Remove OpenMP

.github/workflows/build-and-deploy.yml

@@ -31,6 +31,12 @@ jobs:
     steps:
     - uses: actions/checkout@v4
 
+    - name: Set up Python
+      uses: actions/setup-python@v5
+      with:
+        python-version: '3.12'
+        cache: pip
+
     - name: Build sdist
       run: pipx run build --sdist
 

.pre-commit-config.yaml

@@ -3,7 +3,7 @@ ci:
   autoupdate_schedule: quarterly
 repos:
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.4.8
+  rev: v0.6.7
   hooks:
     - id: ruff
       args: [--fix, --exit-non-zero-on-fix]
@@ -36,6 +36,6 @@ repos:
 
 # this validates our github workflow files
 - repo: https://github.com/python-jsonschema/check-jsonschema
-  rev: 0.28.4
+  rev: 0.29.2
   hooks:
     - id: check-github-workflows

CMakeLists.txt

@@ -1,10 +1,5 @@
 cmake_minimum_required(VERSION 3.15...3.26)
 
-project(nanobind_project LANGUAGES CXX)
-if(NOT APPLE)
-  find_package(OpenMP REQUIRED)
-endif()
-
 # Try to import all Python components potentially needed by nanobind
 find_package(Python 3.9
   REQUIRED COMPONENTS Interpreter Development.Module
@@ -28,23 +23,14 @@ nanobind_add_module(
   src/clustering.cpp
 )
 
-# Link OpenMP
-if(OpenMP_CXX_FOUND)
-  target_link_libraries(_clustering PRIVATE OpenMP::OpenMP_CXX)
-endif()
 
 # Compiler-specific options
 if(MSVC)
-  # Use MSVC optimization levels and OpenMP setup
-  target_compile_options(_clustering PRIVATE /O2 /std:c++17 /openmp:llvm)
-  # /openmp:llvm
+  # Use MSVC optimization levels
+  target_compile_options(_clustering PRIVATE /O2 /std:c++17)
 else()
   # Assuming GCC or Clang
-  if (APPLE)
-    target_compile_options(_clustering PRIVATE -O3)
-  else()
-    target_compile_options(_clustering PRIVATE -O3 -fopenmp)
-  endif()
+  target_compile_options(_clustering PRIVATE -O3)
 
 endif()
 

pyproject.toml

@@ -19,7 +19,7 @@ classifiers = [
 dependencies = [
   "numpy",
   "pyvista>=0.37.0",
-  "pykdtree"
+  'pykdtree',
 ]
 description = "Uniformly remeshes surface meshes"
 keywords = ["vtk", "uniform", "meshing", "remeshing", "acvd"]
@@ -34,6 +34,7 @@ archs = ["auto64"]  # 64-bit only
 skip = "cp38-* cp313-* pp* *musllinux*"  # build Python 3.9 - Python 3.13
 test-command = "pytest {project}/tests"
 test-requires = "pytest"
+test-skip = "*-macosx_arm64"
 
 [tool.cibuildwheel.macos]
 archs = ["native"]

src/clustering.cpp

@@ -7,21 +7,9 @@
 
 #include <nanobind/nanobind.h>
 #include <nanobind/ndarray.h>
-// #include <nanobind/stl/set.h>    // needed to convert c++ sets to python sets
-// #include <nanobind/stl/vector.h> // needed to convert c++ vectors to python lists
 
 #include "array_support.h"
 
-#if !defined(__APPLE__)
-#include <omp.h>
-#endif
-
-#if defined(__linux__) || defined(__APPLE__)
-typedef int64_t vtk_int;
-#else
-typedef int32_t vtk_int;
-#endif
-
 #ifdef _MSC_VER
 #define restrict __restrict
 #elif defined(__GNUC__) || defined(__clang__)
@@ -99,7 +87,6 @@ PointNormals(NDArray<const T, 2> points_arr, NDArray<const int64_t, 2> faces_arr
     const T *v = points_arr.data();
     const int64_t *f = faces_arr.data();
 
