Home | History | Annotate | Download | only in X86
      1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
      2 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl --show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512-KNL
      3 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx --show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512-SKX
      4 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx --show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=AVX
      5 
      6 define float @test_fdiv(float %a, float %b) {
      7 ; AVX512-LABEL: test_fdiv:
      8 ; AVX512:       ## %bb.0:
      9 ; AVX512-NEXT:    vdivss %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x5e,0xc1]
     10 ; AVX512-NEXT:    retq ## encoding: [0xc3]
     11 ;
     12 ; AVX-LABEL: test_fdiv:
     13 ; AVX:       ## %bb.0:
     14 ; AVX-NEXT:    vdivss %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xfa,0x5e,0xc1]
     15 ; AVX-NEXT:    retq ## encoding: [0xc3]
     16   %c = fdiv float %a, %b
     17   ret float %c
     18 }
     19 
     20 define float @test_fsub(float %a, float %b) {
     21 ; AVX512-LABEL: test_fsub:
     22 ; AVX512:       ## %bb.0:
     23 ; AVX512-NEXT:    vsubss %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x5c,0xc1]
     24 ; AVX512-NEXT:    retq ## encoding: [0xc3]
     25 ;
     26 ; AVX-LABEL: test_fsub:
     27 ; AVX:       ## %bb.0:
     28 ; AVX-NEXT:    vsubss %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xfa,0x5c,0xc1]
     29 ; AVX-NEXT:    retq ## encoding: [0xc3]
     30   %c = fsub float %a, %b
     31   ret float %c
     32 }
     33 
     34 define double @test_fadd(double %a, double %b) {
     35 ; AVX512-LABEL: test_fadd:
     36 ; AVX512:       ## %bb.0:
     37 ; AVX512-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfb,0x58,0xc1]
     38 ; AVX512-NEXT:    retq ## encoding: [0xc3]
     39 ;
     40 ; AVX-LABEL: test_fadd:
     41 ; AVX:       ## %bb.0:
     42 ; AVX-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xfb,0x58,0xc1]
     43 ; AVX-NEXT:    retq ## encoding: [0xc3]
     44   %c = fadd double %a, %b
     45   ret double %c
     46 }
     47 
     48 declare float     @llvm.trunc.f32(float  %Val)
     49 declare double    @llvm.trunc.f64(double %Val)
     50 declare float     @llvm.rint.f32(float  %Val)
     51 declare double    @llvm.rint.f64(double %Val)
     52 declare double    @llvm.sqrt.f64(double %Val)
     53 declare float     @llvm.sqrt.f32(float  %Val)
     54 
     55 define float @test_trunc(float %a) {
     56 ; AVX512-LABEL: test_trunc:
     57 ; AVX512:       ## %bb.0:
     58 ; AVX512-NEXT:    vroundss $11, %xmm0, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x79,0x0a,0xc0,0x0b]
     59 ; AVX512-NEXT:    retq ## encoding: [0xc3]
     60 ;
     61 ; AVX-LABEL: test_trunc:
     62 ; AVX:       ## %bb.0:
     63 ; AVX-NEXT:    vroundss $11, %xmm0, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0a,0xc0,0x0b]
     64 ; AVX-NEXT:    retq ## encoding: [0xc3]
     65   %c = call float @llvm.trunc.f32(float %a)
     66   ret float %c
     67 }
     68 
     69 define double @test_sqrt(double %a) {
     70 ; AVX512-LABEL: test_sqrt:
     71 ; AVX512:       ## %bb.0:
     72 ; AVX512-NEXT:    vsqrtsd %xmm0, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfb,0x51,0xc0]
     73 ; AVX512-NEXT:    retq ## encoding: [0xc3]
     74 ;
     75 ; AVX-LABEL: test_sqrt:
     76 ; AVX:       ## %bb.0:
     77 ; AVX-NEXT:    vsqrtsd %xmm0, %xmm0, %xmm0 ## encoding: [0xc5,0xfb,0x51,0xc0]
     78 ; AVX-NEXT:    retq ## encoding: [0xc3]
     79   %c = call double @llvm.sqrt.f64(double %a)
     80   ret double %c
     81 }
     82 
     83 define float @test_rint(float %a) {
     84 ; AVX512-LABEL: test_rint:
     85 ; AVX512:       ## %bb.0:
     86 ; AVX512-NEXT:    vroundss $4, %xmm0, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x79,0x0a,0xc0,0x04]
     87 ; AVX512-NEXT:    retq ## encoding: [0xc3]
