[Mlir-commits] [flang] [mlir] [mlir][flang][OpenMP] Select iterators per locator and allow multi-block regions (PR #227454)

[llvmlistbot at llvm.org]

================
@@ -6101,6 +6119,42 @@ void OmpStructureChecker::CheckArraySection(
   }
 }
 
+// OpenMP 5.2 3.2.5 forbids explicit strides unless the clause permits them.
+// Check only the locator's base chain, not sections in subscript expressions.
+void OmpStructureChecker::CheckArraySectionStrides(
+    const parser::DataRef &dataRef, llvm::omp::Clause clause) {
+  common::visit(
+      common::visitors{
+          [](const parser::Name &) {},
+          [&](const common::Indirection<parser::StructureComponent> &x) {
+            CheckArraySectionStrides(x.value().Base(), clause);
+          },
+          [&](const common::Indirection<parser::ArrayElement> &x) {
+            const auto &arrayElement{x.value()};
+            const auto &name{GetLastName(arrayElement.Base())};
+            // Scalar character substrings can be parsed as ArrayElement.
+            if (name.symbol->Rank() > 0 &&
----------------
MattPD wrote:

An unresolved earlier component leaves `name.symbol` null, so the new walk crashes before Flang emits the component diagnostic:

```fortran
subroutine s(a)
  type t
    integer :: f
  end type
  type(t) :: a
  !$omp task depend(in: a%missing(::2)%f)
  !$omp end task
end
```

At the reviewed commit, `flang -fsyntax-only -fopenmp -fopenmp-version=52 repro.f90` exits 254. At the [parent commit](https://github.com/llvm/llvm-project/commit/5655cabbd7b780c9b3bcfbbb07a619139990f06c), it exits 1 with `Component 'missing' not found in derived type 't'`. AFFINITY and the unstrided `a%missing(:)%f` form have the same regression. Could this walk preserve that diagnostic? Could you also add recovery tests for both clauses?

https://github.com/llvm/llvm-project/pull/227454