[flang][OpenMP] Don't check unlabelled cycle branching for target loops

[ergawy]

Properly handles cycle branching inside target distribute loops.

[llvmbot]

@llvm/pr-subscribers-flang-openmp

@llvm/pr-subscribers-flang-semantics

Author: Kareem Ergawy (ergawy)

Changes

Properly handles cycle branching inside target distribute loops.

---

Full diff: https://github.com/llvm/llvm-project/pull/111656.diff

2 Files Affected:

• (modified) flang/lib/Semantics/check-directive-structure.h (+2)
• (modified) flang/test/Semantics/OpenMP/do05-positivecase.f90 (+15)

diff --git a/flang/lib/Semantics/check-directive-structure.h b/flang/lib/Semantics/check-directive-structure.h
index a1aff52f3a6843..ee360161d351e6 100644
--- a/flang/lib/Semantics/check-directive-structure.h
+++ b/flang/lib/Semantics/check-directive-structure.h
@@ -74,6 +74,8 @@ template <typename D> class NoBranchingEnforce {
         case llvm::omp::Directive::OMPD_distribute_parallel_for:
         case llvm::omp::Directive::OMPD_distribute_simd:
         case llvm::omp::Directive::OMPD_distribute_parallel_for_simd:
+        case llvm::omp::Directive::OMPD_target_teams_distribute_parallel_do:
+        case llvm::omp::Directive::OMPD_target_teams_distribute_parallel_do_simd:
           return;
         default:
           break;
diff --git a/flang/test/Semantics/OpenMP/do05-positivecase.f90 b/flang/test/Semantics/OpenMP/do05-positivecase.f90
index 3b512a5b4f25eb..5e1b1b86f72f61 100644
--- a/flang/test/Semantics/OpenMP/do05-positivecase.f90
+++ b/flang/test/Semantics/OpenMP/do05-positivecase.f90
@@ -42,4 +42,19 @@ program omp_do
   end do
   !$omp end parallel
 
+  !$omp target teams distribute parallel do
+  !DEF:/omp_do/OtherConstruct4/i (OmpPrivate ,OmpPreDetermined) HostAssoc INTEGER(4)
+  do i=1,100
+    !REF:/omp_do/OtherConstruct4/i
+    if(i<10) cycle
+  end do
+  !$omp end target teams distribute parallel do
+
+  !$omp target teams distribute parallel do simd
+  !DEF:/omp_do/OtherConstruct5/i (OmpLinear,OmpPreDetermined) HostAssoc INTEGER(4)
+  do i=1,100
+    !REF:/omp_do/OtherConstruct5/i
+    if(i<10) cycle
+  end do
+  !$omp end target teams distribute parallel do simd
 end program omp_do

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[NimishMishra]

Looks good. Thanks

[mjklemm]

LGTM

[mjklemm]

Stray formatting change?

[ergawy]

Nope, our formatting rules demand this.