [Remote host closed the connection]
Leopold_ has joined #dri-devel
fab has joined #dri-devel
zehortigoza has joined #dri-devel
bgs has joined #dri-devel
<daniels>
kusma: yeah, the stoneys like to fall off the face of the earth, and we've not yet bottomed out why that is
<daniels>
so when they do that, you see a 5min gap of silence, followed by us announcing that the job hung and we're restarting, then we run again
fab has quit [Quit: fab]
rmckeever has joined #dri-devel
fab has joined #dri-devel
fxkamd has quit []
tango_ is now known as Guest383
tango_ has joined #dri-devel
Guest383 has quit [Ping timeout: 480 seconds]
junaid has quit [Ping timeout: 480 seconds]
Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #dri-devel
kts has joined #dri-devel
fab has quit [Quit: fab]
fab has joined #dri-devel
jmadill has joined #dri-devel
Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #dri-devel
junaid has joined #dri-devel
jmadill has quit [Remote host closed the connection]
pcercuei has quit [Read error: Connection reset by peer]
pcercuei has joined #dri-devel
mhenning has joined #dri-devel
ybogdano has joined #dri-devel
gouchi has joined #dri-devel
rasterman has quit [Quit: Gettin' stinky!]
dordoka has joined #dri-devel
alyssa has joined #dri-devel
<alyssa>
kusma: I notice that Zink can't do two-sided polygon modes (i.e. fill_front != fill_back)
<alyssa>
other than a few piglits, does that affect any actual apps?
<alyssa>
AGX bakes that limitation into hardware, Apple's GL driver has a rather complicated way to emulate it
<alyssa>
conceptually doubling the draw with cull_front/cull_back set respectively
<alyssa>
but with a pile of other weird stuff happening (seemingly unrolling each primitive to happen in its own draw?)
<alyssa>
I haven't thought through the API side details of doing that emulation, I don't particularly want to, and if Zink gets away without it I really don't care we'll be shipping Vulkan soon enough anyway
<alyssa>
but if this is going to cause issues for CAD apps or whatever... I'd like to know about that ahead-of-time
haagch has quit [Server closed connection]
haagch has joined #dri-devel
<kusma>
I haven't seen any noteworthy problems as a result. Some piglits fail, but that's about it...
<kusma>
We did implement a work-around for the d3d12 driver for ms for the same issue... Just drawing twice, one with each face. I think that passes the piglits as well. Still not perfect, because of draw order, but good enough imo
Daanct12 has joined #dri-devel
<alyssa>
what's the draw order issue OOI?
<kusma>
The primitives should be interleaved in draw order...
<alyssa>
oh? oh. oh...
Danct12 has quit [Ping timeout: 480 seconds]
<alyssa>
yes, I see why Apple does what it does now
<kusma>
But that requires drawing a primitive at the time, and nobody is going to love the perf
<ajax>
task shader that emits a series of 1-prim meshes...
<airlied>
Lynne: the output is just an image for the client not used for dpb purposes at all
<ajax>
also this is one of the canonical things that differentiates quadros from geforces, is whether front-and-back is allowed to work in hardware
<alyssa>
kusma: I do wonder if there's any value in splitting the d3d12 emulation into a common gallium helper and using in zink at least?
<alyssa>
better than nothing, I would think
<kusma>
The emulation is fairly trivial, but sure, could make sense as common code
srslypascal is now known as Guest387
srslypascal has joined #dri-devel
Guest387 has quit [Ping timeout: 480 seconds]
tursulin has quit [Ping timeout: 480 seconds]
<airlied>
Lynne: where dpb and output are separate the hw encoder does some internal copies
<airlied>
the dpb image allocation shouldnt be touched by the client
Daaanct12 has joined #dri-devel
<airlied>
the output image is normal
<Lynne>
so, with COINCIDE, the output image is in VkVideoDecodeInfoKHR.pSetupReferenceSlot
<Lynne>
the references are images you've previously decoded
<Lynne>
with DISTINCT, the references are blank images, but what's the output?
<Lynne>
*blank opaque images you don't touch
<Lynne>
does the driver automatically copy the output to one of the opaque references?
dordoka has quit [Quit: Konversation terminated!]
