llvm/test/CodeGen/X86/vec-strict-inttofp-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 -O3 | FileCheck %s --check-prefixes=CHECK,X86
   3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx512fp16 -O3 | FileCheck %s --check-prefixes=CHECK,X64
   4
   5 declare <32 x half> @llvm.experimental.constrained.sitofp.v32f16.v32i1(<32 x i1>, metadata, metadata)
   6 declare <32 x half> @llvm.experimental.constrained.uitofp.v32f16.v32i1(<32 x i1>, metadata, metadata)
   7 declare <32 x half> @llvm.experimental.constrained.sitofp.v32f16.v32i8(<32 x i8>, metadata, metadata)
   8 declare <32 x half> @llvm.experimental.constrained.uitofp.v32f16.v32i8(<32 x i8>, metadata, metadata)
   9 declare <32 x half> @llvm.experimental.constrained.sitofp.v32f16.v32i16(<32 x i16>, metadata, metadata)
  10 declare <32 x half> @llvm.experimental.constrained.uitofp.v32f16.v32i16(<32 x i16>, metadata, metadata)
  11 declare <16 x half> @llvm.experimental.constrained.sitofp.v16f16.v16i32(<16 x i32>, metadata, metadata)
  12 declare <16 x half> @llvm.experimental.constrained.uitofp.v16f16.v16i32(<16 x i32>, metadata, metadata)
  13 declare <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i64(<8 x i64>, metadata, metadata)
  14 declare <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i64(<8 x i64>, metadata, metadata)
  15
  16 define <32 x half> @sitofp_v32i1_v32f16(<32 x i1> %x) #0 {
  17 ; CHECK-LABEL: sitofp_v32i1_v32f16:
  18 ; CHECK:       # %bb.0:
  19 ; CHECK-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
  20 ; CHECK-NEXT:    vpsllw $15, %zmm0, %zmm0
  21 ; CHECK-NEXT:    vpsraw $15, %zmm0, %zmm0
  22 ; CHECK-NEXT:    vcvtw2ph %zmm0, %zmm0
  23 ; CHECK-NEXT:    ret{{[l|q]}}
  24  %result = call <32 x half> @llvm.experimental.constrained.sitofp.v32f16.v32i1(<32 x i1> %x,
  25                                                               metadata !"round.dynamic",
  26                                                               metadata !"fpexcept.strict") #0
  27   ret <32 x half> %result
  28 }
  29
  30 define <32 x half> @uitofp_v32i1_v32f16(<32 x i1> %x) #0 {
  31 ; X86-LABEL: uitofp_v32i1_v32f16:
  32 ; X86:       # %bb.0:
  33 ; X86-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %ymm0
  34 ; X86-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
  35 ; X86-NEXT:    vcvtuw2ph %zmm0, %zmm0
  36 ; X86-NEXT:    retl
  37 ;
  38 ; X64-LABEL: uitofp_v32i1_v32f16:
  39 ; X64:       # %bb.0:
  40 ; X64-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
  41 ; X64-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
  42 ; X64-NEXT:    vcvtuw2ph %zmm0, %zmm0
  43 ; X64-NEXT:    retq
  44  %result = call <32 x half> @llvm.experimental.constrained.uitofp.v32f16.v32i1(<32 x i1> %x,
  45                                                               metadata !"round.dynamic",
  46                                                               metadata !"fpexcept.strict") #0
  47   ret <32 x half> %result
  48 }
  49
  50 define <32 x half> @sitofp_v32i8_v32f16(<32 x i8> %x) #0 {
  51 ; CHECK-LABEL: sitofp_v32i8_v32f16:
  52 ; CHECK:       # %bb.0:
  53 ; CHECK-NEXT:    vpmovsxbw %ymm0, %zmm0
  54 ; CHECK-NEXT:    vcvtw2ph %zmm0, %zmm0
  55 ; CHECK-NEXT:    ret{{[l|q]}}
  56  %result = call <32 x half> @llvm.experimental.constrained.sitofp.v32f16.v32i8(<32 x i8> %x,
  57                                                               metadata !"round.dynamic",
  58                                                               metadata !"fpexcept.strict") #0
  59   ret <32 x half> %result
  60 }
  61
  62 define <32 x half> @uitofp_v32i8_v32f16(<32 x i8> %x) #0 {
  63 ; CHECK-LABEL: uitofp_v32i8_v32f16:
  64 ; CHECK:       # %bb.0:
  65 ; CHECK-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
