[IRBuilder] Fixup CreateIntrinsic to allow specifying Types to Mangle.

[sheredom]

Summary

The IRBuilder CreateIntrinsic method wouldn't allow you to specify the types that you wanted the intrinsic to be mangled with. To fix this I've:

• Added an ArrayRef<Type *> member to both CreateIntrinsic overloads.
• Used that array to pass into the Intrinsic::getDeclaration call.
• Added a CreateUnaryIntrinsic to replace the most common use of CreateIntrinsic where the type was auto-deduced from operand 0.
• Added a bunch more unit tests to test Create*Intrinsic calls that weren't being tested (including the FMF flag that wasn't checked).

This was suggested as part of the AMDGPU specific atomic optimizer review (https://reviews.llvm.org/D51969).

[nhaehnle]

include/llvm/IR/IRBuilder.h

689–690

I think it'd be best to remove this overload. It made sense with the "magic but broken" type deduction that was there before, but now anybody wishing to create a no-argument intrinsic can just pass {} to the overload below.

[sheredom]

Removed the no-args overload as per Nicolai's suggestion.

[nhaehnle]

LGTM

[arsenm]

LGTM

include/llvm/IR/IRBuilder.h

public:
   /// start a new statepoint sequence.
   InvokeInst *CreateGCStatepointInvoke(
       uint64_t ID, uint32_t NumPatchBytes, Value *ActualInvokee,
       BasicBlock *NormalDest, BasicBlock *UnwindDest, uint32_t Flags,
       ArrayRef<Use> InvokeArgs, ArrayRef<Use> TransitionArgs,
       ArrayRef<Use> DeoptArgs, ArrayRef<Value *> GCArgs,
       const Twine &Name = "");
 
-  // Conveninence function for the common case when CallArgs are filled in using
+  // Convenience function for the common case when CallArgs are filled in using
   // makeArrayRef(CS.arg_begin(), CS.arg_end()); Use needs to be .get()'ed to
   // get the Value *.
   InvokeInst *
   CreateGCStatepointInvoke(uint64_t ID, uint32_t NumPatchBytes,
                            Value *ActualInvokee, BasicBlock *NormalDest,
                            BasicBlock *UnwindDest, ArrayRef<Use> InvokeArgs,
                            ArrayRef<Value *> DeoptArgs,
                            ArrayRef<Value *> GCArgs, const Twine &Name = "");
 
   /// Create a call to the experimental.gc.result intrinsic to extract
   /// the result from a call wrapped in a statepoint.
   CallInst *CreateGCResult(Instruction *Statepoint,
                            Type *ResultType,
                            const Twine &Name = "");
 
   /// Create a call to the experimental.gc.relocate intrinsics to
   /// project the relocated value of one pointer from the statepoint.
   CallInst *CreateGCRelocate(Instruction *Statepoint,
                              int BaseOffset,
                              int DerivedOffset,
                              Type *ResultType,
                              const Twine &Name = "");
 
-  /// Create a call to intrinsic \p ID with 2 operands which is mangled on the
-  /// first type.
-  CallInst *CreateBinaryIntrinsic(Intrinsic::ID ID,
-                                  Value *LHS, Value *RHS,
-                                  const Twine &Name = "");
+  /// Create a call to intrinsic \p ID with 1 operand which is mangled on its
+  /// type.
+  CallInst *CreateUnaryIntrinsic(Intrinsic::ID ID, Value *V,
+                                 Instruction *FMFSource = nullptr,
 
-  /// Create a call to intrinsic \p ID with no operands.
-  CallInst *CreateIntrinsic(Intrinsic::ID ID,
+  /// Create a call to intrinsic \p ID with 2 operands which is mangled on the
+  /// first type.
+  CallInst *CreateBinaryIntrinsic(Intrinsic::ID ID, Value *LHS, Value *RHS,
                                   Instruction *FMFSource = nullptr,
                                   const Twine &Name = "");
 
