; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=X86
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=X64

;; Use cttz to test if we properly prove never-zero. There is a very
;; simple transform from cttz -> cttz_zero_undef if its operand is
;; known never zero.

define i32 @or_known_nonzero(i32 %x) {
; X86-LABEL: or_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $1, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: or_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    orl $1, %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %z = or i32 %x, 1
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @or_known_nonzero_vec(<4 x i32> %x, ptr %p) {
; X86-LABEL: or_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: or_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = or <4 x i32> %x, <i32 1, i32 0, i32 0, i32 0>
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @or_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: or_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: or_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    orl %esi, %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %z = or i32 %x, %y
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @select_known_nonzero(i1 %c, i32 %x) {
; X86-LABEL: select_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $1, %eax
; X86-NEXT:    testb $1, {{[0-9]+}}(%esp)
; X86-NEXT:    movl $122, %ecx
; X86-NEXT:    cmovnel %eax, %ecx
; X86-NEXT:    rep bsfl %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: select_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    orl $1, %esi
; X64-NEXT:    testb $1, %dil
; X64-NEXT:    movl $122, %eax
; X64-NEXT:    cmovnel %esi, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %y = or i32 %x, 1
  %z = select i1 %c, i32 %y, i32 122
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @select_maybe_zero(i1 %c, i32 %x) {
; X86-LABEL: select_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $1, %eax
; X86-NEXT:    xorl %ecx, %ecx
; X86-NEXT:    testb $1, {{[0-9]+}}(%esp)
; X86-NEXT:    cmovnel %eax, %ecx
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: select_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    orl $1, %esi
; X64-NEXT:    xorl %ecx, %ecx
; X64-NEXT:    testb $1, %dil
; X64-NEXT:    cmovnel %esi, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %y = or i32 %x, 1
  %z = select i1 %c, i32 %y, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @extractelt_nonzero_vec(<4 x i32> %a0, ptr %p1, i32 %a2) {
; X86-LABEL: extractelt_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pcmpgtd %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, %xmm0
; X86-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    por %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    movd %xmm1, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: extractelt_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vmovaps {{.*#+}} xmm1 = [8,4294967295,4294967295,4294967295]
; X64-NEXT:    vblendvps %xmm0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
; X64-NEXT:    vmovaps %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %cmp = icmp sgt <4 x i32> zeroinitializer, %a0
  %sel = select <4 x i1> %cmp, <4 x i32> <i32 4, i32 0, i32 0, i32 0>, <4 x i32> <i32 8, i32 -1, i32 -1, i32 -1>
  store <4 x i32> %sel, ptr %p1
  %shl0 = extractelement <4 x i32> %sel, i32 0
  %res = call i32 @llvm.cttz.i32(i32 %shl0, i1 0)
  ret i32 %res
}

define i32 @extractelt_nonzero_vec_fail0(<4 x i32> %a0, ptr %p1, i32 %a2) {
; X86-LABEL: extractelt_nonzero_vec_fail0:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    andl $3, %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pcmpgtd %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, %xmm0
; X86-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    por %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, (%ecx)
; X86-NEXT:    bsfl (%ecx,%eax,4), %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: extractelt_nonzero_vec_fail0:
; X64:       # %bb.0:
; X64-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-NEXT:    vmovaps {{.*#+}} xmm1 = [8,4294967295,4294967295,4294967295]
; X64-NEXT:    vblendvps %xmm0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
; X64-NEXT:    vmovaps %xmm0, (%rdi)
; X64-NEXT:    andl $3, %esi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl (%rdi,%rsi,4), %eax
; X64-NEXT:    retq
  %cmp = icmp sgt <4 x i32> zeroinitializer, %a0
  %sel = select <4 x i1> %cmp, <4 x i32> <i32 4, i32 0, i32 0, i32 0>, <4 x i32> <i32 8, i32 -1, i32 -1, i32 -1>
  store <4 x i32> %sel, ptr %p1
  %shl0 = extractelement <4 x i32> %sel, i32 %a2
  %res = call i32 @llvm.cttz.i32(i32 %shl0, i1 0)
  ret i32 %res
}

define i32 @shl_known_nonzero_1s_bit_set(i32 %x) {
; X86-LABEL: shl_known_nonzero_1s_bit_set:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $123, %eax
; X86-NEXT:    shll %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: shl_known_nonzero_1s_bit_set:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $123, %eax
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = shl i32 123, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @shl_known_nonzero_1s_bit_set_vec(<4 x i32> %x, ptr %p) {
; X86-LABEL: shl_known_nonzero_1s_bit_set_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pslld $23, %xmm0
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    cvttps2dq %xmm0, %xmm0
; X86-NEXT:    pmuludq {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 # [123,0,0,0]
; X86-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,0,4294967295,0]
; X86-NEXT:    pand %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: shl_known_nonzero_1s_bit_set_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpslld $23, %xmm0, %xmm0
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vcvttps2dq %xmm0, %xmm0
; X64-NEXT:    vpmulld {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [123,0,0,0]
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = shl <4 x i32> <i32 123, i32 0, i32 0, i32 0>, %x
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @shl_known_nonzero_nsw(i32 %x, i32 %yy) {
; X86-LABEL: shl_known_nonzero_nsw:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shll %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: shl_known_nonzero_nsw:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    orl $256, %esi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %esi
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %y = or i32 %yy, 256
  %z = shl nsw i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @shl_known_nonzero_nsw_vec(<4 x i32> %x, <4 x i32> %yy, ptr %p) {
; X86-LABEL: shl_known_nonzero_nsw_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    pslld $23, %xmm0
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    cvttps2dq %xmm0, %xmm0
; X86-NEXT:    pmuludq %xmm0, %xmm1
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[0,2,2,3]
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; X86-NEXT:    pmuludq %xmm2, %xmm0
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; X86-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
; X86-NEXT:    movdqa %xmm3, (%eax)
; X86-NEXT:    movd %xmm1, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: shl_known_nonzero_nsw_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vpslld $23, %xmm0, %xmm0
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vcvttps2dq %xmm0, %xmm0
; X64-NEXT:    vpmulld %xmm0, %xmm1, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %y = or <4 x i32> %yy, <i32 256, i32 0, i32 0, i32 0>
  %z = shl nsw <4 x i32> %y, %x
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @shl_known_nonzero_nuw(i32 %x, i32 %yy) {
; X86-LABEL: shl_known_nonzero_nuw:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shll %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: shl_known_nonzero_nuw:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    orl $256, %esi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %esi
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %y = or i32 %yy, 256
  %z = shl nuw i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @shl_known_nonzero_nuw_vec(<4 x i32> %x, <4 x i32> %yy, ptr %p) {
; X86-LABEL: shl_known_nonzero_nuw_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    pslld $23, %xmm0
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    cvttps2dq %xmm0, %xmm0
; X86-NEXT:    pmuludq %xmm0, %xmm1
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[0,2,2,3]
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; X86-NEXT:    pmuludq %xmm2, %xmm0
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; X86-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
; X86-NEXT:    movdqa %xmm3, (%eax)
; X86-NEXT:    movd %xmm1, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: shl_known_nonzero_nuw_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vpslld $23, %xmm0, %xmm0
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vcvttps2dq %xmm0, %xmm0
; X64-NEXT:    vpmulld %xmm0, %xmm1, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %y = or <4 x i32> %yy, <i32 256, i32 0, i32 0, i32 0>
  %z = shl nuw <4 x i32> %y, %x
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @shl_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: shl_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shll %cl, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: shl_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %esi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %z = shl nuw nsw i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @uaddsat_known_nonzero(i32 %x) {
; X86-LABEL: uaddsat_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    incl %eax
; X86-NEXT:    movl $-1, %ecx
; X86-NEXT:    cmovnel %eax, %ecx
; X86-NEXT:    rep bsfl %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: uaddsat_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    incl %edi
; X64-NEXT:    movl $-1, %eax
; X64-NEXT:    cmovnel %edi, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.uadd.sat.i32(i32 %x, i32 1)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @uaddsat_known_nonzero_vec(<16 x i8> %x, ptr %p) {
; X86-LABEL: uaddsat_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    paddusb {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movzbl (%eax), %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: uaddsat_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpaddusb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vpextrb $0, %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = call <16 x i8> @llvm.uadd.sat.v16i8(<16 x i8> %x, <16 x i8> <i8 1, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>)
  store <16 x i8> %z, ptr %p
  %e = extractelement <16 x i8> %z, i32 0
  %ex = zext i8 %e to i32
  %r = call i32 @llvm.cttz.i32(i32 %ex, i1 false)
  ret i32 %r
}

