test/CodeGen/X86/fma-intrinsics-x86.ll

   1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
   2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
   3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX512VL
   4 ; RUN: llc < %s -mtriple=x86_64-pc-windows -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
   5
   6 ; VFMADD
   7 define <4 x float> @test_x86_fma_vfmadd_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
   8 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ss:
   9 ; CHECK-FMA:       # BB#0:
  10 ; CHECK-FMA-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
  11 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
  12 ;
  13 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_ss:
  14 ; CHECK-AVX512VL:       # BB#0:
  15 ; CHECK-AVX512VL-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
  16 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
  17 ;
  18 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ss:
  19 ; CHECK-FMA-WIN:       # BB#0:
  20 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x0a]
  21 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
  22 ; CHECK-FMA-WIN-NEXT:    vfmadd213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa9,0x00]
  23 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
  24   %res = call <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
  25   ret <4 x float> %res
  26 }
  27
  28 define <4 x float> @test_x86_fma_vfmadd_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
  29 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_bac_ss:
  30 ; CHECK-FMA:       # BB#0:
  31 ; CHECK-FMA-NEXT:    vfmadd213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xa9,0xca]
  32 ; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
  33 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
  34 ;
  35 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_bac_ss:
  36 ; CHECK-AVX512VL:       # BB#0:
  37 ; CHECK-AVX512VL-NEXT:    vfmadd213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xa9,0xca]
  38 ; CHECK-AVX512VL-NEXT:    vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
  39 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
  40 ;
  41 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_bac_ss:
  42 ; CHECK-FMA-WIN:       # BB#0:
  43 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
  44 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
  45 ; CHECK-FMA-WIN-NEXT:    vfmadd213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa9,0x00]
  46 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
  47   %res = call <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
  48   ret <4 x float> %res
  49 }
  50 declare <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>)
  51
  52 define <2 x double> @test_x86_fma_vfmadd_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
  53 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_sd:
  54 ; CHECK-FMA:       # BB#0:
  55 ; CHECK-FMA-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa9,0xc2]
  56 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
  57 ;
  58 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_sd:
  59 ; CHECK-AVX512VL:       # BB#0:
  60 ; CHECK-AVX512VL-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2]
  61 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
  62 ;
  63 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_sd:
  64 ; CHECK-FMA-WIN:       # BB#0:
  65 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x0a]
  66 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
  67 ; CHECK-FMA-WIN-NEXT:    vfmadd213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa9,0x00]
  68 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
  69   %res = call <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
  70   ret <2 x double> %res
  71 }
  72
  73 define <2 x double> @test_x86_fma_vfmadd_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
  74 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_bac_sd:
  75 ; CHECK-FMA:       # BB#0:
  76 ; CHECK-FMA-NEXT:    vfmadd213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xa9,0xca]
  77 ; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
  78 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
  79 ;
  80 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_bac_sd:
  81 ; CHECK-AVX512VL:       # BB#0:
  82 ; CHECK-AVX512VL-NEXT:    vfmadd213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xa9,0xca]
  83 ; CHECK-AVX512VL-NEXT:    vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
  84 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
  85 ;
  86 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_bac_sd:
  87 ; CHECK-FMA-WIN:       # BB#0:
  88 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
  89 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
  90 ; CHECK-FMA-WIN-NEXT:    vfmadd213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa9,0x00]
  91 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
  92   %res = call <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
  93   ret <2 x double> %res
  94 }
  95 declare <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>)
  96
  97 define <4 x float> @test_x86_fma_vfmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
  98 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ps:
  99 ; CHECK-FMA:       # BB#0:
 100 ; CHECK-FMA-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
 101 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 102 ;
 103 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_ps:
 104 ; CHECK-AVX512VL:       # BB#0:
 105 ; CHECK-AVX512VL-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
 106 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 107 ;
 108 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ps:
 109 ; CHECK-FMA-WIN:       # BB#0:
 110 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 111 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 112 ; CHECK-FMA-WIN-NEXT:    vfmadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa8,0x00]
