xref: /llvm-project/llvm/test/CodeGen/AMDGPU/cf-loop-on-constant.ll (revision 6548b6354d1d990e1c98736f5e7c3de876bedc8e) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
3; RUN: llc -mtriple=amdgcn -verify-machineinstrs -O0 < %s | FileCheck -check-prefix=GCN_DBG %s
4
5define amdgpu_kernel void @test_loop(ptr addrspace(3) %ptr, i32 %n) nounwind {
6; GCN-LABEL: test_loop:
7; GCN:       ; %bb.0: ; %entry
8; GCN-NEXT:    s_load_dword s0, s[4:5], 0xa
9; GCN-NEXT:    s_waitcnt lgkmcnt(0)
10; GCN-NEXT:    s_cmp_eq_u32 s0, -1
11; GCN-NEXT:    s_cbranch_scc1 .LBB0_3
12; GCN-NEXT:  ; %bb.1: ; %for.body.preheader
13; GCN-NEXT:    s_load_dword s0, s[4:5], 0x9
14; GCN-NEXT:    s_waitcnt lgkmcnt(0)
15; GCN-NEXT:    s_addk_i32 s0, 0x80
16; GCN-NEXT:    s_and_b64 vcc, exec, -1
17; GCN-NEXT:    s_mov_b32 m0, -1
18; GCN-NEXT:  .LBB0_2: ; %for.body
19; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
20; GCN-NEXT:    v_mov_b32_e32 v0, s0
21; GCN-NEXT:    ds_read_b32 v1, v0
22; GCN-NEXT:    s_waitcnt lgkmcnt(0)
23; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
24; GCN-NEXT:    ds_write_b32 v0, v1
25; GCN-NEXT:    s_add_i32 s0, s0, 4
26; GCN-NEXT:    s_mov_b64 vcc, vcc
27; GCN-NEXT:    s_cbranch_vccnz .LBB0_2
28; GCN-NEXT:  .LBB0_3: ; %for.exit
29; GCN-NEXT:    s_endpgm
30;
31; GCN_DBG-LABEL: test_loop:
32; GCN_DBG:       ; %bb.0: ; %entry
33; GCN_DBG-NEXT:    s_mov_b32 s12, SCRATCH_RSRC_DWORD0
34; GCN_DBG-NEXT:    s_mov_b32 s13, SCRATCH_RSRC_DWORD1
35; GCN_DBG-NEXT:    s_mov_b32 s14, -1
36; GCN_DBG-NEXT:    s_mov_b32 s15, 0xe8f000
37; GCN_DBG-NEXT:    s_add_u32 s12, s12, s11
38; GCN_DBG-NEXT:    s_addc_u32 s13, s13, 0
39; GCN_DBG-NEXT:    s_load_dword s0, s[4:5], 0x9
40; GCN_DBG-NEXT:    ; implicit-def: $vgpr2 : SGPR spill to VGPR lane
41; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
42; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
43; GCN_DBG-NEXT:    s_load_dword s1, s[4:5], 0xa
44; GCN_DBG-NEXT:    s_mov_b32 s0, 0
45; GCN_DBG-NEXT:    s_mov_b32 s2, -1
46; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
47; GCN_DBG-NEXT:    s_cmp_lg_u32 s1, s2
48; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
49; GCN_DBG-NEXT:    s_mov_b64 s[6:7], exec
50; GCN_DBG-NEXT:    s_mov_b64 exec, -1
51; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
52; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
53; GCN_DBG-NEXT:    s_cbranch_scc1 .LBB0_2
54; GCN_DBG-NEXT:  ; %bb.1: ; %for.exit
55; GCN_DBG-NEXT:    s_endpgm
56; GCN_DBG-NEXT:  .LBB0_2: ; %for.body
57; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
58; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
59; GCN_DBG-NEXT:    s_waitcnt expcnt(0)
60; GCN_DBG-NEXT:    buffer_load_dword v2, off, s[12:15], 0 ; 4-byte Folded Reload
61; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
62; GCN_DBG-NEXT:    s_waitcnt vmcnt(0)
63; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
64; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
65; GCN_DBG-NEXT:    s_mov_b32 s1, 2
66; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
67; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
68; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
69; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
70; GCN_DBG-NEXT:    s_mov_b32 m0, -1
71; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
72; GCN_DBG-NEXT:    ds_read_b32 v0, v0
73; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
74; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
75; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
76; GCN_DBG-NEXT:    s_mov_b32 m0, -1
77; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
78; GCN_DBG-NEXT:    ds_write_b32 v0, v1
79; GCN_DBG-NEXT:    s_mov_b32 s1, 1
80; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
81; GCN_DBG-NEXT:    s_mov_b64 s[2:3], -1
82; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
83; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
84; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
85; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
86; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
87; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB0_2
