Update cp2k recipe to use cmake or the current build system (#35718)

* Update cp2k recipe to use cmake or the current build system

Offers the possibility to build cp2k with the new cmake build system. commands like this are now supported

spack install cp2k@master build_system=cmake +.....

the recipe supports the following optional functionalities

- superlu, cosma, sirius, spglib, metis, spglib, libxc, libint, cuda/rocm, mkl/openblas/sci (and others), mpi, openmp, dbcsr
- dbcsr is built separately using the currently available recipe.

Two PRs need to be merged to be fully functional (cosma update in spack + one PR in cp2k github).

* Fix indentation

* Fix indentation

* Update libvori

* More typos

* Simplify BLAS/LAPACK

* Simplify BLAS/LAPACK

* Add A100 gpu value

* Fix typo

* Add the enable_regtests option

if -DCP2K_ENABLE_REGTESTS=ON (+enable_regtests with spack) then the location of the binary executables will be in the cp2k root directory under exe/build-cmake-*. This option is needed to run the regtests afterwards.

* Minor update

* more fixes

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>

* small changes

* Remove any reference to nvidia architecture in the rocm list

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Alberto Invernizzi <9337627+albestro@users.noreply.github.com>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Alberto Invernizzi <9337627+albestro@users.noreply.github.com>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Alberto Invernizzi <9337627+albestro@users.noreply.github.com>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Alberto Invernizzi <9337627+albestro@users.noreply.github.com>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Alberto Invernizzi <9337627+albestro@users.noreply.github.com>

* Update var/spack/repos/builtin/packages/cp2k/package.py

Co-authored-by: Alberto Invernizzi <9337627+albestro@users.noreply.github.com>

* Final reformating

* Update py-fypp

---------

Co-authored-by: Mikael Simberg <mikael.simberg@iki.fi>
Co-authored-by: Alberto Invernizzi <9337627+albestro@users.noreply.github.com>

1 files changed, 210 insertions, 27 deletions

diff --git a/var/spack/repos/builtin/packages/cp2k/package.py b/var/spack/repos/builtin/packages/cp2k/package.py
index f3789b179f..6f012bdf4c 100644
--- a/var/spack/repos/builtin/packages/cp2k/package.py
+++ b/var/spack/repos/builtin/packages/cp2k/package.py
@@ -5,23 +5,33 @@
 import copy
 import os
 import os.path
+import sys
 
+import spack.platforms
 import spack.util.environment
+import spack.util.executable
+from spack.build_environment import dso_suffix
 from spack.package import *
 
 
-class Cp2k(MakefilePackage, CudaPackage):
+class Cp2k(MakefilePackage, CudaPackage, CMakePackage, ROCmPackage):
     """CP2K is a quantum chemistry and solid state physics software package
     that can perform atomistic simulations of solid state, liquid, molecular,
     periodic, material, crystal, and biological systems
     """
 
+    build_system(
+        conditional("cmake", when="@master:"),
+        conditional("makefile", when="@:2023.1"),
+        default="makefile",
+    )
+
     homepage = "https://www.cp2k.org"
     url = "https://github.com/cp2k/cp2k/releases/download/v3.0.0/cp2k-3.0.tar.bz2"
     git = "https://github.com/cp2k/cp2k.git"
     list_url = "https://github.com/cp2k/cp2k/releases"
 
-    maintainers("dev-zero")
+    maintainers("dev-zero", "mtaillefumier")
 
     version("2023.1", sha256="dff343b4a80c3a79363b805429bdb3320d3e1db48e0ff7d20a3dfd1c946a51ce")
     version("2022.2", sha256="1a473dea512fe264bb45419f83de432d441f90404f829d89cbc3a03f723b8354")
@@ -52,7 +62,7 @@ class Cp2k(MakefilePackage, CudaPackage):
     variant(
         "pexsi",
         default=False,
-        description=("Enable the alternative PEXSI method" "for density matrix evaluation"),
+        description="Enable the alternative PEXSI method for density matrix evaluation",
     )
     variant(
         "elpa",
@@ -63,15 +73,31 @@ class Cp2k(MakefilePackage, CudaPackage):
     variant(
         "sirius",
         default=False,
-        description=("Enable planewave electronic structure" " calculations via SIRIUS"),
+        description="Enable planewave electronic structure calculations via SIRIUS",
     )
     variant("cosma", default=False, description="Use COSMA for p?gemm")
     variant(
         "libvori",
         default=False,
-        description=("Enable support for Voronoi integration" " and BQB compression"),
+        description="Enable support for Voronoi integration and BQB compression",
     )
     variant("spglib", default=False, description="Enable support for spglib")
+    variant(
+        "spla",
+        default=False,
+        description="Use SPLA off-loading functionality. Only relevant when CUDA or ROCM"
+        " are enabled",
+    )
+    variant("pytorch", default=False, description="Enable libtorch support")
+    variant("quip", default=False, description="Enable quip support")
+
+    variant(
+        "enable_regtests",
+        default=False,
+        description="Configure cp2k to run the regtests afterwards."
+        " It build cp2k normally but put the executables in exe/cmake-build-* instead of the"
+        " conventional location. This option is only relevant when regtests need to be run.",
+    )
 
     with when("+cuda"):
         variant(
@@ -147,13 +173,13 @@ class Cp2k(MakefilePackage, CudaPackage):
             )
 
     with when("+libxc"):
-        depends_on("pkgconfig", type="build", when="@7.0:")
-        depends_on("libxc@2.2.2:3", when="@:5", type="build")
-        depends_on("libxc@4.0.3:4", when="@6.0:6.9", type="build")
+        depends_on("pkgconfig", when="@7.0:")
+        depends_on("libxc@2.2.2:3", when="@:5")
+        depends_on("libxc@4.0.3:4", when="@6.0:6.9")
         depends_on("libxc@4.0.3:4", when="@7.0:8.1")
         depends_on("libxc@5.1.3:5.1", when="@8.2:8")
-        depends_on("libxc@5.1.7:5.1", when="@9:2022")
-        depends_on("libxc@6:6.1", when="@2023:")
+        depends_on("libxc@5.1.7:5.1", when="@9:2022.2")
+        depends_on("libxc@6.1:", when="@2023.1:")
 
     with when("+mpi"):
         depends_on("mpi@2:")
@@ -163,6 +189,7 @@ class Cp2k(MakefilePackage, CudaPackage):
     with when("+cosma"):
         depends_on("cosma+scalapack")
         depends_on("cosma@2.5.1:", when="@9:")
+        depends_on("cosma@2.6.3:", when="@master:")
         depends_on("cosma+cuda", when="+cuda")
         conflicts("~mpi")
         # COSMA support was introduced in 8+
@@ -198,27 +225,36 @@ class Cp2k(MakefilePackage, CudaPackage):
         depends_on("sirius@7.0.0:7.0", when="@8:8.2")
         depends_on("sirius@7.2", when="@8.3:8.9")
         depends_on("sirius@7.3:", when="@9.1")
-        conflicts("~mpi")
+        depends_on("sirius@7.4:", when="@master")
+        conflicts("~mpi", msg="SIRIUS requires MPI")
         # sirius support was introduced in 7+
         conflicts("@:6")
 
     with when("+libvori"):
-        depends_on("libvori@201219:", when="@8.1", type="build")
-        depends_on("libvori@210412:", when="@8.2:", type="build")
+        depends_on("libvori@201219:", when="@8.1")
+        depends_on("libvori@210412:", when="@8.2:")
+        depends_on("libvori@220621:", when="@2023.1:")
         # libvori support was introduced in 8+
         conflicts("@:7")
 
     # the bundled libcusmm uses numpy in the parameter prediction (v7+)
     # which is written using Python 3
-    depends_on("py-numpy", when="@7:+cuda", type="build")
-    depends_on("python@3.6:", when="@7:+cuda", type="build")
+    depends_on("py-numpy", when="@7:+cuda")
+    depends_on("python@3.6:", when="@7:+cuda")
+    depends_on("py-fypp")
 
     depends_on("spglib", when="+spglib")
 
     # Apparently cp2k@4.1 needs an "experimental" version of libwannier.a
     # which is only available contacting the developer directly. See INSTALL
     # in the stage of cp2k@4.1
-    depends_on("wannier90", when="@3.0+mpi", type="build")
+    depends_on("wannier90", when="@3.0+mpi")
+
+    with when("build_system=cmake"):
+        depends_on("dbcsr")
+        depends_on("dbcsr+openmp", when="+openmp")
+        depends_on("dbcsr+cuda", when="+cuda")
+        depends_on("dbcsr+rocm", when="+rocm")
 
     # CP2K needs compiler specific compilation flags, e.g. optflags
     conflicts("%apple-clang")
@@ -231,14 +267,37 @@ class Cp2k(MakefilePackage, CudaPackage):
     # for optimal kernels. Note that we don't override the cuda_archs property
     # from the parent class, since the parent class defines constraints for all
     # versions. Instead just mark all unsupported cuda archs as conflicting.
-    dbcsr_cuda_archs = ("35", "37", "60", "70")
-    cuda_msg = "cp2k only supports cuda_arch {0}".format(dbcsr_cuda_archs)
 
