Try using scipp.elemwise_func

[SimonHeybrock]

Considering some options for using scipp.elemwise_func. Please see comments in code.

[SimonHeybrock]

@jl-wynen Do you have any preferences on how to proceed?

[jl-wynen]

Branch? We would need to extend the C++ implementation, no?

[SimonHeybrock]

Yes, and branch here to compile two difference numba.cfuncs

[jl-wynen]

Is there a practical need for single precision floats in this case?

[SimonHeybrock]

Could be, inelastic data can be huge.

[jl-wynen]

For now we should keep both because numba is a pretty big dependency. But we need a way to select which implementation is used in order to test both.

[SimonHeybrock]

One way would be to use tox, with two different environments (with and without numba)?

[jl-wynen]

Yes. This would be expensive but thorough because it would ensure that everyhing works without numba.

[jl-wynen]

Looks reasonable. Except, we maybe shouldn't use else with try and group all kernels in there but use something like if numba_availale: ... else: ...

[SimonHeybrock]

Looks reasonable. Except, we maybe shouldn't use else with try

Why?

[jl-wynen]

Why?

• To group the implementation of the numba kernel with the outer python function. When there are multiple of each, it gets harder to read if the numba kernels are all at the top of the file and the others scattered throughout.
• Minor: To avoid try-else (I don't know the precise semantics)

[SimonHeybrock]

• Minor: To avoid try-else (I don't know the precise semantics)

I think it is actually recommended? It avoids putting more than necessary into the try:-block, thus avoiding catching more things than intended.

[jl-wynen]

But I was suggesting

try:
  import numba
except ImportError:
  numba = None

[...]

if numba is not None:
  def kernel():
    ...
else:
  kernel = None