[Costmodel] Add "type-based-intrinsic-cost" cli option

[malharJ]

Summary

This patch adds a command line flag to be able to test the type based cost-model analysis
for Intrinsics.

[malharJ]

• Added updates when printing instruction cost using new pass manager.

[malharJ]

• Updated LIT test
• minor formatting updates

[mgabka]

I think it would be good to explain in the commit message that this patch allows to print cost calculated only based on argument types and that it is useful feature to have for testing purposes.

llvm/include/llvm/Analysis/TargetTransformInfo.h

133

incorrect formatting

llvm/lib/Analysis/TargetTransformInfo.cpp

62

would it be beneficial to keep the TypeBasedOnly as a property of IntrinsicCostAttributes ?

[kiranchandramohan]

A couple of drive-through Nit comments. Feel free to skip.

llvm/include/llvm/CodeGen/BasicTTIImpl.h

1787–1793

Nit: Would this be better for here and below?

llvm/lib/Analysis/CostModel.cpp

118

Nit: Consider using braces everywhere in an if-else construct, here and below.

" For example, readability is also harmed if an if/else chain does not use braced bodies for either all or none of its members, ... etc."
https://llvm.org/docs/CodingStandards.html#don-t-use-braces-on-simple-single-statement-bodies-of-if-else-loop-statements

[malharJ]

• Addressed minor formatting issues

llvm/lib/Analysis/TargetTransformInfo.cpp

62

I think given that there is only one usage right now, ie. in the constructor below,
perhaps it's not worth making into a member variable ?

[malharJ]

ping

[mgabka]

It looks reasonable to me.
The main issue we saw in the past was assertion failure when the cost for gather/scatter wasn't implemented, but you are saying that now it is simply returning invalid cost so the type based cost model for gather/scatter is not needed.
But I guess it does not hurt to add it.
@david-arm you did some work on the cost model for gather/scatter from what I can see, could you have a look if you agree with the proposed changes?

[mgabka]

llvm/include/llvm/CodeGen/BasicTTIImpl.h

1782

@malharJ could you remove this and the case below?

I think this patch could be limited to just adding the new interface to the cost model to allow extra testing.

With the cost for masked_gather and massked_scatter, it needs to be done a bit differently as with this changes you made it will always add in the legalization cost. I think it needs to be done by adding extra AArch64TTIImpl::getTypeBasedIntrinsicInstrCost method

But the main purpose of this patch is to ensure that all paths in the cost model can be unit tested and returning an invalid cost for scalable vectors is still something what compiler should support in other parts.

[malharJ]

• Removed the SVE gather/scatter case as per review comment

[malharJ]

• removed blank line

[malharJ]

ping

[mgabka]

llvm/lib/Analysis/TargetTransformInfo.cpp

70

looks like you added a tabulation here instead of spaces

llvm/include/llvm/Analysis/TargetTransformInfo.h

class IntrinsicCostAttributes {
   FastMathFlags FMF;
   // If ScalarizationCost is UINT_MAX, the cost of scalarizing the
   // arguments and the return value will be computed based on types.
   InstructionCost ScalarizationCost = InstructionCost::getInvalid();
 
 public:
   IntrinsicCostAttributes(
       Intrinsic::ID Id, const CallBase &CI,
-      InstructionCost ScalarCost = InstructionCost::getInvalid());
+      InstructionCost ScalarCost = InstructionCost::getInvalid(),
+      bool TypeBasedOnly=false);
 
   IntrinsicCostAttributes(
       Intrinsic::ID Id, Type *RTy, ArrayRef<Type *> Tys,
       FastMathFlags Flags = FastMathFlags(), const IntrinsicInst *I = nullptr,
       InstructionCost ScalarCost = InstructionCost::getInvalid());
 
   IntrinsicCostAttributes(Intrinsic::ID Id, Type *RTy,
                           ArrayRef<const Value *> Args);