define i32 @uaddsat_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: uaddsat_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl $-1, %ecx
; X86-NEXT:    cmovael %eax, %ecx
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: uaddsat_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    addl %esi, %edi
; X64-NEXT:    movl $-1, %ecx
; X64-NEXT:    cmovael %edi, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.uadd.sat.i32(i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @umax_known_nonzero(i32 %x, i32 %y) {
; X86-LABEL: umax_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $4, %edx
; X86-NEXT:    shll %cl, %edx
; X86-NEXT:    cmpl %edx, %eax
; X86-NEXT:    cmoval %eax, %edx
; X86-NEXT:    rep bsfl %edx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: umax_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    movl $4, %eax
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    cmpl %eax, %edi
; X64-NEXT:    cmoval %edi, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %yy = shl nuw i32 4, %y
  %z = call i32 @llvm.umax.i32(i32 %x, i32 %yy)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @umax_known_nonzero_vec(<16 x i8> %x, ptr %p) {
; X86-LABEL: umax_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movdqa %xmm0, %xmm3
; X86-NEXT:    psllw $5, %xmm3
; X86-NEXT:    pxor %xmm2, %xmm2
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pcmpgtb %xmm3, %xmm1
; X86-NEXT:    movdqa %xmm1, %xmm4
; X86-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}, %xmm4
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    por %xmm4, %xmm1
; X86-NEXT:    paddb %xmm3, %xmm3
; X86-NEXT:    pxor %xmm4, %xmm4
; X86-NEXT:    pcmpgtb %xmm3, %xmm4
; X86-NEXT:    movdqa %xmm4, %xmm5
; X86-NEXT:    pandn %xmm1, %xmm5
; X86-NEXT:    paddb %xmm1, %xmm1
; X86-NEXT:    paddb %xmm1, %xmm1
; X86-NEXT:    pand %xmm4, %xmm1
; X86-NEXT:    por %xmm5, %xmm1
; X86-NEXT:    paddb %xmm3, %xmm3
; X86-NEXT:    pcmpgtb %xmm3, %xmm2
; X86-NEXT:    movdqa %xmm2, %xmm3
; X86-NEXT:    pandn %xmm1, %xmm3
; X86-NEXT:    paddb %xmm1, %xmm1
; X86-NEXT:    pand %xmm2, %xmm1
; X86-NEXT:    por %xmm3, %xmm1
; X86-NEXT:    pmaxub %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    movzbl (%eax), %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: umax_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpsllw $5, %xmm0, %xmm1
; X64-NEXT:    vpmovsxbq {{.*#+}} xmm2 = [4,0]
; X64-NEXT:    vpblendvb %xmm1, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; X64-NEXT:    vpaddb %xmm2, %xmm2, %xmm3
; X64-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
; X64-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpblendvb %xmm1, %xmm3, %xmm2, %xmm2
; X64-NEXT:    vpaddb %xmm2, %xmm2, %xmm3
; X64-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpblendvb %xmm1, %xmm3, %xmm2, %xmm1
; X64-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vpextrb $0, %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %yy = shl nuw <16 x i8> <i8 4, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, %x
  %z = call <16 x i8> @llvm.umax.v16i8(<16 x i8> %x, <16 x i8> %yy)
  store <16 x i8> %z, ptr %p
  %e = extractelement <16 x i8> %z, i32 0
  %ex = zext i8 %e to i32
  %r = call i32 @llvm.cttz.i32(i32 %ex, i1 false)
  ret i32 %r
}

define i32 @umax_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: umax_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    cmpl %eax, %ecx
; X86-NEXT:    cmoval %ecx, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: umax_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    cmpl %esi, %edi
; X64-NEXT:    cmoval %edi, %esi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.umax.i32(i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @umin_known_nonzero(i32 %xx, i32 %yy) {
; X86-LABEL: umin_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $4, %edx
; X86-NEXT:    shll %cl, %edx
; X86-NEXT:    addl $4, %eax
; X86-NEXT:    cmpl %eax, %edx
; X86-NEXT:    cmovbl %edx, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: umin_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $4, %eax
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    addl $4, %esi
; X64-NEXT:    cmpl %esi, %eax
; X64-NEXT:    cmovbl %eax, %esi
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %x = shl nuw i32 4, %xx
  %y = add nuw nsw i32 %yy, 4
  %z = call i32 @llvm.umin.i32(i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @umin_known_nonzero_vec(<16 x i8> %x, ptr %p) {
; X86-LABEL: umin_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movdqa %xmm0, %xmm3
; X86-NEXT:    psllw $5, %xmm3
; X86-NEXT:    pxor %xmm2, %xmm2
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pcmpgtb %xmm3, %xmm1
; X86-NEXT:    movdqa %xmm1, %xmm4
; X86-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}, %xmm4
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    por %xmm4, %xmm1
; X86-NEXT:    paddb %xmm3, %xmm3
; X86-NEXT:    pxor %xmm4, %xmm4
; X86-NEXT:    pcmpgtb %xmm3, %xmm4
; X86-NEXT:    movdqa %xmm4, %xmm5
; X86-NEXT:    pandn %xmm1, %xmm5
; X86-NEXT:    paddb %xmm1, %xmm1
; X86-NEXT:    paddb %xmm1, %xmm1
; X86-NEXT:    pand %xmm4, %xmm1
; X86-NEXT:    por %xmm5, %xmm1
; X86-NEXT:    paddb %xmm3, %xmm3
; X86-NEXT:    pcmpgtb %xmm3, %xmm2
; X86-NEXT:    movdqa %xmm2, %xmm3
; X86-NEXT:    pandn %xmm1, %xmm3
; X86-NEXT:    paddb %xmm1, %xmm1
; X86-NEXT:    pand %xmm2, %xmm1
; X86-NEXT:    por %xmm3, %xmm1
; X86-NEXT:    pminub %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    movzbl (%eax), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: umin_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpsllw $5, %xmm0, %xmm1
; X64-NEXT:    vpmovsxbq {{.*#+}} xmm2 = [4,0]
; X64-NEXT:    vpblendvb %xmm1, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; X64-NEXT:    vpaddb %xmm2, %xmm2, %xmm3
; X64-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
; X64-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpblendvb %xmm1, %xmm3, %xmm2, %xmm2
; X64-NEXT:    vpaddb %xmm2, %xmm2, %xmm3
; X64-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpblendvb %xmm1, %xmm3, %xmm2, %xmm1
; X64-NEXT:    vpminub %xmm1, %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vpextrb $0, %xmm0, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %yy = shl nuw <16 x i8> <i8 4, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, %x
  %z = call <16 x i8> @llvm.umin.v16i8(<16 x i8> %x, <16 x i8> %yy)
  store <16 x i8> %z, ptr %p
  %e = extractelement <16 x i8> %z, i32 0
  %ex = zext i8 %e to i32
  %r = call i32 @llvm.cttz.i32(i32 %ex, i1 false)
  ret i32 %r
}

define i32 @umin_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: umin_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    cmpl $54, %eax
; X86-NEXT:    movl $54, %ecx
; X86-NEXT:    cmovbl %eax, %ecx
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: umin_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    cmpl $54, %edi
; X64-NEXT:    movl $54, %ecx
; X64-NEXT:    cmovbl %edi, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.umin.i32(i32 %x, i32 54)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @smin_known_nonzero(i32 %xx, i32 %yy) {
; X86-LABEL: smin_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $4, %edx
; X86-NEXT:    shll %cl, %edx
; X86-NEXT:    addl $4, %eax
; X86-NEXT:    cmpl %eax, %edx
; X86-NEXT:    cmovll %edx, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smin_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $4, %eax
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    addl $4, %esi
; X64-NEXT:    cmpl %esi, %eax
; X64-NEXT:    cmovll %eax, %esi
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %x = shl nuw i32 4, %xx
  %y = add nuw nsw i32 %yy, 4
  %z = call i32 @llvm.smin.i32(i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @smin_known_zero(i32 %x, i32 %y) {
; X86-LABEL: smin_known_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    cmpl $-54, %eax
; X86-NEXT:    movl $-54, %ecx
; X86-NEXT:    cmovll %eax, %ecx
; X86-NEXT:    rep bsfl %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smin_known_zero:
; X64:       # %bb.0:
; X64-NEXT:    cmpl $-54, %edi
; X64-NEXT:    movl $-54, %eax
; X64-NEXT:    cmovll %edi, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.smin.i32(i32 %x, i32 -54)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define <4 x i32> @smin_known_zero_vec(<4 x i32> %x, <4 x i32> %y) {
; X86-LABEL: smin_known_zero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movdqa {{.*#+}} xmm1 = [4294967242,4294967273,4294967284,4294967295]
; X86-NEXT:    movdqa %xmm1, %xmm2
; X86-NEXT:    pcmpgtd %xmm0, %xmm2
; X86-NEXT:    pand %xmm2, %xmm0
; X86-NEXT:    pandn %xmm1, %xmm2
; X86-NEXT:    por %xmm2, %xmm0
; X86-NEXT:    pcmpeqd %xmm1, %xmm1
; X86-NEXT:    paddd %xmm0, %xmm1
; X86-NEXT:    pand %xmm1, %xmm0
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pcmpeqd %xmm1, %xmm0
; X86-NEXT:    psrld $31, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: smin_known_zero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpminsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpaddd %xmm1, %xmm0, %xmm1
; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
; X64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
; X64-NEXT:    vpsrld $31, %xmm0, %xmm0
; X64-NEXT:    retq
  %z = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %x, <4 x i32> <i32 -54, i32 -23, i32 -12, i32 -1>)
  %r = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %z)
  %3 = icmp eq <4 x i32> %r, <i32 1, i32 1, i32 1, i32 1>
  %ret = zext <4 x i1> %3 to <4 x i32>
  ret <4 x i32> %ret
}

define i32 @smin_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: smin_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    cmpl $54, %eax
; X86-NEXT:    movl $54, %ecx
; X86-NEXT:    cmovll %eax, %ecx
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smin_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    cmpl $54, %edi
; X64-NEXT:    movl $54, %ecx
; X64-NEXT:    cmovll %edi, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.smin.i32(i32 %x, i32 54)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @smin_known_never_zero_vec_element(<4 x i32> %x) {
; X86-LABEL: smin_known_never_zero_vec_element:
; X86:       # %bb.0:
; X86-NEXT:    movdqa {{.*#+}} xmm1 = [54,4294967273,12,1]
; X86-NEXT:    movdqa %xmm1, %xmm2
; X86-NEXT:    pcmpgtd %xmm0, %xmm2
; X86-NEXT:    pand %xmm2, %xmm0
; X86-NEXT:    pandn %xmm1, %xmm2
; X86-NEXT:    por %xmm0, %xmm2
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smin_known_never_zero_vec_element:
; X64:       # %bb.0:
; X64-NEXT:    vpminsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vpextrd $1, %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %x, <4 x i32> <i32 54, i32 -23, i32 12, i32 1>)
  %el = extractelement <4 x i32> %z, i32 1
  %r = call i32 @llvm.cttz.i32(i32 %el, i1 false)
  ret i32 %r
}

define i32 @smax_known_nonzero(i32 %xx, i32 %yy) {
; X86-LABEL: smax_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $4, %edx
; X86-NEXT:    shll %cl, %edx
; X86-NEXT:    addl $4, %eax
; X86-NEXT:    cmpl %eax, %edx
; X86-NEXT:    cmovgl %edx, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smax_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $4, %eax
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    addl $4, %esi
; X64-NEXT:    cmpl %esi, %eax
; X64-NEXT:    cmovgl %eax, %esi
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %x = shl nuw i32 4, %xx
  %y = add nuw nsw i32 %yy, 4
  %z = call i32 @llvm.smax.i32(i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @smax_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: smax_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    cmpl $55, %eax
; X86-NEXT:    movl $54, %ecx
; X86-NEXT:    cmovgel %eax, %ecx
; X86-NEXT:    rep bsfl %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smax_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    cmpl $55, %edi
; X64-NEXT:    movl $54, %eax
; X64-NEXT:    cmovgel %edi, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.smax.i32(i32 %x, i32 54)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define <4 x i32> @smax_known_zero_vec(<4 x i32> %x, <4 x i32> %y) {
; X86-LABEL: smax_known_zero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movdqa {{.*#+}} xmm1 = [54,23,12,1]
; X86-NEXT:    movdqa %xmm0, %xmm2
; X86-NEXT:    pcmpgtd %xmm1, %xmm2
; X86-NEXT:    pand %xmm2, %xmm0
; X86-NEXT:    pandn %xmm1, %xmm2
; X86-NEXT:    por %xmm2, %xmm0
; X86-NEXT:    pcmpeqd %xmm1, %xmm1
; X86-NEXT:    paddd %xmm0, %xmm1
; X86-NEXT:    pand %xmm1, %xmm0
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pcmpeqd %xmm1, %xmm0
; X86-NEXT:    psrld $31, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: smax_known_zero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpmaxsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpaddd %xmm1, %xmm0, %xmm1
; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
; X64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
; X64-NEXT:    vpsrld $31, %xmm0, %xmm0
; X64-NEXT:    retq
  %z = call <4 x i32> @llvm.smax.v4i32(<4 x i32> %x, <4 x i32> <i32 54, i32 23, i32 12, i32 1>)
  %r = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %z)
  %3 = icmp eq <4 x i32> %r, <i32 1, i32 1, i32 1, i32 1>
  %ret = zext <4 x i1> %3 to <4 x i32>
  ret <4 x i32> %ret
}

define i32 @smax_known_zero(i32 %x, i32 %y) {
; X86-LABEL: smax_known_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    testl %eax, %eax
; X86-NEXT:    movl $-1, %ecx
; X86-NEXT:    cmovnsl %eax, %ecx
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smax_known_zero:
; X64:       # %bb.0:
; X64-NEXT:    testl %edi, %edi
; X64-NEXT:    movl $-1, %ecx
; X64-NEXT:    cmovnsl %edi, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.smax.i32(i32 %x, i32 -1)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @smax_known_never_zero_vec_element(<4 x i32> %x) {
; X86-LABEL: smax_known_never_zero_vec_element:
; X86:       # %bb.0:
; X86-NEXT:    movdqa {{.*#+}} xmm1 = [54,4294967273,4294967284,4294967295]
; X86-NEXT:    movdqa %xmm0, %xmm2
; X86-NEXT:    pcmpgtd %xmm1, %xmm2
; X86-NEXT:    pand %xmm2, %xmm0
; X86-NEXT:    pandn %xmm1, %xmm2
; X86-NEXT:    por %xmm0, %xmm2
; X86-NEXT:    movd %xmm2, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: smax_known_never_zero_vec_element:
; X64:       # %bb.0:
; X64-NEXT:    vpmaxsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = call <4 x i32> @llvm.smax.v4i32(<4 x i32> %x, <4 x i32> <i32 54, i32 -23, i32 -12, i32 -1>)
  %el = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %el, i1 false)
  ret i32 %r
}

