[Flang][OpenMP] Compilation error of atomic construct

[ohno-fj]

Version of flang-new : 18.0.0(770dc47659d41a5ca7b7daf5b3134c900ca8c33d)

atomic construct results in a compilation error.

The following are the test program, Flang-new, Gfortran and ifort compilation result.

omp_PL_ATOMIC_001c_3.f90:

      program main
      complex*8 ia, ib
      integer*4 :: loop=10000
      ia = (0,0)
!$omp parallel
!$omp do
      do i=1, loop
!$omp atomic
         ia = ia + (1,1)
      enddo
!$omp end do
!$omp end parallel
      end

$ flang-new -fopenmp omp_PL_ATOMIC_001c_3.f90
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":7:7): no atomic update operation with region argument as operand found inside atomic.update region
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":7:7): LLVM Translation failed for operation: omp.atomic.update
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":6:7): LLVM Translation failed for operation: omp.wsloop
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /lustre/home/ohno/LLVM_20231130/release/bin/flang-new -fc1 -triple aarch64-unknown-linux-gnu -emit-obj -fopenmp -mrelocation-model pic -pic-level 2 -pic-is-pie -target-cpu generic -target-feature +neon -target-feature +v8a -o /tmp/omp_PL_ATOMIC_001c_3-814ed2.o -x f95-cpp-input omp_PL_ATOM\
IC_001c_3.f90
 #0 0x0000000003f7d7b0 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3f7d7b0)
 #1 0x0000000003f7b6b8 llvm::sys::RunSignalHandlers() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3f7b6b8)
 #2 0x0000000003f7b83c SignalHandler(int) Signals.cpp:0:0
 #3 0x00004000249907a0 (linux-vdso.so.1+0x7a0)
 #4 0x0000000007207ba4 llvm::CodeExtractor::findAllocas(llvm::CodeExtractorAnalysisCache const&, llvm::SetVector<llvm::Value*, llvm::SmallVector<llvm::Value*, 0u>, llvm::DenseSet<llvm::Value*, llvm::DenseMapInfo<llvm::Value*, void>>, 0u>&, llvm::SetVector<llvm::Value*, llvm::SmallVector<llvm::Value*, 0u>, llvm::DenseSet<llvm::Value*, llvm::DenseMapInfo<llvm::Value*, void>>, 0u>&, llvm::BasicBlock*&) const (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x7207ba4)
 #5 0x0000000006e341a0 llvm::OpenMPIRBuilder::createParallel(llvm::OpenMPIRBuilder::LocationDescription const&, llvm::IRBuilderBase::InsertPoint, llvm::function_ref<void (llvm::IRBuilderBase::InsertPoint, llvm::IRBuilderBase::InsertPoint)>, llvm::function_ref<llvm::IRBuilderBase::InsertPoint (llvm::IRBuilderBase::InsertPoint, llvm::IRBuilderBase::InsertPoint, llvm::Value&, llvm::Value&, llvm::Value*&)>, std::function<void (llvm::IRBuilderBase::InsertPoint)>, llvm::Value*, llvm::Value*, llvm::omp::ProcBindKind, bool) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x6e341a0)
 #6 0x00000000055b2e2c (anonymous namespace)::OpenMPDialectLLVMIRTranslationInterface::convertOperation(mlir::Operation*, llvm::IRBuilderBase&, mlir::LLVM::ModuleTranslation&) const OpenMPToLLVMIRTranslation.cpp:0:0
 #7 0x0000000005c65114 mlir::LLVM::ModuleTranslation::convertOperation(mlir::Operation&, llvm::IRBuilderBase&) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c65114)
 #8 0x0000000005c70844 mlir::LLVM::ModuleTranslation::convertBlock(mlir::Block&, bool, llvm::IRBuilderBase&) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c70844)
 #9 0x0000000005c71438 mlir::LLVM::ModuleTranslation::convertOneFunction(mlir::LLVM::LLVMFuncOp) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c71438)
#10 0x0000000005c71a58 mlir::LLVM::ModuleTranslation::convertFunctions() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c71a58)
#11 0x0000000005c72864 mlir::translateModuleToLLVMIR(mlir::Operation*, llvm::LLVMContext&, llvm::StringRef) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c72864)
#12 0x00000000045b6b48 Fortran::frontend::CodeGenAction::generateLLVMIR() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x45b6b48)
#13 0x00000000045b86c8 Fortran::frontend::CodeGenAction::executeAction() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x45b86c8)
#14 0x0000000003fae89c Fortran::frontend::FrontendAction::execute() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fae89c)
#15 0x0000000003fa25f4 Fortran::frontend::CompilerInstance::executeAction(Fortran::frontend::FrontendAction&) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fa25f4)
#16 0x0000000003fb3534 Fortran::frontend::executeCompilerInvocation(Fortran::frontend::CompilerInstance*) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fb3534)
#17 0x0000000003bea120 fc1_main(llvm::ArrayRef<char const*>, char const*) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3bea120)
#18 0x0000000003b75674 main (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3b75674)
#19 0x0000400024d74384 __libc_start_main (/lib64/libc.so.6+0x24384)
#20 0x0000000003be8f54 _start (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3be8f54)
flang-new: error: unable to execute command: Segmentation fault (core dumped)
flang-new: error: flang frontend command failed due to signal (use -v to see invocation)
flang-new version 18.0.0 (https://github.com/llvm/llvm-project.git 770dc47659d41a5ca7b7daf5b3134c900ca8c33d)
Target: aarch64-unknown-linux-gnu
Thread model: posix
InstalledDir: /lustre/home/ohno/LLVM_20231130/release/bin
flang-new: note: diagnostic msg:
********************

PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:
Preprocessed source(s) and associated run script(s) are located at:
flang-new: note: diagnostic msg: /tmp/omp_PL_ATOMIC_001c_3-39f821
flang-new: note: diagnostic msg: /tmp/omp_PL_ATOMIC_001c_3-39f821.sh
flang-new: note: diagnostic msg:

********************
$

$ cat /tmp/omp_PL_ATOMIC_001c_3-39f821
#line "./omp_PL_ATOMIC_001c_3.f90" 1
      program main
      complex*8 ia, ib
      integer*4 :: loop=10000
      ia = (0,0)
!$omp parallel
!$omp do
      do i=1, loop
!$omp atomic
         ia = ia + (1,1)
      enddo
!$omp end do
!$omp end parallel
      end
$

$ cat /tmp/omp_PL_ATOMIC_001c_3-39f821.sh
# Crash reproducer for clang version 18.0.0 (https://github.com/llvm/llvm-project.git 770dc47659d41a5ca7b7daf5b3134c900ca8c33d)
# Driver args: "-fopenmp" "omp_PL_ATOMIC_001c_3.f90"
# Original command:  "/lustre/home/ohno/LLVM_20231130/release/bin/flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu" "-emit-obj" "-fopenmp" "-mrelocation-model" "pic" "-pic-level" "2" "-pic-is-pie" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-o" "/tmp/omp_PL_ATOMIC_001c_3-814ed2.o" "-x" "f95-cpp-input" "omp_PL_ATOMIC_001c_3.f90"
 "/lustre/home/ohno/LLVM_20231130/release/bin/flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu" "-emit-obj" "-fopenmp" "-mrelocation-model" "pic" "-pic-level" "2" "-pic-is-pie" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-x" "f95-cpp-input" "omp_PL_ATOMIC_001c_3-39f821"
$

$ gfortran -fopenmp omp_PL_ATOMIC_001c_3.f90
$

$ ifort -qopenmp omp_PL_ATOMIC_001c_3.f90
$

[llvmbot]

