diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -512,14 +512,14 @@
 }
 
 // Load N-element structure into N vectors (scalar base)
-defm SVLD2 : StructLoad<"svld2[_{2}]", "2Pc", "aarch64_sve_ld2">;
-defm SVLD3 : StructLoad<"svld3[_{2}]", "3Pc", "aarch64_sve_ld3">;
-defm SVLD4 : StructLoad<"svld4[_{2}]", "4Pc", "aarch64_sve_ld4">;
+defm SVLD2 : StructLoad<"svld2[_{2}]", "2Pc", "aarch64_sve_ld2_sret">;
+defm SVLD3 : StructLoad<"svld3[_{2}]", "3Pc", "aarch64_sve_ld3_sret">;
+defm SVLD4 : StructLoad<"svld4[_{2}]", "4Pc", "aarch64_sve_ld4_sret">;
 
 // Load N-element structure into N vectors (scalar base, VL displacement)
-defm SVLD2_VNUM : StructLoad<"svld2_vnum[_{2}]", "2Pcl", "aarch64_sve_ld2">;
-defm SVLD3_VNUM : StructLoad<"svld3_vnum[_{2}]", "3Pcl", "aarch64_sve_ld3">;
-defm SVLD4_VNUM : StructLoad<"svld4_vnum[_{2}]", "4Pcl", "aarch64_sve_ld4">;
+defm SVLD2_VNUM : StructLoad<"svld2_vnum[_{2}]", "2Pcl", "aarch64_sve_ld2_sret">;
+defm SVLD3_VNUM : StructLoad<"svld3_vnum[_{2}]", "3Pcl", "aarch64_sve_ld3_sret">;
+defm SVLD4_VNUM : StructLoad<"svld4_vnum[_{2}]", "4Pcl", "aarch64_sve_ld4_sret">;
 
 // Load one octoword and replicate (scalar base)
 let ArchGuard = "defined(__ARM_FEATURE_SVE_MATMUL_FP64)" in {
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -8849,13 +8849,13 @@
 
   unsigned N;
   switch (IntID) {
-  case Intrinsic::aarch64_sve_ld2:
+  case Intrinsic::aarch64_sve_ld2_sret:
     N = 2;
     break;
-  case Intrinsic::aarch64_sve_ld3:
+  case Intrinsic::aarch64_sve_ld3_sret:
     N = 3;
     break;
-  case Intrinsic::aarch64_sve_ld4:
+  case Intrinsic::aarch64_sve_ld4_sret:
     N = 4;
     break;
   default:
@@ -8869,9 +8869,16 @@
   Value *Offset = Ops.size() > 2 ? Ops[2] : Builder.getInt32(0);
   BasePtr = Builder.CreateGEP(VTy, BasePtr, Offset);
   BasePtr = Builder.CreateBitCast(BasePtr, EltPtrTy);
-
-  Function *F = CGM.getIntrinsic(IntID, {RetTy, Predicate->getType()});
-  return Builder.CreateCall(F, { Predicate, BasePtr });
+  Function *F = CGM.getIntrinsic(IntID, {VTy});
+  Value *Call = Builder.CreateCall(F, {Predicate, BasePtr});
+  unsigned MinElts = VTy->getMinNumElements();
+  Value *Ret = llvm::PoisonValue::get(RetTy);
+  for (unsigned I = 0; I < N; I++) {
+    Value *Idx = ConstantInt::get(CGM.Int64Ty, I * MinElts);
+    Value *SRet = Builder.CreateExtractValue(Call, I);
+    Ret = Builder.CreateInsertVector(RetTy, Ret, SRet, Idx);
+  }
+  return Ret;
 }
 
 Value *CodeGenFunction::EmitSVEStructStore(const SVETypeFlags &TypeFlags,
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c
@@ -18,14 +18,22 @@
 // CHECK-LABEL: @test_svld2_bf16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x bfloat> @llvm.aarch64.sve.ld2.nxv16bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svld2_bf16u10__SVBool_tPKu6__bf16(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x bfloat> @llvm.aarch64.sve.ld2.nxv16bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP5]]
 //
 svbfloat16x2_t test_svld2_bf16(svbool_t pg, const bfloat16_t *base)
 {
@@ -38,16 +46,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast bfloat* [[BASE:%.*]] to <vscale x 8 x bfloat>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x bfloat> @llvm.aarch64.sve.ld2.nxv16bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svld2_vnum_bf16u10__SVBool_tPKu6__bf16l(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast bfloat* [[BASE:%.*]] to <vscale x 8 x bfloat>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x bfloat> @llvm.aarch64.sve.ld2.nxv16bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP7]]
 //
 svbfloat16x2_t test_svld2_vnum_bf16(svbool_t pg, const bfloat16_t *base, int64_t vnum)
 {
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c
@@ -17,13 +17,21 @@
 
 // CHECK-LABEL: @test_svld2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld2_s8u10__SVBool_tPKa(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
 //
 svint8x2_t test_svld2_s8(svbool_t pg, const int8_t *base)
 {
@@ -33,14 +41,22 @@
 // CHECK-LABEL: @test_svld2_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_s16u10__SVBool_tPKs(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
 //
 svint16x2_t test_svld2_s16(svbool_t pg, const int16_t *base)
 {
@@ -50,14 +66,22 @@
 // CHECK-LABEL: @test_svld2_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_s32u10__SVBool_tPKi(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
 //
 svint32x2_t test_svld2_s32(svbool_t pg, const int32_t *base)
 {
@@ -67,14 +91,22 @@
 // CHECK-LABEL: @test_svld2_s64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_s64u10__SVBool_tPKl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
 //
 svint64x2_t test_svld2_s64(svbool_t pg, const int64_t *base)
 {
@@ -83,13 +115,21 @@
 
 // CHECK-LABEL: @test_svld2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld2_u8u10__SVBool_tPKh(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
 //
 svuint8x2_t test_svld2_u8(svbool_t pg, const uint8_t *base)
 {
@@ -99,14 +139,22 @@
 // CHECK-LABEL: @test_svld2_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_u16u10__SVBool_tPKt(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
 //
 svuint16x2_t test_svld2_u16(svbool_t pg, const uint16_t *base)
 {
@@ -116,14 +164,22 @@
 // CHECK-LABEL: @test_svld2_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_u32u10__SVBool_tPKj(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
 //
 svuint32x2_t test_svld2_u32(svbool_t pg, const uint32_t *base)
 {
@@ -133,14 +189,22 @@
 // CHECK-LABEL: @test_svld2_u64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_u64u10__SVBool_tPKm(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
 //
 svuint64x2_t test_svld2_u64(svbool_t pg, const uint64_t *base)
 {
@@ -150,14 +214,22 @@
 // CHECK-LABEL: @test_svld2_f16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x half> @llvm.aarch64.sve.ld2.nxv16f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_f16u10__SVBool_tPKDh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x half> @llvm.aarch64.sve.ld2.nxv16f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP5]]
 //
 svfloat16x2_t test_svld2_f16(svbool_t pg, const float16_t *base)
 {
@@ -167,14 +239,22 @@
 // CHECK-LABEL: @test_svld2_f32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x float> @llvm.aarch64.sve.ld2.nxv8f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
+// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_f32u10__SVBool_tPKf(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x float> @llvm.aarch64.sve.ld2.nxv8f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP5]]
 //
 svfloat32x2_t test_svld2_f32(svbool_t pg, const float32_t *base)
 {
@@ -184,14 +264,22 @@
 // CHECK-LABEL: @test_svld2_f64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x double> @llvm.aarch64.sve.ld2.nxv4f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
+// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_f64u10__SVBool_tPKd(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x double> @llvm.aarch64.sve.ld2.nxv4f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP5]]
 //
 svfloat64x2_t test_svld2_f64(svbool_t pg, const float64_t *base)
 {
@@ -202,15 +290,23 @@
 // 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:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld2_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:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
 //
 svint8x2_t test_svld2_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum)
 {
@@ -222,16 +318,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_s16u10__SVBool_tPKsl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP7]]
 //
 svint16x2_t test_svld2_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum)
 {
@@ -243,16 +347,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_s32u10__SVBool_tPKil(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP7]]
 //
 svint32x2_t test_svld2_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum)
 {
@@ -264,16 +376,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_s64u10__SVBool_tPKll(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP7]]
 //
 svint64x2_t test_svld2_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum)
 {
@@ -284,15 +404,23 @@
 // 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:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld2_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:%.*]] = tail call <vscale x 32 x i8> @llvm.aarch64.sve.ld2.nxv32i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
 //
 svuint8x2_t test_svld2_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum)
 {
@@ -304,16 +432,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_u16u10__SVBool_tPKtl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i16> @llvm.aarch64.sve.ld2.nxv16i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP7]]
 //
 svuint16x2_t test_svld2_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum)
 {
@@ -325,16 +461,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_u32u10__SVBool_tPKjl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i32> @llvm.aarch64.sve.ld2.nxv8i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP7]]
 //
 svuint32x2_t test_svld2_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum)
 {
@@ -346,16 +490,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_u64u10__SVBool_tPKml(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 4 x i64> @llvm.aarch64.sve.ld2.nxv4i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP7]]
 //
 svuint64x2_t test_svld2_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum)
 {
@@ -367,16 +519,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x half> @llvm.aarch64.sve.ld2.nxv16f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_f16u10__SVBool_tPKDhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x half> @llvm.aarch64.sve.ld2.nxv16f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP7]]
 //
 svfloat16x2_t test_svld2_vnum_f16(svbool_t pg, const float16_t *base, int64_t vnum)
 {
@@ -388,16 +548,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x float> @llvm.aarch64.sve.ld2.nxv8f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 4)
+// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_f32u10__SVBool_tPKfl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x float> @llvm.aarch64.sve.ld2.nxv8f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP7]]
 //
 svfloat32x2_t test_svld2_vnum_f32(svbool_t pg, const float32_t *base, int64_t vnum)
 {
@@ -409,16 +577,24 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 4 x double> @llvm.aarch64.sve.ld2.nxv4f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 2)
+// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_f64u10__SVBool_tPKdl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 4 x double> @llvm.aarch64.sve.ld2.nxv4f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP7]]
 //
 svfloat64x2_t test_svld2_vnum_f64(svbool_t pg, const float64_t *base, int64_t vnum)
 {
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c
@@ -18,14 +18,26 @@
 // CHECK-LABEL: @test_svld3_bf16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x bfloat> @llvm.aarch64.sve.ld3.nxv24bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svld3_bf16u10__SVBool_tPKu6__bf16(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x bfloat> @llvm.aarch64.sve.ld3.nxv24bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP7]]
 //
 svbfloat16x3_t test_svld3_bf16(svbool_t pg, const bfloat16_t *base)
 {
@@ -37,16 +49,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast bfloat* [[BASE:%.*]] to <vscale x 8 x bfloat>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x bfloat> @llvm.aarch64.sve.ld3.nxv24bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svld3_vnum_bf16u10__SVBool_tPKu6__bf16l(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast bfloat* [[BASE:%.*]] to <vscale x 8 x bfloat>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x bfloat> @llvm.aarch64.sve.ld3.nxv24bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP9]]
 //
 svbfloat16x3_t test_svld3_vnum_bf16(svbool_t pg, const bfloat16_t *base, int64_t vnum)
 {
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c
@@ -17,13 +17,25 @@
 
 // CHECK-LABEL: @test_svld3_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld3_s8u10__SVBool_tPKa(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
 //
 svint8x3_t test_svld3_s8(svbool_t pg, const int8_t *base)
 {
@@ -33,14 +45,26 @@
 // CHECK-LABEL: @test_svld3_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_s16u10__SVBool_tPKs(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
 //
 svint16x3_t test_svld3_s16(svbool_t pg, const int16_t *base)
 {
@@ -50,14 +74,26 @@
 // CHECK-LABEL: @test_svld3_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_s32u10__SVBool_tPKi(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
 //
 svint32x3_t test_svld3_s32(svbool_t pg, const int32_t *base)
 {
@@ -67,14 +103,26 @@
 // CHECK-LABEL: @test_svld3_s64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_s64u10__SVBool_tPKl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
 //
 svint64x3_t test_svld3_s64(svbool_t pg, const int64_t *base)
 {
@@ -83,13 +131,25 @@
 
 // CHECK-LABEL: @test_svld3_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld3_u8u10__SVBool_tPKh(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
 //
 svuint8x3_t test_svld3_u8(svbool_t pg, const uint8_t *base)
 {
@@ -99,14 +159,26 @@
 // CHECK-LABEL: @test_svld3_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_u16u10__SVBool_tPKt(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
 //
 svuint16x3_t test_svld3_u16(svbool_t pg, const uint16_t *base)
 {
@@ -116,14 +188,26 @@
 // CHECK-LABEL: @test_svld3_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_u32u10__SVBool_tPKj(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
 //
 svuint32x3_t test_svld3_u32(svbool_t pg, const uint32_t *base)
 {
@@ -133,14 +217,26 @@
 // CHECK-LABEL: @test_svld3_u64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_u64u10__SVBool_tPKm(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
 //
 svuint64x3_t test_svld3_u64(svbool_t pg, const uint64_t *base)
 {
@@ -150,14 +246,26 @@
 // CHECK-LABEL: @test_svld3_f16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x half> @llvm.aarch64.sve.ld3.nxv24f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 24 x half> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x half> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_f16u10__SVBool_tPKDh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x half> @llvm.aarch64.sve.ld3.nxv24f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x half> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x half> [[TMP7]]
 //
 svfloat16x3_t test_svld3_f16(svbool_t pg, const float16_t *base)
 {
@@ -167,14 +275,26 @@
 // CHECK-LABEL: @test_svld3_f32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x float> @llvm.aarch64.sve.ld3.nxv12f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 12 x float> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 8)
+// CHECK-NEXT:    ret <vscale x 12 x float> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_f32u10__SVBool_tPKf(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x float> @llvm.aarch64.sve.ld3.nxv12f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 12 x float> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 12 x float> [[TMP7]]
 //
 svfloat32x3_t test_svld3_f32(svbool_t pg, const float32_t *base)
 {
@@ -184,14 +304,26 @@
 // CHECK-LABEL: @test_svld3_f64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x double> @llvm.aarch64.sve.ld3.nxv6f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 6 x double> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 4)
+// CHECK-NEXT:    ret <vscale x 6 x double> [[TMP7]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_f64u10__SVBool_tPKd(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x double> @llvm.aarch64.sve.ld3.nxv6f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 6 x double> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 6 x double> [[TMP7]]
 //
 svfloat64x3_t test_svld3_f64(svbool_t pg, const float64_t *base)
 {
@@ -202,15 +334,27 @@
 // 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:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld3_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:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP8]]
 //
 svint8x3_t test_svld3_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum)
 {
@@ -222,16 +366,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_s16u10__SVBool_tPKsl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP9]]
 //
 svint16x3_t test_svld3_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum)
 {
@@ -243,16 +399,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_s32u10__SVBool_tPKil(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP9]]
 //
 svint32x3_t test_svld3_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum)
 {
@@ -264,16 +432,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_s64u10__SVBool_tPKll(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP9]]
 //
 svint64x3_t test_svld3_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum)
 {
@@ -284,15 +464,27 @@
 // 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:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld3_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:%.*]] = tail call <vscale x 48 x i8> @llvm.aarch64.sve.ld3.nxv48i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP8]]
 //
 svuint8x3_t test_svld3_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum)
 {
@@ -304,16 +496,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_u16u10__SVBool_tPKtl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 24 x i16> @llvm.aarch64.sve.ld3.nxv24i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP9]]
 //
 svuint16x3_t test_svld3_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum)
 {
@@ -325,16 +529,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_u32u10__SVBool_tPKjl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 12 x i32> @llvm.aarch64.sve.ld3.nxv12i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP9]]
 //
 svuint32x3_t test_svld3_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum)
 {
@@ -346,16 +562,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_u64u10__SVBool_tPKml(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 6 x i64> @llvm.aarch64.sve.ld3.nxv6i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP9]]
 //
 svuint64x3_t test_svld3_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum)
 {
@@ -367,16 +595,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 24 x half> @llvm.aarch64.sve.ld3.nxv24f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 24 x half> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP7]], <vscale x 8 x half> [[TMP8]], i64 16)
+// CHECK-NEXT:    ret <vscale x 24 x half> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_f16u10__SVBool_tPKDhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 24 x half> @llvm.aarch64.sve.ld3.nxv24f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 24 x half> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP7]], <vscale x 8 x half> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 24 x half> [[TMP9]]
 //
 svfloat16x3_t test_svld3_vnum_f16(svbool_t pg, const float16_t *base, int64_t vnum)
 {
@@ -388,16 +628,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 12 x float> @llvm.aarch64.sve.ld3.nxv12f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 12 x float> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP7]], <vscale x 4 x float> [[TMP8]], i64 8)
+// CHECK-NEXT:    ret <vscale x 12 x float> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_f32u10__SVBool_tPKfl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 12 x float> @llvm.aarch64.sve.ld3.nxv12f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 12 x float> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP7]], <vscale x 4 x float> [[TMP8]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 12 x float> [[TMP9]]
 //
 svfloat32x3_t test_svld3_vnum_f32(svbool_t pg, const float32_t *base, int64_t vnum)
 {
