clang crash targeting aarch64-w64-windows-gnu: Assertion failed: DT.dominates(RHead, LHead) in `CompareSCEVComplexity`

[jeremyd2019]

2022-09-18T17:36:39.8582419Z FAILED: kernel/CMakeFiles/kernel.dir/CMakeFiles/dtrsm_kernel_LN.c.obj 
2022-09-18T17:36:39.8594226Z C:\msys64\clangarm64\bin\clang.exe  -IC:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21 -IC:/M/mingw-w64-openblas/src/build-CLANGARM64-32 -DHAVE_C11 -DMS_ABI -Wall -DSMP_SERVER -DMAX_CPU_NUMBER=64 -DMAX_PARALLEL_NUMBER=1 -DNO_AFFINITY -DVERSION="\"0.3.21\"" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 -O3 -DNDEBUG -MD -MT kernel/CMakeFiles/kernel.dir/CMakeFiles/dtrsm_kernel_LN.c.obj -MF kernel\CMakeFiles\kernel.dir\CMakeFiles\dtrsm_kernel_LN.c.obj.d -o kernel/CMakeFiles/kernel.dir/CMakeFiles/dtrsm_kernel_LN.c.obj -c C:/M/mingw-w64-openblas/src/build-CLANGARM64-32/kernel/CMakeFiles/dtrsm_kernel_LN.c
2022-09-18T17:36:39.8606200Z In file included from C:/M/mingw-w64-openblas/src/build-CLANGARM64-32/kernel/CMakeFiles/dtrsm_kernel_LN.c:11:
2022-09-18T17:36:39.8612214Z In file included from C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/kernel/arm64/../generic/trsm_kernel_LN.c:39:
2022-09-18T17:36:39.8617697Z In file included from C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common.h:466:
2022-09-18T17:36:39.8624037Z C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common_arm64.h:106:65: warning: value size does not match register size specified by the constraint and modifier [-Wasm-operand-widths]
2022-09-18T17:36:39.8628825Z   __asm__ __volatile__ ("mrs %0,cntfrq_el0; clz %w0, %w0":"=&r"(shift));
2022-09-18T17:36:39.8632081Z                                                                 ^
2022-09-18T17:36:39.8636387Z C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common_arm64.h:106:30: note: use constraint modifier "w"
2022-09-18T17:36:39.8639790Z   __asm__ __volatile__ ("mrs %0,cntfrq_el0; clz %w0, %w0":"=&r"(shift));
2022-09-18T17:36:39.8642236Z                              ^~
2022-09-18T17:36:39.8667980Z                              %w0
2022-09-18T17:36:39.8673502Z PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
2022-09-18T17:36:39.8677408Z Stack dump:
2022-09-18T17:36:39.8688157Z 0.	Program arguments: C:\\msys64\\clangarm64\\bin\\clang.exe -IC:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21 -IC:/M/mingw-w64-openblas/src/build-CLANGARM64-32 -DHAVE_C11 -DMS_ABI -Wall -DSMP_SERVER -DMAX_CPU_NUMBER=64 -DMAX_PARALLEL_NUMBER=1 -DNO_AFFINITY -DVERSION=\"0.3.21\" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 -O3 -DNDEBUG -MD -MT kernel/CMakeFiles/kernel.dir/CMakeFiles/dtrsm_kernel_LN.c.obj -MF kernel\\CMakeFiles\\kernel.dir\\CMakeFiles\\dtrsm_kernel_LN.c.obj.d -o kernel/CMakeFiles/kernel.dir/CMakeFiles/dtrsm_kernel_LN.c.obj -c C:/M/mingw-w64-openblas/src/build-CLANGARM64-32/kernel/CMakeFiles/dtrsm_kernel_LN.c
2022-09-18T17:36:39.8696246Z 1.	<eof> parser at end of file
2022-09-18T17:36:39.8698642Z 2.	Code generation
2022-09-18T17:36:39.8703481Z 3.	Running pass 'Function Pass Manager' on module 'C:/M/mingw-w64-openblas/src/build-CLANGARM64-32/kernel/CMakeFiles/dtrsm_kernel_LN.c'.
2022-09-18T17:36:39.8708334Z 4.	Running pass 'Greedy Register Allocator' on function '@dtrsm_kernel_LN'
2022-09-18T17:36:39.8711382Z Exception Code: 0xC0000005
2022-09-18T17:36:39.8717052Z #0 0x00007ffd7a1c3adc llvm::LiveRange::getNextValue(llvm::SlotIndex, llvm::BumpPtrAllocatorImpl<llvm::MallocAllocator, 4096ull, 4096ull, 128ull>&) (C:\msys64\clangarm64\bin\libLLVM-15.dll+0x3f3adc)
2022-09-18T17:36:39.8722234Z #1 0x00007ffd7a3ca024 llvm::RegAllocBase::postOptimization() (C:\msys64\clangarm64\bin\libLLVM-15.dll+0x5fa024)
2022-09-18T17:36:39.8727420Z #2 0x00007ffd7a3e20d4 llvm::RAGreedy::runOnMachineFunction(llvm::MachineFunction&) (C:\msys64\clangarm64\bin\libLLVM-15.dll+0x6120d4)
2022-09-18T17:36:39.8732465Z #3 0x00007ffd7a28d1d8 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (C:\msys64\clangarm64\bin\libLLVM-15.dll+0x4bd1d8)
2022-09-18T17:36:39.8738976Z clang: error: clang frontend command failed due to signal (use -v to see invocation)
2022-09-18T17:36:39.8740086Z clang version 15.0.0
2022-09-18T17:36:39.8741243Z Target: aarch64-w64-windows-gnu
2022-09-18T17:36:39.8742097Z Thread model: posix
2022-09-18T17:36:39.8742991Z InstalledDir: C:/msys64/clangarm64/bin
2022-09-18T17:36:39.8743832Z clang: note: diagnostic msg: 
2022-09-18T17:36:39.8744581Z ********************
2022-09-18T17:36:39.8745004Z 
2022-09-18T17:36:39.8745504Z PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:
2022-09-18T17:36:39.8746586Z Preprocessed source(s) and associated run script(s) are located at:
2022-09-18T17:36:39.8749135Z clang: note: diagnostic msg: C:/msys64/tmp/dtrsm_kernel_LN-36ac6b.c
2022-09-18T17:36:40.3853133Z clang: note: diagnostic msg: C:/msys64/tmp/dtrsm_kernel_LN-36ac6b.sh
2022-09-18T17:36:40.3854715Z 
2022-09-18T17:36:40.3855562Z clang: note: diagnostic msg: 
2022-09-18T17:36:40.3856618Z 
2022-09-18T17:36:40.3856671Z 
2022-09-18T17:36:40.3856703Z 
2022-09-18T17:36:40.3881310Z ********************
2022-09-18T17:36:40.3883186Z 

