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 -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefix=SI %s
4 ; RUN: opt -mtriple=amdgcn-- -S -amdgpu-unify-divergent-exit-nodes -verify -simplifycfg-require-and-preserve-domtree=1 %s | FileCheck -check-prefix=IR %s
6 define 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_waitcnt vmcnt(0)
18 ; SI-NEXT: s_branch BB0_1
19 ; IR-LABEL: @infinite_loop(
21 ; IR-NEXT: br label [[LOOP:%.*]]
23 ; IR-NEXT: store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
24 ; IR-NEXT: br label [[LOOP]]
29 store volatile i32 999, i32 addrspace(1)* %out, align 4
33 define 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_waitcnt vmcnt(0)
50 ; SI-NEXT: s_mov_b64 vcc, vcc
51 ; SI-NEXT: s_cbranch_vccnz BB1_2
52 ; SI-NEXT: BB1_3: ; %UnifiedReturnBlock
54 ; IR-LABEL: @infinite_loop_ret(
56 ; IR-NEXT: [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
57 ; IR-NEXT: [[COND:%.*]] = icmp eq i32 [[TMP]], 1
58 ; IR-NEXT: br i1 [[COND]], label [[LOOP:%.*]], label [[UNIFIEDRETURNBLOCK:%.*]]
60 ; IR-NEXT: store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
61 ; IR-NEXT: br i1 true, label [[LOOP]], label [[UNIFIEDRETURNBLOCK]]
62 ; IR: UnifiedReturnBlock:
65 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
66 %cond = icmp eq i32 %tmp, 1
67 br i1 %cond, label %loop, label %return
70 store volatile i32 999, i32 addrspace(1)* %out, align 4
77 define amdgpu_kernel void @infinite_loops(i32 addrspace(1)* %out) {
78 ; SI-LABEL: infinite_loops:
79 ; SI: ; %bb.0: ; %entry
80 ; SI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
81 ; SI-NEXT: s_mov_b64 s[2:3], -1
82 ; SI-NEXT: s_cbranch_scc1 BB2_4
84 ; SI-NEXT: s_mov_b32 s3, 0xf000
85 ; SI-NEXT: s_mov_b32 s2, -1
86 ; SI-NEXT: v_mov_b32_e32 v0, 0x378
87 ; SI-NEXT: s_and_b64 vcc, exec, -1
88 ; SI-NEXT: BB2_2: ; %loop2
89 ; SI-NEXT: ; =>This Inner Loop Header: Depth=1
90 ; SI-NEXT: s_waitcnt lgkmcnt(0)
91 ; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
92 ; SI-NEXT: s_waitcnt vmcnt(0)
93 ; SI-NEXT: s_mov_b64 vcc, vcc
94 ; SI-NEXT: s_cbranch_vccnz BB2_2
95 ; SI-NEXT: ; %bb.3: ; %Flow
96 ; SI-NEXT: s_mov_b64 s[2:3], 0
97 ; SI-NEXT: BB2_4: ; %Flow2
98 ; SI-NEXT: s_and_b64 vcc, exec, s[2:3]
99 ; SI-NEXT: s_waitcnt lgkmcnt(0)
100 ; SI-NEXT: s_mov_b64 vcc, vcc
101 ; SI-NEXT: s_cbranch_vccz BB2_7
103 ; SI-NEXT: s_mov_b32 s3, 0xf000
104 ; SI-NEXT: s_mov_b32 s2, -1
105 ; SI-NEXT: s_waitcnt expcnt(0)
106 ; SI-NEXT: v_mov_b32_e32 v0, 0x3e7
107 ; SI-NEXT: s_and_b64 vcc, exec, 0
108 ; SI-NEXT: BB2_6: ; %loop1
109 ; SI-NEXT: ; =>This Inner Loop Header: Depth=1
110 ; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
111 ; SI-NEXT: s_waitcnt vmcnt(0)
112 ; SI-NEXT: s_mov_b64 vcc, vcc
113 ; SI-NEXT: s_cbranch_vccz BB2_6
114 ; SI-NEXT: BB2_7: ; %DummyReturnBlock
116 ; IR-LABEL: @infinite_loops(
118 ; IR-NEXT: br i1 undef, label [[LOOP1:%.*]], label [[LOOP2:%.*]]
120 ; IR-NEXT: store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