@@ -409,16 +661,28 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 6 x double> @llvm.aarch64.sve.ld3.nxv6f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 6 x double> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 2)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP7]], <vscale x 2 x double> [[TMP8]], i64 4)
+// CHECK-NEXT:    ret <vscale x 6 x double> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_f64u10__SVBool_tPKdl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 6 x double> @llvm.aarch64.sve.ld3.nxv6f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 6 x double> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP7]], <vscale x 2 x double> [[TMP8]], i64 4)
+// CPP-CHECK-NEXT:    ret <vscale x 6 x double> [[TMP9]]
 //
 svfloat64x3_t test_svld3_vnum_f64(svbool_t pg, const float64_t *base, int64_t vnum)
 {
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c
@@ -18,14 +18,30 @@
 // CHECK-LABEL: @test_svld4_bf16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x bfloat> @llvm.aarch64.sve.ld4.nxv32bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svld4_bf16u10__SVBool_tPKu6__bf16(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x bfloat> @llvm.aarch64.sve.ld4.nxv32bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP9]]
 //
 svbfloat16x4_t test_svld4_bf16(svbool_t pg, const bfloat16_t *base)
 {
@@ -37,16 +53,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast bfloat* [[BASE:%.*]] to <vscale x 8 x bfloat>*
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.aarch64.sve.ld4.nxv32bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 16)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP9]], <vscale x 8 x bfloat> [[TMP10]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svld4_vnum_bf16u10__SVBool_tPKu6__bf16l(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = bitcast bfloat* [[BASE:%.*]] to <vscale x 8 x bfloat>*
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP1]], i64 [[VNUM:%.*]], i64 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.aarch64.sve.ld4.nxv32bf16.nxv8i1(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], bfloat* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP9]], <vscale x 8 x bfloat> [[TMP10]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP11]]
 //
 svbfloat16x4_t test_svld4_vnum_bf16(svbool_t pg, const bfloat16_t *base, int64_t vnum)
 {
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c
@@ -17,13 +17,29 @@
 
 // CHECK-LABEL: @test_svld4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld4_s8u10__SVBool_tPKa(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
 //
 svint8x4_t test_svld4_s8(svbool_t pg, const int8_t *base)
 {
@@ -33,14 +49,30 @@
 // CHECK-LABEL: @test_svld4_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_s16u10__SVBool_tPKs(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP9]]
 //
 svint16x4_t test_svld4_s16(svbool_t pg, const int16_t *base)
 {
@@ -50,14 +82,30 @@
 // CHECK-LABEL: @test_svld4_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 12)
+// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_s32u10__SVBool_tPKi(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 12)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP9]]
 //
 svint32x4_t test_svld4_s32(svbool_t pg, const int32_t *base)
 {
@@ -67,14 +115,30 @@
 // CHECK-LABEL: @test_svld4_s64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 6)
+// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_s64u10__SVBool_tPKl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 6)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP9]]
 //
 svint64x4_t test_svld4_s64(svbool_t pg, const int64_t *base)
 {
@@ -83,13 +147,29 @@
 
 // CHECK-LABEL: @test_svld4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld4_u8u10__SVBool_tPKh(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
 //
 svuint8x4_t test_svld4_u8(svbool_t pg, const uint8_t *base)
 {
@@ -99,14 +179,30 @@
 // CHECK-LABEL: @test_svld4_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_u16u10__SVBool_tPKt(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP9]]
 //
 svuint16x4_t test_svld4_u16(svbool_t pg, const uint16_t *base)
 {
@@ -116,14 +212,30 @@
 // CHECK-LABEL: @test_svld4_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 12)
+// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_u32u10__SVBool_tPKj(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 12)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP9]]
 //
 svuint32x4_t test_svld4_u32(svbool_t pg, const uint32_t *base)
 {
@@ -133,14 +245,30 @@
 // CHECK-LABEL: @test_svld4_u64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 6)
+// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_u64u10__SVBool_tPKm(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 6)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP9]]
 //
 svuint64x4_t test_svld4_u64(svbool_t pg, const uint64_t *base)
 {
@@ -150,14 +278,30 @@
