[SveEmitter] Add builtins for scatter stores

[sdesmalen]

Summary

D77735 only added scatters for the non-temporal variants.

[efriedma]

LGTM

clang/include/clang/Basic/arm_sve.td

 def SVST1_VNUM    : MInst<"svst1_vnum[_{d}]",  "vPpld", "csilUcUsUiUlhfd", [IsStore], MemEltTyDefault, "aarch64_sve_st1">;
 def SVST1B_VNUM_S : MInst<"svst1b_vnum[_{d}]", "vPAld", "sil",             [IsStore], MemEltTyInt8,    "aarch64_sve_st1">;
 def SVST1B_VNUM_U : MInst<"svst1b_vnum[_{d}]", "vPEld", "UsUiUl",          [IsStore], MemEltTyInt8,    "aarch64_sve_st1">;
 def SVST1H_VNUM_S : MInst<"svst1h_vnum[_{d}]", "vPBld", "il",              [IsStore], MemEltTyInt16,   "aarch64_sve_st1">;
 def SVST1H_VNUM_U : MInst<"svst1h_vnum[_{d}]", "vPFld", "UiUl",            [IsStore], MemEltTyInt16,   "aarch64_sve_st1">;
 def SVST1W_VNUM_S : MInst<"svst1w_vnum[_{d}]", "vPCld", "l",               [IsStore], MemEltTyInt32,   "aarch64_sve_st1">;
 def SVST1W_VNUM_U : MInst<"svst1w_vnum[_{d}]", "vPGld", "Ul",              [IsStore], MemEltTyInt32,   "aarch64_sve_st1">;
 
+// Store one vector (vector base)
+def SVST1_SCATTER_BASES_U     : MInst<"svst1_scatter[_{2}base_{d}]",  "vPud",  "ilUiUlfd", [IsScatterStore], MemEltTyDefault, "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1B_SCATTER_BASES_U    : MInst<"svst1b_scatter[_{2}base_{d}]", "vPud",  "ilUiUl",   [IsScatterStore], MemEltTyInt8,    "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1H_SCATTER_BASES_U    : MInst<"svst1h_scatter[_{2}base_{d}]", "vPud",  "ilUiUl",   [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1W_SCATTER_BASES_U    : MInst<"svst1w_scatter[_{2}base_{d}]", "vPud",  "lUl",      [IsScatterStore], MemEltTyInt32,   "aarch64_sve_st1_scatter_scalar_offset">;
+
+// Store one vector (scalar base, signed vector offset in bytes)
+def SVST1_SCATTER_64B_OFFSETS_S   : MInst<"svst1_scatter_[{3}]offset[_{d}]",  "vPpxd", "lUld", [IsScatterStore, IsByteIndexed], MemEltTyDefault, "aarch64_sve_st1_scatter">;
+def SVST1B_SCATTER_64B_OFFSETS_SS : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPAxd", "l",    [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter">;
+def SVST1B_SCATTER_64B_OFFSETS_SU : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPExd", "Ul",   [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter">;
+def SVST1H_SCATTER_64B_OFFSETS_SS : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPBxd", "l",    [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter">;
+def SVST1H_SCATTER_64B_OFFSETS_SU : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPFxd", "Ul",   [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter">;
+def SVST1W_SCATTER_64B_OFFSETS_SS : MInst<"svst1w_scatter_[{3}]offset[_{d}]", "vPCxd", "l",    [IsScatterStore, IsByteIndexed], MemEltTyInt32,   "aarch64_sve_st1_scatter">;
+def SVST1W_SCATTER_64B_OFFSETS_SU : MInst<"svst1w_scatter_[{3}]offset[_{d}]", "vPGxd", "Ul",   [IsScatterStore, IsByteIndexed], MemEltTyInt32,   "aarch64_sve_st1_scatter">;
+
+def SVST1_SCATTER_32B_OFFSETS_S   : MInst<"svst1_scatter_[{3}]offset[_{d}]",  "vPpxd", "iUif", [IsScatterStore, IsByteIndexed], MemEltTyDefault, "aarch64_sve_st1_scatter_sxtw">;
+def SVST1B_SCATTER_32B_OFFSETS_SS : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPAxd", "i",    [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter_sxtw">;
+def SVST1B_SCATTER_32B_OFFSETS_SU : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPExd", "Ui",   [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter_sxtw">;
+def SVST1H_SCATTER_32B_OFFSETS_SS : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPBxd", "i",    [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter_sxtw">;
+def SVST1H_SCATTER_32B_OFFSETS_SU : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPFxd", "Ui",   [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter_sxtw">;
+
+// Store one vector (scalar base, unsigned vector offset in bytes)
+def SVST1_SCATTER_64B_OFFSETS_U   : MInst<"svst1_scatter_[{3}]offset[_{d}]",  "vPpud", "lUld", [IsScatterStore, IsByteIndexed], MemEltTyDefault, "aarch64_sve_st1_scatter">;
+def SVST1B_SCATTER_64B_OFFSETS_US : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPAud", "l",    [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter">;
+def SVST1B_SCATTER_64B_OFFSETS_UU : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPEud", "Ul",   [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter">;
+def SVST1H_SCATTER_64B_OFFSETS_US : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPBud", "l",    [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter">;
+def SVST1H_SCATTER_64B_OFFSETS_UU : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPFud", "Ul",   [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter">;
+def SVST1W_SCATTER_64B_OFFSETS_US : MInst<"svst1w_scatter_[{3}]offset[_{d}]", "vPCud", "l",    [IsScatterStore, IsByteIndexed], MemEltTyInt32,   "aarch64_sve_st1_scatter">;
+def SVST1W_SCATTER_64B_OFFSETS_UU : MInst<"svst1w_scatter_[{3}]offset[_{d}]", "vPGud", "Ul",   [IsScatterStore, IsByteIndexed], MemEltTyInt32,   "aarch64_sve_st1_scatter">;
+
+def SVST1_SCATTER_32B_OFFSETS_U   : MInst<"svst1_scatter_[{3}]offset[_{d}]",  "vPpud", "iUif", [IsScatterStore, IsByteIndexed], MemEltTyDefault, "aarch64_sve_st1_scatter_uxtw">;
+def SVST1B_SCATTER_32B_OFFSETS_US : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPAud", "i",    [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter_uxtw">;
+def SVST1B_SCATTER_32B_OFFSETS_UU : MInst<"svst1b_scatter_[{3}]offset[_{d}]", "vPEud", "Ui",   [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter_uxtw">;
+def SVST1H_SCATTER_32B_OFFSETS_US : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPBud", "i",    [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter_uxtw">;
+def SVST1H_SCATTER_32B_OFFSETS_UU : MInst<"svst1h_scatter_[{3}]offset[_{d}]", "vPFud", "Ui",   [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter_uxtw">;
+
+// Store one vector (vector base, signed scalar offset in bytes)
+def SVST1_SCATTER_OFFSET_S    : MInst<"svst1_scatter[_{2}base]_offset[_{d}]",  "vPuld", "ilUiUlfd", [IsScatterStore, IsByteIndexed], MemEltTyDefault, "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1B_SCATTER_OFFSET_S   : MInst<"svst1b_scatter[_{2}base]_offset[_{d}]", "vPuld", "ilUiUl",   [IsScatterStore, IsByteIndexed], MemEltTyInt8,    "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1H_SCATTER_OFFSET_S   : MInst<"svst1h_scatter[_{2}base]_offset[_{d}]", "vPuld", "ilUiUl",   [IsScatterStore, IsByteIndexed], MemEltTyInt16,   "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1W_SCATTER_OFFSET_S   : MInst<"svst1w_scatter[_{2}base]_offset[_{d}]", "vPuld", "lUl",      [IsScatterStore, IsByteIndexed], MemEltTyInt32,   "aarch64_sve_st1_scatter_scalar_offset">;
+
+// Store one vector (scalar base, signed vector index)
+def SVST1_SCATTER_64B_INDICES_S   : MInst<"svst1_scatter_[{3}]index[_{d}]",  "vPpxd", "lUld", [IsScatterStore], MemEltTyDefault, "aarch64_sve_st1_scatter_index">;
+def SVST1H_SCATTER_64B_INDICES_SS : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPBxd", "l",    [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_index">;
+def SVST1H_SCATTER_64B_INDICES_SU : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPFxd", "Ul",   [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_index">;
+def SVST1W_SCATTER_64B_INDICES_SS : MInst<"svst1w_scatter_[{3}]index[_{d}]", "vPCxd", "l",    [IsScatterStore], MemEltTyInt32,   "aarch64_sve_st1_scatter_index">;
+def SVST1W_SCATTER_64B_INDICES_SU : MInst<"svst1w_scatter_[{3}]index[_{d}]", "vPGxd", "Ul",   [IsScatterStore], MemEltTyInt32,   "aarch64_sve_st1_scatter_index">;
+
+def SVST1_SCATTER_32B_INDICES_S   : MInst<"svst1_scatter_[{3}]index[_{d}]",  "vPpxd", "iUif", [IsScatterStore], MemEltTyDefault, "aarch64_sve_st1_scatter_sxtw_index">;
+def SVST1H_SCATTER_32B_INDICES_SS : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPBxd", "i",    [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_sxtw_index">;
+def SVST1H_SCATTER_32B_INDICES_SU : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPFxd", "Ui",   [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_sxtw_index">;
+
+// Store one vector (scalar base, unsigned vector index)
+def SVST1_SCATTER_64B_INDICES_U   : MInst<"svst1_scatter_[{3}]index[_{d}]",  "vPpud", "lUld", [IsScatterStore], MemEltTyDefault, "aarch64_sve_st1_scatter_index">;
+def SVST1H_SCATTER_64B_INDICES_US : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPBud", "l",    [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_index">;
+def SVST1H_SCATTER_64B_INDICES_UU : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPFud", "Ul",   [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_index">;
+def SVST1W_SCATTER_64B_INDICES_US : MInst<"svst1w_scatter_[{3}]index[_{d}]", "vPCud", "l",    [IsScatterStore], MemEltTyInt32,   "aarch64_sve_st1_scatter_index">;
+def SVST1W_SCATTER_64B_INDICES_UU : MInst<"svst1w_scatter_[{3}]index[_{d}]", "vPGud", "Ul",   [IsScatterStore], MemEltTyInt32,   "aarch64_sve_st1_scatter_index">;
+
+def SVST1_SCATTER_32B_INDICES_U   : MInst<"svst1_scatter_[{3}]index[_{d}]",  "vPpud", "iUif", [IsScatterStore], MemEltTyDefault, "aarch64_sve_st1_scatter_uxtw_index">;
+def SVST1H_SCATTER_32B_INDICES_US : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPBud", "i",    [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_uxtw_index">;
+def SVST1H_SCATTER_32B_INDICES_UU : MInst<"svst1h_scatter_[{3}]index[_{d}]", "vPFud", "Ui",   [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_uxtw_index">;
+
+// Store one vector (vector base, signed scalar index)
+def SVST1_SCATTER_INDEX_S     : MInst<"svst1_scatter[_{2}base]_index[_{d}]",  "vPuld", "ilUiUlfd", [IsScatterStore], MemEltTyDefault, "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1H_SCATTER_INDEX_S    : MInst<"svst1h_scatter[_{2}base]_index[_{d}]", "vPuld", "ilUiUl",   [IsScatterStore], MemEltTyInt16,   "aarch64_sve_st1_scatter_scalar_offset">;
+def SVST1W_SCATTER_INDEX_S    : MInst<"svst1w_scatter[_{2}base]_index[_{d}]", "vPuld", "lUl",      [IsScatterStore], MemEltTyInt32,   "aarch64_sve_st1_scatter_scalar_offset">;
+
 // Store one vector, with no truncation, non-temporal (scalar base)
 def SVSTNT1 : MInst<"svstnt1[_{d}]", "vPpd", "csilUcUsUiUlhfd", [IsStore], MemEltTyDefault, "aarch64_sve_stnt1">;
 
