Skip to content
Discussion options

You must be logged in to vote

There shouldn't be any.

In many cases, cpal directly forwards the hardware buffers provided by the underlying API to the user callback. For instance, on Windows, WASAPI signals a waitable object on every period, and cpal maps WASAPI's buffer as the slice that is passed to the data callback, which means you are reading from/writing to WASAPIs buffer directly. The current ASIO implementation does funnel everything through a staging buffer because it needs to (de-)interleave multi-channel streams, but this happens entirely within the same buffer cycle, so it doesn't affect total latency.

I'm personally not familiar with the other platforms, so I can't make any guarantees there, but if one of…

Replies: 1 comment

Comment options

You must be logged in to vote
0 replies
Answer selected by roderickvd
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Category
Q&A
Labels
None yet
2 participants
Converted from issue

This discussion was converted from issue #560 on September 08, 2026 20:09.