clang/test/CodeGen/X86/fma-builtins-constrained.c

   1 // REQUIRES: x86-registered-target
   2 // RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -O2 -emit-llvm -o - | FileCheck %s --check-prefixes=COMMON,COMMONIR,UNCONSTRAINED
   3 // RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -ffp-exception-behavior=maytrap -DSTRICT=1 -O2 -emit-llvm -o - | FileCheck %s --check-prefixes=COMMON,COMMONIR,CONSTRAINED
   4 // RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -O2 -S -o - | FileCheck %s --check-prefixes=COMMON,CHECK-ASM
   5 // RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -O2 -ffp-exception-behavior=maytrap -DSTRICT=1 -S -o - | FileCheck %s --check-prefixes=COMMON,CHECK-ASM
   6 // RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -O2 -emit-llvm -o - | FileCheck %s --check-prefixes=COMMON,COMMONIR,UNCONSTRAINED
   7 // RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -ffp-exception-behavior=maytrap -DSTRICT=1 -O2 -emit-llvm -o - | FileCheck %s --check-prefixes=COMMON,COMMONIR,CONSTRAINED
   8 // RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -O2 -S -o - | FileCheck %s --check-prefixes=COMMON,CHECK-ASM
   9 // RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +fma -O2 -ffp-exception-behavior=maytrap -DSTRICT=1 -S -o - | FileCheck %s --check-prefixes=COMMON,CHECK-ASM
  10
  11 #ifdef STRICT
  12 // Test that the constrained intrinsics are picking up the exception
  13 // metadata from the AST instead of the global default from the command line.
  14
  15 #pragma float_control(except, on)
  16 #endif
  17
  18 #include <immintrin.h>
  19
  20 __m128 test_mm_fmadd_ps(__m128 a, __m128 b, __m128 c) {
  21   // COMMON-LABEL: test_mm_fmadd_ps
  22   // UNCONSTRAINED: call {{.*}}<4 x float> @llvm.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}})
  23   // CONSTRAINED: call {{.*}}<4 x float> @llvm.experimental.constrained.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}, metadata !{{.*}})
  24   // CHECK-ASM: vfmadd213ps
  25   return _mm_fmadd_ps(a, b, c);
  26 }
  27
  28 __m128d test_mm_fmadd_pd(__m128d a, __m128d b, __m128d c) {
  29   // COMMON-LABEL: test_mm_fmadd_pd
  30   // UNCONSTRAINED: call {{.*}}<2 x double> @llvm.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}})
  31   // CONSTRAINED: call {{.*}}<2 x double> @llvm.experimental.constrained.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}, metadata !{{.*}})
  32   // CHECK-ASM: vfmadd213pd
  33   return _mm_fmadd_pd(a, b, c);
  34 }
  35
  36 __m128 test_mm_fmadd_ss(__m128 a, __m128 b, __m128 c) {
  37   // COMMON-LABEL: test_mm_fmadd_ss
  38   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
  39   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
  40   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
  41   // UNCONSTRAINED: call float @llvm.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}})
  42   // CONSTRAINED: call float @llvm.experimental.constrained.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}}, metadata !{{.*}})
  43   // CHECK-ASM: vfmadd213ss
  44   // COMMONIR: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
  45   return _mm_fmadd_ss(a, b, c);
  46 }
  47
  48 __m128d test_mm_fmadd_sd(__m128d a, __m128d b, __m128d c) {
  49   // COMMON-LABEL: test_mm_fmadd_sd
  50   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
  51   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
  52   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
  53   // UNCONSTRAINED: call double @llvm.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}})
  54   // CONSTRAINED: call double @llvm.experimental.constrained.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}}, metadata !{{.*}})
  55   // CHECK-ASM: vfmadd213sd
  56   // COMMONIR: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
  57   return _mm_fmadd_sd(a, b, c);
  58 }
  59
  60 __m128 test_mm_fmsub_ps(__m128 a, __m128 b, __m128 c) {
  61   // COMMON-LABEL: test_mm_fmsub_ps
  62   // COMMONIR: [[NEG:%.+]] = fneg <4 x float> %{{.+}}
  63   // UNCONSTRAINED: call {{.*}}<4 x float> @llvm.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}})
  64   // CONSTRAINED: call {{.*}}<4 x float> @llvm.experimental.constrained.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}, metadata !{{.*}})
  65   // CHECK-ASM: vfmsub213ps
  66   return _mm_fmsub_ps(a, b, c);
  67 }
  68
  69 __m128d test_mm_fmsub_pd(__m128d a, __m128d b, __m128d c) {
  70   // COMMON-LABEL: test_mm_fmsub_pd
  71   // COMMONIR: [[NEG:%.+]] = fneg <2 x double> %{{.+}}
  72   // UNCONSTRAINED: call {{.*}}<2 x double> @llvm.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}})
  73   // CONSTRAINED: call {{.*}}<2 x double> @llvm.experimental.constrained.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}, metadata !{{.*}})
  74   // CHECK-ASM: vfmsub213pd
  75   return _mm_fmsub_pd(a, b, c);
  76 }
  77
  78 __m128 test_mm_fmsub_ss(__m128 a, __m128 b, __m128 c) {
  79   // COMMON-LABEL: test_mm_fmsub_ss
  80   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
  81   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
  82   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
  83   // COMMONIR: [[NEG:%.+]] = fneg float %{{.+}}
  84   // UNCONSTRAINED: call float @llvm.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}})
  85   // CONSTRAINED: call float @llvm.experimental.constrained.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}}, metadata !{{.*}})
  86   // CHECK-ASM: vfmsub213ss
  87   // COMMONIR: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
  88   return _mm_fmsub_ss(a, b, c);
  89 }
  90
  91 __m128d test_mm_fmsub_sd(__m128d a, __m128d b, __m128d c) {
  92   // COMMON-LABEL: test_mm_fmsub_sd
  93   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
  94   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
  95   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
  96   // COMMONIR: [[NEG:%.+]] = fneg double %{{.+}}
  97   // UNCONSTRAINED: call double @llvm.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}})
  98   // CONSTRAINED: call double @llvm.experimental.constrained.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}}, metadata !{{.*}})
  99   // CHECK-ASM: vfmsub213sd
 100   // COMMONIR: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
 101   return _mm_fmsub_sd(a, b, c);
 102 }
 103
 104 __m128 test_mm_fnmadd_ps(__m128 a, __m128 b, __m128 c) {
 105   // COMMON-LABEL: test_mm_fnmadd_ps
 106   // COMMONIR: [[NEG:%.+]] = fneg <4 x float> %{{.+}}
 107   // UNCONSTRAINED: call {{.*}}<4 x float> @llvm.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}})
 108   // CONSTRAINED: call {{.*}}<4 x float> @llvm.experimental.constrained.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}, metadata !{{.*}})
 109   // CHECK-ASM: vfnmadd213ps
 110   return _mm_fnmadd_ps(a, b, c);
 111 }
 112
 113 __m128d test_mm_fnmadd_pd(__m128d a, __m128d b, __m128d c) {
 114   // COMMON-LABEL: test_mm_fnmadd_pd
 115   // COMMONIR: [[NEG:%.+]] = fneg <2 x double> %{{.+}}
 116   // UNCONSTRAINED: call {{.*}}<2 x double> @llvm.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}})
 117   // CONSTRAINED: call {{.*}}<2 x double> @llvm.experimental.constrained.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}, metadata !{{.*}})
 118   // CHECK-ASM: vfnmadd213pd
 119   return _mm_fnmadd_pd(a, b, c);
 120 }
 121
 122 __m128 test_mm_fnmadd_ss(__m128 a, __m128 b, __m128 c) {
 123   // COMMON-LABEL: test_mm_fnmadd_ss
 124   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
