clang/test/CodeGen/AArch64/sve-intrinsics/acle_sve_ld1sh.c

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
   2 // REQUIRES: aarch64-registered-target
   3 // RUN: %clang_cc1 -triple aarch64 -target-feature +sve -disable-O0-optnone -Werror -emit-llvm -o - %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s
   4 // RUN: %clang_cc1 -triple aarch64 -target-feature +sve -disable-O0-optnone -Werror -emit-llvm -o - -x c++ %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
   5 // RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -disable-O0-optnone -Werror -emit-llvm -o - %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s
   6 // RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -disable-O0-optnone -Werror -emit-llvm -o - -x c++ %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
   7 // RUN: %clang_cc1 -triple aarch64 -target-feature +sve -S -disable-O0-optnone -Werror -o /dev/null %s
   8 // RUN: %clang_cc1 -triple aarch64 -target-feature +sme -S -disable-O0-optnone -Werror -o /dev/null %s
   9
  10 #include <arm_sve.h>
  11
  12 #if defined __ARM_FEATURE_SME
  13 #define MODE_ATTR __arm_streaming
  14 #else
  15 #define MODE_ATTR
  16 #endif
  17
  18 #ifdef SVE_OVERLOADED_FORMS
  19 // A simple used,unused... macro, long enough to represent any SVE builtin.
  20 #define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED) A1##A3
  21 #else
  22 #define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
  23 #endif
  24
  25 // CHECK-LABEL: @test_svld1sh_s32(
  26 // CHECK-NEXT:  entry:
  27 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
  28 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
  29 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
  30 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
  31 //
  32 // CPP-CHECK-LABEL: @_Z16test_svld1sh_s32u10__SVBool_tPKs(
  33 // CPP-CHECK-NEXT:  entry:
  34 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
  35 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
  36 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
  37 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
  38 //
  39 svint32_t test_svld1sh_s32(svbool_t pg, const int16_t *base) MODE_ATTR
  40 {
  41   return svld1sh_s32(pg, base);
  42 }
  43
  44 // CHECK-LABEL: @test_svld1sh_s64(
  45 // CHECK-NEXT:  entry:
  46 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
  47 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
  48 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
  49 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
  50 //
  51 // CPP-CHECK-LABEL: @_Z16test_svld1sh_s64u10__SVBool_tPKs(
  52 // CPP-CHECK-NEXT:  entry:
  53 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
  54 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
  55 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
  56 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
  57 //
  58 svint64_t test_svld1sh_s64(svbool_t pg, const int16_t *base) MODE_ATTR
  59 {
  60   return svld1sh_s64(pg, base);
  61 }
  62
  63 // CHECK-LABEL: @test_svld1sh_u32(
  64 // CHECK-NEXT:  entry:
  65 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
  66 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
  67 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
  68 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
  69 //
  70 // CPP-CHECK-LABEL: @_Z16test_svld1sh_u32u10__SVBool_tPKs(
  71 // CPP-CHECK-NEXT:  entry:
  72 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
  73 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
  74 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
  75 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
  76 //
  77 svuint32_t test_svld1sh_u32(svbool_t pg, const int16_t *base) MODE_ATTR
  78 {
  79   return svld1sh_u32(pg, base);
  80 }
  81
  82 // CHECK-LABEL: @test_svld1sh_u64(
  83 // CHECK-NEXT:  entry:
  84 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
  85 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
  86 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
  87 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
  88 //
  89 // CPP-CHECK-LABEL: @_Z16test_svld1sh_u64u10__SVBool_tPKs(
  90 // CPP-CHECK-NEXT:  entry:
  91 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
  92 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[BASE:%.*]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
  93 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
  94 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
  95 //
  96 svuint64_t test_svld1sh_u64(svbool_t pg, const int16_t *base) MODE_ATTR
  97 {
  98   return svld1sh_u64(pg, base);
  99 }
 100
 101 // CHECK-LABEL: @test_svld1sh_vnum_s32(
 102 // CHECK-NEXT:  entry:
 103 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 104 // CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 105 // CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
 106 // CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 107 // CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 108 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
 109 // CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
 110 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP5]]
 111 //
