; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple armv7-- -mattr=+vfp4 %s -o - | FileCheck %s


; Div whose result is unused should be removed unless we have strict exceptions

define void @unused_div(float %x, float %y) {
; CHECK-LABEL: unused_div:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    bx lr
entry:
  %add = fdiv float %x, %y
  ret void
}

define void @unused_div_fpexcept_strict(float %x, float %y) #0 {
; CHECK-LABEL: unused_div_fpexcept_strict:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmov s0, r1
; CHECK-NEXT:    vmov s2, r0
; CHECK-NEXT:    vdiv.f32 s0, s2, s0
; CHECK-NEXT:    bx lr
entry:
  %add = call float @llvm.experimental.constrained.fdiv.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
  ret void
}

define void @unused_div_round_dynamic(float %x, float %y) #0 {
; CHECK-LABEL: unused_div_round_dynamic:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    bx lr
entry:
  %add = call float @llvm.experimental.constrained.fdiv.f32(float %x, float %y, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
  ret void
}


; Machine CSE should eliminate the second add unless we have strict exceptions

define float @add_twice(float %x, float %y, i32 %n) {
; CHECK-LABEL: add_twice:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmov s0, r1
; CHECK-NEXT:    cmp r2, #0
; CHECK-NEXT:    vmov s2, r0
; CHECK-NEXT:    vadd.f32 s0, s2, s0
; CHECK-NEXT:    vmul.f32 s2, s0, s0
; CHECK-NEXT:    vmoveq.f32 s2, s0
; CHECK-NEXT:    vmov r0, s2
; CHECK-NEXT:    bx lr
entry:
  %add = fadd float %x, %y
  %tobool.not = icmp eq i32 %n, 0
  br i1 %tobool.not, label %if.end, label %if.then

if.then:
  %add1 = fadd float %x, %y
  %mul = fmul float %add, %add1
  br label %if.end

if.end:
  %a.0 = phi float [ %mul, %if.then ], [ %add, %entry ]
  ret float %a.0
}

define float @add_twice_fpexcept_strict(float %x, float %y, i32 %n) #0 {
; CHECK-LABEL: add_twice_fpexcept_strict:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmov s2, r1
; CHECK-NEXT:    cmp r2, #0
; CHECK-NEXT:    vmov s4, r0
; CHECK-NEXT:    vadd.f32 s0, s4, s2
; CHECK-NEXT:    vaddne.f32 s2, s4, s2
; CHECK-NEXT:    vmulne.f32 s0, s0, s2
; CHECK-NEXT:    vmov r0, s0
; CHECK-NEXT:    bx lr
entry:
  %add = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
  %tobool.not = icmp eq i32 %n, 0
  br i1 %tobool.not, label %if.end, label %if.then

if.then:
  %add1 = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
  %mul = call float @llvm.experimental.constrained.fmul.f32(float %add, float %add1, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
  br label %if.end

if.end:
  %a.0 = phi float [ %mul, %if.then ], [ %add, %entry ]
  ret float %a.0
}

define float @add_twice_round_dynamic(float %x, float %y, i32 %n) #0 {
; CHECK-LABEL: add_twice_round_dynamic:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmov s0, r1
; CHECK-NEXT:    cmp r2, #0
; CHECK-NEXT:    vmov s2, r0
; CHECK-NEXT:    vadd.f32 s0, s2, s0
; CHECK-NEXT:    vmulne.f32 s0, s0, s0
; CHECK-NEXT:    vmov r0, s0
; CHECK-NEXT:    bx lr
entry:
  %add = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
  %tobool.not = icmp eq i32 %n, 0
  br i1 %tobool.not, label %if.end, label %if.then

if.then:
  %add1 = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
  %mul = call float @llvm.experimental.constrained.fmul.f32(float %add, float %add1, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
  br label %if.end

if.end:
  %a.0 = phi float [ %mul, %if.then ], [ %add, %entry ]
  ret float %a.0
}

; Two adds separated by llvm.set.rounding should be preserved when rounding is
; dynamic (as they may give different results) or when we have strict exceptions
; (the llvm.set.rounding is irrelevant, but both could trap).

define float @set_rounding(float %x, float %y) {
; CHECK-LABEL: set_rounding:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmrs r2, fpscr
; CHECK-NEXT:    vmov s2, r0
; CHECK-NEXT:    vmov s0, r1
; CHECK-NEXT:    vadd.f32 s0, s2, s0
; CHECK-NEXT:    vsub.f32 s0, s0, s0
; CHECK-NEXT:    orr r0, r2, #12582912
; CHECK-NEXT:    vmsr fpscr, r0
; CHECK-NEXT:    vmov r0, s0
; CHECK-NEXT:    vmrs r1, fpscr
; CHECK-NEXT:    bic r1, r1, #12582912
; CHECK-NEXT:    vmsr fpscr, r1
; CHECK-NEXT:    bx lr
entry:
  %add1 = fadd float %x, %y
  call void @llvm.set.rounding(i32 0)
  %add2 = fadd float %x, %y
  call void @llvm.set.rounding(i32 1)
  %sub = fsub float %add1, %add2
  ret float %sub
}

define float @set_rounding_fpexcept_strict(float %x, float %y) #0 {
; CHECK-LABEL: set_rounding_fpexcept_strict:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmov s0, r1
; CHECK-NEXT:    vmov s2, r0
; CHECK-NEXT:    vadd.f32 s4, s2, s0
; CHECK-NEXT:    vmrs r0, fpscr
; CHECK-NEXT:    orr r0, r0, #12582912
; CHECK-NEXT:    vmsr fpscr, r0
; CHECK-NEXT:    vadd.f32 s0, s2, s0
; CHECK-NEXT:    vmrs r0, fpscr
; CHECK-NEXT:    bic r0, r0, #12582912
; CHECK-NEXT:    vmsr fpscr, r0
; CHECK-NEXT:    vsub.f32 s0, s4, s0
; CHECK-NEXT:    vmov r0, s0
; CHECK-NEXT:    bx lr
entry:
  %add1 = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
  call void @llvm.set.rounding(i32 0) #0
  %add2 = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
  call void @llvm.set.rounding(i32 1) #0
  %sub = call float @llvm.experimental.constrained.fsub.f32(float %add1, float %add2, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
  ret float %sub
}

define float @set_rounding_round_dynamic(float %x, float %y) #0 {
; CHECK-LABEL: set_rounding_round_dynamic:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmov s2, r0
; CHECK-NEXT:    vmrs r0, fpscr
; CHECK-NEXT:    vmov s0, r1
; CHECK-NEXT:    vadd.f32 s4, s2, s0
; CHECK-NEXT:    orr r0, r0, #12582912
; CHECK-NEXT:    vmsr fpscr, r0
; CHECK-NEXT:    vmrs r0, fpscr
; CHECK-NEXT:    vadd.f32 s0, s2, s0
; CHECK-NEXT:    bic r0, r0, #12582912
; CHECK-NEXT:    vmsr fpscr, r0
; CHECK-NEXT:    vsub.f32 s0, s4, s0
; CHECK-NEXT:    vmov r0, s0
; CHECK-NEXT:    bx lr
entry:
  %add1 = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
  call void @llvm.set.rounding(i32 0) #0
  %add2 = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
  call void @llvm.set.rounding(i32 1) #0
  %sub = call float @llvm.experimental.constrained.fsub.f32(float %add1, float %add2, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
  ret float %sub
}

declare float @llvm.experimental.constrained.fadd.f32(float, float, metadata, metadata)
declare float @llvm.experimental.constrained.fsub.f32(float, float, metadata, metadata)
declare float @llvm.experimental.constrained.fmul.f32(float, float, metadata, metadata)
declare float @llvm.experimental.constrained.fdiv.f32(float, float, metadata, metadata)
declare i32 @llvm.get.rounding()
declare void @llvm.set.rounding(i32)

attributes #0 = { strictfp }