 125   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
 126   // COMMONIR: [[NEG:%.+]] = fneg float %{{.+}}
 127   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
 128   // UNCONSTRAINED: call float @llvm.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}})
 129   // CONSTRAINED: call float @llvm.experimental.constrained.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}}, metadata !{{.*}})
 130   // CHECK-ASM: vfnmadd213ss
 131   // COMMONIR: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
 132   return _mm_fnmadd_ss(a, b, c);
 133 }
 134
 135 __m128d test_mm_fnmadd_sd(__m128d a, __m128d b, __m128d c) {
 136   // COMMON-LABEL: test_mm_fnmadd_sd
 137   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
 138   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
 139   // COMMONIR: [[NEG:%.+]] = fneg double %{{.+}}
 140   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
 141   // UNCONSTRAINED: call double @llvm.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}})
 142   // CONSTRAINED: call double @llvm.experimental.constrained.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}}, metadata !{{.*}})
 143   // CHECK-ASM: vfnmadd213sd
 144   // COMMONIR: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
 145   return _mm_fnmadd_sd(a, b, c);
 146 }
 147
 148 __m128 test_mm_fnmsub_ps(__m128 a, __m128 b, __m128 c) {
 149   // COMMON-LABEL: test_mm_fnmsub_ps
 150   // COMMONIR: [[NEG:%.+]] = fneg <4 x float> %{{.+}}
 151   // COMMONIR: [[NEG2:%.+]] = fneg <4 x float> %{{.+}}
 152   // UNCONSTRAINED: call {{.*}}<4 x float> @llvm.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}})
 153   // CONSTRAINED: call {{.*}}<4 x float> @llvm.experimental.constrained.fma.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}, metadata !{{.*}})
 154   // CHECK-ASM: vfnmsub213ps
 155   return _mm_fnmsub_ps(a, b, c);
 156 }
 157
 158 __m128d test_mm_fnmsub_pd(__m128d a, __m128d b, __m128d c) {
 159   // COMMON-LABEL: test_mm_fnmsub_pd
 160   // COMMONIR: [[NEG:%.+]] = fneg <2 x double> %{{.+}}
 161   // COMMONIR: [[NEG2:%.+]] = fneg <2 x double> %{{.+}}
 162   // UNCONSTRAINED: call {{.*}}<2 x double> @llvm.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}})
 163   // CONSTRAINED: call {{.*}}<2 x double> @llvm.experimental.constrained.fma.v2f64(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}, metadata !{{.*}})
 164   // CHECK-ASM: vfnmsub213pd
 165   return _mm_fnmsub_pd(a, b, c);
 166 }
 167
 168 __m128 test_mm_fnmsub_ss(__m128 a, __m128 b, __m128 c) {
 169   // COMMON-LABEL: test_mm_fnmsub_ss
 170   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
 171   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
 172   // COMMONIR: [[NEG:%.+]] = fneg float %{{.+}}
 173   // COMMONIR: extractelement <4 x float> %{{.*}}, i64 0
 174   // COMMONIR: [[NEG2:%.+]] = fneg float %{{.+}}
 175   // UNCONSTRAINED: call float @llvm.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}})
 176   // CONSTRAINED: call float @llvm.experimental.constrained.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}}, metadata !{{.*}})
 177   // CHECK-ASM: vfnmsub213ss
 178   // COMMONIR: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
 179   return _mm_fnmsub_ss(a, b, c);
 180 }
 181
 182 __m128d test_mm_fnmsub_sd(__m128d a, __m128d b, __m128d c) {
 183   // COMMON-LABEL: test_mm_fnmsub_sd
 184   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
 185   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
 186   // COMMONIR: [[NEG:%.+]] = fneg double %{{.+}}
 187   // COMMONIR: extractelement <2 x double> %{{.*}}, i64 0
 188   // COMMONIR: [[NEG2:%.+]] = fneg double %{{.+}}
 189   // UNCONSTRAINED: call double @llvm.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}})