-  /// Create a call to intrinsic \p ID with 1 or more operands assuming the
-  /// intrinsic and all operands have the same type. If \p FMFSource is
-  /// provided, copy fast-math-flags from that instruction to the intrinsic.
-  CallInst *CreateIntrinsic(Intrinsic::ID ID, ArrayRef<Value *> Args,
+  /// Create a call to intrinsic \p ID with \p args, mangled using \p Types. If
+  /// \p FMFSource is provided, copy fast-math-flags from that instruction to
+  /// the intrinsic.
+  CallInst *CreateIntrinsic(Intrinsic::ID ID, ArrayRef<Type *> Types,
+                            ArrayRef<Value *> Args,
                             Instruction *FMFSource = nullptr,
                             const Twine &Name = "");
 
   /// Create call to the minnum intrinsic.
   CallInst *CreateMinNum(Value *LHS, Value *RHS, const Twine &Name = "") {
-    return CreateBinaryIntrinsic(Intrinsic::minnum, LHS, RHS, Name);
+    return CreateBinaryIntrinsic(Intrinsic::minnum, LHS, RHS, nullptr, Name);
   }
 
   /// Create call to the maxnum intrinsic.
   CallInst *CreateMaxNum(Value *LHS, Value *RHS, const Twine &Name = "") {
-    return CreateBinaryIntrinsic(Intrinsic::maxnum, LHS, RHS, Name);
+    return CreateBinaryIntrinsic(Intrinsic::maxnum, LHS, RHS, nullptr, Name);
   }
 
 private:
   /// Create a call to a masked intrinsic with given Id.
   CallInst *CreateMaskedIntrinsic(Intrinsic::ID Id, ArrayRef<Value *> Ops,
                                   ArrayRef<Type *> OverloadedTypes,
                                   const Twine &Name = "");
 

lib/IR/IRBuilder.cpp

Value *FnGCRelocate =
    Intrinsic::getDeclaration(M, Intrinsic::experimental_gc_relocate, Types);
 
  Value *Args[] = {Statepoint,
                   getInt32(BaseOffset),
                   getInt32(DerivedOffset)};
  return createCallHelper(FnGCRelocate, Args, this, Name);
 }
 
-CallInst *IRBuilderBase::CreateBinaryIntrinsic(Intrinsic::ID ID,
-                                               Value *LHS, Value *RHS,
-                                               const Twine &Name) {
+CallInst *IRBuilderBase::CreateUnaryIntrinsic(Intrinsic::ID ID, Value *V,
+                                              Instruction *FMFSource,
   Module *M = BB->getModule();
-  Function *Fn = Intrinsic::getDeclaration(M, ID, { LHS->getType() });
-  return createCallHelper(Fn, { LHS, RHS }, this, Name);
+  Function *Fn = Intrinsic::getDeclaration(M, ID, {V->getType()});
+  return createCallHelper(Fn, {V}, this, Name, FMFSource);
 }
 
-CallInst *IRBuilderBase::CreateIntrinsic(Intrinsic::ID ID,
+CallInst *IRBuilderBase::CreateBinaryIntrinsic(Intrinsic::ID ID, Value *LHS,
+                                               Value *RHS,
                                                Instruction *FMFSource,
                                                const Twine &Name) {
   Module *M = BB->getModule();
-  Function *Fn = Intrinsic::getDeclaration(M, ID);
-  return createCallHelper(Fn, {}, this, Name);
+  Function *Fn = Intrinsic::getDeclaration(M, ID, { LHS->getType() });
+  return createCallHelper(Fn, {LHS, RHS}, this, Name, FMFSource);
 }
 
