yambo: improve Yambo recipe and add new DeviceXlib recipe both with GPU support

repos/spack_repo/builtin/packages/devicexlib/package.py

@@ -0,0 +1,230 @@
+# Copyright Spack Project Developers. See COPYRIGHT file for details.
+#
+# SPDX-License-Identifier: (Apache-2.0 OR MIT)
+
+from spack_repo.builtin.build_systems.autotools import AutotoolsPackage
+from spack_repo.builtin.build_systems.cuda import CudaPackage
+from spack_repo.builtin.build_systems.rocm import ROCmPackage
+
+from spack.package import *
+
+
+class Devicexlib(AutotoolsPackage, CudaPackage, ROCmPackage):
+    """Library wrapping device-oriented routines and utilities.
+
+    deviceXlib provides wrappers for device data allocation and host-device data
+    transfers. It supports CUDA, OpenACC, and OpenMP programming paradigms, and
+    wraps a subset of routines from NVIDIA cuBLAS, Intel oneMKL BLAS, and AMD
+    rocBLAS libraries.
+    """
+
+    homepage = "https://gitlab.com/max-centre/components/devicexlib"
+    url = "https://gitlab.com/max-centre/components/devicexlib/-/archive/0.9.0/devicexlib-0.9.0.tar.gz"
+    git = "https://gitlab.com/max-centre/components/devicexlib"
+
+    license("MIT")
+    maintainers("nicspalla")
+
+    version("develop", branch="develop")
+    version("0.9.1", sha256="900fe8b0849d451e7c541d00a1b92c723e0969bae47ebcabd295e14ebcc17d1e")
+    with default_args(deprecated=True):
+        version("0.9.0", sha256="77c57a31381a69a2eb2a77138b131a553c96aff03ca934c88b8a6d8434b39460")
+        version("0.8.6", sha256="36e6222bc59cf0ed7268cc3652a3661887109f7fe072cefe06884dcd6de2407d")
+        version("0.8.5", sha256="498d5c6804e697123d382d9dd35dedeb4b64228704f84711877c842b851d37df")
+
+    depends_on("c", type="build")
+    depends_on("fortran", type="build")
+
+    variant(
+        "openmp",
+        default=False,
+        description="Enable OpenMP support",
+    )
+
+    variant(
+        "openmp5",
+        default=False,
+        description="Build with OpenMP-GPU support",
+    )
+
+    variant(
+        "openacc",
+        default=False,
+        description="Build with OpenACC",
+    )
+
+    variant(
+        "cuda-fortran",
+        default=False,
+        description="Build with CUDA Fortran",
+    )
+
+    variant(
+        "nvtx",
+        default=False,
+        description="Enable NVTX support",
+        when="+cuda",
+    )
+
+    variant(
+        "roctx",
+        default=False,
+        description="Enable ROCTX support",
+        when="+rocm",
+    )
+
+    variant(
+        "mkl",
+        default=False,
+        description="Enable MKL-GPU support",
+    )
+
+    with when("@0.9.0: +openacc"):
+        variant(
+            "openacc-debug",
+            default=False,
+            description="Enable OpenACC DEBUG macro",
+        )
+
+    with when("+cuda"):
+        variant(
+            "cuda_rt",
+            values=str,
+            default="none",
+            when="%nvhpc",
+            description=(
+                'Specify the CUDA runtime version, e.g. "11.8", only if you '
+                "want the secondary version installed with the NVHPC SDK."
+            ),
+        )
+
+    with when("+cuda-fortran"):
+        requires(
+            "%nvhpc",
+            msg="CUDA Fortran is available only with NVIDIA compilers",
+        )
+
+    with when("+openacc"):
+        requires(
+            "%nvhpc",
+            "%gcc@10:+nvptx",
+            policy="one_of",
