03:11buhman: is the gtx 1070 max q known to work?
03:14mwk: huh, neat
03:14mwk: NV20 uses NaN tagging in one of its internal queues
03:15mwk: to distinguish commands from vertex data
03:15buhman: where's the prebuilt images from https://github.com/hakzsam/archlive-nouveau ?
03:17mwk: so you receive a stream of 4×fp32 vectors from upstream; if the W component is not a NaN, you interpret it as vertex attribute data to interpolate; if it's a NaN, the NaN payload and remaining components are a special command to be executed instead
03:18mwk: I suppose that explains why post-XF processing converts NaNs into 2**a_lot
18:44imirkin: skeggsb_: still something funny going on with vblanks, i think. uptime = 34 days, and glxgears freezes for the first couple of seconds, and then proceeds fine.
18:45imirkin: of course that's only 176M vblanks in that time (at 60hz), so ... dunno
18:57imirkin: WTF. X takes up 75% CPU when glxgears is running (with vblank enabled)??
19:00imirkin: fully taken up by drmmode_event_handler.
19:03imirkin: interesting. https://hastebin.com/bacukoyeti.pl
19:03imirkin: sounds like some kind of BS event keeps getting dispatched
19:03imirkin: skeggsb_: ideas welcome. i don't know anything about drm events
19:04imirkin: [this is with kernel 4.16.4]
20:58imirkin: skeggsb_: i guess xorg_list_for_each_entry is what's taking so long
20:59imirkin: which means that there's a ton of entries in the drmmode_events list