MAINT: interpolate: move an internal utility from cython to python #21539
Conversation
The computational complexity is O(1), so there is no reason for this helper to stay in Cython or move to C++.
ev-br
added
scipy.interpolate
maintenance
| double[::1] wrk = np.empty(2*k+2, dtype=np.float64) | ||
|
|
||
| # derivatives @ xval | ||
| with nogil: |
There was a problem hiding this comment.
just for my understanding, could you explain the relationship between this with nogil code and the Python translation?
There was a problem hiding this comment.
The code is precisely the same, modulo that in python it operates on numpy arrays and in cython it was taking typed memoryviews of arrays and operating on those.
And while it was in a with nogil section, the is doing so little work that releasing the GIL is very unlikely to have any effect.
| # compute and fill in the derivatives @ xval | ||
| for row in range(deriv_ords.shape[0]): | ||
| nu = deriv_ords[row] | ||
| wrk = _bspl.evaluate_all_bspl(t, k, xval, left, nu) |
There was a problem hiding this comment.
I did notice the use of this function instead of _deBoor_D directly, but looks like that still gets called internally, so I don't see any obvious issues on my end either, and CI is happy.
There was a problem hiding this comment.
Yeah, evaluale_all_bspl is a python-side wrapper of C-only _deBoor_D. Both routines are non-public, so can probably unify the naming. Will think about it in the ongoing refactor of the internals (I've a branch, not PR-ready yet)
|
Thanks Evgeni, Lucas |
A follow-up to #21223, chipping small bits and pieces from a cython extension: move the
_handle_lhs_derivativeshelper from Cython to python. The code is basically pure numpy, and the computational complexity is O(1), so there is no reason for this helper to stay in Cython or move to C++.