[llvm] [LV][AArch64] LoopVectorizer allows scalable frem instructions (PR #76247)

[Graham Hunter via llvm-commits]

================
@@ -6951,10 +6951,26 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I, ElementCount VF,
       Op2Info.Kind = TargetTransformInfo::OK_UniformValue;
 
     SmallVector<const Value *, 4> Operands(I->operand_values());
-    return TTI.getArithmeticInstrCost(
+    auto InstrCost = TTI.getArithmeticInstrCost(
         I->getOpcode(), VectorTy, CostKind,
         {TargetTransformInfo::OK_AnyValue, TargetTransformInfo::OP_None},
         Op2Info, Operands, I);
+
+    // Some targets replace frem with vector library calls.
+    if (I->getOpcode() == Instruction::FRem) {
+      LibFunc Func;
+      if (TLI->getLibFunc(I->getOpcode(), I->getType(), Func)) {
+        if (TLI->isFunctionVectorizable(TLI->getName(Func))) {
+          SmallVector<Type *, 4> OpTypes;
+          for (auto &Op : I->operands())
+            OpTypes.push_back(Op->getType());
+          auto CallCost =
+              TTI.getCallInstrCost(nullptr, VectorTy, OpTypes, CostKind);
+          return std::min(InstrCost, CallCost);
----------------
huntergr-arm wrote:

@mgabka If CallCost is invalid because there's no available mapping or very high because it's suboptimal and the target in question (not AArch64, at least at present) has an frem instruction, we would want to return InstrCost.

If replace-with-veclib were to override that, I think it would be a bug in that pass.

https://github.com/llvm/llvm-project/pull/76247