search

ye-luo / openmp-target

OpenMP offload playground
Other
10 stars 6 forks source link

math of complex number in C++ and fortran #3

Open ye-luo opened 4 years ago

ye-luo commented 4 years ago

Test files can be found in https://github.com/ye-luo/openmp-target/tree/master/hands-on/tests/complex

So far XL (xlC, xlf) and GNU (g++, gfortran) pass.

Clang on Power9+V100, bug report

[yeluo@login3.summit complex]$ clang++ -g -O3 -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda -ffast-math complex.cpp 
[yeluo@login3.summit complex]$ ./a.out 
wrong a * b (0,0), correct value (-0.3,0.5)
[yeluo@login3.summit complex]$ clang++ -g -O3 -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda complex.cpp 
nvlink error   : Undefined reference to '__mulsc3' in '/tmp/complex-5ceac3.cubin'
clang-11: error: nvlink command failed with exit code 255 (use -v to see invocation)

AOMP on Radeon VII 0.7-6 compiler crashes at compilation.

ye-luo commented 4 years ago

AOMP crash message

llc: /home/estewart/git/aomp/llvm-project/llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp:269: bool llvm::DwarfExpression::addMachineRegExpression(const llvm::TargetRegisterInfo&, llvm::DIExpressionCursor&, unsigned int, unsigned int): Assertion `DwarfRegs.size() == 1' failed.
Stack dump:
0.  Program arguments: /usr/lib/aomp_0.7-6/bin/llc /tmp/complex-gfx906-3995ea-gfx906-optimized-aec56f.bc -mtriple=amdgcn-amd-amdhsa -filetype=obj -mcpu=gfx906 -o /tmp/complex-gfx906-3995ea-gfx906-2a8bf6.o 
1.  Running pass 'CallGraph Pass Manager' on module '/tmp/complex-gfx906-3995ea-gfx906-optimized-aec56f.bc'.
2.  Running pass 'AMDGPU Assembly Printer' on function '@__omp_offloading_10304_2920c9c_main_l7'
 #0 0x00005616ab744c4a llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/usr/lib/aomp_0.7-6/bin/llc+0x1900c4a)
 #1 0x00005616ab7428f4 llvm::sys::RunSignalHandlers() (/usr/lib/aomp_0.7-6/bin/llc+0x18fe8f4)
 #2 0x00005616ab742a32 (/usr/lib/aomp_0.7-6/bin/llc+0x18fea32)
 #3 0x00007fbd811dd890 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x12890)
 #4 0x00007fbd7fe8ee97 raise /build/glibc-OTsEL5/glibc-2.27/signal/../sysdeps/unix/sysv/linux/raise.c:51:0
 #5 0x00007fbd7fe90801 abort /build/glibc-OTsEL5/glibc-2.27/stdlib/abort.c:81:0
 #6 0x00007fbd7fe8039a __assert_fail_base /build/glibc-OTsEL5/glibc-2.27/assert/assert.c:89:0
 #7 0x00007fbd7fe80412 (/lib/x86_64-linux-gnu/libc.so.6+0x30412)
 #8 0x00005616aab94e7f (/usr/lib/aomp_0.7-6/bin/llc+0xd50e7f)
 #9 0x00005616aab7beb7 llvm::DwarfDebug::emitDebugLocValue(llvm::AsmPrinter const&, llvm::DIBasicType const*, llvm::DbgValueLoc const&, llvm::DwarfExpression&) (/usr/lib/aomp_0.7-6/bin/llc+0xd37eb7)
#10 0x00005616aab7c1eb llvm::DebugLocEntry::finalize(llvm::AsmPrinter const&, llvm::DebugLocStream::ListBuilder&, llvm::DIBasicType const*, llvm::DwarfCompileUnit&) (/usr/lib/aomp_0.7-6/bin/llc+0xd381eb)
#11 0x00005616aab90842 llvm::DwarfDebug::collectEntityInfo(llvm::DwarfCompileUnit&, llvm::DISubprogram const*, llvm::DenseSet<std::pair<llvm::DINode const*, llvm::DILocation const*>, llvm::DenseMapInfo<std::pair<llvm::DINode const*, llvm::DILocation const*> > >&) (/usr/lib/aomp_0.7-6/bin/llc+0xd4c842)
#12 0x00005616aab90cbb llvm::DwarfDebug::endFunctionImpl(llvm::MachineFunction const*) (/usr/lib/aomp_0.7-6/bin/llc+0xd4ccbb)
#13 0x00005616aabda1da llvm::DebugHandlerBase::endFunction(llvm::MachineFunction const*) (/usr/lib/aomp_0.7-6/bin/llc+0xd961da)
#14 0x00005616aab6c41b llvm::AsmPrinter::EmitFunctionBody() (/usr/lib/aomp_0.7-6/bin/llc+0xd2841b)
#15 0x00005616aa432c18 llvm::AMDGPUAsmPrinter::runOnMachineFunction(llvm::MachineFunction&) (/usr/lib/aomp_0.7-6/bin/llc+0x5eec18)
#16 0x00005616aad3ef4e llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/usr/lib/aomp_0.7-6/bin/llc+0xefaf4e)
#17 0x00005616ab0f5ed9 llvm::FPPassManager::runOnFunction(llvm::Function&) (/usr/lib/aomp_0.7-6/bin/llc+0x12b1ed9)
#18 0x00005616aa9ce113 (/usr/lib/aomp_0.7-6/bin/llc+0xb8a113)
#19 0x00005616ab0f5131 llvm::legacy::PassManagerImpl::run(llvm::Module&) (/usr/lib/aomp_0.7-6/bin/llc+0x12b1131)
#20 0x00005616aa3709d5 main (/usr/lib/aomp_0.7-6/bin/llc+0x52c9d5)
#21 0x00007fbd7fe71b97 __libc_start_main /build/glibc-OTsEL5/glibc-2.27/csu/../csu/libc-start.c:344:0
#22 0x00005616aa3c92aa _start (/usr/lib/aomp_0.7-6/bin/llc+0x5852aa)
clang-9: error: unable to execute command: Aborted (core dumped)
clang-9: error: amdgcn-link command failed due to signal (use -v to see invocation)
clang version 9.0.1 (https://github.com/ROCm-Developer-Tools/llvm-project 2d8529565d71c80db00079ea80d37c5814f9ff25)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /usr/lib/aomp/bin
clang-9: note: diagnostic msg: PLEASE open a git issue at https://github.com/ROCm-Developer-Tools/aomp with a detailed description of this problem.
clang-9: note: diagnostic msg: Error generating preprocessed source(s).