1; RUN: opt -basic-aa -loop-idiom < %s -S | FileCheck %s
2
3target datalayout = "e-p:32:32:32-p1:64:64:64-p2:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:32-n8:16:32:64"
4target triple = "x86_64-apple-darwin10.0.0"
5
6; Two dimensional nested loop should be promoted to one big memset.
7define void @test10(i8 addrspace(2)* %X) nounwind ssp {
8; CHECK-LABEL: @test10(
9; CHECK: entry:
10; CHECK-NEXT: call void @llvm.memset.p2i8.i16(i8 addrspace(2)* align 1 %X, i8 0, i16 10000, i1 false)
11; CHECK-NOT: store
12; CHECK: ret void
13
14entry:
15  br label %bb.nph
16
17bb.nph:                                           ; preds = %entry, %for.inc10
18  %i.04 = phi i16 [ 0, %entry ], [ %inc12, %for.inc10 ]
19  br label %for.body5
20
21for.body5:                                        ; preds = %for.body5, %bb.nph
22  %j.02 = phi i16 [ 0, %bb.nph ], [ %inc, %for.body5 ]
23  %mul = mul nsw i16 %i.04, 100
24  %add = add nsw i16 %j.02, %mul
25  %arrayidx = getelementptr inbounds i8, i8 addrspace(2)* %X, i16 %add
26  store i8 0, i8 addrspace(2)* %arrayidx, align 1
27  %inc = add nsw i16 %j.02, 1
28  %cmp4 = icmp eq i16 %inc, 100
29  br i1 %cmp4, label %for.inc10, label %for.body5
30
31for.inc10:                                        ; preds = %for.body5
32  %inc12 = add nsw i16 %i.04, 1
33  %cmp = icmp eq i16 %inc12, 100
34  br i1 %cmp, label %for.end13, label %bb.nph
35
36for.end13:                                        ; preds = %for.inc10
37  ret void
38}
39
40define void @test11_pattern(i32 addrspace(2)* nocapture %P) nounwind ssp {
41; CHECK-LABEL: @test11_pattern(
42; CHECK-NOT: memset_pattern
43entry:
44  br label %for.body
45
46for.body:                                         ; preds = %entry, %for.body
47  %indvar = phi i64 [ 0, %entry ], [ %indvar.next, %for.body ]
48  %arrayidx = getelementptr i32, i32 addrspace(2)* %P, i64 %indvar
49  store i32 1, i32 addrspace(2)* %arrayidx, align 4
50  %indvar.next = add i64 %indvar, 1
51  %exitcond = icmp eq i64 %indvar.next, 10000
52  br i1 %exitcond, label %for.end, label %for.body
53
54for.end:                                          ; preds = %for.body
55  ret void
56}
57
58; PR9815 - This is a partial overlap case that cannot be safely transformed
59; into a memcpy.
60@g_50 = addrspace(2) global [7 x i32] [i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 0], align 16
61
62
63define i32 @test14() nounwind {
64; CHECK-LABEL: @test14(
65; CHECK: for.body:
66; CHECK: load i32
67; CHECK: store i32
68; CHECK: br i1 %cmp
69
70entry:
71  br label %for.body
72
73for.body:                                         ; preds = %for.inc, %for.body.lr.ph
74  %tmp5 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
75  %add = add nsw i32 %tmp5, 4
76  %idxprom = sext i32 %add to i64
77  %arrayidx = getelementptr inbounds [7 x i32], [7 x i32] addrspace(2)* @g_50, i32 0, i64 %idxprom
78  %tmp2 = load i32, i32 addrspace(2)* %arrayidx, align 4
79  %add4 = add nsw i32 %tmp5, 5
80  %idxprom5 = sext i32 %add4 to i64
81  %arrayidx6 = getelementptr inbounds [7 x i32], [7 x i32] addrspace(2)* @g_50, i32 0, i64 %idxprom5
82  store i32 %tmp2, i32 addrspace(2)* %arrayidx6, align 4
83  %inc = add nsw i32 %tmp5, 1
84  %cmp = icmp slt i32 %inc, 2
85  br i1 %cmp, label %for.body, label %for.end
86
87for.end:                                          ; preds = %for.inc
88  %tmp8 = load i32, i32 addrspace(2)* getelementptr inbounds ([7 x i32], [7 x i32] addrspace(2)* @g_50, i32 0, i64 6), align 4
89  ret i32 %tmp8
90}
91
92