# Copyright 2013-2023 Lawrence Livermore National Security, LLC and other # Spack Project Developers. See the top-level COPYRIGHT file for details. # # SPDX-License-Identifier: (Apache-2.0 OR MIT) import os import socket import llnl.util.tty as tty from spack.package import * from spack.pkg.builtin.camp import hip_repair_cache class Umpire(CachedCMakePackage, CudaPackage, ROCmPackage): """An application-focused API for memory management on NUMA & GPU architectures""" homepage = "https://github.com/LLNL/Umpire" git = "https://github.com/LLNL/Umpire.git" tags = ["radiuss", "e4s"] maintainers("davidbeckingsale") version("develop", branch="develop", submodules=False) version("main", branch="main", submodules=False) version("2022.03.1", tag="v2022.03.1", submodules=False) version("2022.03.0", tag="v2022.03.0", submodules=False) version("6.0.0", tag="v6.0.0", submodules=True) version("5.0.1", tag="v5.0.1", submodules=True) version("5.0.0", tag="v5.0.0", submodules=True) version("4.1.2", tag="v4.1.2", submodules=True) version("4.1.1", tag="v4.1.1", submodules=True) version("4.1.0", tag="v4.1.0", submodules=True) version("4.0.1", tag="v4.0.1", submodules=True) version("4.0.0", tag="v4.0.0", submodules=True) version("3.0.0", tag="v3.0.0", submodules=True) version("2.1.0", tag="v2.1.0", submodules=True) version("2.0.0", tag="v2.0.0", submodules=True) version("1.1.0", tag="v1.1.0", submodules=True) version("1.0.1", tag="v1.0.1", submodules=True) version("1.0.0", tag="v1.0.0", submodules=True) version("0.3.5", tag="v0.3.5", submodules=True) version("0.3.4", tag="v0.3.4", submodules=True) version("0.3.3", tag="v0.3.3", submodules=True) version("0.3.2", tag="v0.3.2", submodules=True) version("0.3.1", tag="v0.3.1", submodules=True) version("0.3.0", tag="v0.3.0", submodules=True) version("0.2.4", tag="v0.2.4", submodules=True) version("0.2.3", tag="v0.2.3", submodules=True) version("0.2.2", tag="v0.2.2", submodules=True) version("0.2.1", tag="v0.2.1", submodules=True) version("0.2.0", tag="v0.2.0", submodules=True) version("0.1.4", tag="v0.1.4", submodules=True) version("0.1.3", tag="v0.1.3", submodules=True) patch("std-filesystem-pr784.patch", when="@2022.03.1 +rocm ^blt@0.5.2:") patch("camp_target_umpire_3.0.0.patch", when="@3.0.0") patch("cmake_version_check.patch", when="@4.1") patch("missing_header_for_numeric_limits.patch", when="@4.1:5.0.1") # export targets when building pre-6.0.0 release with BLT 0.4.0+ patch( "https://github.com/LLNL/Umpire/commit/5773ce9af88952c8d23f9bcdcb2e503ceda40763.patch?full_index=1", sha256="f3b21335ce5cf9c0fecc852a94dfec90fb5703032ac97f9fee104af9408d8899", when="@:5.0.1 ^blt@0.4:", ) variant("fortran", default=False, description="Build C/Fortran API") variant("c", default=True, description="Build C API") variant("numa", default=False, description="Enable NUMA support") variant("shared", default=True, description="Enable Shared libs") variant("openmp", default=False, description="Build with OpenMP support") variant("deviceconst", default=False, description="Enables support for constant device memory") variant("examples", default=True, description="Build Umpire Examples") variant( "tests", default="none", values=("none", "basic", "benchmarks"), multi=False, description="Tests to run", ) variant("device_alloc", default=True, description="Build Umpire Device Allocator") depends_on("cmake@3.8:", type="build") depends_on("cmake@3.9:", when="+cuda", type="build") depends_on("cmake@3.14:", when="@2022.03.0:") depends_on("blt@0.5.0:", type="build", when="@2022.03.0:") depends_on("blt@0.4.1", type="build", when="@6.0.0") depends_on("blt@0.4.0:", type="build", when="@4.1.3:5.0.1") depends_on("blt@0.3.6:", type="build", when="@:4.1.2") depends_on("camp", when="@5.0.0:") depends_on("camp@0.2.2:0.2.3", when="@6.0.0") depends_on("camp@0.1.0", when="@5.0.0:5.0.1") depends_on("camp@2022.03.2:", when="@2022.03.0:") depends_on("camp@main", when="@main") depends_on("camp@main", when="@develop") depends_on("camp+openmp", when="+openmp") depends_on("camp~cuda", when="~cuda") with when("@5.0.0:"): with when("+cuda"): depends_on("camp+cuda") for sm_ in CudaPackage.cuda_arch_values: depends_on("camp+cuda cuda_arch={0}".format(sm_), when="cuda_arch={0}".format(sm_)) with when("+rocm"): depends_on("camp+rocm") for arch_ in ROCmPackage.amdgpu_targets: depends_on( "camp+rocm amdgpu_target={0}".format(arch_), when="amdgpu_target={0}".format(arch_), ) conflicts("+numa", when="@:0.3.2") conflicts("~c", when="+fortran", msg="Fortran API requires C API") conflicts("+device_alloc", when="@:2022.03.0") # device allocator exports device code, which requires static libs # currently only available for cuda. conflicts("+shared", when="+cuda") # https://github.com/LLNL/Umpire/issues/653 # This range looks weird, but it ensures the concretizer looks at it as a # range, not as a concrete version, so that it also matches 10.3.