clang/test/CodeGen/X86/fexcess-precision-bfloat16.c

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
   2 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
   3 // RUN: -fbfloat16-excess-precision=fast -emit-llvm -o - %s \
   4 // RUN: | FileCheck -check-prefixes=CHECK-EXT %s
   5
   6 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
   7 // RUN: -fbfloat16-excess-precision=fast -target-feature +fullbf16 \
   8 // RUN: -emit-llvm -o - %s | FileCheck -check-prefixes=CHECK-NO-EXT %s
   9
  10 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  11 // RUN: -fbfloat16-excess-precision=standard -emit-llvm -o - %s \
  12 // RUN: | FileCheck -check-prefixes=CHECK-EXT %s
  13
  14 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  15 // RUN: -fbfloat16-excess-precision=standard -target-feature +fullbf16 \
  16 // RUN: -emit-llvm -o - %s | FileCheck -check-prefixes=CHECK-NO-EXT %s
  17
  18 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  19 // RUN: -fbfloat16-excess-precision=none -emit-llvm -o - %s \
  20 // RUN: | FileCheck -check-prefixes=CHECK-NO-EXT %s
  21
  22 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  23 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
  24 // RUN: -emit-llvm -o - %s \
  25 // RUN: | FileCheck -check-prefixes=CHECK-NO-EXT %s
  26
  27 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  28 // RUN: -fbfloat16-excess-precision=fast \
  29 // RUN: -emit-llvm -ffp-eval-method=source -o - %s \
  30 // RUN: | FileCheck -check-prefixes=CHECK-EXT %s
  31
  32 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  33 // RUN: -fbfloat16-excess-precision=fast -target-feature +fullbf16 \
  34 // RUN: -emit-llvm -ffp-eval-method=source -o - %s \
  35 // RUN: | FileCheck -check-prefixes=CHECK-NO-EXT %s
  36
  37 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  38 // RUN: -fbfloat16-excess-precision=standard \
  39 // RUN: -emit-llvm -ffp-eval-method=source -o - %s \
  40 // RUN: | FileCheck -check-prefixes=CHECK-EXT %s
  41
  42 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  43 // RUN: -fbfloat16-excess-precision=standard  -target-feature +fullbf16 \
  44 // RUN: -emit-llvm -ffp-eval-method=source -o - %s \
  45 // RUN: | FileCheck -check-prefixes=CHECK-NO-EXT %s
  46
  47 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  48 // RUN: -fbfloat16-excess-precision=none \
  49 // RUN: -emit-llvm -ffp-eval-method=source -o - %s \
  50 // RUN: | FileCheck -check-prefixes=CHECK-NO-EXT %s
  51
  52 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  53 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
  54 // RUN: -emit-llvm -ffp-eval-method=source -o - %s \
  55 // RUN: | FileCheck -check-prefixes=CHECK-NO-EXT %s
  56
  57 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  58 // RUN: -fbfloat16-excess-precision=fast \
  59 // RUN: -emit-llvm -ffp-eval-method=double -o - %s \
  60 // RUN: | FileCheck -check-prefixes=CHECK-EXT-DBL %s
  61
  62 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  63 // RUN: -fbfloat16-excess-precision=fast -target-feature +fullbf16 \
  64 // RUN: -emit-llvm -ffp-eval-method=double -o - %s \
  65 // RUN: | FileCheck -check-prefixes=CHECK-EXT-DBL %s
  66
  67 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  68 // RUN: -fbfloat16-excess-precision=standard \
  69 // RUN: -emit-llvm -ffp-eval-method=double -o - %s \
  70 // RUN: | FileCheck -check-prefixes=CHECK-EXT-DBL %s
  71
  72 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  73 // RUN: -fbfloat16-excess-precision=standard -target-feature +fullbf16 \
  74 // RUN: -emit-llvm -ffp-eval-method=double -o - %s \
