clang/test/CodeGen/RISCV/bfloat-abi.c

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
   2 // RUN: %clang_cc1 -triple riscv64 -emit-llvm %s -o - | FileCheck %s --check-prefixes=CHECK-RV64
   3 // RUN: %clang_cc1 -triple riscv32 -emit-llvm %s -o - | FileCheck %s --check-prefixes=CHECK-RV32
   4
   5 struct bfloat1 {
   6   __bf16 a;
   7 };
   8
   9 // CHECK-RV64-LABEL: define dso_local i64 @h1
  10 // CHECK-RV64-SAME: (bfloat noundef [[A:%.*]]) #[[ATTR0:[0-9]+]] {
  11 // CHECK-RV64-NEXT:  entry:
  12 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT1:%.*]], align 2
  13 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
  14 // CHECK-RV64-NEXT:    [[COERCE_DIVE_COERCE:%.*]] = alloca i64, align 8
  15 // CHECK-RV64-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
  16 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  17 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT1]], ptr [[RETVAL]], i32 0, i32 0
  18 // CHECK-RV64-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
  19 // CHECK-RV64-NEXT:    [[COERCE_DIVE:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT1]], ptr [[RETVAL]], i32 0, i32 0
  20 // CHECK-RV64-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[COERCE_DIVE_COERCE]], ptr align 2 [[COERCE_DIVE]], i64 2, i1 false)
  21 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load i64, ptr [[COERCE_DIVE_COERCE]], align 8
  22 // CHECK-RV64-NEXT:    ret i64 [[TMP1]]
  23 //
  24 // CHECK-RV32-LABEL: define dso_local i32 @h1
  25 // CHECK-RV32-SAME: (bfloat noundef [[A:%.*]]) #[[ATTR0:[0-9]+]] {
  26 // CHECK-RV32-NEXT:  entry:
  27 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT1:%.*]], align 2
  28 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
  29 // CHECK-RV32-NEXT:    [[COERCE_DIVE_COERCE:%.*]] = alloca i32, align 4
  30 // CHECK-RV32-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
  31 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  32 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT1]], ptr [[RETVAL]], i32 0, i32 0
  33 // CHECK-RV32-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
  34 // CHECK-RV32-NEXT:    [[COERCE_DIVE:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT1]], ptr [[RETVAL]], i32 0, i32 0
  35 // CHECK-RV32-NEXT:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[COERCE_DIVE_COERCE]], ptr align 2 [[COERCE_DIVE]], i32 2, i1 false)
  36 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load i32, ptr [[COERCE_DIVE_COERCE]], align 4
  37 // CHECK-RV32-NEXT:    ret i32 [[TMP1]]
  38 //
  39 struct bfloat1 h1(__bf16 a) {
  40   struct bfloat1 x;
  41   x.a = a;
  42   return x;
  43 }
  44
  45 struct bfloat2 {
  46   __bf16 a;
  47   __bf16 b;
  48 };
  49
  50 // CHECK-RV64-LABEL: define dso_local i64 @h2
  51 // CHECK-RV64-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]]) #[[ATTR0]] {
  52 // CHECK-RV64-NEXT:  entry:
  53 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT2:%.*]], align 2
  54 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
  55 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
  56 // CHECK-RV64-NEXT:    [[RETVAL_COERCE:%.*]] = alloca i64, align 8
  57 // CHECK-RV64-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
  58 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
  59 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  60 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2]], ptr [[RETVAL]], i32 0, i32 0
  61 // CHECK-RV64-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
  62 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  63 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2]], ptr [[RETVAL]], i32 0, i32 1
  64 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
  65 // CHECK-RV64-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[RETVAL_COERCE]], ptr align 2 [[RETVAL]], i64 4, i1 false)
  66 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load i64, ptr [[RETVAL_COERCE]], align 8
  67 // CHECK-RV64-NEXT:    ret i64 [[TMP2]]
  68 //
  69 // CHECK-RV32-LABEL: define dso_local i32 @h2
  70 // CHECK-RV32-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]]) #[[ATTR0]] {
  71 // CHECK-RV32-NEXT:  entry:
  72 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT2:%.*]], align 2
  73 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
  74 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
  75 // CHECK-RV32-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
  76 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
  77 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  78 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2]], ptr [[RETVAL]], i32 0, i32 0
  79 // CHECK-RV32-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
  80 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  81 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2]], ptr [[RETVAL]], i32 0, i32 1
  82 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
  83 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[RETVAL]], align 2
  84 // CHECK-RV32-NEXT:    ret i32 [[TMP2]]
  85 //
  86 struct bfloat2 h2(__bf16 a, __bf16 b) {
  87   struct bfloat2 x;
  88   x.a = a;
  89   x.b = b;
  90   return x;
  91 }
  92
  93 struct bfloat3 {
  94   __bf16 a;
  95   __bf16 b;
  96   __bf16 c;
  97 };
  98
  99 // CHECK-RV64-LABEL: define dso_local i64 @h3
 100 // CHECK-RV64-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]]) #[[ATTR0]] {
 101 // CHECK-RV64-NEXT:  entry:
 102 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT3:%.*]], align 2
 103 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 104 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 105 // CHECK-RV64-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 106 // CHECK-RV64-NEXT:    [[RETVAL_COERCE:%.*]] = alloca i64, align 8
 107 // CHECK-RV64-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 108 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 109 // CHECK-RV64-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 110 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 111 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT3]], ptr [[RETVAL]], i32 0, i32 0
 112 // CHECK-RV64-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
 113 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 114 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT3]], ptr [[RETVAL]], i32 0, i32 1
 115 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 116 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 117 // CHECK-RV64-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT3]], ptr [[RETVAL]], i32 0, i32 2
 118 // CHECK-RV64-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 119 // CHECK-RV64-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[RETVAL_COERCE]], ptr align 2 [[RETVAL]], i64 6, i1 false)
 120 // CHECK-RV64-NEXT:    [[TMP3:%.*]] = load i64, ptr [[RETVAL_COERCE]], align 8
 121 // CHECK-RV64-NEXT:    ret i64 [[TMP3]]
 122 //
 123 // CHECK-RV32-LABEL: define dso_local [2 x i32] @h3
 124 // CHECK-RV32-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]]) #[[ATTR0]] {
 125 // CHECK-RV32-NEXT:  entry:
 126 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT3:%.*]], align 2
 127 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 128 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 129 // CHECK-RV32-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 130 // CHECK-RV32-NEXT:    [[RETVAL_COERCE:%.*]] = alloca [2 x i32], align 4
 131 // CHECK-RV32-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 132 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 133 // CHECK-RV32-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 134 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 135 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT3]], ptr [[RETVAL]], i32 0, i32 0
 136 // CHECK-RV32-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
 137 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 138 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT3]], ptr [[RETVAL]], i32 0, i32 1
 139 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 140 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 141 // CHECK-RV32-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT3]], ptr [[RETVAL]], i32 0, i32 2
 142 // CHECK-RV32-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 143 // CHECK-RV32-NEXT:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[RETVAL_COERCE]], ptr align 2 [[RETVAL]], i32 6, i1 false)
 144 // CHECK-RV32-NEXT:    [[TMP3:%.*]] = load [2 x i32], ptr [[RETVAL_COERCE]], align 4
 145 // CHECK-RV32-NEXT:    ret [2 x i32] [[TMP3]]
 146 //
 147 struct bfloat3 h3(__bf16 a, __bf16 b, __bf16 c) {
 148   struct bfloat3 x;
 149   x.a = a;
 150   x.b = b;
 151   x.c = c;
 152   return x;
 153 }
 154
 155 struct bfloat4 {
 156   __bf16 a;
 157   __bf16 b;
 158   __bf16 c;
 159   __bf16 d;
 160 };
 161
 162 // CHECK-RV64-LABEL: define dso_local i64 @h4
 163 // CHECK-RV64-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0]] {
 164 // CHECK-RV64-NEXT:  entry:
 165 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT4:%.*]], align 2
 166 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 167 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 168 // CHECK-RV64-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 169 // CHECK-RV64-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 170 // CHECK-RV64-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 171 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 172 // CHECK-RV64-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 173 // CHECK-RV64-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 174 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 175 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 0
