[Types] Define a getWithNewBitWidth for Types and make use of it.

[dmgreen]

Summary

This is designed to change the bitwidth of a type without altering the number of vector lanes. Also useful in D68651. Otherwise an NFC.

[lebedev.ri]

Nice, i like this!
This doesn't change the bitwidth of the type it is called on though, it returns new type,
so i think the name needs a bit more thought (i can't give a better suggestion right now)

llvm/include/llvm/IR/Type.h

386

s/the bitwidth/the lane bitwidth/

[xbolva00]

llvm/include/llvm/IR/Type.h

388

NewBitWidth

[dmgreen]

Hows about a static getWithNewBitWidth(Type*, NewBitWidth)? Similar to Types existing static get methods.

So used like: Type *CondTy = Type::getWithNewBitWidth(RetTy, 1)

[lebedev.ri]

In D69139#1713852, @dmgreen wrote:

Hows about a static getWithNewBitWidth(Type*, NewBitWidth)? Similar to Types existing static get methods.

So used like: Type *CondTy = Type::getWithNewBitWidth(RetTy, 1)

AH yes, i think that would be good.
I don't really see why it should be static, getExtendedType() isn't.

[dmgreen]

Renamed to getWithNewBitWidth

[dmgreen]

Update comment

[lebedev.ri]

LG, thank you for the cleanup!

llvm/include/llvm/CodeGen/BasicTTIImpl.h

case Intrinsic::fshr: {
       // Non-constant shift amounts requires a modulo.
       if (OpKindZ != TTI::OK_UniformConstantValue &&
           OpKindZ != TTI::OK_NonUniformConstantValue)
         Cost += ConcreteTTI->getArithmeticInstrCost(BinaryOperator::URem, RetTy,
                                                     OpKindZ, OpKindBW, OpPropsZ,
                                                     OpPropsBW);
       // For non-rotates (X != Y) we must add shift-by-zero handling costs.
       if (X != Y) {
-        Type *CondTy = Type::getInt1Ty(RetTy->getContext());
-        if (RetVF > 1)
-          CondTy = VectorType::get(CondTy, RetVF);
+        Type *CondTy = RetTy->changeBitWidth(1);
         Cost += ConcreteTTI->getCmpSelInstrCost(BinaryOperator::ICmp, RetTy,
                                                 CondTy, nullptr);
         Cost += ConcreteTTI->getCmpSelInstrCost(BinaryOperator::Select, RetTy,
                                                 CondTy, nullptr);
       }
       return Cost;
     }
     }
   }
 
   /// Get intrinsic cost based on argument types.
   /// If ScalarizationCostPassed is std::numeric_limits<unsigned>::max(), the
   /// cost of scalarizing the arguments and the return value will be computed
   /// based on types.
   unsigned getIntrinsicInstrCost(
       Intrinsic::ID IID, Type *RetTy, ArrayRef<Type *> Tys, FastMathFlags FMF,
       unsigned ScalarizationCostPassed = std::numeric_limits<unsigned>::max()) {
-    unsigned RetVF = (RetTy->isVectorTy() ? RetTy->getVectorNumElements() : 1);
     auto *ConcreteTTI = static_cast<T *>(this);
 
     SmallVector<unsigned, 2> ISDs;
     unsigned SingleCallCost = 10; // Library call cost. Make it expensive.
     switch (IID) {
     default: {
       // Assume that we need to scalarize this intrinsic.
       unsigned ScalarizationCost = ScalarizationCostPassed;
case Intrinsic::experimental_vector_reduce_fmin:
           /*IsUnsigned=*/true);
     case Intrinsic::experimental_vector_reduce_umax:
     case Intrinsic::experimental_vector_reduce_umin:
       return ConcreteTTI->getMinMaxReductionCost(
           Tys[0], CmpInst::makeCmpResultType(Tys[0]), /*IsPairwiseForm=*/false,
           /*IsUnsigned=*/false);
     case Intrinsic::sadd_sat:
     case Intrinsic::ssub_sat: {
-      Type *CondTy = Type::getInt1Ty(RetTy->getContext());
-      if (RetVF > 1)
-        CondTy = VectorType::get(CondTy, RetVF);
+      Type *CondTy = RetTy->changeBitWidth(1);
 
       Type *OpTy = StructType::create({RetTy, CondTy});
       Intrinsic::ID OverflowOp = IID == Intrinsic::sadd_sat
                                      ? Intrinsic::sadd_with_overflow
                                      : Intrinsic::ssub_with_overflow;
 
       // SatMax -> Overflow && SumDiff < 0
       // SatMin -> Overflow && SumDiff >= 0
       unsigned Cost = 0;
       Cost += ConcreteTTI->getIntrinsicInstrCost(
           OverflowOp, OpTy, {RetTy, RetTy}, FMF, ScalarizationCostPassed);
       Cost += ConcreteTTI->getCmpSelInstrCost(BinaryOperator::ICmp, RetTy,
                                               CondTy, nullptr);
       Cost += 2 * ConcreteTTI->getCmpSelInstrCost(BinaryOperator::Select, RetTy,
                                                   CondTy, nullptr);
       return Cost;
     }
     case Intrinsic::uadd_sat:
     case Intrinsic::usub_sat: {
-      Type *CondTy = Type::getInt1Ty(RetTy->getContext());
-      if (RetVF > 1)
-        CondTy = VectorType::get(CondTy, RetVF);
+      Type *CondTy = RetTy->changeBitWidth(1);
 