llvm/include/llvm/CodeGen/BasicTTIImpl.h

case Intrinsic::masked_store: {
                                             CostKind);
     }
     case Intrinsic::masked_load: {
       Type *Ty = RetTy;
       Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
       return thisT()->getMaskedMemoryOpCost(Instruction::Load, Ty, TyAlign, 0,
                                             CostKind);
     }
+
+    case Intrinsic::masked_scatter: {
+      // For type based cost assume worse case scenario.
+      bool VarMask = true;
+      Type *Ty = Tys[0];
+      Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
+      // arguments below are actually not used by
+      // getGatherScatterOpCost
+      const Value *Ptr = nullptr;
+      const Instruction *I = nullptr;
+      return thisT()->getGatherScatterOpCost(Instruction::Store,
+                                             Ty, Ptr, VarMask, TyAlign,
+                                             CostKind, I);
+    }
+
+    case Intrinsic::masked_gather: {
+      // For type based cost assume worse case scenario.
+      bool VarMask = true;
+      Type *Ty = RetTy;
+      Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
+      // arguments below are actually not used by
+      // getGatherScatterOpCost
+      const Value *Ptr = nullptr;
+      const Instruction *I = nullptr;
+      return thisT()->getGatherScatterOpCost(Instruction::Load, RetTy, Ptr,
+                                             VarMask, TyAlign, CostKind, I);
+    }
+
     case Intrinsic::vector_reduce_add:
       return thisT()->getArithmeticReductionCost(Instruction::Add, VecOpTy,
                                                  None, CostKind);
     case Intrinsic::vector_reduce_mul:
       return thisT()->getArithmeticReductionCost(Instruction::Mul, VecOpTy,
                                                  None, CostKind);
     case Intrinsic::vector_reduce_and:
       return thisT()->getArithmeticReductionCost(Instruction::And, VecOpTy,

llvm/lib/Analysis/CostModel.cpp

 #include "llvm/Analysis/Passes.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/IR/IntrinsicInst.h"
 using namespace llvm;
 
 static cl::opt<TargetTransformInfo::TargetCostKind> CostKind(
     "cost-kind", cl::desc("Target cost kind"),
     cl::init(TargetTransformInfo::TCK_RecipThroughput),
     cl::values(clEnumValN(TargetTransformInfo::TCK_RecipThroughput,
                           "throughput", "Reciprocal throughput"),
                clEnumValN(TargetTransformInfo::TCK_Latency,
                           "latency", "Instruction latency"),
                clEnumValN(TargetTransformInfo::TCK_CodeSize,
                           "code-size", "Code size"),
                clEnumValN(TargetTransformInfo::TCK_SizeAndLatency,
                           "size-latency", "Code size and latency")));
 
+static cl::opt<bool> TypeBasedIntrinsicCost("type-based-intrinsic-cost",
+    cl::desc("Calculate intrinsics cost based only on argument types"),
+    cl::init(false));
 
 #define CM_NAME "cost-model"
 #define DEBUG_TYPE CM_NAME
 
 namespace {
   class CostModelAnalysis : public FunctionPass {
 
   public:

 }
 
 void CostModelAnalysis::print(raw_ostream &OS, const Module*) const {
   if (!F)
     return;
 
   for (BasicBlock &B : *F) {
     for (Instruction &Inst : B) {
-      InstructionCost Cost = TTI->getInstructionCost(&Inst, CostKind);
+      InstructionCost Cost;
+      if (TypeBasedIntrinsicCost && isa<IntrinsicInst>(&Inst)) {
+        auto *II = dyn_cast<IntrinsicInst>(&Inst);
+        IntrinsicCostAttributes ICA(II->getIntrinsicID(), *II,
+                                    InstructionCost::getInvalid(), true);
+        Cost = TTI->getIntrinsicInstrCost(ICA, CostKind);
+      }
+      else
+        Cost = TTI->getInstructionCost(&Inst, CostKind);
       if (auto CostVal = Cost.getValue())
         OS << "Cost Model: Found an estimated cost of " << *CostVal;
       else
         OS << "Cost Model: Invalid cost";
 
       OS << " for instruction: " << Inst << "\n";
     }
   }
 }
 
 PreservedAnalyses CostModelPrinterPass::run(Function &F,
                                             FunctionAnalysisManager &AM) {
   auto &TTI = AM.getResult<TargetIRAnalysis>(F);
   OS << "Printing analysis 'Cost Model Analysis' for function '" << F.getName() << "':\n";
   for (BasicBlock &B : F) {
     for (Instruction &Inst : B) {
       // TODO: Use a pass parameter instead of cl::opt CostKind to determine
       // which cost kind to print.
-      InstructionCost Cost = TTI.getInstructionCost(&Inst, CostKind);
+      InstructionCost Cost;
+      if (TypeBasedIntrinsicCost && isa<IntrinsicInst>(&Inst)) {
+        auto *II = dyn_cast<IntrinsicInst>(&Inst);
+        IntrinsicCostAttributes ICA(II->getIntrinsicID(), *II,
+                                    InstructionCost::getInvalid(), true);
+        Cost = TTI.getIntrinsicInstrCost(ICA, CostKind);
+      }
+      else
+        Cost = TTI.getInstructionCost(&Inst, CostKind);
       if (auto CostVal = Cost.getValue())
         OS << "Cost Model: Found an estimated cost of " << *CostVal;
       else
         OS << "Cost Model: Invalid cost";
 