  75 // RUN: | FileCheck -check-prefixes=CHECK-EXT-DBL %s
  76
  77 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  78 // RUN: -fbfloat16-excess-precision=none \
  79 // RUN: -emit-llvm -ffp-eval-method=double -o - %s \
  80 // RUN: | FileCheck -check-prefixes=CHECK-EXT-DBL %s
  81
  82 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  83 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
  84 // RUN: -emit-llvm -ffp-eval-method=double -o - %s \
  85 // RUN: | FileCheck -check-prefixes=CHECK-EXT-DBL %s
  86
  87 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  88 // RUN: -fbfloat16-excess-precision=fast \
  89 // RUN: -emit-llvm -ffp-eval-method=extended -o - %s \
  90 // RUN: | FileCheck -check-prefixes=CHECK-EXT-FP80 %s
  91
  92 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  93 // RUN: -fbfloat16-excess-precision=fast -target-feature +fullbf16 \
  94 // RUN: -emit-llvm -ffp-eval-method=extended -o - %s \
  95 // RUN: | FileCheck -check-prefixes=CHECK-EXT-FP80 %s
  96
  97 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
  98 // RUN: -fbfloat16-excess-precision=standard \
  99 // RUN: -emit-llvm -ffp-eval-method=extended -o - %s \
 100 // RUN: | FileCheck -check-prefixes=CHECK-EXT-FP80 %s
 101
 102 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 103 // RUN: -fbfloat16-excess-precision=standard -target-feature +fullbf16 \
 104 // RUN: -emit-llvm -ffp-eval-method=extended -o - %s \
 105 // RUN: | FileCheck -check-prefixes=CHECK-EXT-FP80 %s
 106
 107 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 108 // RUN: -fbfloat16-excess-precision=none \
 109 // RUN: -emit-llvm -ffp-eval-method=extended -o - %s \
 110 // RUN: | FileCheck -check-prefixes=CHECK-EXT-FP80 %s
 111
 112 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 113 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
 114 // RUN: -emit-llvm -ffp-eval-method=extended -o - %s \
 115 // RUN: | FileCheck -check-prefixes=CHECK-EXT-FP80 %s
 116
 117 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 118 // RUN: -fbfloat16-excess-precision=none \
 119 // RUN: -ffp-contract=on -emit-llvm -o - %s \
 120 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT %s
 121
 122 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 123 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
 124 // RUN: -ffp-contract=on -emit-llvm -o - %s \
 125 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT %s
 126
 127 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 128 // RUN: -fbfloat16-excess-precision=none \
 129 // RUN: -fmath-errno -ffp-contract=on -fno-rounding-math \
 130 // RUN: -ffp-eval-method=source -emit-llvm -o - %s \
 131 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT %s
 132
 133 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 134 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
 135 // RUN: -fmath-errno -ffp-contract=on -fno-rounding-math \
 136 // RUN: -ffp-eval-method=source -emit-llvm -o - %s \
 137 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT %s
 138
 139 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 140 // RUN: -fbfloat16-excess-precision=none \
 141 // RUN: -fmath-errno -ffp-contract=on -fno-rounding-math \
 142 // RUN: -ffp-eval-method=double -emit-llvm -o - %s \
 143 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT-DBL %s
 144
 145 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 146 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
 147 // RUN: -fmath-errno -ffp-contract=on -fno-rounding-math \
 148 // RUN: -ffp-eval-method=double -emit-llvm -o - %s \
 149 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT-DBL %s