Files it requested I attach:
dtrsm_kernel_LN-36ac6b.zip

See also msys2/MINGW-packages#13128

/cc @mati865 @mstorsjo

[mati865]

Also pasting here log from failed assertion on trunk:

$ /e/tmp/build/bin/clang "-cc1" "-triple" "aarch64-w64-windows-gnu" "-emit-obj" "--mrelax-relocations" "-disable-free" "-clear-ast-before-backend" "-disable-llvm-verifier" "-discard-value-names" "-main-file-name" "dtrsm_kernel_LN.c" "-mrelocation-model" "pic" "-pic-level" "2" "-mframe-pointer=none" "-fmath-errno" "-ffp-contract=on" "-fno-rounding-math" "-mconstructor-aliases" "-mms-bitfields" "-funwind-tables=2" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-target-abi" "aapcs" "-fallow-half-arguments-and-returns" "-mllvm" "-treat-scalable-fixed-error-as-warning" "-debugger-tuning=gdb" "-fcoverage-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32" "-sys-header-deps" "-D" "HAVE_C11" "-D" "MS_ABI" "-D" "SMP_SERVER" "-D" "MAX_CPU_NUMBER=64" "-D" "MAX_PARALLEL_NUMBER=1" "-D" "NO_AFFINITY" "-D" "VERSION=\"0.3.21\"" "-D" "BUILD_SINGLE" "-D" "BUILD_DOUBLE" "-D" "BUILD_COMPLEX" "-D" "BUILD_COMPLEX16" "-D" "NDEBUG" "-O3" "-Wall" "-fdebug-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32" "-ferror-limit" "19" "-fno-use-cxa-atexit" "-fgnuc-version=4.2.1" "-exception-model=seh" "-vectorize-loops" "-vectorize-slp" "-faddrsig" "-x" "c" "dtrsm_kernel_LN-36ac6b.c"
In file included from C:/M/mingw-w64-openblas/src/build-CLANGARM64-32/kernel/CMakeFiles/dtrsm_kernel_LN.c:10:
In file included from C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/kernel/arm64/../generic/trsm_kernel_LN.c:39:
In file included from C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common.h:466:
C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common_arm64.h:106:65: warning: value size does not match register size specified by the constraint and modifier [-Wasm-operand-widths]
  __asm__ __volatile__ ("mrs %0,cntfrq_el0; clz %w0, %w0":"=&r"(shift));
                                                                ^
C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common_arm64.h:106:30: note: use constraint modifier "w"
  __asm__ __volatile__ ("mrs %0,cntfrq_el0; clz %w0, %w0":"=&r"(shift));
                             ^~
                             %w0
Assertion failed: DT.dominates(RHead, LHead) && "No dominance between recurrences used by one SCEV?", file E:/tmp/llvm-project/llvm/lib/Analysis/ScalarEvolution.cpp, line 774
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: E:\\tmp\\build\\bin\\clang.exe -cc1 -triple aarch64-w64-windows-gnu -emit-obj --mrelax-relocations -disable-free -clear-ast-before-backend -disable-llvm-verifier -discard-value-names -main-file-name dtrsm_kernel_LN.c -mrelocation-model pic -pic-level 2 -mframe-pointer=none -fmath-errno -ffp-contract=on -fno-rounding-math -mconstructor-aliases -mms-bitfields -funwind-tables=2 -target-cpu generic -target-feature +neon -target-feature +v8a -target-abi aapcs -fallow-half-arguments-and-returns -mllvm -treat-scalable-fixed-error-as-warning -debugger-tuning=gdb -fcoverage-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32 -sys-header-deps -D HAVE_C11 -D MS_ABI -D SMP_SERVER -D MAX_CPU_NUMBER=64 -D MAX_PARALLEL_NUMBER=1 -D NO_AFFINITY -D VERSION=\"0.3.21\" -D BUILD_SINGLE -D BUILD_DOUBLE -D BUILD_COMPLEX -D BUILD_COMPLEX16 -D NDEBUG -O3 -Wall -fdebug-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32 -ferror-limit 19 -fno-use-cxa-atexit -fgnuc-version=4.2.1 -exception-model=seh -vectorize-loops -vectorize-slp -faddrsig -x c dtrsm_kernel_LN-36ac6b.c
1.      <eof> parser at end of file
2.      Optimizer
Exception Code: 0xC000001D
 #0 0x00007ff68f286536 HandleAbort (E:\tmp\build\bin\clang.exe+0x9d6536)
 #1 0x00007ffedfbfcd84 (C:\WINDOWS\System32\ucrtbase.dll+0x7cd84)
 #2 0x00007ffedfbfdd61 (C:\WINDOWS\System32\ucrtbase.dll+0x7dd61)
 #3 0x00007ffedfbff6bc (C:\WINDOWS\System32\ucrtbase.dll+0x7f6bc)
 #4 0x00007ffedfbff8b1 (C:\WINDOWS\System32\ucrtbase.dll+0x7f8b1)
 #5 0x00007ff68ed124e3 CompareSCEVComplexity(llvm::EquivalenceClasses<llvm::SCEV const*, std::__1::less<llvm::SCEV const*>>&, llvm::EquivalenceClasses<llvm::Value const*, std::__1::less<llvm::Value const*>>&, llvm::LoopInfo const*, llvm::SCEV const*, llvm::SCEV const*, llvm::DominatorTree&, unsigned int) (E:\tmp\build\bin\clang.exe+0x4624e3)
 #6 0x00007ff68ed1302a void std::__1::__stable_sort<std::__1::_ClassicAlgPolicy, GroupByComplexity(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::LoopInfo*, llvm::DominatorTree&)::$_57&, llvm::SCEV const**>(llvm::SCEV const**, llvm::SCEV const**, GroupByComplexity(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::LoopInfo*, llvm::DominatorTree&)::$_57&, std::__1::iterator_traits<llvm::SCEV const**>::difference_type, std::__1::iterator_traits<llvm::SCEV const**>::value_type*, long long) (E:\tmp\build\bin\clang.exe+0x46302a)
 #7 0x00007ff68ecc91d1 GroupByComplexity(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::LoopInfo*, llvm::DominatorTree&) (E:\tmp\build\bin\clang.exe+0x4191d1)
 #8 0x00007ff68ecba6b4 llvm::ScalarEvolution::getAddExpr(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::SCEV::NoWrapFlags, unsigned int) (E:\tmp\build\bin\clang.exe+0x40a6b4)
 #9 0x00007ff68fe0b9a5 llvm::SCEVExpander::expandAddToGEP(llvm::SCEV const* const*, llvm::SCEV const* const*, llvm::PointerType*, llvm::Type*, llvm::Value*) (E:\tmp\build\bin\clang.exe+0x155b9a5)
#10 0x00007ff68fe130f2 llvm::SCEVExpander::visitAddRecExpr(llvm::SCEVAddRecExpr const*) (E:\tmp\build\bin\clang.exe+0x15630f2)
#11 0x00007ff68fe0d2ea llvm::SCEVExpander::expand(llvm::SCEV const*) (E:\tmp\build\bin\clang.exe+0x155d2ea)
#12 0x00007ff68fe0c8ad llvm::SCEVExpander::expandCodeForImpl(llvm::SCEV const*, llvm::Type*, bool) (E:\tmp\build\bin\clang.exe+0x155c8ad)
#13 0x00007ff6909fae43 expandBounds(llvm::RuntimeCheckingPtrGroup const*, llvm::Loop*, llvm::Instruction*, llvm::SCEVExpander&) (E:\tmp\build\bin\clang.exe+0x214ae43)
#14 0x00007ff6909f786f llvm::addRuntimeChecks(llvm::Instruction*, llvm::Loop*, llvm::SmallVectorImpl<std::__1::pair<llvm::RuntimeCheckingPtrGroup const*, llvm::RuntimeCheckingPtrGroup const*>> const&, llvm::SCEVExpander&) (E:\tmp\build\bin\clang.exe+0x214786f)
