[CostModel] Remove optional from InstructionCost::getValue() (#135596)

InstructionCost is already an optional value, containing an Invalid
state that can be checked with isValid(). There is little point in
returning another optional from getValue(). Most uses do not make use of
it being a std::optional, dereferencing the value directly (either
isValid has been checked previously or the Cost is assumed to be valid).
The one case that does in AMDGPU used value_or which has been replaced
by a isValid() check.

Author: davemgreen

llvm/include/llvm/CodeGen/BasicTTIImpl.h

@@ -1610,7 +1610,7 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
 
       // Scale the cost of the load by the fraction of legal instructions that
       // will be used.
-      Cost = divideCeil(UsedInsts.count() * *Cost.getValue(), NumLegalInsts);
+      Cost = divideCeil(UsedInsts.count() * Cost.getValue(), NumLegalInsts);
     }
 
     // Then plus the cost of interleave operation.
@@ -2878,7 +2878,7 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
           SubTp && SubTp->getElementType() == FTp->getElementType())
         return divideCeil(FTp->getNumElements(), SubTp->getNumElements());
     }
-    return *LT.first.getValue();
+    return LT.first.getValue();
   }
 
   InstructionCost getAddressComputationCost(Type *Ty, ScalarEvolution *,

llvm/include/llvm/Support/InstructionCost.h

@@ -20,7 +20,6 @@
 
 #include "llvm/Support/MathExtras.h"
 #include <limits>
-#include <optional>
 
 namespace llvm {
 
@@ -84,10 +83,9 @@ class InstructionCost {
   /// This function is intended to be used as sparingly as possible, since the
   /// class provides the full range of operator support required for arithmetic
   /// and comparisons.
-  std::optional<CostType> getValue() const {
-    if (isValid())
-      return Value;
-    return std::nullopt;
+  CostType getValue() const {
+    assert(isValid());
+    return Value;
   }
 
   /// For all of the arithmetic operators provided here any invalid state is

llvm/include/llvm/Transforms/Utils/UnrollLoop.h

@@ -143,7 +143,7 @@ class UnrollCostEstimator {
   /// Whether it is legal to unroll this loop.
   bool canUnroll() const;
 
-  uint64_t getRolledLoopSize() const { return *LoopSize.getValue(); }
+  uint64_t getRolledLoopSize() const { return LoopSize.getValue(); }
 
   /// Returns loop size estimation for unrolled loop, given the unrolling
   /// configuration specified by UP.

llvm/lib/Analysis/CostModel.cpp

@@ -128,8 +128,8 @@ PreservedAnalyses CostModelPrinterPass::run(Function &F,
       } else {
         InstructionCost Cost =
             getCost(Inst, OutputCostKindToTargetCostKind(CostKind), TTI, TLI);
-        if (auto CostVal = Cost.getValue())
-          OS << "Found an estimated cost of " << *CostVal;
+        if (Cost.isValid())
+          OS << "Found an estimated cost of " << Cost.getValue();
         else
           OS << "Invalid cost";
         OS << " for instruction: " << Inst << "\n";

llvm/lib/CodeGen/SelectOptimize.cpp

@@ -206,7 +206,7 @@ class SelectOptimizeImpl {
           getI()->getOpcode(), I->getType(), TargetTransformInfo::TCK_Latency,
           {TargetTransformInfo::OK_AnyValue, TargetTransformInfo::OP_None},
           {TTI::OK_UniformConstantValue, TTI::OP_PowerOf2});
-      auto TotalCost = Scaled64::get(*Cost.getValue());
+      auto TotalCost = Scaled64::get(Cost.getValue());
       if (auto *OpI = dyn_cast<Instruction>(I->getOperand(1 - CondIdx))) {
         auto It = InstCostMap.find(OpI);
         if (It != InstCostMap.end())
@@ -1380,8 +1380,8 @@ std::optional<uint64_t>
 SelectOptimizeImpl::computeInstCost(const Instruction *I) {
   InstructionCost ICost =
       TTI->getInstructionCost(I, TargetTransformInfo::TCK_Latency);
-  if (auto OC = ICost.getValue())
-    return std::optional<uint64_t>(*OC);
+  if (ICost.isValid())
+    return std::optional<uint64_t>(ICost.getValue());
   return std::nullopt;
 }
 