define i32 @rotr_known_nonzero(i32 %xx, i32 %y) {
; X86-LABEL: rotr_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    rorl %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotr_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    orl $256, %edi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    rorl %cl, %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 256
  %shr = lshr i32 %x, %y
  %sub = sub i32 32, %y
  %shl = shl i32 %x, %sub
  %z = or i32 %shl, %shr
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @rotr_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: rotr_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    rorl %cl, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotr_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    rorl %cl, %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %shr = lshr i32 %x, %y
  %sub = sub i32 32, %y
  %shl = shl i32 %x, %sub
  %z = or i32 %shl, %shr
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @rotr_with_fshr_known_nonzero(i32 %xx, i32 %y) {
; X86-LABEL: rotr_with_fshr_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    rorl %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotr_with_fshr_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    orl $256, %edi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    rorl %cl, %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 256
  %z = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @rotr_with_fshr_known_nonzero_vec(<4 x i32> %xx, <4 x i32> %y, ptr %p) {
; X86-LABEL: rotr_with_fshr_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pxor %xmm3, %xmm3
; X86-NEXT:    psubd %xmm1, %xmm3
; X86-NEXT:    pslld $23, %xmm3
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm3
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm3
; X86-NEXT:    cvttps2dq %xmm3, %xmm1
; X86-NEXT:    pmuludq %xmm1, %xmm0
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
; X86-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; X86-NEXT:    pmuludq %xmm2, %xmm1
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
; X86-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; X86-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; X86-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; X86-NEXT:    por %xmm3, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotr_with_fshr_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm2
; X64-NEXT:    vpxor %xmm3, %xmm3, %xmm3
; X64-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
; X64-NEXT:    vpslld $23, %xmm1, %xmm1
; X64-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vcvttps2dq %xmm1, %xmm1
; X64-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; X64-NEXT:    vpmuludq %xmm3, %xmm0, %xmm0
; X64-NEXT:    vpmuludq %xmm1, %xmm2, %xmm1
; X64-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; X64-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,2,2]
; X64-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3],xmm1[4,5],xmm0[6,7]
; X64-NEXT:    vpor %xmm2, %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 256, i32 0, i32 0, i32 0>
  %z = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %y)
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @rotr_with_fshr_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: rotr_with_fshr_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    rorl %cl, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotr_with_fshr_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    rorl %cl, %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @rotl_known_nonzero(i32 %xx, i32 %y) {
; X86-LABEL: rotl_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    roll %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotl_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    orl $256, %edi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    roll %cl, %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 256
  %shl = shl i32 %x, %y
  %sub = sub i32 32, %y
  %shr = lshr i32 %x, %sub
  %z = or i32 %shr, %shl
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @rotl_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: rotl_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    roll %cl, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotl_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    roll %cl, %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %shl = shl i32 %x, %y
  %sub = sub i32 32, %y
  %shr = lshr i32 %x, %sub
  %z = or i32 %shr, %shl
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @rotl_with_fshl_known_nonzero(i32 %xx, i32 %y) {
; X86-LABEL: rotl_with_fshl_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    roll %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotl_with_fshl_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    orl $256, %edi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    roll %cl, %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 256
  %z = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @rotl_with_fshl_known_nonzero_vec(<4 x i32> %xx, <4 x i32> %y, ptr %p) {
; X86-LABEL: rotl_with_fshl_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pslld $23, %xmm1
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    cvttps2dq %xmm1, %xmm1
; X86-NEXT:    pmuludq %xmm1, %xmm0
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
; X86-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; X86-NEXT:    pmuludq %xmm2, %xmm1
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
; X86-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; X86-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; X86-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; X86-NEXT:    por %xmm3, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotl_with_fshl_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm2
; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; X64-NEXT:    vpslld $23, %xmm1, %xmm1
; X64-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vcvttps2dq %xmm1, %xmm1
; X64-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; X64-NEXT:    vpmuludq %xmm3, %xmm0, %xmm0
; X64-NEXT:    vpmuludq %xmm1, %xmm2, %xmm1
; X64-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; X64-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,2,2]
; X64-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3],xmm1[4,5],xmm0[6,7]
; X64-NEXT:    vpor %xmm2, %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 256, i32 0, i32 0, i32 0>
  %z = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %y)
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @rotl_with_fshl_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: rotl_with_fshl_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    roll %cl, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: rotl_with_fshl_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    roll %cl, %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %y)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sra_known_nonzero_sign_bit_set(i32 %x) {
; X86-LABEL: sra_known_nonzero_sign_bit_set:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $-2147360405, %eax # imm = 0x8001E16B
; X86-NEXT:    sarl %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sra_known_nonzero_sign_bit_set:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $-2147360405, %eax # imm = 0x8001E16B
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    sarl %cl, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = ashr i32 2147606891, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sra_known_nonzero_sign_bit_set_vec(<4 x i32> %x, ptr %p) {
; X86-LABEL: sra_known_nonzero_sign_bit_set_vec:
; X86:       # %bb.0:
; X86-NEXT:    xorps %xmm1, %xmm1
; X86-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movdqa {{.*#+}} xmm0 = [2147606891,65535,1,0]
; X86-NEXT:    psrad %xmm1, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sra_known_nonzero_sign_bit_set_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; X64-NEXT:    vmovdqa {{.*#+}} xmm1 = [2147606891,65535,1,0]
; X64-NEXT:    vpsrad %xmm0, %xmm1, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %xx = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer
  %z = ashr <4 x i32> <i32 2147606891, i32 65535, i32 1, i32 0>, %xx
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @sra_known_nonzero_exact(i32 %x, i32 %yy) {
; X86-LABEL: sra_known_nonzero_exact:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    sarl %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sra_known_nonzero_exact:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    orl $256, %esi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    sarl %cl, %esi
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %y = or i32 %yy, 256
  %z = ashr exact i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sra_known_nonzero_exact_vec(<4 x i32> %x, <4 x i32> %yy, ptr %p) {
; X86-LABEL: sra_known_nonzero_exact_vec:
; X86:       # %bb.0:
; X86-NEXT:    xorps %xmm2, %xmm2
; X86-NEXT:    movss {{.*#+}} xmm2 = xmm0[0],xmm2[1,2,3]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    psrad %xmm2, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sra_known_nonzero_exact_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vpsrad %xmm0, %xmm1, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vpextrd $1, %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x.splat = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer
  %y = or <4 x i32> %yy, <i32 0, i32 256, i32 0, i32 0>
  %z = ashr exact <4 x i32> %y, %x.splat
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 1
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @sra_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: sra_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    sarl %cl, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sra_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    sarl %cl, %esi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %z = ashr exact i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @srl_known_nonzero_sign_bit_set(i32 %x) {
; X86-LABEL: srl_known_nonzero_sign_bit_set:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $-2147360405, %eax # imm = 0x8001E16B
; X86-NEXT:    shrl %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: srl_known_nonzero_sign_bit_set:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $-2147360405, %eax # imm = 0x8001E16B
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shrl %cl, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = lshr i32 2147606891, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @srl_known_nonzero_sign_bit_set_vec(<4 x i32> %x, ptr %p) {
; X86-LABEL: srl_known_nonzero_sign_bit_set_vec:
; X86:       # %bb.0:
; X86-NEXT:    xorps %xmm1, %xmm1
; X86-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movdqa {{.*#+}} xmm0 = [0,65535,2147606891,0]
; X86-NEXT:    psrld %xmm1, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: srl_known_nonzero_sign_bit_set_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; X64-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,65535,2147606891,0]
; X64-NEXT:    vpsrld %xmm0, %xmm1, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vpextrd $2, %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x.splat = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer
  %z = lshr <4 x i32> <i32 0, i32 65535, i32 2147606891, i32 0>, %x.splat
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 2
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @srl_known_nonzero_exact(i32 %x, i32 %yy) {
; X86-LABEL: srl_known_nonzero_exact:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: srl_known_nonzero_exact:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    orl $256, %esi # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shrl %cl, %esi
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %y = or i32 %yy, 256
  %z = lshr exact i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @srl_known_nonzero_exact_vec(<4 x i32> %x, <4 x i32> %yy, ptr %p) {
; X86-LABEL: srl_known_nonzero_exact_vec:
; X86:       # %bb.0:
; X86-NEXT:    xorps %xmm2, %xmm2
; X86-NEXT:    movss {{.*#+}} xmm2 = xmm0[0],xmm2[1,2,3]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    psrld %xmm2, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[3,3,3,3]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: srl_known_nonzero_exact_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; X64-NEXT:    vpsrld %xmm0, %xmm1, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vpextrd $3, %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x.splat = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> zeroinitializer
  %y = or <4 x i32> %yy, <i32 0, i32 0, i32 0, i32 256>
  %z = lshr exact <4 x i32> %y, %x.splat
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 3
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @srl_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: srl_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl %cl, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: srl_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shrl %cl, %esi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %z = lshr exact i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @udiv_known_nonzero(i32 %xx, i32 %y) {
; X86-LABEL: udiv_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $64, %eax
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    divl {{[0-9]+}}(%esp)
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: udiv_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    orl $64, %eax
; X64-NEXT:    xorl %edx, %edx
; X64-NEXT:    divl %esi
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 64
  %z = udiv exact i32 %x, %y
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @udiv_known_nonzero_vec(<4 x i32> %xx, <4 x i32> %y, ptr %p) nounwind {
; X86-LABEL: udiv_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    pushl %edi
; X86-NEXT:    pushl %esi
; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
; X86-NEXT:    movd %xmm2, %ecx
; X86-NEXT:    movl $-1, %eax
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    divl %ecx
; X86-NEXT:    movd %eax, %xmm3
; X86-NEXT:    movd %xmm1, %ecx
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    divl %ecx
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    movd %eax, %xmm2
; X86-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[3,3,3,3]
