AMD bug during linking

[AuroraPerego]

When compiling the SYCL version of pixeltrack-standalone for the AMD backend, it crashes during linking.
Command:

/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/bin/clang++ -O3 -fsycl -std=c++17 -O2 -fPIC -fdiagnostics-show-option -felide-constructors -fmessage-length=0 -fno-math-errno -ftree-vectorize -fvisibility-inlines-hidden -msse3 -pipe -pthread -Werror=address -Wall -Werror=array-bounds -Wno-attributes -Werror=conversion-null -Werror=delete-non-virtual-dtor -Wno-deprecated -Werror=format -Wno-long-long -Werror=main -Werror=missing-braces -Werror=narrowing -Wnon-virtual-dtor -Werror=overflow -Werror=overlength-strings -Wparentheses -Werror=pointer-arith -Wno-psabi -Werror=reorder -Wreturn-type -Werror=return-type -Werror=sign-compare -Werror=strict-aliasing -Wstrict-overflow -Werror=switch -Werror=type-limits -Wunused -Wno-unused-local-typedefs -Werror=unused-value -Wno-error=unused-variable -Wno-vla -Werror=write-strings -Wfatal-errors -g  -fsycl-targets=amdgcn-amd-amdhsa -Xsycl-target-backend --offload-arch=gfx900 --rocm-path=/opt/rocm-5.0.2 -Wno-linker-warnings  \ 
   /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/obj/sycl/plugin-PixelTriplets/CAHitNtupletGeneratorKernelsAlloc.cc.o /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/obj/sycl/plugin-PixelTriplets/CAHitNtupletGeneratorOnGPU.cc.o /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/obj/sycl/plugin-PixelTriplets/HelixFitOnGPU.cc.o /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/obj/sycl/plugin-PixelTriplets/RiemannFitOnGPU.cc.o /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/obj/sycl/plugin-PixelTriplets/BrokenLineFitOnGPU.cc.o /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/obj/sycl/plugin-PixelTriplets/CAHitNtupletGeneratorKernels.cc.o /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/obj/sycl/plugin-PixelTriplets/CAHitNtupletSYCL.cc.o \
   -O2 -fPIC -pthread -Wl,-E -lstdc++fs -ldl -shared -Wl,-z,defs -L/afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/lib/sycl -lFramework -lSYCLCore -lSYCLDataFormats -lCondFormats -lDataFormats -L/afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/external/tbb/lib -ltbb  -L/afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/external/boost/lib -L/afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/external/libbacktrace/lib -lbacktrace \
   -o /afs/cern.ch/user/a/aperego/public/pixeltrack-standalone/lib/sycl/pluginPixelTriplets.so

Backtrace:

