cross-unwinding can pass exception across runtimes e.g.
Multi-language applications ( nodejs native modules, python C extensions )
However, cross-unwinding is not default in libcxx for sometime [1]
it requires special handling at runtime by apps who use it
Switch recipe defaults to upstream defaults, which is to turn it off
This means native unwinding ( with in same runtime )
is build by default which is the most common case.
This also fixes libcxx/libunwind build errors
when enabling TC_CXX_RUNTIME = "llvm" on aarch64 particularly
[1] 85624c5de3
(From OE-Core rev: 88e1214e01c48b174aa9516c60e15a0dfb69e10f)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The optimization viewer is a web-based tool that helps developers visualize
and analyze LLVM's optimization passes. It generates reports showing
Which optimizations were applied to specific functions.
Why certain optimizations were or weren't performed.
Performance remarks and missed optimization opportunities
Interactive views of the optimization pipeline.
Its a nice add-on, keep it disabled by default.
(From OE-Core rev: 70c43e35fb7ce0b71f28a91a1864f02338545522)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The environment has variables like LDFLAGS set which CMake will try and
use, so until our cmake class is comprehensive enough to nuke those in
the native toolchain file (if that is possible) do a build inside a
clean (apart from PATH) environment to avoid cross-contamination.
As noticed by Khem Raj, this can fail in nativesdk builds as the target
flags will be passed to the native link resulting in failures like:
Requesting program interpreter: /usr/local/oe-sdk-hardcoded-buildpath/sysroots/x86_64-yoesdk-linux/lib/ld-linux-x86-64.so.2
(From OE-Core rev: 0013f501d1de01606b812d930ef6ca842633f551)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
'cross' dependencies are hard, the code currently handles adding MLPREFIX
in these cases but may not in future depending upon implementation.
Add the correct value explictly.
(From OE-Core rev: 8393be1792cee0fbe76d75fd20a97dda41134880)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The lld linker is a standalone project under the LLVM umbrella that just
depends on libLLVM and nothing else, such as clang.
To reduce the build time of clang if lld is not being used, split it out
into a separate recipe.
To ensure that lld is present if needed, the clang-cross recipe will
depend on lld-native if ld-is-lld is enable.
(From OE-Core rev: 5212b69f892af8c9b080fee2c21533a2bbdd2755)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
LLDB Uses SWIG to generate python bindings, however
LLDB is no longer part of clang anymore, its a
separate recipe.
(From OE-Core rev: deda1e227dbdcb04c7bfb0ad0afef6ff05674c4f)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
None of these recipes need a native build of Python as they don't depend
on native Python modules, nor build Python modules against the target.
(From OE-Core rev: e0da53fca5ec30d883216bfb0be7d96ad597aa69)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Now that lldb has been split out of the clang recipe, there's no need
for the clang recipe to have Python integration beyond the current
Python major version for installing the libclang bindings, which can be
done by simply inheriting python3-dir.
The installation of six.py was done by lldb, and was removed in LLVM 16.
(From OE-Core rev: 5cbe58c604233a0d367e1bb1e3fce0492ae0e836)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Specifically, c-index-test is a 35MB binary which isn't needed if we're
not running the tests.
(From OE-Core rev: a61c3ab08039d382ec2216b8a5deaf256d7bf9c5)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This was added by a certain Mr Burton back in 2016[1] to avoid needing
to depend on perl-native or python3-native, and use the host perl and
python.
However, perl is no longer needed and python3-native is depended on
because the Python bindings are build, so this is now redundant.
[1] meta-clang: 2e003cc ("clang: don't depend on native perl or python")
(From OE-Core rev: 540b02e0fb168fade2afc1d657463c7e0248ae94)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
As a single clang-native build is a cross-compiler for all targets, this
recipe should simply create target-prefixed symlinks.
As such, it doesn't need to pull in the common clang include files, or
set TOOLCHAIN.
Unsetting PACKAGES isn't needed as cross inherits nopackages.
The sysroot preprocessing is used to stage bindir, but cross recipes
already stage this directory.
There doesn't appear to be a need to copy clang, so these are now linked
too.
SSTATE_SCAN_FILES only works on actual files not symlinks, so setting it
is pointless.
(From OE-Core rev: 5d96ed55acf7b9a908241222097ad809355de7cf)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Provide libunwind based on TC_CXX_RUNTIME
When TC_CXX_RUNTIME is set to be 'llvm' distro wants
to use llvm provided runtime elements therefore provide
libunwind when TC_CXX_RUNTIME = "llvm"
(From OE-Core rev: 61cf402d5bf95094a02d4ea3b1a60cecc04aa0f4)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
These functions are useful outside of just the clang recipe, so move
them to a common .inc file so they can be used by other clang-related
recipes.
Also make the function fail if it doesn't recognise the architecture,
instead of returning the empty string and causing mysterious fails later.
(From OE-Core rev: 1d5298533e97dab7636f885ddd740352782395b0)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
LLDB defaults to adding rpaths into the binaries which are then stripped
by CMake on install.
However, this rpath removal is implemented by editing the binary instead
of relinking at install time, so the final binary will have an entry in
the dynstr section which is all nulls but is as long as the build path.
Obviously this breaks reproducibility, so disable the use of rpaths in
LLDB to remove this problem.
(From OE-Core rev: d96e0458b696a7359d310cbe112c5dc2fc60f97d)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
LLDB links to clang libraries, so it really does need clang and not just
LLVM.
(From OE-Core rev: 1a37bb2f0bbdbd599e1d5967eae1ecde4eff6a80)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Set LLVM_HEADERS_TBLGEN so that we don't build another copy of tblgen.
Also remove LLVM_OPTIMIZED_TABLEGEN, this is redundant as we're not
building a tablegen binary in this recipe anymore.
(From OE-Core rev: 259a3e9a8281c4e9ab73dee82738f359f029b78d)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The cmake.bbclass already does this.
(From OE-Core rev: 85c1aecdf847b97409930d915a90ab850e92bc38)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The GPU targets are incredibly slow to build, so if the DISTRO_FEATURES
doesn't include opengl or vulkan assume that the user will not be using
a GPU and disable them.
Alternatively, a distribution could state that they'll only be using
one of the backends, and set LLVM_TARGETS_GPU explicitly.
On my build machine, disabling the GPU targets reduces the build time of
clang-native from 21m to 16m.
(From OE-Core rev: 2273a0685757421f39541a352d77b67e5ba604d3)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Split the libclc subproject out of the clang recipe and into a dedicated
libclc recipe.
This is useful because libclc is the OpenCL runtime library and as such
isn't target-specific and needs a native clang to build, not a target
libllvm.
Verified that nothing is dropped by adding clang and libclc to an image
and verifying that the file list is the same before and after this
change.
We need to patch the libclc CMakeLists to allow it to use an out-of-tree
prepare_builtins binary, discussion is ongoing with upstream to resolve
this properly.
(From OE-Core rev: 33a8742a1280b4c6779a7aa487c2dd4a713babe6)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This ensures that default system linker is used correctly based upon
distro features, current default remain same i.e. uses BFD linker
(From OE-Core rev: 33e7c85e92ddbc35989e7afcaf5fe4c14efdefab)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Neither LLVM_ENABLE_TERMINFO or COMPILER_RT_TERMINFO_LIB are used in the
current CMakeLists.
(From OE-Core rev: 134880baa4a971660e0800f70404d15c758fb7db)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Accidentally duplicated the beginning of the variable name.
(From OE-Core rev: 1484e2cf36428fbe7f5095ec881fdd018eea344b)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Brings following fixes
* 87f0227cb601 [InstCombine] Avoid folding `select(umin(X, Y), X)` with min/max values in false arm (#143020)
* df43f93388b7 [PhaseOrdering] Add test for #139050 (NFC)
* 25bcf1145fd7 [RISCV] Fix assertion failure when using -fstack-clash-protection (#135248)
* 6fb913d3e2ec [RelLookupTableConverter] Drop unnamed_addr for GVs in entries to avoid generating GOTPCREL relocations (#146068)
* 0c9f909b7976 [AArch64][SME] Fix restoring callee-saves from FP with hazard padding (#143371)
* fa792cd4c630 [AsmPrinter] Always emit global equivalents if there is non-global uses (#145648)
* ce455b382c08 [objcopy][MachO] Revert special handling of encryptable binaries (#144058)
* 0de59a293f7a [X86] Ignore NSW when DstSVT is i32 (#131755)
* 9af763f038f7 [gtest] Fix building on OpenBSD/sparc64 (#145225)
* 1daceb20611f [LoongArch] Pass OptLevel to LoongArchDAGToDAGISel correctly
* b21155f97a0a [LoongArch] Precommit test case to show bug in LoongArchISelDagToDag
* da18fb9f04ce [LoongArch] Fix xvshuf instructions lowering (#145868)
* 65ce78f338cf [LoongArch] Pre-commit test for fixing xvshuf instructions. NFC
* 5532d5b745e4 [AArch64] Ensure the LR is preserved if we must call __arm_get_current_vg (#145760)
* 5ac3ce819688 [WebAssembly] Fix inline assembly with vector types (#146574)
* b83658b7e2c8 Bump version to 20.1.8
(From OE-Core rev: 4c6e132d6e5e49426c911d0fa9215957eb43f186)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There are a number of recipes that are part of the LLVM Project, so set
CVE_PRODUCT to llvm:llvm in common.inc to ensure that all of the recipes
are covered.
