Beyond Linux® From Scratch (systemd Edition) - Version r12.2-893

Chapter 13. Programming

LLVM-19.1.0

Introduction to LLVM

The LLVM package contains a collection of modular and reusable compiler and toolchain technologies. The Low Level Virtual Machine (LLVM) Core libraries provide a modern source and target-independent optimizer, along with code generation support for many popular CPUs (as well as some less common ones!). These libraries are built around a well specified code representation known as the LLVM intermediate representation ("LLVM IR").

Clang provides new C, C++, Objective C and Objective C++ front-ends for LLVM and is required by some desktop packages such as firefox and for rust if that is built using the system LLVM.

The Compiler RT package provides runtime sanitizer and profiling libraries for developers who use Clang and LLVM.

[Note]

Note

Development versions of BLFS may not build or run some packages properly if LFS or dependencies have been updated since the most recent stable versions of the books.

Package Information

Additional Downloads

Cmake modules for LLVM

Third-party dependencies for LLVM build system

Recommended Download

Clang

Optional Download

Compiler RT

LLVM Dependencies

Required

CMake-3.31.3

Optional

Doxygen-1.12.0, git-2.47.1, Graphviz-12.2.1, libxml2-2.13.5, psutil-6.1.1 (for tests), Pygments-2.18.0, PyYAML-6.0.2 (for tests), rsync-3.3.0 (for tests), Systemd-257 (rebuilt with PAM, for tests), texlive-20240312 (or install-tl-unx), Valgrind-3.24.0, Zip-3.0, myst-parser (for building documentation), OCaml, and Z3

Installation of LLVM

Two additional tarballs llvm-cmake-19.src.tar.xz and llvm-third-party-19.src.tar.xz are needed by LLVM building system. The upstream expects them extracted at the same level as the llvm-19.1.0.src.tar.xz tarball, and the extracted directories renamed to cmake and third-party. Extract them and modify the build system to avoid creating ambiguously-named directories outside the llvm-19.1.0.src hierarchy: 

tar -xf ../llvm-cmake-19.src.tar.xz                                   &&
tar -xf ../llvm-third-party-19.src.tar.xz                             &&
sed '/LLVM_COMMON_CMAKE_UTILS/s@../cmake@llvm-cmake-19.src@'          \
    -i CMakeLists.txt                                                 &&
sed '/LLVM_THIRD_PARTY_DIR/s@../third-party@llvm-third-party-19.src@' \
    -i cmake/modules/HandleLLVMOptions.cmake

Install clang into the source tree by running the following commands: 

tar -xf ../clang-19.1.0.src.tar.xz -C tools &&
mv tools/clang-19.1.0.src tools/clang

If you have downloaded compiler-rt, install it into the source tree by running the following commands: 

tar -xf ../compiler-rt-19.1.0.src.tar.xz -C projects    &&
mv projects/compiler-rt-19.1.0.src projects/compiler-rt

There are many Python scripts in this package which use /usr/bin/env python to access the system Python which on LFS is Python-3.13.1. Use the following command to fix these scripts: 

grep -rl '#!.*python' | xargs sed -i '1s/python$/python3/'

Ensure installing the FileCheck program which is needed by the test suite of some packages (for example rustc-1.83.0): 

sed 's/utility/tool/' -i utils/FileCheck/CMakeLists.txt

Install LLVM by running the following commands: 

mkdir -v build &&
cd       build &&

CC=gcc CXX=g++                               \
cmake -D CMAKE_INSTALL_PREFIX=/usr           \
      -D CMAKE_SKIP_INSTALL_RPATH=ON         \
      -D LLVM_ENABLE_FFI=ON                  \
      -D CMAKE_BUILD_TYPE=Release            \
      -D LLVM_BUILD_LLVM_DYLIB=ON            \
      -D LLVM_LINK_LLVM_DYLIB=ON             \
      -D LLVM_ENABLE_RTTI=ON                 \
      -D LLVM_TARGETS_TO_BUILD="host;AMDGPU" \
      -D LLVM_BINUTILS_INCDIR=/usr/include   \
      -D LLVM_INCLUDE_BENCHMARKS=OFF         \
      -D CLANG_DEFAULT_PIE_ON_LINUX=ON       \
      -D CLANG_CONFIG_FILE_SYSTEM_DIR=/etc/clang \
      -W no-dev -G Ninja ..                  &&
ninja

If you wish to run the tests, remove a test case known to hang indefinitely on some systems: 

rm -f ../projects/compiler-rt/test/tsan/getline_nohang.cpp

LLVM test suite can produce many core dump files. They will occupy a large amount of disk space, and the core dump process can significantly slow down the testing. To test the results with core dump disabled, ensure Systemd-257 and Shadow-4.16.0 have been rebuilt with Linux-PAM-1.7.0 support (if you are interacting via a SSH or graphical session, also ensure the OpenSSH-9.9p1 server or the desktop manager has been built with Linux-PAM-1.7.0) and the current login session is started after updating the /etc/pam.d/system-session file to include pam_systemd.so, then issue: 

systemctl   --user start dbus &&
systemd-run --user --pty -d -G -p LimitCORE=0 ninja check-all

If -jN (N replaced with a number) is passed to ninja, the tests will be built with N logical cores, but run using all available logical cores. Run the test command in a cgroup (pass the -p AllowedCPUs=... option to the systemd-run command, read the section called “Use Linux Control Group to Limit the Resource Usage” for details) to limit the number of logical cores for running the tests. Two tests related to TestCases/Linux/printf-fortify-5.c are known to fail with Glibc-2.40 or newer. One test named Linux/clone_setns.cpp will fail if CONFIG_USER_NS is not enabled in kernel configuration.

Now, as the root user: 

ninja install

Command Explanations

-D CMAKE_SKIP_INSTALL_RPATH=ON: This switch makes cmake remove hardcoded library search paths (rpath) when installing a binary executable file or a shared library. This package does not need rpath once it's installed into the standard location, and rpath may sometimes cause unwanted effects or even security issues.

-D LLVM_ENABLE_FFI=ON: This switch allows LLVM to use libffi.

-D LLVM_BUILD_LLVM_DYLIB=ON: This switch builds the libraries as static and links all of them into an unique shared one. This is the recommended way of building a shared library.

-D CMAKE_BUILD_TYPE=Release: This switch enables compiler optimizations in order to speed up the code and reduce its size. It also disables some compile checks which are not necessary on a production system.

-D LLVM_TARGETS_TO_BUILD="host;AMDGPU": This switch enables building for the same target as the host, and also for the r600 AMD GPU used by the Mesa r600 and radeonsi drivers. The default is all of the targets. You can use a semicolon separated list. Valid targets are: host, AArch64, AMDGPU, ARM, AVR, BPF, Hexagon, Lanai, LoongArch, Mips, MSP430, NVPTX, PowerPC, RISCV, Sparc, SystemZ, SystemZ, VE, WebAssembly, X86, XCore, or all.

-D LLVM_LINK_LLVM_DYLIB=ON: Used in conjunction with -D LLVM_BUILD_LLVM_DYLIB=ON, this switch enables linking the tools against the shared library instead of the static ones. It slightly reduces their size and also ensures that llvm-config will correctly use libLLVM-19.so.

-D LLVM_ENABLE_RTTI=ON: This switch is used to build LLVM with run-time type information. This is required for building Mesa-24.3.2.

-D LLVM_BINUTILS_INCDIR=/usr/include: This switch is used to tell the build system the location of binutils headers, which were installed in LFS. This allows the building of LLVMgold.so, which is needed for building programs with clang and Link Time Optimization (LTO).

-D LLVM_INCLUDE_BENCHMARKS=OFF: is used to disable generation build targets for the LLVM benchmarks. This option requires additional code that is not currently available.

-D CLANG_DEFAULT_PIE_ON_LINUX=ON: makes -fpie option the default when compiling programs. Together with the ASLR feature enabled in the kernel, this defeats some kind of attacks based on known memory layouts.

-D CLANG_CONFIG_FILE_SYSTEM_DIR=/etc/clang: makes clang and clang++ search /etc/clang for configuration files.

-D BUILD_SHARED_LIBS=ON: if used instead of -D LLVM_BUILD_LLVM_DYLIB=ON and -D LLVM_LINK_LLVM_DYLIB=ON, builds all the LLVM libraries (about 60) as shared libraries instead of static.

-D LLVM_ENABLE_DOXYGEN: Enables the generation of browsable HTML documentation if you have installed Doxygen-1.12.0. You should run make doxygen-html afterwards, and install the generated documentation manually.

Configuring LLVM

Configuration Information

If you've built Clang, as the root user create two configuration files to make SSP enabled by default for clang and clang++, so the default configuration of their SSP feature will be consistent with GCC-14.2.0: 

mkdir -pv /etc/clang &&
for i in clang clang++; do
  echo -fstack-protector-strong > /etc/clang/$i.cfg
done

Contents

Installed Programs: amdgpu-arch, analyze-build, bugpoint, c-index-test, clang, clang++ (symlinks to clang-19), clang-19, clang-check, clang-cl, clang-cpp (last two symlinks to clang), clang-extdef-mapping, clang-format, clang-linker-wrapper, clang-offload-bundler, clang-offload-packager, clang-refactor, clang-rename, clang-repl, clang-scan-deps, clang-tblgen, diagtool, dsymutil, FileCheck, git-clang-format, hmaptool, intercept-build, llc, lli, llvm-addr2line (symlink to llvm-symbolizer), llvm-ar, llvm-as, llvm-bcanalyzer, llvm-bitcode-strip (symlink to llvm-objcopy), llvm-cat, llvm-cfi-verify, llvm-config, llvm-cov, llvm-c-test, llvm-cvtres, llvm-cxxdump, llvm-cxxfilt, llvm-cxxmap, llvm-debuginfo-analyzer, llvm-debuginfod, llvm-debuginfod-find, llvm-diff, llvm-dis, llvm-dlltool (symlink to llvm-ar), llvm-dwarfdump, llvm-dwarfutil, llvm-dwp, llvm-exegesis, llvm-extract, llvm-gsymutil, llvm-ifs, llvm-install-name-tool (symlink to llvm-objcopy), llvm-jitlink, llvm-lib (symlink to llvm-ar), llvm-libtool-darwin, llvm-link, llvm-lipo, llvm-lto, llvm-lto2, llvm-mc, llvm-mca, llvm-ml, llvm-modextract, llvm-mt, llvm-nm, llvm-objcopy, llvm-objdump, llvm-opt-report, llvm-otool (symlink to llv-objdump), llvm-pdbutil, llvm-profdata, llvm-profgen, llvm-ranlib (symlink to llvm-ar), llvm-rc, llvm-readelf (symlink to llvm-readobj), llvm-readobj, llvm-readtapi, llvm-reduce, llvm-remarkutil, llvm-rtdyld, llvm-sim, llvm-size, llvm-split, llvm-stress, llvm-strings, llvm-strip (symlink to llvm-objcopy), llvm-symbolizer, llvm-tblgen, llvm-tli-checker, llvm-undname, llvm-windres (symlink to llvm-rc), llvm-xray, nvptx-arch, opt, sancov, sanstats, scan-build, scan-build-py, scan-view, and verify-uselistorder
Installed Libraries: libLLVM.so, libLLVM*.a (100 libraries), libLTO.so, libRemarks.so, libclang.so, libclang-cpp.so, libclang*.a (42 libraries), and LLVMgold.so
Installed Directories: /usr/include/{clang,clang-c,llvm,llvm-c}, /usr/lib/{clang,cmake/{clang,llvm},libear,libscanbuild}, /usr/share/{clang,opt-viewer,scan-build,scan-view}, and /etc/clang

Short Descriptions

amdgpu-arch

lists AMD GPUs installed; at runtime it needs libhsa-runtime64.so which is not a part of BLFS

analyze-build

is a static analysis tool

bugpoint

is the automatic test case reduction tool

c-index-test

is used to test the libclang API and demonstrate its usage

clang

is the Clang C, C++, and Objective-C compiler

clang-check

is a tool to perform static code analysis and display Abstract Syntax Trees (AST)

clang-extdef-mapping

is a tool to collect the USR name and location of external definitions in a source file

clang-format

is a tool to format C/C++/Java/JavaScript/Objective-C/Protobuf code

clang-linker-wrapper

is a wrapper utility over the host linker

clang-offload-bundler

is a tool to bundle/unbundle OpenMP offloaded files associated with a common source file

clang-offload-packager

is a tool to bundle several object files into a single binary, which can then be used to create a fatbinary containing offloading code

clang-refactor

is a Clang-based refactoring tool for C, C++ and Objective-C

clang-rename

is a tool to rename symbols in C/C++ programs

clang-scan-deps

is a tool to scan for dependencies in a source file

clang-tblgen

is a program that translates compiler-related target description (.td) files into C++ code and other output formats

diagtool

is a combination of tools for dealing with diagnostics in clang

FileCheck

is a tool that reads two files (one from standard input, and one specified on the command line) and uses one to verify the other.