 // Store one vector, with no truncation, non-temporal (scalar base, VL displacement)
 def SVSTNT1_VNUM : MInst<"svstnt1_vnum[_{d}]", "vPpld", "csilUcUsUiUlhfd", [IsStore], MemEltTyDefault, "aarch64_sve_stnt1">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Prefetches

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st1.c

 {
   // CHECK-LABEL: test_svst1_vnum_f64
   // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
   // CHECK-DAG: %[[BASE:.*]] = bitcast double* %base to <vscale x 2 x double>*
   // CHECK-DAG: %[[GEP:.*]] = getelementptr <vscale x 2 x double>, <vscale x 2 x double>* %[[BASE]], i64 %vnum, i64 0
   // CHECK: call void @llvm.aarch64.sve.st1.nxv2f64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], double* %[[GEP]])
   return SVE_ACLE_FUNC(svst1_vnum,_f64,,)(pg, base, vnum, data);
 }
+
+void test_svst1_scatter_u32base_s32(svbool_t pg, svuint32_t bases, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_s32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,,_s32)(pg, bases, data);
+}
+
+void test_svst1_scatter_u64base_s64(svbool_t pg, svuint64_t bases, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_s64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,,_s64)(pg, bases, data);
+}
+
+void test_svst1_scatter_u32base_u32(svbool_t pg, svuint32_t bases, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_u32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,,_u32)(pg, bases, data);
+}
+
+void test_svst1_scatter_u64base_u64(svbool_t pg, svuint64_t bases, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_u64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,,_u64)(pg, bases, data);
+}
+
+void test_svst1_scatter_u32base_f32(svbool_t pg, svuint32_t bases, svfloat32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_f32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x float> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,,_f32)(pg, bases, data);
+}
+
+void test_svst1_scatter_u64base_f64(svbool_t pg, svuint64_t bases, svfloat64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_f64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,,_f64)(pg, bases, data);
+}
+
+void test_svst1_scatter_s32offset_s32(svbool_t pg, int32_t *base, svint32_t offsets, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s32offset_s32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s32,offset,_s32)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_s64offset_s64(svbool_t pg, int64_t *base, svint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s64offset_s64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], i64* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_s32offset_u32(svbool_t pg, uint32_t *base, svint32_t offsets, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s32offset_u32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s32,offset,_u32)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_s64offset_u64(svbool_t pg, uint64_t *base, svint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s64offset_u64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], i64* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_s32offset_f32(svbool_t pg, float32_t *base, svint32_t offsets, svfloat32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s32offset_f32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4f32(<vscale x 4 x float> %data, <vscale x 4 x i1> %[[PG]], float* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s32,offset,_f32)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_s64offset_f64(svbool_t pg, float64_t *base, svint64_t offsets, svfloat64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s64offset_f64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2f64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], double* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s64,offset,_f64)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_u32offset_s32(svbool_t pg, int32_t *base, svuint32_t offsets, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32offset_s32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u32,offset,_s32)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_u64offset_s64(svbool_t pg, int64_t *base, svuint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64offset_s64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], i64* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_u32offset_u32(svbool_t pg, uint32_t *base, svuint32_t offsets, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32offset_u32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u32,offset,_u32)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_u64offset_u64(svbool_t pg, uint64_t *base, svuint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64offset_u64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], i64* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_u32offset_f32(svbool_t pg, float32_t *base, svuint32_t offsets, svfloat32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32offset_f32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4f32(<vscale x 4 x float> %data, <vscale x 4 x i1> %[[PG]], float* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u32,offset,_f32)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_u64offset_f64(svbool_t pg, float64_t *base, svuint64_t offsets, svfloat64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64offset_f64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2f64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], double* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u64,offset,_f64)(pg, base, offsets, data);
+}
+
+void test_svst1_scatter_u32base_offset_s32(svbool_t pg, svuint32_t bases, int64_t offset, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_offset_s32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,_offset,_s32)(pg, bases, offset, data);
+}
+
+void test_svst1_scatter_u64base_offset_s64(svbool_t pg, svuint64_t bases, int64_t offset, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_offset_s64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,_offset,_s64)(pg, bases, offset, data);
+}
+
+void test_svst1_scatter_u32base_offset_u32(svbool_t pg, svuint32_t bases, int64_t offset, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_offset_u32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,_offset,_u32)(pg, bases, offset, data);
+}
+
+void test_svst1_scatter_u64base_offset_u64(svbool_t pg, svuint64_t bases, int64_t offset, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_offset_u64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,_offset,_u64)(pg, bases, offset, data);
