[VPlan] Construct regions from innermost to outermost (NFC).

Flip the region construction order to innermost first from outermost
first. This ensures we only set the final parent region for VPBBs once.

Split off from #137709.

Author: fhahn

llvm/lib/Transforms/Vectorize/VPlanCFG.h

@@ -16,6 +16,7 @@
 #include "VPlanUtils.h"
 #include "llvm/ADT/DepthFirstIterator.h"
 #include "llvm/ADT/GraphTraits.h"
+#include "llvm/ADT/PostOrderIterator.h"
 #include "llvm/ADT/SmallVector.h"
 
 namespace llvm {
@@ -221,6 +222,14 @@ vp_depth_first_shallow(const VPBlockBase *G) {
   return depth_first(VPBlockShallowTraversalWrapper<const VPBlockBase *>(G));
 }
 
+/// Returns an iterator range to traverse the graph starting at \p G in
+/// post order. The iterator won't traverse through region blocks.
+inline iterator_range<
+    po_iterator<VPBlockShallowTraversalWrapper<VPBlockBase *>>>
+vp_post_order_shallow(VPBlockBase *G) {
+  return post_order(VPBlockShallowTraversalWrapper<VPBlockBase *>(G));
+}
+
 /// Returns an iterator range to traverse the graph starting at \p G in
 /// depth-first order while traversing through region blocks.
 inline iterator_range<df_iterator<VPBlockDeepTraversalWrapper<VPBlockBase *>>>

llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp

@@ -559,7 +559,7 @@ void VPlanTransforms::prepareForVectorization(VPlan &Plan, Type *InductionTy,
 void VPlanTransforms::createLoopRegions(VPlan &Plan) {
   VPDominatorTree VPDT;
   VPDT.recalculate(Plan);
-  for (VPBlockBase *HeaderVPB : vp_depth_first_shallow(Plan.getEntry()))
+  for (VPBlockBase *HeaderVPB : vp_post_order_shallow(Plan.getEntry()))
     if (canonicalHeaderAndLatch(HeaderVPB, VPDT))
       createLoopRegion(Plan, HeaderVPB);