xref: /llvm-project/llvm/test/Transforms/LoopInterchange/interchange-flow-dep-outer.ll (revision 456ec1c2f4e487de235c953e8f2832b97372e7b0)
1; Remove 'S' Scalar Dependencies #119345
2; Scalar dependencies are not handled correctly, so they were removed to avoid
3; miscompiles. The loop nest in this test case used to be interchanged, but it's
4; no longer triggering. XFAIL'ing this test to indicate that this test should
5; interchanged if scalar deps are handled correctly.
6;
7; XFAIL: *
8
9; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -verify-dom-info -verify-loop-info -pass-remarks-output=%t -disable-output
10; RUN: FileCheck -input-file %t %s
11
12target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
13
14@A = common global [100 x [100 x i32]] zeroinitializer
15@B = common global [100 x i32] zeroinitializer
16@C = common global [100 x [100 x i32]] zeroinitializer
17@D = common global [100 x [100 x [100 x i32]]] zeroinitializer
18@T = internal global [100 x double] zeroinitializer, align 4
19@Arr = internal global [1000 x [1000 x i32]] zeroinitializer, align 4
20
21; Test that a flow dependency in outer loop doesn't prevent interchange in
22; loops i and j.
23;
24;  for (int k = 0; k < 100; ++k) {
25;    T[k] = fn1();
26;    for (int i = 0; i < 1000; ++i)
27;      for(int j = 1; j < 1000; ++j)
28;        Arr[j][i] = Arr[j][i]+k;
29;    fn2(T[k]);
30;  }
31;
32; So, loops InnerLoopId = 2 and OuterLoopId = 1 should be interchanged,
33; but not InnerLoopId = 1 and OuterLoopId = 0.
34;
35; CHECK:       --- !Passed
36; CHECK-NEXT:  Pass:            loop-interchange
37; CHECK-NEXT:  Name:            Interchanged
38; CHECK-NEXT:  Function:        interchange_09
39; CHECK-NEXT:  Args:
40; CHECK-NEXT:    - String:          Loop interchanged with enclosing loop.
41; CHECK-NEXT:  ...
42; CHECK-NEXT:  --- !Missed
43; CHECK-NEXT:  Pass:            loop-interchange
44; CHECK-NEXT:  Name:            NotTightlyNested
45; CHECK-NEXT:  Function:        interchange_09
46; CHECK-NEXT:  Args:
47; CHECK-NEXT:    - String:          Cannot interchange loops because they are not tightly nested.
48; CHECK-NEXT:  ...
49; CHECK-NEXT:  --- !Missed
50; CHECK-NEXT:  Pass:            loop-interchange
51; CHECK-NEXT:  Name:            InterchangeNotProfitable
52; CHECK-NEXT:  Function:        interchange_09
53; CHECK-NEXT:  Args:
54; CHECK-NEXT:    - String:          Interchanging loops is not considered to improve cache locality nor vectorization.
55; CHECK-NEXT:  ...
56
57define void @interchange_09(i32 %k) {
58entry:
59  br label %for.body
60
61for.cond.cleanup:                                 ; preds = %for.cond.cleanup4
62  ret void
63
64for.body:                                         ; preds = %for.cond.cleanup4, %entry
65  %indvars.iv45 = phi i64 [ 0, %entry ], [ %indvars.iv.next46, %for.cond.cleanup4 ]
66  %call = call double @fn1()
67  %arrayidx = getelementptr inbounds [100 x double], ptr @T, i64 0, i64 %indvars.iv45
68  store double %call, ptr %arrayidx, align 8
69  br label %for.cond6.preheader
70
71for.cond6.preheader:                              ; preds = %for.cond.cleanup8, %for.body
72  %indvars.iv42 = phi i64 [ 0, %for.body ], [ %indvars.iv.next43, %for.cond.cleanup8 ]
73  br label %for.body9
74
75for.cond.cleanup4:                                ; preds = %for.cond.cleanup8
76  %tmp = load double, ptr %arrayidx, align 8
77  call void @fn2(double %tmp)
78  %indvars.iv.next46 = add nuw nsw i64 %indvars.iv45, 1
79  %exitcond47 = icmp ne i64 %indvars.iv.next46, 100
80  br i1 %exitcond47, label %for.body, label %for.cond.cleanup
81
82for.cond.cleanup8:                                ; preds = %for.body9
83  %indvars.iv.next43 = add nuw nsw i64 %indvars.iv42, 1
84  %exitcond44 = icmp ne i64 %indvars.iv.next43, 1000
85  br i1 %exitcond44, label %for.cond6.preheader, label %for.cond.cleanup4
86
87for.body9:                                        ; preds = %for.body9, %for.cond6.preheader
88  %indvars.iv = phi i64 [ 1, %for.cond6.preheader ], [ %indvars.iv.next, %for.body9 ]
89  %arrayidx13 = getelementptr inbounds [1000 x [1000 x i32]], ptr @Arr, i64 0, i64 %indvars.iv, i64 %indvars.iv42
90  %t1 = load i32, ptr %arrayidx13, align 4
91  %t2 = trunc i64 %indvars.iv45 to i32
92  %add = add nsw i32 %t1, %t2
93  store i32 %add, ptr %arrayidx13, align 4
94  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
95  %exitcond = icmp ne i64 %indvars.iv.next, 1000
96  br i1 %exitcond, label %for.body9, label %for.cond.cleanup8
97}
98
99declare double @fn1() readnone
100declare void @fn2(double) readnone
101