nlcglib: add v1.1.0 (#44580)

var/spack/repos/builtin/packages/nlcglib/package.py

@@ -18,8 +18,8 @@ class Nlcglib(CMakePackage, CudaPackage, ROCmPackage):
     license("BSD-3-Clause")
 
     version("develop", branch="develop")
-    version("master", branch="master")
 
+    version("1.1.0", sha256="9e7c2eea84a5ce191bd9af08f6c890717f7b6e88be7bd15cfe774eb0e0dabd8a")
     version("1.0b", sha256="086c46f06a117f267cbdf1df4ad42a8512689a9610885763f463469fb15e82dc")
     version("0.9", sha256="8d5bc6b85ee714fb3d6480f767e7f43e5e7d569116cf60e48f533a7f50a37a08")
 
@@ -32,13 +32,26 @@ class Nlcglib(CMakePackage, CudaPackage, ROCmPackage):
         values=("Debug", "Release", "RelWithDebInfo"),
     )
 
+    with when("@1.1: +cuda"):
+        variant(
+            "gpu_direct",
+            default=False,
+            description="Enable GPU direct. Required to support distributed wave-functions.",
+        )
+
     depends_on("cmake@3.21:", type="build")
     depends_on("mpi")
     depends_on("lapack")
+
     depends_on("kokkos~cuda~rocm", when="~cuda~rocm")
     depends_on("kokkos+openmp", when="+openmp")
+
     depends_on("googletest", type="build", when="+tests")
     depends_on("nlohmann-json")
+    depends_on("kokkos@4:", when="@1.1:")
+
+    # MKLConfig.cmake introduced in 2021.3
+    conflicts("intel-oneapi-mkl@:2021.2", when="^intel-oneapi-mkl")
 
     with when("@:0.9"):
         conflicts("+rocm")
@@ -51,18 +64,18 @@ class Nlcglib(CMakePackage, CudaPackage, ROCmPackage):
         depends_on("rocblas")
         depends_on("rocsolver")
 
-    for arch in CudaPackage.cuda_arch_values:
+    with when("+cuda"):
-        depends_on(
+        depends_on("kokkos+cuda_lambda+wrapper", when="%gcc")
-            f"kokkos+cuda+cuda_lambda+wrapper cuda_arch={arch}",
+        depends_on("kokkos+cuda")
-            when=f"%gcc +cuda cuda_arch={arch}",
+        for arch in CudaPackage.cuda_arch_values:
-        )
+            depends_on(f"kokkos cuda_arch={arch}", when=f"cuda_arch={arch}")
-        depends_on(f"kokkos+cuda cuda_arch={arch}", when=f"+cuda cuda_arch={arch}")
 
     def cmake_args(self):
         options = [
             self.define_from_variant("USE_OPENMP", "openmp"),
             self.define_from_variant("BUILD_TESTS", "tests"),
             self.define_from_variant("USE_ROCM", "rocm"),
+            self.define_from_variant("USE_GPU_DIRECT", "gpu_direct"),
             self.define_from_variant("USE_MAGMA", "magma"),
             self.define_from_variant("USE_CUDA", "cuda"),
         ]
@@ -71,6 +84,29 @@ def cmake_args(self):
             options += [self.define("LAPACK_VENDOR", "MKL")]
         elif self.spec["blas"].name in ["intel-oneapi-mkl"]:
             options += [self.define("LAPACK_VENDOR", "MKLONEAPI")]
+            mkl_mapper = {
+                "threading": {"none": "sequential", "openmp": "gnu_thread", "tbb": "tbb_thread"},
+                "mpi": {"intel-mpi": "intelmpi", "mpich": "mpich", "openmpi": "openmpi"},
+            }
+
+            mkl_threads = mkl_mapper["threading"][
+                self.spec["intel-oneapi-mkl"].variants["threads"].value
+            ]
+
+            mpi_provider = self.spec["mpi"].name
+            if mpi_provider in ["mpich", "cray-mpich", "mvapich", "mvapich2"]:
+                mkl_mpi = mkl_mapper["mpi"]["mpich"]
+            else:
+                mkl_mpi = mkl_mapper["mpi"][mpi_provider]
+
+            options.extend(
+                [
+                    self.define("MKL_INTERFACE", "lp64"),
+                    self.define("MKL_THREADING", mkl_threads),
+                    self.define("MKL_MPI", mkl_mpi),
+                ]
+            )
+
         elif self.spec["blas"].name in ["openblas"]:
             options += [self.define("LAPACK_VENDOR", "OpenBLAS")]
         else: