// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sve2 -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sve2 -S -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -S -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK

// REQUIRES: aarch64-registered-target

#include <arm_sve.h>

#ifdef SVE_OVERLOADED_FORMS
// A simple used,unused... macro, long enough to represent any SVE builtin.
#define SVE_ACLE_FUNC(A1, A2_UNUSED, A3, A4_UNUSED) A1##A3
#else
#define SVE_ACLE_FUNC(A1, A2, A3, A4) A1##A2##A3##A4
#endif

// CHECK-LABEL: @test_svstnt1h_scatter_u32base_s32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z33test_svstnt1h_scatter_u32base_s32u10__SVBool_tu12__SVUint32_tu11__SVInt32_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32base_s32(svbool_t pg, svuint32_t bases, svint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u32base, , _s32)(pg, bases, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64base_s64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z33test_svstnt1h_scatter_u64base_s64u10__SVBool_tu12__SVUint64_tu11__SVInt64_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64base_s64(svbool_t pg, svuint64_t bases, svint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u64base, , _s64)(pg, bases, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u32base_u32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z33test_svstnt1h_scatter_u32base_u32u10__SVBool_tu12__SVUint32_tS0_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32base_u32(svbool_t pg, svuint32_t bases, svuint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u32base, , _u32)(pg, bases, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64base_u64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z33test_svstnt1h_scatter_u64base_u64u10__SVBool_tu12__SVUint64_tS0_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64base_u64(svbool_t pg, svuint64_t bases, svuint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u64base, , _u64)(pg, bases, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_s64offset_s64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z35test_svstnt1h_scatter_s64offset_s64u10__SVBool_tPsu11__SVInt64_tS1_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_s64offset_s64(svbool_t pg, int16_t *base, svint64_t offsets, svint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, s64, offset, _s64)(pg, base, offsets, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_s64offset_u64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z35test_svstnt1h_scatter_s64offset_u64u10__SVBool_tPtu11__SVInt64_tu12__SVUint64_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_s64offset_u64(svbool_t pg, uint16_t *base, svint64_t offsets, svuint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, s64, offset, _u64)(pg, base, offsets, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u32offset_s32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.uxtw.nxv4i16(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z35test_svstnt1h_scatter_u32offset_s32u10__SVBool_tPsu12__SVUint32_tu11__SVInt32_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.uxtw.nxv4i16(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32offset_s32(svbool_t pg, int16_t *base, svuint32_t offsets, svint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, u32, offset, _s32)(pg, base, offsets, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64offset_s64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z35test_svstnt1h_scatter_u64offset_s64u10__SVBool_tPsu12__SVUint64_tu11__SVInt64_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64offset_s64(svbool_t pg, int16_t *base, svuint64_t offsets, svint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, u64, offset, _s64)(pg, base, offsets, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u32offset_u32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.uxtw.nxv4i16(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z35test_svstnt1h_scatter_u32offset_u32u10__SVBool_tPtu12__SVUint32_tS1_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.uxtw.nxv4i16(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32offset_u32(svbool_t pg, uint16_t *base, svuint32_t offsets, svuint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, u32, offset, _u32)(pg, base, offsets, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64offset_u64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z35test_svstnt1h_scatter_u64offset_u64u10__SVBool_tPtu12__SVUint64_tS1_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64offset_u64(svbool_t pg, uint16_t *base, svuint64_t offsets, svuint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, u64, offset, _u64)(pg, base, offsets, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u32base_offset_s32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z40test_svstnt1h_scatter_u32base_offset_s32u10__SVBool_tu12__SVUint32_tlu11__SVInt32_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32base_offset_s32(svbool_t pg, svuint32_t bases, int64_t offset, svint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u32base, _offset, _s32)(pg, bases, offset, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64base_offset_s64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z40test_svstnt1h_scatter_u64base_offset_s64u10__SVBool_tu12__SVUint64_tlu11__SVInt64_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64base_offset_s64(svbool_t pg, svuint64_t bases, int64_t offset, svint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u64base, _offset, _s64)(pg, bases, offset, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u32base_offset_u32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z40test_svstnt1h_scatter_u32base_offset_u32u10__SVBool_tu12__SVUint32_tlS0_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32base_offset_u32(svbool_t pg, svuint32_t bases, int64_t offset, svuint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u32base, _offset, _u32)(pg, bases, offset, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64base_offset_u64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z40test_svstnt1h_scatter_u64base_offset_u64u10__SVBool_tu12__SVUint64_tlS0_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64base_offset_u64(svbool_t pg, svuint64_t bases, int64_t offset, svuint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u64base, _offset, _u64)(pg, bases, offset, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_s64index_s64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z34test_svstnt1h_scatter_s64index_s64u10__SVBool_tPsu11__SVInt64_tS1_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_s64index_s64(svbool_t pg, int16_t *base, svint64_t indices, svint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, s64, index, _s64)(pg, base, indices, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_s64index_u64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z34test_svstnt1h_scatter_s64index_u64u10__SVBool_tPtu11__SVInt64_tu12__SVUint64_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_s64index_u64(svbool_t pg, uint16_t *base, svint64_t indices, svuint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, s64, index, _u64)(pg, base, indices, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64index_s64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z34test_svstnt1h_scatter_u64index_s64u10__SVBool_tPsu12__SVUint64_tu11__SVInt64_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64index_s64(svbool_t pg, int16_t *base, svuint64_t indices, svint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, u64, index, _s64)(pg, base, indices, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64index_u64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z34test_svstnt1h_scatter_u64index_u64u10__SVBool_tPtu12__SVUint64_tS1_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.index.nxv2i16(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64index_u64(svbool_t pg, uint16_t *base, svuint64_t indices, svuint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter_, u64, index, _u64)(pg, base, indices, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u32base_index_s32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP2]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z39test_svstnt1h_scatter_u32base_index_s32u10__SVBool_tu12__SVUint32_tlu11__SVInt32_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP2]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32base_index_s32(svbool_t pg, svuint32_t bases, int64_t index, svint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u32base, _index, _s32)(pg, bases, index, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64base_index_s64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP2]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z39test_svstnt1h_scatter_u64base_index_s64u10__SVBool_tu12__SVUint64_tlu11__SVInt64_t(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP2]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64base_index_s64(svbool_t pg, svuint64_t bases, int64_t index, svint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u64base, _index, _s64)(pg, bases, index, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u32base_index_u32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP2]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z39test_svstnt1h_scatter_u32base_index_u32u10__SVBool_tu12__SVUint32_tlS0_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 4 x i32> [[DATA:%.*]] to <vscale x 4 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP2]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u32base_index_u32(svbool_t pg, svuint32_t bases, int64_t index, svuint32_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u32base, _index, _u32)(pg, bases, index, data);
}

// CHECK-LABEL: @test_svstnt1h_scatter_u64base_index_u64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP2]])
// CHECK-NEXT:    ret void
//
// CPP-CHECK-LABEL: @_Z39test_svstnt1h_scatter_u64base_index_u64u10__SVBool_tu12__SVUint64_tlS0_(
// CPP-CHECK-NEXT:  entry:
// CPP-CHECK-NEXT:    [[TMP0:%.*]] = trunc <vscale x 2 x i64> [[DATA:%.*]] to <vscale x 2 x i16>
// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
// CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[INDEX:%.*]], 1
// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sve.stnt1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP2]])
// CPP-CHECK-NEXT:    ret void
//
void test_svstnt1h_scatter_u64base_index_u64(svbool_t pg, svuint64_t bases, int64_t index, svuint64_t data) {
  return SVE_ACLE_FUNC(svstnt1h_scatter, _u64base, _index, _u64)(pg, bases, index, data);
}