#
✕ nick:aead
#asahi · 2021-11-25 07:11
aeadoh I'll restrict my search to lwn. good point!
#asahi · 2021-11-25 07:10
aead(and smartphone SoC manufacturers are notorious for not upstreaming their device-specific tweaks, AFAIK)
#asahi · 2021-11-25 07:10
aeadit seems it'd have to pull some really clever tricks to do the same thing just based on pure heuristics
#asahi · 2021-11-25 07:09
aeadwhich, if that does exist on Linux now (not sure?) it's unlikely anything would make use of it anyway
#asahi · 2021-11-25 07:09
aeadyeah but I'm curious how smart the scheduler is really. It seems like Apple does a great job of putting the "right" threads on each core because it exposes an API for userspace to tell the kernel what the QoS of the process is
#asahi · 2021-11-25 06:59
aead(I Googled a little before asking it _seems_ like 2 might be the answer based on a paper about teaching CFS how to treat cores proportionally, but concrete info seems kinda thin)
#asahi · 2021-11-25 06:58
aeadIt seems to me the answer could be one of three things: it's good at knowing which threads belong on the big and little cores; or, it's aware of the difference in compute capacity and just schedules all threads proportionally to each core's rough performance; or it doesn't do much special
#asahi · 2021-11-25 06:57
aeadOnce everything is merged and well supported, how well is Linux expected to be able to handle the heterogenous cores in the M1 / M1 Max? Does the scheduler have good support for "big.little"?