[PATCH] D92761: [clang][AArch64][SVE] Avoid going through memory for VLAT <-> VLST casts

Joe Ellis via Phabricator via cfe-commits Wed, 09 Dec 2020 07:23:48 -0800

joechrisellis updated this revision to Diff 310529.
joechrisellis added a comment.
Herald added a subscriber: NickHung.


Rebase.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D92761/new/

https://reviews.llvm.org/D92761

Files:
  clang/lib/CodeGen/CGExprScalar.cpp
  clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
  clang/test/CodeGen/attr-arm-sve-vector-bits-bitcast.c
  clang/test/CodeGen/attr-arm-sve-vector-bits-call.c
  clang/test/CodeGen/attr-arm-sve-vector-bits-cast.c
  clang/test/CodeGen/attr-arm-sve-vector-bits-codegen.c
  clang/test/CodeGen/attr-arm-sve-vector-bits-globals.c
  llvm/include/llvm/IR/IRBuilder.h

Index: llvm/include/llvm/IR/IRBuilder.h
===================================================================
--- llvm/include/llvm/IR/IRBuilder.h
+++ llvm/include/llvm/IR/IRBuilder.h
@@ -923,6 +923,20 @@
     return CreateBinaryIntrinsic(Intrinsic::maximum, LHS, RHS, nullptr, Name);
   }
 
+  CallInst *CreateExtractVector(Type *DstType, Value *SrcVec, Value *Idx,
+                                const Twine &Name = "") {
+    return CreateIntrinsic(Intrinsic::experimental_vector_extract,
+                           {DstType, SrcVec->getType()}, {SrcVec, Idx}, nullptr,
+                           Name);
+  }
+
+  CallInst *CreateInsertVector(Type *DstType, Value *SrcVec, Value *SubVec,
+                               Value *Idx, const Twine &Name = "") {
+    return CreateIntrinsic(Intrinsic::experimental_vector_insert,
+                           {DstType, SubVec->getType()}, {SrcVec, SubVec, Idx},
+                           nullptr, Name);
+  }
+
 private:
   /// Create a call to a masked intrinsic with given Id.
   CallInst *CreateMaskedIntrinsic(Intrinsic::ID Id, ArrayRef<Value *> Ops,
Index: clang/test/CodeGen/attr-arm-sve-vector-bits-globals.c
===================================================================
--- clang/test/CodeGen/attr-arm-sve-vector-bits-globals.c
+++ clang/test/CodeGen/attr-arm-sve-vector-bits-globals.c
@@ -21,40 +21,28 @@
 
 // CHECK-128-LABEL: @write_global_i64(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[V_ADDR:%.*]] = alloca <vscale x 2 x i64>, align 16
-// CHECK-128-NEXT:    store <vscale x 2 x i64> [[V:%.*]], <vscale x 2 x i64>* [[V_ADDR]], align 16, [[TBAA6:!tbaa !.*]]
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x i64>* [[V_ADDR]] to <2 x i64>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 16, [[TBAA10:!tbaa !.*]]
-// CHECK-128-NEXT:    store <2 x i64> [[TMP1]], <2 x i64>* @global_i64, align 16, [[TBAA10]]
+// CHECK-128-NEXT:    [[CASTFIXEDSVE:%.*]] = call <2 x i64> @llvm.experimental.vector.extract.v2i64.nxv2i64(<vscale x 2 x i64> [[V:%.*]], i64 0)
+// CHECK-128-NEXT:    store <2 x i64> [[CASTFIXEDSVE]], <2 x i64>* @global_i64, align 16, [[TBAA6:!tbaa !.*]]
 // CHECK-128-NEXT:    ret void
 //
 // CHECK-512-LABEL: @write_global_i64(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[V_ADDR:%.*]] = alloca <vscale x 2 x i64>, align 16
-// CHECK-512-NEXT:    store <vscale x 2 x i64> [[V:%.*]], <vscale x 2 x i64>* [[V_ADDR]], align 16, [[TBAA6:!tbaa !.*]]
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x i64>* [[V_ADDR]] to <8 x i64>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <8 x i64>, <8 x i64>* [[TMP0]], align 16, [[TBAA10:!tbaa !.*]]
-// CHECK-512-NEXT:    store <8 x i64> [[TMP1]], <8 x i64>* @global_i64, align 16, [[TBAA10]]
+// CHECK-512-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x i64> @llvm.experimental.vector.extract.v8i64.nxv2i64(<vscale x 2 x i64> [[V:%.*]], i64 0)
+// CHECK-512-NEXT:    store <8 x i64> [[CASTFIXEDSVE]], <8 x i64>* @global_i64, align 16, [[TBAA6:!tbaa !.*]]
 // CHECK-512-NEXT:    ret void
 //
 void write_global_i64(svint64_t v) { global_i64 = v; }
 
 // CHECK-128-LABEL: @write_global_bf16(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[V_ADDR:%.*]] = alloca <vscale x 8 x bfloat>, align 16
-// CHECK-128-NEXT:    store <vscale x 8 x bfloat> [[V:%.*]], <vscale x 8 x bfloat>* [[V_ADDR]], align 16, [[TBAA11:!tbaa !.*]]
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 8 x bfloat>* [[V_ADDR]] to <8 x bfloat>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <8 x bfloat>, <8 x bfloat>* [[TMP0]], align 16, [[TBAA10]]
-// CHECK-128-NEXT:    store <8 x bfloat> [[TMP1]], <8 x bfloat>* @global_bf16, align 16, [[TBAA10]]
+// CHECK-128-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x bfloat> @llvm.experimental.vector.extract.v8bf16.nxv8bf16(<vscale x 8 x bfloat> [[V:%.*]], i64 0)
+// CHECK-128-NEXT:    store <8 x bfloat> [[CASTFIXEDSVE]], <8 x bfloat>* @global_bf16, align 16, [[TBAA6]]
 // CHECK-128-NEXT:    ret void
 //
 // CHECK-512-LABEL: @write_global_bf16(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[V_ADDR:%.*]] = alloca <vscale x 8 x bfloat>, align 16
-// CHECK-512-NEXT:    store <vscale x 8 x bfloat> [[V:%.*]], <vscale x 8 x bfloat>* [[V_ADDR]], align 16, [[TBAA11:!tbaa !.*]]
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 8 x bfloat>* [[V_ADDR]] to <32 x bfloat>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <32 x bfloat>, <32 x bfloat>* [[TMP0]], align 16, [[TBAA10]]
-// CHECK-512-NEXT:    store <32 x bfloat> [[TMP1]], <32 x bfloat>* @global_bf16, align 16, [[TBAA10]]
+// CHECK-512-NEXT:    [[CASTFIXEDSVE:%.*]] = call <32 x bfloat> @llvm.experimental.vector.extract.v32bf16.nxv8bf16(<vscale x 8 x bfloat> [[V:%.*]], i64 0)
+// CHECK-512-NEXT:    store <32 x bfloat> [[CASTFIXEDSVE]], <32 x bfloat>* @global_bf16, align 16, [[TBAA6]]
 // CHECK-512-NEXT:    ret void
 //
 void write_global_bf16(svbfloat16_t v) { global_bf16 = v; }
@@ -62,19 +50,19 @@
 // CHECK-128-LABEL: @write_global_bool(
 // CHECK-128-NEXT:  entry:
 // CHECK-128-NEXT:    [[V_ADDR:%.*]] = alloca <vscale x 16 x i1>, align 16
-// CHECK-128-NEXT:    store <vscale x 16 x i1> [[V:%.*]], <vscale x 16 x i1>* [[V_ADDR]], align 16, [[TBAA13:!tbaa !.*]]
+// CHECK-128-NEXT:    store <vscale x 16 x i1> [[V:%.*]], <vscale x 16 x i1>* [[V_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
 // CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 16 x i1>* [[V_ADDR]] to <2 x i8>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* [[TMP0]], align 16, [[TBAA10]]
-// CHECK-128-NEXT:    store <2 x i8> [[TMP1]], <2 x i8>* @global_bool, align 2, [[TBAA10]]
+// CHECK-128-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    store <2 x i8> [[TMP1]], <2 x i8>* @global_bool, align 2, [[TBAA6]]
 // CHECK-128-NEXT:    ret void
 //
 // CHECK-512-LABEL: @write_global_bool(
 // CHECK-512-NEXT:  entry:
 // CHECK-512-NEXT:    [[V_ADDR:%.*]] = alloca <vscale x 16 x i1>, align 16
-// CHECK-512-NEXT:    store <vscale x 16 x i1> [[V:%.*]], <vscale x 16 x i1>* [[V_ADDR]], align 16, [[TBAA13:!tbaa !.*]]
+// CHECK-512-NEXT:    store <vscale x 16 x i1> [[V:%.*]], <vscale x 16 x i1>* [[V_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
 // CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 16 x i1>* [[V_ADDR]] to <8 x i8>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 16, [[TBAA10]]
-// CHECK-512-NEXT:    store <8 x i8> [[TMP1]], <8 x i8>* @global_bool, align 2, [[TBAA10]]
+// CHECK-512-NEXT:    [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    store <8 x i8> [[TMP1]], <8 x i8>* @global_bool, align 2, [[TBAA6]]
 // CHECK-512-NEXT:    ret void
 //
 void write_global_bool(svbool_t v) { global_bool = v; }
@@ -85,36 +73,40 @@
 
 // CHECK-128-LABEL: @read_global_i64(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[TMP0:%.*]] = load <vscale x 2 x i64>, <vscale x 2 x i64>* bitcast (<2 x i64>* @global_i64 to <vscale x 2 x i64>*), align 16, [[TBAA10]]
-// CHECK-128-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-128-NEXT:    [[TMP0:%.*]] = load <2 x i64>, <2 x i64>* @global_i64, align 16, [[TBAA6]]
+// CHECK-128-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x i64> @llvm.experimental.vector.insert.nxv2i64.v2i64(<vscale x 2 x i64> undef, <2 x i64> [[TMP0]], i64 0)
+// CHECK-128-NEXT:    ret <vscale x 2 x i64> [[CASTSCALABLESVE]]
 //
 // CHECK-512-LABEL: @read_global_i64(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[TMP0:%.*]] = load <vscale x 2 x i64>, <vscale x 2 x i64>* bitcast (<8 x i64>* @global_i64 to <vscale x 2 x i64>*), align 16, [[TBAA10]]
-// CHECK-512-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-512-NEXT:    [[TMP0:%.*]] = load <8 x i64>, <8 x i64>* @global_i64, align 16, [[TBAA6]]
+// CHECK-512-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x i64> @llvm.experimental.vector.insert.nxv2i64.v8i64(<vscale x 2 x i64> undef, <8 x i64> [[TMP0]], i64 0)
+// CHECK-512-NEXT:    ret <vscale x 2 x i64> [[CASTSCALABLESVE]]
 //
 svint64_t read_global_i64() { return global_i64; }
 
 // CHECK-128-LABEL: @read_global_bf16(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[TMP0:%.*]] = load <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* bitcast (<8 x bfloat>* @global_bf16 to <vscale x 8 x bfloat>*), align 16, [[TBAA10]]
-// CHECK-128-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-128-NEXT:    [[TMP0:%.*]] = load <8 x bfloat>, <8 x bfloat>* @global_bf16, align 16, [[TBAA6]]
+// CHECK-128-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 8 x bfloat> @llvm.experimental.vector.insert.nxv8bf16.v8bf16(<vscale x 8 x bfloat> undef, <8 x bfloat> [[TMP0]], i64 0)
+// CHECK-128-NEXT:    ret <vscale x 8 x bfloat> [[CASTSCALABLESVE]]
 //
 // CHECK-512-LABEL: @read_global_bf16(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[TMP0:%.*]] = load <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* bitcast (<32 x bfloat>* @global_bf16 to <vscale x 8 x bfloat>*), align 16, [[TBAA10]]
-// CHECK-512-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-512-NEXT:    [[TMP0:%.*]] = load <32 x bfloat>, <32 x bfloat>* @global_bf16, align 16, [[TBAA6]]
+// CHECK-512-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 8 x bfloat> @llvm.experimental.vector.insert.nxv8bf16.v32bf16(<vscale x 8 x bfloat> undef, <32 x bfloat> [[TMP0]], i64 0)
+// CHECK-512-NEXT:    ret <vscale x 8 x bfloat> [[CASTSCALABLESVE]]
 //
 svbfloat16_t read_global_bf16() { return global_bf16; }
 
 // CHECK-128-LABEL: @read_global_bool(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[TMP0:%.*]] = load <vscale x 16 x i1>, <vscale x 16 x i1>* bitcast (<2 x i8>* @global_bool to <vscale x 16 x i1>*), align 2, [[TBAA10]]
+// CHECK-128-NEXT:    [[TMP0:%.*]] = load <vscale x 16 x i1>, <vscale x 16 x i1>* bitcast (<2 x i8>* @global_bool to <vscale x 16 x i1>*), align 2, [[TBAA6]]
 // CHECK-128-NEXT:    ret <vscale x 16 x i1> [[TMP0]]
 //
 // CHECK-512-LABEL: @read_global_bool(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[TMP0:%.*]] = load <vscale x 16 x i1>, <vscale x 16 x i1>* bitcast (<8 x i8>* @global_bool to <vscale x 16 x i1>*), align 2, [[TBAA10]]
+// CHECK-512-NEXT:    [[TMP0:%.*]] = load <vscale x 16 x i1>, <vscale x 16 x i1>* bitcast (<8 x i8>* @global_bool to <vscale x 16 x i1>*), align 2, [[TBAA6]]
 // CHECK-512-NEXT:    ret <vscale x 16 x i1> [[TMP0]]
 //
 svbool_t read_global_bool() { return global_bool; }
Index: clang/test/CodeGen/attr-arm-sve-vector-bits-codegen.c
===================================================================
--- clang/test/CodeGen/attr-arm-sve-vector-bits-codegen.c
+++ clang/test/CodeGen/attr-arm-sve-vector-bits-codegen.c
@@ -17,7 +17,6 @@
 // CHECK-NEXT:    [[PRED_ADDR:%.*]] = alloca <vscale x 16 x i1>, align 2
 // CHECK-NEXT:    [[VEC_ADDR:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[PG:%.*]] = alloca <vscale x 16 x i1>, align 2
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    store <vscale x 16 x i1> [[PRED:%.*]], <vscale x 16 x i1>* [[PRED_ADDR]], align 2
 // CHECK-NEXT:    store <vscale x 4 x i32> [[VEC:%.*]], <vscale x 4 x i32>* [[VEC_ADDR]], align 16
@@ -30,19 +29,17 @@
 // CHECK-NEXT:    store <vscale x 16 x i1> [[TMP5]], <vscale x 16 x i1>* [[PG]], align 2
 // CHECK-NEXT:    [[TMP6:%.*]] = load <vscale x 16 x i1>, <vscale x 16 x i1>* [[PG]], align 2
 // CHECK-NEXT:    [[TMP7:%.*]] = load <16 x i32>, <16 x i32>* @global_vec, align 16
-// CHECK-NEXT:    [[TMP8:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* bitcast (<16 x i32>* @global_vec to <vscale x 4 x i32>*), align 16
-// CHECK-NEXT:    [[TMP9:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[VEC_ADDR]], align 16
-// CHECK-NEXT:    [[TMP10:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[TMP6]])
-// CHECK-NEXT:    [[TMP11:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.add.nxv4i32(<vscale x 4 x i1> [[TMP10]], <vscale x 4 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]])
-// CHECK-NEXT:    store <vscale x 4 x i32> [[TMP11]], <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]], align 16
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP12:%.*]] = load <16 x i32>, <16 x i32>* [[CASTFIXEDSVE]], align 16
-// CHECK-NEXT:    store <16 x i32> [[TMP12]], <16 x i32>* [[RETVAL]], align 16
-// CHECK-NEXT:    [[TMP13:%.*]] = bitcast <vscale x 4 x i32>* [[RETVAL_COERCE]] to i8*
-// CHECK-NEXT:    [[TMP14:%.*]] = bitcast <16 x i32>* [[RETVAL]] to i8*
-// CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 [[TMP13]], i8* align 16 [[TMP14]], i64 64, i1 false)
-// CHECK-NEXT:    [[TMP15:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP15]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[TMP7]], i64 0)
+// CHECK-NEXT:    [[TMP8:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[VEC_ADDR]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[TMP6]])
+// CHECK-NEXT:    [[TMP10:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.add.nxv4i32(<vscale x 4 x i1> [[TMP9]], <vscale x 4 x i32> [[CASTSCALABLESVE]], <vscale x 4 x i32> [[TMP8]])
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[TMP10]], i64 0)
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP11:%.*]] = bitcast <vscale x 4 x i32>* [[RETVAL_COERCE]] to i8*
+// CHECK-NEXT:    [[TMP12:%.*]] = bitcast <16 x i32>* [[RETVAL]] to i8*
+// CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 [[TMP11]], i8* align 16 [[TMP12]], i64 64, i1 false)
+// CHECK-NEXT:    [[TMP13:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP13]]
 //
 fixed_int32_t foo(svbool_t pred, svint32_t vec) {
   svbool_t pg = svand_z(pred, global_pred, global_pred);
Index: clang/test/CodeGen/attr-arm-sve-vector-bits-cast.c
===================================================================
--- clang/test/CodeGen/attr-arm-sve-vector-bits-cast.c
+++ clang/test/CodeGen/attr-arm-sve-vector-bits-cast.c
@@ -14,14 +14,11 @@
 // CHECK-LABEL: @to_svint32_t(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TYPE:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[TYPE_ADDR:%.*]] = alloca <16 x i32>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <16 x i32>* [[TYPE]] to <vscale x 4 x i32>*
 // CHECK-NEXT:    store <vscale x 4 x i32> [[TYPE_COERCE:%.*]], <vscale x 4 x i32>* [[TMP0]], align 16
 // CHECK-NEXT:    [[TYPE1:%.*]] = load <16 x i32>, <16 x i32>* [[TYPE]], align 16, [[TBAA6:!tbaa !.*]]
-// CHECK-NEXT:    store <16 x i32> [[TYPE1]], <16 x i32>* [[TYPE_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i32>* [[TYPE_ADDR]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[TYPE1]], i64 0)
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[CASTSCALABLESVE]]
 //
 svint32_t to_svint32_t(fixed_int32_t type) {
   return type;
@@ -29,15 +26,12 @@
 
