ALSA mmap IO

[ishitatsuyuki]

We should use mmap when trying to achieve the lowest latency, because copy overhead can become a roadblock when using a extremely small buffer size.

(Other platforms often have this implemented as a part of the low-level API, so we have been already using memory mapped IO.)

[sniperrifle2004]

Correct me if I'm wrong, but if you truly want to achieve low latency you also need to expose that back to the application right? Samples have to be written directly into the mapped memory after all. Is that something cpal is doing?

[ishitatsuyuki]

We expose &mut [SampleType] buffers, so it can be zero-copy as long as we don't need to convert (e.g. interleavedness).