07:39 glennk: nv30_transfer_rect_blit does not appear to have any sort of fencing?
12:21 glennk: also is it me or is nv04_gr_idle used as if it was wait for idle in more than one place in the kernel?
19:10 imirkin: glennk: not sure if nv30/40 has simultaneous dispatch of this stuff. might be good enough to be sequential in cmd stream. if not, see what nv50/nvc0 do for this ... there's a bit on the resource which tracks whether it's being drawn to or not, and that influences adding sync's in places.
19:10 imirkin: although the sync might be nv50+ in the first place.
19:10 imirkin: yeah, it's not a subchan-level thing iirc.
19:11 imirkin: no clue about nv04_gr_idle. haven't looked at kernel stuff in that level of detail (esp not for those chips)
19:14 glennk: it appears to at least loop polling checking for idle until timeout, just done in a convoluted way several macros and function calls down
19:15 imirkin: also, i've been out the past week, so if there were questions asked in that timeframe, i missed them. (and if it's from before then, i've long forgotten them.)
23:59 karolherbst: imirkin: any thoughts on https://gitlab.freedesktop.org/mesa/drm/-/merge_requests/188 and https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12189 ?
23:59 karolherbst: imirkin: also.. I wrote https://gitlab.freedesktop.org/drm/nouveau/-/merge_requests/4 :)