; X86-NEXT:    movd %xmm3, %edi
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[3,3,3,3]
; X86-NEXT:    movd %xmm3, %eax
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    divl %edi
; X86-NEXT:    movd %eax, %xmm3
; X86-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
; X86-NEXT:    movd %xmm1, %edi
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    divl %edi
; X86-NEXT:    movd %eax, %xmm0
; X86-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
; X86-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
; X86-NEXT:    movdqa %xmm2, (%esi)
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    popl %esi
; X86-NEXT:    popl %edi
; X86-NEXT:    retl
;
; X64-LABEL: udiv_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vpextrd $1, %xmm1, %ecx
; X64-NEXT:    vpextrd $1, %xmm0, %eax
; X64-NEXT:    xorl %edx, %edx
; X64-NEXT:    divl %ecx
; X64-NEXT:    movl %eax, %ecx
; X64-NEXT:    vmovd %xmm1, %esi
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    xorl %edx, %edx
; X64-NEXT:    divl %esi
; X64-NEXT:    vmovd %eax, %xmm2
; X64-NEXT:    vpinsrd $1, %ecx, %xmm2, %xmm2
; X64-NEXT:    vpextrd $2, %xmm1, %ecx
; X64-NEXT:    vpextrd $2, %xmm0, %eax
; X64-NEXT:    xorl %edx, %edx
; X64-NEXT:    divl %ecx
; X64-NEXT:    vpinsrd $2, %eax, %xmm2, %xmm2
; X64-NEXT:    vpextrd $3, %xmm1, %ecx
; X64-NEXT:    vpextrd $3, %xmm0, %eax
; X64-NEXT:    xorl %edx, %edx
; X64-NEXT:    divl %ecx
; X64-NEXT:    vpinsrd $3, %eax, %xmm2, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 64, i32 -1, i32 0, i32 0>
  %z = udiv exact <4 x i32> %x, %y
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @udiv_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: udiv_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    divl {{[0-9]+}}(%esp)
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: udiv_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    xorl %edx, %edx
; X64-NEXT:    divl %esi
; X64-NEXT:    movl $32, %ecx
; X64-NEXT:    rep bsfl %eax, %ecx
; X64-NEXT:    movl %ecx, %eax
; X64-NEXT:    retq
  %z = udiv exact i32 %x, %y
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sdiv_known_nonzero(i32 %xx, i32 %y) {
; X86-LABEL: sdiv_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $64, %eax
; X86-NEXT:    cltd
; X86-NEXT:    idivl {{[0-9]+}}(%esp)
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sdiv_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    orl $64, %eax
; X64-NEXT:    cltd
; X64-NEXT:    idivl %esi
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 64
  %z = sdiv exact i32 %x, %y
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sdiv_known_nonzero_vec(<4 x i32> %xx, <4 x i32> %y, ptr %p) nounwind {
; X86-LABEL: sdiv_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    pushl %edi
; X86-NEXT:    pushl %esi
; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
; X86-NEXT:    movd %xmm2, %ecx
; X86-NEXT:    movl $-1, %eax
; X86-NEXT:    cltd
; X86-NEXT:    idivl %ecx
; X86-NEXT:    movd %eax, %xmm3
; X86-NEXT:    movd %xmm1, %ecx
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    cltd
; X86-NEXT:    idivl %ecx
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    movd %eax, %xmm2
; X86-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[3,3,3,3]
; X86-NEXT:    movd %xmm3, %edi
; X86-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[3,3,3,3]
; X86-NEXT:    movd %xmm3, %eax
; X86-NEXT:    cltd
; X86-NEXT:    idivl %edi
; X86-NEXT:    movd %eax, %xmm3
; X86-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
; X86-NEXT:    movd %xmm1, %edi
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    cltd
; X86-NEXT:    idivl %edi
; X86-NEXT:    movd %eax, %xmm0
; X86-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
; X86-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
; X86-NEXT:    movdqa %xmm2, (%esi)
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    popl %esi
; X86-NEXT:    popl %edi
; X86-NEXT:    retl
;
; X64-LABEL: sdiv_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vpextrd $1, %xmm1, %ecx
; X64-NEXT:    vpextrd $1, %xmm0, %eax
; X64-NEXT:    cltd
; X64-NEXT:    idivl %ecx
; X64-NEXT:    movl %eax, %ecx
; X64-NEXT:    vmovd %xmm1, %esi
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    cltd
; X64-NEXT:    idivl %esi
; X64-NEXT:    vmovd %eax, %xmm2
; X64-NEXT:    vpinsrd $1, %ecx, %xmm2, %xmm2
; X64-NEXT:    vpextrd $2, %xmm1, %ecx
; X64-NEXT:    vpextrd $2, %xmm0, %eax
; X64-NEXT:    cltd
; X64-NEXT:    idivl %ecx
; X64-NEXT:    vpinsrd $2, %eax, %xmm2, %xmm2
; X64-NEXT:    vpextrd $3, %xmm1, %ecx
; X64-NEXT:    vpextrd $3, %xmm0, %eax
; X64-NEXT:    cltd
; X64-NEXT:    idivl %ecx
; X64-NEXT:    vpinsrd $3, %eax, %xmm2, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 64, i32 -1, i32 0, i32 0>
  %z = sdiv exact <4 x i32> %x, %y
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @sdiv_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: sdiv_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    cltd
; X86-NEXT:    idivl {{[0-9]+}}(%esp)
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sdiv_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    cltd
; X64-NEXT:    idivl %esi
; X64-NEXT:    movl $32, %ecx
; X64-NEXT:    rep bsfl %eax, %ecx
; X64-NEXT:    movl %ecx, %eax
; X64-NEXT:    retq
  %z = sdiv exact i32 %x, %y
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @add_known_nonzero(i32 %xx, i32 %y) {
; X86-LABEL: add_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $1, %eax
; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: add_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    orl $1, %edi
; X64-NEXT:    addl %esi, %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 1
  %z = add nuw i32 %x, %y
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @add_maybe_zero(i32 %xx, i32 %y) {
; X86-LABEL: add_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $1, %eax
; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: add_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    orl $1, %edi
; X64-NEXT:    addl %esi, %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 1
  %z = add nsw i32 %x, %y
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @add_nuw_known_nonzero_vec(<4 x i32> %xx, ptr %p) {
; X86-LABEL: add_nuw_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: add_nuw_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %z = add nuw <4 x i32> %xx, <i32 1, i32 0, i32 0, i32 0>
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @sub_known_nonzero_neg_case(i32 %xx) {
; X86-LABEL: sub_known_nonzero_neg_case:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    shll %cl, %eax
; X86-NEXT:    negl %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sub_known_nonzero_neg_case:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $256, %eax # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    negl %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = shl nuw nsw i32 256, %xx
  %z = sub i32 0, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sub_known_nonzero_ne_case(i32 %xx, i32 %yy) {
; X86-LABEL: sub_known_nonzero_ne_case:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    orl $64, %ecx
; X86-NEXT:    andl $-65, %eax
; X86-NEXT:    subl %ecx, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sub_known_nonzero_ne_case:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    orl $64, %eax
; X64-NEXT:    andl $-65, %edi
; X64-NEXT:    subl %eax, %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 64
  %y = and i32 %xx, -65
  %z = sub i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sub_maybe_zero(i32 %x) {
; X86-LABEL: sub_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    orl $64, %ecx
; X86-NEXT:    subl %eax, %ecx
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sub_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    orl $64, %ecx
; X64-NEXT:    subl %edi, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %y = or i32 %x, 64
  %z = sub i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sub_maybe_zero2(i32 %x) {
; X86-LABEL: sub_maybe_zero2:
; X86:       # %bb.0:
; X86-NEXT:    xorl %eax, %eax
; X86-NEXT:    subl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sub_maybe_zero2:
; X64:       # %bb.0:
; X64-NEXT:    negl %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %z = sub i32 0, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sub_known_nonzero_ne_vec(<4 x i32> %xx, ptr %p) {
; X86-LABEL: sub_known_nonzero_ne_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    movd {{.*#+}} xmm1 = [2,0,0,0]
; X86-NEXT:    psubd %xmm0, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    movd %xmm1, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sub_known_nonzero_ne_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vpmovsxbq {{.*#+}} xmm1 = [2,0]
; X64-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %u = or <4 x i32> %xx, <i32 1, i32 0, i32 0, i32 0>
  %z = sub <4 x i32> <i32 2, i32 0, i32 0, i32 0>, %u
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @mul_known_nonzero_nsw(i32 %x, i32 %yy) {
; X86-LABEL: mul_known_nonzero_nsw:
; X86:       # %bb.0:
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    imull {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: mul_known_nonzero_nsw:
; X64:       # %bb.0:
; X64-NEXT:    orl $256, %esi # imm = 0x100
; X64-NEXT:    imull %edi, %esi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %y = or i32 %yy, 256
  %z = mul nsw i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @mul_known_nonzero_nuw(i32 %x, i32 %yy) {
; X86-LABEL: mul_known_nonzero_nuw:
; X86:       # %bb.0:
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    orl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    imull {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: mul_known_nonzero_nuw:
; X64:       # %bb.0:
; X64-NEXT:    orl $256, %esi # imm = 0x100
; X64-NEXT:    imull %edi, %esi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %esi, %eax
; X64-NEXT:    retq
  %y = or i32 %yy, 256
  %z = mul nuw i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @mul_maybe_zero(i32 %x, i32 %y) {
; X86-LABEL: mul_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    imull {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: mul_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    imull %esi, %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %z = mul nuw nsw i32 %y, %x
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @bitcast_known_nonzero(<2 x i16> %xx) {
; X86-LABEL: bitcast_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT:    pslld $23, %xmm0
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    cvttps2dq %xmm0, %xmm0
; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
; X86-NEXT:    pmullw {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 # [256,256,u,u,u,u,u,u]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bitcast_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
; X64-NEXT:    vpslld $23, %xmm0, %xmm0
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vcvttps2dq %xmm0, %xmm0
; X64-NEXT:    vpackusdw %xmm0, %xmm0, %xmm0
; X64-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [256,256,u,u,u,u,u,u]
; X64-NEXT:    vmovd %xmm0, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %x = shl nuw nsw <2 x i16> <i16 256, i16 256>, %xx
  %z = bitcast <2 x i16> %x to i32
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @bitcast_maybe_zero(<2 x i16> %x) {
; X86-LABEL: bitcast_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bitcast_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    vmovd %xmm0, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = bitcast <2 x i16> %x to i32
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @bitcast_from_float(float %x) {
; X86-LABEL: bitcast_from_float:
; X86:       # %bb.0:
; X86-NEXT:    bsfl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bitcast_from_float:
; X64:       # %bb.0:
; X64-NEXT:    vmovd %xmm0, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = bitcast float %x to i32
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @abs_known_nonzero(i32 %xx) {
; X86-LABEL: abs_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $64, %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    negl %ecx
; X86-NEXT:    cmovsl %eax, %ecx
; X86-NEXT:    rep bsfl %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: abs_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    orl $64, %edi
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    negl %eax
; X64-NEXT:    cmovsl %edi, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 64
  %z = call i32 @llvm.abs.i32(i32 %x, i1 0)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @abs_known_nonzero_vec(<4 x i32> %xx, ptr %p) nounwind {
; X86-LABEL: abs_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    movdqa %xmm0, %xmm1
; X86-NEXT:    psrad $31, %xmm1
; X86-NEXT:    pxor %xmm1, %xmm0
; X86-NEXT:    psubd %xmm1, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: abs_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vpabsd %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 64, i32 -1, i32 0, i32 0>
  %z = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %x, i1 0)
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @abs_maybe_zero(i32 %x) {
; X86-LABEL: abs_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    negl %ecx
; X86-NEXT:    cmovsl %eax, %ecx
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: abs_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    negl %ecx
; X64-NEXT:    cmovsl %edi, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.abs.i32(i32 %x, i1 0)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @bswap_known_nonzero(i32 %xx) {
; X86-LABEL: bswap_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $64, %eax
; X86-NEXT:    bswapl %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bswap_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    orl $64, %edi
; X64-NEXT:    bswapl %edi
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 64
  %z = call i32 @llvm.bswap.i32(i32 %x)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @bswap_known_nonzero_vec(<4 x i32> %xx, ptr %p) nounwind {
; X86-LABEL: bswap_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    movdqa %xmm0, %xmm2
; X86-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
; X86-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
; X86-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