 // CHECK-LABEL: @from_svint32_t(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TYPE_ADDR:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 4 x i32>, align 16
-// CHECK-NEXT:    store <vscale x 4 x i32> [[TYPE:%.*]], <vscale x 4 x i32>* [[TYPE_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 4 x i32>* [[TYPE_ADDR]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, <16 x i32>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[TYPE:%.*]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 4 x i32>* [[RETVAL_COERCE]] to <16 x i32>*
-// CHECK-NEXT:    store <16 x i32> [[TMP1]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
 //
 fixed_int32_t from_svint32_t(svint32_t type) {
   return type;
@@ -46,14 +40,11 @@
 // CHECK-LABEL: @to_svfloat64_t(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TYPE:%.*]] = alloca <8 x double>, align 16
-// CHECK-NEXT:    [[TYPE_ADDR:%.*]] = alloca <8 x double>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x double>* [[TYPE]] to <vscale x 2 x double>*
 // CHECK-NEXT:    store <vscale x 2 x double> [[TYPE_COERCE:%.*]], <vscale x 2 x double>* [[TMP0]], align 16
 // CHECK-NEXT:    [[TYPE1:%.*]] = load <8 x double>, <8 x double>* [[TYPE]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <8 x double> [[TYPE1]], <8 x double>* [[TYPE_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x double>* [[TYPE_ADDR]] to <vscale x 2 x double>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[TMP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x double> @llvm.experimental.vector.insert.nxv2f64.v8f64(<vscale x 2 x double> undef, <8 x double> [[TYPE1]], i64 0)
+// CHECK-NEXT:    ret <vscale x 2 x double> [[CASTSCALABLESVE]]
 //
 svfloat64_t to_svfloat64_t(fixed_float64_t type) {
   return type;
@@ -61,15 +52,12 @@
 
 // CHECK-LABEL: @from_svfloat64_t(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TYPE_ADDR:%.*]] = alloca <vscale x 2 x double>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 2 x double>, align 16
-// CHECK-NEXT:    store <vscale x 2 x double> [[TYPE:%.*]], <vscale x 2 x double>* [[TYPE_ADDR]], align 16, [[TBAA11:!tbaa !.*]]
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x double>* [[TYPE_ADDR]] to <8 x double>*
-// CHECK-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x double> @llvm.experimental.vector.extract.v8f64.nxv2f64(<vscale x 2 x double> [[TYPE:%.*]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 2 x double>* [[RETVAL_COERCE]] to <8 x double>*
-// CHECK-NEXT:    store <8 x double> [[TMP1]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
+// CHECK-NEXT:    store <8 x double> [[CASTFIXEDSVE]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP0]]
 //
 fixed_float64_t from_svfloat64_t(svfloat64_t type) {
   return type;
@@ -95,7 +83,7 @@
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TYPE_ADDR:%.*]] = alloca <vscale x 16 x i1>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 16 x i1>, align 16
-// CHECK-NEXT:    store <vscale x 16 x i1> [[TYPE:%.*]], <vscale x 16 x i1>* [[TYPE_ADDR]], align 16, [[TBAA13:!tbaa !.*]]
+// CHECK-NEXT:    store <vscale x 16 x i1> [[TYPE:%.*]], <vscale x 16 x i1>* [[TYPE_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 16 x i1>* [[TYPE_ADDR]] to <8 x i8>*
 // CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 16, [[TBAA6]]
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 16 x i1>* [[RETVAL_COERCE]] to <8 x i8>*
@@ -109,12 +97,9 @@
 
 // CHECK-LABEL: @to_svint32_t__from_gnu_int32_t(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TYPE_ADDR:%.*]] = alloca <16 x i32>, align 16
 // CHECK-NEXT:    [[TYPE:%.*]] = load <16 x i32>, <16 x i32>* [[TMP0:%.*]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <16 x i32> [[TYPE]], <16 x i32>* [[TYPE_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i32>* [[TYPE_ADDR]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[TYPE]], i64 0)
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[CASTSCALABLESVE]]
 //
 svint32_t to_svint32_t__from_gnu_int32_t(gnu_int32_t type) {
   return type;
@@ -122,11 +107,8 @@
 
 // CHECK-LABEL: @from_svint32_t__to_gnu_int32_t(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TYPE_ADDR:%.*]] = alloca <vscale x 4 x i32>, align 16
-// CHECK-NEXT:    store <vscale x 4 x i32> [[TYPE:%.*]], <vscale x 4 x i32>* [[TYPE_ADDR]], align 16, [[TBAA9]]
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 4 x i32>* [[TYPE_ADDR]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, <16 x i32>* [[TMP0]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <16 x i32> [[TMP1]], <16 x i32>* [[AGG_RESULT:%.*]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[TYPE:%.*]], i64 0)
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[AGG_RESULT:%.*]], align 16, [[TBAA6]]
 // CHECK-NEXT:    ret void
 //
 gnu_int32_t from_svint32_t__to_gnu_int32_t(svint32_t type) {
Index: clang/test/CodeGen/attr-arm-sve-vector-bits-call.c
===================================================================
--- clang/test/CodeGen/attr-arm-sve-vector-bits-call.c
+++ clang/test/CodeGen/attr-arm-sve-vector-bits-call.c
@@ -25,22 +25,16 @@
 // CHECK-LABEL: @fixed_caller(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[X:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[X_ADDR:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <16 x i32>* [[X]] to <vscale x 4 x i32>*
 // CHECK-NEXT:    store <vscale x 4 x i32> [[X_COERCE:%.*]], <vscale x 4 x i32>* [[TMP0]], align 16
 // CHECK-NEXT:    [[X1:%.*]] = load <16 x i32>, <16 x i32>* [[X]], align 16, [[TBAA6:!tbaa !.*]]
-// CHECK-NEXT:    store <16 x i32> [[X1]], <16 x i32>* [[X_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i32>* [[X_ADDR]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA9:!tbaa !.*]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, <16 x i32>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[X1]], i64 0)
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[CASTSCALABLESVE]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 4 x i32>* [[RETVAL_COERCE]] to <16 x i32>*
-// CHECK-NEXT:    store <16 x i32> [[TMP3]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP4:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP1:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP1]]
 //
 fixed_int32_t fixed_caller(fixed_int32_t x) {
   return sizeless_callee(x);
@@ -64,24 +58,18 @@
 