@llvm/issue-subscribers-flang-ir

Author: None (ohno-fj)

``` Version of flang-new : 18.0.0(770dc47) ```

atomic construct results in a compilation error.

The following are the test program, Flang-new, Gfortran and ifort compilation result.

omp_PL_ATOMIC_001c_3.f90:

      program main
      complex*8 ia, ib
      integer*4 :: loop=10000
      ia = (0,0)
!$omp parallel
!$omp do
      do i=1, loop
!$omp atomic
         ia = ia + (1,1)
      enddo
!$omp end do
!$omp end parallel
      end

$ flang-new -fopenmp omp_PL_ATOMIC_001c_3.f90
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":7:7): no atomic update operation with region argument as operand found inside atomic.update region
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":7:7): LLVM Translation failed for operation: omp.atomic.update
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":6:7): LLVM Translation failed for operation: omp.wsloop
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /lustre/home/ohno/LLVM_20231130/release/bin/flang-new -fc1 -triple aarch64-unknown-linux-gnu -emit-obj -fopenmp -mrelocation-model pic -pic-level 2 -pic-is-pie -target-cpu generic -target-feature +neon -target-feature +v8a -o /tmp/omp_PL_ATOMIC_001c_3-814ed2.o -x f95-cpp-input omp_PL_ATOM\
IC_001c_3.f90
 #<!-- -->0 0x0000000003f7d7b0 llvm::sys::PrintStackTrace(llvm::raw_ostream&amp;, int) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3f7d7b0)
 #<!-- -->1 0x0000000003f7b6b8 llvm::sys::RunSignalHandlers() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3f7b6b8)
 #<!-- -->2 0x0000000003f7b83c SignalHandler(int) Signals.cpp:0:0
 #<!-- -->3 0x00004000249907a0 (linux-vdso.so.1+0x7a0)
 #<!-- -->4 0x0000000007207ba4 llvm::CodeExtractor::findAllocas(llvm::CodeExtractorAnalysisCache const&amp;, llvm::SetVector&lt;llvm::Value*, llvm::SmallVector&lt;llvm::Value*, 0u&gt;, llvm::DenseSet&lt;llvm::Value*, llvm::DenseMapInfo&lt;llvm::Value*, void&gt;&gt;, 0u&gt;&amp;, llvm::SetVector&lt;llvm::Value*, llvm::SmallVector&lt;llvm::Value*, 0u&gt;, llvm::DenseSet&lt;llvm::Value*, llvm::DenseMapInfo&lt;llvm::Value*, void&gt;&gt;, 0u&gt;&amp;, llvm::BasicBlock*&amp;) const (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x7207ba4)
 #<!-- -->5 0x0000000006e341a0 llvm::OpenMPIRBuilder::createParallel(llvm::OpenMPIRBuilder::LocationDescription const&amp;, llvm::IRBuilderBase::InsertPoint, llvm::function_ref&lt;void (llvm::IRBuilderBase::InsertPoint, llvm::IRBuilderBase::InsertPoint)&gt;, llvm::function_ref&lt;llvm::IRBuilderBase::InsertPoint (llvm::IRBuilderBase::InsertPoint, llvm::IRBuilderBase::InsertPoint, llvm::Value&amp;, llvm::Value&amp;, llvm::Value*&amp;)&gt;, std::function&lt;void (llvm::IRBuilderBase::InsertPoint)&gt;, llvm::Value*, llvm::Value*, llvm::omp::ProcBindKind, bool) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x6e341a0)
 #<!-- -->6 0x00000000055b2e2c (anonymous namespace)::OpenMPDialectLLVMIRTranslationInterface::convertOperation(mlir::Operation*, llvm::IRBuilderBase&amp;, mlir::LLVM::ModuleTranslation&amp;) const OpenMPToLLVMIRTranslation.cpp:0:0
 #<!-- -->7 0x0000000005c65114 mlir::LLVM::ModuleTranslation::convertOperation(mlir::Operation&amp;, llvm::IRBuilderBase&amp;) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c65114)
 #<!-- -->8 0x0000000005c70844 mlir::LLVM::ModuleTranslation::convertBlock(mlir::Block&amp;, bool, llvm::IRBuilderBase&amp;) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c70844)
 #<!-- -->9 0x0000000005c71438 mlir::LLVM::ModuleTranslation::convertOneFunction(mlir::LLVM::LLVMFuncOp) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c71438)