       OS << " for instruction: " << Inst << "\n";
     }
   }
   return PreservedAnalyses::all();
 }

llvm/lib/Analysis/TargetTransformInfo.cpp

bool HardwareLoopInfo::canAnalyze(LoopInfo &LI) {
   LoopBlocksRPO RPOT(L);
   RPOT.perform(&LI);
   if (containsIrreducibleCFG<const BasicBlock *>(RPOT, LI))
     return false;
   return true;
 }
 
 IntrinsicCostAttributes::IntrinsicCostAttributes(
-    Intrinsic::ID Id, const CallBase &CI, InstructionCost ScalarizationCost)
+    Intrinsic::ID Id, const CallBase &CI, InstructionCost ScalarizationCost,
+    bool TypeBasedOnly)
     : II(dyn_cast<IntrinsicInst>(&CI)), RetTy(CI.getType()), IID(Id),
       ScalarizationCost(ScalarizationCost) {
 
   if (const auto *FPMO = dyn_cast<FPMathOperator>(&CI))
     FMF = FPMO->getFastMathFlags();
 
+  if (!TypeBasedOnly)
     Arguments.insert(Arguments.begin(), CI.arg_begin(), CI.arg_end());
   FunctionType *FTy = CI.getCalledFunction()->getFunctionType();
   ParamTys.insert(ParamTys.begin(), FTy->param_begin(), FTy->param_end());
 }
 
 IntrinsicCostAttributes::IntrinsicCostAttributes(Intrinsic::ID Id, Type *RTy,
                                                  ArrayRef<Type *> Tys,
                                                  FastMathFlags Flags,
                                                  const IntrinsicInst *I,

llvm/test/Analysis/CostModel/AArch64/sve-intrinsics.ll

 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
 ; RUN: opt < %s -passes='print<cost-model>' 2>&1 -disable-output -S -mtriple=aarch64--linux-gnu -mattr=+sve | FileCheck %s
+; RUN: opt < %s -passes='print<cost-model>' 2>&1 -type-based-intrinsic-cost -disable-output -S -mtriple=aarch64--linux-gnu -mattr=+sve | FileCheck %s --check-prefix=TYPE_BASED_ONLY
 
 define void @vector_insert_extract(<vscale x 4 x i32> %v0, <vscale x 16 x i32> %v1, <16 x i32> %v2) {
 ; CHECK-LABEL: 'vector_insert_extract'
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %extract_fixed_from_scalable = call <16 x i32> @llvm.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> %v0, i64 0)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %insert_fixed_into_scalable = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> %v0, <16 x i32> %v2, i64 0)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %extract_scalable_from_scalable = call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> %v1, i64 0)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %insert_scalable_into_scalable = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> %v1, <vscale x 4 x i32> %v0, i64 0)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void

 ;
   call <vscale x 16 x i8> @llvm.fshl.nxv16i8(<vscale x 16 x i8> undef, <vscale x 16 x i8> undef, <vscale x 16 x i8> undef)
   call <vscale x 8 x i16> @llvm.fshl.nxv8i16(<vscale x 8 x i16> undef, <vscale x 8 x i16> undef, <vscale x 8 x i16> undef)
   call <vscale x 4 x i32> @llvm.fshl.nxv4i32(<vscale x 4 x i32> undef, <vscale x 4 x i32> undef, <vscale x 4 x i32> undef)
   call <vscale x 2 x i64> @llvm.fshl.nxv2i64(<vscale x 2 x i64> undef, <vscale x 2 x i64> undef, <vscale x 2 x i64> undef)
   ret void
 }
 
