RISC-V GNU Compiler Toolchain

This is the RISC-V C and C++ cross-compiler. It supports two build modes: a generic ELF/Newlib toolchain and a more sophisticated Linux-ELF/glibc toolchain.

Getting the sources

This repository uses submodules, but submodules will fetch automatically on demand, so --recursive or git submodule update --init --recursive is not needed.

$ git clone https://github.com/riscv/riscv-gnu-toolchain

Warning: git clone takes around 6.65 GB of disk and download size

Prerequisites

Several standard packages are needed to build the toolchain.

On Ubuntu, executing the following command should suffice:

$ sudo apt-get install autoconf automake autotools-dev curl python3 python3-pip libmpc-dev libmpfr-dev libgmp-dev gawk build-essential bison flex texinfo gperf libtool patchutils bc zlib1g-dev libexpat-dev ninja-build git cmake libglib2.0-dev libslirp-dev

On Fedora/CentOS/RHEL OS, executing the following command should suffice:

$ sudo yum install autoconf automake python3 libmpc-devel mpfr-devel gmp-devel gawk  bison flex texinfo patchutils gcc gcc-c++ zlib-devel expat-devel libslirp-devel

On Arch Linux, executing the following command should suffice:

$ sudo pacman -Syyu autoconf automake curl python3 libmpc mpfr gmp gawk base-devel bison flex texinfo gperf libtool patchutils bc zlib expat libslirp

Also available for Arch users on the AUR: https://aur.archlinux.org/packages/riscv-gnu-toolchain-bin

On macOS, you can use Homebrew to install the dependencies:

$ brew install python3 gawk gnu-sed make gmp mpfr libmpc isl zlib expat texinfo flock libslirp

When executing the instructions in this README, please use gmake instead of make to use the newly installed version of make. To build the glibc (Linux) on macOS, you will need to build within a case-sensitive file system. The simplest approach is to create and mount a new disk image with a case sensitive format. Make sure that the mount point does not contain spaces. This is not necessary to build newlib or gcc itself on macOS.

This process will start by downloading about 200 MiB of upstream sources, then will patch, build, and install the toolchain. If a local cache of the upstream sources exists in $(DISTDIR), it will be used; the default location is /var/cache/distfiles. Your computer will need about 8 GiB of disk space to complete the process.

Installation (Newlib)

To build the Newlib cross-compiler, pick an install path (that is writeable). If you choose, say, /opt/riscv, then add /opt/riscv/bin to your PATH. Then, simply run the following command:

./configure --prefix=/opt/riscv
make

You should now be able to use riscv64-unknown-elf-gcc and its cousins.

Note: If you're planning to use an external library that replaces part of newlib (for example libgloss-htif), read the FAQ.

Installation (Linux)

To build the Linux cross-compiler, pick an install path (that is writeable). If you choose, say, /opt/riscv, then add /opt/riscv/bin to your PATH. Then, simply run the following command:

./configure --prefix=/opt/riscv
make linux

The build defaults to targeting RV64GC (64-bit) with glibc, even on a 32-bit build environment. To build the 32-bit RV32GC toolchain, use:

./configure --prefix=/opt/riscv --with-arch=rv32gc --with-abi=ilp32d
make linux

In case you prefer musl libc over glibc, configure just like above and opt for make musl instead of make linux.

Supported architectures are rv32i or rv64i plus standard extensions (a)tomics, (m)ultiplication and division, (f)loat, (d)ouble, or (g)eneral for MAFD.

Supported ABIs are ilp32 (32-bit soft-float), ilp32d (32-bit hard-float), ilp32f (32-bit with single-precision in registers and double in memory, niche use only), lp64 lp64f lp64d (same but with 64-bit long and pointers).

Installation (Newlib/Linux multilib)

To build either cross-compiler with support for both 32-bit and 64-bit, run the following command:

./configure --prefix=/opt/riscv --enable-multilib

And then either make, make linux or make musl for the Newlib, Linux glibc-based or Linux musl libc-based cross-compiler, respectively.

The multilib compiler will have the prefix riscv64-unknown-elf- or riscv64-unknown-linux-gnu- but will be able to target both 32-bit and 64-bit systems. It will support the most common -march/-mabi options, which can be seen by using the --print-multi-lib flag on either cross-compiler.

Linux toolchain has an additional option --enable-default-pie to control the default PIE enablement for GCC, which is disable by default.

To customize the enabled languages, use option --with-languages=. For example, if you want to enable c,c++,fortran, use ./configure --with-languages=c,c++,fortran. This option only takes effect for the GNU toolchain.

