Add some missing CostKind parameters, add debug message.

Author: john-brawn-arm

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -2941,7 +2941,7 @@ LoopVectorizationCostModel::getVectorCallCost(CallInst *CI,
     Tys.push_back(ArgOp->getType());
 
   InstructionCost ScalarCallCost =
-      TTI.getCallInstrCost(CI->getCalledFunction(), RetTy, Tys);
+      TTI.getCallInstrCost(CI->getCalledFunction(), RetTy, Tys, CostKind);
 
   // If this is an intrinsic we may have a lower cost for it.
   if (getVectorIntrinsicIDForCall(CI, TLI)) {
@@ -6679,7 +6679,8 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I,
       InstructionCost MulCost = TTI::TCC_Free;
       ConstantInt *RHS = dyn_cast<ConstantInt>(I->getOperand(1));
       if (!RHS || RHS->getZExtValue() != 1)
-        MulCost = TTI.getArithmeticInstrCost(Instruction::Mul, VectorTy);
+        MulCost =
+            TTI.getArithmeticInstrCost(Instruction::Mul, VectorTy, CostKind);
 
       // Find the cost of the histogram operation itself.
       Type *PtrTy = VectorType::get(HGram->Load->getPointerOperandType(), VF);
@@ -6691,7 +6692,7 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I,
 
       // Add the costs together with the add/sub operation.
       return TTI.getIntrinsicInstrCost(ICA, CostKind) + MulCost +
-             TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy);
+             TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy, CostKind);
     }
     [[fallthrough]];
   }
@@ -7554,6 +7555,16 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
   if (VPlans.size() == 1 && size(FirstPlan.vectorFactors()) == 1)
     return {*FirstPlan.vectorFactors().begin(), 0, 0};
 
+  LLVM_DEBUG(dbgs() << "LV: Computing best VF using cost kind: "
+                    << (CM.CostKind == TTI::TCK_RecipThroughput
+                            ? "Reciprocal Throughput\n"
+                        : CM.CostKind == TTI::TCK_Latency
+                            ? "Instruction Latency\n"
+                        : CM.CostKind == TTI::TCK_CodeSize ? "Code Size\n"
+                        : CM.CostKind == TTI::TCK_SizeAndLatency
+                            ? "Code Size and Latency\n"
+                            : "Unknown\n"));
+
   ElementCount ScalarVF = ElementCount::getFixed(1);
   assert(hasPlanWithVF(ScalarVF) &&
          "More than a single plan/VF w/o any plan having scalar VF");