05:26pmoreau: It would be nice to have a Nouveau talk at FOSDEM, present the NIR backend, OpenCL results, other ongoing work, status update.
05:27pmoreau: Especially since no such update was done at XDC, AFAIK (unless Karol did a lightning talk?).
05:27pmoreau: karolherbst: ^ I guess you were planning on doing one?
07:38karolherbst: pmoreau: pretty much, yes
07:41pmoreau: Cool! :-)
07:44pmoreau: Do you think you would have the OpenCL work (and SVM?) as part of the Nouveau talk, or would that be separate as it is more general and is now being used by other drivers?
07:44karolherbst: mhh, probably
07:45karolherbst: I am reworking the structurizer right now
07:45karolherbst: and my new code is way simplier
07:45pmoreau: (For example I could see the OpenCL part being presented along with the changes to libclc from Dave, as a standalone.)
07:45karolherbst: well.. showing OpenCL demos is always kind of hard
07:45karolherbst: we would need some proper CL application running
07:45karolherbst: and.. that might still be tons of work
07:46pmoreau: If we can get images, maybe we can run the Cycles and show that? But that’s not going to be easy.
07:46karolherbst: if crypto currencies wouldn't be mainly scam, we could also show a miner running on nouveau :D
07:48pmoreau: Maybe we can find some fluid simulation visualisation software that does not use OpenCL images.
07:50karolherbst: or we get images done by then?
07:50karolherbst: I don't know how much has to be done actually
07:51pmoreau: There we go! It doesn’t use images, but the visualisation is done in DirectX. :D https://software.intel.com/en-us/articles/3d-fluid-simulation-sample
07:51karolherbst: although I will probably busy with the structurizer + SVM + random fallout
07:52karolherbst: pmoreau: maybe we should focus on video decoding via OpenCL?
07:52karolherbst: that's always a annoying point
07:53karolherbst: It kind of feels like that no video decoder cares for CL support anymore
07:53pmoreau: They don’t use that much of the API: https://hastebin.com/ociliwiqax.nginx
07:53karolherbst: but I think accelerated video decoding/encoding is what most users will care about
07:54karolherbst: (even though we have dedicated piece of hardware for that, supporting CL is more general purpose)
07:54pmoreau: Though, couldn’t that be done with OpenGL/Vulkan compute shaders instead?
07:54karolherbst: I doubt you want that
07:54pmoreau: Oh wait, right, there is dedicated hardware for that, silly me
07:54karolherbst: is there a CL decoder/encodier for vp9? :D
07:54karolherbst: afaik nv hardware doesn't do vp9
07:55karolherbst: or AV1
07:55pmoreau: https://www.phoronix.com/scan.php?page=news_item&px=OpenCL-VP9-Encoder though, not open-source
07:56karolherbst: and decoder?
07:58karolherbst: we have hw vp9 decoder
07:58karolherbst: gp100 doesn't have vp9 blocks :D
07:59pmoreau: You probably do not care about video decoding/encoding on those cards anyway.
15:02imirkin: danvet: sorry, this weekend was a loss
15:19danvet: imirkin, nw
20:29cosurgi: imirkin: any news on the -EAGAIN front? I had another crash this time on the "work" xserver. Had some casaulties.
21:11imirkin_: cosurgi: not much. however there's a very small chance that an unrelated fix will help your situation too
21:11imirkin_: i'm not particularly hopeful coz it doesn't explain the SIGBUS issue
21:12imirkin_: cosurgi: this is the patch: https://patchwork.freedesktop.org/patch/335616/
21:12imirkin_: it definitely does fix issues for someone else, but i think they're having different issues.
21:12imirkin_: (which end in a SIGSEGV)
21:12imirkin_: i'm going to be doing a release with this patch in the semi-near future
21:13imirkin_: (and a couple others)
21:13imirkin_: the target GPU is mildly older than yours too ... Riva TNT2 :)
21:20cosurgi: Whoa! :)
21:20cosurgi: But thanks. I will try this patch
21:22imirkin_: can't drive 3x 4k on that
21:22imirkin_: but i do have another patch to fix 1920x1200 on there
21:25HdkR: imirkin_: When can I drive 3x 8k though? D:
21:26imirkin_: HdkR: with a TNT2? it'll be tough, they only have a single output
21:26imirkin_: (actually, i bet there were some with s-video, but i don't think they could be driven at the same time - single crtc)