 113 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 114   %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 115   ret <4 x float> %res
 116 }
 117 declare <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float>, <4 x float>, <4 x float>)
 118
 119 define <2 x double> @test_x86_fma_vfmadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 120 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_pd:
 121 ; CHECK-FMA:       # BB#0:
 122 ; CHECK-FMA-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
 123 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 124 ;
 125 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_pd:
 126 ; CHECK-AVX512VL:       # BB#0:
 127 ; CHECK-AVX512VL-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
 128 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 129 ;
 130 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_pd:
 131 ; CHECK-FMA-WIN:       # BB#0:
 132 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 133 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 134 ; CHECK-FMA-WIN-NEXT:    vfmadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa8,0x00]
 135 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 136   %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 137   ret <2 x double> %res
 138 }
 139 declare <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double>, <2 x double>, <2 x double>)
 140
 141 define <8 x float> @test_x86_fma_vfmadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
 142 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ps_256:
 143 ; CHECK-FMA:       # BB#0:
 144 ; CHECK-FMA-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
 145 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 146 ;
 147 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_ps_256:
 148 ; CHECK-AVX512VL:       # BB#0:
 149 ; CHECK-AVX512VL-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
 150 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 151 ;
 152 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ps_256:
 153 ; CHECK-FMA-WIN:       # BB#0:
 154 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
 155 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
 156 ; CHECK-FMA-WIN-NEXT:    vfmadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa8,0x00]
 157 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 158   %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
 159   ret <8 x float> %res
 160 }
 161 declare <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>)
 162
 163 define <4 x double> @test_x86_fma_vfmadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
 164 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_pd_256:
 165 ; CHECK-FMA:       # BB#0:
 166 ; CHECK-FMA-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
 167 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 168 ;
 169 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_pd_256:
 170 ; CHECK-AVX512VL:       # BB#0:
 171 ; CHECK-AVX512VL-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
 172 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 173 ;
 174 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_pd_256:
 175 ; CHECK-FMA-WIN:       # BB#0:
 176 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
 177 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
 178 ; CHECK-FMA-WIN-NEXT:    vfmadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa8,0x00]
 179 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 180   %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
 181   ret <4 x double> %res
 182 }
 183 declare <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>)
 184
 185 ; VFMSUB
 186 define <4 x float> @test_x86_fma_vfmsub_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 187 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ss:
 188 ; CHECK-FMA:       # BB#0:
 189 ; CHECK-FMA-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xab,0xc2]
 190 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 191 ;
 192 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_ss:
 193 ; CHECK-AVX512VL:       # BB#0:
 194 ; CHECK-AVX512VL-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xab,0xc2]
 195 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 196 ;
 197 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ss:
 198 ; CHECK-FMA-WIN:       # BB#0:
 199 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x0a]
 200 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
 201 ; CHECK-FMA-WIN-NEXT:    vfmsub213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xab,0x00]
 202 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 203   %res = call <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 204   ret <4 x float> %res
 205 }
 206
 207 define <4 x float> @test_x86_fma_vfmsub_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 208 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_bac_ss:
 209 ; CHECK-FMA:       # BB#0:
 210 ; CHECK-FMA-NEXT:    vfmsub213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xab,0xca]
 211 ; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
 212 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 213 ;
 214 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_bac_ss:
 215 ; CHECK-AVX512VL:       # BB#0:
 216 ; CHECK-AVX512VL-NEXT:    vfmsub213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xab,0xca]
 217 ; CHECK-AVX512VL-NEXT:    vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
 218 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 219 ;
 220 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_bac_ss:
 221 ; CHECK-FMA-WIN:       # BB#0:
 222 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 223 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 224 ; CHECK-FMA-WIN-NEXT:    vfmsub213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xab,0x00]
 225 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 226   %res = call <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
 227   ret <4 x float> %res
 228 }
