1; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
3
4; This should end with an no-op sequence of exec mask manipulations
5; Mask should be in original state after executed unreachable block
6
7
8; GCN-LABEL: {{^}}uniform_br_trivial_ret_divergent_br_trivial_unreachable:
9; GCN: s_cbranch_scc1 [[RET_BB:BB[0-9]+_[0-9]+]]
10
11; GCN-NEXT: ; %else
12
13; GCN: s_and_saveexec_b64 [[SAVE_EXEC:s\[[0-9]+:[0-9]+\]]], vcc
14
15; GCN: ; %bb.{{[0-9]+}}: ; %unreachable.bb
16; GCN-NEXT: ; divergent unreachable
17
18; GCN-NEXT: ; %bb.{{[0-9]+}}: ; %Flow
19; GCN-NEXT: s_or_b64 exec, exec
20
21; GCN-NEXT: [[RET_BB]]:
22; GCN-NEXT: ; return
23; GCN-NEXT: .Lfunc_end0
24define amdgpu_ps <{ i32, i32, i32, i32, i32, i32, i32, i32, i32, float, float, float, float, float, float, float, float, float, float, float, float, float, float }> @uniform_br_trivial_ret_divergent_br_trivial_unreachable([9 x <4 x i32>] addrspace(4)* inreg %arg, [17 x <4 x i32>] addrspace(4)* inreg %arg1, [17 x <8 x i32>] addrspace(4)* inreg %arg2, i32 addrspace(4)* inreg %arg3, float inreg %arg4, i32 inreg %arg5, <2 x i32> %arg6, <2 x i32> %arg7, <2 x i32> %arg8, <3 x i32> %arg9, <2 x i32> %arg10, <2 x i32> %arg11, <2 x i32> %arg12, float %arg13, float %arg14, float %arg15, float %arg16, i32 inreg %arg17, i32 %arg18, i32 %arg19, float %arg20, i32 %arg21) #0 {
25entry:
26  %i.i = extractelement <2 x i32> %arg7, i32 0
27  %j.i = extractelement <2 x i32> %arg7, i32 1
28  %i.f.i = bitcast i32 %i.i to float
29  %j.f.i = bitcast i32 %j.i to float
30  %p1.i = call float @llvm.amdgcn.interp.p1(float %i.f.i, i32 1, i32 0, i32 %arg5) #2
31  %p2 = call float @llvm.amdgcn.interp.p2(float %p1.i, float %j.f.i, i32 1, i32 0, i32 %arg5) #2
32  %p87 = fmul float %p2, %p2
33  %p88 = fadd float %p87, %p87
34  %p93 = fadd float %p88, %p88
35  %p97 = fmul float %p93, %p93
36  %p102 = fsub float %p97, %p97
37  %p104 = fmul float %p102, %p102
38  %p106 = fadd float 0.000000e+00, %p104
39  %p108 = fadd float %p106, %p106
40  %uniform.cond = icmp slt i32 %arg17, 0
41  br i1 %uniform.cond, label %ret.bb, label %else
42
43else:                                             ; preds = %main_body
44  %p124 = fmul float %p108, %p108
45  %p125 = fsub float %p124, %p124
46  %divergent.cond = fcmp olt float %p125, 0.000000e+00
47  br i1 %divergent.cond, label %ret.bb, label %unreachable.bb
48
49unreachable.bb:                                           ; preds = %else
50  unreachable
51
52ret.bb:                                          ; preds = %else, %main_body
53  ret <{ i32, i32, i32, i32, i32, i32, i32, i32, i32, float, float, float, float, float, float, float, float, float, float, float, float, float, float }> undef
54}
55
56; GCN-LABEL: {{^}}uniform_br_nontrivial_ret_divergent_br_nontrivial_unreachable:
57; GCN: s_cbranch_vccz
58
59; GCN: ; %bb.{{[0-9]+}}: ; %Flow
60; GCN: s_cbranch_execnz [[RETURN:BB[0-9]+_[0-9]+]]
61
62; GCN: ; %UnifiedReturnBlock
63; GCN-NEXT: s_or_b64 exec, exec
64; GCN-NEXT: s_waitcnt
65
66; GCN: BB{{[0-9]+_[0-9]+}}: ; %else
