; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -passes=loop-interchange -cache-line-size=64 -S %s | FileCheck %s

@global = external local_unnamed_addr global [400 x [400 x i32]], align 16

; We need to move %add from the inner loop pre header to the outer loop header
; before interchanging.
;
; outer_red = 0;
; for (i = 0; i < 391; i++) {
;   add = i + 9;
;   inner_red = outer_red;
;   for (j = 0; j < 400; j++) {
;     global[j][add] = 0;
;     inner_red |= 20;
;   }
;   outer_red = inner_red;
; }
;
define void @test1() local_unnamed_addr #0 {
; CHECK-LABEL: define void @test1() local_unnamed_addr {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    br label %[[INNER_PH:.*]]
; CHECK:       [[OUTER_HEADER_PREHEADER:.*]]:
; CHECK-NEXT:    br label %[[OUTER_HEADER:.*]]
; CHECK:       [[OUTER_HEADER]]:
; CHECK-NEXT:    [[OUTER_IV:%.*]] = phi i64 [ [[OUTER_IV_NEXT:%.*]], %[[OUTER_LATCH:.*]] ], [ 0, %[[OUTER_HEADER_PREHEADER]] ]
; CHECK-NEXT:    [[INNER_RED:%.*]] = phi i32 [ [[OUTER_RED:%.*]], %[[OUTER_HEADER_PREHEADER]] ], [ [[RED_NEXT:%.*]], %[[OUTER_LATCH]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = add nsw i64 [[OUTER_IV]], 9
; CHECK-NEXT:    br label %[[INNER_SPLIT1:.*]]
; CHECK:       [[INNER_PH]]:
; CHECK-NEXT:    br label %[[INNER:.*]]
; CHECK:       [[INNER]]:
; CHECK-NEXT:    [[INNER_IV:%.*]] = phi i64 [ 0, %[[INNER_PH]] ], [ [[TMP0:%.*]], %[[INNER_SPLIT:.*]] ]
; CHECK-NEXT:    [[OUTER_RED]] = phi i32 [ [[RED_NEXT_LCSSA:%.*]], %[[INNER_SPLIT]] ], [ 0, %[[INNER_PH]] ]
; CHECK-NEXT:    br label %[[OUTER_HEADER_PREHEADER]]
; CHECK:       [[INNER_SPLIT1]]:
; CHECK-NEXT:    [[PTR:%.*]] = getelementptr inbounds [400 x [400 x i32]], ptr @global, i64 0, i64 [[INNER_IV]], i64 [[TMP4]]
; CHECK-NEXT:    store i32 0, ptr [[PTR]], align 4
; CHECK-NEXT:    [[RED_NEXT]] = or i32 [[INNER_RED]], 20
; CHECK-NEXT:    [[INNER_IV_NEXT:%.*]] = add nsw i64 [[INNER_IV]], 1
; CHECK-NEXT:    [[EC_1:%.*]] = icmp eq i64 [[INNER_IV_NEXT]], 400
; CHECK-NEXT:    br label %[[OUTER_LATCH]]
; CHECK:       [[INNER_SPLIT]]:
; CHECK-NEXT:    [[RED_NEXT_LCSSA]] = phi i32 [ [[RED_NEXT]], %[[OUTER_LATCH]] ]
; CHECK-NEXT:    [[TMP0]] = add nsw i64 [[INNER_IV]], 1
; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i64 [[TMP0]], 400
; CHECK-NEXT:    br i1 [[TMP1]], label %[[EXIT:.*]], label %[[INNER]]
; CHECK:       [[OUTER_LATCH]]:
; CHECK-NEXT:    [[OUTER_IV_NEXT]] = add nsw i64 [[OUTER_IV]], 1
; CHECK-NEXT:    [[EC_2:%.*]] = icmp eq i64 [[OUTER_IV_NEXT]], 391
; CHECK-NEXT:    br i1 [[EC_2]], label %[[INNER_SPLIT]], label %[[OUTER_HEADER]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
bb:
  br label %outer.header

outer.header:                                              ; preds = %bb11, %bb
  %outer.iv = phi i64 [ 0, %bb ], [ %outer.iv.next, %outer.latch ]
  %outer.red = phi i32 [ 0, %bb ], [ %red.next.lcssa, %outer.latch ]
  br label %inner.ph

inner.ph:                                              ; preds = %bb1
  %add = add nsw i64 %outer.iv, 9
  br label %inner

inner:                                              ; preds = %bb5, %bb3
  %inner.iv = phi i64 [ 0, %inner.ph ], [ %inner.iv.next, %inner ]
  %inner.red = phi i32 [ %outer.red, %inner.ph ], [ %red.next, %inner ]
  %ptr = getelementptr inbounds [400 x [400 x i32]], ptr @global, i64 0, i64 %inner.iv, i64 %add
  store i32 0, ptr %ptr
  %red.next = or i32 %inner.red, 20
  %inner.iv.next = add nsw i64 %inner.iv, 1
  %ec.1 = icmp eq i64 %inner.iv.next, 400
  br i1 %ec.1, label %outer.latch, label %inner

outer.latch:                                             ; preds = %bb5
  %red.next.lcssa = phi i32 [ %red.next, %inner ]
  %outer.iv.next = add nsw i64 %outer.iv, 1
  %ec.2 = icmp eq i64 %outer.iv.next, 391
  br i1 %ec.2, label %exit, label %outer.header

exit:                                             ; preds = %bb11
  ret void
}

declare void @side_effect()

