1; RUN: opt -mtriple=thumbv7-unknown-linux-android -arm-parallel-dsp -S %s -o - | FileCheck %s
2
3; CHECK-LABEL: undef_no_return
4; CHECK: [[CAST_A:%[^ ]+]] = bitcast i16* %incdec.ptr21 to i32*
5; CHECK: [[LOAD_A:%[^ ]+]] = load i32, i32* [[CAST_A]], align 2
6; CHECK: %uglygep15 = getelementptr i8, i8* undef, i32 undef
7; CHECK: [[GEP8:%[^ ]+]] = getelementptr i8, i8* undef, i32 undef
8; CHECK: [[CAST_GEP8:%[^ ]+]] = bitcast i8* [[GEP8]] to i16*
9; CHECK: [[GEP16:%[^ ]+]] = getelementptr i16, i16* [[CAST_GEP8]], i32 6
10; CHECK: [[CAST_GEP16:%[^ ]+]] = bitcast i16* [[GEP16]] to i32*
11; CHECK: [[LOAD_UNDEF:%[^ ]+]] = load i32, i32* [[CAST_GEP16]], align 2
12; CHECK: call i32 @llvm.arm.smladx(i32 [[LOAD_UNDEF]], i32 [[LOAD_A]], i32 undef)
13define void @undef_no_return(i16* %a) {
14entry:
15  %incdec.ptr21 = getelementptr inbounds i16, i16* %a, i32 3
16  %incdec.ptr29 = getelementptr inbounds i16, i16* %a, i32 4
17  br label %for.body
18
19for.body:
20  %0 = load i16, i16* %incdec.ptr21, align 2
21  %conv25 = sext i16 %0 to i32
22  %uglygep15 = getelementptr i8, i8* undef, i32 undef
23  %uglygep1516 = bitcast i8* %uglygep15 to i16*
24  %scevgep17 = getelementptr i16, i16* %uglygep1516, i32 7
25  %1 = load i16, i16* %scevgep17, align 2
26  %conv31 = sext i16 %1 to i32
27  %2 = load i16, i16* %incdec.ptr29, align 2
28  %conv33 = sext i16 %2 to i32
29  %uglygep12 = getelementptr i8, i8* undef, i32 undef
30  %uglygep1213 = bitcast i8* %uglygep12 to i16*
31  %scevgep14 = getelementptr i16, i16* %uglygep1213, i32 6
32  %3 = load i16, i16* %scevgep14, align 2
33  %conv39 = sext i16 %3 to i32
34  %mul.i287.neg.neg = mul nsw i32 %conv31, %conv25
35  %mul.i283.neg.neg = mul nsw i32 %conv39, %conv33
36  %reass.add408 = add i32 undef, %mul.i287.neg.neg
37  %reass.add409 = add i32 %reass.add408, %mul.i283.neg.neg
38  br label %for.body
39}
40
41; CHECK-LABEL: return
42; CHECK: phi i32 [ %N, %entry ]
43; CHECK: [[ACC:%[^ ]+]] = phi i32 [ 0, %entry ], [ [[ACC_NEXT:%[^ ]+]], %for.body ]
44; CHECK: [[CAST_A:%[^ ]+]] = bitcast i16* %incdec.ptr21 to i32*
45; CHECK: [[LOAD_A:%[^ ]+]] = load i32, i32* [[CAST_A]], align 2
46; CHECK: [[GEP8:%[^ ]+]] = getelementptr i8, i8* %b, i32 0
47; CHECK: [[CAST_GEP8:%[^ ]+]] = bitcast i8* [[GEP8]] to i16*
48; CHECK: [[GEP16:%[^ ]+]] = getelementptr i16, i16* [[CAST_GEP8]], i32 %iv
49; CHECK: [[CAST_GEP16:%[^ ]+]] = bitcast i16* [[GEP16]] to i32*
50; CHECK: [[LOAD_B:%[^ ]+]] = load i32, i32* [[CAST_GEP16]], align 2
51; CHECK: [[ACC_NEXT]] = call i32 @llvm.arm.smladx(i32 [[LOAD_B]], i32 [[LOAD_A]], i32 [[ACC]])
52define i32 @return(i16* %a, i8* %b, i32 %N) {
53entry:
54  %incdec.ptr21 = getelementptr inbounds i16, i16* %a, i32 3
55  %incdec.ptr29 = getelementptr inbounds i16, i16* %a, i32 4
56  br label %for.body
57
58for.body:
59  %iv = phi i32 [ %N, %entry ], [ %iv.next, %for.body ]
60  %acc = phi i32 [ 0, %entry ], [ %reass.add409, %for.body ]
61  %0 = load i16, i16* %incdec.ptr21, align 2
62  %conv25 = sext i16 %0 to i32
63  %uglygep15 = getelementptr i8, i8* %b, i32 0
64  %uglygep1516 = bitcast i8* %uglygep15 to i16*
65  %b.idx = add nuw nsw i32 %iv, 1
66  %scevgep17 = getelementptr i16, i16* %uglygep1516, i32 %b.idx
67  %scevgep14 = getelementptr i16, i16* %uglygep1516, i32 %iv
68  %1 = load i16, i16* %scevgep17, align 2
69  %conv31 = sext i16 %1 to i32
70  %2 = load i16, i16* %incdec.ptr29, align 2
71  %conv33 = sext i16 %2 to i32
72  %3 = load i16, i16* %scevgep14, align 2
73  %conv39 = sext i16 %3 to i32
74  %mul.i287.neg.neg = mul nsw i32 %conv31, %conv25
75  %mul.i283.neg.neg = mul nsw i32 %conv39, %conv33
76  %reass.add408 = add i32 %acc, %mul.i287.neg.neg
77  %reass.add409 = add i32 %reass.add408, %mul.i283.neg.neg
78  %iv.next = add nuw nsw i32 %iv, -1
79  %cmp = icmp ne i32 %iv.next, 0
80  br i1 %cmp, label %for.body, label %exit
81
82exit:
83  ret i32 %reass.add409
84}
85