#15 0x00007ff690a61974 llvm::LoopVersioning::versionLoop(llvm::SmallVectorImpl<llvm::Instruction*> const&) (E:\tmp\build\bin\clang.exe+0x21b1974)
#16 0x00007ff691735e5e eliminateLoadsAcrossLoops(llvm::Function&, llvm::LoopInfo&, llvm::DominatorTree&, llvm::BlockFrequencyInfo*, llvm::ProfileSummaryInfo*, llvm::ScalarEvolution*, llvm::AssumptionCache*, llvm::function_ref<llvm::LoopAccessInfo const& (llvm::Loop&)>) (E:\tmp\build\bin\clang.exe+0x2e85e5e)
#17 0x00007ff691733287 llvm::LoopLoadEliminationPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0x2e83287)
#18 0x00007ff692575881 llvm::detail::PassModel<llvm::Function, llvm::LoopLoadEliminationPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0x3cc5881)
#19 0x00007ff68fb16465 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0x1266465)
#20 0x00007ff68e951311 llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0xa1311)
#21 0x00007ff68fb1a044 llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (E:\tmp\build\bin\clang.exe+0x126a044)
#22 0x00007ff68e951021 llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (E:\tmp\build\bin\clang.exe+0xa1021)
#23 0x00007ff68fb14ff5 llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (E:\tmp\build\bin\clang.exe+0x1264ff5)
#24 0x00007ff690c2d846 (anonymous namespace)::EmitAssemblyHelper::RunOptimizationPipeline(clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream, std::__1::default_delete<llvm::raw_pwrite_stream>>&, std::__1::unique_ptr<llvm::ToolOutputFile, std::__1::default_delete<llvm::ToolOutputFile>>&) (E:\tmp\build\bin\clang.exe+0x237d846)
#25 0x00007ff690c26309 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream, std::__1::default_delete<llvm::raw_pwrite_stream>>) (E:\tmp\build\bin\clang.exe+0x2376309)
#26 0x00007ff69108de9a clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) (E:\tmp\build\bin\clang.exe+0x27dde9a)
#27 0x00007ff692fa6e3f clang::ParseAST(clang::Sema&, bool, bool) (E:\tmp\build\bin\clang.exe+0x46f6e3f)
#28 0x00007ff690fddae3 clang::FrontendAction::Execute() (E:\tmp\build\bin\clang.exe+0x272dae3)
#29 0x00007ff68f444617 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (E:\tmp\build\bin\clang.exe+0xb94617)
#30 0x00007ff68f4ccf7b clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (E:\tmp\build\bin\clang.exe+0xc1cf7b)
#31 0x00007ff68e8b75e4 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (E:\tmp\build\bin\clang.exe+0x75e4)
#32 0x00007ff68e8b4f1a ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) (E:\tmp\build\bin\clang.exe+0x4f1a)
#33 0x00007ff68e8b4cba clang_main(int, char**) (E:\tmp\build\bin\clang.exe+0x4cba)
#34 0x00007ff68e8b13d7 __tmainCRTStartup C:/M/mingw-w64-crt-git/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:329:13
#35 0x00007ff68e8b1436 .l_start C:/M/mingw-w64-crt-git/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:208:3
#36 0x00007ffee0ae54e0 (C:\WINDOWS\System32\KERNEL32.DLL+0x154e0)
#37 0x00007ffee232485b (C:\WINDOWS\SYSTEM32\ntdll.dll+0x485b)
Illegal instruction

