For some unknown reason, macos-15 runners (which macos-latest
resolves to currently) seem to be consistently 25-40% slower than
macos-14 runners; the "macos" job (doing an universal binary build
for 2 architectures, with LTO) often take over 5 hours, and have
reached the 6 hour timeout once.
Switch this back to macos-14 to speed it up and reduce the risk of
hitting the timeout.
In the future, if/when the macos-14 runners are phased out, and if
the performance reduction in macos-15 runners remains, we'd need to
switch to building each architecture in a separate job, and merge
them together separately if we still want to provide universal
binaries.
If ARCHS itself was set by the caller, it would already have been
implicitly passed to the makefile - but if the caller set
TOOLCHAIN_ARCHS instead, that wasn't acted upon by the makefile.
As the shell script already has the logic for initializing this
(in the same fashion as all other shell scripts), just use that
and pass the value on to the makefile.
This matches what is done in build-all.sh for the PGO step; this
removes rather large (~50 MB) libraries that mostly are needed
for compiling for the host unix platforms. (On Linux it's needed
if building with profiling, on macOS it's needed for the builtin
functions for platform version availability too.)
This matches what is done for the regular builds; we run (build-only)
tests already with the first cross compiler toolchains, not needing
to wait for the cross compiled toolchains to run on Windows.
This is required for properly formatting long doubles on x86
with UCRT (which doesn't use the mingw builtin printfs by default).
This fixes running the LLVM testsuite on x86 after commit
9bf7d04c4386daf1ef0acf95782a59855c98474a in llvm-project.
Relevant changes include:
- Lots of work for supporting ARM64EC and ARM64X
- LLD now produces more compact string tables in executables (only
relevant when including debug symbols)
- llvm-rc now supports multiplication/division expressions, like GNU
windres
- Improved support for armv7 and aarch64 in libunwind (for Itanium
ABI "forced unwinding", which I'm not aware of being used in
practice anywhere)
- Made the libunwind, libcxxabi, libcxx and compiler-rt tests pass on
armv7 and aarch64 mingw
- Made the llvm/clang tests pass on aarch64 mingw
The ARM github actions runners now run on Windows 11 10.0.26100,
which no longer support executing armv7 binaries. (Previously they
were running Windows 11 10.0.22631.)
If using Clang from /usr/bin, this makes it clear to CMake to use
$HOST-strip, as it otherwise may pick plain /usr/bin/strip, which
may not be able to strip the cross built executables.
CMAKE_ASM_COMPILER is set implicitly from CMAKE_C_COMPILER,
but the target flags aren't implied in the same way.
This fixes cross compiling LLVM with clang for an x86_64 target.
This wasn't an issue when cross compiling LLVM with clang for an
aarch64 target, because the BLAKE3 aarch64 SIMD is in the form
of a C file (with intrinsics), not a standalone .S file.
In this case, we want to incrementally build the remaining tools,
not do a full build. In case someone runs build-all.sh with CLEAN
set, the --macos-native-tools invocation of build-llvm.sh shouldn't
wipe the build directory.
This reverts commit 9e628b0f2d.
If LLVM was built with --disable-dylib, then the individual
LLVM libraries may expose a dependency on zlib. By restricting
the lldb-mi build to detect dependencies outside of $LLVM_DIR,
the build of lldb-mi fails to find zlib, making it impossible to
find all the necessary dependencies, ending up with errors like this:
CMake Error at llvm-mingw/llvm-project/llvm/build-asserts/lib/cmake/llvm/LLVMExports.cmake:55 (set_target_properties):
The link interface of target "LLVMSupport" contains:
ZLIB::ZLIB
but the target was not found. Possible reasons include:
* There is a typo in the target name.
* A find_package call is missing for an IMPORTED target.
* An ALIAS target is missing.
Revert this change for now, to allow building lldb-mi against
a non-dylib build of LLVM.
This makes the docker image contain a PGO optimized compiler,
without needing to redo all of the PGO build stages in docker.
This reverts commit 8f42112719,
and fixes building the multiplatform image in a simpler way.
Instead of doing two separate builds, with separate Dockerfiles,
just do one multiplatform build of one Dockerfile, which
docker runs serially, for each one of the included architectures,
just packaging the prebuilt toolchains in the docker image.
Using Frontend instrumentation rather than IR. This gives less
total speedup, but makes the profile much more usable for cross
other build targets.
Build the full Linux toolchain in stage1; this avoids needing to
rebuild runtimes for doing the profiling in stage2, and avoids
needing to wait for stage3 to complete before cross building
for Windows.
Do the same as when cross compiling; disallow detecting dependencies
outside of the given CMAKE_FIND_ROOT_PATH. Don't export LLVM_DIR
as we no longer need to pass it implicitly that way.
This makes the errors clearer, if LLVM isn't found where expected,
instead of misdetecting it from a different installation of LLVM
in the system.
If lldb-mi would try to find other optional dependencies, they would
no longer be found; this was already the case when cross compiling
though.
