#
#asahi-gpu · 2022-11-28 16:18
eiln_My work is on
#asahi-gpu · 2022-11-28 16:18
eiln_I've made progress with the GPU's younger sister, i.e. Apple Neural Engine / ANE / NPU
#asahi-gpu · 2022-11-28 16:18
eiln_My name is Eileen and my handles use @eiln typically.
#asahi-gpu · 2022-11-28 16:17
eiln_Hello! slightly off topic but was pointed to here
#asahi-gpu · 2022-11-28 04:45
alyssaand I have weak evidence against that idea as well
#asahi-gpu · 2022-11-28 04:41
alyssawhich is... possible but unlikely because that would complicate sampling...
#asahi-gpu · 2022-11-28 04:40
alyssaunless twiddling for depth/stencil formats is weirdly different from colour
#asahi-gpu · 2022-11-28 04:17
alyssarendering to 16x16 vs 32x32 (level=1) vs 33x33 (level=1) looks the same
#asahi-gpu · 2022-11-28 04:17
alyssayeah, I can't see how this can possibly work
#asahi-gpu · 2022-11-28 04:13
alyssalike, I'm looking at traces from Metal and i can't see how this can possibly be correct
#asahi-gpu · 2022-11-28 04:12
alyssais AGX broken?!
#asahi-gpu · 2022-11-28 04:12
alyssaam I missing something silly here?
#asahi-gpu · 2022-11-28 04:10
alyssaso how does the hw know to insert padding?
#asahi-gpu · 2022-11-28 04:10
alyssaand setting the base address
#asahi-gpu · 2022-11-28 04:10
alyssafor depth/stencil, Metal is just setting the width/height to that of the rendered level
#asahi-gpu · 2022-11-28 04:10
alyssafor colour render targets, there's a dedicated "level" field in the PBE descriptor and the width/height are of level 0
#asahi-gpu · 2022-11-28 04:10
alyssabut the hw has no way to know whether it needs to add the padding silliness
#asahi-gpu · 2022-11-28 04:09
alyssathat's fine for power-of-two and for levle 0
#asahi-gpu · 2022-11-28 04:09
alyssaWe just specify the size and the base address for twiddled
#asahi-gpu · 2022-11-28 04:08
alyssalina: I'm pretty confused .. I don't see any mechanism for rendering into depth stencil miptrees
#asahi-gpu · 2022-11-28 03:43
alyssaEyyy
#asahi-gpu · 2022-11-28 03:43
alyssa Pass (Pass)
#asahi-gpu · 2022-11-28 03:08
alyssameh let's implement the obvious thing in Mesa and see if anything breaks
#asahi-gpu · 2022-11-28 03:07
alyssathough it has other silliness. delight
#asahi-gpu · 2022-11-28 03:07
alyssaif I don't have any references to per-sample stuff it doesn't do that silliness
#asahi-gpu · 2022-11-28 03:06
alyssathat's the only way I can read this that makes sense
#asahi-gpu · 2022-11-28 03:06
alyssaindicating compiler support for MSAA 16x?!
#asahi-gpu · 2022-11-28 03:06
alyssapretty weird loop getting emitted, calling zs_emit on 8 samples at a time
#asahi-gpu · 2022-11-28 03:05
alyssafrom context, looks like it must set r1 to the mask of active samples
#asahi-gpu · 2022-11-28 03:03
alyssaseen with zs_emit + multisampling
#asahi-gpu · 2022-11-28 03:03
alyssa5905 0000 103c 8000
#asahi-gpu · 2022-11-28 03:03
alyssadougall: <disassembly failed>, ooo, fun
#asahi-gpu · 2022-11-27 22:01
jannauif it was off less than 3 minutes it comes up dark and brightness is restored ~3 minutes after poweroff
#asahi-gpu · 2022-11-27 22:00
jannauit looks like the brightness is restored ~3 minutes after power-off. i.e. the fix works if the display was off for at least 3 mintes
#asahi-gpu · 2022-11-27 21:58
jannauI'll check what macos does on other devices with integrated displays when it powers the display off
#asahi-gpu · 2022-11-27 21:57
jannauyour fix wasn't enough on an imac (maybe m1 laptops as well). I have a fix for clients which do regular swaps but I'm not sure what to for x11
#asahi-gpu · 2022-11-27 21:55
jannaulina: pushed with small fixes to asahi-6.1-rc6-5_dcp-fixes (removed the oob setPowerState and dev_info)
#asahi-gpu · 2022-11-27 18:21
jannaus/muntes/minutes/
#asahi-gpu · 2022-11-27 18:20
jannaubrightness is restored after 2 muntes and 45 seconds without any AP -> DCP call
#asahi-gpu · 2022-11-27 17:53
jannaureproducible but much more than a few seconds. more like 2-3 minutes. it complains after the poweron about idac value out of range, which has to be the 0x0 fromm the poweroff
#asahi-gpu · 2022-11-27 17:48
jannauwtf, it fixed itself after a quite several seconds without setting the brightness
#asahi-gpu · 2022-11-27 17:25
jannaulina: still broken on the imac. I suspect we need to wait after the iomfb has finished it's callbacks after the setPowerState
#asahi-gpu · 2022-11-27 16:35
jannauI can't remember seeing that on devices with external displays or on the j314
#asahi-gpu · 2022-11-27 16:35
jannauI've also noticed that on the imac macos 12.4 shuts dcp downs completely (including a MMIO CPU_RUN=0) when it turns off the display
#asahi-gpu · 2022-11-27 16:19
jannaulina: if you chain the the requests like that there no need to use out of band for dcp_set_power_state(). the true as second parameter
#asahi-gpu · 2022-11-27 16:13
jannaulina: yes, I've come to the same conclusion, seems to be bug on t8103 and t8112. I've never seen seen that on t6001
#asahi-gpu · 2022-11-27 16:06
linaAlong with some cleanup patches and another thing which is probably unrelated but sounded like a good idea?
#asahi-gpu · 2022-11-27 16:06
linaPushed to my usual branch ^^
#asahi-gpu · 2022-11-27 16:05
linaThat took a while to work out... I didn't understand why it was ignoring my requests...
#asahi-gpu · 2022-11-27 16:05
linaSo we not only need to force a set on poweron, but also turn it off on poweroff, otherwise it just ignores the set as a no-op

plus ›