-    for arch in CudaPackage.cuda_arch_values:
-        if arch not in dbcsr_cuda_archs:
-            conflicts("+cuda", when="cuda_arch={0}".format(arch), msg=cuda_msg)
+    supported_cuda_arch_list = ("35", "37", "60", "70", "80")
+    supported_rocm_arch_list = ("gfx906", "gfx908", "gfx90a", "gfx90a:xnack-", "gfx90a:xnack+")
+    gpu_map = {
+        "35": "K40",
+        "37": "K80",
+        "60": "P100",
+        "70": "V100",
+        "80": "A100",
+        "gfx906": "Mi50",
+        "gfx908": "Mi100",
+        "gfx90a": "Mi250",
+        "gfx90a:xnack-": "Mi250",
+        "gfx90a:xnack+": "Mi250",
+    }
+    cuda_msg = "cp2k only supports cuda_arch {0}".format(supported_cuda_arch_list)
+    rocm_msg = "cp2k only supports amdgpu_target {0}".format(supported_rocm_arch_list)
+
+    conflicts("+cuda", when="cuda_arch=none")
+
+    # ROCm already emits an error if +rocm amdgpu_target=none is given
 
-    conflicts("+cuda", when="cuda_arch=none", msg=cuda_msg)
+    with when("+cuda"):
+        for arch in CudaPackage.cuda_arch_values:
+            if arch not in supported_cuda_arch_list:
+                conflicts("+cuda", when="cuda_arch={0}".format(arch), msg=cuda_msg)
+
+    with when("+rocm"):
+        for arch in ROCmPackage.amdgpu_targets:
+            if arch not in supported_rocm_arch_list:
+                conflicts("+rocm", when="amdgpu_target={0}".format(arch), msg=rocm_msg)
 
     # Fix 2- and 3-center integral calls to libint
     patch(
@@ -311,7 +370,6 @@ class Cp2k(MakefilePackage, CudaPackage):
         nvflags = ["-O3"]
         ldflags = []
         libs = []
-        gpuver = ""
 
         # CP2K Makefile doesn't set C standard, but the source code uses
         # C99-style for-loops with inline definition of iterating variable.
@@ -534,6 +592,7 @@ class Cp2k(MakefilePackage, CudaPackage):
             fcflags += ["-I{0}".format(sirius.prefix.include.sirius)]
             libs += list(sirius.libs)
 
+        gpuver = ""
         if spec.satisfies("+cuda"):
             libs += [
                 "-L{}".format(spec["cuda"].libs.directories[0]),
@@ -576,11 +635,25 @@ class Cp2k(MakefilePackage, CudaPackage):
                     libs += ["-lcufft", "-lcublas"]
 
             cuda_arch = spec.variants["cuda_arch"].value[0]
-            if cuda_arch:
-                gpuver = {"35": "K40", "37": "K80", "60": "P100", "70": "V100"}[cuda_arch]
+            gpuver = gpu_map[cuda_arch]
+            if cuda_arch == "35" and spec.satisfies("+cuda_arch_35_k20x"):
+                gpuver = "K20X"
 
-                if cuda_arch == "35" and spec.satisfies("+cuda_arch_35_k20x"):
-                    gpuver = "K20X"
+        if "@2022: +rocm" in spec:
+            libs += [
+                "-L{}".format(spec["rocm"].libs.directories[0]),
+                "-L{}/stubs".format(spec["rocm"].libs.directories[0]),
+                "-lhipblas",
+                "-lhipfft",
+                "-lstdc++",
+            ]
+
+            cppflags += ["-D__OFFLOAD_HIP"]
+            acc_compiler_var = "hipcc"
+            acc_flags_var = "NVFLAGS"
+            cppflags += ["-D__ACC"]
+            cppflags += ["-D__DBCSR_ACC"]
+            gpuver = gpu_map[spec.variants["amdgpu_target"].value[0]]
 