[mstorsjo]

I reduced the triggering input to this:

double *a;
double b, c;
int d, e;
long f;
void g() {
  d = 1;
  for (; d >= 0; d--) {
    e = 0;
    for (; e < f; e++) {
      c = *(&b + e);
      *a = *(&b + d + e) = c;
      a++;
    }
  }
}

Compiled with this command: clang -target aarch64-windows-msvc -c repro.c -O2.

[mstorsjo]

I bisected this failed assert down to this commit:

commit 20d798bd47ec5191de1b2a8a031da06a04e612e1
Author: Florian Hahn <flo@fhahn.com>
Date:   Thu Jun 9 15:20:10 2022 +0100

    Recommit "[SCEV] Look through single value PHIs." (take 3)
    
    This reverts commit 1fbdbb559569641f6d509b569966901c8fb02b63.
    
    All known issues surfaced by this patch should have been fixed now.
    
    The fixes included fixing issues with SCEV expansion in LV and DA's
    reliance on LCSSA phis.

CC @fhahn

[fhahn]

Reduced reproducer: https://llvm.godbolt.org/z/Kcq4Te798

Looks like a SCEV invalidation issue in loop-vectorize.

[lazka]

Could it be that the above errors are two different issues? I backported 623c4a7 but the OP example still crashes as before, unless I missed something.