 // CHECK-LABEL: @test_svld4_f16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x half> @llvm.aarch64.sve.ld4.nxv32f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld4.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 16)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP7]], <vscale x 8 x half> [[TMP8]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_f16u10__SVBool_tPKDh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x half> @llvm.aarch64.sve.ld4.nxv32f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld4.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP7]], <vscale x 8 x half> [[TMP8]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP9]]
 //
 svfloat16x4_t test_svld4_f16(svbool_t pg, const float16_t *base)
 {
@@ -167,14 +311,30 @@
 // CHECK-LABEL: @test_svld4_f32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x float> @llvm.aarch64.sve.ld4.nxv16f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld4.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP7]], <vscale x 4 x float> [[TMP8]], i64 12)
+// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_f32u10__SVBool_tPKf(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x float> @llvm.aarch64.sve.ld4.nxv16f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld4.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP7]], <vscale x 4 x float> [[TMP8]], i64 12)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP9]]
 //
 svfloat32x4_t test_svld4_f32(svbool_t pg, const float32_t *base)
 {
@@ -184,14 +344,30 @@
 // CHECK-LABEL: @test_svld4_f64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x double> @llvm.aarch64.sve.ld4.nxv8f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP1]]
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld4.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 3
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP7]], <vscale x 2 x double> [[TMP8]], i64 6)
+// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_f64u10__SVBool_tPKd(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x double> @llvm.aarch64.sve.ld4.nxv8f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld4.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[BASE:%.*]])
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 3
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP7]], <vscale x 2 x double> [[TMP8]], i64 6)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP9]]
 //
 svfloat64x4_t test_svld4_f64(svbool_t pg, const float64_t *base)
 {
@@ -202,15 +378,31 @@
 // 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:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 3
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 48)
+// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP10]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld4_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:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 3
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 48)
+// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP10]]
 //
 svint8x4_t test_svld4_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum)
 {
@@ -222,16 +414,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP9]], <vscale x 8 x i16> [[TMP10]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_s16u10__SVBool_tPKsl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP9]], <vscale x 8 x i16> [[TMP10]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP11]]
 //
 svint16x4_t test_svld4_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum)
 {
@@ -243,16 +451,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP9]], <vscale x 4 x i32> [[TMP10]], i64 12)
+// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_s32u10__SVBool_tPKil(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP9]], <vscale x 4 x i32> [[TMP10]], i64 12)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP11]]
 //
 svint32x4_t test_svld4_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum)
 {
@@ -264,16 +488,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP9]], <vscale x 2 x i64> [[TMP10]], i64 6)
+// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_s64u10__SVBool_tPKll(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP9]], <vscale x 2 x i64> [[TMP10]], i64 6)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP11]]
 //
 svint64x4_t test_svld4_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum)
 {
@@ -284,15 +524,31 @@
 // 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:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 3
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 48)
+// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP10]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld4_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:%.*]] = tail call <vscale x 64 x i8> @llvm.aarch64.sve.ld4.nxv64i8.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], i8* [[TMP1]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 3
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 48)
+// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP10]]
 //
 svuint8x4_t test_svld4_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum)
 {
