https://bugs.llvm.org/show_bug.cgi?id=50692
Bug ID: 50692
Summary: OpenMP Cuda Offloading Failure with LLVM 12.0.0 (<eof>
parser at end of file error)
Product: OpenMP
Version: unspecified
Hardware: PC
OS: All
Status: NEW
Severity: enhancement
Priority: P
Component: Clang Compiler Support
Assignee: unassignedclangb...@nondot.org
Reporter: asher.mancine...@pnnl.gov
CC: llvm-bugs@lists.llvm.org
With Tesla V100-SXM3 cards and the following source file:
> // example.c
> #include <stdio.h>
>
> #define MAX 32768
>
> int main()
> {
> int x[MAX], i;
> for (i = 0; i < MAX; ++i) {
> x[i] = 42;
> }
>
> #pragma omp target teams distribute parallel for simd map(tofrom: x[0:MAX])
> for (i = 0; i < MAX; ++i) {
> x[i] = x[i] + 1;
> }
> printf("After the target region is executed, x[0] = %d\n", x[0]);
> return 0;
> }
compiled with the following:
> clang --cuda-path=/... \
> -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda \
> -Xopenmp-target -march=sm_70 -o example example.c
Yields the following:
> Program aborted due to an unhandled Error:
> Invalid record (Producer: LLVM13.0.0git Reader: LLVM 12.0.0)
> PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash
> backtrace, preprocessed source, and associated run script.
> Stack dump:
> 0.Program arguments:
> /qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12 -cc1 -triple
> nvptx64-nvidia-cuda -aux-triple x86_64-unknown-linux-gnu -S -disable-free
> -main-file-name example.c -mrelocation-model pic -pic-level 2
> -fhalf-no-semantic-interposition -mframe-pointer=all -fno-rounding-math
> -fno-verbose-asm -no-integrated-as -mlink-builtin-bitcode
> /share/apps/cuda/10.1.243/nvvm/libdevice/libdevice.10.bc -target-feature
> +ptx64 -target-sdk-version=10.1 -mlink-builtin-bitcode
> /qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/lib/libomptarget-nvptx-cuda_101-sm_70.bc
> -target-cpu sm_70 -fno-split-dwarf-inlining -debugger-tuning=gdb
> -resource-dir
> /qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/lib/clang/12.0.0
> -internal-isystem
> /qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/lib/clang/12.0.0/include/openmp_wrappers
> -include __clang_openmp_device_functions.h -internal-isystem
> /usr/local/include -internal-isystem
> /qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/lib/clang/12.0.0/include
> -internal-externc-isystem /include -internal-externc-isystem /usr/include
> -internal-isystem /usr/local/include -internal-isystem
> /qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/lib/clang/12.0.0/include
> -internal-externc-isystem /include -internal-externc-isystem /usr/include
> -fno-dwarf-directory-asm -fdebug-compilation-dir
> /people/manc568/workspace/build-llvm-gcc8-tonga-scratch -ferror-limit 19
> -fopenmp -fopenmp-cuda-parallel-target-regions -fgnuc-version=4.2.1
> -fcolor-diagnostics -fopenmp-is-device -fopenmp-host-ir-file-path
> /tmp/example-1faaa0.bc -o /tmp/example-6d6124.s -x c example.c
> 1.<eof> parser at end of file
> #0 0x0000000001c1a643 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1c1a643)
> #1 0x0000000001c1844e llvm::sys::RunSignalHandlers()
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1c1844e)
> #2 0x0000000001c1ab0f SignalHandler(int)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1c1ab0f)
> #3 0x00002aefd9df2630 __restore_rt (/usr/lib64/libpthread.so.0+0xf630)
> #4 0x00002aefdaf4a387 raise (/usr/lib64/libc.so.6+0x36387)
> #5 0x00002aefdaf4ba78 abort (/usr/lib64/libc.so.6+0x37a78)
> #6 0x0000000001ba0ef0 llvm::Error::fatalUncheckedError() const
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1ba0ef0)
> #7 0x0000000002c2b8d0
> llvm::Expected<llvm::Constant*>::fatalUncheckedExpected() const
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2c2b8d0)
> #8 0x0000000002c2ac8d (anonymous
> namespace)::IRLinker::materialize(llvm::Value*, bool)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2c2ac8d)
> #9 0x0000000001d66c5d (anonymous namespace)::Mapper::mapValue(llvm::Value
> const*)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1d66c5d)
> #10 0x0000000001d67ca2 (anonymous
> namespace)::Mapper::remapInstruction(llvm::Instruction*)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1d67ca2)
> #11 0x0000000001d684c1 (anonymous
