Cycle in an OpenMP Do breaks

[wsmoses]

error: loc("./lbm_pre.f90":7:1): block with no terminator, has omp.wsloop (%arg2) : i32 = (%c0_i32) to (%c43199999_i32) inclusive step (%c20_i32) schedule(static)  {
  fir.store %arg2 to %13 : !fir.ref<i32>
  br ^bb1
^bb1:  // pred: ^bb0
  %c19_i32 = arith.constant 19 : i32
  %14 = fir.load %13 : !fir.ref<i32>
  %15 = fir.no_reassoc %14 : i32
  %16 = arith.addi %c19_i32, %15 : i32
  %17 = fir.convert %16 : (i32) -> i64
  %18 = fir.convert %c0 : (index) -> i64
  %19 = arith.subi %17, %18 : i64
  %20 = fir.coordinate_of %arg0, %19 : (!fir.ref<!fir.array<43200000xf32>>, i64) -> !fir.ref<f32>
  %c1_i32 = arith.constant 1 : i32
  fir.store %c1_i32 to %7 : !fir.ref<i32>
  %21 = fir.call @_QPcond(%20, %7) : (!fir.ref<f32>, !fir.ref<i32>) -> !fir.logical<4>
  %22 = fir.convert %21 : (!fir.logical<4>) -> i1
  cond_br %22, ^bb2, ^bb3
^bb2:  // pred: ^bb1
  br ^bb3
^bb3:  // 2 preds: ^bb1, ^bb2
  omp.yield
}

function cond(g, f)
  real :: g
  integer :: i, f
  logical :: cond
  cond = (0 .ne. iand(transfer(g,i), f))
end function
subroutine lbm_sc_atomic(srcGrid, dstGrid)
  implicit none
  real, dimension(0:(150)*(1*(120))*(1*(120))*20 -1) :: srcGrid, dstGrid
  real :: ux, uy, uz, u2, rho
  integer :: u
  integer :: i
  logical :: cond
  !$AD OMP_ADJ ATOMIC(srcGrid)
  !$omp parallel do private( ux, uy, uz, u2, rho ) schedule(static)
  do i = (0)+20*((0)+(0)*(1*(120))+(0)*(1*(120))*(1*(120))), (0)+20*((0)+(0)*(1*(120))+((150))*(1*(120))*(1*(120)))-1, 20
  if( cond(srcGrid((19)+20*((0)+(0)*(1*(120))+(0)*(1*(120))*(1*(120)))+(i)),1)) then
    cycle
  end if
  end do
  return
end subroutine
 subroutine lbm_sc_atomicENZYME(srcGrid, dsrcGrid, dstGrid, ddstGrid)
     CALL ENZYME_AUTODIFF(lbm_sc_atomic, srcGrid, dsrcGrid, dstGrid, ddstGrid)
    return
 end subroutine

[kiranchandramohan]

@wsmoses Thanks for reporting this issue. We have a patch (#1394) that fixes a similar issue. Could you give that a try?

[wsmoses]

