[flang-commits] [flang] [llvm] [mlir] [flang][OpenMP][MLIR] Add MLIR op for loop directive (PR #113911)

[Sergio Afonso via flang-commits]

================
@@ -382,6 +382,50 @@ def LoopNestOp : OpenMP_Op<"loop_nest", traits = [
 // 2.9.2 Workshare Loop Construct
 //===----------------------------------------------------------------------===//
 
+def LoopOp : OpenMP_Op<"loop", traits = [
+    DeclareOpInterfaceMethods<LoopWrapperInterface>, NoTerminator, SingleBlock,
+    AttrSizedOperandSegments
+  ], clauses = [
+    OpenMP_BindClause, OpenMP_PrivateClause, OpenMP_OrderClause,
+    OpenMP_ReductionClause
+  ], singleRegion = true> {
+  let summary = "loop construct";
+  let description = [{
+    A loop construct specifies that the logical iterations of the associated loops
+    may execute concurrently and permits the encountering threads to execute the
+    loop accordingly. A loop construct is a worksharing construct if its binding
+    region is the innermost enclosing parallel region. Otherwise it is not a work-
+    sharing region. The directive asserts that the iterations of the associated
+    loops may execute in any order, including concurrently. Each logical iteration
+    is executed once per instance of the loop region that is encountered by exactly
+    one thread that is a member of the binding thread set.
----------------
skatrak wrote:

Nit: I find the description of the operation a bit unclear, and it somewhat repeats itself. Maybe keep the first sentence and then briefly introduce the parent constructs to which it may bind to, with just a one sentence description of what the differences between binding to teams or parallel are.

https://github.com/llvm/llvm-project/pull/113911