1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S %s | FileCheck %s
3
4
5; Test case where %gep has multiple uses of %iv.
6define void @multiple_iv_uses_in_same_instruction([100 x [100 x i32]]* %ptr) {
7; CHECK-LABEL: @multiple_iv_uses_in_same_instruction(
8; CHECK-NEXT:  entry:
9; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
10; CHECK:       vector.ph:
11; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
12; CHECK:       vector.body:
13; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
14; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
15; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
16; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
17; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* [[PTR:%.*]], i64 0, i64 [[TMP0]], i64 [[TMP0]]
18; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* [[PTR]], i64 0, i64 [[TMP1]], i64 [[TMP1]]
19; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 0
20; CHECK-NEXT:    store i32 [[TMP4]], i32* [[TMP2]], align 4
21; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 1
22; CHECK-NEXT:    store i32 [[TMP5]], i32* [[TMP3]], align 4
23; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
24; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
25; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
26; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
27; CHECK:       middle.block:
28; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 100, 100
29; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
30; CHECK:       scalar.ph:
31; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 100, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
32; CHECK-NEXT:    br label [[LOOP:%.*]]
33; CHECK:       loop:
34; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
35; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* [[PTR]], i64 0, i64 [[IV]], i64 [[IV]]
36; CHECK-NEXT:    [[T:%.*]] = trunc i64 [[IV]] to i32
37; CHECK-NEXT:    store i32 [[T]], i32* [[GEP]], align 4
38; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
39; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], 100
40; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP2:![0-9]+]]
41; CHECK:       exit:
42; CHECK-NEXT:    ret void
43;
44entry:
45  br label %loop
46
47loop:
48  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
49  %gep = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* %ptr, i64 0, i64 %iv, i64 %iv
50  %t = trunc i64 %iv to i32
51  store i32 %t, i32* %gep, align 4
52  %iv.next = add nuw nsw i64 %iv, 1
53  %exitcond = icmp eq i64 %iv.next, 100
54  br i1 %exitcond, label %exit, label %loop
55
56exit:
57  ret void
58}
59