         if "smm=libsmm" in spec:
             lib_dir = join_path("lib", self.makefile_architecture, self.makefile_version)
@@ -763,3 +836,113 @@ class Cp2k(MakefilePackage, CudaPackage):
         with spack.util.environment.set_env(CP2K_DATA_DIR=data_dir, PWD=self.build_directory):
             with working_dir(self.build_directory):
                 make("test", *self.build_targets)
+
+
+class CMakeBuilder(spack.build_systems.cmake.CMakeBuilder):
+    """Use the new cmake build system to build cp2k. It is the default when
+    building the master branch of cp2k."""
+
+    def cmake_args(self):
+        spec = self.spec
+        args = []
+
+        gpu_map = {
+            "35": "K40",
+            "37": "K80",
+            "60": "P100",
+            "70": "V100",
+            "80": "A100",
+            "gfx906": "Mi50",
+            "gfx908": "Mi100",
+            "gfx90a": "Mi250",
+            "gfx90a:xnack-": "Mi250",
+            "gfx90a:xnack+": "Mi250",
+        }
+
+        if "+cuda" in spec:
+            if (len(spec.variants["cuda_arch"].value) > 1) or spec.satisfies("cuda_arch=none"):
+                raise InstallError("CP2K supports only one cuda_arch at a time.")
+            else:
+                gpu_ver = gpu_map[spec.variants["cuda_arch"].value[0]]
+                args += ["-DCP2K_USE_ACCEL=CUDA"]
+                args += [self.define("CP2K_WITH_GPU", gpu_ver)]
+
+        if "+rocm" in spec:
+            if len(spec.variants["amdgpu_target"].value) > 1:
+                raise InstallError("CP2K supports only one amdgpu_target at a time.")
+            else:
+                gpu_ver = gpu_map[spec.variants["amdgpu_target"].value[0]]
+                args += ["-DCP2K_USE_ACCEL=HIP"]
+                args += [self.define("CP2K_WITH_GPU", gpu_ver)]
+
+        args += [
+            self.define_from_variant("CP2K_ENABLE_REGTESTS", "enable_regtests"),
+            self.define_from_variant("CP2K_USE_ELPA", "elpa"),
+            self.define_from_variant("CP2K_USE_LIBINT2", "libint"),
+            self.define_from_variant("CP2K_USE_SIRIUS", "sirius"),
+            self.define_from_variant("CP2K_USE_SPLA", "spla"),
+            self.define_from_variant("CP2K_USE_COSMA", "cosma"),
+            self.define_from_variant("CP2K_USE_LIBXC", "libxc"),
+            self.define_from_variant("CP2K_USE_LIBTORCH", "pytorch"),
+            self.define_from_variant("CP2K_USE_METIS", "pexsi"),
+            self.define_from_variant("CP2K_USE_SUPERLU", "pexsi"),
+            self.define_from_variant("CP2K_USE_PLUMED", "plumed"),
+            self.define_from_variant("CP2K_USE_SPGLIB", "spglib"),
+            self.define_from_variant("CP2K_USE_VORI", "libvori"),
+            self.define_from_variant("CP2K_USE_SPLA", "spla"),
+            self.define_from_variant("CP2K_USE_QUIP", "quip"),
+        ]
+
+        # we force the use elpa openmp threading support. might need to be revisited though
+        args += [
+            self.define(
+                "CP2K_ENABLE_ELPA_OPENMP_SUPPORT",
+                ("+elpa +openmp" in spec) or ("^elpa +openmp" in spec),
+            )
+        ]
+
+        if "spla" in spec and (spec.satisfies("+cuda") or spec.satisfies("+rocm")):
+            args += ["-DCP2K_USE_SPLA_GEMM_OFFLOADING=ON"]
+
+        args += ["-DCP2K_USE_FFTW3=ON"]
+
+        if spec.satisfies("smm=libxsmm"):
+            args += ["-DCP2K_USE_LIBXSMM=ON"]
+        else:
+            args += ["-DCP2K_USE_LIBXSMM=OFF"]
+
+        lapack = spec["lapack"]
+        blas = spec["blas"]
+
+        if blas.name in ["intel-mkl", "intel-parallel-studio", "intel-oneapi-mkl"]:
+            args += ["-DCP2K_BLAS_VENDOR=MKL"]
+            if sys.platform == "darwin":
+                args += [
+                    self.define("CP2K_BLAS_VENDOR", "CUSTOM"),
+                    self.define("CP2K_SCALAPACK_VENDOR", "GENERIC"),
+                    self.define(
+                        "CP2K_SCALAPACK_LINK_LIBRARIES", spec["scalapack"].libs.joined(";")
+                    ),
+                ]
+            else:
+                args += ["-DCP2K_SCALAPACK_VENDOR=MKL"]
+        else:
+            args.extend(
+                [
+                    self.define("CP2K_LAPACK_FOUND", True),
+                    self.define("CP2K_LAPACK_LINK_LIBRARIES", lapack.libs.joined(";")),
+                    self.define("CP2K_BLAS_FOUND", True),
+                    self.define("CP2K_BLAS_LINK_LIBRARIES", blas.libs.joined(";")),
+                    self.define("CP2K_SCALAPACK_FOUND", True),
+                    self.define("CP2K_SCALAPACK_INCLUDE_DIRS", spec["scalapack"].prefix.include),
+                    self.define("CP2K_BLAS_VENDOR", "CUSTOM"),
+                    self.define("CP2K_SCALAPACK_VENDOR", "GENERIC"),
+                    self.define(
+                        "CP2K_SCALAPACK_LINK_LIBRARIES", spec["scalapack"].libs.joined(";")
+                    ),
+                ]
+            )
+
+        return args
+
+    pass