• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt %loadPolly -polly-scops -polly-invariant-load-hoisting=true -analyze < %s | FileCheck %s
2;
3; Check that we do not consolidate the invariant loads to smp[order - 1] and
4; smp[order - 2] in the blocks %0 and %16. While they have the same pointer
5; operand (SCEV) they do not have the same access relation due to the
6; instantiation of "order" from their domain.
7;
8; CHECK:         Invariant Accesses: {
9; CHECK-NEXT:            ReadAccess :=	[Reduction Type: NONE] [Scalar: 0]
10; CHECK-NEXT:                [order, n] -> { Stmt_bb1[] -> MemRef_smp[1] };
11; CHECK-NEXT:            Execution Context: [order, n] -> {  : order = 2 }
12; CHECK-NEXT:            ReadAccess :=	[Reduction Type: NONE] [Scalar: 0]
13; CHECK-NEXT:                [order, n] -> { Stmt_bb1[] -> MemRef_smp[0] };
14; CHECK-NEXT:            Execution Context: [order, n] -> {  : order = 2 }
15; CHECK-NEXT:            ReadAccess :=	[Reduction Type: NONE] [Scalar: 0]
16; CHECK-NEXT:                [order, n] -> { Stmt_bb16[] -> MemRef_smp[2] };
17; CHECK-NEXT:            Execution Context: [order, n] -> {  : order = 3 }
18; CHECK-NEXT:            ReadAccess :=	[Reduction Type: NONE] [Scalar: 0]
19; CHECK-NEXT:                [order, n] -> { Stmt_bb16[] -> MemRef_smp[1] };
20; CHECK-NEXT:            Execution Context: [order, n] -> {  : order = 3 }
21; CHECK-NEXT:    }
22;
23; ModuleID = '/home/johannes/Downloads/test_case.ll'
24target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
25
26; Function Attrs: nounwind uwtable
27define void @encode_residual_fixed(i32* %res, i32* %smp, i32 %n, i32 %order) {
28bb:
29  br label %.split
30
31.split:                                           ; preds = %bb
32  switch i32 %order, label %bb32 [
33    i32 2, label %bb1
34    i32 3, label %bb16
35  ]
36
37bb1:                                              ; preds = %.split
38  %tmp = add nsw i32 %order, -1
39  %tmp2 = sext i32 %tmp to i64
40  %tmp3 = getelementptr inbounds i32, i32* %smp, i64 %tmp2
41  %tmp4 = load i32, i32* %tmp3, align 4
42  %tmp5 = add nsw i32 %order, -2
43  %tmp6 = sext i32 %tmp5 to i64
44  %tmp7 = getelementptr inbounds i32, i32* %smp, i64 %tmp6
45  %tmp8 = load i32, i32* %tmp7, align 4
46  %tmp9 = sub nsw i32 %tmp4, %tmp8
47  %tmp10 = icmp slt i32 %order, %n
48  br i1 %tmp10, label %.lr.ph, label %.loopexit
49
50.lr.ph:                                           ; preds = %bb1
51  %tmp11 = sext i32 %order to i64
52  br label %bb12
53
54bb12:                                             ; preds = %bb12, %.lr.ph
55  %indvars.iv = phi i64 [ %tmp11, %.lr.ph ], [ %indvars.iv.next, %bb12 ]
56  %i.03 = phi i32 [ %order, %.lr.ph ], [ %tmp14, %bb12 ]
57  %tmp13 = getelementptr inbounds i32, i32* %res, i64 %indvars.iv
58  store i32 %tmp9, i32* %tmp13, align 4
59  %tmp14 = add nsw i32 %i.03, 2
60  %tmp15 = icmp slt i32 %tmp14, %n
61  %indvars.iv.next = add nsw i64 %indvars.iv, 2
62  br i1 %tmp15, label %bb12, label %..loopexit_crit_edge
63
64bb16:                                             ; preds = %.split
65  %tmp17 = add nsw i32 %order, -1
66  %tmp18 = sext i32 %tmp17 to i64
67  %tmp19 = getelementptr inbounds i32, i32* %smp, i64 %tmp18
68  %tmp20 = load i32, i32* %tmp19, align 4
69  %tmp21 = add nsw i32 %order, -2
70  %tmp22 = sext i32 %tmp21 to i64
71  %tmp23 = getelementptr inbounds i32, i32* %smp, i64 %tmp22
72  %tmp24 = load i32, i32* %tmp23, align 4
73  %tmp25 = sub nsw i32 %tmp20, %tmp24
74  %tmp26 = icmp slt i32 %order, %n
75  br i1 %tmp26, label %.lr.ph5, label %.loopexit2
76
77.lr.ph5:                                          ; preds = %bb16
78  %tmp27 = sext i32 %order to i64
79  br label %bb28
80
81bb28:                                             ; preds = %bb28, %.lr.ph5
82  %indvars.iv6 = phi i64 [ %tmp27, %.lr.ph5 ], [ %indvars.iv.next7, %bb28 ]
83  %i.14 = phi i32 [ %order, %.lr.ph5 ], [ %tmp30, %bb28 ]
84  %tmp29 = getelementptr inbounds i32, i32* %res, i64 %indvars.iv6
85  store i32 %tmp25, i32* %tmp29, align 4
86  %tmp30 = add nsw i32 %i.14, 2
87  %tmp31 = icmp slt i32 %tmp30, %n
88  %indvars.iv.next7 = add nsw i64 %indvars.iv6, 2
89  br i1 %tmp31, label %bb28, label %..loopexit2_crit_edge
90
91..loopexit_crit_edge:                             ; preds = %bb12
92  br label %.loopexit
93
94.loopexit:                                        ; preds = %..loopexit_crit_edge, %bb1
95  br label %bb32
96
97..loopexit2_crit_edge:                            ; preds = %bb28
98  br label %.loopexit2
99
100.loopexit2:                                       ; preds = %..loopexit2_crit_edge, %bb16
101  br label %bb32
102
103bb32:                                             ; preds = %.loopexit2, %.loopexit, %.split
104  %tmp33 = getelementptr inbounds i32, i32* %res, i64 2
105  %tmp34 = load i32, i32* %tmp33, align 4
106  %tmp35 = icmp eq i32 %tmp34, 5
107  br i1 %tmp35, label %bb37, label %bb36
108
109bb36:                                             ; preds = %bb32
110  ret void
111
112bb37:                                             ; preds = %bb32
113  ret void
114}
115