llvm/test/CodeGen/X86/vec-strict-256-fp16.ll

   1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
   2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512fp16 -mattr=+avx512vl -O3 | FileCheck %s
   3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512fp16 -mattr=+avx512vl -O3 | FileCheck %s
   4
   5 declare <16 x half> @llvm.experimental.constrained.fadd.v16f16(<16 x half>, <16 x half>, metadata, metadata)
   6 declare <16 x half> @llvm.experimental.constrained.fsub.v16f16(<16 x half>, <16 x half>, metadata, metadata)
   7 declare <16 x half> @llvm.experimental.constrained.fmul.v16f16(<16 x half>, <16 x half>, metadata, metadata)
   8 declare <16 x half> @llvm.experimental.constrained.fdiv.v16f16(<16 x half>, <16 x half>, metadata, metadata)
   9 declare <16 x half> @llvm.experimental.constrained.sqrt.v16f16(<16 x half>, metadata, metadata)
  10 declare <16 x half> @llvm.experimental.constrained.fma.v16f16(<16 x half>, <16 x half>, <16 x half>, metadata, metadata)
  11 declare <4 x double> @llvm.experimental.constrained.fpext.v4f64.v4f16(<4 x half>, metadata)
  12 declare <8 x float> @llvm.experimental.constrained.fpext.v8f32.v8f16(<8 x half>, metadata)
  13 declare <4 x half> @llvm.experimental.constrained.fptrunc.v4f16.v4f64(<4 x double>, metadata, metadata)
  14 declare <8 x half> @llvm.experimental.constrained.fptrunc.v8f16.v8f32(<8 x float>, metadata, metadata)
  15 declare <16 x half> @llvm.experimental.constrained.ceil.v16f16(<16 x half>, metadata)
  16 declare <16 x half> @llvm.experimental.constrained.floor.v16f16(<16 x half>, metadata)
  17 declare <16 x half> @llvm.experimental.constrained.trunc.v16f16(<16 x half>, metadata)
  18 declare <16 x half> @llvm.experimental.constrained.rint.v16f16(<16 x half>, metadata, metadata)
  19 declare <16 x half> @llvm.experimental.constrained.nearbyint.v16f16(<16 x half>, metadata, metadata)
  20
  21 define <16 x half> @f2(<16 x half> %a, <16 x half> %b) #0 {
  22 ; CHECK-LABEL: f2:
  23 ; CHECK:       # %bb.0:
  24 ; CHECK-NEXT:    vaddph %ymm1, %ymm0, %ymm0
  25 ; CHECK-NEXT:    ret{{[l|q]}}
  26   %ret = call <16 x half> @llvm.experimental.constrained.fadd.v16f16(<16 x half> %a, <16 x half> %b,
  27                                                                      metadata !"round.dynamic",
  28                                                                      metadata !"fpexcept.strict") #0
  29   ret <16 x half> %ret
  30 }
  31
  32 define <16 x half> @f4(<16 x half> %a, <16 x half> %b) #0 {
  33 ; CHECK-LABEL: f4:
  34 ; CHECK:       # %bb.0:
  35 ; CHECK-NEXT:    vsubph %ymm1, %ymm0, %ymm0
  36 ; CHECK-NEXT:    ret{{[l|q]}}
  37   %ret = call <16 x half> @llvm.experimental.constrained.fsub.v16f16(<16 x half> %a, <16 x half> %b,
  38                                                                      metadata !"round.dynamic",
  39                                                                      metadata !"fpexcept.strict") #0
  40   ret <16 x half> %ret
  41 }
  42
  43 define <16 x half> @f6(<16 x half> %a, <16 x half> %b) #0 {
  44 ; CHECK-LABEL: f6:
  45 ; CHECK:       # %bb.0:
  46 ; CHECK-NEXT:    vmulph %ymm1, %ymm0, %ymm0
  47 ; CHECK-NEXT:    ret{{[l|q]}}
  48   %ret = call <16 x half> @llvm.experimental.constrained.fmul.v16f16(<16 x half> %a, <16 x half> %b,
  49                                                                      metadata !"round.dynamic",
  50                                                                      metadata !"fpexcept.strict") #0
  51   ret <16 x half> %ret
  52 }
  53
  54 define <16 x half> @f8(<16 x half> %a, <16 x half> %b) #0 {
  55 ; CHECK-LABEL: f8:
  56 ; CHECK:       # %bb.0:
  57 ; CHECK-NEXT:    vdivph %ymm1, %ymm0, %ymm0
  58 ; CHECK-NEXT:    ret{{[l|q]}}
  59   %ret = call <16 x half> @llvm.experimental.constrained.fdiv.v16f16(<16 x half> %a, <16 x half> %b,
  60                                                                      metadata !"round.dynamic",
  61                                                                      metadata !"fpexcept.strict") #0
  62   ret <16 x half> %ret
  63 }
  64
  65
  66 define <16 x half> @f10(<16 x half> %a) #0 {
  67 ; CHECK-LABEL: f10:
  68 ; CHECK:       # %bb.0:
  69 ; CHECK-NEXT:    vsqrtph %ymm0, %ymm0
  70 ; CHECK-NEXT:    ret{{[l|q]}}
  71   %ret = call <16 x half> @llvm.experimental.constrained.sqrt.v16f16(
  72                               <16 x half> %a,
  73                               metadata !"round.dynamic",
  74                               metadata !"fpexcept.strict") #0
  75   ret <16 x half > %ret
  76 }
  77
  78 define <4 x double> @f11(<4 x half> %a) #0 {
  79 ; CHECK-LABEL: f11:
  80 ; CHECK:       # %bb.0:
  81 ; CHECK-NEXT:    vcvtph2pd %xmm0, %ymm0
  82 ; CHECK-NEXT:    ret{{[l|q]}}
  83   %ret = call <4 x double> @llvm.experimental.constrained.fpext.v4f64.v4f16(
  84                                 <4 x half> %a,
  85                                 metadata !"fpexcept.strict") #0
  86   ret <4 x double> %ret
  87 }
  88
  89 define <4 x half> @f12(<4 x double> %a) #0 {
  90 ; CHECK-LABEL: f12:
  91 ; CHECK:       # %bb.0:
  92 ; CHECK-NEXT:    vcvtpd2ph %ymm0, %xmm0
  93 ; CHECK-NEXT:    vzeroupper
  94 ; CHECK-NEXT:    ret{{[l|q]}}
  95   %ret = call <4 x half> @llvm.experimental.constrained.fptrunc.v4f16.v4f64(
  96                                 <4 x double> %a,
  97                                 metadata !"round.dynamic",
  98                                 metadata !"fpexcept.strict") #0
  99   ret <4 x half> %ret
 100 }
 101
 102 define <16 x half> @f13(<16 x half> %a, <16 x half> %b, <16 x half> %c) #0 {
 103 ; CHECK-LABEL: f13:
 104 ; CHECK:       # %bb.0:
 105 ; CHECK-NEXT:    vfmadd213ph %ymm2, %ymm1, %ymm0
 106 ; CHECK-NEXT:    ret{{[l|q]}}
 107   %res = call <16 x half> @llvm.experimental.constrained.fma.v16f16(<16 x half> %a, <16 x half> %b, <16 x half> %c,
 108                                                                     metadata !"round.dynamic",
 109                                                                     metadata !"fpexcept.strict") #0
 110   ret <16 x half> %res
 111 }
 112
 113 define <8 x float> @f14(<8 x half> %a) #0 {
