1; RUN: llc -O0 -verify-machineinstrs -asm-verbose=0 -march=amdgcn < %s | FileCheck -check-prefix=SI -check-prefix=COMMON %s
2; RUN: llc -O0 -verify-machineinstrs -asm-verbose=0 -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=COMMON %s
3; XUN: llc -O0 -verify-machineinstrs -asm-verbose=0 -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=COMMON %s
4
5; SI hits an assertion at -O0, evergreen hits a not implemented unreachable.
6
7; COMMON-LABEL: {{^}}branch_true:
8define amdgpu_kernel void @branch_true(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
9entry:
10  br i1 true, label %for.end, label %for.body.lr.ph
11
12for.body.lr.ph:                                   ; preds = %entry
13  %add.ptr.sum = shl i32 %main_stride, 1
14  %add.ptr1.sum = add i32 %add.ptr.sum, %main_stride
15  %add.ptr4.sum = shl i32 %main_stride, 2
16  br label %for.body
17
18for.body:                                         ; preds = %for.body, %for.body.lr.ph
19  %main.addr.011 = phi i8 addrspace(1)* [ %main, %for.body.lr.ph ], [ %add.ptr6, %for.body ]
20  %0 = bitcast i8 addrspace(1)* %main.addr.011 to i32 addrspace(1)*
21  %1 = load i32, i32 addrspace(1)* %0, align 4
22  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %main_stride
23  %2 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
24  %3 = load i32, i32 addrspace(1)* %2, align 4
25  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr.sum
26  %4 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
27  %5 = load i32, i32 addrspace(1)* %4, align 4
28  %add.ptr2 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr1.sum
29  %6 = bitcast i8 addrspace(1)* %add.ptr2 to i32 addrspace(1)*
30  %7 = load i32, i32 addrspace(1)* %6, align 4
31  %add.ptr3 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr4.sum
32  %8 = bitcast i8 addrspace(1)* %add.ptr3 to i32 addrspace(1)*
33  %9 = load i32, i32 addrspace(1)* %8, align 4
34  %add.ptr6 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 undef
35  br i1 undef, label %for.end, label %for.body
36
37for.end:                                          ; preds = %for.body, %entry
38  ret void
39}
40
41; COMMON-LABEL: {{^}}branch_false:
42; SI: s_cbranch_scc1
43; SI: s_endpgm
44define amdgpu_kernel void @branch_false(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
45entry:
46  br i1 false, label %for.end, label %for.body.lr.ph
47
48for.body.lr.ph:                                   ; preds = %entry
49  %add.ptr.sum = shl i32 %main_stride, 1
50  %add.ptr1.sum = add i32 %add.ptr.sum, %main_stride
51  %add.ptr4.sum = shl i32 %main_stride, 2
52  br label %for.body
53
54for.body:                                         ; preds = %for.body, %for.body.lr.ph
55  %main.addr.011 = phi i8 addrspace(1)* [ %main, %for.body.lr.ph ], [ %add.ptr6, %for.body ]
56  %0 = bitcast i8 addrspace(1)* %main.addr.011 to i32 addrspace(1)*
57  %1 = load i32, i32 addrspace(1)* %0, align 4
58  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %main_stride
59  %2 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
60  %3 = load i32, i32 addrspace(1)* %2, align 4
61  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr.sum
62  %4 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
63  %5 = load i32, i32 addrspace(1)* %4, align 4
64  %add.ptr2 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr1.sum
65  %6 = bitcast i8 addrspace(1)* %add.ptr2 to i32 addrspace(1)*
66  %7 = load i32, i32 addrspace(1)* %6, align 4
67  %add.ptr3 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr4.sum
68  %8 = bitcast i8 addrspace(1)* %add.ptr3 to i32 addrspace(1)*
69  %9 = load i32, i32 addrspace(1)* %8, align 4
70  %add.ptr6 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 undef
71  br i1 undef, label %for.end, label %for.body
72
73for.end:                                          ; preds = %for.body, %entry
74  ret void
75}
76
77; COMMON-LABEL: {{^}}branch_undef:
78; SI: s_cbranch_scc1
79; SI: s_cbranch_scc1
80; SI: s_endpgm
81define amdgpu_kernel void @branch_undef(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
82entry:
83  br i1 undef, label %for.end, label %for.body.lr.ph
84
85for.body.lr.ph:                                   ; preds = %entry
86  %add.ptr.sum = shl i32 %main_stride, 1
87  %add.ptr1.sum = add i32 %add.ptr.sum, %main_stride
88  %add.ptr4.sum = shl i32 %main_stride, 2
89  br label %for.body
90
91for.body:                                         ; preds = %for.body, %for.body.lr.ph
92  %main.addr.011 = phi i8 addrspace(1)* [ %main, %for.body.lr.ph ], [ %add.ptr6, %for.body ]
93  %0 = bitcast i8 addrspace(1)* %main.addr.011 to i32 addrspace(1)*
94  %1 = load i32, i32 addrspace(1)* %0, align 4
95  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %main_stride
96  %2 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
97  %3 = load i32, i32 addrspace(1)* %2, align 4
98  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr.sum
99  %4 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
100  %5 = load i32, i32 addrspace(1)* %4, align 4
101  %add.ptr2 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr1.sum
102  %6 = bitcast i8 addrspace(1)* %add.ptr2 to i32 addrspace(1)*
103  %7 = load i32, i32 addrspace(1)* %6, align 4
104  %add.ptr3 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr4.sum
105  %8 = bitcast i8 addrspace(1)* %add.ptr3 to i32 addrspace(1)*
106  %9 = load i32, i32 addrspace(1)* %8, align 4
107  %add.ptr6 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 undef
108  br i1 undef, label %for.end, label %for.body
109
110for.end:                                          ; preds = %for.body, %entry
111  ret void
112}
113
114attributes #0 = { nounwind }
115