[AArch64] Emit TBAA metadata for SVE load/store intrinsics

[sdesmalen]

Summary

In Clang we can attach TBAA metadata based on the load/store intrinsics
based on the operation's element type.

This also contains changes to InstCombine where the AArch64-specific
intrinsics are transformed into generic LLVM load/store operations,
to ensure that all metadata is transferred to the new instruction.

There will be some further work after this patch to also emit TBAA
metadata for SVE's gather/scatter- and struct load/store intrinsics.

[paulwalker-arm]

I cannot say I fully understand all the connotations of this change but my gut feeling is that if failures occur it's likely the input program is malformed. Either way, we're early in the LLVM 15 development cycle so have plenty of time to react if necessary.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

888–889

If I'm nit picking you use auto here but then choose the explicit return type (i.e. CallInst) for the MaskedStore local. Is there a reason for this or just muscle memory?

[sdesmalen]

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

888–889

Is there a reason for this or just muscle memory?

I initially assumed the IRBuilder's result was a Value* because that's what most of its methods return, so wrote auto *Store = cast<Storeinst>(Builder.CreateStore(...)). When I realised it wasn't required, I remove the cast, but not the auto. I changed the code for LD1 after fixing ST1, hence the inconsistency.

clang/lib/CodeGen/CGBuiltin.cpp

Value *CodeGenFunction::EmitSVEMaskedLoad(const CallExpr *E,
 
   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});
+  auto *Load =
+      cast<llvm::Instruction>(Builder.CreateCall(F, {Predicate, BasePtr}));
+  auto TBAAInfo = CGM.getTBAAAccessInfo(LangPTy->getPointeeType());
+  CGM.DecorateInstructionWithTBAA(Load, TBAAInfo);
 
   return IsZExtReturn ? Builder.CreateZExt(Load, VectorTy)
                      : Builder.CreateSExt(Load, VectorTy);
 }
 
 Value *CodeGenFunction::EmitSVEMaskedStore(const CallExpr *E,
                                            SmallVectorImpl<Value *> &Ops,
                                            unsigned BuiltinID) {
Value *CodeGenFunction::EmitSVEMaskedStore(const CallExpr *E,
   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);
 
   BasePtr = Builder.CreateBitCast(BasePtr, MemEltTy->getPointerTo());
   Function *F = CGM.getIntrinsic(BuiltinID, MemoryTy);
-  return Builder.CreateCall(F, {Val, Predicate, BasePtr});
+  auto *Store =
+      cast<llvm::Instruction>(Builder.CreateCall(F, {Val, Predicate, BasePtr}));
+  auto TBAAInfo = CGM.getTBAAAccessInfo(LangPTy->getPointeeType());
+  CGM.DecorateInstructionWithTBAA(Store, TBAAInfo);
+  return Store;
 }
 
 // Limit the usage of scalable llvm IR generated by the ACLE by using the
 // sve dup.x intrinsic instead of IRBuilder::CreateVectorSplat.
 Value *CodeGenFunction::EmitSVEDupX(Value *Scalar, llvm::Type *Ty) {
   auto F = CGM.getIntrinsic(Intrinsic::aarch64_sve_dup_x, Ty);
   return Builder.CreateCall(F, Scalar);
 }

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld1.c

 #define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED) A1##A3
 #else
 #define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
 #endif
 
 // CHECK-LABEL: @test_svld1_s8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
-// CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP0]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
+// CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP0]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer), !tbaa ![[CHAR:[0-9]+]]
 // CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld1_s8u10__SVBool_tPKa(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP0]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP1]]
 //
 svint8_t test_svld1_s8(svbool_t pg, const int8_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_s8,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer), !tbaa ![[SHORT:[0-9]+]]
 // CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_s16u10__SVBool_tPKs(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 svint16_t test_svld1_s16(svbool_t pg, const int16_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_s16,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer), !tbaa ![[INT:[0-9]+]]
 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_s32u10__SVBool_tPKi(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 svint32_t test_svld1_s32(svbool_t pg, const int32_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_s32,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_s64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer), !tbaa ![[LONG:[0-9]+]]
 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_s64u10__SVBool_tPKl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 svint64_t test_svld1_s64(svbool_t pg, const int64_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_s64,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_u8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
-// CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP0]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
+// CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP0]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer), !tbaa ![[CHAR]]
 // CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld1_u8u10__SVBool_tPKh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP0]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP1]]
 //
 svuint8_t test_svld1_u8(svbool_t pg, const uint8_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_u8,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer), !tbaa ![[SHORT]]
 // CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_u16u10__SVBool_tPKt(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 svuint16_t test_svld1_u16(svbool_t pg, const uint16_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_u16,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer), !tbaa ![[INT]]
 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_u32u10__SVBool_tPKj(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 svuint32_t test_svld1_u32(svbool_t pg, const uint32_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_u32,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_u64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer), !tbaa ![[LONG]]
 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_u64u10__SVBool_tPKm(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 svuint64_t test_svld1_u64(svbool_t pg, const uint64_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_u64,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_f16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast half* [[BASE:%.*]] to <vscale x 8 x half>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x half> @llvm.masked.load.nxv8f16.p0nxv8f16(<vscale x 8 x half>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x half> @llvm.masked.load.nxv8f16.p0nxv8f16(<vscale x 8 x half>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> zeroinitializer), !tbaa ![[HALF:[0-9]+]]
 // CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_f16u10__SVBool_tPKDh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast half* [[BASE:%.*]] to <vscale x 8 x half>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x half> @llvm.masked.load.nxv8f16.p0nxv8f16(<vscale x 8 x half>* [[TMP1]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
 //
 svfloat16_t test_svld1_f16(svbool_t pg, const float16_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_f16,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_f32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast float* [[BASE:%.*]] to <vscale x 4 x float>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> zeroinitializer), !tbaa ![[FLOAT:[0-9]+]]
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_f32u10__SVBool_tPKf(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast float* [[BASE:%.*]] to <vscale x 4 x float>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP1]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP2]]
 //
 svfloat32_t test_svld1_f32(svbool_t pg, const float32_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_f32,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_f64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast double* [[BASE:%.*]] to <vscale x 2 x double>*
-// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.masked.load.nxv2f64.p0nxv2f64(<vscale x 2 x double>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x double> zeroinitializer)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.masked.load.nxv2f64.p0nxv2f64(<vscale x 2 x double>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x double> zeroinitializer), !tbaa ![[DOUBLE:[0-9]+]]
 // CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld1_f64u10__SVBool_tPKd(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast double* [[BASE:%.*]] to <vscale x 2 x double>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.masked.load.nxv2f64.p0nxv2f64(<vscale x 2 x double>* [[TMP1]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x double> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
 //
 svfloat64_t test_svld1_f64(svbool_t pg, const float64_t *base)
 {
   return SVE_ACLE_FUNC(svld1,_f64,,)(pg, base);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_s8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* [[TMP0]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to <vscale x 16 x i8>*
-// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP2]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
+// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP2]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer), !tbaa ![[CHAR]]
 // CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld1_vnum_s8u10__SVBool_tPKal(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* [[TMP0]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to <vscale x 16 x i8>*
