| HN Mirror

Y	Hacker News new \| ask \| show \| jobs

by almostgotcaught 641 days ago

no it won't, because while hitting ioctls in python is cute

it is definitely not shippable

2 comments

Umm, why not?

We wrote entire NVIDIA, AMD, and QCOM drivers in that style.

Because it's slow duh

This sounds like prejudice. Have you benchmarked it?

Yes I literally duplicated your approach for my driver stack last week and surprise surprise the FFI overhead into libc is too high.

FFI? This isn't how GPUs work...they are MMIO (mostly)

Those drivers are faster than anything else when used to run fixed command queues (what neural network runs are)

I can't say anything on the performance, but inline assembly in Python is crazy

It's not inline assembly it's just ioctl through ctypes via libc.