1; RUN: llc -march=hexagon -O2 -fp-contract=fast < %s -pipeliner-experimental-cg=true | FileCheck %s
2
3; Test that the memoperands for instructions in the epilog are updated
4; correctly. Previously, the pipeliner updated the offset for the memoperands
5; in the epilog. But, the value of the offset is incorrect when control flow
6; branches around the kernel.
7
8; In this test, we check that a load and store to the same location are not
9; swapped due to a bad offset in the memoperands. The store and load are both
10; to r29+32. If the memoperands are updated incorrectly, these are swapped.
11
12; CHECK: [[REG0:r([0-9]+)]] = add(r29,#24)
13; CHECK: memw([[REG0]]++#4) = r{{[0-9]+}}
14; CHECK: r{{[0-9]+}} = memw(r29+#{{[0-9]+}})
15
16%s.0 = type { %s.1 }
17%s.1 = type { %s.2 }
18%s.2 = type { %s.3 }
19%s.3 = type { [9 x float] }
20%s.4 = type { %s.5 }
21%s.5 = type { %s.6 }
22%s.6 = type { %s.7 }
23%s.7 = type { [3 x float] }
24
25@g0 = external hidden unnamed_addr constant [29 x i8], align 1
26
27define i32 @f0() unnamed_addr {
28b0:
29  %v0 = alloca %s.0, align 4
30  %v1 = alloca %s.4, align 4
31  %v2 = bitcast %s.0* %v0 to i8*
32  call void @llvm.lifetime.start.p0i8(i64 36, i8* %v2)
33  %v3 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0
34  store float 0x3FEFFF7160000000, float* %v3, align 4
35  %v4 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1
36  store float 0xBF87867F00000000, float* %v4, align 4
37  %v5 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 2
38  store float 0xBF6185CEE0000000, float* %v5, align 4
39  %v6 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 3
40  store float 0x3F8780BAA0000000, float* %v6, align 4
41  %v7 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 4
42  store float 0x3FEFFF5C60000000, float* %v7, align 4
43  %v8 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 5
44  store float 0xBF74717160000000, float* %v8, align 4
45  %v9 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 6
46  store float 0x3F61FF7160000000, float* %v9, align 4
47  %v10 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 7
48  store float 0x3F74573A80000000, float* %v10, align 4
49  %v11 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 8
50  store float 0x3FEFFFE080000000, float* %v11, align 4
51  %v12 = bitcast %s.4* %v1 to i8*
52  call void @llvm.lifetime.start.p0i8(i64 12, i8* %v12)
53  br label %b1
54
55b1:                                               ; preds = %b1, %b0
56  %v13 = phi i32 [ 0, %b0 ], [ %v29, %b1 ]
57  %v14 = mul nuw nsw i32 %v13, 3
58  %v15 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 %v14
59  %v16 = load float, float* %v15, align 4
60  %v17 = fmul float %v16, 0x3FE7B2B120000000
61  %v18 = add nuw nsw i32 %v14, 1
62  %v19 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 %v18
63  %v20 = load float, float* %v19, align 4
64  %v21 = fmul float %v20, 0x3FDA8BC9C0000000
65  %v22 = fsub float %v21, %v17
66  %v23 = add nuw nsw i32 %v14, 2
67  %v24 = getelementptr inbounds %s.0, %s.0* %v0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 %v23
68  %v25 = load float, float* %v24, align 4
69  %v26 = fmul float %v25, 0x40030D6700000000
70  %v27 = fadd float %v22, %v26
71  %v28 = getelementptr inbounds %s.4, %s.4* %v1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 %v13
72  store float %v27, float* %v28, align 4
73  %v29 = add nuw nsw i32 %v13, 1
74  %v30 = icmp eq i32 %v29, 3
75  br i1 %v30, label %b2, label %b1
76
77b2:                                               ; preds = %b1
78  %v31 = getelementptr inbounds %s.4, %s.4* %v1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0
79  %v32 = load float, float* %v31, align 4
80  %v33 = fpext float %v32 to double
81  %v34 = getelementptr inbounds %s.4, %s.4* %v1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1
82  %v35 = load float, float* %v34, align 4
83  %v36 = fpext float %v35 to double
84  %v37 = getelementptr inbounds %s.4, %s.4* %v1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 2
85  %v38 = load float, float* %v37, align 4
86  %v39 = fpext float %v38 to double
87  %v40 = tail call i32 (i8*, ...) @f1(i8* getelementptr inbounds ([29 x i8], [29 x i8]* @g0, i32 0, i32 0), double %v33, double %v36, double %v39)
88  call void @llvm.lifetime.end.p0i8(i64 12, i8* nonnull %v12)
89  call void @llvm.lifetime.end.p0i8(i64 36, i8* nonnull %v2)
90  ret i32 0
91}
92
93declare i32 @f1(i8* nocapture readonly, ...) local_unnamed_addr
94
95; Function Attrs: argmemonly nounwind
96declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #0
97
98; Function Attrs: argmemonly nounwind
99declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #0
100
101attributes #0 = { argmemonly nounwind }
102