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP2]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 svint8_t test_svld1_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_s8,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast i16* [[TMP2]] to <vscale x 8 x i16>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer), !tbaa ![[SHORT]]
 // CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_s16u10__SVBool_tPKsl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast i16* [[TMP2]] to <vscale x 8 x i16>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 svint16_t test_svld1_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_s16,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer), !tbaa ![[INT]]
 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_s32u10__SVBool_tPKil(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <vscale x 4 x i32>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 svint32_t test_svld1_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_s32,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_s64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[TMP2]] to <vscale x 2 x i64>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer), !tbaa ![[LONG]]
 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_s64u10__SVBool_tPKll(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[TMP2]] to <vscale x 2 x i64>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 svint64_t test_svld1_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_s64,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_u8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* [[TMP0]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to <vscale x 16 x i8>*
-// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP2]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
+// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP2]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer), !tbaa ![[CHAR]]
 // CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld1_vnum_u8u10__SVBool_tPKhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[BASE:%.*]] to <vscale x 16 x i8>*
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* [[TMP0]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to <vscale x 16 x i8>*
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* [[TMP2]], i32 1, <vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i8> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 svuint8_t test_svld1_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_u8,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast i16* [[TMP2]] to <vscale x 8 x i16>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer), !tbaa ![[SHORT]]
 // CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_u16u10__SVBool_tPKtl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[BASE:%.*]] to <vscale x 8 x i16>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast i16* [[TMP2]] to <vscale x 8 x i16>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x i16> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 svuint16_t test_svld1_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_u16,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer), !tbaa ![[INT]]
 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_u32u10__SVBool_tPKjl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[BASE:%.*]] to <vscale x 4 x i32>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <vscale x 4 x i32>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 svuint32_t test_svld1_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_u32,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_u64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[TMP2]] to <vscale x 2 x i64>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer), !tbaa ![[LONG]]
 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_u64u10__SVBool_tPKml(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[BASE:%.*]] to <vscale x 2 x i64>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[TMP2]] to <vscale x 2 x i64>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0nxv2i64(<vscale x 2 x i64>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 svuint64_t test_svld1_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_u64,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_f16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast half* [[BASE:%.*]] to <vscale x 8 x half>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x half>, <vscale x 8 x half>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast half* [[TMP2]] to <vscale x 8 x half>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x half> @llvm.masked.load.nxv8f16.p0nxv8f16(<vscale x 8 x half>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x half> @llvm.masked.load.nxv8f16.p0nxv8f16(<vscale x 8 x half>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> zeroinitializer), !tbaa ![[HALF]]
 // CHECK-NEXT:    ret <vscale x 8 x half> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_f16u10__SVBool_tPKDhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast half* [[BASE:%.*]] to <vscale x 8 x half>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x half>, <vscale x 8 x half>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast half* [[TMP2]] to <vscale x 8 x half>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x half> @llvm.masked.load.nxv8f16.p0nxv8f16(<vscale x 8 x half>* [[TMP3]], i32 1, <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP4]]
 //
 svfloat16_t test_svld1_vnum_f16(svbool_t pg, const float16_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_f16,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_f32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast float* [[BASE:%.*]] to <vscale x 4 x float>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 4 x float>, <vscale x 4 x float>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast float* [[TMP2]] to <vscale x 4 x float>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> zeroinitializer), !tbaa ![[FLOAT]]
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_f32u10__SVBool_tPKfl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast float* [[BASE:%.*]] to <vscale x 4 x float>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 4 x float>, <vscale x 4 x float>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast float* [[TMP2]] to <vscale x 4 x float>*
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> zeroinitializer)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP4]]
 //
 svfloat32_t test_svld1_vnum_f32(svbool_t pg, const float32_t *base, int64_t vnum)
 {
   return SVE_ACLE_FUNC(svld1_vnum,_f32,,)(pg, base, vnum);
 }
 