 112 // CPP-CHECK-LABEL: @_Z21test_svld1sh_vnum_s32u10__SVBool_tPKsl(
 113 // CPP-CHECK-NEXT:  entry:
 114 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 115 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 116 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
 117 // CPP-CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 118 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 119 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
 120 // CPP-CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
 121 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP5]]
 122 //
 123 svint32_t test_svld1sh_vnum_s32(svbool_t pg, const int16_t *base, int64_t vnum) MODE_ATTR
 124 {
 125   return svld1sh_vnum_s32(pg, base, vnum);
 126 }
 127
 128 // CHECK-LABEL: @test_svld1sh_vnum_s64(
 129 // CHECK-NEXT:  entry:
 130 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 131 // CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 132 // CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
 133 // CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 134 // CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 135 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
 136 // CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
 137 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP5]]
 138 //
 139 // CPP-CHECK-LABEL: @_Z21test_svld1sh_vnum_s64u10__SVBool_tPKsl(
 140 // CPP-CHECK-NEXT:  entry:
 141 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 142 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 143 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
 144 // CPP-CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 145 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 146 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
 147 // CPP-CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
 148 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP5]]
 149 //
 150 svint64_t test_svld1sh_vnum_s64(svbool_t pg, const int16_t *base, int64_t vnum) MODE_ATTR
 151 {
 152   return svld1sh_vnum_s64(pg, base, vnum);
 153 }
 154
 155 // CHECK-LABEL: @test_svld1sh_vnum_u32(
 156 // CHECK-NEXT:  entry:
 157 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 158 // CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 159 // CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
 160 // CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 161 // CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 162 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
 163 // CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
 164 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP5]]
 165 //
 166 // CPP-CHECK-LABEL: @_Z21test_svld1sh_vnum_u32u10__SVBool_tPKsl(
 167 // CPP-CHECK-NEXT:  entry:
 168 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 169 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 170 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
 171 // CPP-CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 172 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 173 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i16> @llvm.masked.load.nxv4i16.p0(ptr [[TMP3]], i32 1, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i16> zeroinitializer)
 174 // CPP-CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 4 x i16> [[TMP4]] to <vscale x 4 x i32>
 175 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP5]]
 176 //
 177 svuint32_t test_svld1sh_vnum_u32(svbool_t pg, const int16_t *base, int64_t vnum) MODE_ATTR
 178 {
 179   return svld1sh_vnum_u32(pg, base, vnum);
 180 }
 181
 182 // CHECK-LABEL: @test_svld1sh_vnum_u64(
 183 // CHECK-NEXT:  entry:
 184 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 185 // CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 186 // CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
 187 // CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 188 // CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 189 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
 190 // CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
 191 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP5]]
 192 //
 193 // CPP-CHECK-LABEL: @_Z21test_svld1sh_vnum_u64u10__SVBool_tPKsl(
 194 // CPP-CHECK-NEXT:  entry:
 195 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 196 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.vscale.i64()
 197 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
 198 // CPP-CHECK-NEXT:    [[DOTIDX:%.*]] = mul i64 [[VNUM:%.*]], [[TMP2]]
 199 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[DOTIDX]]
 200 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i16> @llvm.masked.load.nxv2i16.p0(ptr [[TMP3]], i32 1, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i16> zeroinitializer)
 201 // CPP-CHECK-NEXT:    [[TMP5:%.*]] = sext <vscale x 2 x i16> [[TMP4]] to <vscale x 2 x i64>
 202 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP5]]
 203 //
 204 svuint64_t test_svld1sh_vnum_u64(svbool_t pg, const int16_t *base, int64_t vnum) MODE_ATTR
 205 {
 206   return svld1sh_vnum_u64(pg, base, vnum);
 207 }
 208
 209 #ifndef __ARM_FEATURE_SME
 210
 211 // CHECK-LABEL: @test_svld1sh_gather_u32base_s32(
 212 // CHECK-NEXT:  entry:
