Merge pull request #105 from nauaneed/bump-cython

Bump cython, numpy

.github/workflows/tests.yml

@@ -12,19 +12,20 @@ on:
 jobs:
   tests:
     strategy:
+      fail-fast: false
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
-        python-version: [3.8, 3.9]
+        python-version: [3.11, 3.12]
 
     runs-on: ${{ matrix.os }}
     defaults:
       run:
         shell: bash -l {0}
 
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - name: Set up Python ${{ matrix.python-version }}
-        uses: conda-incubator/setup-miniconda@v2
+        uses: conda-incubator/setup-miniconda@v3
         with:
           auto-update-conda: true
           python-version: ${{ matrix.python-version }}
@@ -35,6 +36,20 @@ jobs:
           conda install -c conda-forge pocl pyopencl
           python -c 'import pyopencl as cl'
         if: ${{ runner.os != 'Windows' }}
+      - name: Setup compyle config on MacOS to use openmp enabled clang from homebrew
+        run: |
+          brew install libomp
+          mkdir -p ~/.compyle
+          touch ~/.compyle/config.py
+          echo "import os" >> ~/.compyle/config.py
+          echo "os.environ['CC'] = '$(brew --prefix llvm@15)/bin/clang'" >> ~/.compyle/config.py
+          echo "os.environ['CXX'] = '$(brew --prefix llvm@15)/bin/clang++'" >> ~/.compyle/config.py
+          export CPPFLAGS="-I$(brew --prefix libomp)/include -I$(brew --prefix llvm@15)/include -Xclang -fopenmp"
+          export LDFLAGS="-L$(brew --prefix libomp)/lib -L$(brew --prefix llvm@15)/lib -lomp"
+          python -c "import os; OMP_CFLAGS=os.environ.get('CPPFLAGS').split(' '); print(f'{OMP_CFLAGS=}')" >> ~/.compyle/config.py
+          python -c "import os; OMP_LINK=os.environ.get('LDFLAGS').split(' '); print(f'{OMP_LINK=}')" >> ~/.compyle/config.py
+          cat ~/.compyle/config.py
+        if: ${{ runner.os == 'macOS' }}
       - name: Install dependencies
         run: |
           conda info
@@ -46,9 +61,9 @@ jobs:
       # Thanks https://stackoverflow.com/a/60942437
       - name: Get month to use as cache key.
         id: month
-        run: echo "::set-output name=month::$(date +'%m')"
+        run: echo "month==$(date +'%m')" >> $GITHUB_OUTPUT
       - name: Deal with auto-generated code cache
-        uses: actions/cache@v2
+        uses: actions/cache@v4
         with:
           path: |
             ~/.compyle
@@ -61,6 +76,6 @@ jobs:
         if: ${{ success() }}
         run: coverage report
       - name: Upload Coverage to Codecov
-        uses: codecov/codecov-action@v1
+        uses: codecov/codecov-action@v4
         with:
           env_vars: ${{ matrix.os }}, ${{ matrix.python-version }}

compyle/cython_generator.py

@@ -448,7 +448,7 @@ def _get_c_method_spec(self, name, returns, args):
         c_arg_def = ', '.join(c_args)
         if self._config.use_openmp:
             ignore = ['reduce', 'converged']
-            gil = " nogil" if name not in ignore else ""
+            gil = " noexcept nogil" if name not in ignore else ""
         else:
             gil = ""
         cdefn = 'cdef inline {ret} {name}({arg_def}){gil}:'.format(

compyle/opencl.py

@@ -29,10 +29,10 @@ def set_context(ctx):
 def get_queue():
     global _queue
     if _queue is None:
-        properties = None
+        kwargs = dict()
         if get_config().profile:
-            properties = cl.command_queue_properties.PROFILING_ENABLE
-        _queue = cl.CommandQueue(get_context(), properties=properties)
+            kwargs['properties'] = cl.command_queue_properties.PROFILING_ENABLE
+        _queue = cl.CommandQueue(get_context(), **kwargs)
     return _queue
 