 229 declare <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float>, <4 x float>, <4 x float>)
 230
 231 define <2 x double> @test_x86_fma_vfmsub_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 232 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_sd:
 233 ; CHECK-FMA:       # BB#0:
 234 ; CHECK-FMA-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xab,0xc2]
 235 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 236 ;
 237 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_sd:
 238 ; CHECK-AVX512VL:       # BB#0:
 239 ; CHECK-AVX512VL-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xab,0xc2]
 240 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 241 ;
 242 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_sd:
 243 ; CHECK-FMA-WIN:       # BB#0:
 244 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x0a]
 245 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
 246 ; CHECK-FMA-WIN-NEXT:    vfmsub213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xab,0x00]
 247 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 248   %res = call <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 249   ret <2 x double> %res
 250 }
 251
 252 define <2 x double> @test_x86_fma_vfmsub_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 253 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_bac_sd:
 254 ; CHECK-FMA:       # BB#0:
 255 ; CHECK-FMA-NEXT:    vfmsub213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xab,0xca]
 256 ; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
 257 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 258 ;
 259 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_bac_sd:
 260 ; CHECK-AVX512VL:       # BB#0:
 261 ; CHECK-AVX512VL-NEXT:    vfmsub213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xab,0xca]
 262 ; CHECK-AVX512VL-NEXT:    vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
 263 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 264 ;
 265 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_bac_sd:
 266 ; CHECK-FMA-WIN:       # BB#0:
 267 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 268 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 269 ; CHECK-FMA-WIN-NEXT:    vfmsub213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xab,0x00]
 270 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 271   %res = call <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
 272   ret <2 x double> %res
 273 }
 274 declare <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double>, <2 x double>, <2 x double>)
 275
 276 define <4 x float> @test_x86_fma_vfmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 277 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ps:
 278 ; CHECK-FMA:       # BB#0:
 279 ; CHECK-FMA-NEXT:    vfmsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xaa,0xc2]
 280 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 281 ;
 282 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_ps:
 283 ; CHECK-AVX512VL:       # BB#0:
 284 ; CHECK-AVX512VL-NEXT:    vfmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xaa,0xc2]
 285 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 286 ;
 287 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ps:
 288 ; CHECK-FMA-WIN:       # BB#0:
 289 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 290 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 291 ; CHECK-FMA-WIN-NEXT:    vfmsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xaa,0x00]
 292 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 293   %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 294   ret <4 x float> %res
 295 }
 296 declare <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float>, <4 x float>, <4 x float>)
 297
 298 define <2 x double> @test_x86_fma_vfmsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 299 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_pd:
 300 ; CHECK-FMA:       # BB#0:
 301 ; CHECK-FMA-NEXT:    vfmsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xaa,0xc2]
 302 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 303 ;
 304 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_pd:
 305 ; CHECK-AVX512VL:       # BB#0:
 306 ; CHECK-AVX512VL-NEXT:    vfmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xaa,0xc2]
 307 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 308 ;
 309 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_pd:
 310 ; CHECK-FMA-WIN:       # BB#0:
 311 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 312 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 313 ; CHECK-FMA-WIN-NEXT:    vfmsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xaa,0x00]
 314 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 315   %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 316   ret <2 x double> %res
 317 }
 318 declare <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double>, <2 x double>, <2 x double>)
 319
 320 define <8 x float> @test_x86_fma_vfmsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
 321 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ps_256:
 322 ; CHECK-FMA:       # BB#0:
 323 ; CHECK-FMA-NEXT:    vfmsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xaa,0xc2]
 324 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 325 ;
 326 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_ps_256:
 327 ; CHECK-AVX512VL:       # BB#0:
 328 ; CHECK-AVX512VL-NEXT:    vfmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xaa,0xc2]
 329 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 330 ;
 331 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ps_256:
 332 ; CHECK-FMA-WIN:       # BB#0:
 333 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
 334 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
 335 ; CHECK-FMA-WIN-NEXT:    vfmsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xaa,0x00]
 336 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 337   %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
 338   ret <8 x float> %res
 339 }
