• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt -loop-accesses -analyze -enable-new-pm=0 -S < %s | FileCheck %s
2; RUN: opt -passes='require<scalar-evolution>,require<aa>,loop(print-access-info)' -disable-output  < %s 2>&1 | FileCheck %s
3
4; This is the test case from PR26314.
5; When we were retrying dependence checking with memchecks only,
6; the loop-invariant access in the inner loop was incorrectly determined to be wrapping
7; because it was not strided in the inner loop.
8
9; #define Z 32
10; typedef struct s {
11;	int v1[Z];
12;	int v2[Z];
13;	int v3[Z][Z];
14; } s;
15;
16; void slow_function (s* const obj, int z) {
17;    for (int j=0; j<Z; j++) {
18;        for (int k=0; k<z; k++) {
19;            int x = obj->v1[k] + obj->v2[j];
20;            obj->v3[j][k] += x;
21;        }
22;    }
23; }
24
25; CHECK: function 'Test':
26; CHECK:   .inner:
27; CHECK-NEXT:     Memory dependences are safe
28; CHECK-NEXT:     Dependences:
29; CHECK-NEXT:     Run-time memory checks:
30; CHECK:          Check 0:
31; CHECK:          Check 1:
32
33target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
34target triple = "x86_64-unknown-linux-gnu"
35
36%struct.s = type { [32 x i32], [32 x i32], [32 x [32 x i32]] }
37
38define void @Test(%struct.s* nocapture %obj, i64 %z) #0 {
39  br label %.outer.preheader
40
41
42.outer.preheader:
43  %i = phi i64 [ 0, %0 ], [ %i.next, %.outer ]
44  %1 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 1, i64 %i
45  br label %.inner
46
47.exit:
48  ret void
49
50.outer:
51  %i.next = add nuw nsw i64 %i, 1
52  %exitcond.outer = icmp eq i64 %i.next, 32
53  br i1 %exitcond.outer, label %.exit, label %.outer.preheader
54
55.inner:
56  %j = phi i64 [ 0, %.outer.preheader ], [ %j.next, %.inner ]
57  %2 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 0, i64 %j
58  %3 = load i32, i32* %2
59  %4 = load i32, i32* %1
60  %5 = add nsw i32 %4, %3
61  %6 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 2, i64 %i, i64 %j
62  %7 = load i32, i32* %6
63  %8 = add nsw i32 %5, %7
64  store i32 %8, i32* %6
65  %j.next = add nuw nsw i64 %j, 1
66  %exitcond.inner = icmp eq i64 %j.next, %z
67  br i1 %exitcond.inner, label %.outer, label %.inner
68}
69