 176 // CHECK-RV64-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
 177 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 178 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 1
 179 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 180 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 181 // CHECK-RV64-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 2
 182 // CHECK-RV64-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 183 // CHECK-RV64-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 184 // CHECK-RV64-NEXT:    [[D4:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 3
 185 // CHECK-RV64-NEXT:    store bfloat [[TMP3]], ptr [[D4]], align 2
 186 // CHECK-RV64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[RETVAL]], align 2
 187 // CHECK-RV64-NEXT:    ret i64 [[TMP4]]
 188 //
 189 // CHECK-RV32-LABEL: define dso_local [2 x i32] @h4
 190 // CHECK-RV32-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0]] {
 191 // CHECK-RV32-NEXT:  entry:
 192 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT4:%.*]], align 2
 193 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 194 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 195 // CHECK-RV32-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 196 // CHECK-RV32-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 197 // CHECK-RV32-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 198 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 199 // CHECK-RV32-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 200 // CHECK-RV32-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 201 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 202 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 0
 203 // CHECK-RV32-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
 204 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 205 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 1
 206 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 207 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 208 // CHECK-RV32-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 2
 209 // CHECK-RV32-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 210 // CHECK-RV32-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 211 // CHECK-RV32-NEXT:    [[D4:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT4]], ptr [[RETVAL]], i32 0, i32 3
 212 // CHECK-RV32-NEXT:    store bfloat [[TMP3]], ptr [[D4]], align 2
 213 // CHECK-RV32-NEXT:    [[TMP4:%.*]] = load [2 x i32], ptr [[RETVAL]], align 2
 214 // CHECK-RV32-NEXT:    ret [2 x i32] [[TMP4]]
 215 //
 216 struct bfloat4 h4(__bf16 a, __bf16 b, __bf16 c, __bf16 d) {
 217   struct bfloat4 x;
 218   x.a = a;
 219   x.b = b;
 220   x.c = c;
 221   x.d = d;
 222   return x;
 223 }
 224
 225 struct floatbfloat {
 226   float a;
 227   __bf16 b;
 228 };
 229
 230 // CHECK-RV64-LABEL: define dso_local i64 @fh
 231 // CHECK-RV64-SAME: (float noundef [[A:%.*]], bfloat noundef [[B:%.*]]) #[[ATTR0]] {
 232 // CHECK-RV64-NEXT:  entry:
 233 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_FLOATBFLOAT:%.*]], align 4
 234 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca float, align 4
 235 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 236 // CHECK-RV64-NEXT:    store float [[A]], ptr [[A_ADDR]], align 4
 237 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 238 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load float, ptr [[A_ADDR]], align 4
 239 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT]], ptr [[RETVAL]], i32 0, i32 0
 240 // CHECK-RV64-NEXT:    store float [[TMP0]], ptr [[A1]], align 4
 241 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 242 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT]], ptr [[RETVAL]], i32 0, i32 1
 243 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 4
 244 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load i64, ptr [[RETVAL]], align 4
 245 // CHECK-RV64-NEXT:    ret i64 [[TMP2]]
 246 //
 247 // CHECK-RV32-LABEL: define dso_local [2 x i32] @fh
 248 // CHECK-RV32-SAME: (float noundef [[A:%.*]], bfloat noundef [[B:%.*]]) #[[ATTR0]] {
 249 // CHECK-RV32-NEXT:  entry:
 250 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_FLOATBFLOAT:%.*]], align 4
 251 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca float, align 4
 252 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 253 // CHECK-RV32-NEXT:    store float [[A]], ptr [[A_ADDR]], align 4
 254 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 255 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load float, ptr [[A_ADDR]], align 4
 256 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT]], ptr [[RETVAL]], i32 0, i32 0
 257 // CHECK-RV32-NEXT:    store float [[TMP0]], ptr [[A1]], align 4
 258 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 259 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT]], ptr [[RETVAL]], i32 0, i32 1
 260 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 4
 261 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load [2 x i32], ptr [[RETVAL]], align 4
 262 // CHECK-RV32-NEXT:    ret [2 x i32] [[TMP2]]
 263 //
 264 struct floatbfloat fh(float a, __bf16 b) {
 265   struct floatbfloat x;
 266   x.a = a;
 267   x.b = b;
 268   return x;
 269 }
 270
 271 struct floatbfloat2 {
 272   float a;
 273   __bf16 b;
 274   __bf16 c;
 275 };
 276
 277 // CHECK-RV64-LABEL: define dso_local i64 @fh2
 278 // CHECK-RV64-SAME: (float noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]]) #[[ATTR0]] {
 279 // CHECK-RV64-NEXT:  entry:
 280 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_FLOATBFLOAT2:%.*]], align 4
 281 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca float, align 4
 282 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 283 // CHECK-RV64-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 284 // CHECK-RV64-NEXT:    store float [[A]], ptr [[A_ADDR]], align 4
 285 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 286 // CHECK-RV64-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 287 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load float, ptr [[A_ADDR]], align 4
 288 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT2]], ptr [[RETVAL]], i32 0, i32 0
 289 // CHECK-RV64-NEXT:    store float [[TMP0]], ptr [[A1]], align 4
 290 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 291 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT2]], ptr [[RETVAL]], i32 0, i32 1
 292 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 4
 293 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 294 // CHECK-RV64-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT2]], ptr [[RETVAL]], i32 0, i32 2
 295 // CHECK-RV64-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 296 // CHECK-RV64-NEXT:    [[TMP3:%.*]] = load i64, ptr [[RETVAL]], align 4
 297 // CHECK-RV64-NEXT:    ret i64 [[TMP3]]
 298 //
 299 // CHECK-RV32-LABEL: define dso_local [2 x i32] @fh2
 300 // CHECK-RV32-SAME: (float noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]]) #[[ATTR0]] {
 301 // CHECK-RV32-NEXT:  entry:
 302 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_FLOATBFLOAT2:%.*]], align 4
 303 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca float, align 4
 304 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 305 // CHECK-RV32-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 306 // CHECK-RV32-NEXT:    store float [[A]], ptr [[A_ADDR]], align 4
 307 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 308 // CHECK-RV32-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 309 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load float, ptr [[A_ADDR]], align 4
 310 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT2]], ptr [[RETVAL]], i32 0, i32 0
 311 // CHECK-RV32-NEXT:    store float [[TMP0]], ptr [[A1]], align 4
 312 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 313 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT2]], ptr [[RETVAL]], i32 0, i32 1
 314 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 4
 315 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 316 // CHECK-RV32-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT2]], ptr [[RETVAL]], i32 0, i32 2
 317 // CHECK-RV32-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 318 // CHECK-RV32-NEXT:    [[TMP3:%.*]] = load [2 x i32], ptr [[RETVAL]], align 4
 319 // CHECK-RV32-NEXT:    ret [2 x i32] [[TMP3]]
 320 //
 321 struct floatbfloat2 fh2(float a, __bf16 b, __bf16 c) {
 322   struct floatbfloat2 x;
 323   x.a = a;
 324   x.b = b;
 325   x.c = c;
 326   return x;
 327 }
 328
 329 struct bfloatfloat {
 330   __bf16 a;
 331   float b;
 332 };
 333
 334 // CHECK-RV64-LABEL: define dso_local i64 @hf
 335 // CHECK-RV64-SAME: (bfloat noundef [[A:%.*]], float noundef [[B:%.*]]) #[[ATTR0]] {
 336 // CHECK-RV64-NEXT:  entry:
 337 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOATFLOAT:%.*]], align 4
 338 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 339 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca float, align 4
 340 // CHECK-RV64-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 341 // CHECK-RV64-NEXT:    store float [[B]], ptr [[B_ADDR]], align 4
 342 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 343 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOATFLOAT]], ptr [[RETVAL]], i32 0, i32 0
 344 // CHECK-RV64-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 4
 345 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load float, ptr [[B_ADDR]], align 4
 346 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOATFLOAT]], ptr [[RETVAL]], i32 0, i32 1
 347 // CHECK-RV64-NEXT:    store float [[TMP1]], ptr [[B2]], align 4
