06:30 fidtar: is it possible to reliably test and debug nouveau in a vm?
06:52 mupuf: fidtar: what would it bring you?
06:52 mupuf: you can pass the gpu to the vm (passthrough) if you don't need it for your host :)
06:52 mupuf: I would however recommend a second machine and SSH
06:53 mupuf: or, if this is really the code of nouveau you want to check, use the userspace version of nouveau if possible
06:53 fidtar: yeah that's what I mostly run, was just curious of thoughts
06:53 fidtar: ssh to 2nd machine, that is
07:14 mupuf: fidtar: :)
14:36 jbase: hello
14:38 jbase: sorry )
14:42 jbase: Does any1 know if there's a way to obtain the OhGodADump-NVIDIA tool?
16:57 karolherbst: jbase: I suggest you to not use shady tools in the first place
16:58 karolherbst: we have our own tools to do the same stuff, without persons involved with shady backgrounds and software with no source released
21:30 karolherbst: imirkin: I think I came up with a good compromise for the loading immediates as const buffer optimization: we could have a short list (~100, depending on what values I might find in shader-db) of immediates which are often used we always store in the driver const buffer and then up to N immediates per shader. This should hopefully elimiate the need to rebind all that often.. just have to dig around and see if this approach makes
21:30 karolherbst: sense according to the shaders we know of
21:31 karolherbst: or we really only go for the former and only have a fixed list of values... anyway, I think I will analyze the shaders a bit