; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -aa-pipeline=basic-aa -passes='require<opt-remark-emit>,loop-mssa(licm)' < %s -S | FileCheck %s

; Check that we can hoist unordered loads
define i32 @test1(ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test1(
; CHECK-SAME: ptr captures(none) [[Y:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    [[VAL:%.*]] = load atomic i32, ptr [[Y]] unordered, align 4
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[INC:%.*]], %[[LOOP]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT:    [[INC]] = add nsw i32 [[I]], 1
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[VAL]]
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VAL_LCSSA:%.*]] = phi i32 [ [[VAL]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VAL_LCSSA]]
;
entry:
  br label %loop

loop:
  %i = phi i32 [ %inc, %loop ], [ 0, %entry ]
  %val = load atomic i32, ptr %y unordered, align 4
  %inc = add nsw i32 %i, 1
  %exitcond = icmp eq i32 %inc, %val
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %val
}

; Check that we don't sink/hoist monotonic loads
; (Strictly speaking, it's not forbidden, but it's supposed to be possible to
; use monotonic for spinlock-like constructs.)
define i32 @test2(ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test2(
; CHECK-SAME: ptr captures(none) [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VAL:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VAL]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VAL_LCSSA:%.*]] = phi i32 [ [[VAL]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VAL_LCSSA]]
;
entry:
  br label %loop

loop:
  %val = load atomic i32, ptr %y monotonic, align 4
  %exitcond = icmp ne i32 %val, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %val
}

; Check that we hoist unordered around monotonic.
; (The noalias shouldn't be necessary in theory, but LICM isn't quite that
; smart yet.)
define i32 @test3(ptr nocapture noalias %x, ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test3(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[VALB:%.*]] = load atomic i32, ptr [[X]] unordered, align 4
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], [[VALB]]
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  %vala = load atomic i32, ptr %y monotonic, align 4
  %valb = load atomic i32, ptr %x unordered, align 4
  %exitcond = icmp ne i32 %vala, %valb
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}

; We can sink an unordered store
define i32 @test4(ptr nocapture noalias %x, ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test4(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA1:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    store atomic i32 [[VALA_LCSSA1]], ptr [[X]] unordered, align 4
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  %vala = load atomic i32, ptr %y monotonic, align 4
  store atomic i32 %vala, ptr %x unordered, align 4
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}

; We currently don't handle ordered atomics.
define i32 @test5(ptr nocapture noalias %x, ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test5(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    store atomic i32 [[VALA]], ptr [[X]] release, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  %vala = load atomic i32, ptr %y monotonic, align 4
  store atomic i32 %vala, ptr %x release, align 4
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}

; We currently don't touch volatiles
define i32 @test6(ptr nocapture noalias %x, ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test6(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    store volatile i32 [[VALA]], ptr [[X]], align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  %vala = load atomic i32, ptr %y monotonic, align 4
  store volatile i32 %vala, ptr %x, align 4
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}

; We currently don't touch volatiles
define i32 @test6b(ptr nocapture noalias %x, ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test6b(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    store atomic volatile i32 [[VALA]], ptr [[X]] unordered, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  %vala = load atomic i32, ptr %y monotonic, align 4
  store atomic volatile i32 %vala, ptr %x unordered, align 4
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}

; Mixing unorder atomics and normal loads/stores is
; current unimplemented
define i32 @test7(ptr nocapture noalias %x, ptr nocapture %y) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test7(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    store i32 5, ptr [[X]], align 4
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    store atomic i32 [[VALA]], ptr [[X]] unordered, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  store i32 5, ptr %x
  %vala = load atomic i32, ptr %y monotonic, align 4
  store atomic i32 %vala, ptr %x unordered, align 4
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}

; Three provably noalias locations - we can sink normal and unordered, but
;  not monotonic
define i32 @test7b(ptr nocapture noalias %x, ptr nocapture %y, ptr noalias nocapture %z) nounwind uwtable ssp {
; CHECK-LABEL: define i32 @test7b(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]], ptr noalias captures(none) [[Z:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA1:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    store i32 5, ptr [[X]], align 4
; CHECK-NEXT:    store atomic i32 [[VALA_LCSSA1]], ptr [[Z]] unordered, align 4
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  store i32 5, ptr %x
  %vala = load atomic i32, ptr %y monotonic, align 4
  store atomic i32 %vala, ptr %z unordered, align 4
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}


define i32 @test8(ptr nocapture noalias %x, ptr nocapture %y) {
; CHECK-LABEL: define i32 @test8(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    store atomic i32 [[VALA]], ptr [[X]] unordered, align 4
; CHECK-NEXT:    fence release
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  %vala = load atomic i32, ptr %y monotonic, align 4
  store atomic i32 %vala, ptr %x unordered, align 4
  fence release
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}

; Exact semantics of monotonic accesses are a bit vague in the C++ spec,
; for the moment, be conservative and don't touch them.
define i32 @test9(ptr nocapture noalias %x, ptr nocapture %y) {
; CHECK-LABEL: define i32 @test9(
; CHECK-SAME: ptr noalias captures(none) [[X:%.*]], ptr captures(none) [[Y:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[VALA:%.*]] = load atomic i32, ptr [[Y]] monotonic, align 4
; CHECK-NEXT:    store atomic i32 [[VALA]], ptr [[X]] monotonic, align 4
; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[VALA]], 0
; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[END:.*]], label %[[LOOP]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[VALA_LCSSA:%.*]] = phi i32 [ [[VALA]], %[[LOOP]] ]
; CHECK-NEXT:    ret i32 [[VALA_LCSSA]]
;
entry:
  br label %loop

loop:
  %vala = load atomic i32, ptr %y monotonic, align 4
  store atomic i32 %vala, ptr %x monotonic, align 4
  %exitcond = icmp ne i32 %vala, 0
  br i1 %exitcond, label %end, label %loop

end:
  ret i32 %vala
}