; X86-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
; X86-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
; X86-NEXT:    packuswb %xmm2, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bswap_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,u,u,u,u,11,10,9,8,15,14,13,12]
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 64, i32 -1, i32 0, i32 0>
  %z = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %x)
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @bswap_maybe_zero(i32 %x) {
; X86-LABEL: bswap_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bswapl %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bswap_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    bswapl %edi
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %edi, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.bswap.i32(i32 %x)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @bitreverse_known_nonzero(i32 %xx) {
; X86-LABEL: bitreverse_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    orl $64, %eax
; X86-NEXT:    bswapl %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
; X86-NEXT:    shll $4, %ecx
; X86-NEXT:    shrl $4, %eax
; X86-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
; X86-NEXT:    orl %ecx, %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $858993459, %ecx # imm = 0x33333333
; X86-NEXT:    shrl $2, %eax
; X86-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X86-NEXT:    leal (%eax,%ecx,4), %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
; X86-NEXT:    shrl %eax
; X86-NEXT:    andl $1431655765, %eax # imm = 0x55555555
; X86-NEXT:    leal (%eax,%ecx,2), %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bitreverse_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    # kill: def $edi killed $edi def $rdi
; X64-NEXT:    orl $64, %edi
; X64-NEXT:    bswapl %edi
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
; X64-NEXT:    shll $4, %eax
; X64-NEXT:    shrl $4, %edi
; X64-NEXT:    andl $252645135, %edi # imm = 0xF0F0F0F
; X64-NEXT:    orl %eax, %edi
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X64-NEXT:    shrl $2, %edi
; X64-NEXT:    andl $858993459, %edi # imm = 0x33333333
; X64-NEXT:    leal (%rdi,%rax,4), %eax
; X64-NEXT:    movl %eax, %ecx
; X64-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
; X64-NEXT:    shrl %eax
; X64-NEXT:    andl $1431655765, %eax # imm = 0x55555555
; X64-NEXT:    leal (%rax,%rcx,2), %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 64
  %z = call i32 @llvm.bitreverse.i32(i32 %x)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @bitreverse_known_nonzero_vec(<4 x i32> %xx, ptr %p) nounwind {
; X86-LABEL: bitreverse_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    movdqa %xmm0, %xmm2
; X86-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
; X86-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
; X86-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
; X86-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
; X86-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
; X86-NEXT:    packuswb %xmm2, %xmm0
; X86-NEXT:    movdqa %xmm0, %xmm1
; X86-NEXT:    psrlw $4, %xmm1
; X86-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X86-NEXT:    pand %xmm2, %xmm1
; X86-NEXT:    pand %xmm2, %xmm0
; X86-NEXT:    psllw $4, %xmm0
; X86-NEXT:    por %xmm1, %xmm0
; X86-NEXT:    movdqa %xmm0, %xmm1
; X86-NEXT:    psrlw $2, %xmm1
; X86-NEXT:    movdqa {{.*#+}} xmm2 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
; X86-NEXT:    pand %xmm2, %xmm1
; X86-NEXT:    pand %xmm2, %xmm0
; X86-NEXT:    psllw $2, %xmm0
; X86-NEXT:    por %xmm1, %xmm0
; X86-NEXT:    movdqa %xmm0, %xmm1
; X86-NEXT:    psrlw $1, %xmm1
; X86-NEXT:    movdqa {{.*#+}} xmm2 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
; X86-NEXT:    pand %xmm2, %xmm1
; X86-NEXT:    pand %xmm2, %xmm0
; X86-NEXT:    paddb %xmm0, %xmm0
; X86-NEXT:    por %xmm1, %xmm0
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bitreverse_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,u,u,u,u,11,10,9,8,15,14,13,12]
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vbroadcastss {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X64-NEXT:    vpand %xmm1, %xmm0, %xmm2
; X64-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
; X64-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
; X64-NEXT:    vpsrlw $4, %xmm0, %xmm0
; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
; X64-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
; X64-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
; X64-NEXT:    vpor %xmm0, %xmm2, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 64, i32 -1, i32 0, i32 0>
  %z = call <4 x i32> @llvm.bitreverse.v4i32(<4 x i32> %x)
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @bitreverse_maybe_zero(i32 %x) {
; X86-LABEL: bitreverse_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bswapl %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
; X86-NEXT:    shll $4, %ecx
; X86-NEXT:    shrl $4, %eax
; X86-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
; X86-NEXT:    orl %ecx, %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $858993459, %ecx # imm = 0x33333333
; X86-NEXT:    shrl $2, %eax
; X86-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X86-NEXT:    leal (%eax,%ecx,4), %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
; X86-NEXT:    shrl %eax
; X86-NEXT:    andl $1431655765, %eax # imm = 0x55555555
; X86-NEXT:    leal (%eax,%ecx,2), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: bitreverse_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    # kill: def $edi killed $edi def $rdi
; X64-NEXT:    bswapl %edi
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
; X64-NEXT:    shll $4, %eax
; X64-NEXT:    shrl $4, %edi
; X64-NEXT:    andl $252645135, %edi # imm = 0xF0F0F0F
; X64-NEXT:    orl %eax, %edi
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X64-NEXT:    shrl $2, %edi
; X64-NEXT:    andl $858993459, %edi # imm = 0x33333333
; X64-NEXT:    leal (%rdi,%rax,4), %eax
; X64-NEXT:    movl %eax, %ecx
; X64-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
; X64-NEXT:    shrl %eax
; X64-NEXT:    andl $1431655765, %eax # imm = 0x55555555
; X64-NEXT:    leal (%rax,%rcx,2), %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.bitreverse.i32(i32 %x)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @ctpop_known_nonzero(i32 %xx) {
; X86-LABEL: ctpop_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl %ecx, %eax
; X86-NEXT:    orl $64, %eax
; X86-NEXT:    shrl %ecx
; X86-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
; X86-NEXT:    subl %ecx, %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $858993459, %ecx # imm = 0x33333333
; X86-NEXT:    shrl $2, %eax
; X86-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X86-NEXT:    addl %ecx, %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    shrl $4, %ecx
; X86-NEXT:    addl %eax, %ecx
; X86-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
; X86-NEXT:    imull $16843009, %ecx, %eax # imm = 0x1010101
; X86-NEXT:    shrl $24, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: ctpop_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    orl $64, %eax
; X64-NEXT:    shrl %edi
; X64-NEXT:    andl $1431655765, %edi # imm = 0x55555555
; X64-NEXT:    subl %edi, %eax
; X64-NEXT:    movl %eax, %ecx
; X64-NEXT:    andl $858993459, %ecx # imm = 0x33333333
; X64-NEXT:    shrl $2, %eax
; X64-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X64-NEXT:    addl %ecx, %eax
; X64-NEXT:    movl %eax, %ecx
; X64-NEXT:    shrl $4, %ecx
; X64-NEXT:    addl %eax, %ecx
; X64-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
; X64-NEXT:    imull $16843009, %ecx, %eax # imm = 0x1010101
; X64-NEXT:    shrl $24, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or i32 %xx, 64
  %z = call i32 @llvm.ctpop.i32(i32 %x)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @ctpop_known_nonzero_vec(<4 x i32> %xx, ptr %p) nounwind {
; X86-LABEL: ctpop_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    movdqa %xmm0, %xmm1
; X86-NEXT:    psrlw $1, %xmm1
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    psubb %xmm1, %xmm0
; X86-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
; X86-NEXT:    movdqa %xmm0, %xmm2
; X86-NEXT:    pand %xmm1, %xmm2
; X86-NEXT:    psrlw $2, %xmm0
; X86-NEXT:    pand %xmm1, %xmm0
; X86-NEXT:    paddb %xmm2, %xmm0
; X86-NEXT:    movdqa %xmm0, %xmm1
; X86-NEXT:    psrlw $4, %xmm1
; X86-NEXT:    paddb %xmm0, %xmm1
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
; X86-NEXT:    pxor %xmm0, %xmm0
; X86-NEXT:    movdqa %xmm1, %xmm2
; X86-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; X86-NEXT:    psadbw %xmm0, %xmm2
; X86-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; X86-NEXT:    psadbw %xmm0, %xmm1
; X86-NEXT:    packuswb %xmm2, %xmm1
; X86-NEXT:    movdqa %xmm1, (%eax)
; X86-NEXT:    movd %xmm1, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: ctpop_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vbroadcastss {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X64-NEXT:    vpand %xmm1, %xmm0, %xmm2
; X64-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
; X64-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
; X64-NEXT:    vpsrlw $4, %xmm0, %xmm0
; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
; X64-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
; X64-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
; X64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X64-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
; X64-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; X64-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
; X64-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = or <4 x i32> %xx, <i32 64, i32 -1, i32 0, i32 0>
  %z = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %x)
  store <4 x i32> %z, ptr %p
  %e = extractelement <4 x i32> %z, i32 0
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @ctpop_maybe_zero(i32 %x) {
; X86-LABEL: ctpop_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    shrl %ecx
; X86-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
; X86-NEXT:    subl %ecx, %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    andl $858993459, %ecx # imm = 0x33333333
; X86-NEXT:    shrl $2, %eax
; X86-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X86-NEXT:    addl %ecx, %eax
; X86-NEXT:    movl %eax, %ecx
; X86-NEXT:    shrl $4, %ecx
; X86-NEXT:    addl %eax, %ecx
; X86-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
; X86-NEXT:    imull $16843009, %ecx, %eax # imm = 0x1010101
; X86-NEXT:    shrl $24, %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: ctpop_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    shrl %eax
; X64-NEXT:    andl $1431655765, %eax # imm = 0x55555555
; X64-NEXT:    subl %eax, %edi
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    andl $858993459, %eax # imm = 0x33333333
; X64-NEXT:    shrl $2, %edi
; X64-NEXT:    andl $858993459, %edi # imm = 0x33333333
; X64-NEXT:    addl %eax, %edi
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    shrl $4, %eax
; X64-NEXT:    addl %edi, %eax
; X64-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
; X64-NEXT:    imull $16843009, %eax, %ecx # imm = 0x1010101
; X64-NEXT:    shrl $24, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = call i32 @llvm.ctpop.i32(i32 %x)
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @zext_known_nonzero(i16 %xx) {
; X86-LABEL: zext_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    shll %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: zext_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $256, %eax # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = shl nuw nsw i16 256, %xx
  %z = zext i16 %x to i32
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @zext_maybe_zero(i16 %x) {
; X86-LABEL: zext_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: zext_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movzwl %di, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = zext i16 %x to i32
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sext_known_nonzero(i16 %xx) {
; X86-LABEL: sext_known_nonzero:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl $256, %eax # imm = 0x100
; X86-NEXT:    shll %cl, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sext_known_nonzero:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %ecx
; X64-NEXT:    movl $256, %eax # imm = 0x100
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shll %cl, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = shl nuw nsw i16 256, %xx
  %z = sext i16 %x to i32
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @sext_known_nonzero_vec(<8 x i16> %xx, ptr %p) {
; X86-LABEL: sext_known_nonzero_vec:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT:    pslld $23, %xmm0
; X86-NEXT:    paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    cvttps2dq %xmm0, %xmm0
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,2,4,5,6,7]
; X86-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6,4]
; X86-NEXT:    psrad $16, %xmm0
; X86-NEXT:    movdqa %xmm1, 16(%eax)
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; X86-NEXT:    movd %xmm0, %eax
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sext_known_nonzero_vec:
; X64:       # %bb.0:
; X64-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
; X64-NEXT:    vpslld $23, %xmm0, %xmm0
; X64-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; X64-NEXT:    vcvttps2dq %xmm0, %xmm0
; X64-NEXT:    vpackusdw %xmm0, %xmm0, %xmm0
; X64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; X64-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3,4,5,6,7]
; X64-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
; X64-NEXT:    vpmovsxwd %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm1, 16(%rdi)
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vpextrd $1, %xmm0, %eax
; X64-NEXT:    rep bsfl %eax, %eax
; X64-NEXT:    retq
  %x = shl <8 x i16> <i16 0, i16 0, i16 1, i16 0, i16 0, i16 0, i16 0, i16 0>, %xx
  %s = shufflevector <8 x i16> %x, <8 x i16> poison, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7>
  %z = sext <8 x i16> %s to <8 x i32>
  store <8 x i32> %z, ptr %p
  %e = extractelement <8 x i32> %z, i32 1
  %r = call i32 @llvm.cttz.i32(i32 %e, i1 false)
  ret i32 %r
}

