Update umath patching

* Add safety checks to patching
* Add warnings related to multi-threaded programs
* Implement _GlobalPatch wrapper class which implements patching globally
* Rename _patch module to _patch_numpy
* Rename patching functions

Author: ndgrigorian

.github/workflows/conda-package.yml

@@ -133,7 +133,7 @@ jobs:
         run: |
           source "$CONDA/etc/profile.d/conda.sh"
           conda activate test_mkl_umath
-          python -c "import mkl_umath, numpy as np; mkl_umath.use_in_numpy(); np.sin(np.linspace(0, 1, num=10**6));"
+          python -c "import mkl_umath, numpy as np; mkl_umath.patch_numpy_umath(); np.sin(np.linspace(0, 1, num=10**6));"
 
       - name: Run tests
         run: |
@@ -328,7 +328,7 @@ jobs:
         run: |
           @ECHO ON
           conda activate mkl_umath_test
-          python -c "import mkl_umath, numpy as np; mkl_umath.use_in_numpy(); np.sin(np.linspace(0, 1, num=10**6));"
+          python -c "import mkl_umath, numpy as np; mkl_umath.patch_numpy_umath(); np.sin(np.linspace(0, 1, num=10**6));"
 
       - name: Run tests
         shell: cmd /C CALL {0}

AGENTS.md

@@ -7,13 +7,13 @@ Entry point for agent context in this repo.
 
 It provides:
 - `mkl_umath._ufuncs` — OneMKL-backed NumPy ufunc loops
-- `mkl_umath._patch` — runtime patching interface (`use_in_numpy()`, `restore()`, `is_patched()`)
+- `mkl_umath._patch_numpy` — runtime patching interface (`patch_numpy_umath` `restore_numpy_umath`, `is_patched()`)
 - Performance-optimized math operations (sin, cos, exp, log, etc.) using Intel MKL VM
 
 ## Key components
 - **Python interface:** `mkl_umath/__init__.py`, `_init_helper.py`
 - **Core C implementation:** `mkl_umath/src/` (ufuncsmodule.c, mkl_umath_loops.c.src)
-- **Cython patch layer:** `mkl_umath/src/_patch.pyx`
+- **Cython patch layer:** `mkl_umath/src/_patch_numpy.pyx`
 - **Code generation:** `generate_umath.py`, `generate_umath_doc.py`
 - **Build system:** CMake (CMakeLists.txt) + scikit-build
 
@@ -49,9 +49,9 @@ CC=${CC:-icx} pip install --no-build-isolation --no-deps .  # clang is also supp
 ## Usage
 ```python
 import mkl_umath
-mkl_umath.use_in_numpy()  # Patch NumPy to use MKL loops
+mkl_umath.patch_numpy_umath()    # Patch NumPy to use MKL loops
 # ... perform NumPy operations (now accelerated) ...
-mkl_umath.restore()       # Restore original NumPy loops
+mkl_umath.restore_numpy_umath()  # Restore original NumPy loops
 ```
 
 ## How to work in this repo

CMakeLists.txt

@@ -134,13 +134,13 @@ if (UNIX)
 endif()
 install(TARGETS _ufuncs LIBRARY DESTINATION mkl_umath)
 
-add_cython_target(_patch "mkl_umath/src/_patch.pyx" C OUTPUT_VAR _generated_src)
-Python_add_library(_patch MODULE WITH_SOABI ${_generated_src})
-target_include_directories(_patch PRIVATE "mkl_umath/src/" ${Python_NumPy_INCLUDE_DIRS} ${Python_INCLUDE_DIRS})
-target_compile_definitions(_patch PUBLIC NPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION)
-target_link_libraries(_patch PRIVATE mkl_umath_loops)
-set_target_properties(_patch PROPERTIES C_STANDARD 99)
+add_cython_target(_patch_numpy "mkl_umath/src/_patch_numpy.pyx" C OUTPUT_VAR _generated_src)
+Python_add_library(_patch_numpy MODULE WITH_SOABI ${_generated_src})
+target_include_directories(_patch_numpy PRIVATE "mkl_umath/src/" ${Python_NumPy_INCLUDE_DIRS} ${Python_INCLUDE_DIRS})
+target_compile_definitions(_patch_numpy PUBLIC NPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION)
+target_link_libraries(_patch_numpy PRIVATE mkl_umath_loops)
+set_target_properties(_patch_numpy PROPERTIES C_STANDARD 99)
 if (UNIX)
-  set_target_properties(_patch PROPERTIES INSTALL_RPATH "$ORIGIN/../..;$ORIGIN/../../..;$ORIGIN")
+  set_target_properties(_patch_numpy PROPERTIES INSTALL_RPATH "$ORIGIN/../..;$ORIGIN/../../..;$ORIGIN")
 endif()
-install(TARGETS _patch LIBRARY DESTINATION mkl_umath)
+install(TARGETS _patch_numpy LIBRARY DESTINATION mkl_umath)

