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]]
24;
25entry:
26  br label %loop
27
28loop:
29  store volatile i32 999, i32 addrspace(1)* %out, align 4
30  br label %loop
31}
32
33define amdgpu_kernel void @infinite_loop_ret(i32 addrspace(1)* %out) {
34; SI-LABEL: infinite_loop_ret:
35; SI:       ; %bb.0: ; %entry
36; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
37; SI-NEXT:    s_and_saveexec_b64 s[2:3], vcc
38; SI-NEXT:    s_cbranch_execz BB1_3
39; SI-NEXT:  ; %bb.1: ; %loop.preheader
40; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
41; SI-NEXT:    s_mov_b32 s3, 0xf000
42; SI-NEXT:    s_mov_b32 s2, -1
43; SI-NEXT:    v_mov_b32_e32 v0, 0x3e7
44; SI-NEXT:    s_and_b64 vcc, exec, -1
45; SI-NEXT:  BB1_2: ; %loop
46; SI-NEXT:    ; =>This Inner Loop Header: Depth=1
47; SI-NEXT:    s_waitcnt lgkmcnt(0)
48; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
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
62;
63entry:
64  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
65  %cond = icmp eq i32 %tmp, 1
66  br i1 %cond, label %loop, label %return
67
68loop:
69  store volatile i32 999, i32 addrspace(1)* %out, align 4
70  br label %loop
71
72return:
73  ret void
74}
75
76define amdgpu_kernel void @infinite_loops(i32 addrspace(1)* %out) {
77; SI-LABEL: infinite_loops:
78; SI:       ; %bb.0: ; %entry
79; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
80; SI-NEXT:    s_mov_b64 s[2:3], -1
81; SI-NEXT:    s_cbranch_scc1 BB2_4
82; SI-NEXT:  ; %bb.1:
83; SI-NEXT:    s_mov_b32 s3, 0xf000
84; SI-NEXT:    s_mov_b32 s2, -1
85; SI-NEXT:    v_mov_b32_e32 v0, 0x378
86; SI-NEXT:    s_and_b64 vcc, exec, -1
87; SI-NEXT:  BB2_2:
88; SI:         s_waitcnt lgkmcnt(0)
89; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
90; SI-NEXT:    s_cbranch_vccnz BB2_2
91; SI-NEXT:  ; %bb.3:
92; SI-NEXT:    s_mov_b64 s[2:3], 0
93; SI-NEXT:  BB2_4:
94; SI-NEXT:    s_and_b64 vcc, exec, s[2:3]
95; SI-NEXT:    s_waitcnt lgkmcnt(0)
96; SI-NEXT:    s_mov_b64 vcc, vcc
97; SI-NEXT:    s_cbranch_vccz BB2_7
98; SI-NEXT:  ; %bb.5:
99; SI-NEXT:    s_mov_b32 s3, 0xf000
100; SI-NEXT:    s_mov_b32 s2, -1
101; SI-NEXT:    s_waitcnt expcnt(0)
102; SI-NEXT:    v_mov_b32_e32 v0, 0x3e7
103; SI-NEXT:    s_and_b64 vcc, exec, 0
104; SI-NEXT:  BB2_6:
105; SI:         buffer_store_dword v0, off, s[0:3], 0
106; SI-NEXT:    s_cbranch_vccz BB2_6
107; SI-NEXT:  BB2_7:
108; SI-NEXT:    s_endpgm
109
110; IR-LABEL: @infinite_loops(
111; IR-NEXT:  entry:
112; IR-NEXT:    br i1 undef, label [[LOOP1:%.*]], label [[LOOP2:%.*]]
113; IR:       loop1:
114; IR-NEXT:    store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
115; IR-NEXT:    br i1 true, label [[LOOP1]], label [[DUMMYRETURNBLOCK:%.*]]
116; IR:       loop2:
117; IR-NEXT:    store volatile i32 888, i32 addrspace(1)* [[OUT]], align 4
118; IR-NEXT:    br i1 true, label [[LOOP2]], label [[DUMMYRETURNBLOCK]]
119; IR:       DummyReturnBlock:
120; IR-NEXT:    ret void
121;
122entry:
123  br i1 undef, label %loop1, label %loop2
124
125loop1:
126  store volatile i32 999, i32 addrspace(1)* %out, align 4
127  br label %loop1
128
129loop2:
