[llvm] [VPlan] Simplify the computation of the block entry mask. (PR #173265)

[Luke Lau via llvm-commits]

================
@@ -128,16 +138,22 @@ VPValue *VPPredicator::createEdgeMask(VPBasicBlock *Src, VPBasicBlock *Dst) {
   return setEdgeMask(Src, Dst, EdgeMask);
 }
 
-VPValue *VPPredicator::createBlockInMask(VPBasicBlock *VPBB) {
+void VPPredicator::createIncomingEdgeMasks(VPBasicBlock *VPBB) {
   // Start inserting after the block's phis, which be replaced by blends later.
   Builder.setInsertPoint(VPBB, VPBB->getFirstNonPhi());
+  for (auto *Predecessor : SetVector<VPBlockBase *>(
----------------
lukel97 wrote:

Ideally we shouldn't need to eagerly create the edge masks and we can instead just change convertPhisToBlends to use createEdgeMask instead of  getEdgeMask to compute them on demand.

But I see the regression you are mentioning, and I've opened up https://github.com/llvm/llvm-project/pull/180898 which should prevent the need to eagerly create the edge masks. We can land it either before after this PR, it's not blocking.

https://github.com/llvm/llvm-project/pull/173265