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=i686-unknown-unknown -mattr=+avx512vbmi --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
      3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
      4 
      5 declare <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
      6 
      7 define <64 x i8>@test_int_x86_avx512_mask_permvar_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
      8 ; X86-LABEL: test_int_x86_avx512_mask_permvar_qi_512:
      9 ; X86:       # %bb.0:
     10 ; X86-NEXT:    vpermb %zmm0, %zmm1, %zmm3 # encoding: [0x62,0xf2,0x75,0x48,0x8d,0xd8]
     11 ; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
     12 ; X86-NEXT:    vpermb %zmm0, %zmm1, %zmm2 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x8d,0xd0]
     13 ; X86-NEXT:    vpermb %zmm0, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x8d,0xc0]
     14 ; X86-NEXT:    vpaddb %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfc,0xc3]
     15 ; X86-NEXT:    vpaddb %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfc,0xc0]
     16 ; X86-NEXT:    retl # encoding: [0xc3]
     17 ;
     18 ; X64-LABEL: test_int_x86_avx512_mask_permvar_qi_512:
     19 ; X64:       # %bb.0:
     20 ; X64-NEXT:    vpermb %zmm0, %zmm1, %zmm3 # encoding: [0x62,0xf2,0x75,0x48,0x8d,0xd8]
     21 ; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
     22 ; X64-NEXT:    vpermb %zmm0, %zmm1, %zmm2 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x8d,0xd0]
     23 ; X64-NEXT:    vpermb %zmm0, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x8d,0xc0]
     24 ; X64-NEXT:    vpaddb %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfc,0xc3]
     25 ; X64-NEXT:    vpaddb %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfc,0xc0]
     26 ; X64-NEXT:    retq # encoding: [0xc3]
     27  %res = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
     28  %res1 = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> zeroinitializer, i64 %x3)
     29  %res2 = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
     30  %res3 = add <64 x i8> %res, %res1
     31  %res4 = add <64 x i8> %res3, %res2
     32  ret <64 x i8> %res4
     33 }
     34 
     35 declare <64 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
     36 
     37 define <64 x i8>@test_int_x86_avx512_mask_vpermi2var_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
     38 ; X86-LABEL: test_int_x86_avx512_mask_vpermi2var_qi_512:
     39 ; X86:       # %bb.0:
     40 ; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
     41 ; X86-NEXT:    vpermt2b %zmm2, %zmm1, %zmm3 # encoding: [0x62,0xf2,0x75,0x48,0x7d,0xda]
     42 ; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
     43 ; X86-NEXT:    vpermi2b %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x75,0xca]
     44 ; X86-NEXT:    vpxor %xmm4, %xmm4, %xmm4 # encoding: [0xc5,0xd9,0xef,0xe4]
     45 ; X86-NEXT:    vpermi2b %zmm2, %zmm0, %zmm4 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x75,0xe2]
     46 ; X86-NEXT:    vpaddb %zmm3, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfc,0xc3]
     47 ; X86-NEXT:    vpaddb %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfc,0xc0]
     48 ; X86-NEXT:    retl # encoding: [0xc3]
     49 ;
     50 ; X64-LABEL: test_int_x86_avx512_mask_vpermi2var_qi_512:
     51 ; X64:       # %bb.0:
     52 ; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
     53 ; X64-NEXT:    vpermt2b %zmm2, %zmm1, %zmm3 # encoding: [0x62,0xf2,0x75,0x48,0x7d,0xda]
     54 ; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
     55 ; X64-NEXT:    vpermi2b %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x75,0xca]
     56 ; X64-NEXT:    vpxor %xmm4, %xmm4, %xmm4 # encoding: [0xc5,0xd9,0xef,0xe4]
     57 ; X64-NEXT:    vpermi2b %zmm2, %zmm0, %zmm4 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x75,0xe2]
     58 ; X64-NEXT:    vpaddb %zmm3, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfc,0xc3]