; Cannot interchange, as the inner loop preheader contains a call to a function
; with side effects.
;
; outer_red = 0;
; for (i = 0; i < 391; i++) {
;   add = i + 9;
;   side_effect();
;   inner_red = outer_red;
;   for (j = 0; j < 400; j++) {
;     global[j][add] = 0;
;     inner_red |= 20;
;   }
;   outer_red = inner_red;
; }
;

define void @test2() {
; CHECK-LABEL: define void @test2() {
; CHECK-NEXT:  [[BB:.*]]:
; CHECK-NEXT:    br label %[[OUTER_HEADER:.*]]
; CHECK:       [[OUTER_HEADER]]:
; CHECK-NEXT:    [[OUTER_IV:%.*]] = phi i64 [ 0, %[[BB]] ], [ [[OUTER_IV_NEXT:%.*]], %[[OUTER_LATCH:.*]] ]
; CHECK-NEXT:    [[OUTER_RED:%.*]] = phi i32 [ 0, %[[BB]] ], [ [[RED_NEXT_LCSSA:%.*]], %[[OUTER_LATCH]] ]
; CHECK-NEXT:    br label %[[INNER_PH:.*]]
; CHECK:       [[INNER_PH]]:
; CHECK-NEXT:    [[TMP4:%.*]] = add nsw i64 [[OUTER_IV]], 9
; CHECK-NEXT:    call void @side_effect()
; CHECK-NEXT:    br label %[[INNER:.*]]
; CHECK:       [[INNER]]:
; CHECK-NEXT:    [[INNER_IV:%.*]] = phi i64 [ 0, %[[INNER_PH]] ], [ [[INNER_IV_NEXT:%.*]], %[[INNER]] ]
; CHECK-NEXT:    [[INNER_RED:%.*]] = phi i32 [ [[OUTER_RED]], %[[INNER_PH]] ], [ [[RED_NEXT:%.*]], %[[INNER]] ]
; CHECK-NEXT:    [[PTR:%.*]] = getelementptr inbounds [400 x [400 x i32]], ptr @global, i64 0, i64 [[INNER_IV]], i64 [[TMP4]]
; CHECK-NEXT:    store i32 0, ptr [[PTR]], align 4
; CHECK-NEXT:    [[RED_NEXT]] = or i32 [[INNER_RED]], 20
; CHECK-NEXT:    [[INNER_IV_NEXT]] = add nsw i64 [[INNER_IV]], 1
; CHECK-NEXT:    [[EC_1:%.*]] = icmp eq i64 [[INNER_IV_NEXT]], 400
; CHECK-NEXT:    br i1 [[EC_1]], label %[[OUTER_LATCH]], label %[[INNER]]
; CHECK:       [[OUTER_LATCH]]:
; CHECK-NEXT:    [[RED_NEXT_LCSSA]] = phi i32 [ [[RED_NEXT]], %[[INNER]] ]
; CHECK-NEXT:    [[OUTER_IV_NEXT]] = add nsw i64 [[OUTER_IV]], 1
; CHECK-NEXT:    [[EC_2:%.*]] = icmp eq i64 [[OUTER_IV_NEXT]], 391
; CHECK-NEXT:    br i1 [[EC_2]], label %[[EXIT:.*]], label %[[OUTER_HEADER]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
bb:
  br label %outer.header

outer.header:                                              ; preds = %bb11, %bb
  %outer.iv = phi i64 [ 0, %bb ], [ %outer.iv.next, %outer.latch ]
  %outer.red = phi i32 [ 0, %bb ], [ %red.next.lcssa, %outer.latch ]
  br label %inner.ph

inner.ph:                                              ; preds = %bb1
  %add = add nsw i64 %outer.iv, 9
  call void @side_effect()
  br label %inner

inner:                                              ; preds = %bb5, %bb3
  %inner.iv = phi i64 [ 0, %inner.ph ], [ %inner.iv.next, %inner ]
  %inner.red = phi i32 [ %outer.red, %inner.ph ], [ %red.next, %inner ]
  %ptr = getelementptr inbounds [400 x [400 x i32]], ptr @global, i64 0, i64 %inner.iv, i64 %add
  store i32 0, ptr %ptr
  %red.next = or i32 %inner.red, 20
  %inner.iv.next = add nsw i64 %inner.iv, 1
  %ec.1 = icmp eq i64 %inner.iv.next, 400
  br i1 %ec.1, label %outer.latch, label %inner

outer.latch:                                             ; preds = %bb5
  %red.next.lcssa = phi i32 [ %red.next, %inner ]
  %outer.iv.next = add nsw i64 %outer.iv, 1
  %ec.2 = icmp eq i64 %outer.iv.next, 391
  br i1 %ec.2, label %exit, label %outer.header

exit:                                             ; preds = %bb11
  ret void
}