[llvm] [LV] Create block in mask up-front if needed. (PR #76635)

[Florian Hahn via llvm-commits]

================
@@ -8659,23 +8663,28 @@ LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(VFRange &Range) {
   bool HasNUW = Style == TailFoldingStyle::None;
   addCanonicalIVRecipes(*Plan, Legal->getWidestInductionType(), HasNUW, DL);
 
-  // Proactively create header mask. Masks for other blocks are created on
-  // demand.
-  RecipeBuilder.createHeaderMask(*Plan);
-
   // Scan the body of the loop in a topological order to visit each basic block
   // after having visited its predecessor basic blocks.
   LoopBlocksDFS DFS(OrigLoop);
   DFS.perform(LI);
 
   VPBasicBlock *VPBB = HeaderVPBB;
+  bool NeedsMasks = CM.foldTailByMasking() ||
+                    any_of(OrigLoop->blocks(), [this](BasicBlock *BB) {
+                      return Legal->blockNeedsPredication(BB);
+                    });
   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);
 
+    if (VPBB == HeaderVPBB)
+      RecipeBuilder.createHeaderMask(*Plan);
----------------
fhahn wrote:

Unfortunately `prepareToFoldTailByMasking` at the moment is used to check if tail-folding is possible *and* to collect the instructions that need masking with tail-folding. 

So even if tail-folding wasn't picked by the cost-model, loads/stores (and others) will be marked as requiring a mask, which will then lead to looking up a header mask. This could be potentially un-tangled, as in #77612

https://github.com/llvm/llvm-project/pull/76635