67; GCN: s_and_saveexec_b64 [[SAVE_EXEC:s\[[0-9]+:[0-9]+\]]], vcc
68
69; GCN-NEXT:  ; %unreachable.bb
70; GCN: ds_write_b32
71; GCN: ; divergent unreachable
72
73; GCN: ; %ret.bb
74; GCN: store_dword
75define amdgpu_ps <{ i32, i32, i32, i32, i32, i32, i32, i32, i32, float, float, float, float, float, float, float, float, float, float, float, float, float, float }> @uniform_br_nontrivial_ret_divergent_br_nontrivial_unreachable([9 x <4 x i32>] addrspace(4)* inreg %arg, [17 x <4 x i32>] addrspace(4)* inreg %arg1, [17 x <8 x i32>] addrspace(4)* inreg %arg2, i32 addrspace(4)* inreg %arg3, float inreg %arg4, i32 inreg %arg5, <2 x i32> %arg6, <2 x i32> %arg7, <2 x i32> %arg8, <3 x i32> %arg9, <2 x i32> %arg10, <2 x i32> %arg11, <2 x i32> %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, i32 inreg %arg18, i32 %arg19, float %arg20, i32 %arg21) #0 {
76main_body:
77  %i.i = extractelement <2 x i32> %arg7, i32 0
78  %j.i = extractelement <2 x i32> %arg7, i32 1
79  %i.f.i = bitcast i32 %i.i to float
80  %j.f.i = bitcast i32 %j.i to float
81  %p1.i = call float @llvm.amdgcn.interp.p1(float %i.f.i, i32 1, i32 0, i32 %arg5) #2
82  %p2 = call float @llvm.amdgcn.interp.p2(float %p1.i, float %j.f.i, i32 1, i32 0, i32 %arg5) #2
83  %p87 = fmul float %p2, %p2
84  %p88 = fadd float %p87, %p87
85  %p93 = fadd float %p88, %p88
86  %p97 = fmul float %p93, %p93
87  %p102 = fsub float %p97, %p97
88  %p104 = fmul float %p102, %p102
89  %p106 = fadd float 0.000000e+00, %p104
90  %p108 = fadd float %p106, %p106
91  %uniform.cond = icmp slt i32 %arg18, 0
92  br i1 %uniform.cond, label %ret.bb, label %else
93
94else:                                             ; preds = %main_body
95  %p124 = fmul float %p108, %p108
96  %p125 = fsub float %p124, %p124
97  %divergent.cond = fcmp olt float %p125, 0.000000e+00
98  br i1 %divergent.cond, label %ret.bb, label %unreachable.bb
99
100unreachable.bb:                                           ; preds = %else
101  store volatile i32 8, i32 addrspace(3)* undef
102  unreachable
103
104ret.bb:                                          ; preds = %else, %main_body
105  store volatile i32 11, i32 addrspace(1)* undef
106  ret <{ i32, i32, i32, i32, i32, i32, i32, i32, i32, float, float, float, float, float, float, float, float, float, float, float, float, float, float }> undef
107}
108
109; Function Attrs: nounwind readnone
110declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32) #1
111
112; Function Attrs: nounwind readnone
113declare float @llvm.amdgcn.interp.p2(float, float, i32, i32, i32) #1
114
115; Function Attrs: nounwind readnone
116declare float @llvm.amdgcn.interp.mov(i32, i32, i32, i32) #1
117
118; Function Attrs: nounwind readnone
119declare float @llvm.fabs.f32(float) #1
120
121; Function Attrs: nounwind readnone
122declare float @llvm.sqrt.f32(float) #1
123
124; Function Attrs: nounwind readnone
125declare float @llvm.floor.f32(float) #1
126
127attributes #0 = { "InitialPSInputAddr"="36983" }
128attributes #1 = { nounwind readnone }
129attributes #2 = { nounwind }
130