[SVE][NFC] Use ScalableVectorType in CGBuiltin

[ctetreau]

Summary

• Upgrade some usages of VectorType to use ScalableVectorType

[efriedma]

LGTM

[ctetreau]

rebase

[ctetreau]

fix syntax error

clang/lib/CodeGen/CGBuiltin.cpp

llvm::Type *CodeGenFunction::getEltType(SVETypeFlags TypeFlags) {
   case SVETypeFlags::EltTyBool32:
   case SVETypeFlags::EltTyBool64:
     return Builder.getInt1Ty();
   }
 }
 
 // Return the llvm predicate vector type corresponding to the specified element
 // TypeFlags.
-llvm::VectorType* CodeGenFunction::getSVEPredType(SVETypeFlags TypeFlags) {
+llvm::ScalableVectorType *
+CodeGenFunction::getSVEPredType(SVETypeFlags TypeFlags) {
   switch (TypeFlags.getEltType()) {
   default: llvm_unreachable("Unhandled SVETypeFlag!");
 
   case SVETypeFlags::EltTyInt8:
-    return llvm::VectorType::get(Builder.getInt1Ty(), { 16, true });
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16);
   case SVETypeFlags::EltTyInt16:
-    return llvm::VectorType::get(Builder.getInt1Ty(), { 8, true });
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
   case SVETypeFlags::EltTyInt32:
-    return llvm::VectorType::get(Builder.getInt1Ty(), { 4, true });
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4);
   case SVETypeFlags::EltTyInt64:
-    return llvm::VectorType::get(Builder.getInt1Ty(), { 2, true });
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2);
 
   case SVETypeFlags::EltTyFloat16:
-    return llvm::VectorType::get(Builder.getInt1Ty(), { 8, true });
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
   case SVETypeFlags::EltTyFloat32:
-    return llvm::VectorType::get(Builder.getInt1Ty(), { 4, true });
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4);
   case SVETypeFlags::EltTyFloat64:
-    return llvm::VectorType::get(Builder.getInt1Ty(), { 2, true });
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2);
   }
 }
 
 // Return the llvm vector type corresponding to the specified element TypeFlags.
-llvm::VectorType *CodeGenFunction::getSVEType(const SVETypeFlags &TypeFlags) {
+llvm::ScalableVectorType *
+CodeGenFunction::getSVEType(const SVETypeFlags &TypeFlags) {
   switch (TypeFlags.getEltType()) {
   default:
     llvm_unreachable("Invalid SVETypeFlag!");
 
   case SVETypeFlags::EltTyInt8:
-    return llvm::VectorType::get(Builder.getInt8Ty(), {16, true});
+    return llvm::ScalableVectorType::get(Builder.getInt8Ty(), 16);
   case SVETypeFlags::EltTyInt16:
-    return llvm::VectorType::get(Builder.getInt16Ty(), {8, true});
+    return llvm::ScalableVectorType::get(Builder.getInt16Ty(), 8);
   case SVETypeFlags::EltTyInt32:
-    return llvm::VectorType::get(Builder.getInt32Ty(), {4, true});
+    return llvm::ScalableVectorType::get(Builder.getInt32Ty(), 4);
   case SVETypeFlags::EltTyInt64:
-    return llvm::VectorType::get(Builder.getInt64Ty(), {2, true});
+    return llvm::ScalableVectorType::get(Builder.getInt64Ty(), 2);
 
   case SVETypeFlags::EltTyFloat16:
-    return llvm::VectorType::get(Builder.getHalfTy(), {8, true});
+    return llvm::ScalableVectorType::get(Builder.getHalfTy(), 8);
   case SVETypeFlags::EltTyFloat32:
-    return llvm::VectorType::get(Builder.getFloatTy(), {4, true});
+    return llvm::ScalableVectorType::get(Builder.getFloatTy(), 4);
   case SVETypeFlags::EltTyFloat64:
-    return llvm::VectorType::get(Builder.getDoubleTy(), {2, true});
+    return llvm::ScalableVectorType::get(Builder.getDoubleTy(), 2);
 
   case SVETypeFlags::EltTyBool8:
-    return llvm::VectorType::get(Builder.getInt1Ty(), {16, true});
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 16);
   case SVETypeFlags::EltTyBool16:
-    return llvm::VectorType::get(Builder.getInt1Ty(), {8, true});
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 8);
   case SVETypeFlags::EltTyBool32:
-    return llvm::VectorType::get(Builder.getInt1Ty(), {4, true});
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 4);
   case SVETypeFlags::EltTyBool64:
-    return llvm::VectorType::get(Builder.getInt1Ty(), {2, true});
+    return llvm::ScalableVectorType::get(Builder.getInt1Ty(), 2);
   }
 }
 
 constexpr unsigned SVEBitsPerBlock = 128;
 
