Providing an option to use a non-default stream about cudarc HOT 4 CLOSED

This should already be possible via CudaStream. Though you have to manage the additional stream yourself instead of cudarc managing it for you.

from cudarc.

If I want to run, say, CudaDevice::memset_zeros(), I don't see any way to do this without using the default legacy stream to do it, because I don't see any options to customize the stream.

from cudarc.

Ah gotcha, you're right. Would you want an option on creation of the CudaDevice to use a non-null stream?

from cudarc.

I think that could be an interesting way to do it, yeah. My one concern would be if there are any APIs, like candle, that assume that if two devices have the same ordinal, then their operations are on the same stream. But, maybe that would be considered overreliance on implementation details. For candle, it looks like they use a stricter form of checking with a generated unique ID rather than the ordinal, so there wouldn't be any problems with that approach for that library.

from cudarc.