[llvm] [VPlan] Build initial VPlan 0 using HCFGBuilder for inner loops. (NFC) (PR #124432)

[via llvm-commits]

================
@@ -9329,23 +9330,58 @@ LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(VFRange &Range) {
   RecipeBuilder.collectScaledReductions(Range);
 
   auto *MiddleVPBB = Plan->getMiddleBlock();
+
+  // Scan the body of the loop in a topological order to visit each basic block
+  // after having visited its predecessor basic blocks.
+  ReversePostOrderTraversal<VPBlockShallowTraversalWrapper<VPBlockBase *>> RPOT(
+      HeaderVPBB);
+
   VPBasicBlock::iterator MBIP = MiddleVPBB->getFirstNonPhi();
-  for (BasicBlock *BB : make_range(DFS.beginRPO(), DFS.endRPO())) {
-    // Relevant instructions from basic block BB will be grouped into VPRecipe
-    // ingredients and fill a new VPBasicBlock.
-    if (VPBB != HeaderVPBB)
-      VPBB->setName(BB->getName());
-    Builder.setInsertPoint(VPBB);
+  VPBlockBase *PrevVPBB = nullptr;
+  for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(RPOT)) {
+    // Skip VPBBs not corresponding to any input IR basic blocks.
+    if (!HCFGBuilder.getIRBBForVPB(VPBB)) {
+      assert(VPBB == LoopRegion->getExiting() &&
+             "only the latch block shouldn't have a corresponding IRBB");
+      VPBlockUtils::connectBlocks(PrevVPBB, VPBB);
+      break;
+    }
 
-    if (VPBB == HeaderVPBB)
+    // Create mask based on the IR BB corresponding to VPBB.
+    // TODO: Predicate directly based on VPlan.
+    Builder.setInsertPoint(VPBB, VPBB->begin());
----------------
ayalz wrote:

So non-phi recipes that compute a block's mask must be introduced before the phi recipes of the block, because the latter is eventually replaced by blends that use the former? This calls for a FIXME note.
One alternative may be to sink the blends when they replace the phis.
Another is to introduce masking recipes along with replacing phi recipes with blends, as done currently in tryToBuildVPlanWithRecipes().

https://github.com/llvm/llvm-project/pull/124432