 CallInst *IRBuilderBase::CreateIntrinsic(Intrinsic::ID ID,
+                                         ArrayRef<Type *> Types,
                                          ArrayRef<Value *> Args,
                                          Instruction *FMFSource,
                                          const Twine &Name) {
   assert(!Args.empty() && "Expected at least one argument to intrinsic");
   Module *M = BB->getModule();
-  Function *Fn = Intrinsic::getDeclaration(M, ID, { Args.front()->getType() });
+  Function *Fn = Intrinsic::getDeclaration(M, ID, Types);
   return createCallHelper(Fn, Args, this, Name, FMFSource);
 }

lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp

Value* AMDGPUCodeGenPrepare::expandDivRem24(IRBuilder<> &Builder,
   // float fb = (float)ib;
   Value *FB = IsSigned ? Builder.CreateSIToFP(IB,F32Ty)
                        : Builder.CreateUIToFP(IB,F32Ty);
 
   Value *RCP = Builder.CreateFDiv(ConstantFP::get(F32Ty, 1.0), FB);
   Value *FQM = Builder.CreateFMul(FA, RCP);
 
   // fq = trunc(fqm);
-  CallInst* FQ = Builder.CreateIntrinsic(Intrinsic::trunc, { FQM });
+  CallInst *FQ = Builder.CreateUnaryIntrinsic(Intrinsic::trunc, FQM);
   FQ->copyFastMathFlags(Builder.getFastMathFlags());
 
   // float fqneg = -fq;
   Value *FQNeg = Builder.CreateFNeg(FQ);
 
   // float fr = mad(fqneg, fb, fa);
   Value *FR = Builder.CreateIntrinsic(Intrinsic::amdgcn_fmad_ftz,
-                                      { FQNeg, FB, FA }, FQ);
+                                      {FQNeg->getType()}, {FQNeg, FB, FA}, FQ);
 
   // int iq = (int)fq;
   Value *IQ = IsSigned ? Builder.CreateFPToSI(FQ, I32Ty)
                        : Builder.CreateFPToUI(FQ, I32Ty);
 
   // fr = fabs(fr);
-  FR = Builder.CreateIntrinsic(Intrinsic::fabs, { FR }, FQ);
+  FR = Builder.CreateUnaryIntrinsic(Intrinsic::fabs, FR, FQ);
 
   // fb = fabs(fb);
-  FB = Builder.CreateIntrinsic(Intrinsic::fabs, { FB }, FQ);
+  FB = Builder.CreateUnaryIntrinsic(Intrinsic::fabs, FB, FQ);
 
   // int cv = fr >= fb;
   Value *CV = Builder.CreateFCmpOGE(FR, FB);
 
   // jq = (cv ? jq : 0);
   JQ = Builder.CreateSelect(CV, JQ, Builder.getInt32(0));
 
   // dst = iq + jq;

lib/Target/AMDGPU/AMDGPULowerKernelArguments.cpp

bool AMDGPULowerKernelArguments::runOnFunction(Function &F) {
 
   unsigned MaxAlign;
   // FIXME: Alignment is broken broken with explicit arg offset.;
   const uint64_t TotalKernArgSize = ST.getKernArgSegmentSize(F, MaxAlign);
   if (TotalKernArgSize == 0)
     return false;
 
   CallInst *KernArgSegment =
-    Builder.CreateIntrinsic(Intrinsic::amdgcn_kernarg_segment_ptr, nullptr,
-                            F.getName() + ".kernarg.segment");
+      Builder.CreateIntrinsic(Intrinsic::amdgcn_kernarg_segment_ptr, {}, {},
+                              nullptr, F.getName() + ".kernarg.segment");
 
   KernArgSegment->addAttribute(AttributeList::ReturnIndex, Attribute::NonNull);
   KernArgSegment->addAttribute(AttributeList::ReturnIndex,
     Attribute::getWithDereferenceableBytes(Ctx, TotalKernArgSize));
 
   unsigned AS = KernArgSegment->getType()->getPointerAddressSpace();
   uint64_t ExplicitArgOffset = 0;
 

lib/Transforms/InstCombine/InstCombineCalls.cpp

if (IntrinsicID == Intrinsic::x86_sse41_round_ss ||
     } else {
       Dst = Src;
       Mask = ConstantInt::getAllOnesValue(
           Builder.getIntNTy(Src->getType()->getVectorNumElements()));
     }
   }
 