Troubleshooting Build Problems

Builds work best if installing into an empty directory. If you build a hard-float toolchain and then try to build a soft-float toolchain with the same --prefix directory, then the build scripts may get confused and exit with a linker error complaining that hard float code can't be linked with soft float code. Removing the existing toolchain first, or using a different prefix for the second build, avoids the problem. It is OK to build one newlib and one linux toolchain with the same prefix. But you should avoid building two newlib or two linux toolchains with the same prefix.

If building a linux toolchain on a MacOS system, or on a Windows system using the Linux subsystem or cygwin, you must ensure that the filesystem is case-sensitive. A build on a case-insensitive filesystem will fail when building glibc because *.os and *.oS files will clobber each other during the build eventually resulting in confusing link errors.

Centos (and RHEL) provide old GNU tools versions that may be too old to build a RISC-V toolchain. There is an alternate toolset provided that includes current versions of the GNU tools. This is the devtoolset provided as part of the Software Collection service. For more info, see the devtoolset-7 URL. There are various versions of the devtoolset that are available, so you can also try other versions of it, but we have at least one report that devtoolset-7 works.

Advanced Options

There are a number of additional options that may be passed to configure. See './configure --help' for more details.

Also you can define extra flags to pass to specific projects: BINUTILS_NATIVE_FLAGS_EXTRA, BINUTILS_TARGET_FLAGS_EXTRA, GCC_EXTRA_CONFIGURE_FLAGS, GDB_NATIVE_FLAGS_EXTRA, GDB_TARGET_FLAGS_EXTRA, GLIBC_TARGET_FLAGS_EXTRA, NEWLIB_TARGET_FLAGS_EXTRA. Example: GCC_EXTRA_CONFIGURE_FLAGS=--with-gmp=/opt/gmp make linux

Set default ISA spec version

--with-isa-spec= can specify the default version of the RISC-V Unprivileged (formerly User-Level) ISA specification.

Possible options are: 2.2, 20190608 and 20191213.

The default version is 20191213.

More details about this option you can refer this post RISC-V GNU toolchain bumping default ISA spec to 20191213.

Build with customized multi-lib configure.

--with-multilib-generator= can specify what multilibs to build. The argument is a semicolon separated list of values, possibly consisting of a single value. Currently only supported for riscv--elf*. The accepted values and meanings are given below.

Every config is constructed with four components: architecture string, ABI, reuse rule with architecture string and reuse rule with sub-extension.

Re-use part support expansion operator (*) to simplify the combination of different sub-extensions, example 4 demonstrate how it uses and works.

Example 1: Add multi-lib support for rv32i with ilp32.

./configure --with-multilib-generator="rv32i-ilp32--"

Example 2: Add multi-lib support for rv32i with ilp32 and rv32imafd with ilp32.

./configure --with-multilib-generator="rv32i-ilp32--;rv32imafd-ilp32--"

Example 3: Add multi-lib support for rv32i with ilp32; rv32im with ilp32 and rv32ic with ilp32 will reuse this multi-lib set.

./configure --with-multilib-generator="rv32i-ilp32-rv32im-c"

Example 4: Add multi-lib support for rv64ima with lp64; rv64imaf with lp64, rv64imac with lp64 and rv64imafc with lp64 will reuse this multi-lib set.

./configure --with-multilib-generator="rv64ima-lp64--f*c"

Test Suite

The Dejagnu test suite has been ported to RISC-V. This can be run with a simulator for the elf and linux toolchains. The simulator can be selected by the SIM variable in the Makefile, e.g. SIM=qemu, SIM=gdb, or SIM=spike (experimental).In addition, the simulator can also be selected with the configure time option --with-sim=.However, the testsuite allowlist is only mintained for qemu.Other simulators might get extra failures.

Additional Prerequisite

A helper script to setup testing environment requires pyelftools.

On newer versions of Ubuntu, executing the following command should suffice:

$ sudo apt-get install python3-pyelftools

On newer versions of Fedora and CentOS/RHEL OS (9 or later), executing the following command should suffice:

$ sudo yum install python3-pyelftools

On Arch Linux, executing the following command should suffice:

$ sudo pacman -Syyu python-pyelftools

If your distribution/OS does not have pyelftools package, you can install it using PIP.