 

compyle/parallel.py

@@ -755,7 +755,7 @@ def _generate(self, declarations=None):
     def _correct_return_type(self, c_data):
         code = self.tp.blocks[-1].code.splitlines()
         if self._config.use_openmp:
-            gil = " nogil"
+            gil = " noexcept nogil"
         else:
             gil = ""
         code[0] = "cdef inline {type} {name}({args}){gil}:".format(
@@ -904,7 +904,7 @@ def _get_backend_key(self):
     def _correct_return_type(self, c_data, modifier):
         code = self.tp.blocks[-1].code.splitlines()
         if self._config.use_openmp:
-            gil = " nogil"
+            gil = " noexcept nogil"
         else:
             gil = ""
         code[0] = "cdef inline {type} {name}_{modifier}({args}){gil}:".format(

compyle/tests/test_array.py

@@ -274,9 +274,14 @@ def test_sort_by_keys(backend):
     out_array1, out_array2 = array.sort_by_keys([dev_array1, dev_array2])
 
     # Then
-    order = np.argsort(nparr1)
+    if backend == 'opencl': # opencl backend => radix sort which is stable
+        order = np.argsort(nparr1, stable=True)
+    else:
+        order = np.argsort(nparr1)
     act_result1 = np.take(nparr1, order)
     act_result2 = np.take(nparr2, order)
+
+    # find positions of duplicate values in the sorted array
     assert np.all(out_array1.get() == act_result1)
     assert np.all(out_array2.get() == act_result2)
 
@@ -295,7 +300,7 @@ def test_radix_sort_by_keys():
                                                     use_radix_sort=True)
 
         # Then
-        order = np.argsort(nparr1)
+        order = np.argsort(nparr1, stable=True)
         act_result1 = np.take(nparr1, order)
         act_result2 = np.take(nparr2, order)
         assert np.all(out_array1.get() == act_result1)
@@ -322,7 +327,10 @@ def test_sort_by_keys_with_output(backend):
                        out_list=out_arrays, use_radix_sort=False)
 
     # Then
-    order = np.argsort(nparr1)
+    if backend == 'opencl': # opencl backend => radix sort which is stable
+        order = np.argsort(nparr1, stable=True)
+    else:
+        order = np.argsort(nparr1)
     act_result1 = np.take(nparr1, order)
     act_result2 = np.take(nparr2, order)
     assert np.all(out_arrays[0].get() == act_result1)

compyle/tests/test_cython_generator.py

@@ -498,7 +498,7 @@ def __init__(self, **kwargs):
                 for key, value in kwargs.items():
                     setattr(self, key, value)
 
-            cdef inline void func(self, long d_idx, double* d_x) nogil:
+            cdef inline void func(self, long d_idx, double* d_x) noexcept nogil:
                 cdef double tmp
                 tmp = abs(self.rho*self.c)*sin(pi*self.c)
                 d_x[d_idx] = d_x[d_idx]*tmp

compyle/types.py

@@ -188,6 +188,7 @@ def _inject_types_in_module():
     NP_C_TYPE_MAP[np.dtype(np.uint64)] = 'unsigned long long'
     C_NP_TYPE_MAP['long long'] = np.int64
     C_NP_TYPE_MAP['unsigned long long'] = np.uint64
+    TYPES['long long'] = KnownType('long long')
     TYPES['glonglongp'] = KnownType('GLOBAL_MEM long long*', 'long long')
     TYPES['gulonglongp'] = KnownType('GLOBAL_MEM unsigned long long*',
                                      'unsigned long long')

pyproject.toml

@@ -2,7 +2,7 @@
 requires = [
     "wheel>=0.29.0",
     "setuptools>=42.0.0",
-    "oldest-supported-numpy",
+    "numpy>=2.0,<3",
     "Cython>=0.20",
     "mako",
     "pytools"