[llvm] 44a8d9c - Reapply "[VPlan] Use predicate from VPValue VPWidenSelectR::computeCost." (#173170)

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2025-12-22T22:38:31Z
New Revision: 44a8d9c135f2ba0637d3164abafdac0b5e0da012

URL: https://github.com/llvm/llvm-project/commit/44a8d9c135f2ba0637d3164abafdac0b5e0da012
DIFF: https://github.com/llvm/llvm-project/commit/44a8d9c135f2ba0637d3164abafdac0b5e0da012.diff

LOG: Reapply "[VPlan] Use predicate from VPValue VPWidenSelectR::computeCost." (#173170)

This reverts commit f42af14073228 and re-applies
https://github.com/llvm/llvm-project/pull/172915.

It has an additional check if the condition is a live-in,
which makes sure we preserve the original behavior in that case.

This should fix the crash that caused the revert.

Original commit message:

Instead of looking up the predicate from the VPValue condition instead
of the underlying IR.

This improves cost modeling in some cases, e.g. when we can fold
operations like negations in compares. On AArch64, this leads to
additional vectorization in a few cases in practice.

Example lowering for the modified test case:
https://llvm.godbolt.org/z/6nc6jo5eG

Added: 
    

Modified: 
    llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
    llvm/test/Transforms/LoopVectorize/AArch64/select-costs.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 849448a015fe5..b7deed8fc92ab 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -1964,9 +1964,11 @@ InstructionCost VPWidenSelectRecipe::computeCost(ElementCount VF,
   if (!ScalarCond)
     CondTy = VectorType::get(CondTy, VF);
 
-  CmpInst::Predicate Pred = CmpInst::BAD_ICMP_PREDICATE;
-  if (auto *Cmp = dyn_cast<CmpInst>(SI->getCondition()))
-    Pred = Cmp->getPredicate();
+  llvm::CmpPredicate Pred;
+  if (!match(getOperand(0), m_Cmp(Pred, m_VPValue(), m_VPValue())))
+    if (getOperand(0)->isLiveIn())
+      if (auto *Cmp = dyn_cast<CmpInst>(getOperand(0)->getLiveInIRValue()))
+        Pred = Cmp->getPredicate();
   return Ctx.TTI.getCmpSelInstrCost(
       Instruction::Select, VectorTy, CondTy, Pred, Ctx.CostKind,
       {TTI::OK_AnyValue, TTI::OP_None}, {TTI::OK_AnyValue, TTI::OP_None}, SI);

diff  --git a/llvm/test/Transforms/LoopVectorize/AArch64/select-costs.ll b/llvm/test/Transforms/LoopVectorize/AArch64/select-costs.ll
index 8b1b7b32d1f3e..41675e79a0e4d 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/select-costs.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/select-costs.ll
@@ -92,10 +92,10 @@ exit:
 
 define i32 @select_xor_cond(ptr %src, i1 %c.0) {
 ; CHECK: LV: Checking a loop in 'select_xor_cond'
-; CHECK: Cost of 6 for VF 2: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
-; CHECK: Cost of 12 for VF 4: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
-; CHECK: Cost of 24 for VF 8: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
-; CHECK: Cost of 48 for VF 16: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
+; CHECK: Cost of 1 for VF 2: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
+; CHECK: Cost of 1 for VF 4: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
+; CHECK: Cost of 1 for VF 8: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
+; CHECK: Cost of 1 for VF 16: WIDEN-SELECT ir<%sel> = select ir<%c>, ir<false>, ir<%c.0>
 ; CHECK: LV: Selecting VF: 4.
 
 entry: