[CallSite removal] Remove CallSite uses from InstCombine.

[craig.topper]

lib/Transforms/InstCombine/InstCombineCalls.cpp

4263

Wasn't sure what to do here. isMustTailCall is inlined in CallSite, but it can't be inlined in CallBase.

[chandlerc]

Minor nits. I've been using Call instead of CB because when there is a nice word instead of initialism, I prefer it for readability. But that's up to you really.

lib/Transforms/InstCombine/InstCombineCalls.cpp

3984

A const CallSite isn't actually the same as a const CallBase & -- that would be an ImmutableCallSite.

Still, if the const works here, great.

4263

You can just define it out-of-line. I don't think these *need* to be inlined, and in Instructions.cpp we have all the types available.

4530

Can use clang-format here. Also good candidate for auto *NestF.

[chandlerc]

(also LGTM with the updates above and any renaming you want to do)

include/llvm/IR/InstrTypes.h

bool isDataOperand(const Use *U) const {
     assert(this == U->getUser() &&
            "Only valid to query with a use of this instruction!");
     return data_operands_begin() <= U && U < data_operands_end();
   }
   bool isDataOperand(Value::const_user_iterator UI) const {
     return isDataOperand(&UI.getUse());
   }
 
+  /// Given a value use iterator, return the data operand corresponding to it.
+  /// Iterator must actually correspond to a data operand.
+  unsigned getDataOperandNo(Value::const_user_iterator UI) const {
+    return getDataOperandNo(&UI.getUse());
+  }
+
+  /// Given a use for a data operand, get the data operand number that
+  /// corresponds to it.
+  unsigned getDataOperandNo(const Use *U) const {
+    assert(isDataOperand(U) && "Data operand # out of range!");
+    return U - data_operands_begin();
+  }
+
   /// Return the iterator pointing to the beginning of the argument list.
   User::op_iterator arg_begin() { return op_begin(); }
   User::const_op_iterator arg_begin() const {
     return const_cast<CallBase *>(this)->arg_begin();
   }
 
   /// Return the iterator pointing to the end of the argument list.
   User::op_iterator arg_end() {

lib/Transforms/InstCombine/InstCombineCalls.cpp

 #include "llvm/ADT/Twine.h"
 #include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/Attributes.h"
 #include "llvm/IR/BasicBlock.h"
-#include "llvm/IR/CallSite.h"
 #include "llvm/IR/Constant.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalVariable.h"
 #include "llvm/IR/InstrTypes.h"
 #include "llvm/IR/Instruction.h"
if (isa<Constant>(Arg0) && !isa<Constant>(Arg1)) {
     Call.setArgOperand(0, Arg1);
     Call.setArgOperand(1, Arg0);
     return &Call;
   }
   return nullptr;
 }
 
 /// CallInst simplification. This mostly only handles folding of intrinsic
-/// instructions. For normal calls, it allows visitCallSite to do the heavy
+/// instructions. For normal calls, it allows visitCallBase to do the heavy
 /// lifting.
 Instruction *InstCombiner::visitCallInst(CallInst &CI) {
   if (Value *V = SimplifyCall(&CI, SQ.getWithInstruction(&CI)))
     return replaceInstUsesWith(CI, V);
 
   if (isFreeCall(&CI, &TLI))
     return visitFree(CI);
 
   // If the caller function is nounwind, mark the call as nounwind, even if the
   // callee isn't.
   if (CI.getFunction()->doesNotThrow() && !CI.doesNotThrow()) {
     CI.setDoesNotThrow();
     return &CI;
   }
 
   IntrinsicInst *II = dyn_cast<IntrinsicInst>(&CI);
-  if (!II) return visitCallSite(&CI);
+  if (!II) return visitCallBase(CI);
 
   // Intrinsics cannot occur in an invoke, so handle them here instead of in
-  // visitCallSite.
+  // visitCallBase.
   if (auto *MI = dyn_cast<AnyMemIntrinsic>(II)) {
     bool Changed = false;
 
     // memmove/cpy/set of zero bytes is a noop.
     if (Constant *NumBytes = dyn_cast<Constant>(MI->getLength())) {
       if (NumBytes->isNullValue())
         return eraseInstFromFunction(CI);
 
if (match(NextInst,
         Temp->moveBefore(II);
       }
       II->setArgOperand(0, Builder.CreateAnd(CurrCond, NextCond));
       return eraseInstFromFunction(*NextInst);
     }
     break;
   }
   }
-  return visitCallSite(II);
+  return visitCallBase(*II);
 }
 
 // Fence instruction simplification
 Instruction *InstCombiner::visitFenceInst(FenceInst &FI) {
   // Remove identical consecutive fences.
   Instruction *Next = FI.getNextNonDebugInstruction();
   if (auto *NFI = dyn_cast<FenceInst>(Next))
     if (FI.isIdenticalTo(NFI))
       return eraseInstFromFunction(FI);
   return nullptr;
 }
 
 // InvokeInst simplification
 Instruction *InstCombiner::visitInvokeInst(InvokeInst &II) {
-  return visitCallSite(&II);
+  return visitCallBase(II);
 }
 
 /// If this cast does not affect the value passed through the varargs area, we
 /// can eliminate the use of the cast.
-static bool isSafeToEliminateVarargsCast(const CallSite CS,
+static bool isSafeToEliminateVarargsCast(const CallBase &CB,
                                          const DataLayout &DL,
                                          const CastInst *const CI,
                                          const int ix) {
   if (!CI->isLosslessCast())
     return false;
 
   // If this is a GC intrinsic, avoid munging types.  We need types for
   // statepoint reconstruction in SelectionDAG.
   // TODO: This is probably something which should be expanded to all
   // intrinsics since the entire point of intrinsics is that
   // they are understandable by the optimizer.
-  if (isStatepoint(CS) || isGCRelocate(CS) || isGCResult(CS))
+  if (isStatepoint(&CB) || isGCRelocate(&CB) || isGCResult(&CB))
     return false;
 
   // The size of ByVal or InAlloca arguments is derived from the type, so we
   // can't change to a type with a different size.  If the size were
   // passed explicitly we could avoid this check.
-  if (!CS.isByValOrInAllocaArgument(ix))
+  if (!CB.isByValOrInAllocaArgument(ix))
     return true;
 
   Type* SrcTy =
             cast<PointerType>(CI->getOperand(0)->getType())->getElementType();
   Type* DstTy = cast<PointerType>(CI->getType())->getElementType();
   if (!SrcTy->isSized() || !DstTy->isSized())
     return false;
   if (DL.getTypeAllocSize(SrcTy) != DL.getTypeAllocSize(DstTy))
static IntrinsicInst *findInitTrampoline(Value *Callee) {
   if (IntrinsicInst *IT = findInitTrampolineFromAlloca(TrampMem))
     return IT;
   if (IntrinsicInst *IT = findInitTrampolineFromBB(AdjustTramp, TrampMem))
     return IT;
   return nullptr;
 }
 
 /// Improvements for call and invoke instructions.
-Instruction *InstCombiner::visitCallSite(CallSite CS) {
-  if (isAllocLikeFn(CS.getInstruction(), &TLI))
-    return visitAllocSite(*CS.getInstruction());
+Instruction *InstCombiner::visitCallBase(CallBase &CB) {
+  if (isAllocLikeFn(&CB, &TLI))
+    return visitAllocSite(CB);
 
   bool Changed = false;
 
   // Mark any parameters that are known to be non-null with the nonnull
   // attribute.  This is helpful for inlining calls to functions with null
   // checks on their arguments.
   SmallVector<unsigned, 4> ArgNos;
   unsigned ArgNo = 0;
 
-  for (Value *V : CS.args()) {
+  for (Value *V : CB.args()) {
     if (V->getType()->isPointerTy() &&
-        !CS.paramHasAttr(ArgNo, Attribute::NonNull) &&
-        isKnownNonZero(V, DL, 0, &AC, CS.getInstruction(), &DT))
+        !CB.paramHasAttr(ArgNo, Attribute::NonNull) &&
+        isKnownNonZero(V, DL, 0, &AC, &CB, &DT))
       ArgNos.push_back(ArgNo);
     ArgNo++;
   }
 
-  assert(ArgNo == CS.arg_size() && "sanity check");
+  assert(ArgNo == CB.arg_size() && "sanity check");
 
   if (!ArgNos.empty()) {
-    AttributeList AS = CS.getAttributes();
-    LLVMContext &Ctx = CS.getInstruction()->getContext();
+    AttributeList AS = CB.getAttributes();
+    LLVMContext &Ctx = CB.getContext();
     AS = AS.addParamAttribute(Ctx, ArgNos,
                               Attribute::get(Ctx, Attribute::NonNull));
-    CS.setAttributes(AS);
+    CB.setAttributes(AS);
     Changed = true;
   }
 
   // If the callee is a pointer to a function, attempt to move any casts to the
   // arguments of the call/invoke.
-  Value *Callee = CS.getCalledValue();
-  if (!isa<Function>(Callee) && transformConstExprCastCall(CS))
+  Value *Callee = CB.getCalledValue();
+  if (!isa<Function>(Callee) && transformConstExprCastCall(CB))
     return nullptr;
 
   if (Function *CalleeF = dyn_cast<Function>(Callee)) {
     // Remove the convergent attr on calls when the callee is not convergent.
-    if (CS.isConvergent() && !CalleeF->isConvergent() &&
+    if (CB.isConvergent() && !CalleeF->isConvergent() &&
         !CalleeF->isIntrinsic()) {
       LLVM_DEBUG(dbgs() << "Removing convergent attr from instr "
-                        << CS.getInstruction() << "\n");
-      CS.setNotConvergent();
-      return CS.getInstruction();
+                        << CB << "\n");
+      CB.setNotConvergent();
+      return &CB;
     }
 
     // If the call and callee calling conventions don't match, this call must
     // be unreachable, as the call is undefined.
-    if (CalleeF->getCallingConv() != CS.getCallingConv() &&
+    if (CalleeF->getCallingConv() != CB.getCallingConv() &&
         // Only do this for calls to a function with a body.  A prototype may
         // not actually end up matching the implementation's calling conv for a
         // variety of reasons (e.g. it may be written in assembly).
         !CalleeF->isDeclaration()) {
-      Instruction *OldCall = CS.getInstruction();
+      Instruction *OldCall = &CB;
       new StoreInst(ConstantInt::getTrue(Callee->getContext()),
                 UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
                                   OldCall);
       // If OldCall does not return void then replaceAllUsesWith undef.
       // This allows ValueHandlers and custom metadata to adjust itself.
       if (!OldCall->getType()->isVoidTy())
         replaceInstUsesWith(*OldCall, UndefValue::get(OldCall->getType()));
       if (isa<CallInst>(OldCall))
         return eraseInstFromFunction(*OldCall);
 
       // We cannot remove an invoke, because it would change the CFG, just
       // change the callee to a null pointer.
       cast<InvokeInst>(OldCall)->setCalledFunction(
                                     Constant::getNullValue(CalleeF->getType()));
       return nullptr;
     }
   }
 
   if ((isa<ConstantPointerNull>(Callee) &&
-       !NullPointerIsDefined(CS.getInstruction()->getFunction())) ||
+       !NullPointerIsDefined(CB.getFunction())) ||
       isa<UndefValue>(Callee)) {
-    // If CS does not return void then replaceAllUsesWith undef.
+    // If CB does not return void then replaceAllUsesWith undef.
     // This allows ValueHandlers and custom metadata to adjust itself.
-    if (!CS.getInstruction()->getType()->isVoidTy())
-      replaceInstUsesWith(*CS.getInstruction(),
-                          UndefValue::get(CS.getInstruction()->getType()));
+    if (!CB.getType()->isVoidTy())
+      replaceInstUsesWith(CB,
+                          UndefValue::get(CB.getType()));
 
-    if (isa<InvokeInst>(CS.getInstruction())) {
+    if (isa<InvokeInst>(CB)) {
       // Can't remove an invoke because we cannot change the CFG.
       return nullptr;
     }
 
     // This instruction is not reachable, just remove it.  We insert a store to
     // undef so that we know that this code is not reachable, despite the fact
     // that we can't modify the CFG here.
     new StoreInst(ConstantInt::getTrue(Callee->getContext()),
                   UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
-                  CS.getInstruction());
+                  &CB);
 
-    return eraseInstFromFunction(*CS.getInstruction());
+    return eraseInstFromFunction(CB);
   }
 
   if (IntrinsicInst *II = findInitTrampoline(Callee))
-    return transformCallThroughTrampoline(CS, II);
+    return transformCallThroughTrampoline(CB, *II);
 
   PointerType *PTy = cast<PointerType>(Callee->getType());
   FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
   if (FTy->isVarArg()) {
     int ix = FTy->getNumParams();
     // See if we can optimize any arguments passed through the varargs area of
     // the call.
-    for (CallSite::arg_iterator I = CS.arg_begin() + FTy->getNumParams(),
-           E = CS.arg_end(); I != E; ++I, ++ix) {
+    for (auto I = CB.arg_begin() + FTy->getNumParams(),
+           E = CB.arg_end(); I != E; ++I, ++ix) {
       CastInst *CI = dyn_cast<CastInst>(*I);
-      if (CI && isSafeToEliminateVarargsCast(CS, DL, CI, ix)) {
+      if (CI && isSafeToEliminateVarargsCast(CB, DL, CI, ix)) {
         *I = CI->getOperand(0);
         Changed = true;
       }
     }
   }
 
-  if (isa<InlineAsm>(Callee) && !CS.doesNotThrow()) {
+  if (isa<InlineAsm>(Callee) && !CB.doesNotThrow()) {
     // Inline asm calls cannot throw - mark them 'nounwind'.
-    CS.setDoesNotThrow();
+    CB.setDoesNotThrow();
     Changed = true;
   }
 
   // Try to optimize the call if possible, we require DataLayout for most of
   // this.  None of these calls are seen as possibly dead so go ahead and
   // delete the instruction now.
-  if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction())) {
+  if (CallInst *CI = dyn_cast<CallInst>(&CB)) {
     Instruction *I = tryOptimizeCall(CI);
     // If we changed something return the result, etc. Otherwise let
     // the fallthrough check.
     if (I) return eraseInstFromFunction(*I);
   }
 
-  return Changed ? CS.getInstruction() : nullptr;
+  return Changed ? &CB : nullptr;
 }
 
 /// If the callee is a constexpr cast of a function, attempt to move the cast to
 /// the arguments of the call/invoke.
-bool InstCombiner::transformConstExprCastCall(CallSite CS) {
-  auto *Callee = dyn_cast<Function>(CS.getCalledValue()->stripPointerCasts());
+bool InstCombiner::transformConstExprCastCall(CallBase &CB) {
+  auto *Callee = dyn_cast<Function>(CB.getCalledValue()->stripPointerCasts());
   if (!Callee)
     return false;
 
   // If this is a call to a thunk function, don't remove the cast. Thunks are
   // used to transparently forward all incoming parameters and outgoing return
   // values, so it's important to leave the cast in place.
   if (Callee->hasFnAttribute("thunk"))
     return false;
 
   // If this is a musttail call, the callee's prototype must match the caller's
   // prototype with the exception of pointee types. The code below doesn't
   // implement that, so we can't do this transform.
   // TODO: Do the transform if it only requires adding pointer casts.
-  if (CS.isMustTailCall())
+  if (isa<CallInst>(CB) && cast<CallInst>(CB).isMustTailCall())
     return false;
 
-  Instruction *Caller = CS.getInstruction();
-  const AttributeList &CallerPAL = CS.getAttributes();
+  Instruction *Caller = &CB;
+  const AttributeList &CallerPAL = CB.getAttributes();
 
   // Okay, this is a cast from a function to a different type.  Unless doing so
   // would cause a type conversion of one of our arguments, change this call to
   // be a direct call with arguments casted to the appropriate types.
   FunctionType *FT = Callee->getFunctionType();
   Type *OldRetTy = Caller->getType();
   Type *NewRetTy = FT->getReturnType();
 
if (OldRetTy != NewRetTy) {
     }
 
     if (!CallerPAL.isEmpty() && !Caller->use_empty()) {
       AttrBuilder RAttrs(CallerPAL, AttributeList::ReturnIndex);
       if (RAttrs.overlaps(AttributeFuncs::typeIncompatible(NewRetTy)))
         return false;   // Attribute not compatible with transformed value.
     }
 
-    // If the callsite is an invoke instruction, and the return value is used by
+    // If the callbase is an invoke instruction, and the return value is used by
     // a PHI node in a successor, we cannot change the return type of the call
     // because there is no place to put the cast instruction (without breaking
     // the critical edge).  Bail out in this case.
     if (!Caller->use_empty())
       if (InvokeInst *II = dyn_cast<InvokeInst>(Caller))
         for (User *U : II->users())
           if (PHINode *PN = dyn_cast<PHINode>(U))
             if (PN->getParent() == II->getNormalDest() ||
                 PN->getParent() == II->getUnwindDest())
               return false;
   }
 
-  unsigned NumActualArgs = CS.arg_size();
+  unsigned NumActualArgs = CB.arg_size();
   unsigned NumCommonArgs = std::min(FT->getNumParams(), NumActualArgs);
 
   // Prevent us turning:
   // declare void @takes_i32_inalloca(i32* inalloca)
   //  call void bitcast (void (i32*)* @takes_i32_inalloca to void (i32)*)(i32 0)
   //
   // into:
   //  call void @takes_i32_inalloca(i32* null)
   //
   //  Similarly, avoid folding away bitcasts of byval calls.
   if (Callee->getAttributes().hasAttrSomewhere(Attribute::InAlloca) ||
       Callee->getAttributes().hasAttrSomewhere(Attribute::ByVal))
     return false;
 
-  CallSite::arg_iterator AI = CS.arg_begin();
+  auto AI = CB.arg_begin();
   for (unsigned i = 0, e = NumCommonArgs; i != e; ++i, ++AI) {
     Type *ParamTy = FT->getParamType(i);
     Type *ActTy = (*AI)->getType();
 
     if (!CastInst::isBitOrNoopPointerCastable(ActTy, ParamTy, DL))
       return false;   // Cannot transform this parameter value.
 
     if (AttrBuilder(CallerPAL.getParamAttributes(i))
             .overlaps(AttributeFuncs::typeIncompatible(ParamTy)))
       return false;   // Attribute not compatible with transformed value.
 
-    if (CS.isInAllocaArgument(i))
+    if (CB.isInAllocaArgument(i))
       return false;   // Cannot transform to and from inalloca.
 
     // If the parameter is passed as a byval argument, then we have to have a
     // sized type and the sized type has to have the same size as the old type.
     if (ParamTy != ActTy && CallerPAL.hasParamAttribute(i, Attribute::ByVal)) {
       PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
       if (!ParamPTy || !ParamPTy->getElementType()->isSized())
         return false;
 
       Type *CurElTy = ActTy->getPointerElementType();
       if (DL.getTypeAllocSize(CurElTy) !=
           DL.getTypeAllocSize(ParamPTy->getElementType()))
         return false;
     }
   }
 
   if (Callee->isDeclaration()) {
     // Do not delete arguments unless we have a function body.
     if (FT->getNumParams() < NumActualArgs && !FT->isVarArg())
       return false;
 
     // If the callee is just a declaration, don't change the varargsness of the
     // call.  We don't want to introduce a varargs call where one doesn't
     // already exist.
-    PointerType *APTy = cast<PointerType>(CS.getCalledValue()->getType());
+    PointerType *APTy = cast<PointerType>(CB.getCalledValue()->getType());
     if (FT->isVarArg()!=cast<FunctionType>(APTy->getElementType())->isVarArg())
       return false;
 
     // If both the callee and the cast type are varargs, we still have to make
     // sure the number of fixed parameters are the same or we have the same
     // ABI issues as if we introduce a varargs call.
     if (FT->isVarArg() &&
         cast<FunctionType>(APTy->getElementType())->isVarArg() &&
bool InstCombiner::transformConstExprCastCall(CallBase &CB) {
 
   // Get any return attributes.
   AttrBuilder RAttrs(CallerPAL, AttributeList::ReturnIndex);
 
   // If the return value is not being used, the type may not be compatible
   // with the existing attributes.  Wipe out any problematic attributes.
   RAttrs.remove(AttributeFuncs::typeIncompatible(NewRetTy));
 
-  AI = CS.arg_begin();
+  AI = CB.arg_begin();
   for (unsigned i = 0; i != NumCommonArgs; ++i, ++AI) {
     Type *ParamTy = FT->getParamType(i);
 
     Value *NewArg = *AI;
     if ((*AI)->getType() != ParamTy)
       NewArg = Builder.CreateBitOrPointerCast(*AI, ParamTy);
     Args.push_back(NewArg);
 
bool InstCombiner::transformConstExprCastCall(CallBase &CB) {
 
   assert((ArgAttrs.size() == FT->getNumParams() || FT->isVarArg()) &&
          "missing argument attributes");
   LLVMContext &Ctx = Callee->getContext();
   AttributeList NewCallerPAL = AttributeList::get(
       Ctx, FnAttrs, AttributeSet::get(Ctx, RAttrs), ArgAttrs);
 
   SmallVector<OperandBundleDef, 1> OpBundles;
-  CS.getOperandBundlesAsDefs(OpBundles);
+  CB.getOperandBundlesAsDefs(OpBundles);
 
-  CallSite NewCS;
+  CallBase *NewCB;
   if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
-    NewCS = Builder.CreateInvoke(Callee, II->getNormalDest(),
+    NewCB = Builder.CreateInvoke(Callee, II->getNormalDest(),
                                  II->getUnwindDest(), Args, OpBundles);
   } else {
-    NewCS = Builder.CreateCall(Callee, Args, OpBundles);
-    cast<CallInst>(NewCS.getInstruction())
+    NewCB = Builder.CreateCall(Callee, Args, OpBundles);
+    cast<CallInst>(NewCB)
         ->setTailCallKind(cast<CallInst>(Caller)->getTailCallKind());
   }
-  NewCS->takeName(Caller);
-  NewCS.setCallingConv(CS.getCallingConv());
-  NewCS.setAttributes(NewCallerPAL);
+  NewCB->takeName(Caller);
+  NewCB->setCallingConv(CB.getCallingConv());
+  NewCB->setAttributes(NewCallerPAL);
 
   // Preserve the weight metadata for the new call instruction. The metadata
   // is used by SamplePGO to check callsite's hotness.
   uint64_t W;
   if (Caller->extractProfTotalWeight(W))
-    NewCS->setProfWeight(W);
+    NewCB->setProfWeight(W);
 
   // Insert a cast of the return type as necessary.
-  Instruction *NC = NewCS.getInstruction();
+  Instruction *NC = NewCB;
   Value *NV = NC;
   if (OldRetTy != NV->getType() && !Caller->use_empty()) {
     if (!NV->getType()->isVoidTy()) {
       NV = NC = CastInst::CreateBitOrPointerCast(NC, OldRetTy);
       NC->setDebugLoc(Caller->getDebugLoc());
 
       // If this is an invoke instruction, we should insert it after the first
       // non-phi, instruction in the normal successor block.
bool InstCombiner::transformConstExprCastCall(CallBase &CB) {
 
   eraseInstFromFunction(*Caller);
   return true;
 }
 
 /// Turn a call to a function created by init_trampoline / adjust_trampoline
 /// intrinsic pair into a direct call to the underlying function.
 Instruction *
-InstCombiner::transformCallThroughTrampoline(CallSite CS,
-                                             IntrinsicInst *Tramp) {
-  Value *Callee = CS.getCalledValue();
+InstCombiner::transformCallThroughTrampoline(CallBase &CB,
+                                             IntrinsicInst &Tramp) {
+  Value *Callee = CB.getCalledValue();
   PointerType *PTy = cast<PointerType>(Callee->getType());
   FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
-  AttributeList Attrs = CS.getAttributes();
+  AttributeList Attrs = CB.getAttributes();
 
   // If the call already has the 'nest' attribute somewhere then give up -
   // otherwise 'nest' would occur twice after splicing in the chain.
   if (Attrs.hasAttrSomewhere(Attribute::Nest))
     return nullptr;
 
-  assert(Tramp &&
-         "transformCallThroughTrampoline called with incorrect CallSite.");
-
-  Function *NestF =cast<Function>(Tramp->getArgOperand(1)->stripPointerCasts());
+  Function *NestF =cast<Function>(Tramp.getArgOperand(1)->stripPointerCasts());
   FunctionType *NestFTy = cast<FunctionType>(NestF->getValueType());
 
   AttributeList NestAttrs = NestF->getAttributes();
   if (!NestAttrs.isEmpty()) {
     unsigned NestArgNo = 0;
     Type *NestTy = nullptr;
     AttributeSet NestAttr;
 