 348 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load i64, ptr [[RETVAL]], align 4
 349 // CHECK-RV64-NEXT:    ret i64 [[TMP2]]
 350 //
 351 // CHECK-RV32-LABEL: define dso_local [2 x i32] @hf
 352 // CHECK-RV32-SAME: (bfloat noundef [[A:%.*]], float noundef [[B:%.*]]) #[[ATTR0]] {
 353 // CHECK-RV32-NEXT:  entry:
 354 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOATFLOAT:%.*]], align 4
 355 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 356 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca float, align 4
 357 // CHECK-RV32-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 358 // CHECK-RV32-NEXT:    store float [[B]], ptr [[B_ADDR]], align 4
 359 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 360 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOATFLOAT]], ptr [[RETVAL]], i32 0, i32 0
 361 // CHECK-RV32-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 4
 362 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load float, ptr [[B_ADDR]], align 4
 363 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOATFLOAT]], ptr [[RETVAL]], i32 0, i32 1
 364 // CHECK-RV32-NEXT:    store float [[TMP1]], ptr [[B2]], align 4
 365 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load [2 x i32], ptr [[RETVAL]], align 4
 366 // CHECK-RV32-NEXT:    ret [2 x i32] [[TMP2]]
 367 //
 368 struct bfloatfloat hf(__bf16 a, float b) {
 369   struct bfloatfloat x;
 370   x.a = a;
 371   x.b = b;
 372   return x;
 373 }
 374
 375 struct bfloat2float {
 376   __bf16 a;
 377   __bf16 b;
 378   float c;
 379 };
 380
 381 // CHECK-RV64-LABEL: define dso_local i64 @h2f
 382 // CHECK-RV64-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], float noundef [[C:%.*]]) #[[ATTR0]] {
 383 // CHECK-RV64-NEXT:  entry:
 384 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT2FLOAT:%.*]], align 4
 385 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 386 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 387 // CHECK-RV64-NEXT:    [[C_ADDR:%.*]] = alloca float, align 4
 388 // CHECK-RV64-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 389 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 390 // CHECK-RV64-NEXT:    store float [[C]], ptr [[C_ADDR]], align 4
 391 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 392 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2FLOAT]], ptr [[RETVAL]], i32 0, i32 0
 393 // CHECK-RV64-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 4
 394 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 395 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2FLOAT]], ptr [[RETVAL]], i32 0, i32 1
 396 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 397 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load float, ptr [[C_ADDR]], align 4
 398 // CHECK-RV64-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2FLOAT]], ptr [[RETVAL]], i32 0, i32 2
 399 // CHECK-RV64-NEXT:    store float [[TMP2]], ptr [[C3]], align 4
 400 // CHECK-RV64-NEXT:    [[TMP3:%.*]] = load i64, ptr [[RETVAL]], align 4
 401 // CHECK-RV64-NEXT:    ret i64 [[TMP3]]
 402 //
 403 // CHECK-RV32-LABEL: define dso_local [2 x i32] @h2f
 404 // CHECK-RV32-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], float noundef [[C:%.*]]) #[[ATTR0]] {
 405 // CHECK-RV32-NEXT:  entry:
 406 // CHECK-RV32-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT2FLOAT:%.*]], align 4
 407 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 408 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 409 // CHECK-RV32-NEXT:    [[C_ADDR:%.*]] = alloca float, align 4
 410 // CHECK-RV32-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 411 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 412 // CHECK-RV32-NEXT:    store float [[C]], ptr [[C_ADDR]], align 4
 413 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 414 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2FLOAT]], ptr [[RETVAL]], i32 0, i32 0
 415 // CHECK-RV32-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 4
 416 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 417 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2FLOAT]], ptr [[RETVAL]], i32 0, i32 1
 418 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 419 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load float, ptr [[C_ADDR]], align 4
 420 // CHECK-RV32-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT2FLOAT]], ptr [[RETVAL]], i32 0, i32 2
 421 // CHECK-RV32-NEXT:    store float [[TMP2]], ptr [[C3]], align 4
 422 // CHECK-RV32-NEXT:    [[TMP3:%.*]] = load [2 x i32], ptr [[RETVAL]], align 4
 423 // CHECK-RV32-NEXT:    ret [2 x i32] [[TMP3]]
 424 //
 425 struct bfloat2float h2f(__bf16 a, __bf16 b, float c) {
 426   struct bfloat2float x;
 427   x.a = a;
 428   x.b = b;
 429   x.c = c;
 430   return x;
 431 }
 432
 433 struct floatbfloat3 {
 434   float a;