[mati865]

Perhaps the reduction was triggering different issue because it doesn't fail on the trunk anymore.
Original code still hits the same assertion though:

$ /e/tmp/build/bin/clang "-cc1" "-triple" "aarch64-w64-windows-gnu" "-emit-obj" "--mrelax-relocations" "-disable-free" "-clear-ast-before-backend" "-disable-llvm-verifier" "-discard-value-names" "-main-file-name" "dtrsm_kernel_LN.c" "-mrelocation-model" "pic" "-pic-level" "2" "-mframe-pointer=none" "-fmath-errno" "-ffp-contract=on" "-fno-rounding-math" "-mconstructor-aliases" "-mms-bitfields" "-funwind-tables=2" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-target-abi" "aapcs" "-fallow-half-arguments-and-returns" "-mllvm" "-treat-scalable-fixed-error-as-warning" "-debugger-tuning=gdb" "-fcoverage-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32" "-sys-header-deps" "-D" "HAVE_C11" "-D" "MS_ABI" "-D" "SMP_SERVER" "-D" "MAX_CPU_NUMBER=64" "-D" "MAX_PARALLEL_NUMBER=1" "-D" "NO_AFFINITY" "-D" "VERSION=\"0.3.21\"" "-D" "BUILD_SINGLE" "-D" "BUILD_DOUBLE" "-D" "BUILD_COMPLEX" "-D" "BUILD_COMPLEX16" "-D" "NDEBUG" "-O3" "-Wall" "-fdebug-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32" "-ferror-limit" "19" "-fno-use-cxa-atexit" "-fgnuc-version=4.2.1" "-exception-model=seh" "-vectorize-loops" "-vectorize-slp" "-faddrsig" "-x" "c" "dtrsm_kernel_LN-36ac6b.c"
In file included from C:/M/mingw-w64-openblas/src/build-CLANGARM64-32/kernel/CMakeFiles/dtrsm_kernel_LN.c:10:
In file included from C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/kernel/arm64/../generic/trsm_kernel_LN.c:39:
In file included from C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common.h:466:
C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common_arm64.h:106:65: warning: value size does not match register size specified by the constraint and modifier [-Wasm-operand-widths]
  __asm__ __volatile__ ("mrs %0,cntfrq_el0; clz %w0, %w0":"=&r"(shift));
                                                                ^
C:/M/mingw-w64-openblas/src/OpenBLAS-0.3.21/common_arm64.h:106:30: note: use constraint modifier "w"
  __asm__ __volatile__ ("mrs %0,cntfrq_el0; clz %w0, %w0":"=&r"(shift));
                             ^~
                             %w0
Assertion failed: DT.dominates(RHead, LHead) && "No dominance between recurrences used by one SCEV?", file E:/tmp/llvm-project/llvm/lib/Analysis/ScalarEvolution.cpp, line 774
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: E:\\tmp\\build\\bin\\clang.exe -cc1 -triple aarch64-w64-windows-gnu -emit-obj --mrelax-relocations -disable-free -clear-ast-before-backend -disable-llvm-verifier -discard-value-names -main-file-name dtrsm_kernel_LN.c -mrelocation-model pic -pic-level 2 -mframe-pointer=none -fmath-errno -ffp-contract=on -fno-rounding-math -mconstructor-aliases -mms-bitfields -funwind-tables=2 -target-cpu generic -target-feature +neon -target-feature +v8a -target-abi aapcs -fallow-half-arguments-and-returns -mllvm -treat-scalable-fixed-error-as-warning -debugger-tuning=gdb -fcoverage-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32 -sys-header-deps -D HAVE_C11 -D MS_ABI -D SMP_SERVER -D MAX_CPU_NUMBER=64 -D MAX_PARALLEL_NUMBER=1 -D NO_AFFINITY -D VERSION=\"0.3.21\" -D BUILD_SINGLE -D BUILD_DOUBLE -D BUILD_COMPLEX -D BUILD_COMPLEX16 -D NDEBUG -O3 -Wall -fdebug-compilation-dir=C:/M/mingw-w64-openblas/src/build-CLANGARM64-32 -ferror-limit 19 -fno-use-cxa-atexit -fgnuc-version=4.2.1 -exception-model=seh -vectorize-loops -vectorize-slp -faddrsig -x c dtrsm_kernel_LN-36ac6b.c