 // CHECK-LABEL: @sizeless_caller(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[COERCE_COERCE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[COERCE1:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <16 x i32>, align 64
-// CHECK-NEXT:    store <vscale x 4 x i32> [[X:%.*]], <vscale x 4 x i32>* [[X_ADDR]], align 16, [[TBAA9]]
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 4 x i32>* [[X_ADDR]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, <16 x i32>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[X:%.*]], i64 0)
 // CHECK-NEXT:    [[COERCE_0__SROA_CAST:%.*]] = bitcast <vscale x 4 x i32>* [[COERCE_COERCE]] to <16 x i32>*
-// CHECK-NEXT:    store <16 x i32> [[TMP1]], <16 x i32>* [[COERCE_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[COERCE_COERCE]], align 16
-// CHECK-NEXT:    [[CALL:%.*]] = call <vscale x 4 x i32> @fixed_callee(<vscale x 4 x i32> [[TMP2]])
-// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <16 x i32>* [[COERCE1]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    store <vscale x 4 x i32> [[CALL]], <vscale x 4 x i32>* [[TMP3]], align 16
-// CHECK-NEXT:    [[TMP4:%.*]] = load <16 x i32>, <16 x i32>* [[COERCE1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <16 x i32> [[TMP4]], <16 x i32>* [[SAVED_CALL_RVALUE]], align 64, [[TBAA6]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <16 x i32>* [[SAVED_CALL_RVALUE]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP5:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[CASTFIXEDSVE]], align 64, [[TBAA6]]
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP5]]
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[COERCE_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[COERCE_COERCE]], align 16
+// CHECK-NEXT:    [[CALL:%.*]] = call <vscale x 4 x i32> @fixed_callee(<vscale x 4 x i32> [[TMP0]])
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i32>* [[COERCE1]] to <vscale x 4 x i32>*
+// CHECK-NEXT:    store <vscale x 4 x i32> [[CALL]], <vscale x 4 x i32>* [[TMP1]], align 16
+// CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* [[COERCE1]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[TMP2]], i64 0)
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[CASTSCALABLESVE]]
 //
 svint32_t sizeless_caller(svint32_t x) {
   return fixed_callee(x);
@@ -95,9 +83,6 @@
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[OP1:%.*]] = alloca <16 x i32>, align 16
 // CHECK-NEXT:    [[OP2:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[OP1_ADDR:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[OP2_ADDR:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <16 x i32>* [[OP1]] to <vscale x 4 x i32>*
 // CHECK-NEXT:    store <vscale x 4 x i32> [[OP1_COERCE:%.*]], <vscale x 4 x i32>* [[TMP0]], align 16
@@ -105,21 +90,15 @@
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i32>* [[OP2]] to <vscale x 4 x i32>*
 // CHECK-NEXT:    store <vscale x 4 x i32> [[OP2_COERCE:%.*]], <vscale x 4 x i32>* [[TMP1]], align 16
 // CHECK-NEXT:    [[OP22:%.*]] = load <16 x i32>, <16 x i32>* [[OP2]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <16 x i32> [[OP11]], <16 x i32>* [[OP1_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <16 x i32> [[OP22]], <16 x i32>* [[OP2_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <16 x i32>* [[OP1_ADDR]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP3:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP2]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32>* [[OP2_ADDR]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP5:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP4]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP6:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP7:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sel.nxv4i32(<vscale x 4 x i1> [[TMP6]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP5]])
-// CHECK-NEXT:    store <vscale x 4 x i32> [[TMP7]], <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA9]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP8:%.*]] = load <16 x i32>, <16 x i32>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[OP11]], i64 0)
+// CHECK-NEXT:    [[CASTSCALABLESVE3:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[OP22]], i64 0)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sel.nxv4i32(<vscale x 4 x i1> [[TMP2]], <vscale x 4 x i32> [[CASTSCALABLESVE]], <vscale x 4 x i32> [[CASTSCALABLESVE3]])
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[TMP3]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 4 x i32>* [[RETVAL_COERCE]] to <16 x i32>*
-// CHECK-NEXT:    store <16 x i32> [[TMP8]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP9:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP9]]
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP4:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 fixed_int32_t call_int32_ff(svbool_t pg, fixed_int32_t op1, fixed_int32_t op2) {
   return svsel(pg, op1, op2);
@@ -129,9 +108,6 @@
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[OP1:%.*]] = alloca <8 x double>, align 16
 // CHECK-NEXT:    [[OP2:%.*]] = alloca <8 x double>, align 16
-// CHECK-NEXT:    [[OP1_ADDR:%.*]] = alloca <8 x double>, align 16
-// CHECK-NEXT:    [[OP2_ADDR:%.*]] = alloca <8 x double>, align 16
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 2 x double>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 2 x double>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x double>* [[OP1]] to <vscale x 2 x double>*
 // CHECK-NEXT:    store <vscale x 2 x double> [[OP1_COERCE:%.*]], <vscale x 2 x double>* [[TMP0]], align 16
@@ -139,21 +115,15 @@
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x double>* [[OP2]] to <vscale x 2 x double>*
 // CHECK-NEXT:    store <vscale x 2 x double> [[OP2_COERCE:%.*]], <vscale x 2 x double>* [[TMP1]], align 16
 // CHECK-NEXT:    [[OP22:%.*]] = load <8 x double>, <8 x double>* [[OP2]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <8 x double> [[OP11]], <8 x double>* [[OP1_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <8 x double> [[OP22]], <8 x double>* [[OP2_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x double>* [[OP1_ADDR]] to <vscale x 2 x double>*
-// CHECK-NEXT:    [[TMP3:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[TMP2]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x double>* [[OP2_ADDR]] to <vscale x 2 x double>*
-// CHECK-NEXT:    [[TMP5:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[TMP4]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP6:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP7:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.sel.nxv2f64(<vscale x 2 x i1> [[TMP6]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP5]])
-// CHECK-NEXT:    store <vscale x 2 x double> [[TMP7]], <vscale x 2 x double>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA11:!tbaa !.*]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 2 x double>* [[SAVED_CALL_RVALUE]] to <8 x double>*
-// CHECK-NEXT:    [[TMP8:%.*]] = load <8 x double>, <8 x double>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x double> @llvm.experimental.vector.insert.nxv2f64.v8f64(<vscale x 2 x double> undef, <8 x double> [[OP11]], i64 0)
+// CHECK-NEXT:    [[CASTSCALABLESVE3:%.*]] = call <vscale x 2 x double> @llvm.experimental.vector.insert.nxv2f64.v8f64(<vscale x 2 x double> undef, <8 x double> [[OP22]], i64 0)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.sel.nxv2f64(<vscale x 2 x i1> [[TMP2]], <vscale x 2 x double> [[CASTSCALABLESVE]], <vscale x 2 x double> [[CASTSCALABLESVE3]])
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x double> @llvm.experimental.vector.extract.v8f64.nxv2f64(<vscale x 2 x double> [[TMP3]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 2 x double>* [[RETVAL_COERCE]] to <8 x double>*
-// CHECK-NEXT:    store <8 x double> [[TMP8]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP9:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP9]]
+// CHECK-NEXT:    store <8 x double> [[CASTFIXEDSVE]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP4:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP4]]
 //
 fixed_float64_t call_float64_ff(svbool_t pg, fixed_float64_t op1, fixed_float64_t op2) {
   return svsel(pg, op1, op2);
@@ -180,7 +150,7 @@
 // CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i8>* [[OP2_ADDR]] to <vscale x 16 x i1>*
 // CHECK-NEXT:    [[TMP5:%.*]] = load <vscale x 16 x i1>, <vscale x 16 x i1>* [[TMP4]], align 16, [[TBAA6]]
 // CHECK-NEXT:    [[TMP6:%.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.sel.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i1> [[TMP3]], <vscale x 16 x i1> [[TMP5]])
-// CHECK-NEXT:    store <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA13:!tbaa !.*]]
+// CHECK-NEXT:    store <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA9:!tbaa !.*]]
 // CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]] to <8 x i8>*
 // CHECK-NEXT:    [[TMP7:%.*]] = load <8 x i8>, <8 x i8>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 16 x i1>* [[RETVAL_COERCE]] to <8 x i8>*
@@ -199,24 +169,18 @@
 // CHECK-LABEL: @call_int32_fs(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[OP1:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[OP1_ADDR:%.*]] = alloca <16 x i32>, align 16
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <16 x i32>* [[OP1]] to <vscale x 4 x i32>*
 // CHECK-NEXT:    store <vscale x 4 x i32> [[OP1_COERCE:%.*]], <vscale x 4 x i32>* [[TMP0]], align 16
 // CHECK-NEXT:    [[OP11:%.*]] = load <16 x i32>, <16 x i32>* [[OP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <16 x i32> [[OP11]], <16 x i32>* [[OP1_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i32>* [[OP1_ADDR]] to <vscale x 4 x i32>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[TMP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sel.nxv4i32(<vscale x 4 x i1> [[TMP3]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[OP2:%.*]])
-// CHECK-NEXT:    store <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA9]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP5:%.*]] = load <16 x i32>, <16 x i32>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vector.insert.nxv4i32.v16i32(<vscale x 4 x i32> undef, <16 x i32> [[OP11]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sel.nxv4i32(<vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[CASTSCALABLESVE]], <vscale x 4 x i32> [[OP2:%.*]])
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[TMP2]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 4 x i32>* [[RETVAL_COERCE]] to <16 x i32>*
-// CHECK-NEXT:    store <16 x i32> [[TMP5]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP6:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP6]]
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP3:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 //
 fixed_int32_t call_int32_fs(svbool_t pg, fixed_int32_t op1, svint32_t op2) {
   return svsel(pg, op1, op2);
@@ -225,24 +189,18 @@
 // CHECK-LABEL: @call_float64_fs(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[OP1:%.*]] = alloca <8 x double>, align 16
-// CHECK-NEXT:    [[OP1_ADDR:%.*]] = alloca <8 x double>, align 16
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 2 x double>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 2 x double>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x double>* [[OP1]] to <vscale x 2 x double>*
 // CHECK-NEXT:    store <vscale x 2 x double> [[OP1_COERCE:%.*]], <vscale x 2 x double>* [[TMP0]], align 16
 // CHECK-NEXT:    [[OP11:%.*]] = load <8 x double>, <8 x double>* [[OP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    store <8 x double> [[OP11]], <8 x double>* [[OP1_ADDR]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x double>* [[OP1_ADDR]] to <vscale x 2 x double>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[TMP1]], align 16, [[TBAA6]]
-// CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT:    [[TMP4:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.sel.nxv2f64(<vscale x 2 x i1> [[TMP3]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[OP2:%.*]])
-// CHECK-NEXT:    store <vscale x 2 x double> [[TMP4]], <vscale x 2 x double>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA11]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 2 x double>* [[SAVED_CALL_RVALUE]] to <8 x double>*
-// CHECK-NEXT:    [[TMP5:%.*]] = load <8 x double>, <8 x double>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x double> @llvm.experimental.vector.insert.nxv2f64.v8f64(<vscale x 2 x double> undef, <8 x double> [[OP11]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.sel.nxv2f64(<vscale x 2 x i1> [[TMP1]], <vscale x 2 x double> [[CASTSCALABLESVE]], <vscale x 2 x double> [[OP2:%.*]])
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x double> @llvm.experimental.vector.extract.v8f64.nxv2f64(<vscale x 2 x double> [[TMP2]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 2 x double>* [[RETVAL_COERCE]] to <8 x double>*
-// CHECK-NEXT:    store <8 x double> [[TMP5]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP6:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP6]]
+// CHECK-NEXT:    store <8 x double> [[CASTFIXEDSVE]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP3:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP3]]
 //
 fixed_float64_t call_float64_fs(svbool_t pg, fixed_float64_t op1, svfloat64_t op2) {
   return svsel(pg, op1, op2);
@@ -261,7 +219,7 @@
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8>* [[OP1_ADDR]] to <vscale x 16 x i1>*
 // CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 16 x i1>, <vscale x 16 x i1>* [[TMP1]], align 16, [[TBAA6]]
 // CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.sel.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i1> [[TMP2]], <vscale x 16 x i1> [[OP2:%.*]])
-// CHECK-NEXT:    store <vscale x 16 x i1> [[TMP3]], <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA13]]
+// CHECK-NEXT:    store <vscale x 16 x i1> [[TMP3]], <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA9]]
 // CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]] to <8 x i8>*
 // CHECK-NEXT:    [[TMP4:%.*]] = load <8 x i8>, <8 x i8>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 16 x i1>* [[RETVAL_COERCE]] to <8 x i8>*
@@ -279,17 +237,14 @@
 
 // CHECK-LABEL: @call_int32_ss(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 4 x i32>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sel.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[OP1:%.*]], <vscale x 4 x i32> [[OP2:%.*]])
-// CHECK-NEXT:    store <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA9]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 4 x i32>* [[SAVED_CALL_RVALUE]] to <16 x i32>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i32> @llvm.experimental.vector.extract.v16i32.nxv4i32(<vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 4 x i32>* [[RETVAL_COERCE]] to <16 x i32>*
-// CHECK-NEXT:    store <16 x i32> [[TMP2]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP3:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
+// CHECK-NEXT:    store <16 x i32> [[CASTFIXEDSVE]], <16 x i32>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 4 x i32>, <vscale x 4 x i32>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 fixed_int32_t call_int32_ss(svbool_t pg, svint32_t op1, svint32_t op2) {
   return svsel(pg, op1, op2);
@@ -297,17 +252,14 @@
 
 // CHECK-LABEL: @call_float64_ss(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 2 x double>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 2 x double>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.sel.nxv2f64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x double> [[OP1:%.*]], <vscale x 2 x double> [[OP2:%.*]])
-// CHECK-NEXT:    store <vscale x 2 x double> [[TMP1]], <vscale x 2 x double>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA11]]
-// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 2 x double>* [[SAVED_CALL_RVALUE]] to <8 x double>*
-// CHECK-NEXT:    [[TMP2:%.*]] = load <8 x double>, <8 x double>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x double> @llvm.experimental.vector.extract.v8f64.nxv2f64(<vscale x 2 x double> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 2 x double>* [[RETVAL_COERCE]] to <8 x double>*
-// CHECK-NEXT:    store <8 x double> [[TMP2]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
-// CHECK-NEXT:    [[TMP3:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP3]]
+// CHECK-NEXT:    store <8 x double> [[CASTFIXEDSVE]], <8 x double>* [[RETVAL_0__SROA_CAST]], align 16
+// CHECK-NEXT:    [[TMP2:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[RETVAL_COERCE]], align 16
+// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
 //
 fixed_float64_t call_float64_ss(svbool_t pg, svfloat64_t op1, svfloat64_t op2) {
   return svsel(pg, op1, op2);
@@ -318,7 +270,7 @@
 // CHECK-NEXT:    [[SAVED_CALL_RVALUE:%.*]] = alloca <vscale x 16 x i1>, align 16
 // CHECK-NEXT:    [[RETVAL_COERCE:%.*]] = alloca <vscale x 16 x i1>, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.sel.nxv16i1(<vscale x 16 x i1> [[PG:%.*]], <vscale x 16 x i1> [[OP1:%.*]], <vscale x 16 x i1> [[OP2:%.*]])
-// CHECK-NEXT:    store <vscale x 16 x i1> [[TMP0]], <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA13]]
+// CHECK-NEXT:    store <vscale x 16 x i1> [[TMP0]], <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]], align 16, [[TBAA9]]
 // CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = bitcast <vscale x 16 x i1>* [[SAVED_CALL_RVALUE]] to <8 x i8>*
 // CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[CASTFIXEDSVE]], align 16, [[TBAA6]]
 // CHECK-NEXT:    [[RETVAL_0__SROA_CAST:%.*]] = bitcast <vscale x 16 x i1>* [[RETVAL_COERCE]] to <8 x i8>*
Index: clang/test/CodeGen/attr-arm-sve-vector-bits-bitcast.c
===================================================================
--- clang/test/CodeGen/attr-arm-sve-vector-bits-bitcast.c
+++ clang/test/CodeGen/attr-arm-sve-vector-bits-bitcast.c
@@ -30,23 +30,23 @@
 // CHECK-128-LABEL: @read_int64(
 // CHECK-128-NEXT:  entry:
 // CHECK-128-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_INT64:%.*]], %struct.struct_int64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64>* [[ARRAYIDX]] to <vscale x 2 x i64>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP0]], align 16, [[TBAA6:!tbaa !.*]]
-// CHECK-128-NEXT:    ret <vscale x 2 x i64> [[TMP1]]
+// CHECK-128-NEXT:    [[TMP0:%.*]] = load <2 x i64>, <2 x i64>* [[ARRAYIDX]], align 16, [[TBAA6:!tbaa !.*]]
+// CHECK-128-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x i64> @llvm.experimental.vector.insert.nxv2i64.v2i64(<vscale x 2 x i64> undef, <2 x i64> [[TMP0]], i64 0)
+// CHECK-128-NEXT:    ret <vscale x 2 x i64> [[CASTSCALABLESVE]]
 //
 // CHECK-256-LABEL: @read_int64(
 // CHECK-256-NEXT:  entry:
 // CHECK-256-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_INT64:%.*]], %struct.struct_int64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-256-NEXT:    [[TMP0:%.*]] = bitcast <4 x i64>* [[ARRAYIDX]] to <vscale x 2 x i64>*
-// CHECK-256-NEXT:    [[TMP1:%.*]] = load <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP0]], align 16, [[TBAA6:!tbaa !.*]]
-// CHECK-256-NEXT:    ret <vscale x 2 x i64> [[TMP1]]
+// CHECK-256-NEXT:    [[TMP0:%.*]] = load <4 x i64>, <4 x i64>* [[ARRAYIDX]], align 16, [[TBAA6:!tbaa !.*]]
+// CHECK-256-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x i64> @llvm.experimental.vector.insert.nxv2i64.v4i64(<vscale x 2 x i64> undef, <4 x i64> [[TMP0]], i64 0)
+// CHECK-256-NEXT:    ret <vscale x 2 x i64> [[CASTSCALABLESVE]]
 //
 // CHECK-512-LABEL: @read_int64(
 // CHECK-512-NEXT:  entry:
 // CHECK-512-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_INT64:%.*]], %struct.struct_int64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <8 x i64>* [[ARRAYIDX]] to <vscale x 2 x i64>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP0]], align 16, [[TBAA6:!tbaa !.*]]
-// CHECK-512-NEXT:    ret <vscale x 2 x i64> [[TMP1]]
+// CHECK-512-NEXT:    [[TMP0:%.*]] = load <8 x i64>, <8 x i64>* [[ARRAYIDX]], align 16, [[TBAA6:!tbaa !.*]]
+// CHECK-512-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x i64> @llvm.experimental.vector.insert.nxv2i64.v8i64(<vscale x 2 x i64> undef, <8 x i64> [[TMP0]], i64 0)
+// CHECK-512-NEXT:    ret <vscale x 2 x i64> [[CASTSCALABLESVE]]
 //
 svint64_t read_int64(struct struct_int64 *s) {
   return s->y[0];
@@ -54,32 +54,23 @@
 