     59 ; X64-NEXT:    vpaddb %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfc,0xc0]
     60 ; X64-NEXT:    retq # encoding: [0xc3]
     61   %res = call <64 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
     62   %res1 = call <64 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.512(<64 x i8> %x0, <64 x i8> zeroinitializer, <64 x i8> %x2, i64 %x3)
     63   %res2 = call <64 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
     64   %res3 = add <64 x i8> %res, %res1
     65   %res4 = add <64 x i8> %res3, %res2
     66   ret <64 x i8> %res4
     67 }
     68 
     69 declare <64 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
     70 
     71 define <64 x i8>@test_int_x86_avx512_mask_vpermt2var_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
     72 ; X86-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_512:
     73 ; X86:       # %bb.0:
     74 ; X86-NEXT:    vmovdqa64 %zmm1, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd9]
     75 ; X86-NEXT:    vpermt2b %zmm2, %zmm0, %zmm3 # encoding: [0x62,0xf2,0x7d,0x48,0x7d,0xda]
     76 ; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
     77 ; X86-NEXT:    vpermt2b %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x7d,0xca]
     78 ; X86-NEXT:    vpxor %xmm4, %xmm4, %xmm4 # encoding: [0xc5,0xd9,0xef,0xe4]
     79 ; X86-NEXT:    vpermt2b %zmm2, %zmm0, %zmm4 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x7d,0xe2]
     80 ; X86-NEXT:    vpaddb %zmm3, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfc,0xc3]
     81 ; X86-NEXT:    vpaddb %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfc,0xc0]
     82 ; X86-NEXT:    retl # encoding: [0xc3]
     83 ;
     84 ; X64-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_512:
     85 ; X64:       # %bb.0:
     86 ; X64-NEXT:    vmovdqa64 %zmm1, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd9]
     87 ; X64-NEXT:    vpermt2b %zmm2, %zmm0, %zmm3 # encoding: [0x62,0xf2,0x7d,0x48,0x7d,0xda]
     88 ; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
     89 ; X64-NEXT:    vpermt2b %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x7d,0xca]
     90 ; X64-NEXT:    vpxor %xmm4, %xmm4, %xmm4 # encoding: [0xc5,0xd9,0xef,0xe4]
     91 ; X64-NEXT:    vpermt2b %zmm2, %zmm0, %zmm4 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x7d,0xe2]
     92 ; X64-NEXT:    vpaddb %zmm3, %zmm4, %zmm0 # encoding: [0x62,0xf1,0x5d,0x48,0xfc,0xc3]
     93 ; X64-NEXT:    vpaddb %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfc,0xc0]
     94 ; X64-NEXT:    retq # encoding: [0xc3]
     95   %res = call <64 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
     96   %res1 = call <64 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.512(<64 x i8> %x0, <64 x i8> zeroinitializer, <64 x i8> %x2, i64 %x3)
     97   %res2 = call <64 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
     98   %res3 = add <64 x i8> %res, %res1
     99   %res4 = add <64 x i8> %res3, %res2
    100   ret <64 x i8> %res4
    101 }
    102 
    103 declare <64 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
    104 
    105 define <64 x i8>@test_int_x86_avx512_maskz_vpermt2var_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
    106 ; X86-LABEL: test_int_x86_avx512_maskz_vpermt2var_qi_512:
    107 ; X86:       # %bb.0:
    108 ; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
    109 ; X86-NEXT:    vpermi2b %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x75,0xc2]
    110 ; X86-NEXT:    retl # encoding: [0xc3]
    111 ;
    112 ; X64-LABEL: test_int_x86_avx512_maskz_vpermt2var_qi_512:
    113 ; X64:       # %bb.0:
    114 ; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
    115 ; X64-NEXT:    vpermi2b %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x75,0xc2]
    116 ; X64-NEXT:    retq # encoding: [0xc3]
    117   %res = call <64 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
    118   ret <64 x i8> %res
    119 }
    120