-static llvm::VectorType* getSVEVectorForElementType(llvm::Type *EltTy) {
+static llvm::ScalableVectorType *getSVEVectorForElementType(llvm::Type *EltTy) {
   unsigned NumElts = SVEBitsPerBlock / EltTy->getScalarSizeInBits();
-  return llvm::VectorType::get(EltTy, { NumElts, true });
+  return llvm::ScalableVectorType::get(EltTy, NumElts);
 }
 
 // Reinterpret the input predicate so that it can be used to correctly isolate
 // the elements of the specified datatype.
 Value *CodeGenFunction::EmitSVEPredicateCast(Value *Pred,
                                              llvm::VectorType *VTy) {
   llvm::VectorType *RTy = llvm::VectorType::get(
       IntegerType::get(getLLVMContext(), 1), VTy->getElementCount());
Value *CodeGenFunction::EmitSVEPredicateCast(Value *Pred,
   assert(C->getType() == RTy && "Unexpected return type!");
   return C;
 }
 
 Value *CodeGenFunction::EmitSVEGatherLoad(SVETypeFlags TypeFlags,
                                           SmallVectorImpl<Value *> &Ops,
                                           unsigned IntID) {
   auto *ResultTy = getSVEType(TypeFlags);
-  auto *OverloadedTy = llvm::VectorType::get(SVEBuiltinMemEltTy(TypeFlags),
-                                             ResultTy->getElementCount());
+  auto *OverloadedTy =
+      llvm::ScalableVectorType::get(SVEBuiltinMemEltTy(TypeFlags), ResultTy);
 
   // At the ACLE level there's only one predicate type, svbool_t, which is
   // mapped to <n x 16 x i1>. However, this might be incompatible with the
   // actual type being loaded. For example, when loading doubles (i64) the
   // predicated should be <n x 2 x i1> instead. At the IR level the type of
   // the predicate and the data being loaded must match. Cast accordingly.
   Ops[0] = EmitSVEPredicateCast(Ops[0], OverloadedTy);
 
Value *CodeGenFunction::EmitSVEGatherLoad(SVETypeFlags TypeFlags,
   return TypeFlags.isZExtReturn() ? Builder.CreateZExt(Call, ResultTy)
                                   : Builder.CreateSExt(Call, ResultTy);
 }
 
 Value *CodeGenFunction::EmitSVEScatterStore(SVETypeFlags TypeFlags,
                                             SmallVectorImpl<Value *> &Ops,
                                             unsigned IntID) {
   auto *SrcDataTy = getSVEType(TypeFlags);
-  auto *OverloadedTy = llvm::VectorType::get(SVEBuiltinMemEltTy(TypeFlags),
-                                             SrcDataTy->getElementCount());
+  auto *OverloadedTy =
+      llvm::ScalableVectorType::get(SVEBuiltinMemEltTy(TypeFlags), SrcDataTy);
 
   // In ACLE the source data is passed in the last argument, whereas in LLVM IR
   // it's the first argument. Move it accordingly.
   Ops.insert(Ops.begin(), Ops.pop_back_val());
 
   Function *F = nullptr;
   if (Ops[2]->getType()->isVectorTy())
     // This is the "vector base, scalar offset" case. In order to uniquely
Value *CodeGenFunction::EmitSVEScatterStore(SVETypeFlags TypeFlags,
   return Builder.CreateCall(F, Ops);
 }
 
 Value *CodeGenFunction::EmitSVEPrefetchLoad(SVETypeFlags TypeFlags,
                                             SmallVectorImpl<Value *> &Ops,
                                             unsigned BuiltinID) {
   auto *MemEltTy = SVEBuiltinMemEltTy(TypeFlags);
   auto *VectorTy = getSVEVectorForElementType(MemEltTy);
-  auto *MemoryTy = llvm::VectorType::get(MemEltTy, VectorTy->getElementCount());
+  auto *MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy);
 
   Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy);
   Value *BasePtr = Ops[1];
 