* versions. conflicts("%gcc@10.3.0:10.3", when="+cuda") def _get_sys_type(self, spec): sys_type = spec.architecture if "SYS_TYPE" in env: sys_type = env["SYS_TYPE"] return sys_type @property def cache_name(self): hostname = socket.gethostname() if "SYS_TYPE" in env: hostname = hostname.rstrip("1234567890") return "{0}-{1}-{2}@{3}.cmake".format( hostname, self._get_sys_type(self.spec), self.spec.compiler.name, self.spec.compiler.version, ) def initconfig_compiler_entries(self): spec = self.spec entries = super().initconfig_compiler_entries() if "+rocm" in spec: entries.insert(0, cmake_cache_path("CMAKE_CXX_COMPILER", spec["hip"].hipcc)) option_prefix = "UMPIRE_" if spec.satisfies("@2022.03.0:") else "" if "+fortran" in spec and self.compiler.fc is not None: entries.append(cmake_cache_option("ENABLE_FORTRAN", True)) else: entries.append(cmake_cache_option("ENABLE_FORTRAN", False)) entries.append(cmake_cache_option("{}ENABLE_C".format(option_prefix), "+c" in spec)) return entries def initconfig_hardware_entries(self): spec = self.spec entries = super().initconfig_hardware_entries() option_prefix = "UMPIRE_" if spec.satisfies("@2022.03.0:") else "" if "+cuda" in spec: entries.append(cmake_cache_option("ENABLE_CUDA", True)) if not spec.satisfies("cuda_arch=none"): cuda_arch = spec.variants["cuda_arch"].value entries.append(cmake_cache_string("CUDA_ARCH", "sm_{0}".format(cuda_arch[0]))) entries.append( cmake_cache_string("CMAKE_CUDA_ARCHITECTURES", "{0}".format(cuda_arch[0])) ) flag = "-arch sm_{0}".format(cuda_arch[0]) entries.append(cmake_cache_string("CMAKE_CUDA_FLAGS", "{0}".format(flag))) entries.append( cmake_cache_option( "{}ENABLE_DEVICE_CONST".format(option_prefix), spec.satisfies("+deviceconst") ) ) else: entries.append(cmake_cache_option("ENABLE_CUDA", False)) if "+rocm" in spec: entries.append(cmake_cache_option("ENABLE_HIP", True)) entries.append(cmake_cache_path("HIP_ROOT_DIR", "{0}".format(spec["hip"].prefix))) hip_repair_cache(entries, spec) archs = self.spec.variants["amdgpu_target"].value if archs != "none": arch_str = ",".join(archs) entries.append( cmake_cache_string("HIP_HIPCC_FLAGS", "--amdgpu-target={0}".format(arch_str)) ) entries.append( cmake_cache_string("CMAKE_HIP_ARCHITECTURES", "{0}".format(arch_str)) ) else: entries.append(cmake_cache_option("ENABLE_HIP", False)) return entries def initconfig_package_entries(self): spec = self.spec entries = [] option_prefix = "UMPIRE_" if spec.satisfies("@2022.03.0:") else "" # TPL locations entries.append("#------------------{0}".format("-" * 60)) entries.append("# TPLs") entries.append("#------------------{0}\n".format("-" * 60)) entries.append(cmake_cache_path("BLT_SOURCE_DIR", spec["blt"].prefix)) if spec.satisfies("@5.0.0:"): entries.append(cmake_cache_path("camp_DIR", spec["camp"].prefix)) entries.append(cmake_cache_option("{}ENABLE_NUMA".format(option_prefix), "+numa" in spec)) entries.append( cmake_cache_option("{}ENABLE_OPENMP".format(option_prefix), "+openmp" in spec) ) entries.append(cmake_cache_option("ENABLE_BENCHMARKS", "tests=benchmarks" in spec)) entries.append( cmake_cache_option("{}ENABLE_EXAMPLES".format(option_prefix), "+examples" in spec) ) entries.append(cmake_cache_option("{}ENABLE_DOCS".format(option_prefix), False)) entries.append( cmake_cache_option("UMPIRE_ENABLE_DEVICE_ALLOCATOR", "+device_alloc" in spec) ) entries.append(cmake_cache_option("BUILD_SHARED_LIBS", "+shared" in spec)) entries.append(cmake_cache_option("ENABLE_TESTS", "tests=none" not in spec)) return entries def cmake_args(self): options = [] return options def test(self): """Perform stand-alone checks on the installed package.""" if self.spec.satisfies("@:1") or not os.path.isdir(self.prefix.bin): tty.info("Skipping: checks not installed in bin for v{0}".format(self.version)) return # Run a subset of examples PROVIDED installed # tutorials with readily checkable outputs. checks = { "malloc": ["99 should be 99"], "recipe_dynamic_pool_heuristic": ["in the pool", "releas"], "recipe_no_introspection": ["has allocated", "used"], "strategy_example": ["Available allocators", "HOST"], "tut_copy": ["Copied source data"], "tut_introspection": ["Allocator used is HOST", "size of the allocation"], "tut_memset": ["Set data from HOST"], "tut_move": ["Moved source data", "HOST"], "tut_reallocate": ["Reallocated data"], "vector_allocator": [""], } for exe in checks: expected = checks[exe] reason = "test: checking output from {0}".format(exe) self.run_test( exe, [], expected, 0, installed=False, purpose=reason, skip_missing=True, work_dir=self.prefix.bin, )