[llvm] [VPlan] Don't use the legacy cost model for loop conditions (PR #156864)

[Florian Hahn via llvm-commits]

================
@@ -7204,6 +7174,55 @@ static bool planContainsAdditionalSimplifications(VPlan &Plan,
     });
   });
 }
+
+static bool planContainsDifferentCompares(VPlan &Plan, VPCostContext &CostCtx,
+                                          Loop *TheLoop, ElementCount VF) {
+  // Count how many compare instructions there are in the legacy cost model.
+  unsigned NumLegacyScalarCompares = 0, NumLegacyVectorCompares = 0;
+  for (BasicBlock *BB : TheLoop->blocks()) {
+    for (auto &I : *BB) {
+      if (isa<CmpInst>(I)) {
+        if (CostCtx.CM.isUniformAfterVectorization(&I, VF))
+          NumLegacyScalarCompares += 1;
+        else
+          NumLegacyVectorCompares += 1;
+      }
+    }
+  }
+
+  // Count how many compare instructions there are in the VPlan.
+  unsigned NumVPlanScalarCompares = 0, NumVPlanVectorCompares = 0;
+  VPRegionBlock *VectorRegion = Plan.getVectorLoopRegion();
+  auto Iter = vp_depth_first_deep(VectorRegion->getEntry());
----------------
fhahn wrote:

do we need a deep traversal, i.e. do we have test cases where we conditionally execute compares in replicate regions?

https://github.com/llvm/llvm-project/pull/156864