+}
+
+void test_svst1_scatter_u32base_offset_f32(svbool_t pg, svuint32_t bases, int64_t offset, svfloat32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_offset_f32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x float> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,_offset,_f32)(pg, bases, offset, data);
+}
+
+void test_svst1_scatter_u64base_offset_f64(svbool_t pg, svuint64_t bases, int64_t offset, svfloat64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_offset_f64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,_offset,_f64)(pg, bases, offset, data);
+}
+
+void test_svst1_scatter_s32index_s32(svbool_t pg, int32_t *base, svint32_t indices, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s32index_s32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s32,index,_s32)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_s64index_s64(svbool_t pg, int64_t *base, svint64_t indices, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s64index_s64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], i64* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s64,index,_s64)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_s32index_u32(svbool_t pg, uint32_t *base, svint32_t indices, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s32index_u32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s32,index,_u32)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_s64index_u64(svbool_t pg, uint64_t *base, svint64_t indices, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s64index_u64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], i64* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s64,index,_u64)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_s32index_f32(svbool_t pg, float32_t *base, svint32_t indices, svfloat32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s32index_f32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4f32(<vscale x 4 x float> %data, <vscale x 4 x i1> %[[PG]], float* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s32,index,_f32)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_s64index_f64(svbool_t pg, float64_t *base, svint64_t indices, svfloat64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_s64index_f64
+  // CHECK: %[[PG:.*]]  = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2f64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], double* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,s64,index,_f64)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_u32index_s32(svbool_t pg, int32_t *base, svuint32_t indices, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32index_s32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u32,index,_s32)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_u64index_s64(svbool_t pg, int64_t *base, svuint64_t indices, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64index_s64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %0, i64* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u64,index,_s64)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_u32index_u32(svbool_t pg, uint32_t *base, svuint32_t indices, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32index_u32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], i32* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u32,index,_u32)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_u64index_u64(svbool_t pg, uint64_t *base, svuint64_t indices, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64index_u64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], i64* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u64,index,_u64)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_u32index_f32(svbool_t pg, float32_t *base, svuint32_t indices, svfloat32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32index_f32
+  // CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4f32(<vscale x 4 x float> %data, <vscale x 4 x i1> %[[PG]], float* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u32,index,_f32)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_u64index_f64(svbool_t pg, float64_t *base, svuint64_t indices, svfloat64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64index_f64
+  // CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2f64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], double* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter_,u64,index,_f64)(pg, base, indices, data);
+}
+
+void test_svst1_scatter_u32base_index_s32(svbool_t pg, svuint32_t bases, int64_t index, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_index_s32
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 2
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,_index,_s32)(pg, bases, index, data);
+}
+
+void test_svst1_scatter_u64base_index_s64(svbool_t pg, svuint64_t bases, int64_t index, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_index_s64
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 3
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,_index,_s64)(pg, bases, index, data);
+}
+
+void test_svst1_scatter_u32base_index_u32(svbool_t pg, svuint32_t bases, int64_t index, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_index_u32
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 2
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,_index,_u32)(pg, bases, index, data);
+}
+
+void test_svst1_scatter_u64base_index_u64(svbool_t pg, svuint64_t bases, int64_t index, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_index_u64
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 3
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,_index,_u64)(pg, bases, index, data);
+}
+
+void test_svst1_scatter_u32base_index_f32(svbool_t pg, svuint32_t bases, int64_t index, svfloat32_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u32base_index_f32
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 2
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x float> %data, <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u32base,_index,_f32)(pg, bases, index, data);
+}
+
+void test_svst1_scatter_u64base_index_f64(svbool_t pg, svuint64_t bases, int64_t index, svfloat64_t data)
+{
+  // CHECK-LABEL: test_svst1_scatter_u64base_index_f64
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 3
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x double> %data, <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1_scatter,_u64base,_index,_f64)(pg, bases, index, data);
+}

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st1b.c

void test_svst1b_vnum_u64(svbool_t pg, uint8_t *base, int64_t vnum, svuint64_t data)
   // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
   // CHECK-DAG: %[[BASE:.*]] = bitcast i8* %base to <vscale x 2 x i8>*
   // CHECK-DAG: %[[DATA:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
   // CHECK-DAG: %[[GEP:.*]] = getelementptr <vscale x 2 x i8>, <vscale x 2 x i8>* %[[BASE]], i64 %vnum, i64 0
   // CHECK: call void @llvm.aarch64.sve.st1.nxv2i8(<vscale x 2 x i8> %[[DATA]], <vscale x 2 x i1> %[[PG]], i8* %[[GEP]])
   // CHECK: ret void
   return SVE_ACLE_FUNC(svst1b_vnum,_u64,,)(pg, base, vnum, data);
 }
+
+void test_svst1b_scatter_u32base_s32(svbool_t pg, svuint32_t bases, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u32base_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i8> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u32base,,_s32)(pg, bases, data);
+}
+
+void test_svst1b_scatter_u64base_s64(svbool_t pg, svuint64_t bases, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u64base_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u64base,,_s64)(pg, bases, data);
+}
+
+void test_svst1b_scatter_u32base_u32(svbool_t pg, svuint32_t bases, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u32base_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i8> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u32base,,_u32)(pg, bases, data);
+}
+
+void test_svst1b_scatter_u64base_u64(svbool_t pg, svuint64_t bases, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u64base_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u64base,,_u64)(pg, bases, data);
+}
+
+void test_svst1b_scatter_s32offset_s32(svbool_t pg, int8_t *base, svint32_t offsets, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_s32offset_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i8(<vscale x 4 x i8> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i8* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,s32,offset,_s32)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_s64offset_s64(svbool_t pg, int8_t *base, svint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_s64offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i8(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i8* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,s64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_s32offset_u32(svbool_t pg, uint8_t *base, svint32_t offsets, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_s32offset_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i8(<vscale x 4 x i8> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i8* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,s32,offset,_u32)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_s64offset_u64(svbool_t pg, uint8_t *base, svint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_s64offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i8(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i8* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,s64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_u32offset_s32(svbool_t pg, int8_t *base, svuint32_t offsets, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u32offset_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i8(<vscale x 4 x i8> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i8* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,u32,offset,_s32)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_u64offset_s64(svbool_t pg, int8_t *base, svuint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u64offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i8(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i8* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,u64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_u32offset_u32(svbool_t pg, uint8_t *base, svuint32_t offsets, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u32offset_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i8(<vscale x 4 x i8> %0, <vscale x 4 x i1> %1, i8* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,u32,offset,_u32)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_u64offset_u64(svbool_t pg, uint8_t *base, svuint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u64offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i8(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i8* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter_,u64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1b_scatter_u32base_offset_s32(svbool_t pg, svuint32_t bases, int64_t offset, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u32base_offset_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i8> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u32base,_offset,_s32)(pg, bases, offset, data);
+}
+
+void test_svst1b_scatter_u64base_offset_s64(svbool_t pg, svuint64_t bases, int64_t offset, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u64base_offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u64base,_offset,_s64)(pg, bases, offset, data);
+}
+
+void test_svst1b_scatter_u32base_offset_u32(svbool_t pg, svuint32_t bases, int64_t offset, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u32base_offset_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i8> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u32base,_offset,_u32)(pg, bases, offset, data);
+}
+
+void test_svst1b_scatter_u64base_offset_u64(svbool_t pg, svuint64_t bases, int64_t offset, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1b_scatter_u64base_offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i8> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1b_scatter,_u64base,_offset,_u64)(pg, bases, offset, data);
+}