     // Look for a parameter marked with the 'nest' attribute.
     for (FunctionType::param_iterator I = NestFTy->param_begin(),
                                       E = NestFTy->param_end();
          I != E; ++NestArgNo, ++I) {
       AttributeSet AS = NestAttrs.getParamAttributes(NestArgNo);
       if (AS.hasAttribute(Attribute::Nest)) {
         // Record the parameter type and any other attributes.
         NestTy = *I;
         NestAttr = AS;
         break;
       }
     }
 
     if (NestTy) {
-      Instruction *Caller = CS.getInstruction();
+      Instruction *Caller = &CB;
       std::vector<Value*> NewArgs;
       std::vector<AttributeSet> NewArgAttrs;
-      NewArgs.reserve(CS.arg_size() + 1);
-      NewArgAttrs.reserve(CS.arg_size());
+      NewArgs.reserve(CB.arg_size() + 1);
+      NewArgAttrs.reserve(CB.arg_size());
 
       // Insert the nest argument into the call argument list, which may
       // mean appending it.  Likewise for attributes.
 
       {
         unsigned ArgNo = 0;
-        CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
+        auto I = CB.arg_begin(), E = CB.arg_end();
         do {
           if (ArgNo == NestArgNo) {
             // Add the chain argument and attributes.
-            Value *NestVal = Tramp->getArgOperand(2);
+            Value *NestVal = Tramp.getArgOperand(2);
             if (NestVal->getType() != NestTy)
               NestVal = Builder.CreateBitCast(NestVal, NestTy, "nest");
             NewArgs.push_back(NestVal);
             NewArgAttrs.push_back(NestAttr);
           }
 
           if (I == E)
             break;
if (NestTy) {
         NestF->getType() == PointerType::getUnqual(NewFTy) ?
         NestF : ConstantExpr::getBitCast(NestF,
                                          PointerType::getUnqual(NewFTy));
       AttributeList NewPAL =
           AttributeList::get(FTy->getContext(), Attrs.getFnAttributes(),
                              Attrs.getRetAttributes(), NewArgAttrs);
 
       SmallVector<OperandBundleDef, 1> OpBundles;
-      CS.getOperandBundlesAsDefs(OpBundles);
+      CB.getOperandBundlesAsDefs(OpBundles);
 
       Instruction *NewCaller;
       if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
         NewCaller = InvokeInst::Create(NewCallee,
                                        II->getNormalDest(), II->getUnwindDest(),
                                        NewArgs, OpBundles);
         cast<InvokeInst>(NewCaller)->setCallingConv(II->getCallingConv());
         cast<InvokeInst>(NewCaller)->setAttributes(NewPAL);
InstCombiner::transformCallThroughTrampoline(CallBase &CB,
   }
 
   // Replace the trampoline call with a direct call.  Since there is no 'nest'
   // parameter, there is no need to adjust the argument list.  Let the generic
   // code sort out any function type mismatches.
   Constant *NewCallee =
     NestF->getType() == PTy ? NestF :
                               ConstantExpr::getBitCast(NestF, PTy);
-  CS.setCalledFunction(NewCallee);
-  return CS.getInstruction();
+  CB.setCalledFunction(NewCallee);
+  return &CB;
 }

lib/Transforms/InstCombine/InstCombineInternal.h

 #define DEBUG_TYPE "instcombine"
 
 using namespace llvm::PatternMatch;
 
 namespace llvm {
 
 class APInt;
 class AssumptionCache;
-class CallSite;
 class DataLayout;
 class DominatorTree;
 class GEPOperator;
 class GlobalVariable;
 class LoopInfo;
 class OptimizationRemarkEmitter;
 class TargetLibraryInfo;
 class User;
private:
   /// If a simplification is possible, stores the simplified result of the
   /// operation in OperationResult and result of the overflow check in
   /// OverflowResult, and return true.  If no simplification is possible,
   /// returns false.
   bool OptimizeOverflowCheck(OverflowCheckFlavor OCF, Value *LHS, Value *RHS,
                              Instruction &CtxI, Value *&OperationResult,
                              Constant *&OverflowResult);
 