define i32 @sext_maybe_zero(i16 %x) {
; X86-LABEL: sext_maybe_zero:
; X86:       # %bb.0:
; X86-NEXT:    movswl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    bsfl %eax, %ecx
; X86-NEXT:    movl $32, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: sext_maybe_zero:
; X64:       # %bb.0:
; X64-NEXT:    movswl %di, %ecx
; X64-NEXT:    movl $32, %eax
; X64-NEXT:    rep bsfl %ecx, %eax
; X64-NEXT:    retq
  %z = sext i16 %x to i32
  %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
  ret i32 %r
}

define i32 @test_zext_demanded_elts(<4 x i32> %a0, ptr %p) {
; X86-LABEL: test_zext_demanded_elts:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pxor %xmm2, %xmm2
; X86-NEXT:    pcmpgtd %xmm0, %xmm2
; X86-NEXT:    movdqa %xmm2, %xmm0
; X86-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm2
; X86-NEXT:    por %xmm0, %xmm2
; X86-NEXT:    movdqa %xmm2, %xmm0
; X86-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; X86-NEXT:    movd %xmm2, %ecx
; X86-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; X86-NEXT:    movdqa %xmm2, 16(%eax)
; X86-NEXT:    movdqa %xmm0, (%eax)
; X86-NEXT:    bsfl %ecx, %ecx
; X86-NEXT:    movl $64, %eax
; X86-NEXT:    cmovnel %ecx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: test_zext_demanded_elts:
; X64:       # %bb.0:
; X64-NEXT:    vmovaps {{.*#+}} xmm1 = [2,4294967295,4294967295,4294967295]
; X64-NEXT:    vblendvps %xmm0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
; X64-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero
; X64-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; X64-NEXT:    vunpckhps {{.*#+}} xmm2 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
; X64-NEXT:    vmovaps %xmm2, 16(%rdi)
; X64-NEXT:    vmovdqa %xmm1, (%rdi)
; X64-NEXT:    vmovd %xmm0, %eax
; X64-NEXT:    rep bsfq %rax, %rax
; X64-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NEXT:    retq
  %cmp = icmp sgt <4 x i32> zeroinitializer, %a0
  %sel = select <4 x i1> %cmp, <4 x i32> <i32 1, i32 0, i32 0, i32 0>, <4 x i32> <i32 2, i32 -1, i32 -1, i32 -1>

