Check for __CUDA_ARCH_LIST__ for setting GT_CUDA_ARCH #1225

	name: CI

	on:
	push:
	branches: [ master ]
	pull_request:
	branches: [ master ]

	jobs:
	perftests:
	runs-on: ubuntu-latest
	container: ghcr.io/gridtools/gridtools-base:${{ matrix.compiler }}
	strategy:
	matrix:
	compiler: [gcc-8, gcc-9, gcc-10, gcc-11, gcc-12, gcc-13, clang-11, clang-12, clang-13, clang-14, clang-15, clang-16, clang-14-cuda-11, gcc-10-cuda-11.8, gcc-11-cuda-12.0, gcc-12-cuda-12.1, gcc-12-cuda-12.2, gcc-12-cuda-12.3, base-hip, gcc-10-hpx, nvhpc-23.3, nvhpc-23.9]
	build_type: [debug, release]
	exclude:
	- compiler: gcc-8
	build_type: debug
	- compiler: gcc-9
	build_type: debug
	- compiler: gcc-10
	build_type: debug
	- compiler: gcc-11
	build_type: debug
	- compiler: clang-13
	build_type: debug
	- compiler: clang-14
	build_type: debug
	- compiler: gcc-10-hpx
	build_type: debug
	steps:
	- uses: actions/checkout@v2
	- name: setup environment
	run: \|
	echo "GTRUN_BUILD_COMMAND=make -j $(nproc)" >> $GITHUB_ENV
	echo "CTEST_PARALLEL_LEVEL=$(nproc)" >> $GITHUB_ENV
	echo "OMP_NUM_THREADS=$(nproc)" >> $GITHUB_ENV
	- name: build
	run: \|
	python3 pyutils/driver.py -vv build -b ${{ matrix.build_type }} -o $(pwd)/build -i $(pwd)/install -t perftests
	- name: run tests
	# no GPUs available -> no tests (for nvhpc we could run cpu, but we currently don't expose that option in pyutils/driver.py)
	if: (!contains(matrix.compiler, 'cuda') && !contains(matrix.compiler, 'hip') && !contains(matrix.compiler, 'nvhpc'))
	run: \|
	$(pwd)/build/pyutils/driver.py -vv test --perftests-only

Provide feedback