@@ -4,6 +4,7 @@
|
||||
# SPDX-License-Identifier: (Apache-2.0 OR MIT)
|
||||
import collections.abc
|
||||
import os
|
||||
import re
|
||||
from typing import Tuple
|
||||
|
||||
import llnl.util.filesystem as fs
|
||||
@@ -15,6 +16,12 @@
|
||||
from .cmake import CMakeBuilder, CMakePackage
|
||||
|
||||
|
||||
def spec_uses_toolchain(spec):
|
||||
gcc_toolchain_regex = re.compile(".*gcc-toolchain.*")
|
||||
using_toolchain = list(filter(gcc_toolchain_regex.match, spec.compiler_flags["cxxflags"]))
|
||||
return using_toolchain
|
||||
|
||||
|
||||
def cmake_cache_path(name, value, comment="", force=False):
|
||||
"""Generate a string for a cmake cache variable"""
|
||||
force_str = " FORCE" if force else ""
|
||||
@@ -132,6 +139,11 @@ def initconfig_compiler_entries(self):
|
||||
"endif()\n",
|
||||
]
|
||||
|
||||
# We defined hipcc as top-level compiler for packages when +rocm.
|
||||
# This avoid problems coming from rocm flags being applied to another compiler.
|
||||
if "+rocm" in spec:
|
||||
entries.insert(0, cmake_cache_path("CMAKE_CXX_COMPILER", self.spec["hip"].hipcc))
|
||||
|
||||
flags = spec.compiler_flags
|
||||
|
||||
# use global spack compiler flags
|
||||
@@ -213,7 +225,7 @@ def initconfig_mpi_entries(self):
|
||||
else:
|
||||
# starting with cmake 3.10, FindMPI expects MPIEXEC_EXECUTABLE
|
||||
# vs the older versions which expect MPIEXEC
|
||||
if self.pkg.spec["cmake"].satisfies("@3.10:"):
|
||||
if spec["cmake"].satisfies("@3.10:"):
|
||||
entries.append(cmake_cache_path("MPIEXEC_EXECUTABLE", mpiexec))
|
||||
else:
|
||||
entries.append(cmake_cache_path("MPIEXEC", mpiexec))
|
||||
@@ -248,12 +260,17 @@ def initconfig_hardware_entries(self):
|
||||
# Include the deprecated CUDA_TOOLKIT_ROOT_DIR for supporting BLT packages
|
||||
entries.append(cmake_cache_path("CUDA_TOOLKIT_ROOT_DIR", cudatoolkitdir))
|
||||
|
||||
archs = spec.variants["cuda_arch"].value
|
||||
if archs[0] != "none":
|
||||
arch_str = ";".join(archs)
|
||||
entries.append(
|
||||
cmake_cache_string("CMAKE_CUDA_ARCHITECTURES", "{0}".format(arch_str))
|
||||
)
|
||||
# CUDA_FLAGS
|
||||
cuda_flags = []
|
||||
|
||||
if not spec.satisfies("cuda_arch=none"):
|
||||
cuda_archs = ";".join(spec.variants["cuda_arch"].value)
|
||||
entries.append(cmake_cache_string("CMAKE_CUDA_ARCHITECTURES", cuda_archs))
|
||||
|
||||
if spec_uses_toolchain(spec):
|
||||
cuda_flags.append("-Xcompiler {}".format(spec_uses_toolchain(spec)[0]))
|
||||
|
||||
entries.append(cmake_cache_string("CMAKE_CUDA_FLAGS", " ".join(cuda_flags)))
|
||||
|
||||
if "+rocm" in spec:
|
||||
entries.append("#------------------{0}".format("-" * 30))
|
||||
@@ -262,9 +279,6 @@ def initconfig_hardware_entries(self):
|
||||
|
||||
# Explicitly setting HIP_ROOT_DIR may be a patch that is no longer necessary
|
||||
entries.append(cmake_cache_path("HIP_ROOT_DIR", "{0}".format(spec["hip"].prefix)))
|
||||
entries.append(
|
||||
cmake_cache_path("HIP_CXX_COMPILER", "{0}".format(self.spec["hip"].hipcc))
|
||||
)
|
||||
llvm_bin = spec["llvm-amdgpu"].prefix.bin
|
||||
llvm_prefix = spec["llvm-amdgpu"].prefix
|
||||
# Some ROCm systems seem to point to /<path>/rocm-<ver>/ and
|
||||
@@ -277,11 +291,9 @@ def initconfig_hardware_entries(self):
|
||||
archs = self.spec.variants["amdgpu_target"].value
|
||||
if archs[0] != "none":
|
||||
arch_str = ";".join(archs)
|
||||
entries.append(
|
||||
cmake_cache_string("CMAKE_HIP_ARCHITECTURES", "{0}".format(arch_str))
|
||||
)
|
||||
entries.append(cmake_cache_string("AMDGPU_TARGETS", "{0}".format(arch_str)))
|
||||
entries.append(cmake_cache_string("GPU_TARGETS", "{0}".format(arch_str)))
|
||||
entries.append(cmake_cache_string("CMAKE_HIP_ARCHITECTURES", arch_str))
|
||||
entries.append(cmake_cache_string("AMDGPU_TARGETS", arch_str))
|
||||
entries.append(cmake_cache_string("GPU_TARGETS", arch_str))
|
||||
|
||||
return entries
|
||||
|
||||
|
Reference in New Issue
Block a user