lld: /data/user/fwyzard/sycl/llvm/llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp:532: void {anonymous}::SIOptimizeVGPRLiveRange::optimizeLiveRange(llvm::Register, llvm::MachineBasicBlock*, llvm::MachineBasicBlock*, llvm::MachineBasicBlock*, llvm::SmallSetVector<llvm::MachineBasicBlock*, 16>&) const: Assertion `!O.readsReg()' failed.
PLEASE submit a bug report to https://github.com/intel/llvm/issues and include the crash backtrace.
Stack dump:
0.      Program arguments: /cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/bin/lld -flavor gnu -m elf64_amdgpu --no-undefined -shared -plugin-opt=-amdgpu-internalize-symbols -plugin-opt=mcpu=gfx900 -plugin-opt=O2 --lto-CGO2 -o /tmp/aperego/CAHitNtupletGeneratorKernelsAlloc-gfx900-4cc973-50fdf9.out /tmp/aperego/CAHitNtupletGeneratorKernelsAlloc-gfx900-6a6965-5bc87f.o
1.      Running pass 'CallGraph Pass Manager' on module 'ld-temp.o'.
2.      Running pass 'SI Optimize VGPR LiveRange' on function '@_ZN4Rfit10Circle_fitIN5Eigen5BlockINS1_3MapINS1_6MatrixIdLi3ELi3ELi0ELi3ELi3EEELi0ENS1_6StrideILi73728ELi24576EEEEELin1ELin1ELb0EEENS3_INS4_IdLi4ELi1ELi0ELi4ELi1EEELi0ENS1_11InnerStrideILi24576EEEEELi3EEENS_10circle_fitERKT_RKNS4_IdXmlLi2ET1_EXmlLi2ET1_EXorLNS1_14StorageOptionsE0EquaaeqmlLi2ET1_Li1EnemlLi2ET1_Li1ELSI_1EquaaeqmlLi2ET1_Li1EnemlLi2ET1_Li1ELSI_0ELSI_0EEXmlLi2ET1_EXmlLi2ET1_EEERKT0_RKNS4_IdXT1_ELi1EXorLSI_0EquaaeqT1_Li1EneLi1ELi1ELSI_1EquaaeqLi1ELi1EneT1_Li1ELSI_0ELSI_0EEXT1_ELi1EEEdb'
 #0 0x00007fc06e2cf8eb llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/libLLVMSupport.so.17git+0x1b88eb)
 #1 0x00007fc06e2cd2f4 SignalHandler(int) Signals.cpp:0:0
 #2 0x00007fc06e888cf0 __restore_rt (/lib64/libpthread.so.0+0x12cf0)
 #3 0x00007fc06d470aff raise (/lib64/libc.so.6+0x4eaff)
 #4 0x00007fc06d443ea5 abort (/lib64/libc.so.6+0x21ea5)
 #5 0x00007fc06d443d79 _nl_load_domain.cold.0 (/lib64/libc.so.6+0x21d79)
 #6 0x00007fc06d469456 (/lib64/libc.so.6+0x47456)
 #7 0x00007fc06bbe1e2b (anonymous namespace)::SIOptimizeVGPRLiveRange::optimizeLiveRange(llvm::Register, llvm::MachineBasicBlock*, llvm::MachineBasicBlock*, llvm::MachineBasicBlock*, llvm::SmallSetVector<llvm::MachineBasicBlock*, 16u>&) const SIOptimizeVGPRLiveRange.cpp:0:0
 #8 0x00007fc06bbe749d (anonymous namespace)::SIOptimizeVGPRLiveRange::runOnMachineFunction(llvm::MachineFunction&) (.part.0) SIOptimizeVGPRLiveRange.cpp:0:0
 #9 0x00007fc069259ec5 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (.part.0) MachineFunctionPass.cpp:0:0
#10 0x00007fc06a1e8776 llvm::FPPassManager::runOnFunction(llvm::Function&) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/libLLVMCore.so.17git+0x277776)
#11 0x00007fc067626daa (anonymous namespace)::CGPassManager::runOnModule(llvm::Module&) CallGraphSCCPass.cpp:0:0
#12 0x00007fc06a1e9263 llvm::legacy::PassManagerImpl::run(llvm::Module&) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/libLLVMCore.so.17git+0x278263)
#13 0x00007fc06b1734ba codegen(llvm::lto::Config const&, llvm::TargetMachine*, std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>, unsigned int, llvm::Module&, llvm::ModuleSummaryIndex const&) LTOBackend.cpp:0:0
#14 0x00007fc06b173ba9 llvm::lto::backend(llvm::lto::Config const&, std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>, unsigned int, llvm::Module&, llvm::ModuleSummaryIndex&) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/libLLVMLTO.so.17git+0x45ba9)
#15 0x00007fc06b165c33 llvm::lto::LTO::runRegularLTO(std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/libLLVMLTO.so.17git+0x37c33)
#16 0x00007fc06b166227 llvm::lto::LTO::run(std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>, std::function<llvm::Expected<std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>> (unsigned int, llvm::StringRef, llvm::Twine const&)>) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/libLLVMLTO.so.17git+0x38227)
#17 0x00007fc06e6d8b22 lld::elf::BitcodeCompiler::compile() (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/liblldELF.so.17git+0x14bb22)
#18 0x00007fc06e64486f void lld::elf::LinkerDriver::compileBitcodeFiles<llvm::object::ELFType<(llvm::support::endianness)1, true>>(bool) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/liblldELF.so.17git+0xb786f)
#19 0x00007fc06e65b8b5 lld::elf::LinkerDriver::link(llvm::opt::InputArgList&) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/liblldELF.so.17git+0xce8b5)
#20 0x00007fc06e65d2eb lld::elf::LinkerDriver::linkerMain(llvm::ArrayRef<char const*>) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/liblldELF.so.17git+0xd02eb)
#21 0x00007fc06e65eb5a lld::elf::link(llvm::ArrayRef<char const*>, llvm::raw_ostream&, llvm::raw_ostream&, bool, bool) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/lib/liblldELF.so.17git+0xd1b5a)
#22 0x0000000000405e34 lldMain(int, char const**, llvm::raw_ostream&, llvm::raw_ostream&, bool) lld.cpp:0:0
#23 0x00000000004061ee lld_main(int, char**, llvm::ToolContext const&) (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/bin/lld+0x4061ee)
#24 0x0000000000403411 main (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/bin/lld+0x403411)
#25 0x00007fc06d45cd85 __libc_start_main (/lib64/libc.so.6+0x3ad85)
#26 0x000000000040344e _start (/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/bin/lld+0x40344e)
llvm-foreach: Aborted (core dumped)
clang++: fatal error: amdgcn-link command failed with exit code 254 (use -v to see invocation)
clang version 17.0.0 (https://github.com/intel/llvm 6b7424349ee3592909afde25f9921c4e8b83db10)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/bin

How to reproduce:

git clone https://github.com/cms-patatrack/pixeltrack-standalone.git
cd pixeltrack-standalone
# eventually change the SYCL_BASE at line 157 (default is the 2022-12 build)
make environment
source env.sh
make -j`nproc` sycl

or, as suggested in the backtrace, with the file CAHitNtupletGeneratorKernelsAlloc-gfx900-6a6965-5bc87f.o (93MB 😭 ):

/cvmfs/patatrack.cern.ch/externals/x86_64/rhel8/intel/sycl/nightly/20230613/bin/lld -flavor gnu -m elf64_amdgpu --no-undefined \
     -shared -plugin-opt=-amdgpu-internalize-symbols -plugin-opt=mcpu=gfx900 -plugin-opt=O2 --lto-CGO2 \
     -o /tmp/aperego/CAHitNtupletGeneratorKernelsAlloc-gfx900-4cc973-50fdf9.out \
     CAHitNtupletGeneratorKernelsAlloc-gfx900-6a6965-5bc87f.o

Info:
gcc version: 8.5.0 20210514 (Red Hat 8.5.0-16)
clang++ version: 17.0.0 (https://github.com/intel/llvm 6b7424349ee3592909afde25f9921c4e8b83db10) - sycl-nightly/20230613
rocm version used to build clang++: rocm-5.4.0