 // CHECK-128-LABEL: @write_int64(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 2 x i64>, align 16
-// CHECK-128-NEXT:    store <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64>* [[X_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x i64>* [[X_ADDR]] to <2 x i64>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    [[CASTFIXEDSVE:%.*]] = call <2 x i64> @llvm.experimental.vector.extract.v2i64.nxv2i64(<vscale x 2 x i64> [[X:%.*]], i64 0)
 // CHECK-128-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_INT64:%.*]], %struct.struct_int64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-128-NEXT:    store <2 x i64> [[TMP1]], <2 x i64>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    store <2 x i64> [[CASTFIXEDSVE]], <2 x i64>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-128-NEXT:    ret void
 //
 // CHECK-256-LABEL: @write_int64(
 // CHECK-256-NEXT:  entry:
-// CHECK-256-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 2 x i64>, align 16
-// CHECK-256-NEXT:    store <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64>* [[X_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
-// CHECK-256-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x i64>* [[X_ADDR]] to <4 x i64>*
-// CHECK-256-NEXT:    [[TMP1:%.*]] = load <4 x i64>, <4 x i64>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    [[CASTFIXEDSVE:%.*]] = call <4 x i64> @llvm.experimental.vector.extract.v4i64.nxv2i64(<vscale x 2 x i64> [[X:%.*]], i64 0)
 // CHECK-256-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_INT64:%.*]], %struct.struct_int64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-256-NEXT:    store <4 x i64> [[TMP1]], <4 x i64>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    store <4 x i64> [[CASTFIXEDSVE]], <4 x i64>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-256-NEXT:    ret void
 //
 // CHECK-512-LABEL: @write_int64(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 2 x i64>, align 16
-// CHECK-512-NEXT:    store <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64>* [[X_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x i64>* [[X_ADDR]] to <8 x i64>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <8 x i64>, <8 x i64>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x i64> @llvm.experimental.vector.extract.v8i64.nxv2i64(<vscale x 2 x i64> [[X:%.*]], i64 0)
 // CHECK-512-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_INT64:%.*]], %struct.struct_int64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-512-NEXT:    store <8 x i64> [[TMP1]], <8 x i64>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    store <8 x i64> [[CASTFIXEDSVE]], <8 x i64>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-512-NEXT:    ret void
 //
 void write_int64(struct struct_int64 *s, svint64_t x) {
@@ -93,23 +84,23 @@
 // CHECK-128-LABEL: @read_float64(
 // CHECK-128-NEXT:  entry:
 // CHECK-128-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_FLOAT64:%.*]], %struct.struct_float64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <2 x double>* [[ARRAYIDX]] to <vscale x 2 x double>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[TMP0]], align 16, [[TBAA6]]
-// CHECK-128-NEXT:    ret <vscale x 2 x double> [[TMP1]]
+// CHECK-128-NEXT:    [[TMP0:%.*]] = load <2 x double>, <2 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x double> @llvm.experimental.vector.insert.nxv2f64.v2f64(<vscale x 2 x double> undef, <2 x double> [[TMP0]], i64 0)
+// CHECK-128-NEXT:    ret <vscale x 2 x double> [[CASTSCALABLESVE]]
 //
 // CHECK-256-LABEL: @read_float64(
 // CHECK-256-NEXT:  entry:
 // CHECK-256-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_FLOAT64:%.*]], %struct.struct_float64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-256-NEXT:    [[TMP0:%.*]] = bitcast <4 x double>* [[ARRAYIDX]] to <vscale x 2 x double>*
-// CHECK-256-NEXT:    [[TMP1:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[TMP0]], align 16, [[TBAA6]]
-// CHECK-256-NEXT:    ret <vscale x 2 x double> [[TMP1]]
+// CHECK-256-NEXT:    [[TMP0:%.*]] = load <4 x double>, <4 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x double> @llvm.experimental.vector.insert.nxv2f64.v4f64(<vscale x 2 x double> undef, <4 x double> [[TMP0]], i64 0)
+// CHECK-256-NEXT:    ret <vscale x 2 x double> [[CASTSCALABLESVE]]
 //
 // CHECK-512-LABEL: @read_float64(
 // CHECK-512-NEXT:  entry:
 // CHECK-512-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_FLOAT64:%.*]], %struct.struct_float64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <8 x double>* [[ARRAYIDX]] to <vscale x 2 x double>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <vscale x 2 x double>, <vscale x 2 x double>* [[TMP0]], align 16, [[TBAA6]]
-// CHECK-512-NEXT:    ret <vscale x 2 x double> [[TMP1]]
+// CHECK-512-NEXT:    [[TMP0:%.*]] = load <8 x double>, <8 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 2 x double> @llvm.experimental.vector.insert.nxv2f64.v8f64(<vscale x 2 x double> undef, <8 x double> [[TMP0]], i64 0)
+// CHECK-512-NEXT:    ret <vscale x 2 x double> [[CASTSCALABLESVE]]
 //
 svfloat64_t read_float64(struct struct_float64 *s) {
   return s->y[0];
@@ -117,32 +108,23 @@
 
 // CHECK-128-LABEL: @write_float64(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 2 x double>, align 16
-// CHECK-128-NEXT:    store <vscale x 2 x double> [[X:%.*]], <vscale x 2 x double>* [[X_ADDR]], align 16, [[TBAA11:!tbaa !.*]]
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x double>* [[X_ADDR]] to <2 x double>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    [[CASTFIXEDSVE:%.*]] = call <2 x double> @llvm.experimental.vector.extract.v2f64.nxv2f64(<vscale x 2 x double> [[X:%.*]], i64 0)
 // CHECK-128-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_FLOAT64:%.*]], %struct.struct_float64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-128-NEXT:    store <2 x double> [[TMP1]], <2 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    store <2 x double> [[CASTFIXEDSVE]], <2 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-128-NEXT:    ret void
 //
 // CHECK-256-LABEL: @write_float64(
 // CHECK-256-NEXT:  entry:
-// CHECK-256-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 2 x double>, align 16
-// CHECK-256-NEXT:    store <vscale x 2 x double> [[X:%.*]], <vscale x 2 x double>* [[X_ADDR]], align 16, [[TBAA11:!tbaa !.*]]
-// CHECK-256-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x double>* [[X_ADDR]] to <4 x double>*
-// CHECK-256-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    [[CASTFIXEDSVE:%.*]] = call <4 x double> @llvm.experimental.vector.extract.v4f64.nxv2f64(<vscale x 2 x double> [[X:%.*]], i64 0)
 // CHECK-256-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_FLOAT64:%.*]], %struct.struct_float64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-256-NEXT:    store <4 x double> [[TMP1]], <4 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    store <4 x double> [[CASTFIXEDSVE]], <4 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-256-NEXT:    ret void
 //
 // CHECK-512-LABEL: @write_float64(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 2 x double>, align 16
-// CHECK-512-NEXT:    store <vscale x 2 x double> [[X:%.*]], <vscale x 2 x double>* [[X_ADDR]], align 16, [[TBAA11:!tbaa !.*]]
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 2 x double>* [[X_ADDR]] to <8 x double>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x double> @llvm.experimental.vector.extract.v8f64.nxv2f64(<vscale x 2 x double> [[X:%.*]], i64 0)
 // CHECK-512-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_FLOAT64:%.*]], %struct.struct_float64* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-512-NEXT:    store <8 x double> [[TMP1]], <8 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    store <8 x double> [[CASTFIXEDSVE]], <8 x double>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-512-NEXT:    ret void
 //
 void write_float64(struct struct_float64 *s, svfloat64_t x) {
@@ -156,23 +138,23 @@
 // CHECK-128-LABEL: @read_bfloat16(
 // CHECK-128-NEXT:  entry:
 // CHECK-128-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BFLOAT16:%.*]], %struct.struct_bfloat16* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <8 x bfloat>* [[ARRAYIDX]] to <vscale x 8 x bfloat>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP0]], align 16, [[TBAA6]]
-// CHECK-128-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+// CHECK-128-NEXT:    [[TMP0:%.*]] = load <8 x bfloat>, <8 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 8 x bfloat> @llvm.experimental.vector.insert.nxv8bf16.v8bf16(<vscale x 8 x bfloat> undef, <8 x bfloat> [[TMP0]], i64 0)
+// CHECK-128-NEXT:    ret <vscale x 8 x bfloat> [[CASTSCALABLESVE]]
 //
 // CHECK-256-LABEL: @read_bfloat16(
 // CHECK-256-NEXT:  entry:
 // CHECK-256-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BFLOAT16:%.*]], %struct.struct_bfloat16* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-256-NEXT:    [[TMP0:%.*]] = bitcast <16 x bfloat>* [[ARRAYIDX]] to <vscale x 8 x bfloat>*
-// CHECK-256-NEXT:    [[TMP1:%.*]] = load <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP0]], align 16, [[TBAA6]]
-// CHECK-256-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+// CHECK-256-NEXT:    [[TMP0:%.*]] = load <16 x bfloat>, <16 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 8 x bfloat> @llvm.experimental.vector.insert.nxv8bf16.v16bf16(<vscale x 8 x bfloat> undef, <16 x bfloat> [[TMP0]], i64 0)
+// CHECK-256-NEXT:    ret <vscale x 8 x bfloat> [[CASTSCALABLESVE]]
 //
 // CHECK-512-LABEL: @read_bfloat16(
 // CHECK-512-NEXT:  entry:
 // CHECK-512-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BFLOAT16:%.*]], %struct.struct_bfloat16* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <32 x bfloat>* [[ARRAYIDX]] to <vscale x 8 x bfloat>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* [[TMP0]], align 16, [[TBAA6]]
-// CHECK-512-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+// CHECK-512-NEXT:    [[TMP0:%.*]] = load <32 x bfloat>, <32 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 8 x bfloat> @llvm.experimental.vector.insert.nxv8bf16.v32bf16(<vscale x 8 x bfloat> undef, <32 x bfloat> [[TMP0]], i64 0)
+// CHECK-512-NEXT:    ret <vscale x 8 x bfloat> [[CASTSCALABLESVE]]
 //
 svbfloat16_t read_bfloat16(struct struct_bfloat16 *s) {
   return s->y[0];
@@ -180,32 +162,23 @@
 