   Intrinsic::ID ID = (RoundControl == 2) ? Intrinsic::ceil : Intrinsic::floor;
-  Value *Res = Builder.CreateIntrinsic(ID, {Src}, &II);
+  Value *Res = Builder.CreateUnaryIntrinsic(ID, Src, &II);
   if (!IsScalar) {
     if (auto *C = dyn_cast<Constant>(Mask))
       if (C->isAllOnesValue())
         return Res;
     auto *MaskTy = VectorType::get(
         Builder.getInt1Ty(), cast<IntegerType>(Mask->getType())->getBitWidth());
     Mask = Builder.CreateBitCast(Mask, MaskTy);
     unsigned Width = Src->getType()->getVectorNumElements();
case Intrinsic::maxnum: {
     Value *X, *Y;
     if (match(Arg0, m_FNeg(m_Value(X))) && match(Arg1, m_FNeg(m_Value(Y))) &&
         (Arg0->hasOneUse() || Arg1->hasOneUse())) {
       // If both operands are negated, invert the call and negate the result:
       // minnum(-X, -Y) --> -(maxnum(X, Y))
       // maxnum(-X, -Y) --> -(minnum(X, Y))
       Intrinsic::ID NewIID = II->getIntrinsicID() == Intrinsic::maxnum ?
           Intrinsic::minnum : Intrinsic::maxnum;
-      Value *NewCall = Builder.CreateIntrinsic(NewIID, { X, Y }, II);
+      Value *NewCall = Builder.CreateBinaryIntrinsic(NewIID, X, Y, II);
       Instruction *FNeg = BinaryOperator::CreateFNeg(NewCall);
       FNeg->copyIRFlags(II);
       return FNeg;
     }
     break;
   }
   case Intrinsic::fmuladd: {
     // Canonicalize fast fmuladd to the separate fmul + fadd.
Instruction *InstCombiner::visitCallInst(CallInst &CI) {
   case Intrinsic::floor:
   case Intrinsic::round:
   case Intrinsic::nearbyint:
   case Intrinsic::rint:
   case Intrinsic::trunc: {
     Value *ExtSrc;
     if (match(II->getArgOperand(0), m_OneUse(m_FPExt(m_Value(ExtSrc))))) {
       // Narrow the call: intrinsic (fpext x) -> fpext (intrinsic x)
-      Value *NarrowII = Builder.CreateIntrinsic(II->getIntrinsicID(),
-                                                { ExtSrc }, II);
+      Value *NarrowII =
+          Builder.CreateUnaryIntrinsic(II->getIntrinsicID(), ExtSrc, II);
       return new FPExtInst(NarrowII, II->getType());
     }
     break;
   }
   case Intrinsic::cos:
   case Intrinsic::amdgcn_cos: {
     Value *X;
     Value *Src = II->getArgOperand(0);
     if (match(Src, m_FNeg(m_Value(X))) || match(Src, m_FAbs(m_Value(X)))) {
       // cos(-x) -> cos(x)
       // cos(fabs(x)) -> cos(x)
       II->setArgOperand(0, X);
       return II;
     }
     break;
   }
   case Intrinsic::sin: {
     Value *X;
     if (match(II->getArgOperand(0), m_OneUse(m_FNeg(m_Value(X))))) {
       // sin(-x) --> -sin(x)
-      Value *NewSin = Builder.CreateIntrinsic(Intrinsic::sin, { X }, II);
+      Value *NewSin = Builder.CreateUnaryIntrinsic(Intrinsic::sin, X, II);
       Instruction *FNeg = BinaryOperator::CreateFNeg(NewSin);
       FNeg->copyFastMathFlags(II);
       return FNeg;
     }
     break;
   }
   case Intrinsic::ppc_altivec_lvx:
   case Intrinsic::ppc_altivec_lvxl:

lib/Transforms/InstCombine/InstCombineMulDivRem.cpp

if (I.hasAllowReassoc()) {
 
     // sqrt(X) * sqrt(Y) -> sqrt(X * Y)
     // nnan disallows the possibility of returning a number if both operands are
     // negative (in that case, we should return NaN).
     if (I.hasNoNaNs() &&
         match(Op0, m_OneUse(m_Intrinsic<Intrinsic::sqrt>(m_Value(X)))) &&
         match(Op1, m_OneUse(m_Intrinsic<Intrinsic::sqrt>(m_Value(Y))))) {
       Value *XY = Builder.CreateFMulFMF(X, Y, &I);
-      Value *Sqrt = Builder.CreateIntrinsic(Intrinsic::sqrt, { XY }, &I);
+      Value *Sqrt = Builder.CreateUnaryIntrinsic(Intrinsic::sqrt, XY, &I);
       return replaceInstUsesWith(I, Sqrt);
     }
 
     // (X*Y) * X => (X*X) * Y where Y != X
     //  The purpose is two-fold:
     //   1) to form a power expression (of X).
     //   2) potentially shorten the critical path: After transformation, the
     //  latency of the instruction Y is amortized by the expression of X*X,

lib/Transforms/InstCombine/InstCombineSelect.cpp

if (FCmpInst *FCI = dyn_cast<FCmpInst>(CondVal)) {
     FCmpInst::Predicate Pred = FCI->getPredicate();
     if (match(FCI->getOperand(1), m_AnyZeroFP()) && FCI->hasNoNaNs()) {
       // (X <= +/-0.0) ? (0.0 - X) : X --> fabs(X)
       // (X >  +/-0.0) ? X : (0.0 - X) --> fabs(X)
       if ((X == FalseVal && Pred == FCmpInst::FCMP_OLE &&
            match(TrueVal, m_FSub(m_PosZeroFP(), m_Specific(X)))) ||
           (X == TrueVal && Pred == FCmpInst::FCMP_OGT &&
            match(FalseVal, m_FSub(m_PosZeroFP(), m_Specific(X))))) {
-        Value *Fabs = Builder.CreateIntrinsic(Intrinsic::fabs, { X }, FCI);
+        Value *Fabs = Builder.CreateUnaryIntrinsic(Intrinsic::fabs, X, FCI);
         return replaceInstUsesWith(SI, Fabs);
       }
       // With nsz:
       // (X <  +/-0.0) ? -X : X --> fabs(X)
       // (X <= +/-0.0) ? -X : X --> fabs(X)
       // (X >  +/-0.0) ? X : -X --> fabs(X)
       // (X >= +/-0.0) ? X : -X --> fabs(X)
       if (FCI->hasNoSignedZeros() &&
           ((X == FalseVal && match(TrueVal, m_FNeg(m_Specific(X))) &&
             (Pred == FCmpInst::FCMP_OLT || Pred == FCmpInst::FCMP_OLE)) ||
            (X == TrueVal && match(FalseVal, m_FNeg(m_Specific(X))) &&
             (Pred == FCmpInst::FCMP_OGT || Pred == FCmpInst::FCMP_OGE)))) {
-        Value *Fabs = Builder.CreateIntrinsic(Intrinsic::fabs, { X }, FCI);
+        Value *Fabs = Builder.CreateUnaryIntrinsic(Intrinsic::fabs, X, FCI);
         return replaceInstUsesWith(SI, Fabs);
       }
     }
   }
 
   // See if we are selecting two values based on a comparison of the two values.
   if (ICmpInst *ICI = dyn_cast<ICmpInst>(CondVal))
     if (Instruction *Result = foldSelectInstWithICmp(SI, ICI))

unittests/IR/IRBuilderTest.cpp

protected:
   Function *F;
   BasicBlock *BB;
   GlobalVariable *GV;
 };
 
 TEST_F(IRBuilderTest, Intrinsics) {
   IRBuilder<> Builder(BB);
   Value *V;
+  Instruction *I;
   CallInst *Call;
   IntrinsicInst *II;
 