+define <vscale x 4 x i32> @masked_gather_nxv4i32(<vscale x 4 x i32*> %ld, <vscale x 4 x i1> %masks, <vscale x 4 x i32> %passthru) {
+; CHECK-LABEL: 'masked_gather_nxv4i32'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %res = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0i32(<vscale x 4 x i32*> %ld, i32 0, <vscale x 4 x i1> %masks, <vscale x 4 x i32> %passthru)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <vscale x 4 x i32> %res
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_gather_nxv4i32'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %res = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0i32(<vscale x 4 x i32*> %ld, i32 0, <vscale x 4 x i1> %masks, <vscale x 4 x i32> %passthru)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <vscale x 4 x i32> %res
+;
+  %res = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32(<vscale x 4 x i32*> %ld, i32 0, <vscale x 4 x i1> %masks, <vscale x 4 x i32> %passthru)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 8 x i32> @masked_gather_nxv8i32(<vscale x 8 x i32*> %ld, <vscale x 8 x i1> %masks, <vscale x 8 x i32> %passthru) {
+; CHECK-LABEL: 'masked_gather_nxv8i32'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %res = call <vscale x 8 x i32> @llvm.masked.gather.nxv8i32.nxv8p0i32(<vscale x 8 x i32*> %ld, i32 0, <vscale x 8 x i1> %masks, <vscale x 8 x i32> %passthru)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <vscale x 8 x i32> %res
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_gather_nxv8i32'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %res = call <vscale x 8 x i32> @llvm.masked.gather.nxv8i32.nxv8p0i32(<vscale x 8 x i32*> %ld, i32 0, <vscale x 8 x i1> %masks, <vscale x 8 x i32> %passthru)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <vscale x 8 x i32> %res
+;
+  %res = call <vscale x 8 x i32> @llvm.masked.gather.nxv8i32(<vscale x 8 x i32*> %ld, i32 0, <vscale x 8 x i1> %masks, <vscale x 8 x i32> %passthru)
+  ret <vscale x 8 x i32> %res
+}
+
+define <4 x i32> @masked_gather_v4i32(<4 x i32*> %ld, <4 x i1> %masks, <4 x i32> %passthru) {
+; CHECK-LABEL: 'masked_gather_v4i32'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0i32(<4 x i32*> %ld, i32 0, <4 x i1> %masks, <4 x i32> %passthru)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_gather_v4i32'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %res = call <4 x i32> @llvm.masked.gather.v4i32.v4p0i32(<4 x i32*> %ld, i32 0, <4 x i1> %masks, <4 x i32> %passthru)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
+;
+  %res = call <4 x i32> @llvm.masked.gather.v4i32(<4 x i32*> %ld, i32 0, <4 x i1> %masks, <4 x i32> %passthru)
+  ret <4 x i32> %res
+}
+
+define <1 x i128> @masked_gather_v1i128(<1 x i128*> %ld, <1 x i1> %masks, <1 x i128> %passthru) {
+; CHECK-LABEL: 'masked_gather_v1i128'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %res = call <1 x i128> @llvm.masked.gather.v1i128.v1p0i128(<1 x i128*> %ld, i32 0, <1 x i1> %masks, <1 x i128> %passthru)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <1 x i128> %res
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_gather_v1i128'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %res = call <1 x i128> @llvm.masked.gather.v1i128.v1p0i128(<1 x i128*> %ld, i32 0, <1 x i1> %masks, <1 x i128> %passthru)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <1 x i128> %res
+;
+  %res = call <1 x i128> @llvm.masked.gather.v1i128.v1p0i128(<1 x i128*> %ld, i32 0, <1 x i1> %masks, <1 x i128> %passthru)
+  ret <1 x i128> %res
+}
+
+define void @masked_scatter_nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i32*> %ptrs, <vscale x 4 x i1> %masks) {
+; CHECK-LABEL: 'masked_scatter_nxv4i32'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: call void @llvm.masked.scatter.nxv4i32.nxv4p0i32(<vscale x 4 x i32> %data, <vscale x 4 x i32*> %ptrs, i32 0, <vscale x 4 x i1> %masks)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_scatter_nxv4i32'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: call void @llvm.masked.scatter.nxv4i32.nxv4p0i32(<vscale x 4 x i32> %data, <vscale x 4 x i32*> %ptrs, i32 0, <vscale x 4 x i1> %masks)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+
+  call void @llvm.masked.scatter.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i32*> %ptrs, i32 0, <vscale x 4 x i1> %masks)
+  ret void
+}
+
+define void @masked_scatter_nxv8i32(<vscale x 8 x i32> %data, <vscale x 8 x i32*> %ptrs, <vscale x 8 x i1> %masks) {
+; CHECK-LABEL: 'masked_scatter_nxv8i32'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: call void @llvm.masked.scatter.nxv8i32.nxv8p0i32(<vscale x 8 x i32> %data, <vscale x 8 x i32*> %ptrs, i32 0, <vscale x 8 x i1> %masks)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_scatter_nxv8i32'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: call void @llvm.masked.scatter.nxv8i32.nxv8p0i32(<vscale x 8 x i32> %data, <vscale x 8 x i32*> %ptrs, i32 0, <vscale x 8 x i1> %masks)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+
+  call void @llvm.masked.scatter.nxv8i32(<vscale x 8 x i32> %data, <vscale x 8 x i32*> %ptrs, i32 0, <vscale x 8 x i1> %masks)
+  ret void
+}
+
+define void @masked_scatter_v4i32(<4 x i32> %data, <4 x i32*> %ptrs, <4 x i1> %masks) {
+; CHECK-LABEL: 'masked_scatter_v4i32'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %data, <4 x i32*> %ptrs, i32 0, <4 x i1> %masks)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_scatter_v4i32'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %data, <4 x i32*> %ptrs, i32 0, <4 x i1> %masks)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+
+  call void @llvm.masked.scatter.v4i32(<4 x i32> %data, <4 x i32*> %ptrs, i32 0, <4 x i1> %masks)
+  ret void
+}
+
+define void @masked_scatter_v1i128(<1 x i128> %data, <1 x i128*> %ptrs, <1 x i1> %masks) {
+; CHECK-LABEL: 'masked_scatter_v1i128'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: call void @llvm.masked.scatter.v1i128.v1p0i128(<1 x i128> %data, <1 x i128*> %ptrs, i32 0, <1 x i1> %masks)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; TYPE_BASED_ONLY-LABEL: 'masked_scatter_v1i128'
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: call void @llvm.masked.scatter.v1i128.v1p0i128(<1 x i128> %data, <1 x i128*> %ptrs, i32 0, <1 x i1> %masks)
+; TYPE_BASED_ONLY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+
+  call void @llvm.masked.scatter.v1i128.v1p0i128(<1 x i128> %data, <1 x i128*> %ptrs, i32 0, <1 x i1> %masks)
+  ret void
+}
+
 declare <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i64(i64, i64)
 declare <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i64(i64, i64)
 declare <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i64(i64, i64)
 declare <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64, i64)
 declare <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i32(i32, i32)
 declare <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i32(i32, i32)
 declare <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i32(i32, i32)
 declare <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i32(i32, i32)