+            msg="OpenACC is available only with NVIDIA or GCC compilers",
+        )
+
+    with when("+openmp5"):
+        requires(
+            "%oneapi",
+            "%cce",
+            "%gcc@10:+nvptx",
+            policy="one_of",
+            msg=("OpenMP offloading is available only with GCC, oneAPI, or Cray compilers"),
+        )
+
+    depends_on("blas")
+
+    depends_on("intel-oneapi-mkl", when="+mkl")
+
+    with when("+openmp"):
+        depends_on("openblas threads=openmp", when="^[virtuals=blas] openblas")
+
+    conflicts(
+        "cuda_arch=none",
+        when="+cuda",
+        msg="CUDA architecture is required",
+    )
+
+    conflicts(
+        "cuda_arch=none",
+        when="+cuda-fortran",
+        msg="CUDA architecture is required",
+    )
+
+    conflicts(
+        "cuda_rt=none",
+        when="@:0.8.5 +cuda",
+        msg="CUDA runtime version is required",
+    )
+
+    def enable_or_disable_cuda(self, activated):
+        return "--enable-cublas=yes" if activated else "--enable-cublas=no"
+
+    def enable_or_disable_rocm(self, activated):
+        return "--enable-rocblas=yes" if activated else "--enable-rocblas=no"
+
+    def enable_or_disable_mkl(self, activated):
+        return "--enable-mkl-gpu=yes" if activated else "--enable-mkl-gpu=no"
+
+    def setup_build_environment(self, env):
+        spec = self.spec
+
+        if "%nvhpc" in spec:
+            env.set("CC", "nvc")
+            env.set("FC", "nvfortran")
+            env.set("F90", "nvfortran")
+            env.set("CPP", "cpp -E")
+            env.set("FPP", "nvfortran -Mpreprocess -E")
+            env.set("F90SUFFIX", ".f90")
+
+        if "%gcc" in spec:
+            env.set("CC", "gcc")
+            env.set("FC", "gfortran")
+            env.set("F90", "gfortran")
+            env.set("CPP", "gcc -E -P")
+            env.set("FPP", "gfortran -E -P")
+            env.set("F90SUFFIX", ".f90")
+
+    def configure_args(self):
+        spec = self.spec
+
+        args = [
+            "--enable-cuda-env-check=no",
+            "--enable-parallel=no",
+        ]
+
+        # OpenMP
+        args.extend(self.enable_or_disable("openmp"))
+
+        # GPU offloading
+        if "+cuda-fortran" in spec:
+            args.append("--enable-cuda-fortran")
+
+        if "+openacc" in spec:
+            args.append("--enable-openacc")
+
+        if "+openacc-debug" in spec:
+            args.append("--enable-openacc-debug")
+
+        if "+openmp5" in spec:
+            args.append("--enable-openmp5")
+
+        # BLAS
+        args.append(f"--with-blas-libs={spec['blas'].libs}")
+
+        # CUDA
+        args.extend(self.enable_or_disable("cuda"))
+
+        if "+cuda" in spec:
+            cuda_arch = spec.variants["cuda_arch"].value[0]
+            args.append(f"--with-cuda-cc={cuda_arch}")
+
+            if spec.variants["cuda_rt"].value != "none":
+                args.append(f"--with-cuda-runtime={spec.variants['cuda_rt'].value}")
+
+            if "%nvhpc" not in spec:
+                args.append(f"--with-cuda-path={spec['cuda'].home}")
+
+        # ROCm
+        args.extend(self.enable_or_disable("rocm"))
+
+        if "+rocm" in spec:
+            args.append(f"--with-rocm-path={spec['hip'].home}")
+
+        # MKL
+        args.extend(self.enable_or_disable("mkl"))
+
+        return args
+
+    @property
+    def build_targets(self):
+        return ["all"]