       Type *OpTy = StructType::create({RetTy, CondTy});
       Intrinsic::ID OverflowOp = IID == Intrinsic::uadd_sat
                                      ? Intrinsic::uadd_with_overflow
                                      : Intrinsic::usub_with_overflow;
 
       unsigned Cost = 0;
       Cost += ConcreteTTI->getIntrinsicInstrCost(
           OverflowOp, OpTy, {RetTy, RetTy}, FMF, ScalarizationCostPassed);
       Cost += ConcreteTTI->getCmpSelInstrCost(BinaryOperator::Select, RetTy,
                                               CondTy, nullptr);
       return Cost;
     }
     case Intrinsic::smul_fix:
     case Intrinsic::umul_fix: {
       unsigned ExtSize = RetTy->getScalarSizeInBits() * 2;
-      Type *ExtTy = Type::getIntNTy(RetTy->getContext(), ExtSize);
-      if (RetVF > 1)
-        ExtTy = VectorType::get(ExtTy, RetVF);
+      Type *ExtTy = RetTy->changeBitWidth(ExtSize);
 
       unsigned ExtOp =
           IID == Intrinsic::smul_fix ? Instruction::SExt : Instruction::ZExt;
 
       unsigned Cost = 0;
       Cost += 2 * ConcreteTTI->getCastInstrCost(ExtOp, ExtTy, RetTy);
       Cost += ConcreteTTI->getArithmeticInstrCost(Instruction::Mul, ExtTy);
       Cost +=
case Intrinsic::usub_with_overflow: {
                                               OverflowTy, nullptr);
       return Cost;
     }
     case Intrinsic::smul_with_overflow:
     case Intrinsic::umul_with_overflow: {
       Type *MulTy = RetTy->getContainedType(0);
       Type *OverflowTy = RetTy->getContainedType(1);
       unsigned ExtSize = MulTy->getScalarSizeInBits() * 2;
-      Type *ExtTy = Type::getIntNTy(RetTy->getContext(), ExtSize);
-      if (MulTy->isVectorTy())
-        ExtTy = VectorType::get(ExtTy, MulTy->getVectorNumElements() );
+      Type *ExtTy = MulTy->changeBitWidth(ExtSize);
 
       unsigned ExtOp =
           IID == Intrinsic::smul_fix ? Instruction::SExt : Instruction::ZExt;
 
       unsigned Cost = 0;
       Cost += 2 * ConcreteTTI->getCastInstrCost(ExtOp, ExtTy, MulTy);
       Cost += ConcreteTTI->getArithmeticInstrCost(Instruction::Mul, ExtTy);
       Cost +=

llvm/include/llvm/IR/DerivedTypes.h

 unsigned Type::getVectorNumElements() const {
   return cast<VectorType>(this)->getNumElements();
 }
 
 bool Type::getVectorIsScalable() const {
   return cast<VectorType>(this)->isScalable();
 }
 
+ElementCount Type::getVectorElementCount() const {
+  return cast<VectorType>(this)->getElementCount();
+}
+
 /// Class to represent pointers.
 class PointerType : public Type {
   explicit PointerType(Type *ElType, unsigned AddrSpace);
 
   Type *PointeeTy;
 
 public:
   PointerType(const PointerType &) = delete;
assert(
       isIntOrIntVectorTy() &&
       "Original type expected to be a vector of integers or a scalar integer.");
   if (auto *VTy = dyn_cast<VectorType>(this))
     return VectorType::getExtendedElementVectorType(
         const_cast<VectorType *>(VTy));
   return cast<IntegerType>(this)->getExtendedType();
 }
 
+Type *Type::changeBitWidth(unsigned NewBitWidth) const {
+  assert(
+      isIntOrIntVectorTy() &&
+      "Original type expected to be a vector of integers or a scalar integer.");
+  Type *NewType = getIntNTy(getContext(), NewBitWidth);
+  if (isVectorTy())
+    NewType = VectorType::get(NewType, getVectorElementCount());
+  return NewType;
+}
+
 unsigned Type::getPointerAddressSpace() const {
   return cast<PointerType>(getScalarType())->getAddressSpace();
 }
 
 } // end namespace llvm
 
 #endif // LLVM_IR_DERIVEDTYPES_H

llvm/include/llvm/IR/Type.h

public:
 
   Type *getArrayElementType() const {
     assert(getTypeID() == ArrayTyID);
     return ContainedTys[0];
   }
 
   inline bool getVectorIsScalable() const;
   inline unsigned getVectorNumElements() const;
+  inline ElementCount getVectorElementCount() const;
   Type *getVectorElementType() const {
     assert(getTypeID() == VectorTyID);
     return ContainedTys[0];
   }
 
   Type *getPointerElementType() const {
     assert(getTypeID() == PointerTyID);
     return ContainedTys[0];
   }
 
+  /// Given an integer or vector type, change the bitwidth to NewBitwidth,
+  /// whilst keeping the old number of lanes.
+  inline Type *changeBitWidth(unsigned NewBitwidth) const;
+
   /// Given scalar/vector integer type, returns a type with elements twice as
   /// wide as in the original type. For vectors, preserves element count.
   inline Type *getExtendedType() const;
 
   /// Get the address space of this pointer or pointer vector type.
   inline unsigned getPointerAddressSpace() const;
 
   //===--------------------------------------------------------------------===//