 // CHECK-LABEL: @test_svld1_vnum_f64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast double* [[BASE:%.*]] to <vscale x 2 x double>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 2 x double>, <vscale x 2 x double>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast double* [[TMP2]] to <vscale x 2 x double>*
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x double> @llvm.masked.load.nxv2f64.p0nxv2f64(<vscale x 2 x double>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x double> zeroinitializer)
+// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x double> @llvm.masked.load.nxv2f64.p0nxv2f64(<vscale x 2 x double>* [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x double> zeroinitializer), !tbaa ![[DOUBLE]]
 // CHECK-NEXT:    ret <vscale x 2 x double> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld1_vnum_f64u10__SVBool_tPKdl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast double* [[BASE:%.*]] to <vscale x 2 x double>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 2 x double>, <vscale x 2 x double>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = bitcast double* [[TMP2]] to <vscale x 2 x double>*

 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 3
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP1]])
 // CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
 //
 svfloat64_t test_svld1_gather_u64base_index_f64(svbool_t pg, svuint64_t bases, int64_t index) {
   return SVE_ACLE_FUNC(svld1_gather, _u64base, _index_f64, )(pg, bases, index);
 }
+
+// CHECK-DAG: ![[CHAR_TY:[0-9]+]] = !{!"omnipotent char", !8, i64 0}
+// CHECK-DAG: ![[CHAR]] = !{![[CHAR_TY]], ![[CHAR_TY]], i64 0}
+// CHECK-DAG: ![[SHORT_TY:[0-9]+]] = !{!"short", !{{.*}}, i64 0}
+// CHECK-DAG: ![[SHORT]] = !{![[SHORT_TY]], ![[SHORT_TY]], i64 0}
+// CHECK-DAG: ![[INT_TY:[0-9]+]] = !{!"int", !7, i64 0}
+// CHECK-DAG: ![[INT]] = !{![[INT_TY]], ![[INT_TY]], i64 0}
+// CHECK-DAG: ![[LONG_TY:[0-9]+]] = !{!"long", !7, i64 0}
+// CHECK-DAG: ![[LONG]] = !{![[LONG_TY]], ![[LONG_TY]], i64 0}
+// CHECK-DAG: ![[HALF_TY:[0-9]+]] = !{!"__fp16", !7, i64 0}
+// CHECK-DAG: ![[HALF]] = !{![[HALF_TY]], ![[HALF_TY]], i64 0}
+// CHECK-DAG: ![[FLOAT_TY:[0-9]+]] = !{!"float", !7, i64 0}
+// CHECK-DAG: ![[FLOAT]] = !{![[FLOAT_TY]], ![[FLOAT_TY]], i64 0}
+// CHECK-DAG: ![[DOUBLE_TY:[0-9]+]] = !{!"double", !7, i64 0}
+// CHECK-DAG: ![[DOUBLE]] = !{![[DOUBLE_TY]], ![[DOUBLE_TY]], i64 0}

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

instCombineSVELD1(InstCombiner &IC, IntrinsicInst &II, const DataLayout &DL) {
 
   Value *Pred = II.getOperand(0);
   Value *PtrOp = II.getOperand(1);
   Type *VecTy = II.getType();
   Value *VecPtr = Builder.CreateBitCast(PtrOp, VecTy->getPointerTo());
 
   if (isAllActivePredicate(Pred)) {
     LoadInst *Load = Builder.CreateLoad(VecTy, VecPtr);
+    Load->copyMetadata(II);
     return IC.replaceInstUsesWith(II, Load);
   }
 
   CallInst *MaskedLoad =
       Builder.CreateMaskedLoad(VecTy, VecPtr, PtrOp->getPointerAlignment(DL),
                                Pred, ConstantAggregateZero::get(VecTy));
+  MaskedLoad->copyMetadata(II);
   return IC.replaceInstUsesWith(II, MaskedLoad);
 }
 
 static Optional<Instruction *>
 instCombineSVEST1(InstCombiner &IC, IntrinsicInst &II, const DataLayout &DL) {
   IRBuilder<> Builder(II.getContext());
   Builder.SetInsertPoint(&II);
 
   Value *VecOp = II.getOperand(0);
   Value *Pred = II.getOperand(1);
   Value *PtrOp = II.getOperand(2);
   Value *VecPtr =
       Builder.CreateBitCast(PtrOp, VecOp->getType()->getPointerTo());
 
   if (isAllActivePredicate(Pred)) {
-    Builder.CreateStore(VecOp, VecPtr);
+    StoreInst *Store = Builder.CreateStore(VecOp, VecPtr);
+    Store->copyMetadata(II);
     return IC.eraseInstFromFunction(II);
   }
 
-  Builder.CreateMaskedStore(VecOp, VecPtr, PtrOp->getPointerAlignment(DL),
-                            Pred);
+  CallInst *MaskedStore = Builder.CreateMaskedStore(
+      VecOp, VecPtr, PtrOp->getPointerAlignment(DL), Pred);
+  MaskedStore->copyMetadata(II);
   return IC.eraseInstFromFunction(II);
 }
 
 static Instruction::BinaryOps intrinsicIDToBinOpCode(unsigned Intrinsic) {
   switch (Intrinsic) {
   case Intrinsic::aarch64_sve_fmul:
     return Instruction::BinaryOps::FMul;
   case Intrinsic::aarch64_sve_fadd:

llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-loadstore.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S -passes=instcombine < %s | FileCheck %s
 
 target triple = "aarch64-unknown-linux-gnu"
 
 define <vscale x 4 x i32> @combine_ld1(i32* %ptr) #0 {
 ; CHECK-LABEL: @combine_ld1(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[PTR:%.*]] to <vscale x 4 x i32>*
-; CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16
+; CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16, !annotation !0
 ; CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 ;
   %1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
-  %2 = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1.nxv4i32(<vscale x 4 x i1> %1, i32* %ptr)
+  %2 = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1.nxv4i32(<vscale x 4 x i1> %1, i32* %ptr), !annotation !0
   ret <vscale x 4 x i32> %2
 }
 
 define <vscale x 4 x i32> @combine_ld1_casted_predicate(i32* %ptr) #0 {
 ; CHECK-LABEL: @combine_ld1_casted_predicate(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[PTR:%.*]] to <vscale x 4 x i32>*
-; CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16
+; CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16, !annotation !0
 ; CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 ;
   %1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
   %2 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %1)
   %3 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %2)
-  %4 = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1.nxv4i32(<vscale x 4 x i1> %3, i32* %ptr)
+  %4 = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1.nxv4i32(<vscale x 4 x i1> %3, i32* %ptr), !annotation !0
   ret <vscale x 4 x i32> %4
 }
 
 define <vscale x 4 x i32> @combine_ld1_masked(i32* %ptr) #0 {
 ; CHECK-LABEL: @combine_ld1_masked(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 16)
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i32* [[PTR:%.*]] to <vscale x 4 x i32>*
-; CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP2]], i32 1, <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> zeroinitializer)
+; CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP2]], i32 1, <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> zeroinitializer), !annotation !0
 ; CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 ;
   %1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 16)
-  %2 = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1.nxv4i32(<vscale x 4 x i1> %1, i32* %ptr)
+  %2 = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1.nxv4i32(<vscale x 4 x i1> %1, i32* %ptr), !annotation !0
   ret <vscale x 4 x i32> %2
 }
 
 define <vscale x 8 x i16> @combine_ld1_masked_casted_predicate(i16* %ptr) #0 {
 ; CHECK-LABEL: @combine_ld1_masked_casted_predicate(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
 ; CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[TMP1]])
 ; CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[TMP2]])
 ; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i16* [[PTR:%.*]] to <vscale x 8 x i16>*
-; CHECK-NEXT:    [[TMP5:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP4]], i32 1, <vscale x 8 x i1> [[TMP3]], <vscale x 8 x i16> zeroinitializer)
+; CHECK-NEXT:    [[TMP5:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0nxv8i16(<vscale x 8 x i16>* [[TMP4]], i32 1, <vscale x 8 x i1> [[TMP3]], <vscale x 8 x i16> zeroinitializer), !annotation !0
 ; CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP5]]
 ;
   %1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
   %2 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
   %3 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %2)
-  %4 = call <vscale x 8 x i16> @llvm.aarch64.sve.ld1.nxv8i16(<vscale x 8 x i1> %3, i16* %ptr)
+  %4 = call <vscale x 8 x i16> @llvm.aarch64.sve.ld1.nxv8i16(<vscale x 8 x i1> %3, i16* %ptr), !annotation !0
   ret <vscale x 8 x i16> %4
 }
 
 define void @combine_st1(<vscale x 4 x i32> %vec, i32* %ptr) #0 {
 ; CHECK-LABEL: @combine_st1(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[PTR:%.*]] to <vscale x 4 x i32>*
-; CHECK-NEXT:    store <vscale x 4 x i32> [[VEC:%.*]], <vscale x 4 x i32>* [[TMP1]], align 16
+; CHECK-NEXT:    store <vscale x 4 x i32> [[VEC:%.*]], <vscale x 4 x i32>* [[TMP1]], align 16, !annotation !0
 ; CHECK-NEXT:    ret void
 ;
   %1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
-  call void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32> %vec, <vscale x 4 x i1> %1, i32* %ptr)
+  call void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32> %vec, <vscale x 4 x i1> %1, i32* %ptr), !annotation !0
   ret void
 }
 
