Expand description
dspk: multi-channel signal processing written once in CubeCL, run on any GPU (wgpu, CUDA, HIP) or the CPU, with the device chosen at run time.
This crate re-exports the dspk crates; features choose which are included and which runtimes
are compiled in (default: wgpu, cuda, cpu, zarr).
| Module | Crate | Contents |
|---|---|---|
core | dspk-core | recordings, buffers, time, windows, compute targets |
io | dspk-io | file formats (raw binary, Zarr), detection, chunk cache |
ops | dspk-ops | filters, resampling, linear algebra, statistics, peaks, spatial operators |
view (feature view) | dspk-view | min/max envelopes and pyramids |
stream (feature stream) | dspk-stream | QUIC + TLS sessions for continuous signals |
synapse (feature synapse) | dspk-synapse | electrophysiology: formats, probes, detection, clustering, metrics |
hub (feature hub) | dspk-hub | verified downloads of published artifacts |
The spike sorters are separate crates, each under its upstream’s licence: dspk-sorter-kilosort
(GPL-3.0), dspk-sorter-ms5 (Apache-2.0), dspk-sorter-si (MIT OR Apache-2.0).
§Start here
ops::Pipeline chains operations on the device; the guides and every operation are in the
documentation.
Early development (0.0.x): APIs change between releases.