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -28530,7 +28530,7 @@ bool AArch64TargetLowering::shouldLocalize(
         Imm, CI->getType(), TargetTransformInfo::TCK_CodeSize);
     assert(Cost.isValid() && "Expected a valid imm cost");
 
-    unsigned RematCost = *Cost.getValue();
+    unsigned RematCost = Cost.getValue();
     RematCost += AdditionalCost;
     Register Reg = MI.getOperand(0).getReg();
     unsigned MaxUses = maxUses(RematCost);

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

@@ -4618,7 +4618,7 @@ static bool isLoopSizeWithinBudget(Loop *L, const AArch64TTIImpl &TTI,
   }
 
   if (FinalSize)
-    *FinalSize = *LoopCost.getValue();
+    *FinalSize = LoopCost.getValue();
   return true;
 }
 

llvm/lib/Target/AMDGPU/AMDGPUSplitModule.cpp

@@ -205,8 +205,8 @@ static CostType calculateFunctionCosts(GetTTIFn GetTTI, Module &M,
             TTI.getInstructionCost(&I, TargetTransformInfo::TCK_CodeSize);
         assert(Cost != InstructionCost::getMax());
         // Assume expensive if we can't tell the cost of an instruction.
-        CostType CostVal =
-            Cost.getValue().value_or(TargetTransformInfo::TCC_Expensive);
+        CostType CostVal = Cost.isValid() ? Cost.getValue()
+                                          : TargetTransformInfo::TCC_Expensive;
         assert((FnCost + CostVal) >= FnCost && "Overflow!");
         FnCost += CostVal;
       }

llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp

@@ -1277,9 +1277,9 @@ static unsigned adjustInliningThresholdUsingCallee(const CallBase *CB,
   // The penalty cost is computed relative to the cost of instructions and does
   // not model any storage costs.
   adjustThreshold += std::max(0, SGPRsInUse - NrOfSGPRUntilSpill) *
-                     *ArgStackCost.getValue() * InlineConstants::getInstrCost();
+                     ArgStackCost.getValue() * InlineConstants::getInstrCost();
   adjustThreshold += std::max(0, VGPRsInUse - NrOfVGPRUntilSpill) *
-                     *ArgStackCost.getValue() * InlineConstants::getInstrCost();
+                     ArgStackCost.getValue() * InlineConstants::getInstrCost();
   return adjustThreshold;
 }
 

llvm/lib/Target/PowerPC/PPCTargetTransformInfo.cpp

@@ -1096,7 +1096,7 @@ InstructionCost PPCTTIImpl::getVPMemoryOpCost(unsigned Opcode, Type *Src,
     float AlignmentProb = ((float)Alignment.value()) / DesiredAlignment.value();
     float MisalignmentProb = 1.0 - AlignmentProb;
     return (MisalignmentProb * P9PipelineFlushEstimate) +
-           (AlignmentProb * *Cost.getValue());
+           (AlignmentProb * Cost.getValue());
   }
 
   // Usually we should not get to this point, but the following is an attempt to

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -2909,7 +2909,7 @@ InstructionCost RISCVTargetLowering::getVRGatherVVCost(MVT VT) const {
   bool Log2CostModel =
       Subtarget.getVRGatherCostModel() == llvm::RISCVSubtarget::NLog2N;
   if (Log2CostModel && LMULCost.isValid()) {
-    unsigned Log = Log2_64(*LMULCost.getValue());
+    unsigned Log = Log2_64(LMULCost.getValue());
     if (Log > 0)
       return LMULCost * Log;
   }

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

@@ -483,7 +483,7 @@ costShuffleViaVRegSplitting(const RISCVTTIImpl &TTI, MVT LegalVT,
   auto *SingleOpTy = FixedVectorType::get(Tp->getElementType(),
                                           LegalVT.getVectorNumElements());
 
-  unsigned E = *NumOfDests.getValue();
+  unsigned E = NumOfDests.getValue();
   unsigned NormalizedVF =
       LegalVT.getVectorNumElements() * std::max(NumOfSrcs, E);
   unsigned NumOfSrcRegs = NormalizedVF / LegalVT.getVectorNumElements();

llvm/lib/Target/SystemZ/SystemZTargetTransformInfo.cpp

@@ -380,7 +380,7 @@ void SystemZTTIImpl::getUnrollingPreferences(
   // The z13 processor will run out of store tags if too many stores
   // are fed into it too quickly. Therefore make sure there are not
   // too many stores in the resulting unrolled loop.
-  unsigned const NumStoresVal = *NumStores.getValue();
+  unsigned const NumStoresVal = NumStores.getValue();
   unsigned const Max = (NumStoresVal ? (12 / NumStoresVal) : UINT_MAX);
 
   if (HasCall) {

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

@@ -1748,7 +1748,7 @@ InstructionCost X86TTIImpl::getShuffleCost(
             getTypeLegalizationCost(
                 FixedVectorType::get(BaseTp->getElementType(), Mask.size()))
                 .first;
-        unsigned E = *NumOfDests.getValue();
+        unsigned E = NumOfDests.getValue();
         unsigned NormalizedVF =
             LegalVT.getVectorNumElements() * std::max(NumOfSrcs, E);
         unsigned NumOfSrcRegs = NormalizedVF / LegalVT.getVectorNumElements();
@@ -4931,7 +4931,7 @@ InstructionCost X86TTIImpl::getScalarizationOverhead(
           (LegalVectorBitWidth % LaneBitWidth) == 0) &&
          "Illegal vector");
 
-  const int NumLegalVectors = *LT.first.getValue();
+  const int NumLegalVectors = LT.first.getValue();
   assert(NumLegalVectors >= 0 && "Negative cost!");
 
   // For insertions, a ISD::BUILD_VECTOR style vector initialization can be much
@@ -6164,7 +6164,7 @@ InstructionCost X86TTIImpl::getGSVectorCost(unsigned Opcode,
   std::pair<InstructionCost, MVT> IdxsLT = getTypeLegalizationCost(IndexVTy);
   std::pair<InstructionCost, MVT> SrcLT = getTypeLegalizationCost(SrcVTy);
   InstructionCost::CostType SplitFactor =
-      *std::max(IdxsLT.first, SrcLT.first).getValue();
+      std::max(IdxsLT.first, SrcLT.first).getValue();
   if (SplitFactor > 1) {
     // Handle splitting of vector of pointers
     auto *SplitSrcTy =

llvm/lib/Transforms/IPO/FunctionSpecialization.cpp

@@ -662,7 +662,7 @@ FunctionSpecializer::~FunctionSpecializer() {
 /// non-negative, which is true for both TCK_CodeSize and TCK_Latency, and
 /// always Valid.
 static unsigned getCostValue(const Cost &C) {
-  int64_t Value = *C.getValue();
+  int64_t Value = C.getValue();
 
   assert(Value >= 0 && "CodeSize and Latency cannot be negative");
   // It is safe to down cast since we know the arguments cannot be negative and
@@ -713,7 +713,7 @@ bool FunctionSpecializer::run() {
     if (!SpecializeLiteralConstant && !Inserted && !Metrics.isRecursive)
       continue;
 
-    int64_t Sz = *Metrics.NumInsts.getValue();
+    int64_t Sz = Metrics.NumInsts.getValue();
     assert(Sz > 0 && "CodeSize should be positive");
     // It is safe to down cast from int64_t, NumInsts is always positive.
     unsigned FuncSize = static_cast<unsigned>(Sz);

llvm/lib/Transforms/IPO/PartialInlining.cpp

@@ -1320,7 +1320,7 @@ bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
     RelativeToEntryFreq = BranchProbability(0, 1);
 
   BlockFrequency WeightedRcost =
-      BlockFrequency(*NonWeightedRcost.getValue()) * RelativeToEntryFreq;
+      BlockFrequency(NonWeightedRcost.getValue()) * RelativeToEntryFreq;
 
   // The call sequence(s) to the outlined function(s) are larger than the sum of
   // the original outlined region size(s), it does not increase the chances of

llvm/lib/Transforms/Scalar/ConstantHoisting.cpp

@@ -386,7 +386,7 @@ void ConstantHoistingPass::collectConstantCandidates(
       ConstIntCandVec.push_back(ConstantCandidate(ConstInt));
       Itr->second = ConstIntCandVec.size() - 1;
     }
-    ConstIntCandVec[Itr->second].addUser(Inst, Idx, *Cost.getValue());
+    ConstIntCandVec[Itr->second].addUser(Inst, Idx, Cost.getValue());
     LLVM_DEBUG(if (isa<ConstantInt>(Inst->getOperand(Idx))) dbgs()
                    << "Collect constant " << *ConstInt << " from " << *Inst
                    << " with cost " << Cost << '\n';
@@ -446,7 +446,7 @@ void ConstantHoistingPass::collectConstantCandidates(
         ConstExpr));
     Itr->second = ExprCandVec.size() - 1;
   }
-  ExprCandVec[Itr->second].addUser(Inst, Idx, *Cost.getValue());
+  ExprCandVec[Itr->second].addUser(Inst, Idx, Cost.getValue());
 }
 
 /// Check the operand for instruction Inst at index Idx.

llvm/lib/Transforms/Scalar/LoopDataPrefetch.cpp

@@ -304,7 +304,7 @@ bool LoopDataPrefetch::runOnLoop(Loop *L) {
   if (!Metrics.NumInsts.isValid())
     return MadeChange;
 
-  unsigned LoopSize = *Metrics.NumInsts.getValue();
+  unsigned LoopSize = Metrics.NumInsts.getValue();
   if (!LoopSize)
     LoopSize = 1;
 

llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp

@@ -1535,7 +1535,7 @@ void Cost::RateFormula(const Formula &F,
   C.NumBaseAdds += (F.UnfoldedOffset.isNonZero());
 
   // Accumulate non-free scaling amounts.
-  C.ScaleCost += *getScalingFactorCost(*TTI, LU, F, *L).getValue();
+  C.ScaleCost += getScalingFactorCost(*TTI, LU, F, *L).getValue();
 
   // Tally up the non-zero immediates.
   for (const LSRFixup &Fixup : LU.Fixups) {

llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp

@@ -677,8 +677,8 @@ static std::optional<EstimatedUnrollCost> analyzeLoopUnrollCost(
   LLVM_DEBUG(dbgs() << "Analysis finished:\n"
                     << "UnrolledCost: " << UnrolledCost << ", "
                     << "RolledDynamicCost: " << RolledDynamicCost << "\n");
-  return {{unsigned(*UnrolledCost.getValue()),
-           unsigned(*RolledDynamicCost.getValue())}};
+  return {{unsigned(UnrolledCost.getValue()),
+           unsigned(RolledDynamicCost.getValue())}};
 }
 
 UnrollCostEstimator::UnrollCostEstimator(
@@ -729,7 +729,7 @@ bool UnrollCostEstimator::canUnroll() const {
 uint64_t UnrollCostEstimator::getUnrolledLoopSize(
     const TargetTransformInfo::UnrollingPreferences &UP,
     unsigned CountOverwrite) const {
-  unsigned LS = *LoopSize.getValue();
+  unsigned LS = LoopSize.getValue();
   assert(LS >= UP.BEInsns && "LoopSize should not be less than BEInsns!");
   if (CountOverwrite)
     return static_cast<uint64_t>(LS - UP.BEInsns) * CountOverwrite + UP.BEInsns;

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -2002,7 +2002,7 @@ class GeneratedRTChecks {
           InstructionCost NewMemCheckCost = MemCheckCost / BestTripCount;
 
           // Let's ensure the cost is always at least 1.
-          NewMemCheckCost = std::max(*NewMemCheckCost.getValue(),
+          NewMemCheckCost = std::max(NewMemCheckCost.getValue(),
                                      (InstructionCost::CostType)1);
 
           if (BestTripCount > 1)
@@ -5314,7 +5314,7 @@ LoopVectorizationCostModel::selectInterleaveCount(VPlan &Plan, ElementCount VF,
     // to estimate the cost of the loop and interleave until the cost of the
     // loop overhead is about 5% of the cost of the loop.
     unsigned SmallIC = std::min(IC, (unsigned)llvm::bit_floor<uint64_t>(
-                                        SmallLoopCost / *LoopCost.getValue()));
+                                        SmallLoopCost / LoopCost.getValue()));
 
     // Interleave until store/load ports (estimated by max interleave count) are
     // saturated.
@@ -7659,7 +7659,7 @@ InstructionCost LoopVectorizationPlanner::cost(VPlan &Plan,
   LLVM_DEBUG(dbgs() << "Cost for VF " << VF << ": " << Cost
                     << " (Estimated cost per lane: ");
   if (Cost.isValid()) {
-    double CostPerLane = double(*Cost.getValue()) / EstimatedWidth;
+    double CostPerLane = double(Cost.getValue()) / EstimatedWidth;
     LLVM_DEBUG(dbgs() << format("%.1f", CostPerLane));
   } else /* No point dividing an invalid cost - it will still be invalid */
     LLVM_DEBUG(dbgs() << "Invalid");
@@ -10478,7 +10478,7 @@ static bool isOutsideLoopWorkProfitable(GeneratedRTChecks &Checks,
 
   // The scalar cost should only be 0 when vectorizing with a user specified
   // VF/IC. In those cases, runtime checks should always be generated.
-  uint64_t ScalarC = *VF.ScalarCost.getValue();
+  uint64_t ScalarC = VF.ScalarCost.getValue();
   if (ScalarC == 0)
     return true;
 
@@ -10513,8 +10513,8 @@ static bool isOutsideLoopWorkProfitable(GeneratedRTChecks &Checks,
   // the computations are performed on doubles, not integers and the result
   // is rounded up, hence we get an upper estimate of the TC.
   unsigned IntVF = getEstimatedRuntimeVF(VF.Width, VScale);
-  uint64_t RtC = *TotalCost.getValue();
-  uint64_t Div = ScalarC * IntVF - *VF.Cost.getValue();
+  uint64_t RtC = TotalCost.getValue();
+  uint64_t Div = ScalarC * IntVF - VF.Cost.getValue();
   uint64_t MinTC1 = Div == 0 ? 0 : divideCeil(RtC * IntVF, Div);
 
   // Second, compute a minimum iteration count so that the cost of the

llvm/unittests/Support/InstructionCostTest.cpp

@@ -23,7 +23,7 @@ TEST_F(CostTest, DefaultCtor) {
   InstructionCost DefaultCost;
 
   ASSERT_TRUE(DefaultCost.isValid());
-  EXPECT_EQ(*(DefaultCost.getValue()), 0);
+  EXPECT_EQ(DefaultCost.getValue(), 0);
 }
 
 TEST_F(CostTest, Operators) {
@@ -70,8 +70,7 @@ TEST_F(CostTest, Operators) {
   EXPECT_FALSE(TmpCost.isValid());
 
   // Test value extraction
-  EXPECT_EQ(*(VThree.getValue()), 3);
-  EXPECT_EQ(IThreeA.getValue(), std::nullopt);
+  EXPECT_EQ(VThree.getValue(), 3);
 
   EXPECT_EQ(std::min(VThree, VNegTwo), -2);
   EXPECT_EQ(std::max(VThree, VSix), 6);