Merge pull request numpy#24174 from charris/backport-24162

ENH: Improve clang-cl compliance

Author: charris

.github/workflows/windows_clangcl.yml

@@ -0,0 +1,91 @@
+name: Test Clang-CL Build (Windows)
+
+on:
+  pull_request:
+    branches:
+      - main
+      - maintenance/**
+
+env:
+  PYTHON_VERSION: 3.11
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read # to fetch code (actions/checkout)
+
+jobs:
+  meson:
+    name: Meson windows build/test
+    runs-on: windows-2019
+    # if: "github.repository == 'numpy/numpy'"
+    steps:
+    - name: Checkout
+      uses: actions/checkout@c85c95e3d7251135ab7dc9ce3241c5835cc595a9 # v3.5.3
+      with:
+        submodules: recursive
+        fetch-depth: 0
+    - name: Setup Python
+      uses: actions/setup-python@bd6b4b6205c4dbad673328db7b31b7fab9e241c0 # v4.6.1
+      with:
+        python-version: ${{ env.PYTHON_VERSION }}
+
+    - name: Install dependencies
+      run: |
+        pip install -r build_requirements.txt
+    - name: openblas-libs
+      run: |
+        # Download and install pre-built OpenBLAS library
+        # with 32-bit interfaces
+        # Unpack it in the pkg-config hardcoded path
+        choco install unzip -y
+        choco install wget -y
+        # Install llvm, which contains clang-cl
+        choco install llvm -y --version=16.0.6 
+        choco install -y --checksum 6004DF17818F5A6DBF19CB335CC92702 pkgconfiglite
+        wget https://anaconda.org/multibuild-wheels-staging/openblas-libs/v0.3.21/download/openblas-v0.3.21-win_amd64-gcc_10_3_0.zip
+        unzip -d c:\opt openblas-v0.3.21-win_amd64-gcc_10_3_0.zip
+        echo "PKG_CONFIG_PATH=c:\opt\64\lib\pkgconfig;" >> $env:GITHUB_ENV
+    - name: meson-configure
+      run: |
+        "[binaries]","c = 'clang-cl'","cpp = 'clang-cl'","ar = 'llvm-lib'","c_ld = 'lld-link'","cpp_ld = 'lld-link'"  | Out-File $PWD/clang-cl-build.ini -Encoding ascii
+        meson setup build --prefix=$PWD\build-install --native-file=$PWD/clang-cl-build.ini -Ddebug=false --optimization 2 --vsenv
+    - name: meson-build
+      run: |
+        meson compile -C build -v
+
+    - name: meson-install
+      run: |
+        cd build
+        meson install --no-rebuild
+    - name: build-path
+      run: |
+        echo "installed_path=$PWD\build-install\Lib\site-packages" >> $env:GITHUB_ENV
+    - name: post-install
+      run: |
+        $numpy_path = "${env:installed_path}\numpy"
+        $libs_path = "${numpy_path}\.libs"
+        mkdir ${libs_path}
+        $ob_path = "C:/opt/64/bin/"
+        cp $ob_path/*.dll $libs_path
+        # Write _distributor_init.py to load .libs DLLs.
+        python -c "from tools import openblas_support; openblas_support.make_init(r'${numpy_path}')"
+
+    - name: prep-test
+      run: |
+        echo "PYTHONPATH=${env:installed_path}" >> $env:GITHUB_ENV
+        python -m pip install -r test_requirements.txt
+        python -m pip install threadpoolctl
+
+    - name: test
+      run: |
+        mkdir tmp
+        cd tmp
+        echo "============================================"
+        python -c "import numpy; print(numpy.show_runtime())"
+        echo "============================================"
+        echo "LASTEXITCODE is '$LASTEXITCODE'"
+        python -c "import numpy, sys; sys.exit(numpy.test(verbose=3) is False)"
+        echo "LASTEXITCODE is '$LASTEXITCODE'"

build_requirements.txt

@@ -2,4 +2,4 @@ meson-python>=0.10.0
 Cython>=0.29.34,<3.0
 wheel==0.38.1
 ninja
-spin==0.3
+spin==0.4

doc_requirements.txt

@@ -11,3 +11,4 @@ breathe>4.33.0
 
 # needed to build release notes
 towncrier
+toml

numpy/core/src/multiarray/_multiarray_tests.c.src

@@ -2054,13 +2054,13 @@ get_fpu_mode(PyObject *NPY_UNUSED(self), PyObject *args)
         return NULL;
     }
 