 define void @combine_st1_casted_predicate(<vscale x 4 x i32> %vec, i32* %ptr) #0 {
 ; CHECK-LABEL: @combine_st1_casted_predicate(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[PTR:%.*]] to <vscale x 4 x i32>*
-; CHECK-NEXT:    store <vscale x 4 x i32> [[VEC:%.*]], <vscale x 4 x i32>* [[TMP1]], align 16
+; CHECK-NEXT:    store <vscale x 4 x i32> [[VEC:%.*]], <vscale x 4 x i32>* [[TMP1]], align 16, !annotation !0
 ; CHECK-NEXT:    ret void
 ;
   %1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
   %2 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %1)
   %3 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %2)
-  call void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32> %vec, <vscale x 4 x i1> %3, i32* %ptr)
+  call void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32> %vec, <vscale x 4 x i1> %3, i32* %ptr), !annotation !0
   ret void
 }
 
 define void @combine_st1_masked(<vscale x 4 x i32> %vec, i32* %ptr) #0 {
 ; CHECK-LABEL: @combine_st1_masked(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 16)
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i32* [[PTR:%.*]] to <vscale x 4 x i32>*
-; CHECK-NEXT:    call void @llvm.masked.store.nxv4i32.p0nxv4i32(<vscale x 4 x i32> [[VEC:%.*]], <vscale x 4 x i32>* [[TMP2]], i32 1, <vscale x 4 x i1> [[TMP1]])
+; CHECK-NEXT:    call void @llvm.masked.store.nxv4i32.p0nxv4i32(<vscale x 4 x i32> [[VEC:%.*]], <vscale x 4 x i32>* [[TMP2]], i32 1, <vscale x 4 x i1> [[TMP1]]), !annotation !0
 ; CHECK-NEXT:    ret void
 ;
   %1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 16)
-  call void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32> %vec, <vscale x 4 x i1> %1, i32* %ptr)
+  call void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32> %vec, <vscale x 4 x i1> %1, i32* %ptr), !annotation !0
   ret void
 }
 
 define void @combine_st1_masked_casted_predicate(<vscale x 8 x i16> %vec, i16* %ptr) #0 {
 ; CHECK-LABEL: @combine_st1_masked_casted_predicate(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
 ; CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[TMP1]])
 ; CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[TMP2]])
 ; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i16* [[PTR:%.*]] to <vscale x 8 x i16>*
-; CHECK-NEXT:    call void @llvm.masked.store.nxv8i16.p0nxv8i16(<vscale x 8 x i16> [[VEC:%.*]], <vscale x 8 x i16>* [[TMP4]], i32 1, <vscale x 8 x i1> [[TMP3]])
+; CHECK-NEXT:    call void @llvm.masked.store.nxv8i16.p0nxv8i16(<vscale x 8 x i16> [[VEC:%.*]], <vscale x 8 x i16>* [[TMP4]], i32 1, <vscale x 8 x i1> [[TMP3]]), !annotation !0
 ; CHECK-NEXT:    ret void
 ;
   %1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
   %2 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
   %3 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %2)
-  call void @llvm.aarch64.sve.st1.nxv8i16(<vscale x 8 x i16> %vec, <vscale x 8 x i1> %3, i16* %ptr)
+  call void @llvm.aarch64.sve.st1.nxv8i16(<vscale x 8 x i16> %vec, <vscale x 8 x i1> %3, i16* %ptr), !annotation !0
   ret void
 }
 
 declare <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1>)
 declare <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1>)
 
 declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1>)
 declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1>)
 
 declare <vscale x 4 x i32> @llvm.aarch64.sve.ld1.nxv4i32(<vscale x 4 x i1>, i32*)
 declare <vscale x 8 x i16> @llvm.aarch64.sve.ld1.nxv8i16(<vscale x 8 x i1>, i16*)
 
 declare <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32)
 declare <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32)
 declare <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32)
 
 declare void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, i32*)
 declare void @llvm.aarch64.sve.st1.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i1>, i16*)
 
 attributes #0 = { "target-features"="+sve" }
+
+!0 = !{!"some metadata"}