• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt %loadPolly -analyze -polly-scops \
2; RUN: -polly-invariant-load-hoisting=true < %s | FileCheck %s
3; RUN: opt %loadPolly -S -polly-codegen -polly-overflow-tracking=never \
4; RUN: -polly-invariant-load-hoisting=true < %s | FileCheck %s --check-prefix=IR
5;
6; As (p + q) can overflow we have to check that we load from
7; I[p + q] only if it does not.
8;
9; CHECK:         Invariant Accesses: {
10; CHECK-NEXT:            ReadAccess :=	[Reduction Type: NONE] [Scalar: 0]
11; CHECK-NEXT:                [N, p, q] -> { Stmt_for_body[i0] -> MemRef_I[p + q] };
12; CHECK-NEXT:            Execution Context: [N, p, q] -> {  : N > 0 and -2147483648 - p <= q <= 2147483647 - p }
13; CHECK-NEXT:            ReadAccess :=	[Reduction Type: NONE] [Scalar: 0]
14; CHECK-NEXT:                [N, p, q] -> { Stmt_for_body[i0] -> MemRef_tmp1[0] };
15; CHECK-NEXT:            Execution Context: [N, p, q] -> {  : N > 0 }
16; CHECK-NEXT:    }
17;
18; IR:      polly.preload.merge:
19; IR-NEXT:   %polly.preload.tmp1.merge = phi i32* [ %polly.access.I.load, %polly.preload.exec ], [ null, %polly.preload.cond ]
20; IR-NEXT:   store i32* %polly.preload.tmp1.merge, i32** %tmp1.preload.s2a
21; IR-NEXT:   %12 = sext i32 %N to i64
22; IR-NEXT:   %13 = icmp sge i64 %12, 1
23; IR-NEXT:   %14 = sext i32 %q to i64
24; IR-NEXT:   %15 = sext i32 %p to i64
25; IR-NEXT:   %16 = add nsw i64 %15, %14
26; IR-NEXT:   %17 = icmp sle i64 %16, 2147483647
27; IR-NEXT:   %18 = and i1 %13, %17
28; IR-NEXT:   %19 = sext i32 %q to i64
29; IR-NEXT:   %20 = sext i32 %p to i64
30; IR-NEXT:   %21 = add nsw i64 %20, %19
31; IR-NEXT:   %22 = icmp sge i64 %21, -2147483648
32; IR-NEXT:   %23 = and i1 %18, %22
33; IR-NEXT:   br label %polly.preload.cond1
34;
35; IR:      polly.preload.cond1:
36; IR-NEXT:   br i1 %23
37;
38; IR:      polly.preload.exec3:
39; IR-NEXT:   %polly.access.polly.preload.tmp1.merge = getelementptr i32, i32* %polly.preload.tmp1.merge, i64 0
40; IR-NEXT:   %polly.access.polly.preload.tmp1.merge.load = load i32, i32* %polly.access.polly.preload.tmp1.merge, align 4
41;
42;    void f(int **I, int *A, int N, int p, int q) {
43;      for (int i = 0; i < N; i++)
44;        A[i] = *(I[p + q]);
45;    }
46;
47target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
48
49define void @f(i32** %I, i32* %A, i32 %N, i32 %p, i32 %q) {
50entry:
51  %tmp = sext i32 %N to i64
52  br label %for.cond
53
54for.cond:                                         ; preds = %for.inc, %entry
55  %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ]
56  %cmp = icmp slt i64 %indvars.iv, %tmp
57  br i1 %cmp, label %for.body, label %for.end
58
59for.body:                                         ; preds = %for.cond
60  %add = add i32 %p, %q
61  %idxprom = sext i32 %add to i64
62  %arrayidx = getelementptr inbounds i32*, i32** %I, i64 %idxprom
63  %tmp1 = load i32*, i32** %arrayidx, align 8
64  %tmp2 = load i32, i32* %tmp1, align 4
65  %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
66  store i32 %tmp2, i32* %arrayidx2, align 4
67  br label %for.inc
68
69for.inc:                                          ; preds = %for.body
70  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
71  br label %for.cond
72
73for.end:                                          ; preds = %for.cond
74  ret void
75}
76