 150
 151 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 152 // RUN: -fbfloat16-excess-precision=none \
 153 // RUN: -fmath-errno -ffp-contract=on -fno-rounding-math \
 154 // RUN: -ffp-eval-method=extended -emit-llvm -o - %s \
 155 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT-EXT %s
 156
 157 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 158 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
 159 // RUN: -fmath-errno -ffp-contract=on -fno-rounding-math \
 160 // RUN: -ffp-eval-method=extended -emit-llvm -o - %s \
 161 // RUN: | FileCheck -check-prefixes=CHECK-CONTRACT-EXT %s
 162
 163 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 164 // RUN: -fbfloat16-excess-precision=none \
 165 // RUN: -fapprox-func -fmath-errno -fno-signed-zeros -mreassociate \
 166 // RUN: -freciprocal-math -ffp-contract=on -fno-rounding-math \
 167 // RUN: -funsafe-math-optimizations -emit-llvm -o - %s \
 168 // RUN: | FileCheck -check-prefixes=CHECK-UNSAFE %s
 169
 170 // RUN: %clang_cc1 -triple x86_64-unknown-unknown \
 171 // RUN: -fbfloat16-excess-precision=none -target-feature +fullbf16 \
 172 // RUN: -fapprox-func -fmath-errno -fno-signed-zeros -mreassociate \
 173 // RUN: -freciprocal-math -ffp-contract=on -fno-rounding-math \
 174 // RUN: -funsafe-math-optimizations -emit-llvm -o - %s \
 175 // RUN: | FileCheck -check-prefixes=CHECK-UNSAFE %s
 176
 177 // CHECK-EXT-LABEL: define dso_local bfloat @f
 178 // CHECK-EXT-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 179 // CHECK-EXT-NEXT:  entry:
 180 // CHECK-EXT-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 181 // CHECK-EXT-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 182 // CHECK-EXT-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 183 // CHECK-EXT-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 184 // CHECK-EXT-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 185 // CHECK-EXT-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 186 // CHECK-EXT-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 187 // CHECK-EXT-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 188 // CHECK-EXT-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 189 // CHECK-EXT-NEXT:    [[EXT:%.*]] = fpext bfloat [[TMP0]] to float
 190 // CHECK-EXT-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 191 // CHECK-EXT-NEXT:    [[EXT1:%.*]] = fpext bfloat [[TMP1]] to float
 192 // CHECK-EXT-NEXT:    [[MUL:%.*]] = fmul float [[EXT]], [[EXT1]]
 193 // CHECK-EXT-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 194 // CHECK-EXT-NEXT:    [[EXT2:%.*]] = fpext bfloat [[TMP2]] to float
 195 // CHECK-EXT-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 196 // CHECK-EXT-NEXT:    [[EXT3:%.*]] = fpext bfloat [[TMP3]] to float
 197 // CHECK-EXT-NEXT:    [[MUL4:%.*]] = fmul float [[EXT2]], [[EXT3]]
 198 // CHECK-EXT-NEXT:    [[ADD:%.*]] = fadd float [[MUL]], [[MUL4]]
 199 // CHECK-EXT-NEXT:    [[UNPROMOTION:%.*]] = fptrunc float [[ADD]] to bfloat
 200 // CHECK-EXT-NEXT:    ret bfloat [[UNPROMOTION]]
 201 //
 202 // CHECK-NO-EXT-LABEL: define dso_local bfloat @f
 203 // CHECK-NO-EXT-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 204 // CHECK-NO-EXT-NEXT:  entry:
 205 // CHECK-NO-EXT-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 206 // CHECK-NO-EXT-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 207 // CHECK-NO-EXT-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 208 // CHECK-NO-EXT-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 209 // CHECK-NO-EXT-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 210 // CHECK-NO-EXT-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 211 // CHECK-NO-EXT-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 212 // CHECK-NO-EXT-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 213 // CHECK-NO-EXT-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 214 // CHECK-NO-EXT-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 215 // CHECK-NO-EXT-NEXT:    [[MUL:%.*]] = fmul bfloat [[TMP0]], [[TMP1]]
