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-unknown-unknown -mattr=+sse2 | FileCheck %s
      3 
      4 declare <2 x i64> @llvm.cttz.v2i64(<2 x i64>, i1)
      5 declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1)
      6 declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>)
      7 
      8 define <2 x i64> @footz(<2 x i64> %a) nounwind {
      9 ; CHECK-LABEL: footz:
     10 ; CHECK:       # BB#0:
     11 ; CHECK-NEXT:    movd %xmm0, %rax
     12 ; CHECK-NEXT:    bsfq %rax, %rax
     13 ; CHECK-NEXT:    movd %rax, %xmm1
     14 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
     15 ; CHECK-NEXT:    movd %xmm0, %rax
     16 ; CHECK-NEXT:    bsfq %rax, %rax
     17 ; CHECK-NEXT:    movd %rax, %xmm0
     18 ; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
     19 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
     20 ; CHECK-NEXT:    retq
     21   %c = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
     22   ret <2 x i64> %c
     23 
     24 }
     25 define <2 x i64> @foolz(<2 x i64> %a) nounwind {
     26 ; CHECK-LABEL: foolz:
     27 ; CHECK:       # BB#0:
     28 ; CHECK-NEXT:    movd %xmm0, %rax
     29 ; CHECK-NEXT:    bsrq %rax, %rax
     30 ; CHECK-NEXT:    xorq $63, %rax
     31 ; CHECK-NEXT:    movd %rax, %xmm1
     32 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
     33 ; CHECK-NEXT:    movd %xmm0, %rax
     34 ; CHECK-NEXT:    bsrq %rax, %rax
     35 ; CHECK-NEXT:    xorq $63, %rax
     36 ; CHECK-NEXT:    movd %rax, %xmm0
     37 ; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
     38 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
     39 ; CHECK-NEXT:    retq
     40   %c = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
     41   ret <2 x i64> %c
     42 
     43 }
     44 
     45 define <2 x i64> @foopop(<2 x i64> %a) nounwind {
     46 ; CHECK-LABEL: foopop:
     47 ; CHECK:       # BB#0:
     48 ; CHECK-NEXT:    movdqa %xmm0, %xmm1
     49 ; CHECK-NEXT:    psrlq $1, %xmm1
     50 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
     51 ; CHECK-NEXT:    psubq %xmm1, %xmm0
     52 ; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [3689348814741910323,3689348814741910323]
     53 ; CHECK-NEXT:    movdqa %xmm0, %xmm2
     54 ; CHECK-NEXT:    pand %xmm1, %xmm2
     55 ; CHECK-NEXT:    psrlq $2, %xmm0
     56 ; CHECK-NEXT:    pand %xmm1, %xmm0
     57 ; CHECK-NEXT:    paddq %xmm2, %xmm0
     58 ; CHECK-NEXT:    movdqa %xmm0, %xmm1
     59 ; CHECK-NEXT:    psrlq $4, %xmm1
     60 ; CHECK-NEXT:    paddq %xmm0, %xmm1
     61 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
     62 ; CHECK-NEXT:    pxor %xmm0, %xmm0
     63 ; CHECK-NEXT:    psadbw %xmm0, %xmm1
     64 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
     65 ; CHECK-NEXT:    retq
     66   %c = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
     67   ret <2 x i64> %c
     68 }
     69 
     70 declare <2 x i32> @llvm.cttz.v2i32(<2 x i32>, i1)
     71 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1)
     72 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32>)
     73 
     74 define <2 x i32> @promtz(<2 x i32> %a) nounwind {
     75 ; CHECK-LABEL: promtz:
     76 ; CHECK:       # BB#0:
     77 ; CHECK-NEXT:    por {{.*}}(%rip), %xmm0
     78 ; CHECK-NEXT:    movd %xmm0, %rax
     79 ; CHECK-NEXT:    bsfq %rax, %rax
     80 ; CHECK-NEXT:    movl $64, %ecx
     81 ; CHECK-NEXT:    cmoveq %rcx, %rax
     82 ; CHECK-NEXT:    movd %rax, %xmm1
     83 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
     84 ; CHECK-NEXT:    movd %xmm0, %rax
     85 ; CHECK-NEXT:    bsfq %rax, %rax
     86 ; CHECK-NEXT:    cmoveq %rcx, %rax
     87 ; CHECK-NEXT:    movd %rax, %xmm0
     88 ; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
     89 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
     90 ; CHECK-NEXT:    retq
     91   %c = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %a, i1 false)
     92   ret <2 x i32> %c
     93 
     94 }
     95 define <2 x i32> @promlz(<2 x i32> %a) nounwind {
     96 ; CHECK-LABEL: promlz:
     97 ; CHECK:       # BB#0:
     98 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
     99 ; CHECK-NEXT:    movd %xmm0, %rax
    100 ; CHECK-NEXT:    bsrq %rax, %rax
    101 ; CHECK-NEXT:    movl $127, %ecx
    102 ; CHECK-NEXT:    cmoveq %rcx, %rax
    103 ; CHECK-NEXT:    xorq $63, %rax
    104 ; CHECK-NEXT:    movd %rax, %xmm1
    105 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
    106 ; CHECK-NEXT:    movd %xmm0, %rax
    107 ; CHECK-NEXT:    bsrq %rax, %rax
    108 ; CHECK-NEXT:    cmoveq %rcx, %rax
    109 ; CHECK-NEXT:    xorq $63, %rax
    110 ; CHECK-NEXT:    movd %rax, %xmm0
    111 ; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
    112 ; CHECK-NEXT:    psubq {{.*}}(%rip), %xmm1
    113 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
    114 ; CHECK-NEXT:    retq
    115   %c = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false)
    116   ret <2 x i32> %c
    117 
    118 }
    119 
    120 define <2 x i32> @prompop(<2 x i32> %a) nounwind {
    121 ; CHECK-LABEL: prompop:
    122 ; CHECK:       # BB#0:
    123 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
    124 ; CHECK-NEXT:    pxor %xmm2, %xmm2
    125 ; CHECK-NEXT:    movdqa %xmm0, %xmm1
    126 ; CHECK-NEXT:    psrlq $1, %xmm1
    127 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
    128 ; CHECK-NEXT:    psubq %xmm1, %xmm0
    129 ; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [3689348814741910323,3689348814741910323]
    130 ; CHECK-NEXT:    movdqa %xmm0, %xmm3
    131 ; CHECK-NEXT:    pand %xmm1, %xmm3
    132 ; CHECK-NEXT:    psrlq $2, %xmm0
    133 ; CHECK-NEXT:    pand %xmm1, %xmm0
    134 ; CHECK-NEXT:    paddq %xmm3, %xmm0
    135 ; CHECK-NEXT:    movdqa %xmm0, %xmm1
    136 ; CHECK-NEXT:    psrlq $4, %xmm1
    137 ; CHECK-NEXT:    paddq %xmm0, %xmm1
    138 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
    139 ; CHECK-NEXT:    psadbw %xmm2, %xmm1
    140 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
    141 ; CHECK-NEXT:    retq
    142   %c = call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %a)
    143   ret <2 x i32> %c
    144 }
    145