 213 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 214 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
 215 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 216 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 217 //
 218 // CPP-CHECK-LABEL: @_Z31test_svld1sh_gather_u32base_s32u10__SVBool_tu12__SVUint32_t(
 219 // CPP-CHECK-NEXT:  entry:
 220 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 221 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
 222 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 223 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 224 //
 225 svint32_t test_svld1sh_gather_u32base_s32(svbool_t pg, svuint32_t bases) {
 226   return SVE_ACLE_FUNC(svld1sh_gather, _u32base, _s32, )(pg, bases);
 227 }
 228
 229 // CHECK-LABEL: @test_svld1sh_gather_u64base_s64(
 230 // CHECK-NEXT:  entry:
 231 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 232 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
 233 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 234 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 235 //
 236 // CPP-CHECK-LABEL: @_Z31test_svld1sh_gather_u64base_s64u10__SVBool_tu12__SVUint64_t(
 237 // CPP-CHECK-NEXT:  entry:
 238 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 239 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
 240 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 241 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 242 //
 243 svint64_t test_svld1sh_gather_u64base_s64(svbool_t pg, svuint64_t bases) {
 244   return SVE_ACLE_FUNC(svld1sh_gather, _u64base, _s64, )(pg, bases);
 245 }
 246
 247 // CHECK-LABEL: @test_svld1sh_gather_u32base_u32(
 248 // CHECK-NEXT:  entry:
 249 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 250 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
 251 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 252 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 253 //
 254 // CPP-CHECK-LABEL: @_Z31test_svld1sh_gather_u32base_u32u10__SVBool_tu12__SVUint32_t(
 255 // CPP-CHECK-NEXT:  entry:
 256 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 257 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 0)
 258 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 259 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 260 //
 261 svuint32_t test_svld1sh_gather_u32base_u32(svbool_t pg, svuint32_t bases) {
 262   return SVE_ACLE_FUNC(svld1sh_gather, _u32base, _u32, )(pg, bases);
 263 }
 264
 265 // CHECK-LABEL: @test_svld1sh_gather_u64base_u64(
 266 // CHECK-NEXT:  entry:
 267 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 268 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
 269 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 270 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 271 //
 272 // CPP-CHECK-LABEL: @_Z31test_svld1sh_gather_u64base_u64u10__SVBool_tu12__SVUint64_t(
 273 // CPP-CHECK-NEXT:  entry:
 274 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 275 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 0)
 276 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 277 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 278 //
 279 svuint64_t test_svld1sh_gather_u64base_u64(svbool_t pg, svuint64_t bases) {
 280   return SVE_ACLE_FUNC(svld1sh_gather, _u64base, _u64, )(pg, bases);
 281 }
 282
 283 // CHECK-LABEL: @test_svld1sh_gather_s32offset_s32(
 284 // CHECK-NEXT:  entry:
 285 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 286 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 287 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 288 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 289 //
 290 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_s32offset_s32u10__SVBool_tPKsu11__SVInt32_t(
 291 // CPP-CHECK-NEXT:  entry:
 292 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 293 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 294 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 295 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 296 //
 297 svint32_t test_svld1sh_gather_s32offset_s32(svbool_t pg, const int16_t *base, svint32_t offsets) {
 298   return SVE_ACLE_FUNC(svld1sh_gather_, s32, offset_s32, )(pg, base, offsets);
 299 }
 300
 301 // CHECK-LABEL: @test_svld1sh_gather_s64offset_s64(
 302 // CHECK-NEXT:  entry:
 303 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 304 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 305 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 306 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 307 //
 308 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_s64offset_s64u10__SVBool_tPKsu11__SVInt64_t(
 309 // CPP-CHECK-NEXT:  entry:
 310 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 311 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 312 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 313 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 314 //
 315 svint64_t test_svld1sh_gather_s64offset_s64(svbool_t pg, const int16_t *base, svint64_t offsets) {
 316   return SVE_ACLE_FUNC(svld1sh_gather_, s64, offset_s64, )(pg, base, offsets);
 317 }
 318
 319 // CHECK-LABEL: @test_svld1sh_gather_s32offset_u32(
 320 // CHECK-NEXT:  entry:
 321 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 322 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 323 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 324 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 325 //
 326 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_s32offset_u32u10__SVBool_tPKsu11__SVInt32_t(
 327 // CPP-CHECK-NEXT:  entry:
 328 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 329 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 330 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 331 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 332 //
 333 svuint32_t test_svld1sh_gather_s32offset_u32(svbool_t pg, const int16_t *base, svint32_t offsets) {
 334   return SVE_ACLE_FUNC(svld1sh_gather_, s32, offset_u32, )(pg, base, offsets);
 335 }
 336
 337 // CHECK-LABEL: @test_svld1sh_gather_s64offset_u64(
 338 // CHECK-NEXT:  entry:
 339 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 340 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 341 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 342 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 343 //
 344 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_s64offset_u64u10__SVBool_tPKsu11__SVInt64_t(
 345 // CPP-CHECK-NEXT:  entry:
 346 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 347 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 348 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 349 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 350 //
 351 svuint64_t test_svld1sh_gather_s64offset_u64(svbool_t pg, const int16_t *base, svint64_t offsets) {
 352   return SVE_ACLE_FUNC(svld1sh_gather_, s64, offset_u64, )(pg, base, offsets);
 353 }
 354
 355 // CHECK-LABEL: @test_svld1sh_gather_u32offset_s32(
 356 // CHECK-NEXT:  entry:
 357 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 358 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 359 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 360 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 361 //
 362 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_u32offset_s32u10__SVBool_tPKsu12__SVUint32_t(
 363 // CPP-CHECK-NEXT:  entry:
 364 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 365 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 366 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 367 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 368 //
 369 svint32_t test_svld1sh_gather_u32offset_s32(svbool_t pg, const int16_t *base, svuint32_t offsets) {
 370   return SVE_ACLE_FUNC(svld1sh_gather_, u32, offset_s32, )(pg, base, offsets);
 371 }
 372
 373 // CHECK-LABEL: @test_svld1sh_gather_u64offset_s64(
 374 // CHECK-NEXT:  entry:
 375 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 376 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 377 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 378 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 379 //
 380 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_u64offset_s64u10__SVBool_tPKsu12__SVUint64_t(
 381 // CPP-CHECK-NEXT:  entry:
 382 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 383 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 384 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 385 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 386 //
 387 svint64_t test_svld1sh_gather_u64offset_s64(svbool_t pg, const int16_t *base, svuint64_t offsets) {
 388   return SVE_ACLE_FUNC(svld1sh_gather_, u64, offset_s64, )(pg, base, offsets);
 389 }
 390
 391 // CHECK-LABEL: @test_svld1sh_gather_u32offset_u32(
 392 // CHECK-NEXT:  entry:
 393 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 394 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 395 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 396 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 397 //
 398 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_u32offset_u32u10__SVBool_tPKsu12__SVUint32_t(
 399 // CPP-CHECK-NEXT:  entry:
 400 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 401 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[OFFSETS:%.*]])
 402 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 403 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 404 //
 405 svuint32_t test_svld1sh_gather_u32offset_u32(svbool_t pg, const int16_t *base, svuint32_t offsets) {
 406   return SVE_ACLE_FUNC(svld1sh_gather_, u32, offset_u32, )(pg, base, offsets);
 407 }
 408
 409 // CHECK-LABEL: @test_svld1sh_gather_u64offset_u64(
 410 // CHECK-NEXT:  entry:
 411 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 412 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 413 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 414 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 415 //
 416 // CPP-CHECK-LABEL: @_Z33test_svld1sh_gather_u64offset_u64u10__SVBool_tPKsu12__SVUint64_t(
 417 // CPP-CHECK-NEXT:  entry:
 418 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 419 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[OFFSETS:%.*]])
 420 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 421 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 422 //
 423 svuint64_t test_svld1sh_gather_u64offset_u64(svbool_t pg, const int16_t *base, svuint64_t offsets) {
 424   return SVE_ACLE_FUNC(svld1sh_gather_, u64, offset_u64, )(pg, base, offsets);
 425 }
 426
 427 // CHECK-LABEL: @test_svld1sh_gather_u32base_offset_s32(
 428 // CHECK-NEXT:  entry:
 429 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 430 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 431 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 432 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 433 //
 434 // CPP-CHECK-LABEL: @_Z38test_svld1sh_gather_u32base_offset_s32u10__SVBool_tu12__SVUint32_tl(
 435 // CPP-CHECK-NEXT:  entry:
 436 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 437 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 438 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 439 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 440 //
 441 svint32_t test_svld1sh_gather_u32base_offset_s32(svbool_t pg, svuint32_t bases, int64_t offset) {
 442   return SVE_ACLE_FUNC(svld1sh_gather, _u32base, _offset_s32, )(pg, bases, offset);
 443 }
 444
 445 // CHECK-LABEL: @test_svld1sh_gather_u64base_offset_s64(
 446 // CHECK-NEXT:  entry:
 447 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 448 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 449 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 450 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 451 //
 452 // CPP-CHECK-LABEL: @_Z38test_svld1sh_gather_u64base_offset_s64u10__SVBool_tu12__SVUint64_tl(
 453 // CPP-CHECK-NEXT:  entry:
 454 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 455 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 456 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 457 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 458 //
 459 svint64_t test_svld1sh_gather_u64base_offset_s64(svbool_t pg, svuint64_t bases, int64_t offset) {
 460   return SVE_ACLE_FUNC(svld1sh_gather, _u64base, _offset_s64, )(pg, bases, offset);
 461 }
 462
 463 // CHECK-LABEL: @test_svld1sh_gather_u32base_offset_u32(
 464 // CHECK-NEXT:  entry:
 465 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 466 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 467 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 468 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 469 //
 470 // CPP-CHECK-LABEL: @_Z38test_svld1sh_gather_u32base_offset_u32u10__SVBool_tu12__SVUint32_tl(
 471 // CPP-CHECK-NEXT:  entry:
 472 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 473 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 474 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 475 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 476 //
 477 svuint32_t test_svld1sh_gather_u32base_offset_u32(svbool_t pg, svuint32_t bases, int64_t offset) {
 478   return SVE_ACLE_FUNC(svld1sh_gather, _u32base, _offset_u32, )(pg, bases, offset);
 479 }
 480
 481 // CHECK-LABEL: @test_svld1sh_gather_u64base_offset_u64(
 482 // CHECK-NEXT:  entry:
 483 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 484 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 485 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 486 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 487 //
 488 // CPP-CHECK-LABEL: @_Z38test_svld1sh_gather_u64base_offset_u64u10__SVBool_tu12__SVUint64_tl(
 489 // CPP-CHECK-NEXT:  entry:
 490 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 491 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[OFFSET:%.*]])
 492 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 493 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 494 //
 495 svuint64_t test_svld1sh_gather_u64base_offset_u64(svbool_t pg, svuint64_t bases, int64_t offset) {
 496   return SVE_ACLE_FUNC(svld1sh_gather, _u64base, _offset_u64, )(pg, bases, offset);
 497 }
 498
 499 // CHECK-LABEL: @test_svld1sh_gather_s32index_s32(
 500 // CHECK-NEXT:  entry:
 501 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 502 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 503 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 504 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 505 //
 506 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_s32index_s32u10__SVBool_tPKsu11__SVInt32_t(
 507 // CPP-CHECK-NEXT:  entry:
 508 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 509 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 510 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 511 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 512 //
 513 svint32_t test_svld1sh_gather_s32index_s32(svbool_t pg, const int16_t *base, svint32_t indices) {
 514   return SVE_ACLE_FUNC(svld1sh_gather_, s32, index_s32, )(pg, base, indices);
 515 }
 516
 517 // CHECK-LABEL: @test_svld1sh_gather_s64index_s64(
 518 // CHECK-NEXT:  entry:
 519 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 520 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 521 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 522 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 523 //
 524 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_s64index_s64u10__SVBool_tPKsu11__SVInt64_t(
 525 // CPP-CHECK-NEXT:  entry:
 526 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 527 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 528 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 529 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 530 //
