hpcpony
commented
1 year ago Tried to build the same LLVM on a different host (Red Hat 8.7) using the same cmake and same compiler (exactly the same off a shared NFS mount) as before. This attempt seems to get through the problematic section of the build.
Aside from being Red Hat 8.7 vs. CentOS 7.9 this machine does have GPU cards and the CUDA software installed (though I'm not sure if that's significant).
Anyway, maybe this will help someone see what the problem is.
cmake version 3.23.2 GNU bash, version 4.4.20(1)-release (x86_64-redhat-linux-gnu) GNU Make 4.2.1 Red Hat Enterprise Linux release 8.7 (Ootpa) Linux host2 4.18.0-425.3.1.el8.x86_64 #1 SMP Fri Sep 30 11:45:06 EDT 2022 x86_64 x86_64 x86_64 GNU/Linux
[user@host2 build.host2.nocuda]$ cmake \
? -DLLVM_TARGETS_TO_BUILD=X86 \
? -DCMAKE_C_COMPILER=${HOST_GCC}/bin/gcc \
? -DCMAKE_CXX_COMPILER=${HOST_GCC}/bin/g++ \
? -DCMAKE_BUILD_TYPE="Release" \
? -DLIBOMP_ARCH="x86_64" \
? -DLIBOMPTARGET_BUILD_AMDGPU_PLUGIN="OFF" -DLIBOMPTARGET_BUILD_CUDA_PLUGIN="OFF" \
? -DCMAKE_INSTALL_PREFIX=${INSTALL_PREFIX} \
? -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra;cross-project-tests;flang;libclc;lld;lldb;mlir;openmp;polly;pstl" \
? -DLLVM_ENABLE_RUNTIMES="compiler-rt;libc;libcxx;libcxxabi;libunwind" \
? -DGCC_INSTALL_PREFIX=${HOST_GCC} \
? -DCMAKE_CXX_LINK_FLAGS="-L${HOST_GCC}/lib64 -Wl,-rpath,${HOST_GCC}/lib64" \
? ../llvm
-- The C compiler identification is GNU 11.3.0
-- The CXX compiler identification is GNU 11.3.0
.
.
.
-- Found LIBOMPTARGET_DEP_LIBFFI: /usr/lib64/libffi.so
-- Found LIBOMPTARGET_DEP_CUDA_DRIVER: /usr/lib64/libcuda.so
-- Could NOT find LIBOMPTARGET_DEP_VEO (missing: LIBOMPTARGET_DEP_VEO_LIBRARIES LIBOMPTARGET_DEP_VEOSINFO_LIBRARIES LIBOMPTARGET_DEP_VEO_INCLUDE_DIRS)
-- OMPT target enabled
-- OpenMP tools dir in libomptarget: /scratch/src/llvm-project-16.0.0.src/build.host2.nocuda/projects/openmp/runtime/src
-- LIBOMPTARGET: Building offloading runtime library libomptarget.
-- LIBOMPTARGET: Not building aarch64 offloading plugin: machine not found in the system.
-- LIBOMPTARGET: Not building AMDGPU offloading plugin: LIBOMPTARGET_BUILD_AMDGPU_PLUGIN is false
-- LIBOMPTARGET: Not building CUDA offloading plugin: LIBOMPTARGET_BUILD_CUDA_PLUGIN is false
-- LIBOMPTARGET: Not building PPC64 offloading plugin: machine not found in the system.
-- LIBOMPTARGET: Not building PPC64le offloading plugin: machine not found in the system.
-- LIBOMPTARGET: Not building nec-aurora plugin: libveo or libveosinfo not found.
-- LIBOMPTARGET: Building x86_64 offloading plugin.
-- LIBOMPTARGET: Not building aarch64 NextGen offloading plugin: machine not found in the system.
-- LIBOMPTARGET: Not building AMDGPU NextGen offloading plugin: LIBOMPTARGET_BUILD_AMDGPU_PLUGIN is false
-- LIBOMPTARGET: Not building CUDA NextGen offloading plugin: LIBOMPTARGET_BUILD_CUDA_PLUGIN is false
-- LIBOMPTARGET: Not building PPC64 NextGen offloading plugin: machine not found in the system.
-- LIBOMPTARGET: Not building PPC64le NextGen offloading plugin: machine not found in the system.
-- LIBOMPTARGET: Building x86_64 NextGen offloading plugin.
-- LIBOMPTARGET: Building DeviceRTL. Using clang from in-tree build
-- LIBOMPTARGET: Building the llvm-omp-device-info tool
-- LIBOMPTARGET: Building the llvm-omp-kernel-replay tool
.
.
.
[ 33%] Built target omptarget-nvptx-sm_37-bc
Scanning dependencies of target omptarget-nvptx-sm_50-bc
[ 33%] Building LLVM bitcode Configuration.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Debug.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Kernel.cpp-sm_50.bc
[ 33%] Building LLVM bitcode LibC.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Mapping.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Misc.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Parallelism.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Reduction.cpp-sm_50.bc
[ 33%] Building LLVM bitcode State.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Synchronization.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Tasking.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Utils.cpp-sm_50.bc
[ 33%] Building LLVM bitcode Workshare.cpp-sm_50.bc
[ 33%] Linking LLVM bitcode libomptarget-nvptx-sm_50.bc
[ 33%] Internalizing LLVM bitcode libomptarget-nvptx-sm_50.bc
[ 33%] Optimizing LLVM bitcode libomptarget-nvptx-sm_50.bc
[ 33%] Built target omptarget-nvptx-sm_50-bc
Scanning dependencies of target omptarget-nvptx-sm_52-bc
[ 33%] Building LLVM bitcode Configuration.cpp-sm_52.bc
[ 33%] Building LLVM bitcode Debug.cpp-sm_52.bc
.
.
.
bjodah
jhuber6
Yaraslaut
bryanpkc
ahrbe1
llvmbot
fyrie6870
tardigradus
I'm getting Segmentation faults trying to build 16.0.0 from source.
If I try to run the "opt" line by hand I do not get a Segmentation fault.
I tried this 4 different times (select combinations of 2 different hosts, 2 different compiler gcc 11.3.0, 12.2.0, with and without LIBOMPTARGET_BUILD_X). They all get the Segmentation fault, but it happens on different versions of libomptarget-nvptx-sm_X.bc (X = 35, 37, 50, 70).
I'd be happy to just turn off anything related to NVidia (and GPUs in general) because I don't have any of that hardware or software anyway.
OS: CentOS 7.9.2009 gcc: 11.3.0 cmake: 3.23.2 /bin/sh: 4.2.46(2)-release gmake: 3.82