04:01imirkin: ok. i think i have the load/store pack/unpack logic down. need to add more emitCVT handling though... using some that aren't in our emitter =/
04:02imirkin: ah yeah. none of the TYPE_U16/U8 stuff is there when that's the dst type
04:17imirkin: mwk: would you expect there to be problems with e.g. "cvt u32 $r2 u8 $r2l"?
04:18imirkin: (i.e. due to register conflicts or whatever)
06:01imirkin: mwk: ignore that question. there were about 100 different issues. all resolved now
06:01imirkin: made good progress ... almost everything works now
06:02imirkin: something wrong with CAS
06:02imirkin: probably just messed something simple up
06:04imirkin: oh. cas is just encoding fail. gr
06:32imirkin: making progress. https://paste.debian.net/plainh/ffbdc39b
09:59uis: Hi. Any news? What about register allocation bug?
10:01uis: On nve0
10:03uis: And what about power managment?
10:14RSpliet: what about it? nothing changed with power management, is still all manual and experimental-but-working-alright-for-the-majority
10:21uis: What about register allocation bug?
10:25karolherbst: uis: please don't be annoying. Most of us are doing this in their spare time and others have more important/fundamental issues to work on. One game not working is nothing compared basic applications not working (like everything chromium based these days) because of other more fundamental bugs. I know it is frustrating, but at the moment we have more demand for random bug fixes _and_ new features than we can work on
13:50dviola: skeggsb: I saw your patch, thanks for replying and for sending the patch
13:50dviola: I'll test it today
13:53skeggsb: dviola: hopefully it'll have the same result for you, i wanted to avoid changing the behaviour of earlier GPUs as i can't recall why that's even there
13:54skeggsb: there's been a few other weird reports of issues on some GPUs running ACR unload ucode on pmu, wondering if this could be part of the cause there
13:54skeggsb: it's not an issue i've encountered, personally
13:55dviola: I will let you know
13:55skeggsb: thanks :)
13:55karolherbst: ehh gp108 again
13:56skeggsb: yeah, mine doesn't do it :P
13:56karolherbst: mine gp107 was also fine, and we have this other stupid workaround for both :/
13:56karolherbst: ohh, I actually also have a gp108
13:56karolherbst: and I think this was fine anyway
13:57karolherbst: skeggsb: ohh btw, do you have a plan on when we can have this "super" race condition fixed inside the kernel? hitting it regulary when working on heavily threaded applications :/
13:58karolherbst: and makes it hard to pinpoint causes of random errors
14:01karolherbst: because I think I have a mesa state which threading fixed for real by now and would like to get both sides fixed :)
14:01skeggsb: yeah, working on it, it's worse than just that obvious problem. i've got two things to hopefully push out somewhere by the end of the next week before PTO, and i expect maaaybe in the week i get back it might be ready to push that out
14:02skeggsb: (with ampere host stuff)
14:02karolherbst: ahh, cool!
14:52imirkin: skeggsb: fwiw i'm running on a GP108 and never had that pmu problem. but it's desktop, not mobile.
15:13imirkin: emersion: i just cc'd you on a thread where someone's having trouble with 256x256 cursors with a GK104. if you get a chance, could you test with your GK208?
15:20emersion: i'll try!
15:21imirkin: emersion: thanks =]
19:21imirkin: pmoreau: fyi, this is my branch atm. need to split that huge "WIP" commit of course. https://gitlab.freedesktop.org/imirkin/mesa/-/commits/compute/
19:21imirkin: and there's a lot more stuff to do
19:21imirkin: i expect i'll have more updates
19:21imirkin: but in case you wanted to peruse
19:22imirkin: (like need to pass in a bunch of image parameters to the shader to make arrays work, i need to make 3d work, still haven't done image size, etc)
23:25imirkin: pmoreau: some of your changes are missing S-o-b's -- can i just add them, or is there more going on?