00:15skeggsb: imirkin: i'm actually not 100% sure why it should matter which you use there actually
00:15skeggsb: i think they should be equal at that point
00:16skeggsb: obviously they're not, but i don't know why
00:28imirkin: skeggsb: hm ok. well the person was also testing on an older kernel
00:28imirkin: pre dispnv50 split
00:28imirkin: although afaik you didn't actually restructure anything, just moved things around
00:29imirkin: skeggsb: the bug filer is pretty responsive, so if you want him to try some stuff, i suspect he'd oblige
00:31skeggsb: i think it's probably ok to just use mode
00:31skeggsb: though mode_fixup() in drm_atomic_helper.c should probably have copied mode->adjusted_mode in the cases we'd need this already, from my reading
00:33endrift: turns out now that I'm using uniforms for palettes my palette-pathological case wasn't actually palette-pathological. I fixed the thing that was making it pathological without realizing it and now it's fast :x
00:34imirkin: make it slow again!
00:34endrift: well this actually tells me that frame-splitting isn't as slow anymore
00:34imirkin: i was using the GPU as a space heater. it is no longer heating up. bug!
00:34endrift: since I'm still frame-splitting on palette changes
00:34endrift: imirkin: I've legitimately done that before
00:34endrift: when I was in college
00:34endrift: I would use BOINC as my space heater
00:35endrift: probably only because I didn't have to pay for electricity though *shrug*
00:35imirkin: leaky dorm windows?
00:36endrift: nah, I just didn't turn on my radiator until it actually got to be snowy out
00:36endrift: it was hard to control how hot it got my room
00:36imirkin: that's why you put the heater full-blast and then open your windows :)
00:36endrift: anyway this means that the reason frame-splitting was slow was due to texture uploads which I've now eliminated all of
00:37endrift: during scanline splits at least
00:37endrift: VRAM texture uploads are still a thing
00:37endrift: and they are still annoying
00:38endrift: HdkR: you should hack your switch and use my emulator on it to play those GBA games you haven't been dying to play :P
00:38imirkin: INTEL_map_texture to the rescue
00:38endrift: I'm morbidly curious to see how badly this'll break on your average Mali GPU
00:38endrift: in Android
00:38endrift: i.e. not using Panfrost
00:39endrift: I'm looking forward to the future of Panfrost, but right now I hear it's not generally usable
00:39imirkin: HdkR: time for an updated "graphics drivers on android" follow-up article?
00:40endrift: also I want a Pinebook Pro already gosh dang
00:41HdkR: imirkin: I think most of the results would end up the same. Nvidia Blob and Mesa being highest rated
00:41imirkin: HdkR: they haven't cleaned up their act? =/
00:41imirkin: hopefully mesa does better than it did before.
00:41imirkin: hey - freedreno has UBO's now
00:41imirkin: and dual-source blend on a3xx =]
00:42imirkin: (i'm still disproportionately happy that i managed to get that working... and equally unhappy i couldn't work it out on a4xx)
00:42HdkR: Dolphin implements framebuffer_fetch now for those chips that can use it anyway
00:44HdkR: endrift: For Midgard it is almost usable for regular desktop use :P
14:52imirkin_: skeggsb: should this go the other way and null out ilut unconditionally for fp16? https://github.com/skeggsb/nouveau/commit/28dd125e669bcf61c5414b95fc403b476167d6af
14:52imirkin_: i.e. even if ilut is set explicitly, it should be blanked out?
14:53imirkin_: and/or a return -EINVAL in that case?
14:53imirkin_: (erm, there's no return value there. but probably should be?)
23:28skeggsb: imirkin_: that can't happen at this point anyway
23:28imirkin_: no, but in the future, when the luts are configurable
23:29imirkin_: (which is something i'd like to work on)
23:29imirkin_: i can fix it then, i suppose. just a bit of a landmine :)
23:29skeggsb: you'd just reject that configuration instead, earlier in atomic_check
23:30imirkin_: fair enough
23:31imirkin_: presumably that was the bit required to make fp16 work on turing? or it's still broken?
23:32skeggsb: no, everything i tested works now, i tested/fixed the issues on earlier and later gpus than you have available, so, when you're happy with the rest of it, should be good to go i expect
23:32imirkin_: i'm happy...
23:33skeggsb: alright, well, i'll merge your patch as-is in that case :)
23:39imirkin_: thanks =]
23:39imirkin_: btw, ksm/gv100 -> kms/gv100...
23:43imirkin_: skeggsb: fyi, my hope is to try to bang out most of the HDR stuff over the next month or two -- basically keeping up with intel / amdgpu.
23:44imirkin_: my next thing is going to be supporting higher-bpc modes, esp on hdmi
23:45imirkin_: let me know if you plan on working on related items, since i don't want to collide with your work