 216 // CHECK-NO-EXT-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 217 // CHECK-NO-EXT-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 218 // CHECK-NO-EXT-NEXT:    [[MUL1:%.*]] = fmul bfloat [[TMP2]], [[TMP3]]
 219 // CHECK-NO-EXT-NEXT:    [[ADD:%.*]] = fadd bfloat [[MUL]], [[MUL1]]
 220 // CHECK-NO-EXT-NEXT:    ret bfloat [[ADD]]
 221 //
 222 // CHECK-EXT-DBL-LABEL: define dso_local bfloat @f
 223 // CHECK-EXT-DBL-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 224 // CHECK-EXT-DBL-NEXT:  entry:
 225 // CHECK-EXT-DBL-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 226 // CHECK-EXT-DBL-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 227 // CHECK-EXT-DBL-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 228 // CHECK-EXT-DBL-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 229 // CHECK-EXT-DBL-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 230 // CHECK-EXT-DBL-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 231 // CHECK-EXT-DBL-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 232 // CHECK-EXT-DBL-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 233 // CHECK-EXT-DBL-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 234 // CHECK-EXT-DBL-NEXT:    [[CONV:%.*]] = fpext bfloat [[TMP0]] to double
 235 // CHECK-EXT-DBL-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 236 // CHECK-EXT-DBL-NEXT:    [[CONV1:%.*]] = fpext bfloat [[TMP1]] to double
 237 // CHECK-EXT-DBL-NEXT:    [[MUL:%.*]] = fmul double [[CONV]], [[CONV1]]
 238 // CHECK-EXT-DBL-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 239 // CHECK-EXT-DBL-NEXT:    [[CONV2:%.*]] = fpext bfloat [[TMP2]] to double
 240 // CHECK-EXT-DBL-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 241 // CHECK-EXT-DBL-NEXT:    [[CONV3:%.*]] = fpext bfloat [[TMP3]] to double
 242 // CHECK-EXT-DBL-NEXT:    [[MUL4:%.*]] = fmul double [[CONV2]], [[CONV3]]
 243 // CHECK-EXT-DBL-NEXT:    [[ADD:%.*]] = fadd double [[MUL]], [[MUL4]]
 244 // CHECK-EXT-DBL-NEXT:    [[CONV5:%.*]] = fptrunc double [[ADD]] to bfloat
 245 // CHECK-EXT-DBL-NEXT:    ret bfloat [[CONV5]]
 246 //
 247 // CHECK-EXT-FP80-LABEL: define dso_local bfloat @f
 248 // CHECK-EXT-FP80-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 249 // CHECK-EXT-FP80-NEXT:  entry:
 250 // CHECK-EXT-FP80-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 251 // CHECK-EXT-FP80-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 252 // CHECK-EXT-FP80-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 253 // CHECK-EXT-FP80-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 254 // CHECK-EXT-FP80-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 255 // CHECK-EXT-FP80-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 256 // CHECK-EXT-FP80-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 257 // CHECK-EXT-FP80-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 258 // CHECK-EXT-FP80-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 259 // CHECK-EXT-FP80-NEXT:    [[CONV:%.*]] = fpext bfloat [[TMP0]] to x86_fp80
 260 // CHECK-EXT-FP80-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 261 // CHECK-EXT-FP80-NEXT:    [[CONV1:%.*]] = fpext bfloat [[TMP1]] to x86_fp80
 262 // CHECK-EXT-FP80-NEXT:    [[MUL:%.*]] = fmul x86_fp80 [[CONV]], [[CONV1]]