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st1h.c

void test_svst1h_vnum_u64(svbool_t pg, uint16_t *base, int64_t vnum, svuint64_t data)
   // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
   // CHECK-DAG: %[[BASE:.*]] = bitcast i16* %base to <vscale x 2 x i16>*
   // CHECK-DAG: %[[DATA:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
   // CHECK-DAG: %[[GEP:.*]] = getelementptr <vscale x 2 x i16>, <vscale x 2 x i16>* %[[BASE]], i64 %vnum, i64 0
   // CHECK: call void @llvm.aarch64.sve.st1.nxv2i16(<vscale x 2 x i16> %[[DATA]], <vscale x 2 x i1> %[[PG]], i16* %[[GEP]])
   // CHECK: ret void
   return SVE_ACLE_FUNC(svst1h_vnum,_u64,,)(pg, base, vnum, data);
 }
+
+void test_svst1h_scatter_u32base_s32(svbool_t pg, svuint32_t bases, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32base_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u32base,,_s32)(pg, bases, data);
+}
+
+void test_svst1h_scatter_u64base_s64(svbool_t pg, svuint64_t bases, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64base_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u64base,,_s64)(pg, bases, data);
+}
+
+void test_svst1h_scatter_u32base_u32(svbool_t pg, svuint32_t bases, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32base_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u32base,,_u32)(pg, bases, data);
+}
+
+void test_svst1h_scatter_u64base_u64(svbool_t pg, svuint64_t bases, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64base_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u64base,,_u64)(pg, bases, data);
+}
+
+void test_svst1h_scatter_s32offset_s32(svbool_t pg, int16_t *base, svint32_t offsets, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s32offset_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s32,offset,_s32)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_s64offset_s64(svbool_t pg, int16_t *base, svint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s64offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_s32offset_u32(svbool_t pg, uint16_t *base, svint32_t offsets, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s32offset_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s32,offset,_u32)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_s64offset_u64(svbool_t pg, uint16_t *base, svint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s64offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_u32offset_s32(svbool_t pg, int16_t *base, svuint32_t offsets, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32offset_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u32,offset,_s32)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_u64offset_s64(svbool_t pg, int16_t *base, svuint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_u32offset_u32(svbool_t pg, uint16_t *base, svuint32_t offsets, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32offset_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u32,offset,_u32)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_u64offset_u64(svbool_t pg, uint16_t *base, svuint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1h_scatter_u32base_offset_s32(svbool_t pg, svuint32_t bases, int64_t offset, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32base_offset_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u32base,_offset,_s32)(pg, bases, offset, data);
+}
+
+void test_svst1h_scatter_u64base_offset_s64(svbool_t pg, svuint64_t bases, int64_t offset, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64base_offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u64base,_offset,_s64)(pg, bases, offset, data);
+}
+
+void test_svst1h_scatter_u32base_offset_u32(svbool_t pg, svuint32_t bases, int64_t offset, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32base_offset_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u32base,_offset,_u32)(pg, bases, offset, data);
+}
+
+void test_svst1h_scatter_u64base_offset_u64(svbool_t pg, svuint64_t bases, int64_t offset, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64base_offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u64base,_offset,_u64)(pg, bases, offset, data);
+}
+
+void test_svst1h_scatter_s32index_s32(svbool_t pg, int16_t *base, svint32_t indices, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s32index_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s32,index,_s32)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_s64index_s64(svbool_t pg, int16_t *base, svint64_t indices, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s64index_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s64,index,_s64)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_s32index_u32(svbool_t pg, uint16_t *base, svint32_t indices, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s32index_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s32,index,_u32)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_s64index_u64(svbool_t pg, uint16_t *base, svint64_t indices, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_s64index_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,s64,index,_u64)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_u32index_s32(svbool_t pg, int16_t *base, svuint32_t indices, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32index_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u32,index,_s32)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_u64index_s64(svbool_t pg, int16_t *base, svuint64_t indices, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64index_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u64,index,_s64)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_u32index_u32(svbool_t pg, uint16_t *base, svuint32_t indices, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32index_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i16(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], i16* %base, <vscale x 4 x i32> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u32,index,_u32)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_u64index_u64(svbool_t pg, uint16_t *base, svuint64_t indices, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64index_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i16(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i16* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter_,u64,index,_u64)(pg, base, indices, data);
+}
+
+void test_svst1h_scatter_u32base_index_s32(svbool_t pg, svuint32_t bases, int64_t index, svint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32base_index_s32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]]  = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 1
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u32base,_index,_s32)(pg, bases, index, data);
+}
+
+void test_svst1h_scatter_u64base_index_s64(svbool_t pg, svuint64_t bases, int64_t index, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64base_index_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]]  = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 1
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u64base,_index,_s64)(pg, bases, index, data);
+}
+
+void test_svst1h_scatter_u32base_index_u32(svbool_t pg, svuint32_t bases, int64_t index, svuint32_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u32base_index_u32
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  // CHECK-DAG: %[[PG:.*]]  = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 1
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i16> %[[TRUNC]], <vscale x 4 x i1> %[[PG]], <vscale x 4 x i32> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u32base,_index,_u32)(pg, bases, index, data);
+}
+
+void test_svst1h_scatter_u64base_index_u64(svbool_t pg, svuint64_t bases, int64_t index, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1h_scatter_u64base_index_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  // CHECK-DAG: %[[PG:.*]]  = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 1
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i16> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1h_scatter,_u64base,_index,_u64)(pg, bases, index, data);
+}