  %ext = zext <4 x i32> %sel to <4 x i64>
  store <4 x i64> %ext, ptr %p

  %lane0 = extractelement <4 x i64> %ext, i32 0
  %tz = call i64 @llvm.cttz.i64(i64 %lane0, i1 false)
  %res = trunc i64 %tz to i32
  ret i32 %res
}

define i32 @test_sext_demanded_elts(<4 x i32> %a0, ptr %p) {
; X86-LABEL: test_sext_demanded_elts:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    pxor %xmm1, %xmm1
; X86-NEXT:    pxor %xmm2, %xmm2
; X86-NEXT:    pcmpgtd %xmm0, %xmm2
; X86-NEXT:    movdqa %xmm2, %xmm0
; X86-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm2
; X86-NEXT:    por %xmm0, %xmm2
; X86-NEXT:    pcmpgtd %xmm2, %xmm1
; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,2,3,3]
; X86-NEXT:    movd %xmm2, %ecx
; X86-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; X86-NEXT:    movdqa %xmm0, 16(%eax)
; X86-NEXT:    movdqa %xmm2, (%eax)
; X86-NEXT:    movd %xmm1, %eax
; X86-NEXT:    rep bsfl %ecx, %edx
; X86-NEXT:    rep bsfl %eax, %eax
; X86-NEXT:    addl $32, %eax
; X86-NEXT:    testl %ecx, %ecx
; X86-NEXT:    cmovnel %edx, %eax
; X86-NEXT:    retl
;
; X64-LABEL: test_sext_demanded_elts:
; X64:       # %bb.0:
; X64-NEXT:    vmovaps {{.*#+}} xmm1 = [2,4294967295,4294967295,4294967295]
; X64-NEXT:    vblendvps %xmm0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
; X64-NEXT:    vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
; X64-NEXT:    vpmovsxdq %xmm1, %xmm1
; X64-NEXT:    vpmovsxdq %xmm0, %xmm0
; X64-NEXT:    vmovdqa %xmm0, (%rdi)
; X64-NEXT:    vmovdqa %xmm1, 16(%rdi)
; X64-NEXT:    vmovq %xmm0, %rax
; X64-NEXT:    rep bsfq %rax, %rax
; X64-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NEXT:    retq
  %cmp = icmp sgt <4 x i32> zeroinitializer, %a0
  %sel = select <4 x i1> %cmp, <4 x i32> <i32 1, i32 0, i32 0, i32 0>, <4 x i32> <i32 2, i32 -1, i32 -1, i32 -1>

  %ext = sext <4 x i32> %sel to <4 x i64>
  store <4 x i64> %ext, ptr %p

  %lane0 = extractelement <4 x i64> %ext, i32 0
  %tz = call i64 @llvm.cttz.i64(i64 %lane0, i1 false)
  %res = trunc i64 %tz to i32
  ret i32 %res
}