Closed Meinersbur closed 2 years ago
@llvm/issue-subscribers-mlir-core
@joker-eph I "assigned" the bug to you, not sure what the equivalent of adding someone's email to the CC field in bugzilla is. Maybe just mention someone via @
in comment, but did not know what to write. Well, now I do.
Thanks! Our gcc5 and gcc10 bots are happy, so something seems specific to these particular gcc versions.
I built locally with gcc 11.2 and gcc 9.4 and I can't reproduce right now. There has to be something else involved.
Invocation looks like:
g++-9 -DBUILD_EXAMPLES -DGTEST_HAS_RTTI=0 -DMLIR_CUDA_CONVERSIONS_ENABLED=1 -DMLIR_INCLUDE_TESTS -DMLIR_ROCM_CONVERSIONS_ENABLED=0 -D_DEBUG -D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -Itools/mlir/unittests/Support -Imlir/unittests/Support -Iinclude -Illvm/include -Imlir/include -Itools/mlir/include -Illvm/utils/unittest/googletest/include -Illvm/utils/unittest/googlemock/include -fPIC -fno-semantic-interposition -fvisibility-inlines-hidden -Werror=date-time -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wno-missing-field-initializers -pedantic -Wno-long-long -Wimplicit-fallthrough -Wno-maybe-uninitialized -Wno-class-memaccess -Wno-redundant-move -Wno-pessimizing-move -Wno-noexcept-type -Wdelete-non-virtual-dtor -Wsuggest-override -Wno-comment -Wmisleading-indentation -fdiagnostics-color -ffunction-sections -fdata-sections -O3 -DNDEBUG -Wno-variadic-macros -fno-exceptions -fno-rtti -UNDEBUG -Wno-suggest-override -std=c++14 -MD -MT tools/mlir/unittests/Support/CMakeFiles/MLIRSupportTests.dir/DebugCounterTest.cpp.o -MF tools/mlir/unittests/Support/CMakeFiles/MLIRSupportTests.dir/DebugCounterTest.cpp.o.d -o tools/mlir/unittests/Support/CMakeFiles/MLIRSupportTests.dir/DebugCounterTest.cpp.o -c mlir/unittests/Support/DebugCounterTest.cpp
Tried to run the invocation of the bot, this is only a C++17 problem apparently : only flang hits this because the project builds in c++14 mode by default.
We build with C++17 at Google, but we're using Clang, this is likely why we haven't hit this right now.
What is the outlook to get this fixed?
Someone will have to prioritize trying to debug this, unfortunately this isn't a combination of a config that most core developers are focusing on at the moment I think.
Maybe if you can bisect to the pointer when the bug was introduced that could help by hinting more at the issue?
Bijection points to 4fe5cfe53e86dbfb4ad4ad05f7c19c0929433e16, which unfortunately does not help to identify the cause.
@gkistanova contacted me to move the flang-x86_64-knl-linux builder to staging because of its ongoing fail status.
Cause is 72d5afa4acc3cb7fdeb6ead90cc8f446a48c1f9d where DebugActionTest.cpp
and DebugAction.h
were first introduced. Because of the reasons mentioned in 4fe5cfe53e86dbfb4ad4ad05f7c19c0929433e16, the unittests never ran. This is also the reason why D116027 was already once reverted in 2132906836cf0618e76485c67a60305bf1557ffc.
It complains about an ambiguity of the following two functions:
template <typename T>
static TypeID get();
template <template <typename> class Trait>
static TypeID get();
which were introduced in https://reviews.llvm.org/D77768. The second indeed looks redundant to me (any template-instantiated class should also match template T
). After removing it, unfortunately it again fails with gcc.
This is also the reason why D116027 was already once reverted in https://github.com/llvm/llvm-project/commit/2132906836cf0618e76485c67a60305bf1557ffc.
The revert linked here was bogus in itself, and the commit was re-landed as-is if I followed correctly.
MLIR fails to compile on the alcf-theta-flang. The builder was unavailble for a while, the problem must have been introduced during its downtime. Other buildbots still green seem to use either clang or msvc.
This occurs using gcc 9.3 as well as gcc 11.2
https://lab.llvm.org/buildbot/#/builders/132/builds/7871
https://lab.llvm.org/buildbot/#/builders/132/builds/7895