-#if defined(_MSC_VER)
+#if defined(_MSC_VER) && !defined(__clang__)
     {
         unsigned int result = 0;
         result = _controlfp(0, 0);
         return PyLong_FromLongLong(result);
     }
-#elif defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
+#elif (defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))) || (defined(_MSC_VER) && defined(__clang__))
     {
         unsigned short cw = 0;
         __asm__("fstcw %w0" : "=m" (cw));

numpy/core/tests/test_einsum.py

@@ -1,4 +1,6 @@
 import itertools
+import sys
+import platform
 
 import pytest
 
@@ -8,6 +10,12 @@
     assert_raises, suppress_warnings, assert_raises_regex, assert_allclose
     )
 
+try:
+    COMPILERS = np.show_config(mode="dicts")["Compilers"]
+    USING_CLANG_CL = COMPILERS["c"]["name"] == "clang-cl"
+except TypeError:
+    USING_CLANG_CL = False
+
 # Setup for optimize einsum
 chars = 'abcdefghij'
 sizes = np.array([2, 3, 4, 5, 4, 3, 2, 6, 5, 4, 3])
@@ -102,7 +110,7 @@ def test_einsum_errors(self):
 
     def test_einsum_object_errors(self):
         # Exceptions created by object arithmetic should
-        # successfully propogate
+        # successfully propagate
 
         class CustomException(Exception):
             pass
@@ -609,9 +617,23 @@ def check_einsum_sums(self, dtype, do_opt=False):
                            [2.])  # contig_stride0_outstride0_two
 
     def test_einsum_sums_int8(self):
+        if (
+                (sys.platform == 'darwin' and platform.machine() == 'x86_64')
+                or
+                USING_CLANG_CL
+        ):
+            pytest.xfail('Fails on macOS x86-64 and when using clang-cl '
+                         'with Meson, see gh-23838')
         self.check_einsum_sums('i1')
 
     def test_einsum_sums_uint8(self):
+        if (
+                (sys.platform == 'darwin' and platform.machine() == 'x86_64')
+                or
+                USING_CLANG_CL
+        ):
+            pytest.xfail('Fails on macOS x86-64 and when using clang-cl '
+                         'with Meson, see gh-23838')
         self.check_einsum_sums('u1')
 
     def test_einsum_sums_int16(self):

numpy/core/tests/test_scalarmath.py

@@ -17,6 +17,12 @@
     assert_warns, _SUPPORTS_SVE,
     )
 
+try:
+    COMPILERS = np.show_config(mode="dicts")["Compilers"]
+    USING_CLANG_CL = COMPILERS["c"]["name"] == "clang-cl"
+except TypeError:
+    USING_CLANG_CL = False
+
 types = [np.bool_, np.byte, np.ubyte, np.short, np.ushort, np.intc, np.uintc,
          np.int_, np.uint, np.longlong, np.ulonglong,
          np.single, np.double, np.longdouble, np.csingle,
@@ -798,7 +804,13 @@ class TestBitShifts:
     @pytest.mark.parametrize('op',
         [operator.rshift, operator.lshift], ids=['>>', '<<'])
     def test_shift_all_bits(self, type_code, op):
-        """ Shifts where the shift amount is the width of the type or wider """
+        """Shifts where the shift amount is the width of the type or wider """
+        if (
+                USING_CLANG_CL and
+                type_code in ("l", "L") and
+                op is operator.lshift
+        ):
+            pytest.xfail("Failing on clang-cl builds")
         # gh-2449
         dt = np.dtype(type_code)
         nbits = dt.itemsize * 8

test_requirements.txt

@@ -1,10 +1,12 @@
 cython>=0.29.34,<3.0
 wheel==0.38.1
-setuptools==59.2.0
-hypothesis==6.24.1
-pytest==6.2.5
-pytz==2021.3
-pytest-cov==3.0.0
+setuptools==59.2.0 ; python_version < '3.12'
+setuptools         ; python_version >= '3.12'
+hypothesis==6.81.1
+pytest==7.4.0
+pytz==2023.3
+pytest-cov==4.1.0
+pytest-xdist
 # for numpy.random.test.test_extending
 cffi; python_version < '3.10'
 # For testing types. Notes on the restrictions: