ci: Add icx workflow

Author: georgebisbas

.github/workflows/pytest-core-nompi.yml

@@ -34,7 +34,8 @@ jobs:
            pytest-ubuntu-py39-gcc9-omp,
            pytest-osx-py37-clang-omp,
            pytest-docker-py37-gcc-omp,
-           pytest-docker-py37-icc-omp
+           pytest-docker-py37-icc-omp,
+           pytest-docker-py38-icx-omp
         ]
         set: [base, adjoint]
         include:
@@ -101,6 +102,13 @@ jobs:
           language: "openmp"
           sympy: "1.11"
 
+        - name: pytest-docker-py37-icx-omp
+          python-version: '3.8'
+          os: ubuntu-22.04
+          arch: "icx"
+          language: "openmp"
+          sympy: "1.11"
+
         - set: base
           test-set: 'not adjoint'
 
@@ -135,7 +143,7 @@ jobs:
           fi
       id: set-run
 
-    - name: Install GCC ${{ matrix.arch }}
+    - name: Install ${{ matrix.arch }} compiler
       if: "runner.os == 'linux' && !contains(matrix.name, 'docker')"
       run : |
         sudo apt-get install -y ${{ matrix.arch }}

devito/arch/compiler.py

@@ -552,6 +552,8 @@ def __init__(self, *args, **kwargs):
                 self.cflags.append('-fsycl-targets=spir64')
 
         if language == 'openmp':
+            # To be switched to `-fiopenmp` or `-qopenmp` as soon as
+            # it is fixed in the new Intel OneAPI release
             self.cflags.append('-fopenmp')
             if platform is NVIDIAX:
                 self.cflags.append('-fopenmp-targets=nvptx64-cuda')
@@ -566,8 +568,8 @@ def __init__(self, *args, **kwargs):
 
         # Make sure the MPI compiler uses `icx` underneath -- whatever the MPI distro is
         if kwargs.get('mpi'):
-            ver = check_output([self.MPICC, "--version"]).decode("utf-8")
-            if not ver.startswith("Intel(R) oneAPI"):
+            mpi_distro = sniff_mpi_distro('mpiexec')
+            if not mpi_distro.startswith("Intel(R) oneAPI"):
                 warning("The MPI compiler `%s` doesn't use the Intel(R) oneAPI "
                         "DPC++/C++ compiler underneath" % self.MPICC)