1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
3; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
4; RUN: opt -mtriple=amdgcn-- -S -amdgpu-unify-divergent-exit-nodes -verify %s | FileCheck -check-prefix=IR %s
5
6define amdgpu_kernel void @infinite_loop(i32 addrspace(1)* %out) {
7; SI-LABEL: infinite_loop:
8; SI:       ; %bb.0: ; %entry
9; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
10; SI-NEXT:    s_mov_b32 s3, 0xf000
11; SI-NEXT:    s_mov_b32 s2, -1
12; SI-NEXT:    v_mov_b32_e32 v0, 0x3e7
13; SI-NEXT:  BB0_1: ; %loop
14; SI-NEXT:    ; =>This Inner Loop Header: Depth=1
15; SI-NEXT:    s_waitcnt lgkmcnt(0)
16; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
17; SI-NEXT:    s_branch BB0_1
18; IR-LABEL: @infinite_loop(
19; IR-NEXT:  entry:
20; IR-NEXT:    br label [[LOOP:%.*]]
21; IR:       loop:
22; IR-NEXT:    store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
23; IR-NEXT:    br label [[LOOP]]
24entry:
25  br label %loop
26
27loop:
28  store volatile i32 999, i32 addrspace(1)* %out, align 4
29  br label %loop
30}
31
32define amdgpu_kernel void @infinite_loop_ret(i32 addrspace(1)* %out) {
33; SI-LABEL: infinite_loop_ret:
34; SI:       ; %bb.0: ; %entry
35; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
36; SI-NEXT:    s_and_saveexec_b64 s[2:3], vcc
37; SI-NEXT:    s_cbranch_execz BB1_3
38; SI-NEXT:  ; %bb.1: ; %loop.preheader
39; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
40; SI-NEXT:    s_mov_b32 s3, 0xf000
41; SI-NEXT:    s_mov_b32 s2, -1
42; SI-NEXT:    v_mov_b32_e32 v0, 0x3e7
43; SI-NEXT:    s_and_b64 vcc, exec, -1
44; SI-NEXT:  BB1_2: ; %loop
45; SI-NEXT:    ; =>This Inner Loop Header: Depth=1
46; SI-NEXT:    s_waitcnt lgkmcnt(0)
47; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
48; SI-NEXT:    s_mov_b64 vcc, vcc
49; SI-NEXT:    s_cbranch_vccnz BB1_2
50; SI-NEXT:  BB1_3: ; %UnifiedReturnBlock
51; SI-NEXT:    s_endpgm
52; IR-LABEL: @infinite_loop_ret(
53; IR-NEXT:  entry:
54; IR-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
55; IR-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP]], 1
56; IR-NEXT:    br i1 [[COND]], label [[LOOP:%.*]], label [[UNIFIEDRETURNBLOCK:%.*]]
57; IR:       loop:
58; IR-NEXT:    store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
59; IR-NEXT:    br i1 true, label [[LOOP]], label [[UNIFIEDRETURNBLOCK]]
60; IR:       UnifiedReturnBlock:
61; IR-NEXT:    ret void
62entry:
63  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
64  %cond = icmp eq i32 %tmp, 1
65  br i1 %cond, label %loop, label %return
66
67loop:
68  store volatile i32 999, i32 addrspace(1)* %out, align 4
69  br label %loop
70
71return:
72  ret void
73}
74
75define amdgpu_kernel void @infinite_loops(i32 addrspace(1)* %out) {
76; SI-LABEL: infinite_loops:
77; SI:       ; %bb.0: ; %entry
78; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
79; SI-NEXT:    s_mov_b64 s[2:3], -1
80; SI-NEXT:    s_cbranch_scc1 BB2_4
81; SI-NEXT:  ; %bb.1:
82; SI-NEXT:    s_mov_b32 s3, 0xf000
83; SI-NEXT:    s_mov_b32 s2, -1
84; SI-NEXT:    v_mov_b32_e32 v0, 0x378
85; SI-NEXT:    s_and_b64 vcc, exec, -1
86; SI-NEXT:  BB2_2: ; %loop2
87; SI-NEXT:    ; =>This Inner Loop Header: Depth=1
88; SI-NEXT:    s_waitcnt lgkmcnt(0)
89; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
90; SI-NEXT:    s_mov_b64 vcc, vcc
91; SI-NEXT:    s_cbranch_vccnz BB2_2
92; SI-NEXT:  ; %bb.3: ; %Flow
93; SI-NEXT:    s_mov_b64 s[2:3], 0
94; SI-NEXT:  BB2_4: ; %Flow2
95; SI-NEXT:    s_and_b64 vcc, exec, s[2:3]
96; SI-NEXT:    s_waitcnt lgkmcnt(0)
97; SI-NEXT:    s_mov_b64 vcc, vcc
98; SI-NEXT:    s_cbranch_vccz BB2_7
99; SI-NEXT:  ; %bb.5:
100; SI-NEXT:    s_mov_b32 s3, 0xf000
101; SI-NEXT:    s_mov_b32 s2, -1
102; SI-NEXT:    s_waitcnt expcnt(0)
103; SI-NEXT:    v_mov_b32_e32 v0, 0x3e7
104; SI-NEXT:    s_and_b64 vcc, exec, 0
105; SI-NEXT:  BB2_6: ; %loop1
106; SI-NEXT:    ; =>This Inner Loop Header: Depth=1
107; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
108; SI-NEXT:    s_mov_b64 vcc, vcc
109; SI-NEXT:    s_cbranch_vccz BB2_6
110; SI-NEXT:  BB2_7: ; %DummyReturnBlock
111; SI-NEXT:    s_endpgm
112; IR-LABEL: @infinite_loops(
113; IR-NEXT:  entry:
114; IR-NEXT:    br i1 undef, label [[LOOP1:%.*]], label [[LOOP2:%.*]]
115; IR:       loop1:
116; IR-NEXT:    store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
117; IR-NEXT:    br i1 true, label [[LOOP1]], label [[DUMMYRETURNBLOCK:%.*]]
118; IR:       loop2:
119; IR-NEXT:    store volatile i32 888, i32 addrspace(1)* [[OUT]], align 4
120; IR-NEXT:    br i1 true, label [[LOOP2]], label [[DUMMYRETURNBLOCK]]
121; IR:       DummyReturnBlock:
122; IR-NEXT:    ret void
123entry:
124  br i1 undef, label %loop1, label %loop2
125
126loop1:
127  store volatile i32 999, i32 addrspace(1)* %out, align 4
128  br label %loop1
129
130loop2:
131  store volatile i32 888, i32 addrspace(1)* %out, align 4
132  br label %loop2
133}
134
135define amdgpu_kernel void @infinite_loop_nest_ret(i32 addrspace(1)* %out) {
136; SI-LABEL: infinite_loop_nest_ret:
137; SI:       ; %bb.0: ; %entry
138; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
139; SI-NEXT:    s_and_saveexec_b64 s[2:3], vcc
140; SI-NEXT:    s_cbranch_execz BB3_5
141; SI-NEXT:  ; %bb.1: ; %outer_loop.preheader
142; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
143; SI-NEXT:    v_cmp_ne_u32_e64 s[0:1], 3, v0
144; SI-NEXT:    s_mov_b32 s7, 0xf000
145; SI-NEXT:    s_mov_b32 s6, -1
146; SI-NEXT:  BB3_2: ; %outer_loop
147; SI-NEXT:    ; =>This Loop Header: Depth=1
148; SI-NEXT:    ; Child Loop BB3_3 Depth 2
149; SI-NEXT:    s_mov_b64 s[2:3], 0
150; SI-NEXT:  BB3_3: ; %inner_loop
151; SI-NEXT:    ; Parent Loop BB3_2 Depth=1
152; SI-NEXT:    ; => This Inner Loop Header: Depth=2
153; SI-NEXT:    s_and_b64 s[8:9], exec, s[0:1]
154; SI-NEXT:    s_or_b64 s[2:3], s[8:9], s[2:3]
155; SI-NEXT:    s_waitcnt expcnt(0)
156; SI-NEXT:    v_mov_b32_e32 v0, 0x3e7
157; SI-NEXT:    s_waitcnt lgkmcnt(0)
158; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
159; SI-NEXT:    s_andn2_b64 exec, exec, s[2:3]
160; SI-NEXT:    s_cbranch_execnz BB3_3
161; SI-NEXT:  ; %bb.4: ; %loop.exit.guard
162; SI-NEXT:    ; in Loop: Header=BB3_2 Depth=1
163; SI-NEXT:    s_or_b64 exec, exec, s[2:3]
164; SI-NEXT:    s_mov_b64 vcc, 0
165; SI-NEXT:    s_branch BB3_2
166; SI-NEXT:  BB3_5: ; %UnifiedReturnBlock
167; SI-NEXT:    s_endpgm
168; IR-LABEL: @infinite_loop_nest_ret(
169; IR-NEXT:  entry:
170; IR-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
171; IR-NEXT:    [[COND1:%.*]] = icmp eq i32 [[TMP]], 1
172; IR-NEXT:    br i1 [[COND1]], label [[OUTER_LOOP:%.*]], label [[UNIFIEDRETURNBLOCK:%.*]]
173; IR:       outer_loop:
174; IR-NEXT:    br label [[INNER_LOOP:%.*]]
175; IR:       inner_loop:
176; IR-NEXT:    store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
177; IR-NEXT:    [[COND3:%.*]] = icmp eq i32 [[TMP]], 3
178; IR-NEXT:    br i1 true, label [[TRANSITIONBLOCK:%.*]], label [[UNIFIEDRETURNBLOCK]]
179; IR:       TransitionBlock:
180; IR-NEXT:    br i1 [[COND3]], label [[INNER_LOOP]], label [[OUTER_LOOP]]
181; IR:       UnifiedReturnBlock:
182; IR-NEXT:    ret void
183entry:
184  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
185  %cond1 = icmp eq i32 %tmp, 1
186  br i1 %cond1, label %outer_loop, label %return
187
188outer_loop:
189  ; %cond2 = icmp eq i32 %tmp, 2
190  ; br i1 %cond2, label %outer_loop, label %inner_loop
191  br label %inner_loop
192
193inner_loop:                                     ; preds = %LeafBlock, %LeafBlock1
194  store volatile i32 999, i32 addrspace(1)* %out, align 4
195  %cond3 = icmp eq i32 %tmp, 3
196  br i1 %cond3, label %inner_loop, label %outer_loop
197
198return:
199  ret void
200}
201
202declare i32 @llvm.amdgcn.workitem.id.x()
203