llvm/test/CodeGen/X86/vec-strict-512-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 <32 x half> @llvm.experimental.constrained.fadd.v32f16(<32 x half>, <32 x half>, metadata, metadata)
   6 declare <32 x half> @llvm.experimental.constrained.fsub.v32f16(<32 x half>, <32 x half>, metadata, metadata)
   7 declare <32 x half> @llvm.experimental.constrained.fmul.v32f16(<32 x half>, <32 x half>, metadata, metadata)
   8 declare <32 x half> @llvm.experimental.constrained.fdiv.v32f16(<32 x half>, <32 x half>, metadata, metadata)
   9 declare <32 x half> @llvm.experimental.constrained.sqrt.v32f16(<32 x half>, metadata, metadata)
  10 declare <8 x double> @llvm.experimental.constrained.fpext.v8f64.v8f16(<8 x half>, metadata)
  11 declare <16 x float> @llvm.experimental.constrained.fpext.v16f32.v16f16(<16 x half>, metadata)
  12 declare <8 x half> @llvm.experimental.constrained.fptrunc.v8f16.v8f64(<8 x double>, metadata, metadata)
  13 declare <16 x half> @llvm.experimental.constrained.fptrunc.v16f16.v16f32(<16 x float>, metadata, metadata)
  14 declare <32 x half> @llvm.experimental.constrained.fma.v32f16(<32 x half>, <32 x half>, <32 x half>, metadata, metadata)
  15 declare <32 x half> @llvm.experimental.constrained.ceil.v32f16(<32 x half>, metadata)
  16 declare <32 x half> @llvm.experimental.constrained.floor.v32f16(<32 x half>, metadata)
  17 declare <32 x half> @llvm.experimental.constrained.trunc.v32f16(<32 x half>, metadata)
  18 declare <32 x half> @llvm.experimental.constrained.rint.v32f16(<32 x half>, metadata, metadata)
  19 declare <32 x half> @llvm.experimental.constrained.nearbyint.v32f16(<32 x half>, metadata, metadata)
  20
  21 define <32 x half> @f2(<32 x half> %a, <32 x half> %b) #0 {
  22 ; CHECK-LABEL: f2:
  23 ; CHECK:       # %bb.0:
  24 ; CHECK-NEXT:    vaddph %zmm1, %zmm0, %zmm0
  25 ; CHECK-NEXT:    ret{{[l|q]}}
  26   %ret = call <32 x half> @llvm.experimental.constrained.fadd.v32f16(<32 x half> %a, <32 x half> %b,
  27                                                                      metadata !"round.dynamic",
  28                                                                      metadata !"fpexcept.strict") #0
  29   ret <32 x half> %ret
  30 }
  31
  32 define <32 x half> @f4(<32 x half> %a, <32 x half> %b) #0 {
  33 ; CHECK-LABEL: f4:
  34 ; CHECK:       # %bb.0:
  35 ; CHECK-NEXT:    vsubph %zmm1, %zmm0, %zmm0
  36 ; CHECK-NEXT:    ret{{[l|q]}}
  37   %ret = call <32 x half> @llvm.experimental.constrained.fsub.v32f16(<32 x half> %a, <32 x half> %b,
  38                                                                      metadata !"round.dynamic",
  39                                                                      metadata !"fpexcept.strict") #0
  40   ret <32 x half> %ret
  41 }
  42
  43 define <32 x half> @f6(<32 x half> %a, <32 x half> %b) #0 {
  44 ; CHECK-LABEL: f6:
  45 ; CHECK:       # %bb.0:
  46 ; CHECK-NEXT:    vmulph %zmm1, %zmm0, %zmm0
  47 ; CHECK-NEXT:    ret{{[l|q]}}