> namespace)::Mapper::remapFunction(llvm::Function&)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1d684c1)
> #12 0x0000000001d6618b (anonymous
> namespace)::FlushingMapper::~FlushingMapper()
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1d6618b)
> #13 0x0000000001d66b10 llvm::ValueMapper::mapValue(llvm::Value const&)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x1d66b10)
> #14 0x0000000002c25e37 (anonymous namespace)::IRLinker::run()
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2c25e37)
> #15 0x0000000002c241f2 llvm::IRMover::move(std::unique_ptr<llvm::Module,
> std::default_delete<llvm::Module> >, llvm::ArrayRef<llvm::GlobalValue*>,
> std::function<void (llvm::GlobalValue&, std::function<void
> (llvm::GlobalValue&)>)>, bool)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2c241f2)
> #18 0x0000000002a79ede clang::BackendConsumer::LinkInModules()
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2a79ede)
> #19 0x0000000002a78f82
> clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2a78f82)
> #20 0x0000000003533184 clang::ParseAST(clang::Sema&, bool, bool)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x3533184)
> #21 0x0000000002471c30 clang::FrontendAction::Execute()
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2471c30)
> #22 0x00000000023f34ea
> clang::CompilerInstance::ExecuteAction(clang::FrontendAction&)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x23f34ea)
> #23 0x0000000002527bb4
> clang::ExecuteCompilerInvocation(clang::CompilerInstance*)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x2527bb4)
> #24 0x000000000096c827 cc1_main(llvm::ArrayRef<char const*>, char const*,
> void*)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x96c827)
> #25 0x000000000096a3a2 ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&)
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x96a3a2)
> #26 0x000000000096a0bd main
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x96a0bd)
> #27 0x00002aefdaf36555 __libc_start_main (/usr/lib64/libc.so.6+0x22555)
> #28 0x000000000096704d _start
> (/qfs/projects/ops/rh7_gpu/llvm/12.0.0-cuda10.1.243/bin/clang-12+0x96704d)
> clang-12: error: unable to execute command: Aborted (core dumped)
> clang-12: error: clang frontend command failed due to signal (use -v to see
> invocation)
> clang version 12.0.0 (g...@github.com:llvm/llvm-project.git
> d28af7c654d8db0b68c175db5ce212d74fb5e9bc)
> Target: x86_64-unknown-linux-gnu
> Thread model: posix
> InstalledDir: /share/apps/llvm/12.0.0-cuda10.1.243/bin
> clang-12: note: diagnostic msg: Error generating preprocessed source(s).
> make: *** [all] Error 25
Is this a user error? These examples are from [NERSC documentation on OMP
offloading](https://docs-dev.nersc.gov/cgpu/examples/), and anything that used
the omp `target` pragma resulted in the same error.
I built LLVM git tag llvmorg-12.0.0 with the following options:
> CC=clang CXX=clang++ \
> cmake \
> -GNinja \
>
> -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra;libcxx;libunwind;libcxxabi;lld;openmp;compiler-rt"
> \
> -DLLVM_TARGETS_TO_BUILD="X86;NVPTX" \
> -DCLANG_OPENMP_NVPTX_DEFAULT_ARCH=sm_70 \
> -DLIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES=37,60,70 \
> -DCMAKE_BUILD_TYPE=Release \
> -DLLVM_ENABLE_ASSERTIONS=ON \
> -DFLANG_BUILD_NEW_DRIVER=ON \
> -DCMAKE_INSTALL_PREFIX=$final_install \
> -DCMAKE_CXX_FLAGS='--gcc-toolchain=/share/apps/gcc/9.1.0' \
> -DLIBCXX_CXX_ABI=libcxxabi \
> -DLIBCXX_USE_COMPILER_RT=On \
> -DLIBCXXABI_USE_LLVM_UNWINDER=On \
> -DLIBCXXABI_USE_COMPILER_RT=On \
> -DLIBUNWIND_USE_COMPILER_RT=On \
> -DCLANG_DEFAULT_CXX_STDLIB=libc++ \
> -DCLANG_DEFAULT_RTLIB=compiler-rt \
> -DCLANG_DEFAULT_LINKER=lld \
> $srcdir/llvm
I built llvmorg-12.0.0 with this clang:
> $ clang --version
> clang version 13.0.0 (g...@github.com:llvm/llvm-project.git
> 7714b405a0de47e461c77fa8dbd2c21f0d34bbf2)
> Target: x86_64-unknown-linux-gnu
> Thread model: posix
> InstalledDir: /...
and GCC 9.1.0 and CUDA 10.1.243.
Resulting clang install:
> $ clang --version
> clang version 12.0.0 (g...@github.com:llvm/llvm-project.git
> d28af7c654d8db0b68c175db5ce212d74fb5e9bc)
> Target: x86_64-unknown-linux-gnu
> Thread model: posix
> InstalledDir: /...
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