#<!-- -->10 0x0000000005c71a58 mlir::LLVM::ModuleTranslation::convertFunctions() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c71a58)
#<!-- -->11 0x0000000005c72864 mlir::translateModuleToLLVMIR(mlir::Operation*, llvm::LLVMContext&amp;, llvm::StringRef) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c72864)
#<!-- -->12 0x00000000045b6b48 Fortran::frontend::CodeGenAction::generateLLVMIR() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x45b6b48)
#<!-- -->13 0x00000000045b86c8 Fortran::frontend::CodeGenAction::executeAction() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x45b86c8)
#<!-- -->14 0x0000000003fae89c Fortran::frontend::FrontendAction::execute() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fae89c)
#<!-- -->15 0x0000000003fa25f4 Fortran::frontend::CompilerInstance::executeAction(Fortran::frontend::FrontendAction&amp;) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fa25f4)
#<!-- -->16 0x0000000003fb3534 Fortran::frontend::executeCompilerInvocation(Fortran::frontend::CompilerInstance*) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fb3534)
#<!-- -->17 0x0000000003bea120 fc1_main(llvm::ArrayRef&lt;char const*&gt;, char const*) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3bea120)
#<!-- -->18 0x0000000003b75674 main (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3b75674)
#<!-- -->19 0x0000400024d74384 __libc_start_main (/lib64/libc.so.6+0x24384)
#<!-- -->20 0x0000000003be8f54 _start (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3be8f54)
flang-new: error: unable to execute command: Segmentation fault (core dumped)
flang-new: error: flang frontend command failed due to signal (use -v to see invocation)
flang-new version 18.0.0 (https://github.com/llvm/llvm-project.git 770dc47659d41a5ca7b7daf5b3134c900ca8c33d)
Target: aarch64-unknown-linux-gnu
Thread model: posix
InstalledDir: /lustre/home/ohno/LLVM_20231130/release/bin
flang-new: note: diagnostic msg:
********************

PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:
Preprocessed source(s) and associated run script(s) are located at:
flang-new: note: diagnostic msg: /tmp/omp_PL_ATOMIC_001c_3-39f821
flang-new: note: diagnostic msg: /tmp/omp_PL_ATOMIC_001c_3-39f821.sh
flang-new: note: diagnostic msg:

********************
$

$ cat /tmp/omp_PL_ATOMIC_001c_3-39f821
#line "./omp_PL_ATOMIC_001c_3.f90" 1
      program main
      complex*8 ia, ib
      integer*4 :: loop=10000
      ia = (0,0)
!$omp parallel
!$omp do
      do i=1, loop
!$omp atomic
         ia = ia + (1,1)
      enddo
!$omp end do
!$omp end parallel
      end
$

$ cat /tmp/omp_PL_ATOMIC_001c_3-39f821.sh
# Crash reproducer for clang version 18.0.0 (https://github.com/llvm/llvm-project.git 770dc47659d41a5ca7b7daf5b3134c900ca8c33d)
# Driver args: "-fopenmp" "omp_PL_ATOMIC_001c_3.f90"
# Original command:  "/lustre/home/ohno/LLVM_20231130/release/bin/flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu" "-emit-obj" "-fopenmp" "-mrelocation-model" "pic" "-pic-level" "2" "-pic-is-pie" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-o" "/tmp/omp_PL_ATOMIC_001c_3-814ed2.o" "-x" "f95-cpp-input" "omp_PL_ATOMIC_001c_3.f90"
 "/lustre/home/ohno/LLVM_20231130/release/bin/flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu" "-emit-obj" "-fopenmp" "-mrelocation-model" "pic" "-pic-level" "2" "-pic-is-pie" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-x" "f95-cpp-input" "omp_PL_ATOMIC_001c_3-39f821"
$

$ gfortran -fopenmp omp_PL_ATOMIC_001c_3.f90
$

$ ifort -qopenmp omp_PL_ATOMIC_001c_3.f90
$

[llvmbot]

@llvm/issue-subscribers-openmp

Author: None (ohno-fj)