Nope, instead gets an internal error

flang-new: warning: argument unused during compilation: '-O3'
flang-new: warning: argument unused during compilation: '-ffree-line-length-none'
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.
Stack dump:
0.	Program arguments: /mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new -fc1 -triple x86_64-unknown-linux-gnu -emit-obj -I ../../../ADFirstAidKit -fopenmp -o lbm_pre.o lbm_pre.f90
 #0 0x000055aa9174b9ff PrintStackTraceSignalHandler(void*) Signals.cpp:0:0
 #1 0x000055aa917492ad SignalHandler(int) Signals.cpp:0:0
 #2 0x00007f736a167980 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x12980)
 #3 0x000055aa934ce680 deleteDeadness(mlir::RewriterBase&, llvm::MutableArrayRef<mlir::Region>, (anonymous namespace)::LiveMap&) RegionUtils.cpp:0:0
 #4 0x000055aa934cdef6 deleteDeadness(mlir::RewriterBase&, llvm::MutableArrayRef<mlir::Region>, (anonymous namespace)::LiveMap&) RegionUtils.cpp:0:0
 #5 0x000055aa934cdef6 deleteDeadness(mlir::RewriterBase&, llvm::MutableArrayRef<mlir::Region>, (anonymous namespace)::LiveMap&) RegionUtils.cpp:0:0
 #6 0x000055aa934d7af0 mlir::simplifyRegions(mlir::RewriterBase&, llvm::MutableArrayRef<mlir::Region>) (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x4aaaaf0)
 #7 0x000055aa91e08957 (anonymous namespace)::FirConverter::lowerFunc(Fortran::lower::pft::FunctionLikeUnit&) Bridge.cpp:0:0
 #8 0x000055aa91dff43a Fortran::lower::LoweringBridge::lower(Fortran::parser::Program const&, Fortran::semantics::SemanticsContext const&) (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x33d243a)
 #9 0x000055aa9177affd Fortran::frontend::CodeGenAction::BeginSourceFileAction() (.part.7736) FrontendActions.cpp:0:0
#10 0x000055aa9176ccae Fortran::frontend::FrontendAction::BeginSourceFile(Fortran::frontend::CompilerInstance&, Fortran::frontend::FrontendInputFile const&) (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x2d3fcae)
#11 0x000055aa91760ae7 Fortran::frontend::CompilerInstance::ExecuteAction(Fortran::frontend::FrontendAction&) (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x2d33ae7)
#12 0x000055aa91840013 Fortran::frontend::ExecuteCompilerInvocation(Fortran::frontend::CompilerInstance*) (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x2e13013)
#13 0x000055aa913b9650 fc1_main(llvm::ArrayRef<char const*>, char const*) (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x298c650)
#14 0x000055aa9135fe0b main (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x2932e0b)
#15 0x00007f7368dfbbf7 __libc_start_main /build/glibc-S9d2JN/glibc-2.27/csu/../csu/libc-start.c:344:0
#16 0x000055aa913b840a _start (/mnt/pci4/wmdata/Enzyme/enzyme/flangdocker/f18-llvm-project/build/bin/flang-new+0x298b40a)
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 14.0.0 (https://github.com/flang-compiler/f18-llvm-project 213452875b08a3d98f3aaeb9862285da4e64c44d)
Target: x86_64-unknown-linux-gnu
Thread model: posix

[kiranchandramohan]

A smaller reproducer from the reported issue can be found below. It is probably the existence of the ContinueStmt (with a new block) that is causing problems. Commenting out the code (given below) introduced in #1178 seems to fix the issue for the small reproducer. Not sure whether it will work for @wsmoses. I will have a look in detail later.

CC: @vdonaldson

diff --git a/flang/lib/Lower/PFTBuilder.cpp b/flang/lib/Lower/PFTBuilder.cpp
index 4c3064f5c60d..be7be8c60549 100644
--- a/flang/lib/Lower/PFTBuilder.cpp
+++ b/flang/lib/Lower/PFTBuilder.cpp
@@ -328,17 +328,17 @@ private:
   void exitConstructOrDirective() {
     rewriteIfGotos();
     auto *eval = constructAndDirectiveStack.back();
-    if (eval->isExecutableDirective()) {
-      // A construct at the end of an (unstructured) OpenACC or OpenMP
-      // construct region must have an exit target inside the region.
-      Fortran::lower::pft::EvaluationList &evaluationList =
-          *eval->evaluationList;
-      if (!evaluationList.empty() && evaluationList.back().isConstruct()) {
-        static const parser::ContinueStmt exitTarget{};
-        addEvaluation(
-            lower::pft::Evaluation{exitTarget, pftParentStack.back(), {}, {}});
-      }
-    }
+    //if (eval->isExecutableDirective()) {
+    //  // A construct at the end of an (unstructured) OpenACC or OpenMP
+    //  // construct region must have an exit target inside the region.
+    //  Fortran::lower::pft::EvaluationList &evaluationList =
+    //      *eval->evaluationList;
+    //  if (!evaluationList.empty() && evaluationList.back().isConstruct()) {
+    //    static const parser::ContinueStmt exitTarget{};
+    //    addEvaluation(
+    //        lower::pft::Evaluation{exitTarget, pftParentStack.back(), {}, {}});
+    //  }
+    //}
     popEvaluationList();
     pftParentStack.pop_back();
     constructAndDirectiveStack.pop_back();

subroutine lbm_sc_atomic(cond)
  implicit none
  logical :: cond
  integer :: i
  !$omp parallel do
  do i = 1, 20
    if( cond) then
      cycle
    end if
  end do
  return
end subroutine

[kiranchandramohan]

This works after llvm@52ca6ad.