 114 ; CHECK-LABEL: f14:
 115 ; CHECK:       # %bb.0:
 116 ; CHECK-NEXT:    vcvtph2psx %xmm0, %ymm0
 117 ; CHECK-NEXT:    ret{{[l|q]}}
 118   %ret = call <8 x float> @llvm.experimental.constrained.fpext.v8f32.v8f16(
 119                                 <8 x half> %a,
 120                                 metadata !"fpexcept.strict") #0
 121   ret <8 x float> %ret
 122 }
 123
 124 define <8 x half> @f15(<8 x float> %a) #0 {
 125 ; CHECK-LABEL: f15:
 126 ; CHECK:       # %bb.0:
 127 ; CHECK-NEXT:    vcvtps2phx %ymm0, %xmm0
 128 ; CHECK-NEXT:    vzeroupper
 129 ; CHECK-NEXT:    ret{{[l|q]}}
 130   %ret = call <8 x half> @llvm.experimental.constrained.fptrunc.v8f16.v8f32(
 131                                 <8 x float> %a,
 132                                 metadata !"round.dynamic",
 133                                 metadata !"fpexcept.strict") #0
 134   ret <8 x half> %ret
 135 }
 136
 137 define <16 x half> @fceilv16f16(<16 x half> %f) #0 {
 138 ; CHECK-LABEL: fceilv16f16:
 139 ; CHECK:       # %bb.0:
 140 ; CHECK-NEXT:    vrndscaleph $10, %ymm0, %ymm0
 141 ; CHECK-NEXT:    ret{{[l|q]}}
 142   %res = call <16 x half> @llvm.experimental.constrained.ceil.v16f16(
 143                           <16 x half> %f, metadata !"fpexcept.strict") #0
 144   ret <16 x half> %res
 145 }
 146
 147 define <16 x half> @ffloorv16f16(<16 x half> %f) #0 {
 148 ; CHECK-LABEL: ffloorv16f16:
 149 ; CHECK:       # %bb.0:
 150 ; CHECK-NEXT:    vrndscaleph $9, %ymm0, %ymm0
 151 ; CHECK-NEXT:    ret{{[l|q]}}
 152   %res = call <16 x half> @llvm.experimental.constrained.floor.v16f16(
 153                           <16 x half> %f, metadata !"fpexcept.strict") #0
 154   ret <16 x half> %res
 155 }
 156
 157
 158 define <16 x half> @ftruncv16f16(<16 x half> %f) #0 {
 159 ; CHECK-LABEL: ftruncv16f16:
 160 ; CHECK:       # %bb.0:
 161 ; CHECK-NEXT:    vrndscaleph $11, %ymm0, %ymm0
 162 ; CHECK-NEXT:    ret{{[l|q]}}
 163   %res = call <16 x half> @llvm.experimental.constrained.trunc.v16f16(
 164                           <16 x half> %f, metadata !"fpexcept.strict") #0
 165   ret <16 x half> %res
 166 }
 167
 168 define <16 x half> @frintv16f16(<16 x half> %f) #0 {
 169 ; CHECK-LABEL: frintv16f16:
 170 ; CHECK:       # %bb.0:
 171 ; CHECK-NEXT:    vrndscaleph $4, %ymm0, %ymm0
 172 ; CHECK-NEXT:    ret{{[l|q]}}
 173   %res = call <16 x half> @llvm.experimental.constrained.rint.v16f16(
 174                           <16 x half> %f,
 175                           metadata !"round.dynamic", metadata !"fpexcept.strict") #0
 176   ret <16 x half> %res
 177 }
 178
 179 define <16 x half> @fnearbyintv16f16(<16 x half> %f) #0 {
 180 ; CHECK-LABEL: fnearbyintv16f16:
 181 ; CHECK:       # %bb.0:
 182 ; CHECK-NEXT:    vrndscaleph $12, %ymm0, %ymm0
 183 ; CHECK-NEXT:    ret{{[l|q]}}
 184   %res = call <16 x half> @llvm.experimental.constrained.nearbyint.v16f16(
 185                           <16 x half> %f,
 186                           metadata !"round.dynamic", metadata !"fpexcept.strict") #0
 187   ret <16 x half> %res
 188 }
 189
 190 attributes #0 = { strictfp }