  48   %ret = call <32 x half> @llvm.experimental.constrained.fmul.v32f16(<32 x half> %a, <32 x half> %b,
  49                                                                      metadata !"round.dynamic",
  50                                                                      metadata !"fpexcept.strict") #0
  51   ret <32 x half> %ret
  52 }
  53
  54 define <32 x half> @f8(<32 x half> %a, <32 x half> %b) #0 {
  55 ; CHECK-LABEL: f8:
  56 ; CHECK:       # %bb.0:
  57 ; CHECK-NEXT:    vdivph %zmm1, %zmm0, %zmm0
  58 ; CHECK-NEXT:    ret{{[l|q]}}
  59   %ret = call <32 x half> @llvm.experimental.constrained.fdiv.v32f16(<32 x half> %a, <32 x half> %b,
  60                                                                      metadata !"round.dynamic",
  61                                                                      metadata !"fpexcept.strict") #0
  62   ret <32 x half> %ret
  63 }
  64
  65 define <32 x half> @f10(<32 x half> %a) #0 {
  66 ; CHECK-LABEL: f10:
  67 ; CHECK:       # %bb.0:
  68 ; CHECK-NEXT:    vsqrtph %zmm0, %zmm0
  69 ; CHECK-NEXT:    ret{{[l|q]}}
  70   %ret = call <32 x half> @llvm.experimental.constrained.sqrt.v32f16(
  71                               <32 x half> %a,
  72                               metadata !"round.dynamic",
  73                               metadata !"fpexcept.strict") #0
  74   ret <32 x half > %ret
  75 }
  76
  77 define <8 x double> @f11(<8 x half> %a) #0 {
  78 ; CHECK-LABEL: f11:
  79 ; CHECK:       # %bb.0:
  80 ; CHECK-NEXT:    vcvtph2pd %xmm0, %zmm0
  81 ; CHECK-NEXT:    ret{{[l|q]}}
  82   %ret = call <8 x double> @llvm.experimental.constrained.fpext.v8f64.v8f16(
  83                                 <8 x half> %a,
  84                                 metadata !"fpexcept.strict") #0
  85   ret <8 x double> %ret
  86 }
  87
  88 define <8 x half> @f12(<8 x double> %a) #0 {
  89 ; CHECK-LABEL: f12:
  90 ; CHECK:       # %bb.0:
  91 ; CHECK-NEXT:    vcvtpd2ph %zmm0, %xmm0
  92 ; CHECK-NEXT:    vzeroupper
  93 ; CHECK-NEXT:    ret{{[l|q]}}
  94   %ret = call <8 x half> @llvm.experimental.constrained.fptrunc.v8f16.v8f64(
  95                                 <8 x double> %a,
  96                                 metadata !"round.dynamic",
  97                                 metadata !"fpexcept.strict") #0
  98   ret <8 x half> %ret
  99 }
 100
 101 define <32 x half> @f13(<32 x half> %a, <32 x half> %b, <32 x half> %c) #0 {
 102 ; CHECK-LABEL: f13:
 103 ; CHECK:       # %bb.0:
 104 ; CHECK-NEXT:    vfmadd213ph %zmm2, %zmm1, %zmm0
 105 ; CHECK-NEXT:    ret{{[l|q]}}
 106   %res = call <32 x half> @llvm.experimental.constrained.fma.v32f16(<32 x half> %a, <32 x half> %b, <32 x half> %c,
 107                                                                     metadata !"round.dynamic",
 108                                                                     metadata !"fpexcept.strict") #0
 109   ret <32 x half> %res
 110 }
 111
 112 define <16 x float> @f14(<16 x half> %a) #0 {
 113 ; CHECK-LABEL: f14:
 114 ; CHECK:       # %bb.0:
 115 ; CHECK-NEXT:    vcvtph2psx %ymm0, %zmm0
 116 ; CHECK-NEXT:    ret{{[l|q]}}
 117   %ret = call <16 x float> @llvm.experimental.constrained.fpext.v16f32.v16f16(
 118                                 <16 x half> %a,
 119                                 metadata !"fpexcept.strict") #0
 120   ret <16 x float> %ret
 121 }
 122
 123 define <16 x half> @f15(<16 x float> %a) #0 {
 124 ; CHECK-LABEL: f15:
 125 ; CHECK:       # %bb.0:
 126 ; CHECK-NEXT:    vcvtps2phx %zmm0, %ymm0
 127 ; CHECK-NEXT:    ret{{[l|q]}}
 128   %ret = call <16 x half> @llvm.experimental.constrained.fptrunc.v16f16.v16f32(
 129                                 <16 x float> %a,
 130                                 metadata !"round.dynamic",
 131                                 metadata !"fpexcept.strict") #0
 132   ret <16 x half> %ret
 133 }
 134
 135 define <32 x half> @strict_vector_fceil_v32f16(<32 x half> %f) #0 {
 136 ; CHECK-LABEL: strict_vector_fceil_v32f16:
 137 ; CHECK:       # %bb.0:
 138 ; CHECK-NEXT:    vrndscaleph $10, %zmm0, %zmm0
 139 ; CHECK-NEXT:    ret{{[l|q]}}
 140   %res = call <32 x half> @llvm.experimental.constrained.ceil.v32f16(<32 x half> %f, metadata !"fpexcept.strict") #0
 141   ret <32 x half> %res
 142 }
 143
 144 define <32 x half> @strict_vector_ffloor_v32f16(<32 x half> %f) #0 {
 145 ; CHECK-LABEL: strict_vector_ffloor_v32f16:
 146 ; CHECK:       # %bb.0:
 147 ; CHECK-NEXT:    vrndscaleph $9, %zmm0, %zmm0
 148 ; CHECK-NEXT:    ret{{[l|q]}}
 149   %res = call <32 x half> @llvm.experimental.constrained.floor.v32f16(<32 x half> %f, metadata !"fpexcept.strict") #0
 150   ret <32 x half> %res
 151 }
 152
 153 define <32 x half> @strict_vector_ftrunc_v32f16(<32 x half> %f) #0 {
 154 ; CHECK-LABEL: strict_vector_ftrunc_v32f16:
 155 ; CHECK:       # %bb.0:
 156 ; CHECK-NEXT:    vrndscaleph $11, %zmm0, %zmm0
 157 ; CHECK-NEXT:    ret{{[l|q]}}
 158   %res = call <32 x half> @llvm.experimental.constrained.trunc.v32f16(<32 x half> %f, metadata !"fpexcept.strict") #0
 159   ret <32 x half> %res
 160 }
 161
 162 define <32 x half> @strict_vector_frint_v32f16(<32 x half> %f) #0 {
 163 ; CHECK-LABEL: strict_vector_frint_v32f16:
 164 ; CHECK:       # %bb.0:
 165 ; CHECK-NEXT:    vrndscaleph $4, %zmm0, %zmm0
 166 ; CHECK-NEXT:    ret{{[l|q]}}
 167   %res = call <32 x half> @llvm.experimental.constrained.rint.v32f16(<32 x half> %f,
 168                              metadata !"round.dynamic", metadata !"fpexcept.strict") #0
 169   ret <32 x half> %res
 170 }
 171
 172 define <32 x half> @strict_vector_fnearbyint_v32f16(<32 x half> %f) #0 {
 173 ; CHECK-LABEL: strict_vector_fnearbyint_v32f16:
 174 ; CHECK:       # %bb.0:
 175 ; CHECK-NEXT:    vrndscaleph $12, %zmm0, %zmm0
 176 ; CHECK-NEXT:    ret{{[l|q]}}
 177   %res = call <32 x half> @llvm.experimental.constrained.nearbyint.v32f16(<32 x half> %f,
 178                              metadata !"round.dynamic", metadata !"fpexcept.strict") #0
 179   ret <32 x half> %res
 180 }
 181
 182 attributes #0 = { strictfp }