 531 svint64_t test_svld1sh_gather_s64index_s64(svbool_t pg, const int16_t *base, svint64_t indices) {
 532   return SVE_ACLE_FUNC(svld1sh_gather_, s64, index_s64, )(pg, base, indices);
 533 }
 534
 535 // CHECK-LABEL: @test_svld1sh_gather_s32index_u32(
 536 // CHECK-NEXT:  entry:
 537 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 538 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 539 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 540 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 541 //
 542 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_s32index_u32u10__SVBool_tPKsu11__SVInt32_t(
 543 // CPP-CHECK-NEXT:  entry:
 544 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 545 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.sxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 546 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 547 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 548 //
 549 svuint32_t test_svld1sh_gather_s32index_u32(svbool_t pg, const int16_t *base, svint32_t indices) {
 550   return SVE_ACLE_FUNC(svld1sh_gather_, s32, index_u32, )(pg, base, indices);
 551 }
 552
 553 // CHECK-LABEL: @test_svld1sh_gather_s64index_u64(
 554 // CHECK-NEXT:  entry:
 555 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 556 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 557 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 558 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 559 //
 560 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_s64index_u64u10__SVBool_tPKsu11__SVInt64_t(
 561 // CPP-CHECK-NEXT:  entry:
 562 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 563 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 564 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 565 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 566 //
 567 svuint64_t test_svld1sh_gather_s64index_u64(svbool_t pg, const int16_t *base, svint64_t indices) {
 568   return SVE_ACLE_FUNC(svld1sh_gather_, s64, index_u64, )(pg, base, indices);
 569 }
 570
 571 // CHECK-LABEL: @test_svld1sh_gather_u32index_s32(
 572 // CHECK-NEXT:  entry:
 573 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 574 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 575 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 576 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 577 //
 578 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_u32index_s32u10__SVBool_tPKsu12__SVUint32_t(
 579 // CPP-CHECK-NEXT:  entry:
 580 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 581 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 582 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 583 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 584 //
 585 svint32_t test_svld1sh_gather_u32index_s32(svbool_t pg, const int16_t *base, svuint32_t indices) {
 586   return SVE_ACLE_FUNC(svld1sh_gather_, u32, index_s32, )(pg, base, indices);
 587 }
 588
 589 // CHECK-LABEL: @test_svld1sh_gather_u64index_s64(
 590 // CHECK-NEXT:  entry:
 591 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 592 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 593 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 594 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 595 //
 596 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_u64index_s64u10__SVBool_tPKsu12__SVUint64_t(
 597 // CPP-CHECK-NEXT:  entry:
 598 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 599 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 600 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 601 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 602 //
 603 svint64_t test_svld1sh_gather_u64index_s64(svbool_t pg, const int16_t *base, svuint64_t indices) {
 604   return SVE_ACLE_FUNC(svld1sh_gather_, u64, index_s64, )(pg, base, indices);
 605 }
 606
 607 // CHECK-LABEL: @test_svld1sh_gather_u32index_u32(
 608 // CHECK-NEXT:  entry:
 609 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 610 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 611 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 612 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 613 //
 614 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_u32index_u32u10__SVBool_tPKsu12__SVUint32_t(
 615 // CPP-CHECK-NEXT:  entry:
 616 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 617 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.uxtw.index.nxv4i16(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 4 x i32> [[INDICES:%.*]])
 618 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 4 x i16> [[TMP1]] to <vscale x 4 x i32>
 619 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 620 //
 621 svuint32_t test_svld1sh_gather_u32index_u32(svbool_t pg, const int16_t *base, svuint32_t indices) {
 622   return SVE_ACLE_FUNC(svld1sh_gather_, u32, index_u32, )(pg, base, indices);
 623 }
 624
 625 // CHECK-LABEL: @test_svld1sh_gather_u64index_u64(
 626 // CHECK-NEXT:  entry:
 627 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 628 // CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 629 // CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 630 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 631 //
 632 // CPP-CHECK-LABEL: @_Z32test_svld1sh_gather_u64index_u64u10__SVBool_tPKsu12__SVUint64_t(
