[Mlir-commits] [mlir] [mlir][OpenMP] don't add compiler-generated barrier in single threaded code (PR #174105)

[Tom Eccles]

================
@@ -1678,6 +1678,25 @@ allocatePrivateVars(llvm::IRBuilderBase &builder,
   return afterAllocas;
 }
 
+// This can't always be determined statically, but when we can, it is good to
+// avoid generating compiler-added barriers which will deadlock the program.
+static bool opIsInSingleThread(mlir::Operation *op) {
+  while (mlir::Operation *parent = op->getParentOp()) {
+    if (mlir::isa<omp::SingleOp, omp::CriticalOp>(parent))
+      return true;
+
+    // e.g.
+    // omp.single {
+    //   omp.parallel {
+    //     op
+    //   }
+    // }
+    if (mlir::isa<omp::ParallelOp>(parent))
----------------
tblah wrote:

No. Because otherwise the example in the comment above would return true.

https://github.com/llvm/llvm-project/pull/174105