@@ -304,16 +560,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP9]], <vscale x 8 x i16> [[TMP10]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_u16u10__SVBool_tPKtl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 32 x i16> @llvm.aarch64.sve.ld4.nxv32i16.nxv8i1(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], i16* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP9]], <vscale x 8 x i16> [[TMP10]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP11]]
 //
 svuint16x4_t test_svld4_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum)
 {
@@ -325,16 +597,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP9]], <vscale x 4 x i32> [[TMP10]], i64 12)
+// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_u32u10__SVBool_tPKjl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x i32> @llvm.aarch64.sve.ld4.nxv16i32.nxv4i1(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], i32* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP8]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP9]], <vscale x 4 x i32> [[TMP10]], i64 12)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP11]]
 //
 svuint32x4_t test_svld4_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum)
 {
@@ -346,16 +634,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP9]], <vscale x 2 x i64> [[TMP10]], i64 6)
+// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_u64u10__SVBool_tPKml(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x i64> @llvm.aarch64.sve.ld4.nxv8i64.nxv2i1(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], i64* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP9]], <vscale x 2 x i64> [[TMP10]], i64 6)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP11]]
 //
 svuint64x4_t test_svld4_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum)
 {
@@ -367,16 +671,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 32 x half> @llvm.aarch64.sve.ld4.nxv32f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld4.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 8)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP7]], <vscale x 8 x half> [[TMP8]], i64 16)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP9]], <vscale x 8 x half> [[TMP10]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_f16u10__SVBool_tPKDhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 32 x half> @llvm.aarch64.sve.ld4.nxv32f16.nxv8i1(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld4.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], half* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP7]], <vscale x 8 x half> [[TMP8]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP9]], <vscale x 8 x half> [[TMP10]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP11]]
 //
 svfloat16x4_t test_svld4_vnum_f16(svbool_t pg, const float16_t *base, int64_t vnum)
 {
@@ -388,16 +708,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x float> @llvm.aarch64.sve.ld4.nxv16f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld4.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 4)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP7]], <vscale x 4 x float> [[TMP8]], i64 8)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP9]], <vscale x 4 x float> [[TMP10]], i64 12)
+// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_f32u10__SVBool_tPKfl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 16 x float> @llvm.aarch64.sve.ld4.nxv16f32.nxv4i1(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld4.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], float* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP7]], <vscale x 4 x float> [[TMP8]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP9]], <vscale x 4 x float> [[TMP10]], i64 12)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP11]]
 //
 svfloat32x4_t test_svld4_vnum_f32(svbool_t pg, const float32_t *base, int64_t vnum)
 {
@@ -409,16 +745,32 @@
 // CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x double> @llvm.aarch64.sve.ld4.nxv8f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP3]]
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld4.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP4]], i64 0)
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 2)
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP7]], <vscale x 2 x double> [[TMP8]], i64 4)
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP9]], <vscale x 2 x double> [[TMP10]], i64 6)
+// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP11]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld4_vnum_f64u10__SVBool_tPKdl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail 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:%.*]] = tail call <vscale x 8 x double> @llvm.aarch64.sve.ld4.nxv8f64.nxv2i1(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld4.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], double* [[TMP2]])
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP4]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP7]], <vscale x 2 x double> [[TMP8]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP9]], <vscale x 2 x double> [[TMP10]], i64 6)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP11]]
 //
 svfloat64x4_t test_svld4_vnum_f64(svbool_t pg, const float64_t *base, int64_t vnum)
 {