Open bmaranville opened 4 years ago
If a program is written in Python and has a well-written (compliant) kernel, then numba
package is a very good way of compiling Python code with LLVM and then also parallelising.
The reflectivity calculation can be vectorised quite easily, so I'm not sure how much improvement you'd see when using numba over numpy (if one takes care to avoid temp arrays). I've messed around a little bit with numba and could never get it to outperform numpy.
Sure, you're right! When I was working on off-specular calculation I had to use the proposed solution.
Comments from the workshop: Do you parallelize this over the model, or over the NR?