Check your device

dsp-kitchen runs the same code on several runtimes. Being compiled in doesn’t mean a runtime works on your machine: it also needs a device and a driver.

Command line

dspk doctor

dspk 0.0.8 (linux / x86_64)
Features:   wgpu, cpu, cuda, hub
Runtimes (compiled in, in preference order):
  wgpu  usable wgpu<wgsl> (plane 32, ≤ 1024 units per cube)
  cuda  usable cuda (plane 32, ≤ 1024 units per cube, 36 multiprocessors)
  cpu   usable cpu (plane 1, ≤ 12 units per cube, 12 cores)
Selected:   wgpu (DSPK_RUNTIME or the first compiled in)

A runtime that can’t work here says why instead of usable.

Python

import dspk

dspk.runtime.available()   # ['wgpu', 'cuda', 'cpu']: compiled in, GPUs first
dspk.runtime.current()     # what calls use without runtime=

Choose one

In order of precedence:

1. A runtime= argument on the call, or --runtime on the command line.
2. dspk.runtime.set("cuda") in Python.
3. The DSPK_RUNTIME environment variable.
4. The first one compiled in.

Note

The first run on a device is slow, because kernels are compiled and tuned. Later runs start in about a second. See the kernel cache.

Next: a first filter.