llvm-project

1
; RUN: opt %loadNPMPolly '-passes=polly-import-jscop,polly-codegen' \
2
; RUN: -polly-import-jscop-postfix=transformed \
3
; RUN: -verify-dom-info -polly-allow-nonaffine-loops \
4
; RUN: -S < %s | FileCheck %s
5

6
; This test verifies that partial writes within non-affine loops are code
7
; generated correctly.
8

9
; CHECK:polly.stmt.bb3:
10
; CHECK-NEXT:  %polly.subregion.iv = phi i32 [ %polly.subregion.iv.inc, %polly.stmt.bb5.cont ], [ 0, %polly.stmt.bb3.entry ]
11
; CHECK-NEXT:  %polly.j.0 = phi i64 [ %j.0.phiops.reload, %polly.stmt.bb3.entry ], [ %p_tmp10, %polly.stmt.bb5.cont ]
12
; CHECK-NEXT:  %8 = zext nneg i64 %polly.indvar to i65
13
; CHECK-NEXT:  %9 = add nsw i64 %polly.indvar, -1
14
; CHECK-NEXT:  %10 = zext i64 %9 to i65
15
; CHECK-NEXT:  %11 = mul i65 %8, %10
16
; CHECK-NEXT:  %12 = lshr i65 %11, 1
17
; CHECK-NEXT:  %13 = trunc i65 %12 to i64
18
; CHECK-NEXT:  %14 = shl i64 %13, 1
19
; CHECK-NEXT:  %15 = add i64 %polly.indvar, %14
20
; CHECK-NEXT:  %p_tmp4 = icmp slt i64 %polly.j.0, %15
21
; CHECK-NEXT:  %polly.subregion.iv.inc = add i32 %polly.subregion.iv, 1
22
; CHECK-NEXT:  br i1 %p_tmp4, label %polly.stmt.bb5, label %polly.stmt.bb11.exit
23

24
; CHECK:polly.stmt.bb5:
25
; CHECK-NEXT:  %p_tmp6 = getelementptr inbounds float, ptr %B, i64 42
26
; CHECK-NEXT:  %tmp7_p_scalar_ = load float, ptr %p_tmp6
27
; CHECK-NEXT:  %p_tmp8 = fadd float %tmp7_p_scalar_, 1.000000e+00
28
; CHECK-NEXT:  %16 = icmp sle i64 %polly.indvar, 9
29
; CHECK-NEXT:  %polly.Stmt_bb3__TO__bb11_MayWrite2.cond = icmp ne i1 %16, false
30
; CHECK-NEXT:  br i1 %polly.Stmt_bb3__TO__bb11_MayWrite2.cond, label %polly.stmt.bb5.Stmt_bb3__TO__bb11_MayWrite2.partial, label %polly.stmt.bb5.cont
31

32
; CHECK:polly.stmt.bb5.Stmt_bb3__TO__bb11_MayWrite2.partial: ; preds = %polly.stmt.bb5
33
; CHECK-NEXT:  %polly.access.B3 = getelementptr float, ptr %B, i64 42
34
; CHECK-NEXT:  store float %p_tmp8, ptr %polly.access.B3
35
; CHECK-NEXT:  br label %polly.stmt.bb5.cont
36

37
; CHECK:polly.stmt.bb5.cont:
38
; CHECK-NEXT:  %p_tmp10 = add nuw nsw i64 %polly.j.0, 1
39
; CHECK-NEXT:  br label %polly.stmt.bb3
40

41

42

43
;    void foo(long A[], float B[], long *x) {
44
;      for (long i = 0; i < 1024; i++)
45
;        for (long j = *x; j < i * i; j++)
46
;          B[42]++;
47
;    }
48
;
49
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
50

51
define void @partial_write_in_region_with_loop(ptr %A, ptr %B, ptr %xptr) {
52
bb:
53
  br label %bb1
54

55
bb1:                                              ; preds = %bb12, %bb
56
  %i.0 = phi i64 [ 0, %bb ], [ %tmp13, %bb12 ]
57
  %exitcond = icmp ne i64 %i.0, 1024
58
  br i1 %exitcond, label %bb2, label %bb14
59

60
bb2:                                              ; preds = %bb1
61
  %x = load i64, ptr %xptr
62
  br label %bb3
63

64
bb3:                                              ; preds = %bb9, %bb2
65
  %j.0 = phi i64 [ %x, %bb2 ], [ %tmp10, %bb5 ]
66
  %tmp = mul nsw i64 %i.0, %i.0
67
  %tmp4 = icmp slt i64 %j.0, %tmp
68
  br i1 %tmp4, label %bb5, label %bb11
69

70
bb5:                                              ; preds = %bb3
71
  %tmp6 = getelementptr inbounds float, ptr %B, i64 42
72
  %tmp7 = load float, ptr %tmp6, align 4
73
  %tmp8 = fadd float %tmp7, 1.000000e+00
74
  store float %tmp8, ptr %tmp6, align 4
75
  %tmp10 = add nuw nsw i64 %j.0, 1
76
  br label %bb3
77

78
bb11:                                             ; preds = %bb3
79
  br label %bb12
80

81
bb12:                                             ; preds = %bb11
82
  %tmp13 = add nuw nsw i64 %i.0, 1
83
  br label %bb1
84

85
bb14:                                             ; preds = %bb1
86
  ret void
87
}
88