130  store volatile i32 888, i32 addrspace(1)* %out, align 4
131  br label %loop2
132}
133
134define amdgpu_kernel void @infinite_loop_nest_ret(i32 addrspace(1)* %out) {
135; SI-LABEL: infinite_loop_nest_ret:
136; SI:       ; %bb.0: ; %entry
137; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
138; SI-NEXT:    s_and_saveexec_b64 s[2:3], vcc
139; SI-NEXT:    s_cbranch_execz BB3_5
140; SI-NEXT:  ; %bb.1: ; %outer_loop.preheader
141; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
142; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
143; SI-NEXT:    v_cmp_ne_u32_e64 s[0:1], 3, v0
144; SI-NEXT:    s_mov_b64 s[2:3], 0
145; SI-NEXT:    s_mov_b32 s7, 0xf000
146; SI-NEXT:    s_mov_b32 s6, -1
147; SI-NEXT:  BB3_2: ; %outer_loop
148; SI-NEXT:    ; =>This Loop Header: Depth=1
149; SI-NEXT:    ; Child Loop BB3_3 Depth 2
150; SI-NEXT:    s_and_b64 s[8:9], exec, vcc
151; SI-NEXT:    s_or_b64 s[2:3], s[8:9], s[2:3]
152; SI-NEXT:    s_mov_b64 s[8:9], 0
153; SI-NEXT:  BB3_3: ; %inner_loop
154; SI-NEXT:    ; Parent Loop BB3_2 Depth=1
155; SI-NEXT:    ; => This Inner Loop Header: Depth=2
156; SI-NEXT:    s_and_b64 s[10:11], exec, s[0:1]
157; SI-NEXT:    s_or_b64 s[8:9], s[10:11], s[8:9]
158; SI-NEXT:    s_waitcnt expcnt(0)
159; SI-NEXT:    v_mov_b32_e32 v0, 0x3e7
160; SI-NEXT:    s_waitcnt lgkmcnt(0)
161; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
162; SI-NEXT:    s_andn2_b64 exec, exec, s[8:9]
163; SI-NEXT:    s_cbranch_execnz BB3_3
164; SI-NEXT:  ; %bb.4: ; %Flow
165; SI-NEXT:    ; in Loop: Header=BB3_2 Depth=1
166; SI-NEXT:    s_or_b64 exec, exec, s[8:9]
167; SI-NEXT:    s_andn2_b64 exec, exec, s[2:3]
168; SI-NEXT:    s_cbranch_execnz BB3_2
169; SI-NEXT:  BB3_5: ; %UnifiedReturnBlock
170; SI-NEXT:    s_endpgm
171; IR-LABEL: @infinite_loop_nest_ret(
172; IR-NEXT:  entry:
173; IR-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
174; IR-NEXT:    [[COND1:%.*]] = icmp eq i32 [[TMP]], 1
175; IR-NEXT:    br i1 [[COND1]], label [[OUTER_LOOP:%.*]], label [[UNIFIEDRETURNBLOCK:%.*]]
176; IR:       outer_loop:
177; IR-NEXT:    br label [[INNER_LOOP:%.*]]
178; IR:       inner_loop:
179; IR-NEXT:    store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
180; IR-NEXT:    [[COND3:%.*]] = icmp eq i32 [[TMP]], 3
181; IR-NEXT:    br i1 true, label [[TRANSITIONBLOCK:%.*]], label [[UNIFIEDRETURNBLOCK]]
182; IR:       TransitionBlock:
183; IR-NEXT:    br i1 [[COND3]], label [[INNER_LOOP]], label [[OUTER_LOOP]]
184; IR:       UnifiedReturnBlock:
185; IR-NEXT:    ret void
186;
187entry:
188  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
189  %cond1 = icmp eq i32 %tmp, 1
190  br i1 %cond1, label %outer_loop, label %return
191
192outer_loop:
193  ; %cond2 = icmp eq i32 %tmp, 2
194  ; br i1 %cond2, label %outer_loop, label %inner_loop
195  br label %inner_loop
196
197inner_loop:                                     ; preds = %LeafBlock, %LeafBlock1
198  store volatile i32 999, i32 addrspace(1)* %out, align 4
199  %cond3 = icmp eq i32 %tmp, 3
200  br i1 %cond3, label %inner_loop, label %outer_loop
201
202return:
203  ret void
204}
205
206declare i32 @llvm.amdgcn.workitem.id.x()
207