1.      <eof> parser at end of file
2.      Optimizer
Exception Code: 0xC000001D
 #0 0x00007ff6ef8103e6 HandleAbort (E:\tmp\build\bin\clang.exe+0x9f03e6)
 #1 0x00007ffb38d3e5f5 (C:\WINDOWS\System32\ucrtbase.dll+0x7e5f5)
 #2 0x00007ffb38d3f601 (C:\WINDOWS\System32\ucrtbase.dll+0x7f601)
 #3 0x00007ffb38d40f39 (C:\WINDOWS\System32\ucrtbase.dll+0x80f39)
 #4 0x00007ffb38d41121 (C:\WINDOWS\System32\ucrtbase.dll+0x81121)
 #5 0x00007ff6ef298b13 CompareSCEVComplexity(llvm::EquivalenceClasses<llvm::SCEV const*, std::__1::less<llvm::SCEV const*>>&, llvm::EquivalenceClasses<llvm::Value const*, std::__1::less<llvm::Value const*>>&, llvm::LoopInfo const*, llvm::SCEV const*, llvm::SCEV const*, llvm::DominatorTree&, unsigned int) (E:\tmp\build\bin\clang.exe+0x478b13)
 #6 0x00007ff6ef29965a void std::__1::__stable_sort<std::__1::_ClassicAlgPolicy, GroupByComplexity(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::LoopInfo*, llvm::DominatorTree&)::$_57&, llvm::SCEV const**>(llvm::SCEV const**, llvm::SCEV const**, GroupByComplexity(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::LoopInfo*, llvm::DominatorTree&)::$_57&, std::__1::iterator_traits<llvm::SCEV const**>::difference_type, std::__1::iterator_traits<llvm::SCEV const**>::value_type*, long long) (E:\tmp\build\bin\clang.exe+0x47965a)
 #7 0x00007ff6ef24f381 GroupByComplexity(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::LoopInfo*, llvm::DominatorTree&) (E:\tmp\build\bin\clang.exe+0x42f381)
 #8 0x00007ff6ef240864 llvm::ScalarEvolution::getAddExpr(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::SCEV::NoWrapFlags, unsigned int) (E:\tmp\build\bin\clang.exe+0x420864)
 #9 0x00007ff6f0397e75 llvm::SCEVExpander::expandAddToGEP(llvm::SCEV const* const*, llvm::SCEV const* const*, llvm::PointerType*, llvm::Type*, llvm::Value*) (E:\tmp\build\bin\clang.exe+0x1577e75)
#10 0x00007ff6f039f5c2 llvm::SCEVExpander::visitAddRecExpr(llvm::SCEVAddRecExpr const*) (E:\tmp\build\bin\clang.exe+0x157f5c2)
#11 0x00007ff6f03997ba llvm::SCEVExpander::expand(llvm::SCEV const*) (E:\tmp\build\bin\clang.exe+0x15797ba)
#12 0x00007ff6f0398d7d llvm::SCEVExpander::expandCodeForImpl(llvm::SCEV const*, llvm::Type*, bool) (E:\tmp\build\bin\clang.exe+0x1578d7d)
#13 0x00007ff6f0f834ba expandBounds(llvm::RuntimeCheckingPtrGroup const*, llvm::Loop*, llvm::Instruction*, llvm::SCEVExpander&) (E:\tmp\build\bin\clang.exe+0x21634ba)
#14 0x00007ff6f0f7feff llvm::addRuntimeChecks(llvm::Instruction*, llvm::Loop*, llvm::SmallVectorImpl<std::__1::pair<llvm::RuntimeCheckingPtrGroup const*, llvm::RuntimeCheckingPtrGroup const*>> const&, llvm::SCEVExpander&) (E:\tmp\build\bin\clang.exe+0x215feff)
#15 0x00007ff6f0fea3a4 llvm::LoopVersioning::versionLoop(llvm::SmallVectorImpl<llvm::Instruction*> const&) (E:\tmp\build\bin\clang.exe+0x21ca3a4)
#16 0x00007ff6f1cc2d4e eliminateLoadsAcrossLoops(llvm::Function&, llvm::LoopInfo&, llvm::DominatorTree&, llvm::BlockFrequencyInfo*, llvm::ProfileSummaryInfo*, llvm::ScalarEvolution*, llvm::AssumptionCache*, llvm::function_ref<llvm::LoopAccessInfo const& (llvm::Loop&)>) (E:\tmp\build\bin\clang.exe+0x2ea2d4e)