 340 declare <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x float>)
 341
 342 define <4 x double> @test_x86_fma_vfmsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
 343 ; CHECK-FMA-LABEL: test_x86_fma_vfmsub_pd_256:
 344 ; CHECK-FMA:       # BB#0:
 345 ; CHECK-FMA-NEXT:    vfmsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xaa,0xc2]
 346 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 347 ;
 348 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_pd_256:
 349 ; CHECK-AVX512VL:       # BB#0:
 350 ; CHECK-AVX512VL-NEXT:    vfmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xaa,0xc2]
 351 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 352 ;
 353 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_pd_256:
 354 ; CHECK-FMA-WIN:       # BB#0:
 355 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
 356 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
 357 ; CHECK-FMA-WIN-NEXT:    vfmsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xaa,0x00]
 358 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 359   %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
 360   ret <4 x double> %res
 361 }
 362 declare <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double>, <4 x double>, <4 x double>)
 363
 364 ; VFNMADD
 365 define <4 x float> @test_x86_fma_vfnmadd_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 366 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ss:
 367 ; CHECK-FMA:       # BB#0:
 368 ; CHECK-FMA-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xad,0xc2]
 369 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 370 ;
 371 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_ss:
 372 ; CHECK-AVX512VL:       # BB#0:
 373 ; CHECK-AVX512VL-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xad,0xc2]
 374 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 375 ;
 376 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ss:
 377 ; CHECK-FMA-WIN:       # BB#0:
 378 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x0a]
 379 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
 380 ; CHECK-FMA-WIN-NEXT:    vfnmadd213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xad,0x00]
 381 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 382   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 383   ret <4 x float> %res
 384 }
 385
 386 define <4 x float> @test_x86_fma_vfnmadd_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 387 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_bac_ss:
 388 ; CHECK-FMA:       # BB#0:
 389 ; CHECK-FMA-NEXT:    vfnmadd213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xad,0xca]
 390 ; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
 391 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 392 ;
 393 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_bac_ss:
 394 ; CHECK-AVX512VL:       # BB#0:
 395 ; CHECK-AVX512VL-NEXT:    vfnmadd213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xad,0xca]
 396 ; CHECK-AVX512VL-NEXT:    vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
 397 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 398 ;
 399 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_bac_ss:
 400 ; CHECK-FMA-WIN:       # BB#0:
 401 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 402 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 403 ; CHECK-FMA-WIN-NEXT:    vfnmadd213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xad,0x00]
 404 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 405   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
 406   ret <4 x float> %res
 407 }
 408 declare <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float>, <4 x float>, <4 x float>)
 409
 410 define <2 x double> @test_x86_fma_vfnmadd_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 411 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_sd:
 412 ; CHECK-FMA:       # BB#0:
 413 ; CHECK-FMA-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xad,0xc2]
 414 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 415 ;
 416 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_sd:
 417 ; CHECK-AVX512VL:       # BB#0:
 418 ; CHECK-AVX512VL-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xad,0xc2]
 419 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 420 ;
 421 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_sd:
 422 ; CHECK-FMA-WIN:       # BB#0:
 423 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x0a]
 424 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
 425 ; CHECK-FMA-WIN-NEXT:    vfnmadd213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xad,0x00]
 426 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 427   %res = call <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 428   ret <2 x double> %res
 429 }
 430
 431 define <2 x double> @test_x86_fma_vfnmadd_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 432 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_bac_sd:
 433 ; CHECK-FMA:       # BB#0:
 434 ; CHECK-FMA-NEXT:    vfnmadd213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xad,0xca]
 435 ; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
 436 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 437 ;
 438 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_bac_sd:
 439 ; CHECK-AVX512VL:       # BB#0:
 440 ; CHECK-AVX512VL-NEXT:    vfnmadd213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xad,0xca]
 441 ; CHECK-AVX512VL-NEXT:    vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
 442 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 443 ;
 444 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_bac_sd:
 445 ; CHECK-FMA-WIN:       # BB#0:
 446 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 447 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 448 ; CHECK-FMA-WIN-NEXT:    vfnmadd213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xad,0x00]
 449 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 450   %res = call <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