 // CHECK-128-LABEL: @write_bfloat16(
 // CHECK-128-NEXT:  entry:
-// CHECK-128-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 8 x bfloat>, align 16
-// CHECK-128-NEXT:    store <vscale x 8 x bfloat> [[X:%.*]], <vscale x 8 x bfloat>* [[X_ADDR]], align 16, [[TBAA13:!tbaa !.*]]
-// CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 8 x bfloat>* [[X_ADDR]] to <8 x bfloat>*
-// CHECK-128-NEXT:    [[TMP1:%.*]] = load <8 x bfloat>, <8 x bfloat>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    [[CASTFIXEDSVE:%.*]] = call <8 x bfloat> @llvm.experimental.vector.extract.v8bf16.nxv8bf16(<vscale x 8 x bfloat> [[X:%.*]], i64 0)
 // CHECK-128-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BFLOAT16:%.*]], %struct.struct_bfloat16* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-128-NEXT:    store <8 x bfloat> [[TMP1]], <8 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-128-NEXT:    store <8 x bfloat> [[CASTFIXEDSVE]], <8 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-128-NEXT:    ret void
 //
 // CHECK-256-LABEL: @write_bfloat16(
 // CHECK-256-NEXT:  entry:
-// CHECK-256-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 8 x bfloat>, align 16
-// CHECK-256-NEXT:    store <vscale x 8 x bfloat> [[X:%.*]], <vscale x 8 x bfloat>* [[X_ADDR]], align 16, [[TBAA13:!tbaa !.*]]
-// CHECK-256-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 8 x bfloat>* [[X_ADDR]] to <16 x bfloat>*
-// CHECK-256-NEXT:    [[TMP1:%.*]] = load <16 x bfloat>, <16 x bfloat>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x bfloat> @llvm.experimental.vector.extract.v16bf16.nxv8bf16(<vscale x 8 x bfloat> [[X:%.*]], i64 0)
 // CHECK-256-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BFLOAT16:%.*]], %struct.struct_bfloat16* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-256-NEXT:    store <16 x bfloat> [[TMP1]], <16 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-256-NEXT:    store <16 x bfloat> [[CASTFIXEDSVE]], <16 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-256-NEXT:    ret void
 //
 // CHECK-512-LABEL: @write_bfloat16(
 // CHECK-512-NEXT:  entry:
-// CHECK-512-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 8 x bfloat>, align 16
-// CHECK-512-NEXT:    store <vscale x 8 x bfloat> [[X:%.*]], <vscale x 8 x bfloat>* [[X_ADDR]], align 16, [[TBAA13:!tbaa !.*]]
-// CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 8 x bfloat>* [[X_ADDR]] to <32 x bfloat>*
-// CHECK-512-NEXT:    [[TMP1:%.*]] = load <32 x bfloat>, <32 x bfloat>* [[TMP0]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    [[CASTFIXEDSVE:%.*]] = call <32 x bfloat> @llvm.experimental.vector.extract.v32bf16.nxv8bf16(<vscale x 8 x bfloat> [[X:%.*]], i64 0)
 // CHECK-512-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BFLOAT16:%.*]], %struct.struct_bfloat16* [[S:%.*]], i64 0, i32 1, i64 0
-// CHECK-512-NEXT:    store <32 x bfloat> [[TMP1]], <32 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
+// CHECK-512-NEXT:    store <32 x bfloat> [[CASTFIXEDSVE]], <32 x bfloat>* [[ARRAYIDX]], align 16, [[TBAA6]]
 // CHECK-512-NEXT:    ret void
 //
 void write_bfloat16(struct struct_bfloat16 *s, svbfloat16_t x) {
@@ -244,7 +217,7 @@
 // CHECK-128-LABEL: @write_bool(
 // CHECK-128-NEXT:  entry:
 // CHECK-128-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 16 x i1>, align 16
-// CHECK-128-NEXT:    store <vscale x 16 x i1> [[X:%.*]], <vscale x 16 x i1>* [[X_ADDR]], align 16, [[TBAA15:!tbaa !.*]]
+// CHECK-128-NEXT:    store <vscale x 16 x i1> [[X:%.*]], <vscale x 16 x i1>* [[X_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
 // CHECK-128-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 16 x i1>* [[X_ADDR]] to <2 x i8>*
 // CHECK-128-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* [[TMP0]], align 16, [[TBAA6]]
 // CHECK-128-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BOOL:%.*]], %struct.struct_bool* [[S:%.*]], i64 0, i32 1, i64 0
@@ -254,7 +227,7 @@
 // CHECK-256-LABEL: @write_bool(
 // CHECK-256-NEXT:  entry:
 // CHECK-256-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 16 x i1>, align 16
-// CHECK-256-NEXT:    store <vscale x 16 x i1> [[X:%.*]], <vscale x 16 x i1>* [[X_ADDR]], align 16, [[TBAA15:!tbaa !.*]]
+// CHECK-256-NEXT:    store <vscale x 16 x i1> [[X:%.*]], <vscale x 16 x i1>* [[X_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
 // CHECK-256-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 16 x i1>* [[X_ADDR]] to <4 x i8>*
 // CHECK-256-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 16, [[TBAA6]]
 // CHECK-256-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BOOL:%.*]], %struct.struct_bool* [[S:%.*]], i64 0, i32 1, i64 0
@@ -264,7 +237,7 @@
 // CHECK-512-LABEL: @write_bool(
 // CHECK-512-NEXT:  entry:
 // CHECK-512-NEXT:    [[X_ADDR:%.*]] = alloca <vscale x 16 x i1>, align 16
-// CHECK-512-NEXT:    store <vscale x 16 x i1> [[X:%.*]], <vscale x 16 x i1>* [[X_ADDR]], align 16, [[TBAA15:!tbaa !.*]]
+// CHECK-512-NEXT:    store <vscale x 16 x i1> [[X:%.*]], <vscale x 16 x i1>* [[X_ADDR]], align 16, [[TBAA9:!tbaa !.*]]
 // CHECK-512-NEXT:    [[TMP0:%.*]] = bitcast <vscale x 16 x i1>* [[X_ADDR]] to <8 x i8>*
 // CHECK-512-NEXT:    [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 16, [[TBAA6]]
 // CHECK-512-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [[STRUCT_STRUCT_BOOL:%.*]], %struct.struct_bool* [[S:%.*]], i64 0, i32 1, i64 0
Index: clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
===================================================================
--- clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
+++ clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
@@ -51,34 +51,22 @@
 typedef int8_t vec_int8 __attribute__((vector_size(N / 8)));
 // CHECK128-LABEL: define <16 x i8> @f2(<16 x i8> %x)
 // CHECK128-NEXT:  entry:
-// CHECK128-NEXT:    %x.addr = alloca <16 x i8>, align 16
-// CHECK128-NEXT:    %saved-call-rvalue = alloca <vscale x 16 x i8>, align 16
-// CHECK128-NEXT:    store <16 x i8> %x, <16 x i8>* %x.addr, align 16
-// CHECK128-NEXT:    %0 = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
-// CHECK128-NEXT:    %1 = bitcast <16 x i8>* %x.addr to <vscale x 16 x i8>*
-// CHECK128-NEXT:    %2 = load <vscale x 16 x i8>, <vscale x 16 x i8>* %1, align 16
-// CHECK128-NEXT:    %3 = call <vscale x 16 x i8> @llvm.aarch64.sve.asrd.nxv16i8(<vscale x 16 x i1> %0, <vscale x 16 x i8> %2, i32 1)
-// CHECK128-NEXT:    store <vscale x 16 x i8> %3, <vscale x 16 x i8>* %saved-call-rvalue, align 16
-// CHECK128-NEXT:    %castFixedSve = bitcast <vscale x 16 x i8>* %saved-call-rvalue to <16 x i8>*
-// CHECK128-NEXT:    %4 = load <16 x i8>, <16 x i8>* %castFixedSve, align 16
-// CHECK128-NEXT:    ret <16 x i8> %4
+// CHECK128-NEXT:    [[TMP0:%.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
+// CHECK128-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vector.insert.nxv16i8.v16i8(<vscale x 16 x i8> undef, <16 x i8> [[X:%.*]], i64 0)
+// CHECK128-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i8> @llvm.aarch64.sve.asrd.nxv16i8(<vscale x 16 x i1> [[TMP0]], <vscale x 16 x i8> [[CASTSCALABLESVE]], i32 1)
+// CHECK128-NEXT:    [[CASTFIXEDSVE:%.*]] = call <16 x i8> @llvm.experimental.vector.extract.v16i8.nxv16i8(<vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK128-NEXT:    ret <16 x i8> [[CASTFIXEDSVE]]
 
