Fix cython performance regression (#726)

* Add legacy_implicit_noexcept = True

* [skipci] changelog

CHANGELOG.rst

@@ -7,8 +7,12 @@
 Changelog
 =========
 
-2.7.0 - 2023-10-xx
-------------------
+Unreleased
+----------
+
+**Bug fix:**
+
+- Added cython compiler directive legacy_implicit_noexcept = True to fix performance regression with cython 3.
 
 **Other changes:**
 

setup.py

@@ -95,6 +95,18 @@
         glm_benchmarks_run = glum_benchmarks.cli_run:cli_run
         glm_benchmarks_analyze = glum_benchmarks.cli_analyze:cli_analyze
     """,
-    ext_modules=cythonize(ext_modules, annotate=False),
+    ext_modules=cythonize(
+        ext_modules,
+        annotate=False,
+        compiler_directives={
+            "language_level": "3",
+            "boundscheck": False,
+            "wraparound": False,
+            "initializedcheck": False,
+            "nonecheck": False,
+            "cdivision": True,
+            "legacy_implicit_noexcept": True,
+        },
+    ),
     zip_safe=False,
 )

src/glum/_cd_fast.pyx

@@ -7,7 +7,6 @@
 # License: BSD 3 clause
 #         Substantial modifications by Ben Thompson <[email protected]>
 #
-# cython: boundscheck=False, wraparound=False, cdivision=True
 from libc.math cimport fabs
 cimport numpy as np
 import numpy as np

src/glum/_functions.pyx

@@ -1,5 +1,3 @@
-# cython: boundscheck=False, wraparound=False, cdivision=True
-
 from cython cimport floating
 from cython.parallel import prange