  66 ; CHECK-NEXT:    vcvtuw2ph %zmm0, %zmm0
  67 ; CHECK-NEXT:    ret{{[l|q]}}
  68  %result = call <32 x half> @llvm.experimental.constrained.uitofp.v32f16.v32i8(<32 x i8> %x,
  69                                                               metadata !"round.dynamic",
  70                                                               metadata !"fpexcept.strict") #0
  71   ret <32 x half> %result
  72 }
  73
  74 define <32 x half> @sitofp_v32i16_v32f16(<32 x i16> %x) #0 {
  75 ; CHECK-LABEL: sitofp_v32i16_v32f16:
  76 ; CHECK:       # %bb.0:
  77 ; CHECK-NEXT:    vcvtw2ph %zmm0, %zmm0
  78 ; CHECK-NEXT:    ret{{[l|q]}}
  79  %result = call <32 x half> @llvm.experimental.constrained.sitofp.v32f16.v32i16(<32 x i16> %x,
  80                                                               metadata !"round.dynamic",
  81                                                               metadata !"fpexcept.strict") #0
  82   ret <32 x half> %result
  83 }
  84
  85 define <32 x half> @uitofp_v32i16_v32f16(<32 x i16> %x) #0 {
  86 ; CHECK-LABEL: uitofp_v32i16_v32f16:
  87 ; CHECK:       # %bb.0:
  88 ; CHECK-NEXT:    vcvtuw2ph %zmm0, %zmm0
  89 ; CHECK-NEXT:    ret{{[l|q]}}
  90  %result = call <32 x half> @llvm.experimental.constrained.uitofp.v32f16.v32i16(<32 x i16> %x,
  91                                                               metadata !"round.dynamic",
  92                                                               metadata !"fpexcept.strict") #0
  93   ret <32 x half> %result
  94 }
  95
  96 define <16 x half> @sitofp_v16i32_v16f16(<16 x i32> %x) #0 {
  97 ; CHECK-LABEL: sitofp_v16i32_v16f16:
  98 ; CHECK:       # %bb.0:
  99 ; CHECK-NEXT:    vcvtdq2ph %zmm0, %ymm0
 100 ; CHECK-NEXT:    ret{{[l|q]}}
 101  %result = call <16 x half> @llvm.experimental.constrained.sitofp.v16f16.v16i32(<16 x i32> %x,
 102                                                               metadata !"round.dynamic",
 103                                                               metadata !"fpexcept.strict") #0
 104   ret <16 x half> %result
 105 }
 106
 107 define <16 x half> @uitofp_v16i32_v16f16(<16 x i32> %x) #0 {
 108 ; CHECK-LABEL: uitofp_v16i32_v16f16:
 109 ; CHECK:       # %bb.0:
 110 ; CHECK-NEXT:    vcvtudq2ph %zmm0, %ymm0
 111 ; CHECK-NEXT:    ret{{[l|q]}}
 112  %result = call <16 x half> @llvm.experimental.constrained.uitofp.v16f16.v16i32(<16 x i32> %x,
 113                                                               metadata !"round.dynamic",
 114                                                               metadata !"fpexcept.strict") #0
 115   ret <16 x half> %result
 116 }
 117
 118 define <8 x half> @sitofp_v8i64_v8f16(<8 x i64> %x) #0 {
 119 ; CHECK-LABEL: sitofp_v8i64_v8f16:
 120 ; CHECK:       # %bb.0:
 121 ; CHECK-NEXT:    vcvtqq2ph %zmm0, %xmm0
 122 ; CHECK-NEXT:    vzeroupper
 123 ; CHECK-NEXT:    ret{{[l|q]}}
 124  %result = call <8 x half> @llvm.experimental.constrained.sitofp.v8f16.v8i64(<8 x i64> %x,
 125                                                               metadata !"round.dynamic",
 126                                                               metadata !"fpexcept.strict") #0
 127   ret <8 x half> %result
 128 }
 129
 130 define <8 x half> @uitofp_v8i64_v8f16(<8 x i64> %x) #0 {
 131 ; CHECK-LABEL: uitofp_v8i64_v8f16:
 132 ; CHECK:       # %bb.0:
 133 ; CHECK-NEXT:    vcvtuqq2ph %zmm0, %xmm0
 134 ; CHECK-NEXT:    vzeroupper
 135 ; CHECK-NEXT:    ret{{[l|q]}}
 136  %result = call <8 x half> @llvm.experimental.constrained.uitofp.v8f16.v8i64(<8 x i64> %x,
 137                                                               metadata !"round.dynamic",
 138                                                               metadata !"fpexcept.strict") #0
 139   ret <8 x half> %result
 140 }
 141
 142 attributes #0 = { strictfp }