   // Implement the index operand if not omitted.
   if (Ops.size() > 3) {
     BasePtr = Builder.CreateBitCast(BasePtr, MemoryTy->getPointerTo());
     BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Ops[2]);
Value *CodeGenFunction::EmitSVEMaskedLoad(const CallExpr *E,
                                           unsigned BuiltinID,
                                           bool IsZExtReturn) {
   QualType LangPTy = E->getArg(1)->getType();
   llvm::Type *MemEltTy = CGM.getTypes().ConvertType(
       LangPTy->getAs<PointerType>()->getPointeeType());
 
   // The vector type that is returned may be different from the
   // eventual type loaded from memory.
-  auto VectorTy = cast<llvm::VectorType>(ReturnTy);
-  auto MemoryTy = llvm::VectorType::get(MemEltTy, VectorTy->getElementCount());
+  auto VectorTy = cast<llvm::ScalableVectorType>(ReturnTy);
+  auto MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy);
 
   Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy);
   Value *BasePtr = Builder.CreateBitCast(Ops[1], MemoryTy->getPointerTo());
   Value *Offset = Ops.size() > 2 ? Ops[2] : Builder.getInt32(0);
   BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Offset);
 
   BasePtr = Builder.CreateBitCast(BasePtr, MemEltTy->getPointerTo());
   Function *F = CGM.getIntrinsic(BuiltinID, MemoryTy);
   Value *Load = Builder.CreateCall(F, {Predicate, BasePtr});
 
   return IsZExtReturn ? Builder.CreateZExt(Load, VectorTy)
                      : Builder.CreateSExt(Load, VectorTy);
 }
 
 Value *CodeGenFunction::EmitSVEMaskedStore(const CallExpr *E,
                                            SmallVectorImpl<Value *> &Ops,
                                            unsigned BuiltinID) {
   QualType LangPTy = E->getArg(1)->getType();
   llvm::Type *MemEltTy = CGM.getTypes().ConvertType(
       LangPTy->getAs<PointerType>()->getPointeeType());
 
   // The vector type that is stored may be different from the
   // eventual type stored to memory.
-  auto VectorTy = cast<llvm::VectorType>(Ops.back()->getType());
-  auto MemoryTy = llvm::VectorType::get(MemEltTy, VectorTy->getElementCount());
+  auto VectorTy = cast<llvm::ScalableVectorType>(Ops.back()->getType());
+  auto MemoryTy = llvm::ScalableVectorType::get(MemEltTy, VectorTy);
 
   Value *Predicate = EmitSVEPredicateCast(Ops[0], MemoryTy);
   Value *BasePtr = Builder.CreateBitCast(Ops[1], MemoryTy->getPointerTo());
   Value *Offset = Ops.size() == 4 ? Ops[2] : Builder.getInt32(0);
   BasePtr = Builder.CreateGEP(MemoryTy, BasePtr, Offset);
 
   // Last value is always the data
   llvm::Value *Val = Builder.CreateTrunc(Ops.back(), MemoryTy);

clang/lib/CodeGen/CodeGenFunction.h

public:
   /// SVEBuiltinMemEltTy - Returns the memory element type for this memory
   /// access builtin.  Only required if it can't be inferred from the base
   /// pointer operand.
   llvm::Type *SVEBuiltinMemEltTy(SVETypeFlags TypeFlags);
 
   SmallVector<llvm::Type *, 2> getSVEOverloadTypes(SVETypeFlags TypeFlags,
                                                    ArrayRef<llvm::Value *> Ops);
   llvm::Type *getEltType(SVETypeFlags TypeFlags);
-  llvm::VectorType *getSVEType(const SVETypeFlags &TypeFlags);
-  llvm::VectorType *getSVEPredType(SVETypeFlags TypeFlags);
+  llvm::ScalableVectorType *getSVEType(const SVETypeFlags &TypeFlags);
+  llvm::ScalableVectorType *getSVEPredType(SVETypeFlags TypeFlags);
   llvm::Value *EmitSVEDupX(llvm::Value *Scalar);
   llvm::Value *EmitSVEPredicateCast(llvm::Value *Pred, llvm::VectorType *VTy);
   llvm::Value *EmitSVEGatherLoad(SVETypeFlags TypeFlags,
                                  llvm::SmallVectorImpl<llvm::Value *> &Ops,
                                  unsigned IntID);
   llvm::Value *EmitSVEScatterStore(SVETypeFlags TypeFlags,
                                    llvm::SmallVectorImpl<llvm::Value *> &Ops,
                                    unsigned IntID);