conda-recipe-cf/meta.yaml

@@ -43,7 +43,7 @@ test:
     imports:
       - mkl_umath
       - mkl_umath._ufuncs
-      - mkl_umath._patch
+      - mkl_umath._patch_numpy
 
 about:
     home: http://github.com/IntelPython/mkl_umath

conda-recipe/meta.yaml

@@ -47,7 +47,7 @@ test:
     imports:
       - mkl_umath
       - mkl_umath._ufuncs
-      - mkl_umath._patch
+      - mkl_umath._patch_numpy
 
 about:
     home: http://github.com/IntelPython/mkl_umath

mkl_umath/AGENTS.md

@@ -14,9 +14,9 @@ Core MKL-backed ufunc implementation: Python interface, Cython patching, and C/M
 
 ## Patching API
 ```python
-mkl_umath.use_in_numpy()  # Replace NumPy loops with MKL
-mkl_umath.restore()       # Restore original NumPy loops
-mkl_umath.is_patched()    # Check patch status
+mkl_umath.patch_numpy_umath()    # Replace NumPy loops with MKL
+mkl_umath.restore_numpy_umath()  # Restore original NumPy loops
+mkl_umath.is_patched()           # Check patch status
 ```
 
 ## Development guardrails
@@ -31,6 +31,6 @@ mkl_umath.is_patched()    # Check patch status
 - Docstrings: dual NumPy 1.x/2.x support via separate docstring modules
 
 ## Notes
-- `_patch.pyx` is Cython; changes require Cython rebuild
+- `_patch_numpy.pyx` is Cython; changes require Cython rebuild
 - MKL VM loops in `src/mkl_umath_loops.c.src`
 - `src/ufuncsmodule.c` — NumPy ufunc registration and dispatch

mkl_umath/__init__.py

@@ -29,10 +29,13 @@
 """
 
 from . import _init_helper
-from ._patch import is_patched, mkl_umath, restore, use_in_numpy
+from ._patch_numpy import (
+    is_patched,
+    mkl_umath,
+    patch_numpy_umath,
+    restore_numpy_umath,
+)
 from ._ufuncs import *
 from ._version import __version__
 
-# TODO: add __all__ with public API and remove star imports
-
 del _init_helper

mkl_umath/src/AGENTS.md

@@ -7,7 +7,7 @@ C/Cython implementation layer: MKL VM integration, ufunc loops, and NumPy patchi
 - **ufuncsmodule.h** — ufunc module public headers
 - **mkl_umath_loops.c.src** — MKL VM loop implementations (template, ~60k LOC)
 - **mkl_umath_loops.h.src** — loop function declarations (template)
-- **_patch.pyx** — Cython patching layer (runtime NumPy loop replacement)
+- **_patch_numpy.pyx** — Cython patching layer (runtime NumPy loop replacement)
 - **fast_loop_macros.h** — loop generation macros
 - **blocking_utils.h** — blocking/chunking utilities for large arrays
 
@@ -21,15 +21,16 @@ C/Cython implementation layer: MKL VM integration, ufunc loops, and NumPy patchi
 - Blocking strategy: chunk large arrays for cache efficiency
 - Error handling: MKL VM status → NumPy error state
 
-## Patching mechanism (_patch.pyx)
-- Cython extension exposing `use_in_numpy()`, `restore()`, `is_patched()`
+## Patching mechanism (_patch_numpy.pyx)
+- Cython extension exposing `patch_numpy_umath()`, `restore_numpy_umath()`,
+  `is_patched()`
 - Replaces function pointers in NumPy's ufunc loop tables
 - Thread-safe: guards against concurrent patching
 - Reversible: stores original pointers for restoration
 
 ## Build output
 - `mkl_umath_loops.c` → shared library (libmkl_umath_loops.so/.dll)
-- `_patch.pyx` → Python extension (_patch.*.so)
+- `_patch_numpy.pyx` → Python extension (_patch.*.so)
 - `ufuncsmodule.c` + `__umath_generated.c` → `_ufuncs` extension
 
 ## Development notes