00:02 airlied: nope definitely not that, just doing prep work for any possible vulkan driver
00:03 airlied: in theory an ogk gallium driver shouldn't need any codegen changes
00:06 Wally: airlied: Other than moving it to src/nouveau right?
00:07 Wally: (it would simplify things a lot)
00:07 Wally: Question: Does anyone thing gv100 codegen would work with Ampere or is a new target needed
00:08 Wally: *think
00:08 airlied: not really, since in theory an ogk gallium driver should work as a new winsys layer
00:08 airlied: or libdrm_nouveau replacement
00:08 airlied: the rest of the gallium code should mostly stay the ame
00:08 airlied: though there might be some funky changes
00:10 Wally: I dont seem to understand...how would a new winsys driver support ogk?
00:10 Wally: * add support
00:13 airlied: you'd start writing an abstraction on top of the ogk API that mostly matched the nouveau ones
00:14 Wally: mostly but there would still be a lot of libdrm_nouveau calls in the gallium driver
00:14 Wally: wouldnt there?
00:15 Wally: perhaps im missing something
00:20 airlied: yeah so you'd design a common abstraction layer, and hook up two backends
00:20 airlied: we generally call that abstraction layer the winsys
00:20 airlied: though you could just fork to a new driver
00:39 airlied: getting TGSI_SEMANTICS out of the interace is going to be trickiest
00:45 Wally: I see
00:45 Wally: welp ill try to wrap up pushbuf support on harambae then
07:40 airlied:wonders how much we care about nouveau mesa supporting Linux prior to 4.4 from 2016
07:41 HdkR: Isn't it reasonable that old software should be delegated to other old software?
07:44 airlied: it usually is :-)
07:51 airlied: bleh appears nv30 still uses those paths
10:33 mupuf: same take as HdkR. 4.4 is ridiculously old!
10:33 mupuf:'s take on dependencies is: if it is in debian stable, it's outdated-enough
12:51 karolherbst: airlied: yeah.. nv30...
12:52 karolherbst: I don't mind if we have a proper winsys abstraction layer for those of things, but I would design it different from what libdrm currently is
12:53 karolherbst: airlied: in regards to dropping the TGSI bits, we could convert the TGSI names to mesa ones inside gallium
12:53 karolherbst: and use the mesa ones instead
12:54 karolherbst: (inside codegen I mean)
12:54 karolherbst: that _could_ be painless