pub fn wait_readback_map(
device: &Device,
submission: SubmissionIndex,
_map_rx: &Receiver<Result<(), BufferAsyncError>>,
_total_bytes: usize,
)Expand description
Block until the submission that produced the readback has finished and its buffer-map callback has fired, then return.
A single submission-index Wait replaces the old 64–256 poll(Poll)
busy-spin followed by a full Wait. Each Poll maintain pass costs tens of
microseconds on Metal, so the spin alone added ~3 ms of pure CPU overhead per
run — the dominant cost for small graphs (an MNIST CNN forward dropped from
~3.4 ms to sub-millisecond). wgpu invokes the buffer-map callback as part of
the poll, so the mapped range is ready when this returns.