As build-llvm now can create a larger variety of build dir names,
enumerating all possible suffixes becomes problematic (especially
as they don't necessarily appear in one canonical order).
Instead, iterate over potential matches. If cross compiling, it
is easy as we can require the directory to contain the expected
suffix. If not cross compiling, we have no explicit suffix to look
for, but we can check that the directory doesn't match common
cross compilation triples.
Passing --instrumented=IR or --instrumented=Frontend builds
instrumented tool binaries. After running them and merging the
output profile files, run build-llvm.sh with
--profile=profile.profdata.
When cross compiling with llvm-mingw, we already implicitly use
clang and lld, but we don't need to specify these options as they
are set implicitly via the llvm-mingw wrappers.
Don't use tools from $PREFIX/bin automatically; this can be
confusing and problematic. If running with $PREFIX/bin in path,
the build at this stage gets reconfigured if rebuilding after
installing clang.
For multistage builds, install into separate toolchain
directories and add the right one to $PATH before invoking
build-llvm.sh. If one really wants to use the tools from $PREFIX
for a new build into the same directory, one has to add that
directory to the path manually.
This is an option I set outside of this script in a number of
build cases where I build with ccache; move this into the script.
In practice it has little effect though; one has to set
CCACHE_BASEDIR anyway in order to be able to share a ccache
between build directories.
If enabling ccache by adding e.g. /usr/lib/ccache to the front of
PATH, without using the COMPILER_LAUNCHER variable, this won't
get enabled.
This allows specifying either a commit or a branch in the same
field, allowing simplifying both the workflow inputs, and
the expressions for passing commit/branch to both
dawidd6/action-download-artifact and actions/checkout.
This new option is available since
https://github.com/dawidd6/action-download-artifact/pull/329.
As the branch field has a default value "master", this default
not only applies when the dialog box for starting the manual
workflow is presented. But even if the prefilled branch field
is cleared, the default value also gets substituted when submitting.
In this case, when the commit and branch fields are mutually
exclusive, we can't easily have a default. Instead change the
uses of the fields, ignoring the branch field if the commit field
is filled.
If the dawidd6/action-download-artifact action would support
specifying either a commit or a branch in the same field, like
actions/checkout does, we wouldn't need this distinction at all;
this request is tracked upstream at
https://github.com/dawidd6/action-download-artifact/issues/242.
This requires setting CMAKE_CXX_COMPILER_TARGET manually; the LLVM
build system infers the wrong architecture when CMake is an
x86_64 binary.
(The wrong inferred architecture doesn't affect building libcxx,
but the wrong architecture gets passed as --target= parameter
when running the tests.)
Since 9c67ae861ccb0312fc680414a8441892d169d14a in llvm-project
(May 9th), these tests should pass on aarch64 too.
Co-authored-by: Jeremy Drake <github@jdrake.com>
Currently, "macos-latest" gets "macos-14", while there also are
"macos-15" runners available. Switch back to the latest variant
in preparation for that being updated at some point.
This reverts 7bd4b284f9.
As the preinstalled CMake on the windows-11-arm runners currently
is an x86_64 build, we need to explicitly specify
CMAKE_SYSTEM_PROCESSOR, and CMAKE_SYSTEM_NAME to make CMake
consider it a cross compilation (otherwise CMAKE_SYSTEM_PROCESSOR
gets overridden by CMAKE_HOST_SYSTEM_PROCESSOR, which gets influenced
by the x86_64 as CMake is running emulated).
We also need to set LLVM_HOST_TRIPLE, as the autodetection of
it relies on CMAKE_HOST_SYSTEM_PROCESSOR. The host triple needs
to be in normalized form with "-windows-gnu", otherwise a couple
dozens of tests fail.
We also include the AArch64 target, to make sure the built compiler
can generate code for the host.
Since 07bc54bf4554398b199f4dc849e5193b98422f23 in llvm-project
(May 5th), these tests should pass on aarch64 too.
Since 3e56f5f41843204b6e612362542b77110f8f8188 in llvm-project,
this should pass for all 4 architectures. (This doesn't pass
in the pinned release version right now.)
This avoids relying on finding these dependencies in $PATH while
running the tests. Some tests may set PATH="" while executing them,
which previously caused those tests to fail in this build setup.
The same is not an issue with MSVC builds with
-DCMAKE_MSVC_RUNTIME_LIBRARY=MultiThreadedDLL, because those
DLLs are also available in global system directories (including
DLLs such as msvcp140.dll and vcruntime140.dll).
When removing them, also accept that they are missing to begin
with.
Since 655933070219f2b6f3a457c7e5af7edd4b5291b4 in llvm-project
(Apr 25th), asan only gets built for x86.
This matches what 28851edf164a337c334755ae33fd58f03cffd5a2 in
llvm-project did.
When setting up the built libcxx files for testing, the built
files are installed into a temporary test root, and the install of
them litters the build log with mostly uninteresting output.
This was not converted in 5fe2e05419,
probably because those (automake based) makefiles work fine with
BSD make.
Use $MAKE here anyway, for consistency.
Martin Storsjö