These are a set of steps to add support for a new target. There are numerous end states and paths to get there, so not all sections may be relevant to your desired goal.
For very new targets, you may need to use a different fork of LLVM
than what is currently shipped with Rust. In that case, navigate to
the src/llvm-project
git submodule (you might need to run ./x.py check
at least once so the submodule is updated), check out the
appropriate commit for your fork, then commit that new submodule
reference in the main Rust repository.
An example would be:
cd src/llvm-project
git remote add my-target-llvm some-llvm-repository
git checkout my-target-llvm/my-branch
cd ..
git add llvm-project
git commit -m 'Use my custom LLVM'
If you have a local LLVM checkout that is already built, you may be able to configure Rust to treat your build as the system LLVM to avoid redundant builds.
You can tell Rust to use a pre-built version of LLVM using the target
section
of config.toml
:
[target.x86_64-unknown-linux-gnu]
llvm-config = "/path/to/llvm/llvm-7.0.1/bin/llvm-config"
If you are attempting to use a system LLVM, we have observed the following paths before, though they may be different from your system:
/usr/bin/llvm-config-8
/usr/lib/llvm-8/bin/llvm-config
Note that you need to have the LLVM FileCheck
tool installed, which is used
for codegen tests. This tool is normally built with LLVM, but if you use your
own preinstalled LLVM, you will need to provide FileCheck
in some other way.
On Debian-based systems, you can install the llvm-N-tools
package (where N
is the LLVM version number, e.g. llvm-8-tools
). Alternately, you can specify
the path to FileCheck
with the llvm-filecheck
config item in config.toml
or you can disable codegen test with the codegen-tests
item in config.toml
.
You should start with a target JSON file. You can see the specification
for an existing target using --print target-spec-json
:
rustc -Z unstable-options --target=wasm32-unknown-unknown --print target-spec-json
Save that JSON to a file and modify it as appropriate for your target.
Once you have filled out a JSON specification and been able to compile somewhat successfully, you can copy the specification into the compiler itself.
You will need to add a line to the big table inside of the
supported_targets
macro in the rustc_target::spec
module. You
will then add a corresponding file for your new target containing a
target
function.
Look for existing targets to use as examples.
After adding your target to the rustc_target
crate you may want to add
core
, std
, ... with support for your new target. In that case you will
probably need access to some target_*
cfg. Unfortunately when building with
stage0 (the beta compiler), you'll get an error that the target cfg is
unexpected because stage0 doesn't know about the new target specification and
we pass --check-cfg
in order to tell it to check.
To fix the errors you will need to manually add the unexpected value to the
EXTRA_CHECK_CFGS
list in src/bootstrap/lib.rs
. Here is an example for
adding NEW_TARGET_OS
as target_os
:
- (Some(Mode::Std), "target_os", Some(&["watchos"])),
+ // #[cfg(bootstrap)] NEW_TARGET_OS
+ (Some(Mode::Std), "target_os", Some(&["watchos", "NEW_TARGET_OS"])),
You may need to make changes to crates that the compiler depends on,
such as libc
or cc
. If so, you can use Cargo's
[patch]
ability. For example, if you want to use an
unreleased version of libc
, you can add it to the top-level
Cargo.toml
file:
diff --git a/Cargo.toml b/Cargo.toml
index be15e50e2bc..4fb1248ba99 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -66,10 +66,11 @@ cargo = { path = "src/tools/cargo" }
[patch.crates-io]
# Similar to Cargo above we want the RLS to use a vendored version of `rustfmt`
# that we're shipping as well (to ensure that the rustfmt in RLS and the
# `rustfmt` executable are the same exact version).
rustfmt-nightly = { path = "src/tools/rustfmt" }
+libc = { git = "https://github.com/rust-lang/libc", rev = "0bf7ce340699dcbacabdf5f16a242d2219a49ee0" }
# See comments in `src/tools/rustc-workspace-hack/README.md` for what's going on
# here
rustc-workspace-hack = { path = 'src/tools/rustc-workspace-hack' }
After this, run cargo update -p libc
to update the lockfiles.
Beware that if you patch to a local path
dependency, this will enable
warnings for that dependency. Some dependencies are not warning-free, and due
to the deny-warnings
setting in config.toml
, the build may suddenly start
to fail. To work around the warnings, you may want to disable deny-warnings
in the config, or modify the dependency to remove the warnings.
Once you have a target specification in JSON and in the code, you can
cross-compile rustc
:
DESTDIR=/path/to/install/in \
./x.py install -i --stage 1 --host aarch64-apple-darwin.json --target aarch64-apple-darwin \
compiler/rustc library/std
If your target specification is already available in the bootstrap compiler, you can use it instead of the JSON file for both arguments.
There are two levels of tier 2 targets:
a) Targets that are only cross-compiled (rustup target add
)
b) Targets that have a native toolchain (rustup toolchain install
)
For an example of promoting a target from cross-compiled to native, see #75914.