[llvm] [VPlan] Dispatch to multiple exit blocks via middle blocks. (PR #112138)

[via llvm-commits]

================
@@ -7850,16 +7859,18 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan(
 
   ILV.printDebugTracesAtEnd();
 
-  // 4. Adjust branch weight of the branch in the middle block.
-  auto *MiddleTerm =
-      cast<BranchInst>(State.CFG.VPBB2IRBB[ExitVPBB]->getTerminator());
-  if (MiddleTerm->isConditional() &&
-      hasBranchWeightMD(*OrigLoop->getLoopLatch()->getTerminator())) {
-    // Assume that `Count % VectorTripCount` is equally distributed.
-    unsigned TripCount = BestVPlan.getUF() * State.VF.getKnownMinValue();
-    assert(TripCount > 0 && "trip count should not be zero");
-    const uint32_t Weights[] = {1, TripCount - 1};
-    setBranchWeights(*MiddleTerm, Weights, /*IsExpected=*/false);
+  // 4. Adjust branch weight of the branch in the middle block if it exists.
----------------
ayalz wrote:

Thanks for adding "if it exists". Comment also meant to question when would the middle block not exist? Should this refer to cases where the middle block (which conceptually is always there, after vector loop and whatever comes nest (exit(s) and/or scalar loop)), rather than being a single block (that ends with a conditional or unconditional branch), is split into multiple blocks as it has more than two successors (rather than targeting them with a switch)?
A null ExitVPBB, OTOH, corresponds to a scalar preheader having more than one predecessor, i.e., it also has runtime guards as predecessors.

https://github.com/llvm/llvm-project/pull/112138