xref: /llvm-project/llvm/test/Transforms/LoopVectorize/fixed-order-recurrences-memory-instructions.ll (revision fd5bccb8b15a58232d4ffb81cb801a32ef1a2716) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S %s | FileCheck %s
3
4define i32 @can_sink_load_that_uses_for_past_other_loads(ptr noalias %A, ptr noalias %B) {
5; CHECK-LABEL: @can_sink_load_that_uses_for_past_other_loads(
6; CHECK-NEXT:  entry:
7; CHECK-NEXT:    br label [[LOOP:%.*]]
8; CHECK:       loop:
9; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
10; CHECK-NEXT:    [[FOR:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[FOR_NEXT:%.*]], [[LOOP]] ]
11; CHECK-NEXT:    [[RED:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[RED_NEXT:%.*]], [[LOOP]] ]
12; CHECK-NEXT:    [[GEP_A:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[FOR]]
13; CHECK-NEXT:    [[L_A:%.*]] = load i32, ptr [[GEP_A]], align 4
14; CHECK-NEXT:    [[GEP_B:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 [[IV]]
15; CHECK-NEXT:    [[FOR_NEXT]] = load i32, ptr [[GEP_B]], align 4
16; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[FOR]], [[L_A]]
17; CHECK-NEXT:    [[RED_NEXT]] = add i32 [[RED]], [[ADD]]
18; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
19; CHECK-NEXT:    [[EC:%.*]] = icmp eq i32 [[IV_NEXT]], 1000
20; CHECK-NEXT:    br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP]]
21; CHECK:       exit:
22; CHECK-NEXT:    [[RED_NEXT_LCSSA:%.*]] = phi i32 [ [[RED_NEXT]], [[LOOP]] ]
23; CHECK-NEXT:    ret i32 [[RED_NEXT_LCSSA]]
24;
25entry:
26  br label %loop
27
28loop:
29  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
30  %for = phi i32 [ 0, %entry ], [ %for.next, %loop ]
31  %red = phi i32 [ 0, %entry ], [ %red.next, %loop ]
32  %gep.a = getelementptr inbounds i32, ptr %A, i32 %for
33  %l.a = load i32, ptr %gep.a
34  %gep.b = getelementptr inbounds i32, ptr %B, i32 %iv
35  %for.next = load i32, ptr %gep.b
36  %add = add i32 %for, %l.a
37  %red.next = add i32 %red, %add
38  %iv.next = add nuw nsw i32 %iv, 1
39  %ec = icmp eq i32 %iv.next, 1000
40  br i1 %ec, label %exit, label %loop
41
42exit:
43  ret i32 %red.next
44}
45
46define void @cannot_sink_load_past_mayalias_store(ptr noalias %A, ptr noalias %B) {
47; CHECK-LABEL: @cannot_sink_load_past_mayalias_store(
48; CHECK-NEXT:  entry:
49; CHECK-NEXT:    br label [[LOOP:%.*]]
50; CHECK:       loop:
51; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
52; CHECK-NEXT:    [[FOR:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[FOR_NEXT:%.*]], [[LOOP]] ]
53; CHECK-NEXT:    [[GEP_A:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[FOR]]
54; CHECK-NEXT:    [[L_A:%.*]] = load i32, ptr [[GEP_A]], align 4
55; CHECK-NEXT:    [[GEP_A_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[IV]]
56; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[FOR]], [[L_A]]
57; CHECK-NEXT:    store i32 [[ADD]], ptr [[GEP_A_2]], align 4
58; CHECK-NEXT:    [[FOR_NEXT]] = load i32, ptr [[GEP_A_2]], align 4
59; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
60; CHECK-NEXT:    [[EC:%.*]] = icmp eq i32 [[IV_NEXT]], 1000
61; CHECK-NEXT:    br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP]]
62; CHECK:       exit:
63; CHECK-NEXT:    ret void
64;
65entry:
66  br label %loop
67
68loop:
69  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
70  %for = phi i32 [ 0, %entry ], [ %for.next, %loop ]
71  %gep.a = getelementptr inbounds i32, ptr %A, i32 %for
72  %l.a = load i32, ptr %gep.a
73  %gep.a.2 = getelementptr inbounds i32, ptr %A, i32 %iv
74  %add = add i32 %for, %l.a
75  store i32 %add, ptr %gep.a.2
76  %for.next = load i32, ptr %gep.a.2
77  %iv.next = add nuw nsw i32 %iv, 1
78  %ec = icmp eq i32 %iv.next, 1000
79  br i1 %ec, label %exit, label %loop
80
81exit:
82  ret void
83}
84