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st1w.c

void test_svst1w_vnum_u64(svbool_t pg, uint32_t *base, int64_t vnum, svuint64_t data)
   // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
   // CHECK-DAG: %[[BASE:.*]] = bitcast i32* %base to <vscale x 2 x i32>*
   // CHECK-DAG: %[[DATA:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
   // CHECK-DAG: %[[GEP:.*]] = getelementptr <vscale x 2 x i32>, <vscale x 2 x i32>* %[[BASE]], i64 %vnum, i64 0
   // CHECK: call void @llvm.aarch64.sve.st1.nxv2i32(<vscale x 2 x i32> %[[DATA]], <vscale x 2 x i1> %[[PG]], i32* %[[GEP]])
   // CHECK: ret void
   return SVE_ACLE_FUNC(svst1w_vnum,_u64,,)(pg, base, vnum, data);
 }
+
+void test_svst1w_scatter_u64base_s64(svbool_t pg, svuint64_t bases, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64base_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter,_u64base,,_s64)(pg, bases, data);
+}
+
+void test_svst1w_scatter_u64base_u64(svbool_t pg, svuint64_t bases, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64base_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 0)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter,_u64base,,_u64)(pg, bases, data);
+}
+
+void test_svst1w_scatter_s64offset_s64(svbool_t pg, int32_t *base, svint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_s64offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i32(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i32* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,s64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1w_scatter_s64offset_u64(svbool_t pg, uint32_t *base, svint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_s64offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i32(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i32* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,s64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1w_scatter_u64offset_s64(svbool_t pg, int32_t *base, svuint64_t offsets, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i32(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i32* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,u64,offset,_s64)(pg, base, offsets, data);
+}
+
+void test_svst1w_scatter_u64offset_u64(svbool_t pg, uint32_t *base, svuint64_t offsets, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.nxv2i32(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i32* %base, <vscale x 2 x i64> %offsets)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,u64,offset,_u64)(pg, base, offsets, data);
+}
+
+void test_svst1w_scatter_u64base_offset_s64(svbool_t pg, svuint64_t bases, int64_t offset, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64base_offset_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter,_u64base,_offset,_s64)(pg, bases, offset, data);
+}
+
+void test_svst1w_scatter_u64base_offset_u64(svbool_t pg, svuint64_t bases, int64_t offset, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64base_offset_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %offset)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter,_u64base,_offset,_u64)(pg, bases, offset, data);
+}
+
+void test_svst1w_scatter_s64index_s64(svbool_t pg, int32_t *base, svint64_t indices, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_s64index_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i32(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i32* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,s64,index,_s64)(pg, base, indices, data);
+}
+
+void test_svst1w_scatter_s64index_u64(svbool_t pg, uint32_t *base, svint64_t indices, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_s64index_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i32(<vscale x 2 x i32> %0, <vscale x 2 x i1> %1, i32* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,s64,index,_u64)(pg, base, indices, data);
+}
+
+void test_svst1w_scatter_u64index_s64(svbool_t pg, int32_t *base, svuint64_t indices, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64index_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i32(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i32* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,u64,index,_s64)(pg, base, indices, data);
+}
+
+void test_svst1w_scatter_u64index_u64(svbool_t pg, uint32_t *base, svuint64_t indices, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64index_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.index.nxv2i32(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], i32* %base, <vscale x 2 x i64> %indices)
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter_,u64,index,_u64)(pg, base, indices, data);
+}
+
+void test_svst1w_scatter_u64base_index_s64(svbool_t pg, svuint64_t bases, int64_t index, svint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64base_index_s64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 2
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter,_u64base,_index,_s64)(pg, bases, index, data);
+}
+
+void test_svst1w_scatter_u64base_index_u64(svbool_t pg, svuint64_t bases, int64_t index, svuint64_t data)
+{
+  // CHECK-LABEL: test_svst1w_scatter_u64base_index_u64
+  // CHECK-DAG: %[[TRUNC:.*]] = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  // CHECK-DAG: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
+  // CHECK-DAG: %[[SHL:.*]] = shl i64 %index, 2
+  // CHECK: call void @llvm.aarch64.sve.st1.scatter.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i32> %[[TRUNC]], <vscale x 2 x i1> %[[PG]], <vscale x 2 x i64> %bases, i64 %[[SHL]])
+  // CHECK: ret void
+  return SVE_ACLE_FUNC(svst1w_scatter,_u64base,_index,_u64)(pg, bases, index, data);
+}