Cythonize get_coinc_indices

[spxiwh]

This continues work in #5148 by Cythonizing the get_coinc_indices function. For now this is still draft, and I put it here to demonstrate how this might work (especially if anyone in the GRB team is interested in learning some Cython, it would be great to have more people with that knowledge).

Particular things that need doing

• Writing Cython functions for other common use cases
• Making the one-detector case a no-op if (as I think) it should be
• Creating a new .pyx file for GRB Cython functions and moving the new functions there (expecting this will not be the only time we Cythonize bits).
• Testing (I ran Tito's test case 2 and obtained the same number of triggers (within 1)), I've also ran the code on the same test case generating indices with both the old and new methods and set to raise an error if they differ ... It did not fail (after I fixed the bugs that test uncovered).

Standard information about the request

This is a: efficiency update

This change affects: PyGRB

This change changes: Core scientific code, but only optimization.

This change: has appropriate unit tests, follows style guidelines (See e.g. PEP8), has been proposed using the contribution guidelines

Motivation

#5148 has all the motivation. PyGRB can be slow in some cases. This function was written in "nice" python, but not "fast" python.

Contents

For the 2-detector case this is just finding duplicated values in 2 sorted arrays. There's a well known algorithm for that problem, and I implement it here.

Additional notes

Profile (for test case 2) with this PR. It's not as great as I would like, but the dominant cost of get_coinc_indexes remains outside of the Cython function (even though it's doing very little!)

• The author of this pull request confirms they will adhere to the code of conduct

[spxiwh]

... I also wonder if caching the output array might fix the 10% of time spent in get_coinc_indexes and not in the Cython function??

[titodalcanton]

Please rebase on master, we would like to get this on master soon to include it in the latest analyses.

[spxiwh]

This has been rebased now. I note that there were a bunch of additional potential things here listed at the top that would be good for the GRB group to look at (especially in terms of getting more people touching our Cython code).

However, the patch will work fine on it's own especially if the 2-detector case is the main place where slowdown is observed.

[Copilot]

Pull request overview

This pull request optimizes the get_coinc_indexes function by implementing a Cythonized version for the common two-detector, two-coincidence case. This addresses performance issues identified in issue #5148, where PyGRB workflows were experiencing long run times due to inefficient trigger coincidence finding.

Changes:

• Added a Cythonized implementation using a merge algorithm to find coincident triggers between two sorted detector arrays
• Introduced a fast path in get_coinc_indexes for the two-detector case when min_nifos == 2
• Maintained backward compatibility by falling back to the original implementation for other cases

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 8 comments.

File	Description
pycbc/events/eventmgr_cython.pyx	Adds the Cython implementation get_coinc_indexes_cython_twodet_twocoinc using a fused type to support both int and long arrays, implementing a standard merge algorithm for finding array intersections
pycbc/events/coherent.py	Adds a Python wrapper and integrates the fast path into get_coinc_indexes, allocating output arrays and calling the Cython function when applicable

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[spxiwh]

Thanks copilot. There were some useful comments there (and it's worth reading through how you analysed the code to see how you are checking everything). I've resolved the comments you raised now.