xref: /llvm-project/llvm/test/Transforms/LoopRotate/pr51981-scev-problem.ll (revision 6afcc54ac7d68fa2b28f0e7cbf9dc1d4ac7fb95e)
1; RUN: opt < %s -passes='print<scalar-evolution>,loop(loop-rotate),invalidate<scalar-evolution>,print<scalar-evolution>' -disable-output 2>&1 | FileCheck -check-prefixes CHECK-SCEV %s
2; RUN: opt < %s -passes='print<scalar-evolution>,loop(loop-rotate),print<scalar-evolution>' -disable-output 2>&1 | FileCheck -check-prefixes CHECK-SCEV %s
3; RUN: opt < %s -passes='loop(canon-freeze),loop(loop-rotate),print<scalar-evolution>' -disable-output
4
5; In the first two RUN lines print<scalar-evolution> is used to populate the
6; analysis cache before loop-rotate. That was enough to see the problem by
7; examining print<scalar-evolution> printouts after loop-rotate. However, the
8; crashes where only observed when using canon-freeze as a trigger to populate
9; the analysis cache, so that is why canon-freeze is used in the third RUN
10; line.
11
12; Verify that we get the same SCEV expressions after loop-rotate, regardless
13; if we invalidate scalar-evolution before the final printing or not.
14;
15; This used to fail as described by PR51981 (some expressions still referred
16; to (trunc i32 %div210 to i16) but after the rotation it should be (trunc i32
17; %div2102 to i16).
18;
19; CHECK-SCEV: Classifying expressions for: @test_function
20; CHECK-SCEV:   %wide = load i32, ptr @offset, align 1
21; CHECK-SCEV:   -->  %wide U: full-set S: full-set          Exits: <<Unknown>>              LoopDispositions: { %loop.outer.header: Variant, %loop.inner: Invariant }
22; CHECK-SCEV:   %narrow = trunc i32 %wide to i16
23; CHECK-SCEV:   -->  (trunc i32 %wide to i16) U: full-set S: full-set               Exits: <<Unknown>>              LoopDispositions: { %loop.outer.header: Variant, %loop.inner: Invariant }
24; CHECK-SCEV:   %iv = phi i16 [ %narrow, %loop.inner.ph ], [ %iv.plus, %loop.inner ]
25; CHECK-SCEV:   -->  {(trunc i32 %wide to i16),+,1}<nw><%loop.inner> U: full-set S: full-set           Exits: (-1 + (700 umax (1 + (trunc i32 %wide to i16))))               LoopDispositions: { %loop.inner: Computable, %loop.outer.header: Variant }
26;
27; CHECK-SCEV: Classifying expressions for: @test_function
28; CHECK-SCEV:   %wide1 = load i32, ptr @offset, align 1
29; CHECK-SCEV:   -->  %wide1 U: full-set S: full-set
30; CHECK-SCEV:   %wide2 = phi i32 [ %wide1, %loop.inner.ph.lr.ph ], [ %wide, %loop.outer.latch ]
31; CHECK-SCEV:   -->  %wide2 U: full-set S: full-set         Exits: <<Unknown>>              LoopDispositions: { %loop.inner.ph: Variant, %loop.inner: Invariant }
32; CHECK-SCEV:   %narrow = trunc i32 %wide2 to i16
33; CHECK-SCEV:   -->  (trunc i32 %wide2 to i16) U: full-set S: full-set               Exits: <<Unknown>>              LoopDispositions: { %loop.inner.ph: Variant, %loop.inner: Invariant }
34; CHECK-SCEV:   %iv = phi i16 [ %narrow, %loop.inner.ph ], [ %iv.plus, %loop.inner ]
35; CHECK-SCEV:   -->  {(trunc i32 %wide2 to i16),+,1}<nw><%loop.inner> U: full-set S: full-set           Exits: (-1 + (700 umax (1 + (trunc i32 %wide2 to i16))))               LoopDispositions: { %loop.inner: Computable, %loop.inner.ph: Variant }
36
37
38@offset = external dso_local global i32, align 1
39@array = internal global [11263 x i32] zeroinitializer, align 1
40
41define void @test_function(i1 %cond) {
42entry:
43  br label %loop.outer.header
44
45loop.outer.header:                                ; preds = %loop.outer.latch, %entry
46  %wide = load i32, ptr @offset, align 1
47  br i1 %cond, label %exit, label %loop.inner.ph
48
49loop.inner.ph:                                    ; preds = %loop.outer.header
50  %narrow = trunc i32 %wide to i16
51  br label %loop.inner
52
53loop.inner:                                       ; preds = %loop.inner, %loop.inner.ph
54  %iv = phi i16 [ %narrow, %loop.inner.ph ], [ %iv.plus, %loop.inner ]
55  %iv.promoted = zext i16 %iv to i32
56  %gep = getelementptr inbounds [11263 x i32], ptr @array, i32 0, i32 %iv.promoted
57  store i32 7, ptr %gep, align 1
58  %iv.plus = add i16 %iv, 1
59  %cmp = icmp ult i16 %iv.plus, 700
60  br i1 %cmp, label %loop.inner, label %loop.outer.latch
61
62loop.outer.latch:                                 ; preds = %loop.inner
63  br label %loop.outer.header
64
65exit:                                             ; preds = %loop.outer.header
66  ret void
67}
68
69