 451   ret <2 x double> %res
 452 }
 453 declare <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double>, <2 x double>, <2 x double>)
 454
 455 define <4 x float> @test_x86_fma_vfnmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 456 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ps:
 457 ; CHECK-FMA:       # BB#0:
 458 ; CHECK-FMA-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xac,0xc2]
 459 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 460 ;
 461 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_ps:
 462 ; CHECK-AVX512VL:       # BB#0:
 463 ; CHECK-AVX512VL-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xac,0xc2]
 464 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 465 ;
 466 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ps:
 467 ; CHECK-FMA-WIN:       # BB#0:
 468 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 469 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 470 ; CHECK-FMA-WIN-NEXT:    vfnmadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xac,0x00]
 471 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 472   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 473   ret <4 x float> %res
 474 }
 475 declare <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float>, <4 x float>, <4 x float>)
 476
 477 define <2 x double> @test_x86_fma_vfnmadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 478 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_pd:
 479 ; CHECK-FMA:       # BB#0:
 480 ; CHECK-FMA-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
 481 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 482 ;
 483 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_pd:
 484 ; CHECK-AVX512VL:       # BB#0:
 485 ; CHECK-AVX512VL-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
 486 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 487 ;
 488 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_pd:
 489 ; CHECK-FMA-WIN:       # BB#0:
 490 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 491 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 492 ; CHECK-FMA-WIN-NEXT:    vfnmadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xac,0x00]
 493 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 494   %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 495   ret <2 x double> %res
 496 }
 497 declare <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double>, <2 x double>, <2 x double>)
 498
 499 define <8 x float> @test_x86_fma_vfnmadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
 500 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ps_256:
 501 ; CHECK-FMA:       # BB#0:
 502 ; CHECK-FMA-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xac,0xc2]
 503 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 504 ;
 505 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_ps_256:
 506 ; CHECK-AVX512VL:       # BB#0:
 507 ; CHECK-AVX512VL-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xac,0xc2]
 508 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 509 ;
 510 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ps_256:
 511 ; CHECK-FMA-WIN:       # BB#0:
 512 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
 513 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
 514 ; CHECK-FMA-WIN-NEXT:    vfnmadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xac,0x00]
 515 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 516   %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
 517   ret <8 x float> %res
 518 }
 519 declare <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float>, <8 x float>, <8 x float>)
 520
 521 define <4 x double> @test_x86_fma_vfnmadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
 522 ; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_pd_256:
 523 ; CHECK-FMA:       # BB#0:
 524 ; CHECK-FMA-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
 525 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 526 ;
 527 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_pd_256:
 528 ; CHECK-AVX512VL:       # BB#0:
 529 ; CHECK-AVX512VL-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
 530 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 531 ;
 532 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_pd_256:
 533 ; CHECK-FMA-WIN:       # BB#0:
 534 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
 535 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
 536 ; CHECK-FMA-WIN-NEXT:    vfnmadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xac,0x00]
 537 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 538   %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
 539   ret <4 x double> %res
 540 }
 541 declare <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double>, <4 x double>, <4 x double>)
 542
 543 ; VFNMSUB
 544 define <4 x float> @test_x86_fma_vfnmsub_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 545 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ss:
 546 ; CHECK-FMA:       # BB#0:
 547 ; CHECK-FMA-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xaf,0xc2]
 548 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 549 ;
 550 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_ss:
 551 ; CHECK-AVX512VL:       # BB#0:
 552 ; CHECK-AVX512VL-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xaf,0xc2]
 553 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 554 ;
 555 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ss:
 556 ; CHECK-FMA-WIN:       # BB#0:
 557 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x0a]
 558 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
 559 ; CHECK-FMA-WIN-NEXT:    vfnmsub213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xaf,0x00]
 560 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 561   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 562   ret <4 x float> %res
 563 }
 564
 565 define <4 x float> @test_x86_fma_vfnmsub_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 566 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_bac_ss:
 567 ; CHECK-FMA:       # BB#0:
 568 ; CHECK-FMA-NEXT:    vfnmsub213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xaf,0xca]
 569 ; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
 570 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 571 ;
 572 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_bac_ss:
 573 ; CHECK-AVX512VL:       # BB#0:
 574 ; CHECK-AVX512VL-NEXT:    vfnmsub213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xaf,0xca]
 575 ; CHECK-AVX512VL-NEXT:    vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
 576 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 577 ;
 578 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_bac_ss:
 579 ; CHECK-FMA-WIN:       # BB#0:
 580 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 581 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 582 ; CHECK-FMA-WIN-NEXT:    vfnmsub213ss (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xaf,0x00]
 583 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 584   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
 585   ret <4 x float> %res
 586 }
 587 declare <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float>, <4 x float>, <4 x float>)
 588
 589 define <2 x double> @test_x86_fma_vfnmsub_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 590 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_sd:
 591 ; CHECK-FMA:       # BB#0:
 592 ; CHECK-FMA-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xaf,0xc2]
 593 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 594 ;
 595 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_sd:
 596 ; CHECK-AVX512VL:       # BB#0:
 597 ; CHECK-AVX512VL-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xaf,0xc2]
 598 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 599 ;
 600 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_sd:
 601 ; CHECK-FMA-WIN:       # BB#0:
 602 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x0a]
 603 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
 604 ; CHECK-FMA-WIN-NEXT:    vfnmsub213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xaf,0x00]
 605 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 606   %res = call <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 607   ret <2 x double> %res
 608 }
 609
 610 define <2 x double> @test_x86_fma_vfnmsub_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 611 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_bac_sd:
 612 ; CHECK-FMA:       # BB#0:
 613 ; CHECK-FMA-NEXT:    vfnmsub213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xaf,0xca]
 614 ; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
 615 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 616 ;
 617 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_bac_sd:
 618 ; CHECK-AVX512VL:       # BB#0:
 619 ; CHECK-AVX512VL-NEXT:    vfnmsub213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xaf,0xca]
 620 ; CHECK-AVX512VL-NEXT:    vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
 621 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 622 ;
 623 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_bac_sd:
 624 ; CHECK-FMA-WIN:       # BB#0:
 625 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 626 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 627 ; CHECK-FMA-WIN-NEXT:    vfnmsub213sd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xaf,0x00]
 628 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 629   %res = call <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
 630   ret <2 x double> %res
 631 }
 632 declare <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double>, <2 x double>, <2 x double>)
 633
 634 define <4 x float> @test_x86_fma_vfnmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 635 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ps:
 636 ; CHECK-FMA:       # BB#0:
 637 ; CHECK-FMA-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xae,0xc2]
 638 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 639 ;
 640 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_ps:
 641 ; CHECK-AVX512VL:       # BB#0:
 642 ; CHECK-AVX512VL-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xae,0xc2]
 643 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 644 ;
 645 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ps:
 646 ; CHECK-FMA-WIN:       # BB#0:
 647 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 648 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 649 ; CHECK-FMA-WIN-NEXT:    vfnmsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xae,0x00]
 650 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 651   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 652   ret <4 x float> %res
 653 }
 654 declare <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float>, <4 x float>, <4 x float>)
 655
 656 define <2 x double> @test_x86_fma_vfnmsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 657 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_pd:
 658 ; CHECK-FMA:       # BB#0:
 659 ; CHECK-FMA-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
 660 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 661 ;
 662 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_pd:
 663 ; CHECK-AVX512VL:       # BB#0:
 664 ; CHECK-AVX512VL-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
 665 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 666 ;
 667 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_pd:
 668 ; CHECK-FMA-WIN:       # BB#0:
 669 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 670 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 671 ; CHECK-FMA-WIN-NEXT:    vfnmsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xae,0x00]
 672 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 673   %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 674   ret <2 x double> %res
 675 }
 676 declare <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double>, <2 x double>, <2 x double>)
 677
 678 define <8 x float> @test_x86_fma_vfnmsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