``` Version of flang-new : 18.0.0(770dc47) ```

atomic construct results in a compilation error.

The following are the test program, Flang-new, Gfortran and ifort compilation result.

omp_PL_ATOMIC_001c_3.f90:

      program main
      complex*8 ia, ib
      integer*4 :: loop=10000
      ia = (0,0)
!$omp parallel
!$omp do
      do i=1, loop
!$omp atomic
         ia = ia + (1,1)
      enddo
!$omp end do
!$omp end parallel
      end

$ flang-new -fopenmp omp_PL_ATOMIC_001c_3.f90
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":7:7): no atomic update operation with region argument as operand found inside atomic.update region
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":7:7): LLVM Translation failed for operation: omp.atomic.update
error: loc("/work/home/ohno/CT/test/fort/tp/reproducerJ/fomp_add00/omp_PL_ATOMIC_001c_3.f90":6:7): LLVM Translation failed for operation: omp.wsloop
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /lustre/home/ohno/LLVM_20231130/release/bin/flang-new -fc1 -triple aarch64-unknown-linux-gnu -emit-obj -fopenmp -mrelocation-model pic -pic-level 2 -pic-is-pie -target-cpu generic -target-feature +neon -target-feature +v8a -o /tmp/omp_PL_ATOMIC_001c_3-814ed2.o -x f95-cpp-input omp_PL_ATOM\
IC_001c_3.f90
 #<!-- -->0 0x0000000003f7d7b0 llvm::sys::PrintStackTrace(llvm::raw_ostream&amp;, int) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3f7d7b0)
 #<!-- -->1 0x0000000003f7b6b8 llvm::sys::RunSignalHandlers() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3f7b6b8)
 #<!-- -->2 0x0000000003f7b83c SignalHandler(int) Signals.cpp:0:0
 #<!-- -->3 0x00004000249907a0 (linux-vdso.so.1+0x7a0)
 #<!-- -->4 0x0000000007207ba4 llvm::CodeExtractor::findAllocas(llvm::CodeExtractorAnalysisCache const&amp;, llvm::SetVector&lt;llvm::Value*, llvm::SmallVector&lt;llvm::Value*, 0u&gt;, llvm::DenseSet&lt;llvm::Value*, llvm::DenseMapInfo&lt;llvm::Value*, void&gt;&gt;, 0u&gt;&amp;, llvm::SetVector&lt;llvm::Value*, llvm::SmallVector&lt;llvm::Value*, 0u&gt;, llvm::DenseSet&lt;llvm::Value*, llvm::DenseMapInfo&lt;llvm::Value*, void&gt;&gt;, 0u&gt;&amp;, llvm::BasicBlock*&amp;) const (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x7207ba4)
 #<!-- -->5 0x0000000006e341a0 llvm::OpenMPIRBuilder::createParallel(llvm::OpenMPIRBuilder::LocationDescription const&amp;, llvm::IRBuilderBase::InsertPoint, llvm::function_ref&lt;void (llvm::IRBuilderBase::InsertPoint, llvm::IRBuilderBase::InsertPoint)&gt;, llvm::function_ref&lt;llvm::IRBuilderBase::InsertPoint (llvm::IRBuilderBase::InsertPoint, llvm::IRBuilderBase::InsertPoint, llvm::Value&amp;, llvm::Value&amp;, llvm::Value*&amp;)&gt;, std::function&lt;void (llvm::IRBuilderBase::InsertPoint)&gt;, llvm::Value*, llvm::Value*, llvm::omp::ProcBindKind, bool) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x6e341a0)