dsymutil

is a tool used to manipulate archived DWARF debug symbol files, compatible with the Darwin command dsymutil

git-clang-format

runs clang-format on git generated patches (requires git-2.47.1)

hmaptool

is a Python tool to dump and construct header maps

intercept-build

generates a database of build commands for a project

llc

is the LLVM static compiler

lli

is used to directly execute programs from LLVM bitcode

llvm-addr2line

is a tool used to convert addresses into file names and line numbers

llvm-ar

is the LLVM archiver

llvm-as

is the LLVM assembler

llvm-bcanalyzer

is the LLVM bitcode analyzer

llvm-bitcode-strip

strips LLVM bitcode from an object

llvm-cat

is a tool to concatenate llvm modules

llvm-cfi-verify

identifies whether Control Flow Integrity protects all indirect control flow instructions in the provided object file, DSO, or binary

llvm-config

Prints LLVM compilation options

llvm-cov

is used to emit coverage information

llvm-c-test

is a bytecode disassembler

llvm-cvtres

is a tool to convert Microsoft resource files to COFF

llvm-cxxdump

is used as a C++ ABI Data Dumper

llvm-cxxfilt

is used to demangle C++ symbols in llvm code

llvm-cxxmap

is used to remap C++ mangled symbols

llvm-debuginfo-analyzer

prints a logical representation of low-level debug information

llvm-debuginfod

is a service providing debug information over an HTTP API for analyzing stripped binaries

llvm-debuginfod-find

is an interface to the llvm-debuginfod daemon for finding debuginfod artifacts

llvm-diff

is the LLVM structural 'diff'

llvm-dis

is the LLVM disassembler

llvm-dwarfdump

prints the content of DWARF sections in object files

llvm-dwarfutil

is a tool to copy and manipulate debug info

llvm-dwp

merges split DWARF files

llvm-elfabi

is used to read information about an ELF binary's ABI

llvm-exegesis

is a benchmarking tool that uses information available in LLVM to measure host machine instruction characteristics like latency or port decomposition

llvm-extract

is used to extract a function from an LLVM module

llvm-gsymutil

is used to process GSYM Symbolication Format files which convert memory addresses to function name and source file line. These files are smaller than DWARF or Breakpad files

llvm-ifs

is used to merge interface stubs with object files

llvm-install-name-tool

is used to rewrite load commands into MachO binary format

llvm-jitlink

is used to parse relocatable object files to make their contents executable in a target process

llvm-libtool-darwin

provides basic libtool functionality on Darwin-based systems. This is mostly useful if you are generating binaries for macOS systems

llvm-link

is the LLVM linker

llvm-lipo

is used to create universal binaries from MachO files

llvm-lto

is the LLVM LTO (link time optimization) linker

llvm-lto2

is a test harness for the resolution based LTO interface

llvm-mc

is a standalone machine code assembler/disassembler

llvm-mca

is a performance analysis tool to statically measure the performance of machine code

llvm-ml

is a playground for machine code provided by LLVM

llvm-modextract

is a tool to extract one module from multimodule bitcode files

llvm-mt

is a tool to generate signed files and catalogs from a side-by-side assembly manifest (used for Microsoft SDK)

llvm-nm

is used to list LLVM bitcode and object file's symbol table

llvm-objcopy

is LLVM's version of an objcopy tool

llvm-objdump

is an LLVM object file dumper

llvm-opt-report

is a tool to generate an optimization report from YAML optimization record files

llvm-pdbutil

is a PDB (Program Database) dumper. PDB is a Microsoft format

llvm-profdata

is a small tool to manipulate and print profile data files

llvm-profgen

generates LLVM SPGO profiling information

llvm-ranlib

is used to generate an index for a LLVM archive

llvm-rc

is a platform-independent tool to compile resource scripts into binary resource files

llvm-readobj

displays low-level format-specific information about object files

llvm-readtapi

is the LLVM TAPI file reader and transformer

llvm-reduce

is used to automatically reduce testcases when running a test suite

llvm-remarkutil

converts remark files between bitstream and YAML; or prints function instruction count information in remark files

llvm-rtdyld

is the LLVM MC-JIT tool

llvm-size

is the LLVM object size dumper

llvm-split

is the LLVM module splitter

llvm-stress

is used to generate random .ll files

llvm-strings

print strings found in a binary (object file, executable, or archive library)

llvm-symbolizer

converts addresses into source code locations

llvm-tblgen

is the LLVM Target Description To C++ Code Generator

llvm-tli-checker

is the LLVM TargetLibraryInfo versus SDK checker

llvm-undname

is a tool to demangle names

llvm-xray

is an implementation of Google's XRay function call tracing system

nvptx-arch

lists NVIDIA GPUs installed; at runtime it needs libcuda.so which is not a part of BLFS

opt

is the LLVM optimizer

sancov

is the sanitizer coverage processing tool

sanstats

is the sanitizer statistics processing tool

scan-build

is a Perl script that invokes the Clang static analyzer

scan-build-py

is a Python script that invokes the Clang static analyzer

scan-view

is a viewer for Clang static analyzer results

verify-uselistorder

is the LLVM tool to verify use-list order