Skip to content

Instantly share code, notes, and snippets.

@scivision
Last active September 24, 2024 15:16
Show Gist options
  • Save scivision/33bd9e17c9520d07be0448fe61541605 to your computer and use it in GitHub Desktop.
Save scivision/33bd9e17c9520d07be0448fe61541605 to your computer and use it in GitHub Desktop.
build script for LLVM & Flang-f18 Fortran compiler

Build script for Flang-f18 (flang-new) + LLVM

This is a Bash script (macOS, Linux, ...) for building Flang-f18 and LLVM from source. It is adapted from Jeff Hammond

Ninja: recommended for best build efficiency and speed.

In general, a recent GCC would work. On macOS, system AppleClang compiler can be used as well.

bash build-flang-f18.sh

To specify the source URL, for example to build the latest LLVM 19.x release:

bash build-flang-f18.sh https://github.com/llvm/llvm-project/archive/refs/heads/release/19.x.zip

For a specific version of LLVM:

bash build-flang-f18.sh https://github.com/llvm/llvm-project/archive/refs/tags/llvmorg-19.1.0.zip

The source download is about 300 MB. LLVM takes about a half-hour to build on a powerful laptop or workstation.

To use the compiler after install:

source clang-latest.sh

LLVM projects

Troubleshooting

See ulimit settings if link failures occur.

Disk usage

The installed binaries take a few gigabytes of disk space.

du -sh ~/llvm-latest  # the install prefix

3.5G

About 3.5 Gigabytes are used for the installed LLVM binaries.

The source directory can be deleted after the build / install is complete.

du -sh $TMPDIR/llvm-src

1.8G

The build directory can be deleted after the build / install is complete.

du -sh $TMPDIR/llvm-build

5.0G

#!/bin/bash -xe
# -x: verbose
# -e: exit on error
# Ninja is recommended for best build efficiency and speed
# always use the ".zip" source file
# adapted from https://github.com/jeffhammond/HPCInfo/blob/master/buildscripts/llvm-git.sh
remote="${1:-https://github.com/llvm/llvm-project/archive/refs/heads/main.zip}"
[[ -z $remote ]] && { echo "Usage: $0 [archive_url]" && exit 1; }
# use TMPDIR if not empty, else /tmp
TMPDIR=${TMPDIR:-/tmp}
llvm_src=${TMPDIR}/llvm-src
llvm_build=${TMPDIR}/llvm-build
prefix=$HOME/llvm-latest
stem=$(basename ${remote} .zip)
cmake_root=${llvm_src}/llvm-project-${stem}/llvm
llvm_projects="lld;mlir;clang;flang;openmp;pstl"
echo "LLVM projects: $llvm_projects"
echo "LLVM source: $llvm_src"
echo "LLVM build: $llvm_build"
echo "LLVM install: $prefix"
mkdir -p $prefix
mkdir -p $llvm_src
mkdir -p $llvm_build
# Git not used as it's so slow for a huge project history like LLVM.
# git clone --recursive https://github.com/llvm/llvm-project.git $llvm_src
# ~300 MB
archive=${TMPDIR}/llvm_main.zip
# Download/update the source
[[ -f $archive ]] || curl --location --output ${archive} ${remote}
[[ -f ${cmake_root}/CMakeLists.txt ]] || unzip -d $llvm_src $archive
[[ $(which ninja) ]] && CMAKE_GENERATOR="Ninja" || CMAKE_GENERATOR="Unix Makefiles"
case "$(uname -m)" in
arm64|aarch64)
llvm_arch=AArch64
;;
*)
llvm_arch=X86
;;
esac
# helpful system parameters
case "$OSTYPE" in
darwin*)
macos_sysroot=-DDEFAULT_SYSROOT="$(xcrun --show-sdk-path)"
;;
*)
llvm_linker=-DLLVM_USE_LINKER=gold
;;
esac\
# lldb busted on MacOS
# libcxx requires libcxxabi
cmake \
-G"$CMAKE_GENERATOR" \
-DCMAKE_BUILD_TYPE=Release \
-DLLVM_TARGETS_TO_BUILD=$llvm_arch \
-DLLVM_ENABLE_RUNTIMES="libcxxabi;libcxx;libunwind" \
-DLLVM_ENABLE_PROJECTS=${llvm_projects} \
$macos_sysroot \
$llvm_linker \
--install-prefix=$prefix \
-S${cmake_root} \
-B${llvm_build}
cmake --build ${llvm_build}
cmake --install ${llvm_build}
r=$HOME/llvm-latest
b=${r}/bin
export CC=${b}/clang CXX=${b}/clang++ FC=${b}/flang-new
# flags below not generally needed
#export LDFLAGS="-L${r}/lib"
#export CPPFLAGS="-I${r}/include"
#case "$OSTYPE" in
#darwin*)
# avoid missing -lSystem
#export LIBRARY_PATH="/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/usr/lib"
# avoid missing libc++.1.dylib
#export DYLD_FALLBACK_LIBRARY_PATH=${r}/lib
#;;
#esac
$FC --version
@scivision
Copy link
Author

Interesting. A linux system I used also had 1024 and macOS has 256. I will note it in the Readme.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment