clang 10.0.0_rc1 standalone fails to build with polly #120

Keruspe · 2020-02-09T12:39:15Z

When llvm is built with polly, clang now fails to build with undefined symbol: getPollyPluginInfo().

9.0.1 builds fine

The text was updated successfully, but these errors were encountered:

zmodem · 2020-02-09T12:53:13Z

Can you please provide steps to reproduce, i.e. with what cmake invocations etc. did you build llvm, polly, and clang?

Keruspe · 2020-02-09T13:52:09Z

llvm + polly (polly symlinked into the tools directory) were built with:

cmake -DCMAKE_C_FLAGS:STRING=-march=native -pipe -O3 -fgnuc-version=6.5.0 -flto=thin -DCMAKE_CXX_FLAGS:STRING=-march=native -pipe -O3 -fgnuc-version=6.5.0 -flto=thin -DCMAKE_AR:PATH=x86_64-pc-linux-gnu-ar -DCMAKE_RANLIB:PATH=x86_64-pc-linux-gnu-ranlib -DCMAKE_NM:PATH=x86_64-pc-linux-gnu-nm -DCMAKE_C_COMPILER:PATH=x86_64-pc-linux-gnu-cc -DCMAKE_CXX_COMPILER:PATH=x86_64-pc-linux-gnu-c++ -DCMAKE_INSTALL_PREFIX:PATH=/usr/x86_64-pc-linux-gnu -DCMAKE_FIND_ROOT_PATH=/usr/x86_64-pc-linux-gnu -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM:STRING=NEVER -DCMAKE_SYSTEM_PREFIX_PATH:PATH=/usr/x86_64-pc-linux-gnu -DCMAKE_INSTALL_LIBDIR:STRING=lib -DCMAKE_INSTALL_DATAROOTDIR:PATH=/usr/share/ -DPYTHON_EXECUTABLE:PATH=/usr/x86_64-pc-linux-gnu/bin/python3.8 -DLLVM_MAIN_SRC_DIR=/var/tmp/paludis/build/dev-lang-llvm-10.0.0_rc1/work/llvm-project/llvm -DLLVM_LIT_ARGS:STRING=-sv -DLLVM_LINK_LLVM_DYLIB:BOOL=ON -DLLVM_ENABLE_RTTI:BOOL=ON -DLLVM_ENABLE_EH:BOOL=ON -DLLVM_ENABLE_PIC:BOOL=ON -DCMAKE_BUILD_TYPE=Release -DSUPPORTS_FVISIBILITY_INLINES_HIDDEN_FLAG:BOOL=OFF -DLLVM_BINUTILS_INCDIR:STRING=/usr/x86_64-pc-linux-gnu/include/ -DLLVM_DEFAULT_TARGET_TRIPLE:STRING=x86_64-pc-linux-gnu -DLLVM_INCLUDE_TESTS:BOOL=TRUE -DLLVM_BUILD_LLVM_DYLIB:BOOL=TRUE -DCMAKE_INSTALL_PREFIX:STRING=/usr/x86_64-pc-linux-gnu/lib/llvm/10 -DCMAKE_INSTALL_MANDIR:STRING=/usr/x86_64-pc-linux-gnu/lib/llvm/10/share/man -DLLVM_INSTALL_UTILS:BOOL=TRUE -DLLVM_ENABLE_LIBPFM:BOOL=FALSE -DLLVM_ENABLE_LIBXML2:BOOL=TRUE -DLLVM_ENABLE_WERROR:BOOL=FALSE -DLLVM_ENABLE_ASSERTIONS:BOOL=FALSE -DLLVM_ENABLE_LIBEDIT:BOOL=TRUE -DLLVM_POLLY_LINK_INTO_TOOLS:BOOL=TRUE -DLLVM_TOOL_POLLY_BUILD:BOOL=TRUE -DLLVM_BUILD_TESTS:BOOL=OFF -DLLVM_ENABLE_EXPENSIVE_CHECKS:BOOL=OFF -GNinja -DPOLLY_BUNDLED_ISL:BOOL=ON -DPOLLY_ENABLE_GPGPU_CODEGEN:BOOL=OFF /var/tmp/paludis/build/dev-lang-llvm-10.0.0_rc1/work/llvm-project/llvm

Clang is built with

cmake -DCMAKE_C_FLAGS:STRING=-march=native -pipe -O3 -fgnuc-version=6.5.0 -flto=thin -DCMAKE_CXX_FLAGS:STRING=-march=native -pipe -O3 -fgnuc-version=6.5.0 -flto=thin -DCMAKE_AR:PATH=x86_64-pc-linux-gnu-ar -DCMAKE_RANLIB:PATH=x86_64-pc-linux-gnu-ranlib -DCMAKE_NM:PATH=x86_64-pc-linux-gnu-nm -DCMAKE_C_COMPILER:PATH=x86_64-pc-linux-gnu-cc -DCMAKE_CXX_COMPILER:PATH=x86_64-pc-linux-gnu-c++ -DCMAKE_INSTALL_PREFIX:PATH=/usr/x86_64-pc-linux-gnu -DCMAKE_FIND_ROOT_PATH=/usr/x86_64-pc-linux-gnu -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM:STRING=NEVER -DCMAKE_SYSTEM_PREFIX_PATH:PATH=/usr/x86_64-pc-linux-gnu -DCMAKE_INSTALL_LIBDIR:STRING=lib -DCMAKE_INSTALL_DATAROOTDIR:PATH=/usr/share/ -DPYTHON_EXECUTABLE:PATH=/usr/x86_64-pc-linux-gnu/bin/python3.8 -DLLVM_MAIN_SRC_DIR=/var/tmp/paludis/build/dev-lang-clang-10.0.0_rc1/work/llvm-project/llvm -DLLVM_LIT_ARGS:STRING=-sv -DLLVM_LINK_LLVM_DYLIB:BOOL=ON -DLLVM_ENABLE_RTTI:BOOL=ON -DLLVM_ENABLE_EH:BOOL=ON -DLLVM_ENABLE_PIC:BOOL=ON -DCMAKE_BUILD_TYPE=Release -DLLVM_CONFIG:STRING=/usr/x86_64-pc-linux-gnu/lib/llvm/10/bin/llvm-config -DCMAKE_INSTALL_PREFIX:STRING=/usr/x86_64-pc-linux-gnu/lib/llvm/10 -DCMAKE_INSTALL_MANDIR:STRING=/usr/x86_64-pc-linux-gnu/lib/llvm/10/share/man -DSUPPORTS_FVISIBILITY_INLINES_HIDDEN_FLAG:BOOL=NO -DCLANG_BUILD_TOOLS:BOOL=ON -DCLANG_ENABLE_ARCMT:BOOL=ON -DCLANG_ENABLE_CLANGD:BOOL=ON -DCLANG_ENABLE_PROTO_FUZZER:BOOL=OFF -DCLANG_ENABLE_STATIC_ANALYZER:BOOL=ON -DCLANG_INSTALL_SCANBUILD:BOOL=ON -DCLANG_INSTALL_SCANVIEW:BOOL=ON -DCLANG_LINK_CLANG_DYLIB:BOOL=ON -DCLANG_PLUGIN_SUPPORT:BOOL=ON -DCLANG_RESOURCE_DIR:STRING=../lib/clang/10 -DCLANG_INCLUDE_TESTS:BOOL=OFF -GNinja -DCLANG_DEFAULT_CXX_STDLIB:STRING=libc++ -DCLANG_DEFAULT_RTLIB:STRING=compiler-rt -DCLANG_DEFAULT_UNWINDLIB:STRING=libunwind /var/tmp/paludis/build/dev-lang-clang-10.0.0_rc1/work/llvm-project/clang

zmodem · 2020-02-10T10:11:43Z

I spent some time trying to reproduce this, but with no luck so far. Can you provide more detailed repro steps?

This is what I tried:

$ wget https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0-rc1/llvm-10.0.0rc1.src.tar.xz
$ wget https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0-rc1/polly-10.0.0rc1.src.tar.xz
$ wget https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0-rc1/clang-10.0.0rc1.src.tar.xz

$ mkdir llvm
$ tar xf llvm-10.0.0rc1.src.tar.xz --strip-components=1 -C llvm
$ mkdir llvm/tools/polly
$ tar xf polly-10.0.0rc1.src.tar.xz --strip-components=1 -C llvm/tools/polly/

$ mkdir build1 && cd build1
$ cmake -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/inst -DLLVM_LINK_LLVM_DYLIB=ON -DLLVM_BUILD_LLVM_DYLIB=ON   -DDLLVM_POLLY_LINK_INTO_TOOLS=ON -DDLLVM_TOOL_POLLY_BUILD=ON -DDPOLLY_BUNDLED_ISL=ON ../llvm && ninja && ninja install
$ cd ..

$ mkdir clang
$ tar xf clang-10.0.0rc1.src.tar.xz --strip-components=1 -C clang

$ mkdir build2 && cd build2
$ cmake -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/inst -DLLVM_LINK_LLVM_DYLIB=ON -DCLANG_LINK_CLANG_DYLIB=ON -DLLVM_MAIN_SRC_DIR=/tmp/foo/llvm -DLLVM_CONFIG=/tmp/inst/bin/llvm-config ../clang && ninja clang

Keruspe · 2020-02-10T10:19:29Z

I wonder if that could be because of ThinLTO.
Will do some more testing and will try to give more pointers towards reproduction, but here the way to reproduce is basically

enable polly option
reinstall llvm package
reinstall clang package

Will try to do that outside of the package manager in a temporary location.

Keruspe · 2020-02-10T14:04:21Z

First test:

I unpacked llvm, polly and clang as you did, but only built clang, using the same command as you, apart from llvm-config (using the system one instead, as llvm 10.0.0_rc1 with polly is installed system-wide here). Got the same error:

ld.lld: error: undefined symbol: getPollyPluginInfo()
>>> referenced by BackendUtil.cpp
>>>               lib/CodeGen/CMakeFiles/obj.clangCodeGen.dir/BackendUtil.cpp.o:((anonymous namespace)::EmitAssemblyHelper::EmitAssemblyWithNewPassManager(clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream, std::__1::default_delete<llvm::raw_pwrite_stream> >))
clang-10: error: linker command failed with exit code 1 (use -v to see invocation)
ninja: build stopped: subcommand failed.

Fwiw, the getPollyPluginInfo() calls are from the included /usr/lib/llvm/10/include/llvm/Support/Extension.def which contains

//extension handlers
HANDLE_EXTENSION(Polly)
#undef HANDLE_EXTENSION

Keruspe · 2020-02-10T14:06:48Z

I think I know why you weren't able to reproduce. When I wanted to reproduce your full scenario, I just noticed that the three cmake params for polly start with -DD instead of just -D and are thus ignored

  Manually-specified variables were not used by the project:

    DLLVM_POLLY_LINK_INTO_TOOLS
    DLLVM_TOOL_POLLY_BUILD
    DPOLLY_BUNDLED_ISL

Keruspe · 2020-02-10T15:19:28Z

Yup, I confirm that with these typo fixed, I can reproduce the problem with your commands

Meinersbur · 2020-02-10T18:18:58Z

The change causing this is probably 24ab9b5. I suspect that -DLLVM_POLLY_LINK_INTO_TOOLS=ON does not cause Polly to be linked into libLLVM, which clang links against. It's a bit unusual that you build clang separately from LLVM (But I think it's still supposed to work). Ping @serge-sans-paille.

Note that -DLLVM_ENABLE_PROJECTS=polly mechanism is newer than -DLLVM_TOOL_POLLY_BUILD=ON . Our primary bug tracker is https://bugs.llvm.org/.

Keruspe · 2020-02-11T08:43:39Z

Building clang separately is at least how all of the source-based distribution do it AFAIK.

Do you want me to resubmit the bug to bugs.llvm.org?

zmodem · 2020-02-11T13:25:53Z

Yup, I confirm that with these typo fixed, I can reproduce the problem with your commands

D'oh! Thanks for confirming.

Do you want me to resubmit the bug to bugs.llvm.org?

Yes, it would be good to have it there so we can track it as a release blocker. I've filed https://bugs.llvm.org/show_bug.cgi?id=44870

Do you have a bugzilla account so I can cc you?

Keruspe · 2020-02-11T13:34:12Z

I don't (yet)

zmodem · 2020-02-11T13:58:57Z

The change causing this is probably 24ab9b5.

Yes, it seems after this change (at 0d27543 which has a build fix), it starts failing to link with

undefined reference to `getPollyPluginInfo()'

However, I'm also not able to build before that. For example at 24ab9b5^1 my build fails with

undefined reference to `polly::initializePollyPasses(llvm::PassRegistry&)'

So maybe I'm still not holding the build quite right. My reproduction command was (from a monorepo checkout):

$ rm -rf /tmp/issue120 && mkdir /tmp/issue120 && cp -r llvm clang /tmp/issue120/ && cp -r polly /tmp/issue120/llvm/tools/ && ( cd /tmp/issue120 && mkdir build1 && cd build1 && cmake -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/inst -DLLVM_LINK_LLVM_DYLIB=ON -DLLVM_BUILD_LLVM_DYLIB=ON -DLLVM_POLLY_LINK_INTO_TOOLS=ON -DLLVM_TOOL_POLLY_BUILD=ON -DPOLLY_BUNDLED_ISL=ON ../llvm && ninja && ninja install && cd .. && mkdir build2 && cd build2 && cmake -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/inst -DLLVM_LINK_LLVM_DYLIB=ON -DCLANG_LINK_CLANG_DYLIB=ON -DLLVM_MAIN_SRC_DIR=/tmp/issue120/llvm -DLLVM_CONFIG=/tmp/inst/bin/llvm-config ../clang && ninja clang )

Keruspe · 2020-02-11T14:08:31Z

For building 9.0.1, we use LLVM_POLLY_BUILD instead of LLVM_TOOL_POLLY_BUILD for llvm (option was renamed for 10), and we add WITH_POLLY=TRUE (which is gone in 10) and LINK_POLLY_INTO_TOOLS=TRUE (which was renamed in 10)

serge-sans-paille · 2020-02-11T15:11:09Z

I managed to reproduce the issue and found the actual problem: the extension mechanism is not aware of the DYLIB process, and linked-in extensions are not selected when selecting all components.

I'm currently trying a patch, if it works I'll submit it for review.

Meinersbur · 2020-02-11T17:21:15Z

The change causing this is probably 24ab9b5.

Yes, it seems after this change (at 0d27543 which has a build fix), it starts failing to link with

undefined reference to `getPollyPluginInfo()'

However, I'm also not able to build before that. For example at 24ab9b5^1 my build fails with

undefined reference to `polly::initializePollyPasses(llvm::PassRegistry&)'

This confirms my suspicion that libPolly is just not added into libLLVM.so (the dylib). Before 24ab9b5, the clang executable pulled-in Polly into the executable by calling initializePollyPasses. After 24ab9b5 it is pulled-in by the registered plugin (which registered getPollyPluginInfo as entry point). Try adding Polly to -DLLVM_DYLIB_COMPONENTS=. If that works, we can solve it my adding registered plugins to the all list (see llvm/cmake/modules/LLVM-Config.cmake).

serge-sans-paille · 2020-02-12T19:21:28Z

https://reviews.llvm.org/D74464 fixes the issue

serge-sans-paille · 2020-02-13T13:22:19Z

Closed by d21664c

Keruspe · 2020-02-13T21:47:42Z

@serge-sans-paille I don't think that's quite enough.

When llvm 10 is installed system-wide and I try to build clang against it, the new added line tries to write to /usr/lib/llvm/10/include/llvm/Support/Extension.def.tmp which it's obviously not allowed to do.
It should only create Extension.def when called from llvm, not when called from clang

serge-sans-paille · 2020-02-14T12:53:19Z

Correct! should be fixed by https://reviews.llvm.org/D74602

Keruspe · 2020-02-14T21:54:01Z

So, just retried the full scenario from above, but using 10.0.0rc2 + applying D74464 and D74602

I still get the initial error:

ld.lld: error: undefined symbol: getPollyPluginInfo()
>>> referenced by BackendUtil.cpp
>>>               lib/CodeGen/CMakeFiles/obj.clangCodeGen.dir/BackendUtil.cpp.o:((anonymous namespace)::EmitAssemblyHelper::EmitAssemblyWithNewPassManager(clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream, std::__1::default_delete<llvm::raw_pwrite_stream> >))

Here is what I ran:

mkdir /tmp/foo
cd /tmp/foo

wget https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0-rc2/llvm-10.0.0rc2.src.tar.xz
wget https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0-rc2/polly-10.0.0rc2.src.tar.xz
wget https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0-rc2/clang-10.0.0rc2.src.tar.xz

mkdir llvm
tar xf llvm-10.0.0rc2.src.tar.xz --strip-components=1 -C llvm
mkdir llvm/tools/polly
tar xf polly-10.0.0rc2.src.tar.xz --strip-components=1 -C llvm/tools/polly/
mkdir clang
tar xf clang-10.0.0rc2.src.tar.xz --strip-components=1 -C clang

curl https://github.com/llvm/llvm-project/commit/d21664cce1db8debe2528f36b1fbd2b8af9c9401.patch | patch -p1
curl https://reviews.llvm.org/file/data/nrlsy72arffezufkqttt/PHID-FILE-lslvurprq7y74emgjb7w/D74602.diff | patch -p1

mkdir build1 && cd build1
cmake -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/inst -DLLVM_LINK_LLVM_DYLIB=ON -DLLVM_BUILD_LLVM_DYLIB=ON   -DLLVM_POLLY_LINK_INTO_TOOLS=ON -DLLVM_TOOL_POLLY_BUILD=ON -DPOLLY_BUNDLED_ISL=ON ../llvm && ninja && ninja install
cd ..

mkdir build2 && cd build2
cmake -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/inst -DLLVM_LINK_LLVM_DYLIB=ON -DCLANG_LINK_CLANG_DYLIB=ON -DLLVM_MAIN_SRC_DIR=/tmp/foo/llvm -DLLVM_CONFIG=/tmp/inst/bin/llvm-config ../clang && ninja clang

Keruspe · 2020-02-14T22:23:35Z

I guess this is because LLVM_COMPILE_EXTENSIONS doesn't contain Polly since add_llvm_pass_plugin from Polly isn't called in standalone clang?

Keruspe · 2020-02-15T06:55:44Z

So, obviously not the right fix, but adding set_property(GLOBAL APPEND PROPERTY LLVM_COMPILE_EXTENSIONS Polly) just before the call to process_llvm_pass_plugins(NO_GEN) allows me to pass -DLLVM_POLLY_LINK_INTO_TOOLS=ON to clang and the build then passes.

Maybe the final value of LLVM_COMPILE_EXTENSIONS should be retrieved somehow with llvm-config or whatever when building standalong clang?

As suggested in #120, don't try to generate the extension file from clang, only do the linking step. Fixes the regression introduced in D74464 when running cmake inside the clang directory. Differential Revision: https://reviews.llvm.org/D74602

serge-sans-paille · 2020-02-15T10:38:08Z

Maybe the final value of LLVM_COMPILE_EXTENSIONS should be retrieved somehow with llvm-config or whatever when building standalong clang?

I like that idea, yeah. Let me dig into it later one. I still commitedD74602 as a small step forward.

serge-sans-paille · 2020-02-18T17:12:46Z

Approach implemented in https://reviews.llvm.org/D74757

Keruspe · 2020-02-19T10:10:20Z

👍 Testing in progress

Keruspe · 2020-02-19T13:15:17Z

Any chance of a backport of these patches to the llvm 10 branch?

serge-sans-paille · 2020-02-19T15:10:34Z

@Keruspe I think it's feasible, @zmodem ?

zmodem · 2020-02-20T14:09:56Z

I'm a little bit worried that this is landing so late in the process, and because it's non-trivial cmake changes it has the potential of breaking lots of people.

On the other hand, the use of Polly is very limited iiuc, so I'm not sure how bad the bug is.

What revisions would need to be ported, and what's your confidence level regarding their correctness?

serge-sans-paille · 2020-02-20T16:40:57Z

I'm a little bit worried that this is landing so late in the process, and because it's non-trivial cmake changes it has the potential of breaking lots of people.

So am I. On the other hand there's no regression on master branch as of now, so we already have some positive feedback :-)

Concerning the revision that need to be ported, I'd say

3a0f6e6
87dac7d

Concerning their correctness, I can't go further than: bots validation is OK, I tested several configuration locally and @Keruspe did too.

Plus this should only impact Polly users, so it should have limited impact.

zmodem · 2020-02-21T09:09:57Z

I looked at the changes again, and decided not to merge them.

It seems to me the risk (cmake breaking somehow) is too high for the reward (standalone clang builds with polly, which isn't widely used).

Maybe it can be a candidate for 10.0.1.

slacka · 2020-02-25T18:35:59Z

@serge-sans-paille
Since @zmodem is averse to the risk of moving forward with the fixes, what are your thoughts on reverting 24ab9b5 for the 10.x series?

serge-sans-paille · 2020-02-26T09:01:40Z

@slacka / @zmodem : considering the multiple patch applied to the original commit, I'd understand if you totally delay it to 10.0.1 or 11.x

zmodem · 2020-02-26T14:00:16Z

@serge-sans-paille
Since @zmodem is averse to the risk of moving forward with the fixes, what are your thoughts on reverting 24ab9b5 for the 10.x series?

Backing out this whole thing would be my preferred approach, but there seem to be too many changes on top to be able to do that cleanly.

As suggested in llvm/llvm-project#120, don't try to generate the extension file from clang, only do the linking step. Fixes the regression introduced in D74464 when running cmake inside the clang directory. Differential Revision: https://reviews.llvm.org/D74602 (cherry picked from commit 87dac7d)

llvmbot · 2022-04-12T12:25:08Z

@llvm/issue-subscribers-polly

As suggested in llvm/llvm-project#120, don't try to generate the extension file from clang, only do the linking step. Fixes the regression introduced in D74464 when running cmake inside the clang directory. Differential Revision: https://reviews.llvm.org/D74602 (cherry picked from commit 7663149)

…624.3 (llvm#120) [release/11.x] Update dependencies from dotnet/arcade

As suggested in llvm/llvm-project#120, don't try to generate the extension file from clang, only do the linking step. Fixes the regression introduced in D74464 when running cmake inside the clang directory. Differential Revision: https://reviews.llvm.org/D74602

This reverts commit 997f33c. Breaks check-mlir ******************** TEST 'MLIR :: IR/attribute.mlir' FAILED ******************** Script: -- : 'RUN: at line 1'; mlir-opt llvm-project/mlir/test/IR/attribute.mlir -split-input-file -verify-diagnostics | /FileCheck llvm-project/mlir/test/IR/attribute.mlir -- Exit Code: 1 Command Output (stderr): -- llvm-project/mlir/test/IR/attribute.mlir split at line llvm#1:19:3: error: unexpected error: 'test.int_attrs' op requires attribute 'index_attr' "test.int_attrs"() { ^ llvm-project/mlir/test/IR/attribute.mlir split at line llvm#120:6:3: error: unexpected error: 'test.int_attrs' op requires attribute 'index_attr' "test.int_attrs"() { ^ llvm-project/mlir/test/IR/attribute.mlir split at line llvm#120:5:6: error: expected error "'si32_attr' failed to satisfy constraint: 32-bit signed integer attribute" was not produced // expected-error @+1 {{'si32_attr' failed to satisfy constraint: 32-bit signed integer attribute}} ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ llvm-project/mlir/test/IR/attribute.mlir split at line llvm#133:5:3: error: unexpected error: 'test.int_attrs' op requires attribute 'index_attr' "test.int_attrs"() { ^ llvm-project/mlir/test/IR/attribute.mlir split at line llvm#133:4:6: error: expected error "'ui32_attr' failed to satisfy constraint: 32-bit unsigned integer attribute" was not produced // expected-error @+1 {{'ui32_attr' failed to satisfy constraint: 32-bit unsigned integer attribute}} ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ llvm-project/mlir/test/IR/attribute.mlir:9:12: error: CHECK: expected string not found in input // CHECK: any_i32_attr = 5 : ui32 ^ <stdin>:3:1: note: scanning from here module { ^ <stdin>:21:28: note: possible intended match here "test.non_negative_int_attr"() {i32attr = 5 : i32, i64attr = 10 : i64} : () -> ()

arsenm · 2023-08-13T19:51:19Z

Old build issue on old release

…or mask operations (llvm#120) * [LLVM][XTHeadVector] support `vbool16/32/64` for vector mask operations * [LLVM][XTHeadVector] correctly expand pseudos for `vmset/vmclr` * [LLVM][XTHeadVector] update corresponding tests * [LLVM][XTHeadVector] update tests for `vmsof/vmsbf/vmsif` * [LLVM][XTHeadVector] update tests for `vmfirst/vmpopc` * [LLVM][XTHeadVector] update tests for `vmfirst/vmpopc`

Conditionally skip docs PR build Change-Id: I36e1db0edf58bc2f4ac95686675b315a23fa232a

…464)" (#120511) This reverts commit 2691b96. This reapply fixes the buildbot breakage of the original patch, by updating clang/test/CodeGen/ubsan-trap-debugloc.c to specify -fsanitize-merge (the default, which is merge, is applied by the driver but not clang_cc1). This reapply also expands clang/test/CodeGen/ubsan-trap-merge.c. ---- Original commit message: '-mllvm -ubsan-unique-traps' (#65972) applies to all UBSan checks. This patch introduces -fsanitize-merge (defaults to on, maintaining the status quo behavior) and -fno-sanitize-merge (equivalent to '-mllvm -ubsan-unique-traps'), with the option to selectively applying non-merged handlers to a subset of UBSan checks (e.g., -fno-sanitize-merge=bool,enum). N.B. we do not use "trap" in the argument name since #119302 has generalized -ubsan-unique-traps to work for non-trap modes (min-rt and regular rt). This patch does not remove the -ubsan-unique-traps flag; that will override -f(no-)sanitize-merge.

serge-sans-paille closed this as completed Feb 13, 2020

serge-sans-paille reopened this Feb 15, 2020

DieGoldeneEnte mentioned this issue Feb 25, 2020

llvmPackages_10: init at rc2 NixOS/nixpkgs#78482

Merged

12 tasks

Meinersbur mentioned this issue Mar 6, 2020

llvmPackages_10: rc2 -> rc3 NixOS/nixpkgs#81819

Merged

10 tasks

RKSimon added build-problem polly labels Apr 12, 2022

doocman mentioned this issue Jun 5, 2022

clang-cl.exe (windows) crash in front end #55879

Open

mmitche pushed a commit to mmitche/llvm-project that referenced this issue Aug 3, 2022

Update dependencies from https://github.com/dotnet/arcade build 20210…

51c3228

…624.3 (llvm#120) [release/11.x] Update dependencies from dotnet/arcade

Bouncner mentioned this issue Jun 12, 2023

[Clang-Tidy] misc-const-correctness crash due to seemingly infinite recursion #60895

Closed

arsenm closed this as not planned Won't fix, can't repro, duplicate, stale Aug 13, 2023

EugeneZelenko added the obsolete Issues with old (unsupported) versions of LLVM label Aug 13, 2023

nilres mentioned this issue May 28, 2024

Crash when expanding over members_of bloomberg/clang-p2996#54

Closed

yxsamliu pushed a commit to yxsamliu/llvm-project that referenced this issue Sep 1, 2024

Merge pull request llvm#120 from samjwu/rtd-condition

24f9282

Conditionally skip docs PR build Change-Id: I36e1db0edf58bc2f4ac95686675b315a23fa232a

usx95 mentioned this issue Sep 6, 2024

[clang] Crash in TreeTransform<TemplateInstantiator>::TransformExpr with invalid expr #107560

Closed

piyushhingu mentioned this issue Apr 2, 2025

optimization failed for llvm IR using mlir-opt, mlir-translate, opt #134041

Open

damyanp mentioned this issue Apr 1, 2025

[milestone] Compile a broad selection of compute shaders llvm/wg-hlsl#11

Open

53 tasks

clang 10.0.0_rc1 standalone fails to build with polly #120

clang 10.0.0_rc1 standalone fails to build with polly #120

Comments

Keruspe commented Feb 9, 2020

zmodem commented Feb 9, 2020

Keruspe commented Feb 9, 2020

zmodem commented Feb 10, 2020

Keruspe commented Feb 10, 2020

Keruspe commented Feb 10, 2020

Keruspe commented Feb 10, 2020

Keruspe commented Feb 10, 2020

Meinersbur commented Feb 10, 2020 • edited Loading

Keruspe commented Feb 11, 2020

zmodem commented Feb 11, 2020

Keruspe commented Feb 11, 2020

zmodem commented Feb 11, 2020

Keruspe commented Feb 11, 2020

serge-sans-paille commented Feb 11, 2020

Meinersbur commented Feb 11, 2020

serge-sans-paille commented Feb 12, 2020

serge-sans-paille commented Feb 13, 2020

Keruspe commented Feb 13, 2020

serge-sans-paille commented Feb 14, 2020

Keruspe commented Feb 14, 2020 • edited Loading

Keruspe commented Feb 14, 2020

Keruspe commented Feb 15, 2020

serge-sans-paille commented Feb 15, 2020

serge-sans-paille commented Feb 18, 2020

Keruspe commented Feb 19, 2020

Keruspe commented Feb 19, 2020

serge-sans-paille commented Feb 19, 2020

zmodem commented Feb 20, 2020

serge-sans-paille commented Feb 20, 2020

zmodem commented Feb 21, 2020

slacka commented Feb 25, 2020

serge-sans-paille commented Feb 26, 2020

zmodem commented Feb 26, 2020

llvmbot commented Apr 12, 2022

arsenm commented Aug 13, 2023

Meinersbur commented Feb 10, 2020 •

edited

Loading

Keruspe commented Feb 14, 2020 •

edited

Loading