   V = Builder.CreateLoad(GV);
+  I = cast<Instruction>(Builder.CreateFAdd(V, V));
+  I->setHasNoInfs(true);
+  I->setHasNoNaNs(false);
 
   Call = Builder.CreateMinNum(V, V);
   II = cast<IntrinsicInst>(Call);
   EXPECT_EQ(II->getIntrinsicID(), Intrinsic::minnum);
 
   Call = Builder.CreateMaxNum(V, V);
   II = cast<IntrinsicInst>(Call);
   EXPECT_EQ(II->getIntrinsicID(), Intrinsic::maxnum);
 
-  Call = Builder.CreateIntrinsic(Intrinsic::readcyclecounter);
+  Call = Builder.CreateIntrinsic(Intrinsic::readcyclecounter, {}, {});
   II = cast<IntrinsicInst>(Call);
   EXPECT_EQ(II->getIntrinsicID(), Intrinsic::readcyclecounter);
+
+  Call = Builder.CreateUnaryIntrinsic(Intrinsic::fabs, V);
+  II = cast<IntrinsicInst>(Call);
+  EXPECT_EQ(II->getIntrinsicID(), Intrinsic::fabs);
+  EXPECT_FALSE(II->hasNoInfs());
+  EXPECT_FALSE(II->hasNoNaNs());
+
+  Call = Builder.CreateUnaryIntrinsic(Intrinsic::fabs, V, I);
+  II = cast<IntrinsicInst>(Call);
+  EXPECT_EQ(II->getIntrinsicID(), Intrinsic::fabs);
+  EXPECT_TRUE(II->hasNoInfs());
+  EXPECT_FALSE(II->hasNoNaNs());
+
+  Call = Builder.CreateBinaryIntrinsic(Intrinsic::pow, V, V);
+  II = cast<IntrinsicInst>(Call);
+  EXPECT_EQ(II->getIntrinsicID(), Intrinsic::pow);
+  EXPECT_FALSE(II->hasNoInfs());
+  EXPECT_FALSE(II->hasNoNaNs());
+
+  Call = Builder.CreateBinaryIntrinsic(Intrinsic::pow, V, V, I);
+  II = cast<IntrinsicInst>(Call);
+  EXPECT_EQ(II->getIntrinsicID(), Intrinsic::pow);
+  EXPECT_TRUE(II->hasNoInfs());
+  EXPECT_FALSE(II->hasNoNaNs());
+
+  Call = Builder.CreateIntrinsic(Intrinsic::fma, {V->getType()}, {V, V, V});
+  II = cast<IntrinsicInst>(Call);
+  EXPECT_EQ(II->getIntrinsicID(), Intrinsic::fma);
+  EXPECT_FALSE(II->hasNoInfs());
+  EXPECT_FALSE(II->hasNoNaNs());
+
+  Call = Builder.CreateIntrinsic(Intrinsic::fma, {V->getType()}, {V, V, V}, I);
+  II = cast<IntrinsicInst>(Call);
+  EXPECT_EQ(II->getIntrinsicID(), Intrinsic::fma);
+  EXPECT_TRUE(II->hasNoInfs());
+  EXPECT_FALSE(II->hasNoNaNs());
+
+  Call = Builder.CreateIntrinsic(Intrinsic::fma, {V->getType()}, {V, V, V}, I);
+  II = cast<IntrinsicInst>(Call);
+  EXPECT_EQ(II->getIntrinsicID(), Intrinsic::fma);
+  EXPECT_TRUE(II->hasNoInfs());
+  EXPECT_FALSE(II->hasNoNaNs());
 }
 
 TEST_F(IRBuilderTest, Lifetime) {
   IRBuilder<> Builder(BB);
   AllocaInst *Var1 = Builder.CreateAlloca(Builder.getInt8Ty());
   AllocaInst *Var2 = Builder.CreateAlloca(Builder.getInt32Ty());
   AllocaInst *Var3 = Builder.CreateAlloca(Builder.getInt8Ty(),
                                           Builder.getInt32(123));