Speed up Cube.subset/Coord.intersect

[bouweandela]

🚀 Pull Request

Description

This speeds up Cube.subset / Coord.intersect by using a dict lookup instead of list.index to find the indices of the coordinate that should be used for subsetting the cube.

Here is some benchmark code:

import iris.cube
import iris.coords
import numpy as np

lon = iris.coords.DimCoord(np.arange(0, 360, 0.5, dtype=np.float64), standard_name='longitude')
cube = iris.cube.Cube(np.arange(720, dtype=np.float32))
cube.add_dim_coord(lon, 0)
sublon = lon[(lon.points > 60) & (lon.points < 300)]

%timeit cube.subset(sublon)

The current main branch result is:

93.9 ms ± 4.51 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

and the result of running that code with this branch is:

13.5 ms ± 1.54 ms per loop (mean ± std. dev. of 7 runs, 100 loops each)

---

Consult Iris pull request check list

[lbdreyer]

This is looking great, thanks @bouweandela ! Do you mind adding a whatsnew entry to cover this change?

[bouweandela]

Thanks for the review @lbdreyer ! I added the entry.

[lbdreyer]

Perfect! Thanks @bouweandela !