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