#17 0x00007ff6f1cc0177 llvm::LoopLoadEliminationPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0x2ea0177)
#18 0x00007ff6f2b06c91 llvm::detail::PassModel<llvm::Function, llvm::LoopLoadEliminationPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0x3ce6c91)
#19 0x00007ff6f00a27d5 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0x12827d5)
#20 0x00007ff6eeec1341 llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (E:\tmp\build\bin\clang.exe+0xa1341)
#21 0x00007ff6f00a63b4 llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (E:\tmp\build\bin\clang.exe+0x12863b4)
#22 0x00007ff6eeec1051 llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (E:\tmp\build\bin\clang.exe+0xa1051)
#23 0x00007ff6f00a1365 llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (E:\tmp\build\bin\clang.exe+0x1281365)
#24 0x00007ff6f11b6736 (anonymous namespace)::EmitAssemblyHelper::RunOptimizationPipeline(clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream, std::__1::default_delete<llvm::raw_pwrite_stream>>&, std::__1::unique_ptr<llvm::ToolOutputFile, std::__1::default_delete<llvm::ToolOutputFile>>&) (E:\tmp\build\bin\clang.exe+0x2396736)
#25 0x00007ff6f11af1a9 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, std::__1::unique_ptr<llvm::raw_pwrite_stream, std::__1::default_delete<llvm::raw_pwrite_stream>>) (E:\tmp\build\bin\clang.exe+0x238f1a9)
#26 0x00007ff6f161b53a clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) (E:\tmp\build\bin\clang.exe+0x27fb53a)
#27 0x00007ff6f353afeb clang::ParseAST(clang::Sema&, bool, bool) (E:\tmp\build\bin\clang.exe+0x471afeb)
#28 0x00007ff6f1568213 clang::FrontendAction::Execute() (E:\tmp\build\bin\clang.exe+0x2748213)
#29 0x00007ff6ef9cf277 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (E:\tmp\build\bin\clang.exe+0xbaf277)
#30 0x00007ff6efa57e6b clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (E:\tmp\build\bin\clang.exe+0xc37e6b)
#31 0x00007ff6eee27614 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (E:\tmp\build\bin\clang.exe+0x7614)
#32 0x00007ff6eee24f1a ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) (E:\tmp\build\bin\clang.exe+0x4f1a)
#33 0x00007ff6eee24cba clang_main(int, char**) (E:\tmp\build\bin\clang.exe+0x4cba)
#34 0x00007ff6eee213d7 __tmainCRTStartup C:/M/mingw-w64-crt-git/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:329:13
#35 0x00007ff6eee21436 .l_start C:/M/mingw-w64-crt-git/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:208:3
#36 0x00007ffb3a5e244d (C:\WINDOWS\System32\KERNEL32.DLL+0x1244d)
#37 0x00007ffb3b60df78 (C:\WINDOWS\SYSTEM32\ntdll.dll+0x5df78)
Illegal instruction

[mstorsjo]

Yes, apparently there’s multiple similar issues nearby. I re-reduced the original case down to a new one with latest git main:

double a;
double *b, *c;
int d, e;
long f;
void g() {
  while (4) {
    d = 4;
    for (; d; d--) {
      e = 0;
      for (; e < f; e++) {
        *b = *b++;
        *c = a;
      }
    }
  }
}

[mati865]

Interestingly this reduction triggers the assertion (if enabled) but doesn't cause ICE on LLVM 15 without assertions so indeed there are multiple issues here. So this fixing this reduction still likely won't solve ICE on LLVM 15.