[LV][NFC] Fix the condition for printing debug messages

[fhahn]

Would it be possible to add a test case?

[Mel-Chen]

Add a test case

[fhahn]

Thanks for the update! Just a few comments with regard to the test.

LGTM, with the suggested cleanup for the test. It would also be good to pre-commit the test and then have the diff only show the change caused by the patch.

llvm/test/Transforms/LoopVectorize/RISCV/force-vect-msg.ll

14

Given that the LLVM IR is very compact I don't think the C code adds much and IMO it would be better to not include it.

27

Keep the comment next to the check?

34

this should not be needed, you can branch directly to either %for.body.preheader or %for.body.

40

IMO the test is easier to read if this exit block is at the end of the function

[Mel-Chen]

Update test case according to the comments.

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

#endif
   if (!EnableCondStoresVectorization && NumPredStores) {
     reportVectorizationFailure("There are conditional stores.",
         "store that is conditionally executed prevents vectorization",
         "ConditionalStore", ORE, TheLoop);
     ChosenFactor = ScalarCost;
   }
 
   LLVM_DEBUG(if (ForceVectorization && !ChosenFactor.Width.isScalar() &&
-                 ChosenFactor.Cost >= ScalarCost.Cost) dbgs()
+                 !isMoreProfitable(ChosenFactor, ScalarCost)) dbgs()
              << "LV: Vectorization seems to be not beneficial, "
              << "but was forced by a user.\n");
   LLVM_DEBUG(dbgs() << "LV: Selecting VF: " << ChosenFactor.Width << ".\n");
   return ChosenFactor;
 }
 
 bool LoopVectorizationCostModel::isCandidateForEpilogueVectorization(
     const Loop &L, ElementCount VF) const {

llvm/test/Transforms/LoopVectorize/RISCV/force-vect-msg.ll

 ; REQUIRES: asserts
 ; RUN: opt < %s -loop-vectorize -mtriple riscv64 -riscv-v-vector-bits-min=128 -mattr="+v" -debug-only=loop-vectorize -S 2>&1 | FileCheck %s
 
 ; CHECK: LV: Loop hints: force=enabled
 ; CHECK: LV: Scalar loop costs: 7.
+; ChosenFactor.Cost is 8, but the real cost will be divided by the width, which is 4.
 ; CHECK: LV: Vector loop of width 2 costs: 4.
-; CHECK: LV: Vectorization seems to be not beneficial, but was forced by a user.
+; Regardless of force vectorization or not, this loop will eventually be vectorized because of the cost model.
+; Therefore, the following message does not need to be printed even if vectorization is explicitly forced in the metadata.
+; CHECK-NOT: LV: Vectorization seems to be not beneficial, but was forced by a user.
 
 target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128"
 target triple = "riscv64-unknown-unknown"
 
 define i64 @foo(ptr nocapture noundef readonly %a, i64 noundef %N, i64 noundef %init) {
 entry:
   br label %for.body
 
 for.body:                                         ; preds = %entry, %for.body
   %i.06 = phi i64 [ %inc, %for.body ], [ 0, %entry ]
   %rd.05 = phi i64 [ %add, %for.body ], [ %init, %entry ]
   %arrayidx = getelementptr inbounds i64, ptr %a, i64 %i.06
   %0 = load i64, ptr %arrayidx, align 8
   %add = add nsw i64 %0, %rd.05
   %inc = add nuw i64 %i.06, 1
   %exitcond.not = icmp eq i64 %inc, %N
   br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0
 
 for.end:                                          ; preds = %for.body
   ret i64 %add
 }
 
 !0 = !{!0, !1}
 !1 = !{!"llvm.loop.vectorize.enable", i1 true}