 435   __bf16 b;
 436   __bf16 c;
 437   __bf16 d;
 438 };
 439
 440 // CHECK-RV64-LABEL: define dso_local [2 x i64] @fh3
 441 // CHECK-RV64-SAME: (float noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0]] {
 442 // CHECK-RV64-NEXT:  entry:
 443 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_FLOATBFLOAT3:%.*]], align 4
 444 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca float, align 4
 445 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 446 // CHECK-RV64-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 447 // CHECK-RV64-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 448 // CHECK-RV64-NEXT:    [[RETVAL_COERCE:%.*]] = alloca [2 x i64], align 8
 449 // CHECK-RV64-NEXT:    store float [[A]], ptr [[A_ADDR]], align 4
 450 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 451 // CHECK-RV64-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 452 // CHECK-RV64-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 453 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load float, ptr [[A_ADDR]], align 4
 454 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[RETVAL]], i32 0, i32 0
 455 // CHECK-RV64-NEXT:    store float [[TMP0]], ptr [[A1]], align 4
 456 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 457 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[RETVAL]], i32 0, i32 1
 458 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 4
 459 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 460 // CHECK-RV64-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[RETVAL]], i32 0, i32 2
 461 // CHECK-RV64-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 462 // CHECK-RV64-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 463 // CHECK-RV64-NEXT:    [[D4:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[RETVAL]], i32 0, i32 3
 464 // CHECK-RV64-NEXT:    store bfloat [[TMP3]], ptr [[D4]], align 4
 465 // CHECK-RV64-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[RETVAL_COERCE]], ptr align 4 [[RETVAL]], i64 12, i1 false)
 466 // CHECK-RV64-NEXT:    [[TMP4:%.*]] = load [2 x i64], ptr [[RETVAL_COERCE]], align 8
 467 // CHECK-RV64-NEXT:    ret [2 x i64] [[TMP4]]
 468 //
 469 // CHECK-RV32-LABEL: define dso_local void @fh3
 470 // CHECK-RV32-SAME: (ptr noalias sret([[STRUCT_FLOATBFLOAT3:%.*]]) align 4 [[AGG_RESULT:%.*]], float noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0]] {
 471 // CHECK-RV32-NEXT:  entry:
 472 // CHECK-RV32-NEXT:    [[RESULT_PTR:%.*]] = alloca ptr, align 4
 473 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca float, align 4
 474 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 475 // CHECK-RV32-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 476 // CHECK-RV32-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 477 // CHECK-RV32-NEXT:    store ptr [[AGG_RESULT]], ptr [[RESULT_PTR]], align 4
 478 // CHECK-RV32-NEXT:    store float [[A]], ptr [[A_ADDR]], align 4
 479 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 480 // CHECK-RV32-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 481 // CHECK-RV32-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 482 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load float, ptr [[A_ADDR]], align 4
 483 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[AGG_RESULT]], i32 0, i32 0
 484 // CHECK-RV32-NEXT:    store float [[TMP0]], ptr [[A1]], align 4
 485 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 486 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[AGG_RESULT]], i32 0, i32 1
 487 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 4
 488 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 489 // CHECK-RV32-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[AGG_RESULT]], i32 0, i32 2
 490 // CHECK-RV32-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 491 // CHECK-RV32-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 492 // CHECK-RV32-NEXT:    [[D4:%.*]] = getelementptr inbounds [[STRUCT_FLOATBFLOAT3]], ptr [[AGG_RESULT]], i32 0, i32 3
 493 // CHECK-RV32-NEXT:    store bfloat [[TMP3]], ptr [[D4]], align 4
 494 // CHECK-RV32-NEXT:    ret void
 495 //
 496 struct floatbfloat3 fh3(float a, __bf16 b, __bf16 c, __bf16 d) {
 497   struct floatbfloat3 x;
 498   x.a = a;
 499   x.b = b;
 500   x.c = c;
 501   x.d = d;
 502   return x;
 503 }
 504
 505 struct bfloat5 {
 506   __bf16 a;
 507   __bf16 b;
 508   __bf16 c;
 509   __bf16 d;
 510   __bf16 e;
 511 };
 512
 513 // CHECK-RV64-LABEL: define dso_local [2 x i64] @h5
 514 // CHECK-RV64-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]], bfloat noundef [[E:%.*]]) #[[ATTR0]] {
 515 // CHECK-RV64-NEXT:  entry:
 516 // CHECK-RV64-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_BFLOAT5:%.*]], align 2