 190   // CONSTRAINED: call double @llvm.experimental.constrained.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}}, metadata !{{.*}})
 191   // CHECK-ASM: vfnmsub213sd
 192   // COMMONIR: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
 193   return _mm_fnmsub_sd(a, b, c);
 194 }
 195
 196 __m128 test_mm_fmaddsub_ps(__m128 a, __m128 b, __m128 c) {
 197   // COMMON-LABEL: test_mm_fmaddsub_ps
 198   // COMMONIR-NOT: fneg
 199   // COMMONIR: tail call {{.*}}<4 x float> @llvm.x86.fma.vfmaddsub.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}})
 200   // CHECK-ASM: vfmaddsub213ps
 201   return _mm_fmaddsub_ps(a, b, c);
 202 }
 203
 204 __m128d test_mm_fmaddsub_pd(__m128d a, __m128d b, __m128d c) {
 205   // COMMON-LABEL: test_mm_fmaddsub_pd
 206   // COMMONIR-NOT: fneg
 207   // COMMONIR: tail call {{.*}}<2 x double> @llvm.x86.fma.vfmaddsub.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}})
 208   // CHECK-ASM: vfmaddsub213pd
 209   return _mm_fmaddsub_pd(a, b, c);
 210 }
 211
 212 __m128 test_mm_fmsubadd_ps(__m128 a, __m128 b, __m128 c) {
 213   // COMMON-LABEL: test_mm_fmsubadd_ps
 214   // COMMONIR: [[FNEG:%.+]] = fneg <4 x float> %{{.*}}
 215   // COMMONIR: tail call {{.*}}<4 x float> @llvm.x86.fma.vfmaddsub.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x float> [[FNEG]])
 216   // CHECK-ASM: vfmsubadd213ps
 217   return _mm_fmsubadd_ps(a, b, c);
 218 }
 219
 220 __m128d test_mm_fmsubadd_pd(__m128d a, __m128d b, __m128d c) {
 221   // COMMON-LABEL: test_mm_fmsubadd_pd
 222   // COMMONIR: [[FNEG:%.+]] = fneg <2 x double> %{{.*}}
 223   // COMMONIR: tail call {{.*}}<2 x double> @llvm.x86.fma.vfmaddsub.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x double> [[FNEG]])
 224   // CHECK-ASM: vfmsubadd213pd
 225   return _mm_fmsubadd_pd(a, b, c);
 226 }
 227
 228 __m256 test_mm256_fmadd_ps(__m256 a, __m256 b, __m256 c) {
 229   // COMMON-LABEL: test_mm256_fmadd_ps
 230   // UNCONSTRAINED: call {{.*}}<8 x float> @llvm.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}})
 231   // CONSTRAINED: call {{.*}}<8 x float> @llvm.experimental.constrained.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}, metadata !{{.*}})
 232   // CHECK-ASM: vfmadd213ps
 233   return _mm256_fmadd_ps(a, b, c);
 234 }
 235
 236 __m256d test_mm256_fmadd_pd(__m256d a, __m256d b, __m256d c) {
 237   // COMMON-LABEL: test_mm256_fmadd_pd
 238   // UNCONSTRAINED: call {{.*}}<4 x double> @llvm.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}})
 239   // CONSTRAINED: call {{.*}}<4 x double> @llvm.experimental.constrained.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}, metadata !{{.*}})
 240   // CHECK-ASM: vfmadd213pd
 241   return _mm256_fmadd_pd(a, b, c);
 242 }
 243
 244 __m256 test_mm256_fmsub_ps(__m256 a, __m256 b, __m256 c) {
 245   // COMMON-LABEL: test_mm256_fmsub_ps
 246   // COMMONIR: [[NEG:%.+]] = fneg <8 x float> %{{.*}}
 247   // UNCONSTRAINED: call {{.*}}<8 x float> @llvm.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}})
 248   // CONSTRAINED: call {{.*}}<8 x float> @llvm.experimental.constrained.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}, metadata !{{.*}})
 249   // CHECK-ASM: vfmsub213ps
 250   return _mm256_fmsub_ps(a, b, c);
 251 }
 252
 253 __m256d test_mm256_fmsub_pd(__m256d a, __m256d b, __m256d c) {
 254   // COMMON-LABEL: test_mm256_fmsub_pd
 255   // COMMONIR: [[NEG:%.+]] = fneg <4 x double> %{{.+}}
 256   // UNCONSTRAINED: call {{.*}}<4 x double> @llvm.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}})
 257   // CONSTRAINED: call {{.*}}<4 x double> @llvm.experimental.constrained.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}, metadata !{{.*}})
 258   // CHECK-ASM: vfmsub213pd
 259   return _mm256_fmsub_pd(a, b, c);
 260 }
 261