kts has quit [Quit: Leaving]
Daanct12 has quit [Ping timeout: 480 seconds]
<airlied>
Lynne: not the driver, the hw decoder does it
<airlied>
or rather the fw
Daaanct12 has quit [Ping timeout: 480 seconds]
<airlied>
Lynne: tbe dstpictureresource
* airlied
isnt near spec
<airlied>
but you have two ptrs in a decode, one to dst output and one to dpb
rellla has quit [Quit: ZNC 1.6.5+deb1+deb9u2 - http://znc.in]
Danct12 has joined #dri-devel
<Lynne>
the spec doesn't say it
<Lynne>
the spec literally says "output and dpb are distinct", and lets you figure it out on your own
<Lynne>
oh, so the pSetupReferenceSlot is the reference, and dstPictureResource is the destination
<alyssa>
kusma: Maybe I'll open an issue for "pull out a common helper, convert d3d12, add zink support, add asahi support" as a good-first-task
<alyssa>
seems like a nice "student getting oriented in mesa" task
<airlied>
Lynne: yup
rellla has joined #dri-devel
<airlied>
and you dont do anything with the contents of dpb imgs
<airlied>
so from client side its just different img allocation behaviour
MajorBiscuit has quit [Ping timeout: 480 seconds]
<eric_engestrom>
has there been any recent mesa changes that would make `llvmpipe-piglit-clover` fail because `artifacts/install.tar` (presumably `dependency`d from a build job) doesn't exist? https://gitlab.freedesktop.org/mesa/mesa/-/jobs/32667152
<eric_engestrom>
I tried grepping around in the git history but I can't find anything relevant
ybogdano has quit [Read error: Connection reset by peer]
heat__ has quit [Read error: Connection reset by peer]
linusw has quit [Quit: Connection closed for inactivity]
jkrzyszt has quit [Remote host closed the connection]
jkrzyszt has joined #dri-devel
mhenning has quit [Quit: mhenning]
<eric_engestrom>
gallo[m]: ah thanks, I didn't think to check if the var was defined for that job
reductum has quit [Quit: WeeChat 2.8]
reductum has joined #dri-devel
fab has quit [Quit: fab]
<DavidHeidelberg[m]>
btw talking about d3d12, where is the native Linux d3d12 😉? d3d9 still killing dxvk in terms of efficiency, so I guess there would be value
<DavidHeidelberg[m]>
* s/d3d9/gallium-nine/
gouchi has quit [Remote host closed the connection]
bgs has quit [Remote host closed the connection]
<jenatali>
David Heidelberg: We've talked about it
<DavidHeidelberg[m]>
I must missed it :'(
<jenatali>
Sorry no not an announcement, I just mean we've talked about building it
<DavidHeidelberg[m]>
oh
junaid has joined #dri-devel
anholt has quit [Remote host closed the connection]
apinheiro has joined #dri-devel
<glehmann>
David Heidelberg: what makes you think that nine is killing dxvk?
<Lynne>
airlied: updated branch, last commit adds support for DISTINCT
<Lynne>
I think it's correct, but it's still not fixed the references bug
<Lynne>
if you could take a look at it would be nice
<Lynne>
btw your patchset still doesn't signal a mode, it should
ahajda has quit [Quit: Going offline, see ya! (www.adiirc.com)]
junaid has quit [Remote host closed the connection]
junaid has joined #dri-devel
dcz_ has quit [Ping timeout: 480 seconds]
<DavidHeidelberg[m]>
glehmann: in terms of effectivity for sure. If you have mid-high end cpu, lot of memory and modern gpu, you can run anything older without looking at "cost", but when you running on low-power configuration, low memory or older/weaker GPU it's visible, I think zmike did some benchmark some time ago
<airlied>
"VK_VIDEO_CAPABILITY_SEPARATE_REFERENCE_IMAGES_BIT_KHR - the DPB or Reconstructed Video Picture Resources for the video session may be created as a separate VkImage for each DPB picture. If not supported, the DPB must be created as single multi-layered image where each layer represents one of the DPB Video Picture Resources."
<jekstrand>
cmarcelo: I'd still like to but it's a big cross-tree rework.
<Lynne>
airlied: that's, uh, a bit insane
<cmarcelo>
jekstrand: got it
<airlied>
Lynne: them's the rules :-P
<Lynne>
it's very insane, for starters, all layers have to be allocated upfront
<Lynne>
at 8k or so, 17 references, it's gigabytes of vram
<airlied>
Lynne: newer hw should hopefully do away with it, don't you know the video size that early?
<airlied>
seem crazy to always be doing 8kx8k
<airlied>
I already noticed that in the current code
<Lynne>
you know the video size, but still, you don't know the number of references
<airlied>
Lynne: it's how the older hw gens mandate things, I think for some of the newer ones it can be moved to separate
<Lynne>
8k frames at 10bits are 200mb a pop
<airlied>
Lynne: the current code has been doing it as a workaround using one of the session memory reqs
<airlied>
Lynne: and the radeonsi vaapi driver has always done it
<Lynne>
can't the driver massage the data and not require users to use layered textures?
<airlied>
Lynne: nope
<airlied>
so you are just getting to see the horrors that are already happening
<airlied>
so it really makes no difference
<jekstrand>
cmarcelo: I'd still like to one day
<airlied>
Lynne: just try not to pass maxCoded directly back in please like you do now
<airlied>
for CreateVideoSession
<Lynne>
why, it shouldn't matter, the client allocates all data
<Lynne>
plus, uh, streams can dynamically change size
<airlied>
Lynne: you don't know the max stream size before hand?
<airlied>
I thought for h264 it was pretty fixed
<Lynne>
nope, fixed or not, streams can do whatever they want, especially in broadcasting
<airlied>
Lynne: okay actually maybe it doesn't matter if you are doing the allocating now
<airlied>
I can remove the code that cares once I get an array texture
Akari has joined #dri-devel
<Lynne>
that works
<Lynne>
the plumbing and/or code duplication... ugh
<Lynne>
I've never worked with layered textures, what's special about them?
<Lynne>
is it a single call to alloc and multiple calls to bind or?
<airlied>
Lynne: just a single call to alloc and bind
<airlied>
you just stick number of layers into the create image call
<airlied>
I think for the DPB you might also have to put the layer number into the image view, but I'm not 100% sure on that
<airlied>
I don't think at least radv will care
<Lynne>
oh, it's less of a nightmare
<Lynne>
maybe I can plumb that
<Lynne>
is VK_REMAINING_ARRAY_LAYERS shorthand for "all layers"?
<airlied>
yes if baseLayer is 0
mbrost has joined #dri-devel
<Lynne>
VK_VIDEO_CAPABILITY_SEPARATE_REFERENCE_IMAGES_BIT_KHR can't show up if the COINCIDE bit flag is set, right?
<Lynne>
err, isn't set, rather
<airlied>
Lynne: yeah I think they should be mutually exclusive
<Lynne>
spec doesn't say it, but yeah, it doesn't really need to mention it I guess (though it probably should)