summaryrefslogtreecommitdiff
blob: 980686f9e8422e74a10c316d14ac48336edafbf5 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
# Copyright 1999-2021 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2

EAPI=7

CMAKE_ECLASS=cmake
PYTHON_COMPAT=( python3_{8..10} )
inherit cmake-multilib linux-info llvm llvm.org python-any-r1

DESCRIPTION="OpenMP runtime library for LLVM/clang compiler"
HOMEPAGE="https://openmp.llvm.org"

LICENSE="Apache-2.0-with-LLVM-exceptions || ( UoI-NCSA MIT )"
SLOT="0"
KEYWORDS="amd64 arm arm64 ~ppc ppc64 ~riscv ~x86 ~amd64-linux ~x64-macos"
IUSE="cuda debug hwloc kernel_linux offload ompt test
	llvm_targets_AMDGPU llvm_targets_NVPTX"
# CUDA works only with the x86_64 ABI
REQUIRED_USE="
	cuda? ( llvm_targets_NVPTX )
	offload? ( cuda? ( abi_x86_64 ) )"
RESTRICT="!test? ( test )"

RDEPEND="
	hwloc? ( sys-apps/hwloc:0=[${MULTILIB_USEDEP}] )
	offload? (
		virtual/libelf:=[${MULTILIB_USEDEP}]
		dev-libs/libffi:=[${MULTILIB_USEDEP}]
		~sys-devel/llvm-${PV}[${MULTILIB_USEDEP}]
		cuda? ( dev-util/nvidia-cuda-toolkit:= )
	)"
# tests:
# - dev-python/lit provides the test runner
# - sys-devel/llvm provide test utils (e.g. FileCheck)
# - sys-devel/clang provides the compiler to run tests
DEPEND="${RDEPEND}"
BDEPEND="dev-lang/perl
	offload? (
		llvm_targets_AMDGPU? ( sys-devel/clang )
		llvm_targets_NVPTX? ( sys-devel/clang )
		virtual/pkgconfig
	)
	test? (
		$(python_gen_any_dep 'dev-python/lit[${PYTHON_USEDEP}]')
		sys-devel/clang
	)"

LLVM_COMPONENTS=( openmp llvm/include )
LLVM_PATCHSET=${PV/_/-}-r1
llvm.org_set_globals

python_check_deps() {
	has_version "dev-python/lit[${PYTHON_USEDEP}]"
}

kernel_pds_check() {
	if use kernel_linux && kernel_is -lt 4 15 && kernel_is -ge 4 13; then
		local CONFIG_CHECK="~!SCHED_PDS"
		local ERROR_SCHED_PDS="\
PDS scheduler versions >= 0.98c < 0.98i (e.g. used in kernels >= 4.13-pf11
< 4.14-pf9) do not implement sched_yield() call which may result in horrible
performance problems with libomp. If you are using one of the specified
kernel versions, you may want to disable the PDS scheduler."

		check_extra_config
	fi
}

pkg_pretend() {
	kernel_pds_check
}

pkg_setup() {
	use offload && LLVM_MAX_SLOT=${PV%%.*} llvm_pkg_setup
	use test && python-any-r1_pkg_setup
}

multilib_src_configure() {
	# LLVM_ENABLE_ASSERTIONS=NO does not guarantee this for us, #614844
	use debug || local -x CPPFLAGS="${CPPFLAGS} -DNDEBUG"

	local libdir="$(get_libdir)"
	local mycmakeargs=(
		-DOPENMP_LIBDIR_SUFFIX="${libdir#lib}"

		-DLIBOMP_USE_HWLOC=$(usex hwloc)
		-DLIBOMP_OMPT_SUPPORT=$(usex ompt)

		-DOPENMP_ENABLE_LIBOMPTARGET=$(usex offload)

		# do not install libgomp.so & libiomp5.so aliases
		-DLIBOMP_INSTALL_ALIASES=OFF
		# disable unnecessary hack copying stuff back to srcdir
		-DLIBOMP_COPY_EXPORTS=OFF
	)
	use offload && mycmakeargs+=(
		# this is non-fatal and libomp checks for CUDA conditionally
		# to ABI, so we can just ignore passing the wrong value
		# on non-amd64 ABIs
		-DCMAKE_DISABLE_FIND_PACKAGE_CUDA=$(usex !cuda)

		-DLIBOMPTARGET_BUILD_AMDGCN_BCLIB=$(usex llvm_targets_AMDGPU)
		-DLIBOMPTARGET_BUILD_NVPTX_BCLIB=$(usex llvm_targets_NVPTX)
		# a cheap hack to force clang
		-DLIBOMPTARGET_NVPTX_CUDA_COMPILER="$(type -P "${CHOST}-clang")"
	)
	use test && mycmakeargs+=(
		# this project does not use standard LLVM cmake macros
		-DOPENMP_LLVM_LIT_EXECUTABLE="${EPREFIX}/usr/bin/lit"
		-DOPENMP_LIT_ARGS="$(get_lit_flags)"

		-DOPENMP_TEST_C_COMPILER="$(type -P "${CHOST}-clang")"
		-DOPENMP_TEST_CXX_COMPILER="$(type -P "${CHOST}-clang++")"
	)
	addpredict /dev/nvidiactl
	cmake_src_configure
}

multilib_src_test() {
	# respect TMPDIR!
	local -x LIT_PRESERVES_TMP=1

	cmake_build check-libomp
}