 517 // CHECK-RV64-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 518 // CHECK-RV64-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 519 // CHECK-RV64-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 520 // CHECK-RV64-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 521 // CHECK-RV64-NEXT:    [[E_ADDR:%.*]] = alloca bfloat, align 2
 522 // CHECK-RV64-NEXT:    [[RETVAL_COERCE:%.*]] = alloca [2 x i64], align 8
 523 // CHECK-RV64-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 524 // CHECK-RV64-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 525 // CHECK-RV64-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 526 // CHECK-RV64-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 527 // CHECK-RV64-NEXT:    store bfloat [[E]], ptr [[E_ADDR]], align 2
 528 // CHECK-RV64-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 529 // CHECK-RV64-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[RETVAL]], i32 0, i32 0
 530 // CHECK-RV64-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
 531 // CHECK-RV64-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 532 // CHECK-RV64-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[RETVAL]], i32 0, i32 1
 533 // CHECK-RV64-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 534 // CHECK-RV64-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 535 // CHECK-RV64-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[RETVAL]], i32 0, i32 2
 536 // CHECK-RV64-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 537 // CHECK-RV64-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 538 // CHECK-RV64-NEXT:    [[D4:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[RETVAL]], i32 0, i32 3
 539 // CHECK-RV64-NEXT:    store bfloat [[TMP3]], ptr [[D4]], align 2
 540 // CHECK-RV64-NEXT:    [[TMP4:%.*]] = load bfloat, ptr [[E_ADDR]], align 2
 541 // CHECK-RV64-NEXT:    [[E5:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[RETVAL]], i32 0, i32 4
 542 // CHECK-RV64-NEXT:    store bfloat [[TMP4]], ptr [[E5]], align 2
 543 // CHECK-RV64-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[RETVAL_COERCE]], ptr align 2 [[RETVAL]], i64 10, i1 false)
 544 // CHECK-RV64-NEXT:    [[TMP5:%.*]] = load [2 x i64], ptr [[RETVAL_COERCE]], align 8
 545 // CHECK-RV64-NEXT:    ret [2 x i64] [[TMP5]]
 546 //
 547 // CHECK-RV32-LABEL: define dso_local void @h5
 548 // CHECK-RV32-SAME: (ptr noalias sret([[STRUCT_BFLOAT5:%.*]]) align 2 [[AGG_RESULT:%.*]], bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]], bfloat noundef [[E:%.*]]) #[[ATTR0]] {
 549 // CHECK-RV32-NEXT:  entry:
 550 // CHECK-RV32-NEXT:    [[RESULT_PTR:%.*]] = alloca ptr, align 4
 551 // CHECK-RV32-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 552 // CHECK-RV32-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 553 // CHECK-RV32-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 554 // CHECK-RV32-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 555 // CHECK-RV32-NEXT:    [[E_ADDR:%.*]] = alloca bfloat, align 2
 556 // CHECK-RV32-NEXT:    store ptr [[AGG_RESULT]], ptr [[RESULT_PTR]], align 4
 557 // CHECK-RV32-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 558 // CHECK-RV32-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 559 // CHECK-RV32-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 560 // CHECK-RV32-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 561 // CHECK-RV32-NEXT:    store bfloat [[E]], ptr [[E_ADDR]], align 2
 562 // CHECK-RV32-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 563 // CHECK-RV32-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[AGG_RESULT]], i32 0, i32 0
 564 // CHECK-RV32-NEXT:    store bfloat [[TMP0]], ptr [[A1]], align 2
 565 // CHECK-RV32-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 566 // CHECK-RV32-NEXT:    [[B2:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[AGG_RESULT]], i32 0, i32 1
 567 // CHECK-RV32-NEXT:    store bfloat [[TMP1]], ptr [[B2]], align 2
 568 // CHECK-RV32-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 569 // CHECK-RV32-NEXT:    [[C3:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[AGG_RESULT]], i32 0, i32 2
 570 // CHECK-RV32-NEXT:    store bfloat [[TMP2]], ptr [[C3]], align 2
 571 // CHECK-RV32-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 572 // CHECK-RV32-NEXT:    [[D4:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[AGG_RESULT]], i32 0, i32 3
 573 // CHECK-RV32-NEXT:    store bfloat [[TMP3]], ptr [[D4]], align 2
 574 // CHECK-RV32-NEXT:    [[TMP4:%.*]] = load bfloat, ptr [[E_ADDR]], align 2
 575 // CHECK-RV32-NEXT:    [[E5:%.*]] = getelementptr inbounds [[STRUCT_BFLOAT5]], ptr [[AGG_RESULT]], i32 0, i32 4
 576 // CHECK-RV32-NEXT:    store bfloat [[TMP4]], ptr [[E5]], align 2
 577 // CHECK-RV32-NEXT:    ret void
 578 //
 579 struct bfloat5 h5(__bf16 a, __bf16 b, __bf16 c, __bf16 d, __bf16 e) {
 580   struct bfloat5 x;
 581   x.a = a;
 582   x.b = b;
 583   x.c = c;
 584   x.d = d;
 585   x.e = e;
 586   return x;
 587 }