 633 // CPP-CHECK-NEXT:  entry:
 634 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 635 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.index.nxv2i16(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]], <vscale x 2 x i64> [[INDICES:%.*]])
 636 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = sext <vscale x 2 x i16> [[TMP1]] to <vscale x 2 x i64>
 637 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 638 //
 639 svuint64_t test_svld1sh_gather_u64index_u64(svbool_t pg, const int16_t *base, svuint64_t indices) {
 640   return SVE_ACLE_FUNC(svld1sh_gather_, u64, index_u64, )(pg, base, indices);
 641 }
 642
 643 // CHECK-LABEL: @test_svld1sh_gather_u32base_index_s32(
 644 // CHECK-NEXT:  entry:
 645 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 646 // CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 647 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP1]])
 648 // CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 4 x i16> [[TMP2]] to <vscale x 4 x i32>
 649 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 650 //
 651 // CPP-CHECK-LABEL: @_Z37test_svld1sh_gather_u32base_index_s32u10__SVBool_tu12__SVUint32_tl(
 652 // CPP-CHECK-NEXT:  entry:
 653 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 654 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 655 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP1]])
 656 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 4 x i16> [[TMP2]] to <vscale x 4 x i32>
 657 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 658 //
 659 svint32_t test_svld1sh_gather_u32base_index_s32(svbool_t pg, svuint32_t bases, int64_t index) {
 660   return SVE_ACLE_FUNC(svld1sh_gather, _u32base, _index_s32, )(pg, bases, index);
 661 }
 662
 663 // CHECK-LABEL: @test_svld1sh_gather_u64base_index_s64(
 664 // CHECK-NEXT:  entry:
 665 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 666 // CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 667 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP1]])
 668 // CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 2 x i16> [[TMP2]] to <vscale x 2 x i64>
 669 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 670 //
 671 // CPP-CHECK-LABEL: @_Z37test_svld1sh_gather_u64base_index_s64u10__SVBool_tu12__SVUint64_tl(
 672 // CPP-CHECK-NEXT:  entry:
 673 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 674 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 675 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP1]])
 676 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 2 x i16> [[TMP2]] to <vscale x 2 x i64>
 677 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 678 //
 679 svint64_t test_svld1sh_gather_u64base_index_s64(svbool_t pg, svuint64_t bases, int64_t index) {
 680   return SVE_ACLE_FUNC(svld1sh_gather, _u64base, _index_s64, )(pg, bases, index);
 681 }
 682
 683 // CHECK-LABEL: @test_svld1sh_gather_u32base_index_u32(
 684 // CHECK-NEXT:  entry:
 685 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 686 // CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 687 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP1]])
 688 // CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 4 x i16> [[TMP2]] to <vscale x 4 x i32>
 689 // CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 690 //
 691 // CPP-CHECK-LABEL: @_Z37test_svld1sh_gather_u32base_index_u32u10__SVBool_tu12__SVUint32_tl(
 692 // CPP-CHECK-NEXT:  entry:
 693 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 694 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 695 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x i32> [[BASES:%.*]], i64 [[TMP1]])
 696 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 4 x i16> [[TMP2]] to <vscale x 4 x i32>
 697 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 698 //
 699 svuint32_t test_svld1sh_gather_u32base_index_u32(svbool_t pg, svuint32_t bases, int64_t index) {
 700   return SVE_ACLE_FUNC(svld1sh_gather, _u32base, _index_u32, )(pg, bases, index);
 701 }
 702
 703 // CHECK-LABEL: @test_svld1sh_gather_u64base_index_u64(
 704 // CHECK-NEXT:  entry:
 705 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 706 // CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 707 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP1]])
 708 // CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 2 x i16> [[TMP2]] to <vscale x 2 x i64>
 709 // CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 710 //
 711 // CPP-CHECK-LABEL: @_Z37test_svld1sh_gather_u64base_index_u64u10__SVBool_tu12__SVUint64_tl(
 712 // CPP-CHECK-NEXT:  entry:
 713 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 714 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[INDEX:%.*]], 1
 715 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> [[TMP0]], <vscale x 2 x i64> [[BASES:%.*]], i64 [[TMP1]])
 716 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = sext <vscale x 2 x i16> [[TMP2]] to <vscale x 2 x i64>
 717 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 718 //
 719 svuint64_t test_svld1sh_gather_u64base_index_u64(svbool_t pg, svuint64_t bases, int64_t index) {
 720   return SVE_ACLE_FUNC(svld1sh_gather, _u64base, _index_u64, )(pg, bases, index);
 721 }
 722
 723 #endif