 // CHECK-LABEL: define void @f2(
 // CHECK-SAME:   <[[#div(VBITS,8)]] x i8>* noalias nocapture sret(<[[#div(VBITS,8)]] x i8>) align 16 %agg.result, <[[#div(VBITS,8)]] x i8>* nocapture readonly %0)
-// CHECK-NEXT: entry:
-// CHECK-NEXT:   %x.addr = alloca <[[#div(VBITS,8)]] x i8>, align 16
-// CHECK-NEXT:   %saved-call-rvalue = alloca <vscale x 16 x i8>, align 16
-// CHECK-NEXT:   %x = load <[[#div(VBITS,8)]] x i8>, <[[#div(VBITS,8)]] x i8>* %0, align 16
-// CHECK-NEXT:   store <[[#div(VBITS,8)]] x i8> %x, <[[#div(VBITS,8)]] x i8>* %x.addr, align 16
-// CHECK-NEXT:   %1 = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
-// CHECK-NEXT:   %2 = bitcast <[[#div(VBITS,8)]] x i8>* %x.addr to <vscale x 16 x i8>*
-// CHECK-NEXT:   %3 = load <vscale x 16 x i8>, <vscale x 16 x i8>* %2, align 16
-// CHECK-NEXT:   %4 = call <vscale x 16 x i8> @llvm.aarch64.sve.asrd.nxv16i8(<vscale x 16 x i1> %1, <vscale x 16 x i8> %3, i32 1)
-// CHECK-NEXT:   store <vscale x 16 x i8> %4, <vscale x 16 x i8>* %saved-call-rvalue, align 16
-// CHECK-NEXT:   %castFixedSve = bitcast <vscale x 16 x i8>* %saved-call-rvalue to <[[#div(VBITS,8)]] x i8>*
-// CHECK-NEXT:   %5 = load <[[#div(VBITS,8)]] x i8>, <[[#div(VBITS,8)]] x i8>* %castFixedSve, align 16
-// CHECK-NEXT:   store <[[#div(VBITS,8)]] x i8> %5, <[[#div(VBITS,8)]] x i8>* %agg.result, align 16
-// CHECK-NEXT:   ret void
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[X:%.*]] = load <[[#div(VBITS,8)]] x i8>, <[[#div(VBITS,8)]] x i8>* [[TMP0:%.*]], align 16, [[TBAA6:!tbaa !.*]]
+// CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
+// CHECK-NEXT:    [[CASTSCALABLESVE:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vector.insert.nxv16i8.v[[#div(VBITS,8)]]i8(<vscale x 16 x i8> undef, <[[#div(VBITS,8)]] x i8> [[X]], i64 0)
+// CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 16 x i8> @llvm.aarch64.sve.asrd.nxv16i8(<vscale x 16 x i1> [[TMP1]], <vscale x 16 x i8> [[CASTSCALABLESVE]], i32 1)
+// CHECK-NEXT:    [[CASTFIXEDSVE:%.*]] = call <[[#div(VBITS,8)]] x i8> @llvm.experimental.vector.extract.v[[#div(VBITS,8)]]i8.nxv16i8(<vscale x 16 x i8> [[TMP2]], i64 0)
+// CHECK-NEXT:    store <[[#div(VBITS,8)]] x i8> [[CASTFIXEDSVE]], <[[#div(VBITS,8)]] x i8>* [[AGG_RESULT:%.*]], align 16, [[TBAA6]]
+// CHECK-NEXT:    ret void
 vec_int8 f2(vec_int8 x) { return svasrd_x(svptrue_b8(), x, 1); }
 #endif
 
@@ -90,24 +78,24 @@
 typedef svint8_t vec2 __attribute__((arm_sve_vector_bits(N)));
 
 // CHECK128-LABEL: define void @g(<vscale x 16 x i8> %x.coerce)
-// CHECK128-NEXT: entry:
-// CHECK128-NEXT:  %x = alloca <16 x i8>, align 16
-// CHECK128-NEXT:      %0 = bitcast <16 x i8>* %x to <vscale x 16 x i8>*
-// CHECK128-NEXT:    store <vscale x 16 x i8> %x.coerce, <vscale x 16 x i8>* %0, align 16
-// CHECK128-NEXT:    %x1 = load <16 x i8>, <16 x i8>* %x, align 16,
-// CHECK128-NEXT:    call void @f3(<16 x i8> %x1) #4
-// CHECK128-NEXT:      ret void
+// CHECK128-NEXT:  entry:
+// CHECK128-NEXT:    [[X:%.*]] = alloca <16 x i8>, align 16
+// CHECK128-NEXT:    [[TMP0:%.*]] = bitcast <16 x i8>* [[X]] to <vscale x 16 x i8>*
+// CHECK128-NEXT:    store <vscale x 16 x i8> [[X_COERCE:%.*]], <vscale x 16 x i8>* [[TMP0]], align 16
+// CHECK128-NEXT:    [[X1:%.*]] = load <16 x i8>, <16 x i8>* [[X]], align 16, [[TBAA6:!tbaa !.*]]
+// CHECK128-NEXT:    call void @f3(<16 x i8> [[X1]]) [[ATTR5:#.*]]
+// CHECK128-NEXT:    ret void
 
 // CHECK-LABEL: define void @g(<vscale x 16 x i8> %x.coerce)
-// CHECK-NEXT: entry:
-// CHECK-NEXT:   %x = alloca <[[#div(VBITS,8)]] x i8>, align 16
-// CHECK-NEXT:   %indirect-arg-temp = alloca <[[#div(VBITS,8)]] x i8>, align 16
-// CHECK-NEXT:   %0 = bitcast <[[#div(VBITS,8)]] x i8>* %x to <vscale x 16 x i8>*
-// CHECK-NEXT:   store <vscale x 16 x i8> %x.coerce, <vscale x 16 x i8>* %0
-// CHECK-NEXT:   %x1 = load <[[#div(VBITS,8)]] x i8>, <[[#div(VBITS,8)]] x i8>* %x, align 16
-// CHECK-NEXT:   store <[[#div(VBITS,8)]] x i8> %x1, <[[#div(VBITS,8)]] x i8>* %indirect-arg-temp
-// CHECK-NEXT:   call void @f3(<[[#div(VBITS,8)]] x i8>* nonnull %indirect-arg-temp)
-// CHECK-NEXT:   ret void
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[X:%.*]] = alloca <[[#div(VBITS,8)]] x i8>, align 16
+// CHECK-NEXT:    [[INDIRECT_ARG_TEMP:%.*]] = alloca <[[#div(VBITS,8)]] x i8>, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <[[#div(VBITS,8)]] x i8>* [[X]] to <vscale x 16 x i8>*
+// CHECK-NEXT:    store <vscale x 16 x i8> [[X_COERCE:%.*]], <vscale x 16 x i8>* [[TMP0]], align 16
+// CHECK-NEXT:    [[X1:%.*]] = load <[[#div(VBITS,8)]] x i8>, <[[#div(VBITS,8)]] x i8>* [[X]], align 16, [[TBAA6]]
+// CHECK-NEXT:    store <[[#div(VBITS,8)]] x i8> [[X1]], <[[#div(VBITS,8)]] x i8>* [[INDIRECT_ARG_TEMP]], align 16, [[TBAA6]]
+// CHECK-NEXT:    call void @f3(<[[#div(VBITS,8)]] x i8>* nonnull [[INDIRECT_ARG_TEMP]]) [[ATTR5:#.*]]
+// CHECK-NEXT:    ret void
 
 // CHECK128-LABEL: declare void @f3(<16 x i8>)
 
Index: clang/lib/CodeGen/CGExprScalar.cpp
===================================================================
--- clang/lib/CodeGen/CGExprScalar.cpp
+++ clang/lib/CodeGen/CGExprScalar.cpp
@@ -1996,6 +1996,32 @@
       }
     }
 
+    // If Src is a fixed vector and Dst is a scalable vector, and both have the
+    // same element type, use the llvm.experimental.vector.insert intrinsic to
+    // perform the bitcast.
+    if (const auto *FixedSrc = dyn_cast<llvm::FixedVectorType>(SrcTy)) {
+      if (const auto *ScalableDst = dyn_cast<llvm::ScalableVectorType>(DstTy)) {
+        if (FixedSrc->getElementType() == ScalableDst->getElementType()) {
+          llvm::Value *UndefVec = llvm::UndefValue::get(DstTy);
+          llvm::Value *Zero = llvm::Constant::getNullValue(CGF.CGM.Int64Ty);
+          return Builder.CreateInsertVector(DstTy, UndefVec, Src, Zero,
+                                            "castScalableSve");
+        }
+      }
+    }
+
+    // If Src is a scalable vector and Dst is a fixed vector, and both have the
+    // same element type, use the llvm.experimental.vector.extract intrinsic to
+    // perform the bitcast.
+    if (const auto *ScalableSrc = dyn_cast<llvm::ScalableVectorType>(SrcTy)) {
+      if (const auto *FixedDst = dyn_cast<llvm::FixedVectorType>(DstTy)) {
+        if (ScalableSrc->getElementType() == FixedDst->getElementType()) {
+          llvm::Value *Zero = llvm::Constant::getNullValue(CGF.CGM.Int64Ty);
+          return Builder.CreateExtractVector(DstTy, Src, Zero, "castFixedSve");
+        }
+      }
+    }
+
     // Perform VLAT <-> VLST bitcast through memory.
     if ((isa<llvm::FixedVectorType>(SrcTy) &&
          isa<llvm::ScalableVectorType>(DstTy)) ||

_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D92761: [clang][AArch64][SVE] Avoid going through memory for VLAT <-> VLST casts

Reply via email to