#
✕ nick:marcan
#asahi · 2021-10-04 10:42
marcanit can be worked around but not perfectly
#asahi · 2021-10-04 10:42
marcanyes, it will
#asahi · 2021-10-04 10:41
marcanhttps://mrcn.st/p/tAj5V85R did I miss anything?
#asahi · 2021-10-04 10:39
marcanIIRC I switched when testing the hv PAC mask stuff
#asahi · 2021-10-04 10:38
marcanlooks like 16K right now :)
#asahi · 2021-10-04 10:38
marcanI don't even remember what I have now, I think it used to be 4K and I changed it?
#asahi · 2021-10-04 10:31
marcanI'm putting together a list of all the in-flux kernel drivers, I'll post it here so you all can double check if you don't mind
#asahi · 2021-10-04 10:31
marcangoing to finish the progress report today (I get myself into too many loops...)
#asahi · 2021-10-04 10:30
marcanyeah, just saw that!
#asahi · 2021-10-04 10:30
marcanj_ey: if it's TBT then USB A ports would be behind a host controller; if it's just USB3 then it'd be a hub
#asahi · 2021-10-04 09:57
marcanTB3 dock with DP out should be ~equivalent
#asahi · 2021-10-04 09:57
marcannah but actually, I'm thinking a dock first
#asahi · 2021-10-04 09:56
marcanif only I had a good place to use it! :D
#asahi · 2021-10-04 09:56
marcanI'll be picking up some test hardware for this :)
#asahi · 2021-10-04 04:09
marcanI don't have a proper rootfs set up yet, but I wonder about things like compiling
#asahi · 2021-10-04 04:07
marcan690 by doing repeated `find /` calls (syscall-heavy)
#asahi · 2021-10-04 04:04
marcan876 with a better gzip benchmark (binary data, not urandom)
#asahi · 2021-10-04 04:03
marcan(modulo frequency)
#asahi · 2021-10-04 04:03
marcanfunny enough zcat is almost ~identical perf on both cores
#asahi · 2021-10-04 04:01
marcan662 with a simple dd copy
#asahi · 2021-10-04 04:00
marcanI get 754 using gzip as a benchmark
#asahi · 2021-10-04 04:00
marcankdrag0n[m]: re performance, yeah, I think you're right that dhrystone seems to be an outlier
#asahi · 2021-10-04 03:52
marcankdrag0n[m]: works for me, is it a timing thing? it has some timeout stuff that might fire due to the latency introduced depending on how you're doing it
#asahi · 2021-10-04 02:44
marcanso you're saying you're signing up to write the atc driver? :D
#asahi · 2021-10-04 02:43
marcandisplayport != thunderbolt
#asahi · 2021-10-04 02:43
marcanof course those are distinct
#asahi · 2021-10-04 02:43
marcan08:22:11 < alyssa> displayport over usb4/thunderbolt over usb type-c
#asahi · 2021-10-04 02:43
marcan08:22:02 < alyssa> displayport over usb type-c
#asahi · 2021-10-03 05:25
marcane.g. if you're running a spinloop both CPU cores have exactly the same performance (one iteration per cycle)
#asahi · 2021-10-03 05:25
marcanthe whole dmips/mhz thing is massively application-dependent anyway
#asahi · 2021-10-03 05:24
marcanthough it's a <10% difference anyway so it shouldn't be a massive problem
#asahi · 2021-10-03 05:24
marcanyeah
#asahi · 2021-10-03 05:24
marcanI'm not sure there's a way to describe the "1 core 3.2, 2 cores 3.1, 3+ cores 3.0" relationship that the chips have
#asahi · 2021-10-03 05:24
marcanif the scheduler does something useful with that, then yes
#asahi · 2021-10-03 05:24
marcanyeah, though I'm not sure if that makes a difference?
#asahi · 2021-10-03 05:22
marcanI can try again on bare metal
#asahi · 2021-10-03 05:22
marcanI was running on the hypervisor, so that might skew things a bit (timer IRQs galore), but it shouldn't be enough to skew the numbers majorly I'd hope
#asahi · 2021-10-03 05:20
marcankdrag0n[m]: https://youtu.be/Yqc-sDg5YZE?t=11619 this is where I do the benchmarks
#asahi · 2021-10-03 05:15
marcanand also I might just move the DVMR thing in there because it's one fewer thing for linux to worry about, and 12.0 does it anyway so even *Apple* thinks that belongs in the bootloader
#asahi · 2021-10-03 05:15
marcannot full throttle, something reasonable that should speed up boot
#asahi · 2021-10-03 05:15
marcanin that case I will probably have m1n1 put the p-cores into 2GHz state or so
#asahi · 2021-10-03 05:15
marcanalyssa: okay, so everything is working as intended then
#asahi · 2021-10-03 05:13
marcanthis might be because CoreMark does a worse job exercising the pcores' wide instruction dispatch than Dhrystone?
#asahi · 2021-10-03 05:12
marcanI did consider the frequency to be 2988 when doing the math
#asahi · 2021-10-03 05:12
marcan713 for the e-cores looks like a way too small difference
#asahi · 2021-10-03 05:11
marcanI calculated those on my last stream
#asahi · 2021-10-03 05:11
marcankdrag0n[m]: the capacity-dmips-mhz numbers are based on dhrystone benchmarks
#asahi · 2021-10-03 05:11
marcan(feel free to comment out those pstates in the devicetree if you don't want it to "lie")
#asahi · 2021-10-03 05:11
marcanbut not in Linux, so for the time being, even though you can request 3.2GHz, the hardware will cap you at 3GHz
#asahi · 2021-10-03 05:10
marcanm1n1 currently does that properly in its internal SMP code, which is why you can benchmark 3.2GHz there

plus ›