-  Instruction *visitCallSite(CallSite CS);
+  Instruction *visitCallBase(CallBase &CB);
   Instruction *tryOptimizeCall(CallInst *CI);
-  bool transformConstExprCastCall(CallSite CS);
-  Instruction *transformCallThroughTrampoline(CallSite CS,
-                                              IntrinsicInst *Tramp);
+  bool transformConstExprCastCall(CallBase &CB);
+  Instruction *transformCallThroughTrampoline(CallBase &CB,
+                                              IntrinsicInst &Tramp);
 
   /// Transform (zext icmp) to bitwise / integer operations in order to
   /// eliminate it.
   ///
   /// \param ICI The icmp of the (zext icmp) pair we are interested in.
   /// \parem CI The zext of the (zext icmp) pair we are interested in.
   /// \param DoTransform Pass false to just test whether the given (zext icmp)
   /// would be transformed. Pass true to actually perform the transformation.

lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

for (auto &U : ValuePair.first->uses()) {
       }
       if (auto *GEP = dyn_cast<GetElementPtrInst>(I)) {
         // If the GEP has all zero indices, it doesn't offset the pointer. If it
         // doesn't, it does.
         ValuesToInspect.emplace_back(I, IsOffset || !GEP->hasAllZeroIndices());
         continue;
       }
 
-      if (auto CS = CallSite(I)) {
+      if (auto *CB = dyn_cast<CallBase>(I)) {
         // If this is the function being called then we treat it like a load and
         // ignore it.
-        if (CS.isCallee(&U))
+        if (CB->isCallee(&U))
           continue;
 
-        unsigned DataOpNo = CS.getDataOperandNo(&U);
-        bool IsArgOperand = CS.isArgOperand(&U);
+        unsigned DataOpNo = CB->getDataOperandNo(&U);
+        bool IsArgOperand = CB->isArgOperand(&U);
 
         // Inalloca arguments are clobbered by the call.
-        if (IsArgOperand && CS.isInAllocaArgument(DataOpNo))
+        if (IsArgOperand && CB->isInAllocaArgument(DataOpNo))
           return false;
 
         // If this is a readonly/readnone call site, then we know it is just a
         // load (but one that potentially returns the value itself), so we can
         // ignore it if we know that the value isn't captured.
-        if (CS.onlyReadsMemory() &&
-            (CS.getInstruction()->use_empty() || CS.doesNotCapture(DataOpNo)))
+        if (CB->onlyReadsMemory() &&
+            (CB->use_empty() || CB->doesNotCapture(DataOpNo)))
           continue;
 
         // If this is being passed as a byval argument, the caller is making a
         // copy, so it is only a read of the alloca.
-        if (IsArgOperand && CS.isByValArgument(DataOpNo))
+        if (IsArgOperand && CB->isByValArgument(DataOpNo))
           continue;
       }
 
       // Lifetime intrinsics can be handled by the caller.
       if (I->isLifetimeStartOrEnd()) {
         assert(I->use_empty() && "Lifetime markers have no result to use!");
         ToDelete.push_back(I);
         continue;

lib/Transforms/InstCombine/InstCombineMulDivRem.cpp

bool IsCot =
         !IsTan && match(Op0, m_Intrinsic<Intrinsic::cos>(m_Value(X))) &&
                   match(Op1, m_Intrinsic<Intrinsic::sin>(m_Specific(X)));
 
     if ((IsTan || IsCot) && hasUnaryFloatFn(&TLI, I.getType(), LibFunc_tan,
                                             LibFunc_tanf, LibFunc_tanl)) {
       IRBuilder<> B(&I);
       IRBuilder<>::FastMathFlagGuard FMFGuard(B);
       B.setFastMathFlags(I.getFastMathFlags());
-      AttributeList Attrs = CallSite(Op0).getCalledFunction()->getAttributes();
+      AttributeList Attrs =
+          cast<CallBase>(Op0)->getCalledFunction()->getAttributes();
       Value *Res = emitUnaryFloatFnCall(X, &TLI, LibFunc_tan, LibFunc_tanf,
                                         LibFunc_tanl, B, Attrs);
       if (IsCot)
         Res = B.CreateFDiv(ConstantFP::get(I.getType(), 1.0), Res);
       return replaceInstUsesWith(I, Res);
     }
   }