 declare <vscale x 16 x i8> @llvm.fshr.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>)
 declare <vscale x 8 x i16> @llvm.fshr.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>)
 declare <vscale x 4 x i32> @llvm.fshr.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>)
 declare <vscale x 2 x i64> @llvm.fshr.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>)
 declare <vscale x 16 x i8> @llvm.fshl.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>)
 declare <vscale x 8 x i16> @llvm.fshl.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>)
 declare <vscale x 4 x i32> @llvm.fshl.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>)
 declare <vscale x 2 x i64> @llvm.fshl.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 4 x i32> @llvm.masked.gather.nxv4i32(<vscale x 4 x i32*> %ptrs, i32 %align, <vscale x 4 x i1> %masks, <vscale x 4 x i32> %passthru)
+declare <vscale x 8 x i32> @llvm.masked.gather.nxv8i32(<vscale x 8 x i32*> %ptrs, i32 %align, <vscale x 8 x i1> %masks, <vscale x 8 x i32> %passthru)
+declare <4 x i32> @llvm.masked.gather.v4i32(<4 x i32*> %ptrs, i32 %align, <4 x i1> %masks, <4 x i32> %passthru)
+declare <1 x i128> @llvm.masked.gather.v1i128.v1p0i128(<1 x i128*>, i32, <1 x i1>, <1 x i128>)
+declare void @llvm.masked.scatter.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i32*> %ptrs, i32 %align, <vscale x 4 x i1> %masks)
+declare void @llvm.masked.scatter.nxv8i32(<vscale x 8 x i32> %data, <vscale x 8 x i32*> %ptrs, i32 %align, <vscale x 8 x i1> %masks)
+declare void @llvm.masked.scatter.v4i32(<4 x i32> %data, <4 x i32*> %ptrs, i32 %align, <4 x i1> %masks)
+declare void @llvm.masked.scatter.v1i128.v1p0i128(<1 x i128> %data, <1 x i128*> %ptrs, i32 %align, <1 x i1> %masks)
 
 attributes #0 = { "target-features"="+sve,+bf16" }