repos/spack_repo/builtin/packages/yambo/cuda_runtime.patch

@@ -0,0 +1,90 @@
+--- a/configure
++++ b/configure
+@@ -1860,7 +1860,8 @@ Optional Packages:
+   --with-cuda-path=<path>       Path to libcuda install directory
+   --with-cuda-cc=VAL      GPU architecture (Kepler: 35, Pascal: 60, Volta: 70,
+                           Ampere: 80) [default=70]
+-  --with-cuda-runtime=VAL CUDA runtime (Pascal: 8+, Volta: 9+) [default=10.1]
++  --with-cuda-runtime=VAL CUDA runtime (Pascal: 8+, Volta: 9+) [default=none,
++                          checks if the NVHPC_CUDA_HOME variable is set.]
+   --with-cuda-int-libs=VAL
+                           CUDA internal libraries ()
+                           [default=cuda,cufft,cublas,cusolver,cudart]
+@@ -16049,7 +16050,7 @@ if test ${with_cuda_runtime+y}
+ then :
+   withval=$with_cuda_runtime;
+ else $as_nop
+-  with_cuda_runtime=10.1
++  with_cuda_runtime=none
+ fi
+
+ #
+@@ -16194,17 +16195,26 @@ if test x"$enable_cuda_fortran" != "xno" ; then
+    #
+    # Flags to be passed to the devicexlib library
+    #
+-   DEVXLIB_FLAGS="--enable-openmp --enable-cuda-fortran --with-cuda-cc=${with_cuda_cc} --with-cuda-runtime=${with_cuda_runtime}"
++   DEVXLIB_FLAGS="--enable-openmp --enable-cuda-fortran --with-cuda-cc=${with_cuda_cc}"
++   if test "x$with_cuda_runtime" != "xnone" ; then
++     DEVXLIB_FLAGS+=" --with-cuda-runtime=${with_cuda_runtime}"
++   fi
+    #
+    case "${FCVERSION}" in
+     *nvfortran*)
+-      GPU_FLAGS="-cuda -gpu=cc${with_cuda_cc},cuda${with_cuda_runtime}"
++      GPU_FLAGS="-cuda -gpu=cc${with_cuda_cc}"
++      if test "x$with_cuda_runtime" != "xnone" ; then
++        GPU_FLAGS+=" -gpu=cuda${with_cuda_runtime}";
++      fi
+       if test x"$use_int_cuda_libs" = "xyes" ; then
+         GPU_FLAGS+=" -cudalib=${with_cuda_int_libs}";
+       fi
+       ;;
+     *)
+-      GPU_FLAGS="-Mcuda=cc${with_cuda_cc},cuda${with_cuda_runtime}"
++      GPU_FLAGS="-Mcuda=cc${with_cuda_cc}"
++      if test "x$with_cuda_runtime" != "xnone" ; then
++        GPU_FLAGS+=" -gpu=cuda${with_cuda_runtime}"
++      fi
+       if test x"$use_int_cuda_libs" = "xyes" ; then
+         GPU_FLAGS+=" -Mcudalib=${with_cuda_int_libs}"
+       fi
+@@ -16338,14 +16348,17 @@ if test x"$enable_openacc" != "xno" ; then
+    case "${FCVERSION}" in
+     *nvfortran*)
+       DEVXLIB_FLAGS+="--enable-openmp"
+-      GPU_FLAGS="-acc=gpu,multicore -acclibs -gpu=cc${with_cuda_cc},cuda${with_cuda_runtime} " # -gpu=cc${with_cuda_cc},cuda${with_cuda_runtime}"
++      GPU_FLAGS="-acc=gpu,multicore -acclibs -gpu=cc${with_cuda_cc}"
++      if test "x$with_cuda_runtime" != "xnone" ; then
++        GPU_FLAGS+=" -gpu=cuda${with_cuda_runtime}"
++      fi
+       if test x"$use_int_cuda_libs" = "xyes" ; then
+         GPU_FLAGS+=" -cudalib=${with_cuda_int_libs}";
+       fi
+       ;;
+     *pgfortran*)
+       DEVXLIB_FLAGS+="--enable-openmp"
+-      GPU_FLAGS="-acc -acclibs -ta=tesla:cc${with_cuda_cc} "           # -gpu=cc${with_cuda_cc},cuda${with_cuda_runtime}"
++      GPU_FLAGS="-acc -acclibs -ta=tesla:cc${with_cuda_cc} "
+       if test x"$use_int_cuda_libs" = "xyes" ; then
+         GPU_FLAGS+=" -cudalib=${with_cuda_int_libs}";
+       fi
+@@ -16354,7 +16367,6 @@ if test x"$enable_openacc" != "xno" ; then
+       GPU_FLAGS="-fopenacc -foffload=-lm  -foffload=-lgfortran"
+       # -foffload=nvptx-none
+       # -foffload=amdgcn-amdhsa  -foffload=-march=gfx908
+-      # -gpu=cc${with_cuda_cc},cuda${with_cuda_runtime}
+    esac
+    #
+ fi
+@@ -16366,8 +16378,8 @@ if test x"$enable_openmp5" != "xno" ; then
+    # Flags to be passed to the devicexlib library
+    #
+    def_gpu="-D_GPU -D_OPENMP_GPU"
+-   GPU_FLAGS="-fopenmp" # -gpu=cc${with_cuda_cc},cuda${with_cuda_runtime}"
+-   DEVXLIB_FLAGS="--enable-openmp5" # --with-cuda-cc=${with_cuda_cc} --with-cuda-runtime=${with_cuda_runtime}"
++   GPU_FLAGS="-fopenmp"
++   DEVXLIB_FLAGS="--enable-openmp5"
+    if test x"$LIBROCM_LIBS" != "x" ; then
+      DEVXLIB_FLAGS+=" --enable-rocblas --with-rocm-libs=$LIBROCM_LIBS";
+      def_gpu="$def_gpu -D_HIP"

repos/spack_repo/builtin/packages/yambo/hdf5.patch

@@ -0,0 +1,11 @@
+--- a/configure
++++ b/configure
+@@ -12970,7 +12970,7 @@ ac_compiler_gnu=$ac_cv_fc_compiler_gnu
+   #
+   if test x"$with_hdf5_libs" != "x" ; then try_HDF5_LIBS="$with_hdf5_libs" ; fi
+   #
+-  if test -d "$try_hdf5_libdir" ; then try_HDF5_LIBS="-L$try_hdf5_libdir -lhdf5hl_fortran -lhdf5_fortran -lhdf5_hl -lhdf5" ; fi
++  if test -d "$try_hdf5_libdir" ; then try_HDF5_LIBS="-L$try_hdf5_libdir -lhdf5_hl_fortran -lhdf5_fortran -lhdf5_hl -lhdf5" ; fi
+   #
+   if test -d "$try_hdf5_incdir" ; then try_HDF5_INCS="$IFLAG$try_hdf5_incdir" ; fi
+   #