121 ; IR-NEXT: br i1 true, label [[LOOP1]], label [[DUMMYRETURNBLOCK:%.*]]
123 ; IR-NEXT: store volatile i32 888, i32 addrspace(1)* [[OUT]], align 4
124 ; IR-NEXT: br i1 true, label [[LOOP2]], label [[DUMMYRETURNBLOCK]]
125 ; IR: DummyReturnBlock:
128 br i1 undef, label %loop1, label %loop2
131 store volatile i32 999, i32 addrspace(1)* %out, align 4
135 store volatile i32 888, i32 addrspace(1)* %out, align 4
139 define amdgpu_kernel void @infinite_loop_nest_ret(i32 addrspace(1)* %out) {
140 ; SI-LABEL: infinite_loop_nest_ret:
141 ; SI: ; %bb.0: ; %entry
142 ; SI-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
143 ; SI-NEXT: s_and_saveexec_b64 s[2:3], vcc
144 ; SI-NEXT: s_cbranch_execz BB3_5
145 ; SI-NEXT: ; %bb.1: ; %outer_loop.preheader
146 ; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
147 ; SI-NEXT: v_cmp_ne_u32_e64 s[0:1], 3, v0
148 ; SI-NEXT: s_mov_b32 s7, 0xf000
149 ; SI-NEXT: s_mov_b32 s6, -1
150 ; SI-NEXT: BB3_2: ; %outer_loop
151 ; SI-NEXT: ; =>This Loop Header: Depth=1
152 ; SI-NEXT: ; Child Loop BB3_3 Depth 2
153 ; SI-NEXT: s_mov_b64 s[2:3], 0
154 ; SI-NEXT: BB3_3: ; %inner_loop
155 ; SI-NEXT: ; Parent Loop BB3_2 Depth=1
156 ; SI-NEXT: ; => This Inner Loop Header: Depth=2
157 ; SI-NEXT: s_and_b64 s[8:9], exec, s[0:1]
158 ; SI-NEXT: s_or_b64 s[2:3], s[8:9], s[2:3]
159 ; SI-NEXT: s_waitcnt expcnt(0)
160 ; SI-NEXT: v_mov_b32_e32 v0, 0x3e7
161 ; SI-NEXT: s_waitcnt lgkmcnt(0)
162 ; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
163 ; SI-NEXT: s_waitcnt vmcnt(0)
164 ; SI-NEXT: s_andn2_b64 exec, exec, s[2:3]
165 ; SI-NEXT: s_cbranch_execnz BB3_3
166 ; SI-NEXT: ; %bb.4: ; %loop.exit.guard
167 ; SI-NEXT: ; in Loop: Header=BB3_2 Depth=1
168 ; SI-NEXT: s_or_b64 exec, exec, s[2:3]
169 ; SI-NEXT: s_mov_b64 vcc, 0
170 ; SI-NEXT: s_branch BB3_2
171 ; SI-NEXT: BB3_5: ; %UnifiedReturnBlock
173 ; IR-LABEL: @infinite_loop_nest_ret(
175 ; IR-NEXT: [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
176 ; IR-NEXT: [[COND1:%.*]] = icmp eq i32 [[TMP]], 1
177 ; IR-NEXT: br i1 [[COND1]], label [[OUTER_LOOP:%.*]], label [[UNIFIEDRETURNBLOCK:%.*]]
179 ; IR-NEXT: br label [[INNER_LOOP:%.*]]
181 ; IR-NEXT: store volatile i32 999, i32 addrspace(1)* [[OUT:%.*]], align 4
182 ; IR-NEXT: [[COND3:%.*]] = icmp eq i32 [[TMP]], 3
183 ; IR-NEXT: br i1 true, label [[TRANSITIONBLOCK:%.*]], label [[UNIFIEDRETURNBLOCK]]
184 ; IR: TransitionBlock:
185 ; IR-NEXT: br i1 [[COND3]], label [[INNER_LOOP]], label [[OUTER_LOOP]]
186 ; IR: UnifiedReturnBlock:
189 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
190 %cond1 = icmp eq i32 %tmp, 1
191 br i1 %cond1, label %outer_loop, label %return
194 ; %cond2 = icmp eq i32 %tmp, 2
195 ; br i1 %cond2, label %outer_loop, label %inner_loop
198 inner_loop: ; preds = %LeafBlock, %LeafBlock1
199 store volatile i32 999, i32 addrspace(1)* %out, align 4
200 %cond3 = icmp eq i32 %tmp, 3
201 br i1 %cond3, label %inner_loop, label %outer_loop
207 declare i32 @llvm.amdgcn.workitem.id.x()