-#pragma omp parallel for
     for (size_t i = 0; i < n_faces; i++) {
         int64_t point0 = f[i * 3 + 0];
         int64_t point1 = f[i * 3 + 1];
@@ -163,7 +150,6 @@ PointNormals(NDArray<const T, 2> points_arr, NDArray<const int64_t, 2> faces_arr
     }
 
     // Normalize point normals
-#pragma omp parallel for
     for (size_t i = 0; i < n_points; i++) {
         T plen = sqrt(
             pnorm[i * 3 + 0] * pnorm[i * 3 + 0] + pnorm[i * 3 + 1] * pnorm[i * 3 + 1] +
@@ -192,7 +178,6 @@ FaceCentroid(const NDArray<const T, 2> points, const NDArray<const int64_t, 2> f
     auto fmean_arr = MakeNDArray<T, 2>({n_faces, 3});
     T *fmean = fmean_arr.data();
 
-#pragma omp parallel for
     for (size_t i = 0; i < n_faces; i++) {
         const int64_t point0 = f[i * 3 + 0];
         const int64_t point1 = f[i * 3 + 1];
@@ -221,7 +206,6 @@ FaceNormals(const NDArray<const T, 2> points, const NDArray<const int64_t, 2> fa
     const T *v = points.data();
     const int64_t *f = faces.data();
 
-#pragma omp parallel for
     for (size_t i = 0; i < n_faces; i++) {
         int64_t point0 = f[i * 3 + 0];
         int64_t point1 = f[i * 3 + 1];
@@ -299,7 +283,6 @@ nb::tuple RayTrace(
     int *near_ind = near_ind_arr.data();
 
     // Loop through each face and determine intersections
-#pragma omp parallel for num_threads(num_threads)
     for (size_t i = 0; i < npoints; i++) {
         T prev_dist = std::numeric_limits<T>::infinity();
         int near_idx = -1;
@@ -516,11 +499,8 @@ nb::tuple PointWeights(
     const T *v = points_arr.data();
     const int64_t *f = faces_arr.data();
 
-#pragma omp parallel num_threads(n_threads)
-    {
-        T *local_pweight = AllocateArray<T>(n_points, true);
+    T *local_pweight = AllocateArray<T>(n_points, true);
 
-#pragma omp for
         for (size_t i = 0; i < n_faces; i++) {
             int64_t point0 = f[i * 3 + 0];
             int64_t point1 = f[i * 3 + 1];
@@ -544,27 +524,23 @@ nb::tuple PointWeights(
             local_pweight[point2] += farea_l;
         }
 
-#pragma omp critical
         for (size_t i = 0; i < n_points; i++) {
             pweight[i] += local_pweight[i];
         }
 
         delete[] local_pweight;
-    }
 
     // ensure this actually helps
     const T *pweight_const = pweight;
 
     if (n_add_weights) {
-#pragma omp parallel num_threads(n_threads)
         for (size_t i = 0; i < n_points; i++) {
             const T wgt = aweights[i] * pweight_const[i];
             wvertex[i * 3 + 0] = wgt * v[i * 3 + 0];
             wvertex[i * 3 + 1] = wgt * v[i * 3 + 1];
             wvertex[i * 3 + 2] = wgt * v[i * 3 + 2];
         }
     } else {
-#pragma omp parallel num_threads(n_threads)
         for (size_t i = 0; i < n_points; i++) {
             const T wgt = pweight[i];
             wvertex[i * 3 + 0] = wgt * v[i * 3 + 0];
@@ -1315,7 +1291,6 @@ template <typename T> NDArray<T, 1> TriArea(NDArray<T, 2> points, NDArray<int64_
     auto v = points.view();
     auto f = faces.view();
 
-#pragma omp parallel for firstprivate(v, f, tria_view)
     for (size_t i = 0; i < n_faces; i++) {
         int64_t point0 = f(i, 0);
         int64_t point1 = f(i, 1);