     88 ;
     89 ; AVX-LABEL: test_rint:
     90 ; AVX:       ## %bb.0:
     91 ; AVX-NEXT:    vroundss $4, %xmm0, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0a,0xc0,0x04]
     92 ; AVX-NEXT:    retq ## encoding: [0xc3]
     93   %c = call float @llvm.rint.f32(float %a)
     94   ret float %c
     95 }
     96 
     97 define float @test_vmax(float %i, float %j) {
     98 ; AVX512-LABEL: test_vmax:
     99 ; AVX512:       ## %bb.0:
    100 ; AVX512-NEXT:    vmaxss %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x5f,0xc1]
    101 ; AVX512-NEXT:    retq ## encoding: [0xc3]
    102 ;
    103 ; AVX-LABEL: test_vmax:
    104 ; AVX:       ## %bb.0:
    105 ; AVX-NEXT:    vmaxss %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xfa,0x5f,0xc1]
    106 ; AVX-NEXT:    retq ## encoding: [0xc3]
    107   %cmp_res = fcmp ogt float %i, %j
    108   %max = select i1 %cmp_res, float %i, float %j
    109   ret float %max
    110 }
    111 
    112 define float @test_mov(float %a, float %b, float %i, float %j) {
    113 ; AVX512-LABEL: test_mov:
    114 ; AVX512:       ## %bb.0:
    115 ; AVX512-NEXT:    vcmpltss %xmm2, %xmm3, %k1 ## encoding: [0x62,0xf1,0x66,0x08,0xc2,0xca,0x01]
    116 ; AVX512-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1} ## encoding: [0x62,0xf1,0x7e,0x09,0x10,0xc1]
    117 ; AVX512-NEXT:    retq ## encoding: [0xc3]
    118 ;
    119 ; AVX-LABEL: test_mov:
    120 ; AVX:       ## %bb.0:
    121 ; AVX-NEXT:    vcmpltss %xmm2, %xmm3, %xmm2 ## encoding: [0xc5,0xe2,0xc2,0xd2,0x01]
    122 ; AVX-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x4a,0xc1,0x20]
    123 ; AVX-NEXT:    retq ## encoding: [0xc3]
    124   %cmp_res = fcmp ogt float %i, %j
    125   %max = select i1 %cmp_res, float %b, float %a
    126   ret float %max
    127 }
    128 
    129 define float @zero_float(float %a) {
    130 ; AVX512-KNL-LABEL: zero_float:
    131 ; AVX512-KNL:       ## %bb.0:
    132 ; AVX512-KNL-NEXT:    vxorps %xmm1, %xmm1, %xmm1 ## encoding: [0xc5,0xf0,0x57,0xc9]
    133 ; AVX512-KNL-NEXT:    vaddss %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x58,0xc1]
    134 ; AVX512-KNL-NEXT:    retq ## encoding: [0xc3]
    135 ;
    136 ; AVX512-SKX-LABEL: zero_float:
    137 ; AVX512-SKX:       ## %bb.0:
    138 ; AVX512-SKX-NEXT:    vxorps %xmm1, %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf0,0x57,0xc9]
    139 ; AVX512-SKX-NEXT:    vaddss %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x58,0xc1]
    140 ; AVX512-SKX-NEXT:    retq ## encoding: [0xc3]
    141 ;
    142 ; AVX-LABEL: zero_float:
    143 ; AVX:       ## %bb.0:
    144 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1 ## encoding: [0xc5,0xf0,0x57,0xc9]
    145 ; AVX-NEXT:    vaddss %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xfa,0x58,0xc1]
    146 ; AVX-NEXT:    retq ## encoding: [0xc3]
    147   %b = fadd float %a, 0.0
    148   ret float %b
    149 }
    150 
    151 define double @zero_double(double %a) {
    152 ; AVX512-KNL-LABEL: zero_double:
    153 ; AVX512-KNL:       ## %bb.0:
    154 ; AVX512-KNL-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x57,0xc9]
    155 ; AVX512-KNL-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfb,0x58,0xc1]
    156 ; AVX512-KNL-NEXT:    retq ## encoding: [0xc3]
    157 ;
    158 ; AVX512-SKX-LABEL: zero_double:
    159 ; AVX512-SKX:       ## %bb.0:
    160 ; AVX512-SKX-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0x57,0xc9]
    161 ; AVX512-SKX-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfb,0x58,0xc1]
    162 ; AVX512-SKX-NEXT:    retq ## encoding: [0xc3]
    163 ;
    164 ; AVX-LABEL: zero_double:
    165 ; AVX:       ## %bb.0:
    166 ; AVX-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x57,0xc9]
    167 ; AVX-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xfb,0x58,0xc1]
    168 ; AVX-NEXT:    retq ## encoding: [0xc3]
    169   %b = fadd double %a, 0.0
    170   ret double %b
    171 }
    172