 #<!-- -->6 0x00000000055b2e2c (anonymous namespace)::OpenMPDialectLLVMIRTranslationInterface::convertOperation(mlir::Operation*, llvm::IRBuilderBase&amp;, mlir::LLVM::ModuleTranslation&amp;) const OpenMPToLLVMIRTranslation.cpp:0:0
 #<!-- -->7 0x0000000005c65114 mlir::LLVM::ModuleTranslation::convertOperation(mlir::Operation&amp;, llvm::IRBuilderBase&amp;) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c65114)
 #<!-- -->8 0x0000000005c70844 mlir::LLVM::ModuleTranslation::convertBlock(mlir::Block&amp;, bool, llvm::IRBuilderBase&amp;) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c70844)
 #<!-- -->9 0x0000000005c71438 mlir::LLVM::ModuleTranslation::convertOneFunction(mlir::LLVM::LLVMFuncOp) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c71438)
#<!-- -->10 0x0000000005c71a58 mlir::LLVM::ModuleTranslation::convertFunctions() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c71a58)
#<!-- -->11 0x0000000005c72864 mlir::translateModuleToLLVMIR(mlir::Operation*, llvm::LLVMContext&amp;, llvm::StringRef) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x5c72864)
#<!-- -->12 0x00000000045b6b48 Fortran::frontend::CodeGenAction::generateLLVMIR() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x45b6b48)
#<!-- -->13 0x00000000045b86c8 Fortran::frontend::CodeGenAction::executeAction() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x45b86c8)
#<!-- -->14 0x0000000003fae89c Fortran::frontend::FrontendAction::execute() (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fae89c)
#<!-- -->15 0x0000000003fa25f4 Fortran::frontend::CompilerInstance::executeAction(Fortran::frontend::FrontendAction&amp;) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fa25f4)
#<!-- -->16 0x0000000003fb3534 Fortran::frontend::executeCompilerInvocation(Fortran::frontend::CompilerInstance*) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3fb3534)
#<!-- -->17 0x0000000003bea120 fc1_main(llvm::ArrayRef&lt;char const*&gt;, char const*) (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3bea120)
#<!-- -->18 0x0000000003b75674 main (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3b75674)
#<!-- -->19 0x0000400024d74384 __libc_start_main (/lib64/libc.so.6+0x24384)
#<!-- -->20 0x0000000003be8f54 _start (/lustre/home/ohno/LLVM_20231130/release/bin/flang-new+0x3be8f54)
flang-new: error: unable to execute command: Segmentation fault (core dumped)
flang-new: error: flang frontend command failed due to signal (use -v to see invocation)
flang-new version 18.0.0 (https://github.com/llvm/llvm-project.git 770dc47659d41a5ca7b7daf5b3134c900ca8c33d)
Target: aarch64-unknown-linux-gnu
Thread model: posix
InstalledDir: /lustre/home/ohno/LLVM_20231130/release/bin
flang-new: note: diagnostic msg:
********************

PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:
Preprocessed source(s) and associated run script(s) are located at:
flang-new: note: diagnostic msg: /tmp/omp_PL_ATOMIC_001c_3-39f821
flang-new: note: diagnostic msg: /tmp/omp_PL_ATOMIC_001c_3-39f821.sh
flang-new: note: diagnostic msg:

********************
$

$ cat /tmp/omp_PL_ATOMIC_001c_3-39f821
#line "./omp_PL_ATOMIC_001c_3.f90" 1
      program main
      complex*8 ia, ib
      integer*4 :: loop=10000
      ia = (0,0)
!$omp parallel
!$omp do
      do i=1, loop
!$omp atomic
         ia = ia + (1,1)
      enddo
!$omp end do
!$omp end parallel
      end
$

$ cat /tmp/omp_PL_ATOMIC_001c_3-39f821.sh
# Crash reproducer for clang version 18.0.0 (https://github.com/llvm/llvm-project.git 770dc47659d41a5ca7b7daf5b3134c900ca8c33d)
# Driver args: "-fopenmp" "omp_PL_ATOMIC_001c_3.f90"
# Original command:  "/lustre/home/ohno/LLVM_20231130/release/bin/flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu" "-emit-obj" "-fopenmp" "-mrelocation-model" "pic" "-pic-level" "2" "-pic-is-pie" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-o" "/tmp/omp_PL_ATOMIC_001c_3-814ed2.o" "-x" "f95-cpp-input" "omp_PL_ATOMIC_001c_3.f90"
 "/lustre/home/ohno/LLVM_20231130/release/bin/flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu" "-emit-obj" "-fopenmp" "-mrelocation-model" "pic" "-pic-level" "2" "-pic-is-pie" "-target-cpu" "generic" "-target-feature" "+neon" "-target-feature" "+v8a" "-x" "f95-cpp-input" "omp_PL_ATOMIC_001c_3-39f821"
$

$ gfortran -fopenmp omp_PL_ATOMIC_001c_3.f90
$

$ ifort -qopenmp omp_PL_ATOMIC_001c_3.f90
$

[kiranchandramohan]

@NimishMishra could you have a look?

[NimishMishra]

@kiranchandramohan Thanks. I'm picking this up.

[NimishMishra]

Hello @kiranchandramohan.
I have done an initial analysis of this issue. The FIR for complex types is generated fine:

omp.atomic.update %1 : !fir.ref<!fir.complex<4>> {
      ^bb0(%arg0 : !fir.complex<4>):
               %2 = fir.addc %arg0, %13 .....
               omp.yield(%14 : !fir.complex<4>)
}

However, during translation, the block argument is the complex type as it is. But the arguments in the operations within the block are simplified using llvm.extract instructions. All in all, the check in

llvm-project/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

Line 1307 in 8e24bc0

if (!llvm::is_contained(innerOp.getOperands(),

fails. Notice that if I use an intrinsic type (like integer), then the block argument is used as-is in the operations, and this check succeeds.

From my end, I tried to analyse the operations that are using the block argument (through getUses()). The returned operations are actually llvm.extractvalue. The final addition is not using the block argument directly, which means we are unable to perform this check. For instance, following is the translation for the given example. It is not completely clear where and how to generate the atomic operation with this.

%13 = llvm.extractvalue %arg0[0] : !llvm.struct<(f32, f32)> 
%14 = llvm.extractvalue %arg0[1] : !llvm.struct<(f32, f32)>
%15 = llvm.extractvalue %12[0] : !llvm.struct<(f32, f32)>
%16 = llvm.extractvalue %12[1] : !llvm.struct<(f32, f32)>
%17 = llvm.fadd %13, %15  {fastmathFlags = #llvm.fastmath<contract>} : f32
%18 = llvm.fadd %14, %16  {fastmathFlags = #llvm.fastmath<contract>} : f32
%19 = llvm.mlir.undef : !llvm.struct<(f32, f32)> 
%20 = llvm.insertvalue %17, %19[0] : !llvm.struct<(f32, f32)>
%21 = llvm.insertvalue %18, %20[1] : !llvm.struct<(f32, f32)>                                                                                                                            
omp.yield(%21 : !llvm.struct<(f32, f32)>)

I also tried to see if we can reuse some the logic we discussed for lowering that essentially ports non-dependent operations outside the block region (%13, %14, %15, and %16 here). But still, it is not clear how to handle %19, %20, and %21. Can you suggest something?

[kiranchandramohan]

The region is going to have more than one operation for Complex types, this will also be the case for intrinsics like min/max that lower to compares and select. So we will need a mechanism to handle when there are more than one operation. I do not have a suggestion now, but will have a look later.

[NimishMishra]

Hello @kiranchandramohan

Thanks for the reply. I was wondering whether it is possible to consider a possible alternative that is functionally equivalent to atomic execution. For example, if we are forced to translate multiple operations, then instead of trying to create an atomic operation for it (which I am not sure how to do at the moment), could we consider enclosing the entire region within runtime locks (like CriticalSection)?

Concretely, during translation, should we detect a correctly formed atomic region, we create an atomic instruction as normal. Otherwise, we rely upon the flang runtime for CriticalSection locks that enclose the entire region, and guarantee functional equivalent to atomic execution.