; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc < %s -mtriple=x86_64-- -mcpu=btver2 | FileCheck %s

define i64 @PR81136(i32 %a0, i32 %a1, ptr %a2) {
; CHECK-LABEL: PR81136:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovd %edi, %xmm0
; CHECK-NEXT:    vmovd %esi, %xmm1
; CHECK-NEXT:    vmovdqa (%rdx), %ymm2
; CHECK-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpmovsxwq %xmm0, %xmm0
; CHECK-NEXT:    vpalignr {{.*#+}} xmm0 = mem[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
; CHECK-NEXT:    vpcmpeqq %xmm3, %xmm0, %xmm0
; CHECK-NEXT:    vpxor %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; CHECK-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1
; CHECK-NEXT:    vextractf128 $1, %ymm2, %xmm2
; CHECK-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; CHECK-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
; CHECK-NEXT:    vandnpd %ymm0, %ymm1, %ymm0
; CHECK-NEXT:    vmovmskpd %ymm0, %eax
; CHECK-NEXT:    popcntl %eax, %eax
; CHECK-NEXT:    negq %rax
; CHECK-NEXT:    retq
  %v0 = bitcast i32 %a0 to <2 x i16>
  %v1 = bitcast i32 %a1 to <2 x i16>
  %cmp15 = icmp sle <2 x i16> %v1, %v0
  %conv16 = sext <2 x i1> %cmp15 to <2 x i64>
  %shuffle29 = shufflevector <2 x i64> %conv16, <2 x i64> <i64 128, i64 1>, <4 x i32> <i32 2, i32 3, i32 3, i32 0>
  %data = load volatile <4 x i64>, ptr %a2, align 32
  %cmp65 = icmp ne <4 x i64> %data, <i64 -2071602529, i64 -1537047284, i64 717942021, i64 597457239>
  %cmp67 = icmp ne <4 x i64> %shuffle29, zeroinitializer
  %and = and <4 x i1> %cmp65, %cmp67
  %mask = bitcast <4 x i1> %and to i4
  %cnt = tail call i4 @llvm.ctpop.i4(i4 %mask)
  %cntz = zext i4 %cnt to i64
  %res = sub nsw i64 0, %cntz
  ret i64 %res
}
declare i4 @llvm.ctpop.i4(i4)