# Assuming that PIP is installed
$ pip3 install --user pyelftools

Testing GCC

To test GCC, run the following commands:

./configure --prefix=$RISCV --disable-linux --with-arch=rv64ima # or --with-arch=rv32ima
make newlib
make report-newlib SIM=gdb # Run with gdb simulator

./configure --prefix=$RISCV
make linux
make report-linux SIM=qemu # Run with qemu

./configure --prefix=$RISCV --with-sim=spike
make linux
make report               # Run with spike

Note: - spike only support rv64* bare-metal/elf toolchain. - gdb simulator only support bare-metal/elf toolchain.

Selecting the tests to run in GCC's regression test suite

By default GCC will execute all tests of its regression test suite. While running them in parallel (e.g. make -j$(nproc) report) will significanlty speed up the execution time on multi-processor systems, the required time for executing all tests is usually too high for typical development cycles. Therefore GCC allows to select the tests that are being executed using the environment variable RUNTESTFLAGS.

To restrict a test run to only RISC-V specific tests the following command can be used:

RUNTESTFLAGS="riscv.exp" make report

To restrict a test run to only RISC-V specific tests with match the pattern "zb.c" and "sm.c" the following command can be used:

RUNTESTFLAGS="riscv.exp=zb.c\ sm.c" make report

Testing GCC, Binutils, and glibc of a Linux toolchain

The default Makefile target to run toolchain tests is report. This will run all tests of the GCC regression test suite. Alternatively, the following command can be used to do the same:

make check-gcc

The following command can be used to run the Binutils tests:

make check-binutils

The command below can be used to run the glibc tests:

make check-glibc-linux

Adding more arch/abi combination for testing without introducing multilib

--with-extra-multilib-test can be used when you want to test more combination of arch/ABI, for example: built a linux toolchain with multilib with rv64gc/lp64d and rv64imac/lp64, but you want to test more configuration like rv64gcv/lp64d or rv64gcv_zba/lp64d, then you can use --with-extra-multilib-test to specify that via --with-extra-multilib-test="rv64gcv-lp64d;rv64gcv_zba-lp64d", then the testing will run for rv64gc/lp64d, rv64imac/lp64, rv64gcv/lp64d and rv64gcv_zba/lp64d.

--with-extra-multilib-test support bare-metal and linux toolchain and support even multilib is disable, but the user must ensure extra multilib test configuration can be work with existing lib/multilib, e.g. rv32gcv/ilp32 test can't work if multilib didn't have any rv32 multilib.

--with-extra-multilib-test also support more complicated format to fit the requirements of end-users. First of all, the argument is a list of test configurations. Each test configuration are separated by ;. For example:

rv64gcv-lp64d;rv64_zvl256b_zvfh-lp64d

For each test configuration, it has two parts, aka required arch-abi part and optional build flags. We leverage : to separate them with some restrictions.

arch-abi should be required and there must be only one at the begining of the test configuration.
build flags is a array-like flags after the arch-abi, there will be two ways to arrange them, aka AND, OR operation.
If you would like the flags in build flags array acts on arch-abi simultaneously, you can use : to separate them. For example:

rv64gcv-lp64d:--param=riscv-autovec-lmul=dynamic:--param=riscv-autovec-preference=fixed-vlmax

will be consider as one target board same as below:

riscv-sim/-march=rv64gcv/-mabi=lp64d/-mcmodel=medlow/--param=riscv-autovec-lmul=dynamic/--param=riscv-autovec-preference=fixed-vlmax

If you would like the flags in build flags array acts on arch-abi respectively, you can use ',' to separate them. For example:

rv64gcv-lp64d:--param=riscv-autovec-lmul=dynamic,--param=riscv-autovec-preference=fixed-vlmax

will be consider as two target boards same as below:

riscv-sim/-march=rv64gcv/-mabi=lp64d/-mcmodel=medlow/--param=riscv-autovec-preference=fixed-vlmax riscv-sim/-march=rv64gcv/-mabi=lp64d/-mcmodel=medlow/--param=riscv-autovec-lmul=dynamic

However, you can also leverage AND(:), OR(,) operator together but the OR(,) will always have the higher priority. For example:

rv64gcv-lp64d:--param=riscv-autovec-lmul=dynamic:--param=riscv-autovec-preference=fixed-vlmax,--param=riscv-autovec-lmul=m2

will be consider as tow target boars same as below:

riscv-sim/-march=rv64gcv/-mabi=lp64d/-mcmodel=medlow/--param=riscv-autovec-lmul=dynamic/--param=riscv-autovec-preference=fixed-vlmax riscv-sim/-march=rv64gcv/-mabi=lp64d/-mcmodel=medlow/--param=riscv-autovec-lmul=m2

LLVM / clang

LLVM can be used in combination with the RISC-V GNU Compiler Toolchain to build RISC-V applications. To build LLVM with C and C++ support the configure flag --enable-llvm can be used.