 263 // CHECK-EXT-FP80-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 264 // CHECK-EXT-FP80-NEXT:    [[CONV2:%.*]] = fpext bfloat [[TMP2]] to x86_fp80
 265 // CHECK-EXT-FP80-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 266 // CHECK-EXT-FP80-NEXT:    [[CONV3:%.*]] = fpext bfloat [[TMP3]] to x86_fp80
 267 // CHECK-EXT-FP80-NEXT:    [[MUL4:%.*]] = fmul x86_fp80 [[CONV2]], [[CONV3]]
 268 // CHECK-EXT-FP80-NEXT:    [[ADD:%.*]] = fadd x86_fp80 [[MUL]], [[MUL4]]
 269 // CHECK-EXT-FP80-NEXT:    [[CONV5:%.*]] = fptrunc x86_fp80 [[ADD]] to bfloat
 270 // CHECK-EXT-FP80-NEXT:    ret bfloat [[CONV5]]
 271 //
 272 // CHECK-CONTRACT-LABEL: define dso_local bfloat @f
 273 // CHECK-CONTRACT-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 274 // CHECK-CONTRACT-NEXT:  entry:
 275 // CHECK-CONTRACT-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 276 // CHECK-CONTRACT-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 277 // CHECK-CONTRACT-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 278 // CHECK-CONTRACT-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 279 // CHECK-CONTRACT-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 280 // CHECK-CONTRACT-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 281 // CHECK-CONTRACT-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 282 // CHECK-CONTRACT-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 283 // CHECK-CONTRACT-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 284 // CHECK-CONTRACT-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 285 // CHECK-CONTRACT-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 286 // CHECK-CONTRACT-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 287 // CHECK-CONTRACT-NEXT:    [[MUL1:%.*]] = fmul bfloat [[TMP2]], [[TMP3]]
 288 // CHECK-CONTRACT-NEXT:    [[TMP4:%.*]] = call bfloat @llvm.fmuladd.bf16(bfloat [[TMP0]], bfloat [[TMP1]], bfloat [[MUL1]])
 289 // CHECK-CONTRACT-NEXT:    ret bfloat [[TMP4]]
 290 //
 291 // CHECK-CONTRACT-DBL-LABEL: define dso_local bfloat @f
 292 // CHECK-CONTRACT-DBL-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 293 // CHECK-CONTRACT-DBL-NEXT:  entry:
 294 // CHECK-CONTRACT-DBL-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 295 // CHECK-CONTRACT-DBL-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 296 // CHECK-CONTRACT-DBL-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 297 // CHECK-CONTRACT-DBL-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 298 // CHECK-CONTRACT-DBL-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 299 // CHECK-CONTRACT-DBL-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 300 // CHECK-CONTRACT-DBL-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 301 // CHECK-CONTRACT-DBL-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 302 // CHECK-CONTRACT-DBL-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 303 // CHECK-CONTRACT-DBL-NEXT:    [[CONV:%.*]] = fpext bfloat [[TMP0]] to double
 304 // CHECK-CONTRACT-DBL-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 305 // CHECK-CONTRACT-DBL-NEXT:    [[CONV1:%.*]] = fpext bfloat [[TMP1]] to double
 306 // CHECK-CONTRACT-DBL-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 307 // CHECK-CONTRACT-DBL-NEXT:    [[CONV2:%.*]] = fpext bfloat [[TMP2]] to double
 308 // CHECK-CONTRACT-DBL-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 309 // CHECK-CONTRACT-DBL-NEXT:    [[CONV3:%.*]] = fpext bfloat [[TMP3]] to double
 310 // CHECK-CONTRACT-DBL-NEXT:    [[MUL4:%.*]] = fmul double [[CONV2]], [[CONV3]]
 311 // CHECK-CONTRACT-DBL-NEXT:    [[TMP4:%.*]] = call double @llvm.fmuladd.f64(double [[CONV]], double [[CONV1]], double [[MUL4]])