Also add llvm:clang in the clang recipe, as there are a number of CVEs
with that product name.
(From OE-Core rev: 319e97643c1e342491931b6274996d1c8caa7d33)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
v1 of the lldb patchset was accidentally merged, which didn't remove
this package from the clang recipe.
(From OE-Core rev: f389f14983cf87238f9a073b50837583596735ea)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
LLDB takes a reasonable amount of time to compile, but also isn't an
essential component of the LLVM suite.
Instead of always building it when we build clang, split it out into a
separate recipe.
On my build machine where clang takes 21 minutes to build with lldb, it
takes 19 minutes without lldb.
(From OE-Core rev: 9da4900aa5a37718bd42f277d5a1805ec897b1b4)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Building the LLVM projects often means using the TableGen tools
(llvm-tblgen etc).
We currently build them as part of clang-native, but I am teasing the
clang recipe into its component parts and having to build llvm-native
or lldb-native simply for one tool isn't ideal.
Instead, add a native recipe that simply builds the tablegen binaries
for llvm, clang, and lldb
(From OE-Core rev: fbf63e03fe09ca74022c9d06442b4f1021b71d57)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There's a BPN assignment in common.inc which means all recipes need to
either be called clang, or set BPN themselves.
Move the assignment to the clang recipes. For now I'm leaving the
existing BPN assignments in the other recipes, in case there are complex
multilib-related reasons to retain them.
(From OE-Core rev: fc7e8c3e5c19a1885bec564c8fc07df5a13c8bd4)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Currently, clang when used for native builds and uses --rtlib=compiler-rt
does not find the LLVM runtime because its installed in different
directory than where compiler expects it to be. As a result, build fails
for packages using clang in both capacity ( native and cross )
e.g. qtwebengine
Make the build work across native sanitizers as well.
(From OE-Core rev: 4d55698330ce6720cab5dfe9b4e00efeb40ba6e6)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
compiler-rt configure calls for c++ compiler which can cause C++ runtime to
not be detected on some Yocto autobuilder workers running ubuntu 24.04
therefore let it use gcc for native version
Set TOOLCHAIN_NATIVE for using clang for libcxx
(From OE-Core rev: d54512c1c0a98516077b77d5414af47d8c2b8c39)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some of them were introduced by mass-removal of S = WORKDIR/git assignments;
rather than try to fix up (or redo) just these, I've run this sed command over
the whole tree:
sed -i -z -E 's/([ \t\f\v\r]*\n){3,}/\n\n/g' `find . -name *.bb -o -name *.inc`
The rationale is that more than one empty line is wasting vertical screen space, and
does nothing for readability.
(From OE-Core rev: cedc4ff7c9bcfb22a20e43e47f9759f4007a4f1a)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This and some further commits allows removing the 'backwards
compatibility magic' in do_unpack that moves unpacked sources
to where S is set to.
(From OE-Core rev: 6b2f14cd2d110113f1065bdc818ec4e9dbd4b054)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This change implements a toolchain selection mechanism. Selection is
made using a set of variables, primarily PREFERRED_TOOLCHAIN_TARGET which
defaults to gcc.
It uses the familiar name for toolchain e.g. "gcc" which selects GNU
compiler + binutils as default C/C++ toolchain or "clang" which will
use LLVM/Clang Compiler. Layers an add their own toolchain definitions
too.
There are also PREFERRED_TOOLCHAIN_NATIVE and PREFERRED_TOOLCHAIN_SDK
which will ulitmately allow selection of the toolchain used for the
native/cross and nativesdk/crosssdk compilers. This currently isn't
functional but is essential to the patch to ensure things are set
to the existing gcc support in those cases.
Users would most commonly want to set:
PREFERRED_TOOLCHAIN_TARGET ?= "clang"
in local.conf or other distro specific global configuration metadata.
It is also selectable at recipe scope, since not all packages are
buildable with either clang or gcc, a recipe can explicitly require
a given toolchain using the TOOLCAHIN variable, e.g. glibc can not
be built with clang therefore glibc recipe sets:
TOOLCHAIN = "gcc"
The TOOLCHAIN variable is distinct from the user preference so recipes
with specific requirements can be identified. This also allows different
polcies to be be specified for native/SDK cases in the future.
(From OE-Core rev: 45bdedd213aff8df3214b95ef2a8551c0abd93a0)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Brings following fixes
* 7b09d7b44638 [analyzer] Workaround for slowdown spikes (unintended scope increase) (#136720)
* a708fb737a78 [RISCV] Allow `Zicsr`/`Zifencei` to duplicate with `g` (#136842)
* 1c0368417f55 [clang][analysis] Fix flaky clang/test/Analysis/live-stmts.cpp test (2nd attempt) (#127406)
* 0439d1d36312 [Clang] Fix handling of reference types in tryEvaluateBuiltinObjectSize (#138247)
* 74ed1ac61104 [sanitizer_common] Fix build on ppc64+musl (#120036)
* 2d079b96a5fb release/20.x: [clang-format] Fix a crash on formatting missing r_paren/r_brace (#138230)
* 2cacf46f35c8 [X86][TargetLowering] Avoid deleting temporary nodes in `getNegatedExpression` (#139029)
* f233430d977b [AArch64] Fix feature list for FUJITSU-MONAKA processor (#139212)
* 41c36d940804 [clang] Fix unused variable warning in MS mangler from constant matrix patch
* 72ad9be1e337 [Clang][MicrosoftMangle] Implement mangling for ConstantMatrixType (#134930)
* 0019b7d0ae0b [wasm-ld] Refactor WasmSym from static globals to per-link context (#134970)
* b7b834e2a20e [RTSan][Darwin] Adjust OSSpinLock/_os_nospin_lock interceptor and tests (#132867)
* f811c7df0a10 [rtsan][Apple] Add interceptor for _os_nospin_lock_lock (#131034)
* 4370072022e5 [clang] Forward TPL of NestedNameSpecifier
* d34d5296095b Support z17 processor name and scheduler description
* a7166c373946 release/20.x: [clang-format] RemoveParentheses shouldn't remove empty parentheses (#138229)
* 5429418cb064 [clang] Add support for Debian 14 Forky and Debian 15 Duke (#138460)
* be087ab35970 [libc++] Re-introduce _LIBCPP_DISABLE_AVAILABILITY (#134158)
* 2b34040173f7 [clang-repl] Fix destructor for interpreter for the cuda negation case (#138091)
* ae97a56d363f [Hexagon] Add missing patterns to select PFALSE and PTRUE (#138712)
* 2386c377db4f [BasicAA] Gracefully handle large LocationSize (#138528)
* 961ce35e2957 [OpenMP] Add pre sm_70 load hack back in (#138589)
* 009f3c10d1c1 [LLD][COFF] Don't dllimport from static libraries (#134443)
* 70eed33971d9 [InstCombine] Do not combine shuffle+bitcast if the bitcast is eliminable. (#135769)
* 6ddf2e5d10f8 [clang-tidy] Do not pass any file when listing checks in run_clang_ti… (#137286)
* 8272e451613d [flang] Exempt construct entities from SAVE check for PURE (#131383)
* 069ef671e0ab [AArch64][SME] Allow spills of ZT0 around SME ABI routines again (#136726)
* a38e1ae2041d [AArch64][SME2] Don't preserve ZT0 around SME ABI routines (#132722)
* be4097b6ee57 Fix crash lowering stack guard on OpenBSD/aarch64. (#125416)
* aecbb2364a7c [Clang] Fix the trailing comma regression (#136273)
* ebfae55af454 Bump version to 20.1.5
(From OE-Core rev: 07a1c1d765f220b1f75e503404d689524a8c14ef)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Building of spirv-mesa and spirv64-mesa CLC targets (which are required
for Mesa to work) requires a working llvm-spirv tool (provided by the
spirv-llvm-translator-native package). Make clang build-depend on the
corresponding recipe in order to be able to build requried targets.
Fixes: 4178fe97371b ("clang: split SPIRV-LLVM-Translator to its own recipe")
(From OE-Core rev: 177aaa7912f317da4a17a57081eb4f5667ef2c02)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add logic to deduce compiler-rt target architecture based on tune
features
Cleanup the cmake options for native/nativesdk/target recipes
(From OE-Core rev: 579d4d85574624e39140fb74f6d37dcd7bf363a6)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