E.g. to build LLVM on top of a RV64 Linux toolchain the following commands can be used:

./configure --prefix=$RISCV --enable-llvm --enable-linux make

Note, that a combination of --enable-llvm and multilib configuration flags is not supported.

Below are examples how to build a rv64gc Linux/newlib toolchain with LLVM support, how to use it to build a C and a C++ application using clang, and how to execute the generated binaries using QEMU.

Build Linux toolchain and run examples:

# Build rv64gc toolchain with LLVM
./configure --prefix=$RISCV --enable-llvm --enable-linux --with-arch=rv64gc --with-abi=lp64d
make -j$(nproc) all build-sim SIM=qemu
# Build C application with clang
$RISCV/bin/clang -march=rv64imafdc -o hello_world hello_world.c
$RISCV/bin/qemu-riscv64 -L $RISCV/sysroot ./hello_world
# Build C++ application with clang
$RISCV/bin/clang++ -march=rv64imafdc -stdlib=libc++ -o hello_world_cpp hello_world_cpp.cxx
$RISCV/bin/qemu-riscv64 -L $RISCV/sysroot ./hello_world_cpp

Build newlib toolchain and run examples (don't work with --with-multilib-generator=):

# Build rv64gc bare-metal toolchain with LLVM
./configure --prefix=$RISCV --enable-llvm --disable-linux --with-arch=rv64gc --with-abi=lp64d
make -j$(nproc) all build-sim SIM=qemu
# Build C application with clang
$RISCV/bin/clang -march=rv64imafdc -o hello_world hello_world.c
$RISCV/bin/qemu-riscv64 -L $RISCV/sysroot ./hello_world
# Build C++ application with clang using static link
$RISCV/bin/clang++ -march=rv64imafdc -static -o hello_world_cpp hello_world_cpp.cxx
$RISCV/bin/qemu-riscv64 -L $RISCV/sysroot ./hello_world_cpp

Development

This section is only for developer or advanced user, or you want to build toolchain with your own source tree.

Update Source Tree

riscv-gnu-toolchain contain stable but not latest source for each submodule, in case you want to using latest develoment tree, you can use following command to upgrade all submodule.

git submodule update --remote

Or you can upgrade specific submodule only.

git submodule update --remote <component>

For example, upgrade gcc only, you can using following command:

git submodule update --remote gcc

How to Check Which Branch are Used for Specific submodule

The branch info has recorded in .gitmodules file, which can set or update via git submodule add -b or git submodule set-branch.

However the only way to check which branch are using is to check .gitmodules file, here is the example for gcc, it's using releases/gcc-12 branch, so it will has a section named gcc and has a field branch is releases/gcc-12.

[submodule "gcc"]
        path = gcc
        url = ../gcc.git
        branch = releases/gcc-12

Use Source Tree Other Than `riscv-gnu-toolchain`

riscv-gnu-toolchain also supports using out-of-tree source to build the toolchain. There are several configure options for specifying the source tree of each submodule/component.

For example, if you have GCC sources in $HOME/gcc, use --with-gcc-src to build the toolchain using those sources:

./configure ... --with-gcc-src=$HOME/gcc

Here is the list of configure options for specifying alternative sources for the various submodules/components:

--with-binutils-src
--with-dejagnu-src
--with-gcc-src
--with-gdb-src
--with-glibc-src
--with-linux-headers-src
--with-llvm-src
--with-musl-src
--with-newlib-src
--with-pk-src
--with-qemu-src
--with-spike-src
--with-uclibc-src

Build host GCC to check for compiler warnings

GCC contributions have to meet several requirements to qualify for upstream inclusion. Warning free compilation with a compiler build from the same sources is among them. The flag --enable-host-gcc does exaclty that:

Initially a host GCC will be built
This host GCC is then used to build the cross compiler
The cross compiler will be built with -Werror to identify code issues

FAQ

Ensuring Code Model Consistency

If parts of newlib are going to be replaced with an external library (such as with libgloss-htif for Berkeley Host-Target Interface), you should take care to ensure that both newlib and the external library are built using the same code model. For more information about RISC-V code models, read this SiFive blog article.

Errors that indicate a code model mismatch include "relocation overflow" or "relocation truncated" errors from the linker being unable to successfully relocate symbols in the executable.

By default, riscv-gnu-toolchain builds newlib with -mcmodel=medlow. You can use the alternative medany code model (as used in libgloss-htif) by passing --with-cmodel=medany to the configure script.