 312 // CHECK-CONTRACT-DBL-NEXT:    [[CONV5:%.*]] = fptrunc double [[TMP4]] to bfloat
 313 // CHECK-CONTRACT-DBL-NEXT:    ret bfloat [[CONV5]]
 314 //
 315 // CHECK-CONTRACT-EXT-LABEL: define dso_local bfloat @f
 316 // CHECK-CONTRACT-EXT-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 317 // CHECK-CONTRACT-EXT-NEXT:  entry:
 318 // CHECK-CONTRACT-EXT-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 319 // CHECK-CONTRACT-EXT-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 320 // CHECK-CONTRACT-EXT-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 321 // CHECK-CONTRACT-EXT-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 322 // CHECK-CONTRACT-EXT-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 323 // CHECK-CONTRACT-EXT-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 324 // CHECK-CONTRACT-EXT-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 325 // CHECK-CONTRACT-EXT-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 326 // CHECK-CONTRACT-EXT-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 327 // CHECK-CONTRACT-EXT-NEXT:    [[CONV:%.*]] = fpext bfloat [[TMP0]] to x86_fp80
 328 // CHECK-CONTRACT-EXT-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 329 // CHECK-CONTRACT-EXT-NEXT:    [[CONV1:%.*]] = fpext bfloat [[TMP1]] to x86_fp80
 330 // CHECK-CONTRACT-EXT-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 331 // CHECK-CONTRACT-EXT-NEXT:    [[CONV2:%.*]] = fpext bfloat [[TMP2]] to x86_fp80
 332 // CHECK-CONTRACT-EXT-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 333 // CHECK-CONTRACT-EXT-NEXT:    [[CONV3:%.*]] = fpext bfloat [[TMP3]] to x86_fp80
 334 // CHECK-CONTRACT-EXT-NEXT:    [[MUL4:%.*]] = fmul x86_fp80 [[CONV2]], [[CONV3]]
 335 // CHECK-CONTRACT-EXT-NEXT:    [[TMP4:%.*]] = call x86_fp80 @llvm.fmuladd.f80(x86_fp80 [[CONV]], x86_fp80 [[CONV1]], x86_fp80 [[MUL4]])
 336 // CHECK-CONTRACT-EXT-NEXT:    [[CONV5:%.*]] = fptrunc x86_fp80 [[TMP4]] to bfloat
 337 // CHECK-CONTRACT-EXT-NEXT:    ret bfloat [[CONV5]]
 338 //
 339 // CHECK-UNSAFE-LABEL: define dso_local bfloat @f
 340 // CHECK-UNSAFE-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]], bfloat noundef [[C:%.*]], bfloat noundef [[D:%.*]]) #[[ATTR0:[0-9]+]] {
 341 // CHECK-UNSAFE-NEXT:  entry:
 342 // CHECK-UNSAFE-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2
 343 // CHECK-UNSAFE-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
 344 // CHECK-UNSAFE-NEXT:    [[C_ADDR:%.*]] = alloca bfloat, align 2
 345 // CHECK-UNSAFE-NEXT:    [[D_ADDR:%.*]] = alloca bfloat, align 2
 346 // CHECK-UNSAFE-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
 347 // CHECK-UNSAFE-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
 348 // CHECK-UNSAFE-NEXT:    store bfloat [[C]], ptr [[C_ADDR]], align 2
 349 // CHECK-UNSAFE-NEXT:    store bfloat [[D]], ptr [[D_ADDR]], align 2
 350 // CHECK-UNSAFE-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
 351 // CHECK-UNSAFE-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
 352 // CHECK-UNSAFE-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[C_ADDR]], align 2
 353 // CHECK-UNSAFE-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[D_ADDR]], align 2
 354 // CHECK-UNSAFE-NEXT:    [[MUL1:%.*]] = fmul reassoc nsz arcp afn bfloat [[TMP2]], [[TMP3]]
 355 // CHECK-UNSAFE-NEXT:    [[TMP4:%.*]] = call reassoc nsz arcp afn bfloat @llvm.fmuladd.bf16(bfloat [[TMP0]], bfloat [[TMP1]], bfloat [[MUL1]])
 356 // CHECK-UNSAFE-NEXT:    ret bfloat [[TMP4]]
 357 //
 358 __bf16 f(__bf16 a, __bf16 b, __bf16 c, __bf16 d) {
 359     return a * b + c * d;
 360 }