[flang] [llvm] [Flang][OpenMP] Fix allocating arrays with size intrinisic (PR #119226)

[Sergio Afonso via llvm-commits]

================
@@ -922,32 +922,70 @@ static void genBodyOfTargetOp(
   while (!valuesDefinedAbove.empty()) {
     for (mlir::Value val : valuesDefinedAbove) {
       mlir::Operation *valOp = val.getDefiningOp();
-      if (mlir::isMemoryEffectFree(valOp)) {
+      assert(valOp != nullptr);
+
+      // NOTE: We skip BoxDimsOp's as the lesser of two evils is to map the
+      // indices separately, as the alternative is to eventually map the Box,
+      // which comes with a fairly large overhead comparatively. We could be
+      // more robust about this and check using a BackWardsSlice to see if we
----------------
skatrak wrote:

```suggestion
      // more robust about this and check using a BackwardsSlice to see if we
```

https://github.com/llvm/llvm-project/pull/119226