; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -S -passes=loop-fusion < %s | FileCheck %s

; The expected output keeps body1 and body2 as separate loops.
; body2 must execute only after body1 finishes so every store to %out
; uses the final value left in %ptr.

define void @fill(i32 %N, ptr %out) {
; CHECK-LABEL: define void @fill(
; CHECK-SAME: i32 [[N:%.*]], ptr [[OUT:%.*]]) {
; CHECK-NEXT:  [[PRE1:.*]]:
; CHECK-NEXT:    [[PTR:%.*]] = alloca i32, align 4
; CHECK-NEXT:    br label %[[BODY1:.*]]
; CHECK:       [[BODY1]]:
; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 0, %[[PRE1]] ], [ [[I_NEXT:%.*]], %[[BODY1]] ]
; CHECK-NEXT:    store i32 [[I]], ptr [[PTR]], align 4
; CHECK-NEXT:    [[I_NEXT]] = add i32 [[I]], 1
; CHECK-NEXT:    [[COND_NOT:%.*]] = icmp eq i32 [[I_NEXT]], [[N]]
; CHECK-NEXT:    br i1 [[COND_NOT]], label %[[BODY2_PREHEADER:.*]], label %[[BODY1]]
; CHECK:       [[BODY2_PREHEADER]]:
; CHECK-NEXT:    br label %[[BODY2:.*]]
; CHECK:       [[BODY2]]:
; CHECK-NEXT:    [[J:%.*]] = phi i32 [ [[J_NEXT:%.*]], %[[BODY2]] ], [ 0, %[[BODY2_PREHEADER]] ]
; CHECK-NEXT:    [[VAL:%.*]] = load i32, ptr [[PTR]], align 4
; CHECK-NEXT:    [[TMP0:%.*]] = sext i32 [[J]] to i64
; CHECK-NEXT:    [[OUT_PTR:%.*]] = getelementptr [4 x i8], ptr [[OUT]], i64 [[TMP0]]
; CHECK-NEXT:    store i32 [[VAL]], ptr [[OUT_PTR]], align 4
; CHECK-NEXT:    [[J_NEXT]] = add i32 [[J]], 1
; CHECK-NEXT:    [[COND2_NOT:%.*]] = icmp eq i32 [[J_NEXT]], [[N]]
; CHECK-NEXT:    br i1 [[COND2_NOT]], label %[[EXIT:.*]], label %[[BODY2]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
pre1:
  %ptr = alloca i32, align 4
  br label %body1

body1:                                            ; preds = %body1, %pre1
  %i = phi i32 [ 0, %pre1 ], [ %i_next, %body1 ]
  store i32 %i, ptr %ptr, align 4
  %i_next = add i32 %i, 1
  %cond.not = icmp eq i32 %i_next, %N
  br i1 %cond.not, label %body2, label %body1

body2:                                            ; preds = %body2, %body1
  %j = phi i32 [ %j_next, %body2 ], [ 0, %body1 ]
  %val = load i32, ptr %ptr, align 4
  %0 = sext i32 %j to i64
  %out_ptr = getelementptr [4 x i8], ptr %out, i64 %0
  store i32 %val, ptr %out_ptr, align 4
  %j_next = add i32 %j, 1
  %cond2.not = icmp eq i32 %j_next, %N
  br i1 %cond2.not, label %exit, label %body2

exit:                                             ; preds = %body2
  ret void
}