 679 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ps_256:
 680 ; CHECK-FMA:       # BB#0:
 681 ; CHECK-FMA-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xae,0xc2]
 682 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 683 ;
 684 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_ps_256:
 685 ; CHECK-AVX512VL:       # BB#0:
 686 ; CHECK-AVX512VL-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xae,0xc2]
 687 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 688 ;
 689 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ps_256:
 690 ; CHECK-FMA-WIN:       # BB#0:
 691 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
 692 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
 693 ; CHECK-FMA-WIN-NEXT:    vfnmsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xae,0x00]
 694 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 695   %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
 696   ret <8 x float> %res
 697 }
 698 declare <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>)
 699
 700 define <4 x double> @test_x86_fma_vfnmsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
 701 ; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_pd_256:
 702 ; CHECK-FMA:       # BB#0:
 703 ; CHECK-FMA-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
 704 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 705 ;
 706 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_pd_256:
 707 ; CHECK-AVX512VL:       # BB#0:
 708 ; CHECK-AVX512VL-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
 709 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 710 ;
 711 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_pd_256:
 712 ; CHECK-FMA-WIN:       # BB#0:
 713 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
 714 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
 715 ; CHECK-FMA-WIN-NEXT:    vfnmsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xae,0x00]
 716 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 717   %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
 718   ret <4 x double> %res
 719 }
 720 declare <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double>, <4 x double>, <4 x double>)
 721
 722 ; VFMADDSUB
 723 define <4 x float> @test_x86_fma_vfmaddsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 724 ; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_ps:
 725 ; CHECK-FMA:       # BB#0:
 726 ; CHECK-FMA-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
 727 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 728 ;
 729 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_ps:
 730 ; CHECK-AVX512VL:       # BB#0:
 731 ; CHECK-AVX512VL-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
 732 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 733 ;
 734 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_ps:
 735 ; CHECK-FMA-WIN:       # BB#0:
 736 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 737 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 738 ; CHECK-FMA-WIN-NEXT:    vfmaddsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa6,0x00]
 739 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 740   %res = call <4 x float> @llvm.x86.fma.vfmaddsub.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 741   ret <4 x float> %res
 742 }
 743 declare <4 x float> @llvm.x86.fma.vfmaddsub.ps(<4 x float>, <4 x float>, <4 x float>)
 744
 745 define <2 x double> @test_x86_fma_vfmaddsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 746 ; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_pd:
 747 ; CHECK-FMA:       # BB#0:
 748 ; CHECK-FMA-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
 749 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 750 ;
 751 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_pd:
 752 ; CHECK-AVX512VL:       # BB#0:
 753 ; CHECK-AVX512VL-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
 754 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 755 ;
 756 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_pd:
 757 ; CHECK-FMA-WIN:       # BB#0:
 758 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 759 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 760 ; CHECK-FMA-WIN-NEXT:    vfmaddsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa6,0x00]
 761 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 762   %res = call <2 x double> @llvm.x86.fma.vfmaddsub.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 763   ret <2 x double> %res
 764 }
 765 declare <2 x double> @llvm.x86.fma.vfmaddsub.pd(<2 x double>, <2 x double>, <2 x double>)
 766
 767 define <8 x float> @test_x86_fma_vfmaddsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
 768 ; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_ps_256:
 769 ; CHECK-FMA:       # BB#0:
 770 ; CHECK-FMA-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
 771 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 772 ;
 773 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_ps_256:
 774 ; CHECK-AVX512VL:       # BB#0:
 775 ; CHECK-AVX512VL-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
 776 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 777 ;
 778 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_ps_256:
 779 ; CHECK-FMA-WIN:       # BB#0:
 780 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
 781 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
 782 ; CHECK-FMA-WIN-NEXT:    vfmaddsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa6,0x00]
 783 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 784   %res = call <8 x float> @llvm.x86.fma.vfmaddsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
 785   ret <8 x float> %res
 786 }
 787 declare <8 x float> @llvm.x86.fma.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>)
 788
 789 define <4 x double> @test_x86_fma_vfmaddsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
 790 ; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_pd_256:
 791 ; CHECK-FMA:       # BB#0:
 792 ; CHECK-FMA-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
 793 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 794 ;
 795 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_pd_256:
 796 ; CHECK-AVX512VL:       # BB#0:
 797 ; CHECK-AVX512VL-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
 798 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 799 ;
 800 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_pd_256:
 801 ; CHECK-FMA-WIN:       # BB#0:
 802 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
 803 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
 804 ; CHECK-FMA-WIN-NEXT:    vfmaddsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa6,0x00]
 805 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 806   %res = call <4 x double> @llvm.x86.fma.vfmaddsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
 807   ret <4 x double> %res
 808 }
 809 declare <4 x double> @llvm.x86.fma.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>)
 810
 811 ; VFMSUBADD
 812 define <4 x float> @test_x86_fma_vfmsubadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
 813 ; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_ps:
 814 ; CHECK-FMA:       # BB#0:
 815 ; CHECK-FMA-NEXT:    vfmsubadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa7,0xc2]
 816 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 817 ;
 818 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_ps:
 819 ; CHECK-AVX512VL:       # BB#0:
 820 ; CHECK-AVX512VL-NEXT:    vfmsubadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa7,0xc2]
 821 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 822 ;
 823 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_ps:
 824 ; CHECK-FMA-WIN:       # BB#0:
 825 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
 826 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
 827 ; CHECK-FMA-WIN-NEXT:    vfmsubadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa7,0x00]
 828 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 829   %res = call <4 x float> @llvm.x86.fma.vfmsubadd.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
 830   ret <4 x float> %res
 831 }
 832 declare <4 x float> @llvm.x86.fma.vfmsubadd.ps(<4 x float>, <4 x float>, <4 x float>)
 833
 834 define <2 x double> @test_x86_fma_vfmsubadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
 835 ; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_pd:
 836 ; CHECK-FMA:       # BB#0:
 837 ; CHECK-FMA-NEXT:    vfmsubadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa7,0xc2]
 838 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 839 ;
 840 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_pd:
 841 ; CHECK-AVX512VL:       # BB#0:
 842 ; CHECK-AVX512VL-NEXT:    vfmsubadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa7,0xc2]
 843 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 844 ;
 845 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_pd:
 846 ; CHECK-FMA-WIN:       # BB#0:
 847 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
 848 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
 849 ; CHECK-FMA-WIN-NEXT:    vfmsubadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa7,0x00]
 850 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 851   %res = call <2 x double> @llvm.x86.fma.vfmsubadd.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
 852   ret <2 x double> %res
 853 }
 854 declare <2 x double> @llvm.x86.fma.vfmsubadd.pd(<2 x double>, <2 x double>, <2 x double>)
 855
 856 define <8 x float> @test_x86_fma_vfmsubadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
 857 ; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_ps_256:
 858 ; CHECK-FMA:       # BB#0:
 859 ; CHECK-FMA-NEXT:    vfmsubadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa7,0xc2]
 860 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 861 ;
 862 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_ps_256:
 863 ; CHECK-AVX512VL:       # BB#0:
 864 ; CHECK-AVX512VL-NEXT:    vfmsubadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa7,0xc2]
 865 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 866 ;
 867 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_ps_256:
 868 ; CHECK-FMA-WIN:       # BB#0:
 869 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
 870 ; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
 871 ; CHECK-FMA-WIN-NEXT:    vfmsubadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa7,0x00]
 872 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 873   %res = call <8 x float> @llvm.x86.fma.vfmsubadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
 874   ret <8 x float> %res
 875 }
 876 declare <8 x float> @llvm.x86.fma.vfmsubadd.ps.256(<8 x float>, <8 x float>, <8 x float>)
 877
 878 define <4 x double> @test_x86_fma_vfmsubadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
 879 ; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_pd_256:
 880 ; CHECK-FMA:       # BB#0:
 881 ; CHECK-FMA-NEXT:    vfmsubadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa7,0xc2]
 882 ; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
 883 ;
 884 ; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_pd_256:
 885 ; CHECK-AVX512VL:       # BB#0:
 886 ; CHECK-AVX512VL-NEXT:    vfmsubadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa7,0xc2]
 887 ; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
 888 ;
 889 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_pd_256:
 890 ; CHECK-FMA-WIN:       # BB#0:
 891 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
 892 ; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
 893 ; CHECK-FMA-WIN-NEXT:    vfmsubadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa7,0x00]
 894 ; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
 895   %res = call <4 x double> @llvm.x86.fma.vfmsubadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
 896   ret <4 x double> %res
 897 }
 898 declare <4 x double> @llvm.x86.fma.vfmsubadd.pd.256(<4 x double>, <4 x double>, <4 x double>)
 899
 900 attributes #0 = { nounwind }