 262 __m256 test_mm256_fnmadd_ps(__m256 a, __m256 b, __m256 c) {
 263   // COMMON-LABEL: test_mm256_fnmadd_ps
 264   // COMMONIR: [[NEG:%.+]] = fneg <8 x float> %{{.*}}
 265   // UNCONSTRAINED: call {{.*}}<8 x float> @llvm.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}})
 266   // CONSTRAINED: call {{.*}}<8 x float> @llvm.experimental.constrained.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}, metadata !{{.*}})
 267   // CHECK-ASM: vfnmadd213ps
 268   return _mm256_fnmadd_ps(a, b, c);
 269 }
 270
 271 __m256d test_mm256_fnmadd_pd(__m256d a, __m256d b, __m256d c) {
 272   // COMMON-LABEL: test_mm256_fnmadd_pd
 273   // COMMONIR: [[NEG:%.+]] = fneg <4 x double> %{{.+}}
 274   // UNCONSTRAINED: call {{.*}}<4 x double> @llvm.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}})
 275   // CONSTRAINED: call {{.*}}<4 x double> @llvm.experimental.constrained.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}, metadata !{{.*}})
 276   // CHECK-ASM: vfnmadd213pd
 277   return _mm256_fnmadd_pd(a, b, c);
 278 }
 279
 280 __m256 test_mm256_fnmsub_ps(__m256 a, __m256 b, __m256 c) {
 281   // COMMON-LABEL: test_mm256_fnmsub_ps
 282   // COMMONIR: [[NEG:%.+]] = fneg <8 x float> %{{.*}}
 283   // COMMONIR: [[NEG2:%.+]] = fneg <8 x float> %{{.*}}
 284   // UNCONSTRAINED: call {{.*}}<8 x float> @llvm.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}})
 285   // CONSTRAINED: call {{.*}}<8 x float> @llvm.experimental.constrained.fma.v8f32(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}, metadata !{{.*}})
 286   // CHECK-ASM: vfnmsub213ps
 287   return _mm256_fnmsub_ps(a, b, c);
 288 }
 289
 290 __m256d test_mm256_fnmsub_pd(__m256d a, __m256d b, __m256d c) {
 291   // COMMON-LABEL: test_mm256_fnmsub_pd
 292   // COMMONIR: [[NEG:%.+]] = fneg <4 x double> %{{.+}}
 293   // COMMONIR: [[NEG2:%.+]] = fneg <4 x double> %{{.+}}
 294   // UNCONSTRAINED: call {{.*}}<4 x double> @llvm.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}})
 295   // CONSTRAINED: call {{.*}}<4 x double> @llvm.experimental.constrained.fma.v4f64(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}, metadata !{{.*}})
 296   // CHECK-ASM: vfnmsub213pd
 297   return _mm256_fnmsub_pd(a, b, c);
 298 }
 299
 300 __m256 test_mm256_fmaddsub_ps(__m256 a, __m256 b, __m256 c) {
 301   // COMMON-LABEL: test_mm256_fmaddsub_ps
 302   // COMMONIR-NOT: fneg
 303   // COMMONIR: tail call {{.*}}<8 x float> @llvm.x86.fma.vfmaddsub.ps.256(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}})
 304   // CHECK-ASM: vfmaddsub213ps
 305   return _mm256_fmaddsub_ps(a, b, c);
 306 }
 307
 308 __m256d test_mm256_fmaddsub_pd(__m256d a, __m256d b, __m256d c) {
 309   // COMMON-LABEL: test_mm256_fmaddsub_pd
 310   // COMMONIR-NOT: fneg
 311   // COMMONIR: tail call {{.*}}<4 x double> @llvm.x86.fma.vfmaddsub.pd.256(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}})
 312   // CHECK-ASM: vfmaddsub213pd
 313   return _mm256_fmaddsub_pd(a, b, c);
 314 }
 315
 316 __m256 test_mm256_fmsubadd_ps(__m256 a, __m256 b, __m256 c) {
 317   // COMMON-LABEL: test_mm256_fmsubadd_ps
 318   // COMMONIR: [[FNEG:%.+]] = fneg <8 x float> %{{.*}}
 319   // COMMONIR: tail call {{.*}}<8 x float> @llvm.x86.fma.vfmaddsub.ps.256(<8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x float> [[FNEG]])
 320   // CHECK-ASM: vfmsubadd213ps
 321   return _mm256_fmsubadd_ps(a, b, c);
 322 }
 323
 324 __m256d test_mm256_fmsubadd_pd(__m256d a, __m256d b, __m256d c) {
 325   // COMMON-LABEL: test_mm256_fmsubadd_pd
 326   // COMMONIR: [[FNEG:%.+]] = fneg <4 x double> %{{.*}}
 327   // COMMONIR: tail call {{.*}}<4 x double> @llvm.x86.fma.vfmaddsub.pd.256(<4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x double> [[FNEG]])
 328   // CHECK-ASM: vfmsubadd213pd
 329   return _mm256_fmsubadd_pd(a, b, c);
 330 }