#asahi-gpu · 2022-12-27 16:07
linaalyssa: I should just expose these properly as a sysfs file... it'd be nice to at least have an agxtop showing power consumption and idle times and such...#asahi-gpu · 2022-12-27 16:03
linaSo really, what I'm calling the stats channel seems to be quite specifically power stats. It's all pstates, power on/off events, power consumption, temperatures, and the like.#asahi-gpu · 2022-12-27 16:02
linaAnd indeed glxgears goes to 17-23%, which is what I'd expect for something vsynced that actually lets the GPU power down between frames#asahi-gpu · 2022-12-27 16:01
linaSo for anything >500 FPS or so, that's 100%#asahi-gpu · 2022-12-27 16:01
linaalyssa: Ah, I bet I know what this is. GPU active %, meaning powered on.#asahi-gpu · 2022-12-27 15:56
linaalyssa: What I have identified as "utilization" in the metrics does show 100%, but that can't be right with the CPU stalls... so now I wonder what that really measures...#asahi-gpu · 2022-12-27 05:04
linahowever, 1 (which is the base on laptops) is 396MHz, so if the laptops are stuck there that's quite a bit worse...#asahi-gpu · 2022-12-27 05:03
linaalyssa: 3 is 720MHz and 6 is 1287 MHz, so I would expect 78% faster if it scales with clock#asahi-gpu · 2022-12-26 03:43
linaI think the pstate machinery isn't running at all...#asahi-gpu · 2022-12-26 03:42
linaI'll have to look into that...#asahi-gpu · 2022-12-26 03:42
linaBut why? The python experiments did boost all the way up...#asahi-gpu · 2022-12-26 03:41
linaApparently we're stuck at pstate 3... that explains the worse than expected performance...#asahi-gpu · 2022-12-26 03:41
linaAnd speaking of that...#asahi-gpu · 2022-12-26 03:41
linacorion: It's not really implemented "properly" yet, but you can try `echo 0x40000 > /sys/module/asahi/parameters/debug_flags` and look at dmesg. It will be spammy though...#asahi-gpu · 2022-12-24 04:29
linaalyssa: Do it!!! ^^#asahi-gpu · 2022-12-21 15:16
linaYeah... maybe that endpoint automatically sets that conversion factor?#asahi-gpu · 2022-12-21 15:15
linaMaybe you need to use the matrix stuff to correct it then?#asahi-gpu · 2022-12-21 15:10
linaIt looks like that pixel format itself is inherently wide gamut, so what you need is another pixel format#asahi-gpu · 2022-12-21 15:10
linakCVPixelFormatType_30RGBLEPackedWideGamut = 'w30r', /* little-endian RGB101010, 2 MSB are zero, wide-gamut (384-895) */#asahi-gpu · 2022-12-21 15:09
linaYou were wondering why 10-bit mode broke on the macbooks, right?#asahi-gpu · 2022-12-21 15:09
linajannau: I just found something interesting#asahi-gpu · 2022-12-21 09:15
linamkurz: Sorry about that, it's fixed now! ^^#asahi-gpu · 2022-12-19 12:15
linaalyssa: Nice! ^^#asahi-gpu · 2022-12-16 04:23
linaAssuming a nightly build happened since the change was merged last night#asahi-gpu · 2022-12-16 04:22
linabluetail: yes, should be#asahi-gpu · 2022-12-16 04:22
linaI don't think we can blame you for Firefox binding a NULL VBO because they forgot to save/restore it properly...#asahi-gpu · 2022-12-16 04:22
lina... the previous breakage was a Firefox bug#asahi-gpu · 2022-12-16 04:21
linaI mean I tend to taskswitch, I'm not going to block on this (and also apparently Firefox is still broken in other ways...)#asahi-gpu · 2022-12-16 04:21
linaOr you mean if you did?#asahi-gpu · 2022-12-16 04:21
linaWait do you know already?#asahi-gpu · 2022-12-16 04:20
linaalyssa: ^^;;#asahi-gpu · 2022-12-16 04:19
linaPlus we got you to 100% dEQP2 thanks to app debugging <3#asahi-gpu · 2022-12-16 04:18
linaHey I enjoy this! Especially when the app developer is being really nice and helpful ^^#asahi-gpu · 2022-12-16 04:17
linaThere are very few diffs in state between my metal tests yesterday, although a preshader did pop up and a uniform block...#asahi-gpu · 2022-12-16 04:17
linaOK, then I don't know what flat shading fields I might be missing to make this work...#asahi-gpu · 2022-12-16 04:16
linaMaybe it is? But then I don't know what fields I'm missing...#asahi-gpu · 2022-12-16 04:15
linaShould we have something like a catchall 32-bit field conditional on the decoder build or some other way to actually get full dumps when tracing macos? I think I'm missing bits...#asahi-gpu · 2022-12-16 04:14
linaI think the unpacker complains in stderr but not in the agxdump file?#asahi-gpu · 2022-12-16 04:14
linaalyssa: Right now the decode dylib does not actually print unknown fields, right?#asahi-gpu · 2022-12-16 04:09
linaI'm just experimenting right now, I'll fix the XML once it makes sense#asahi-gpu · 2022-12-16 04:09
linaYes of course ^^#asahi-gpu · 2022-12-16 03:59
linaGuess I need more macOS investigation ^^#asahi-gpu · 2022-12-16 03:57
linaStill doesn't work but I think we're getting somewhere#asahi-gpu · 2022-12-16 03:57
linaOooh so with your patch, I can set the varying <<8 field and *not* crash. That's new I think!#asahi-gpu · 2022-12-16 03:49
linaBut now I wonder if combining them with some of the other things I tried last time would...#asahi-gpu · 2022-12-16 03:49
linaAs for the flat shading ones... interestingly, they don't actually fix Darwinia!#asahi-gpu · 2022-12-16 03:49
linaalyssa: Apparently I cherry-picked the depth/stencil export one but forgot to push...#asahi-gpu · 2022-12-15 08:05
linaYay! ^^#asahi-gpu · 2022-12-08 02:25
linajannau: For gnome terminal you need to unset the use system color scheme thing, else the background breaks. I don't know what's up with that...