88; GCN_DBG-NEXT:  ; %bb.3: ; %DummyReturnBlock
89; GCN_DBG-NEXT:    s_endpgm
90entry:
91  %cmp = icmp eq i32 %n, -1
92  br i1 %cmp, label %for.exit, label %for.body
93
94for.exit:
95  ret void
96
97for.body:
98  %indvar = phi i32 [ %inc, %for.body ], [ 0, %entry ]
99  %tmp = add i32 %indvar, 32
100  %arrayidx = getelementptr float, ptr addrspace(3) %ptr, i32 %tmp
101  %vecload = load float, ptr addrspace(3) %arrayidx, align 4
102  %add = fadd float %vecload, 1.0
103  store float %add, ptr addrspace(3) %arrayidx, align 8
104  %inc = add i32 %indvar, 1
105  br label %for.body
106}
107
108define amdgpu_kernel void @loop_const_true(ptr addrspace(3) %ptr, i32 %n) nounwind {
109; GCN-LABEL: loop_const_true:
110; GCN:       ; %bb.0: ; %entry
111; GCN-NEXT:    s_load_dword s0, s[4:5], 0x9
112; GCN-NEXT:    s_waitcnt lgkmcnt(0)
113; GCN-NEXT:    s_addk_i32 s0, 0x80
114; GCN-NEXT:    s_and_b64 vcc, exec, -1
115; GCN-NEXT:    s_mov_b32 m0, -1
116; GCN-NEXT:  .LBB1_1: ; %for.body
117; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
118; GCN-NEXT:    v_mov_b32_e32 v0, s0
119; GCN-NEXT:    ds_read_b32 v1, v0
120; GCN-NEXT:    s_waitcnt lgkmcnt(0)
121; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
122; GCN-NEXT:    ds_write_b32 v0, v1
123; GCN-NEXT:    s_add_i32 s0, s0, 4
124; GCN-NEXT:    s_mov_b64 vcc, vcc
125; GCN-NEXT:    s_cbranch_vccnz .LBB1_1
126; GCN-NEXT:  ; %bb.2: ; %DummyReturnBlock
127; GCN-NEXT:    s_endpgm
128;
129; GCN_DBG-LABEL: loop_const_true:
130; GCN_DBG:       ; %bb.0: ; %entry
131; GCN_DBG-NEXT:    s_mov_b32 s12, SCRATCH_RSRC_DWORD0
132; GCN_DBG-NEXT:    s_mov_b32 s13, SCRATCH_RSRC_DWORD1
133; GCN_DBG-NEXT:    s_mov_b32 s14, -1
134; GCN_DBG-NEXT:    s_mov_b32 s15, 0xe8f000
135; GCN_DBG-NEXT:    s_add_u32 s12, s12, s11
136; GCN_DBG-NEXT:    s_addc_u32 s13, s13, 0
137; GCN_DBG-NEXT:    s_load_dword s0, s[4:5], 0x9
138; GCN_DBG-NEXT:    ; implicit-def: $vgpr2 : SGPR spill to VGPR lane
139; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
140; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
141; GCN_DBG-NEXT:    s_mov_b32 s0, 0
142; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
143; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
144; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
145; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
146; GCN_DBG-NEXT:    s_branch .LBB1_2
147; GCN_DBG-NEXT:  .LBB1_1: ; %for.exit
148; GCN_DBG-NEXT:    s_endpgm
149; GCN_DBG-NEXT:  .LBB1_2: ; %for.body
150; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
151; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
152; GCN_DBG-NEXT:    s_waitcnt expcnt(0)
153; GCN_DBG-NEXT:    buffer_load_dword v2, off, s[12:15], 0 ; 4-byte Folded Reload
154; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
155; GCN_DBG-NEXT:    s_waitcnt vmcnt(0)
156; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
157; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
158; GCN_DBG-NEXT:    s_mov_b32 s1, 2
159; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
160; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
161; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
162; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
163; GCN_DBG-NEXT:    s_mov_b32 m0, -1
164; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
165; GCN_DBG-NEXT:    ds_read_b32 v0, v0
166; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
167; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
168; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
169; GCN_DBG-NEXT:    s_mov_b32 m0, -1
170; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
171; GCN_DBG-NEXT:    ds_write_b32 v0, v1
172; GCN_DBG-NEXT:    s_mov_b32 s1, 1
173; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
174; GCN_DBG-NEXT:    s_mov_b64 s[2:3], 0
175; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
176; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
177; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
178; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
179; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
180; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB1_1
181; GCN_DBG-NEXT:    s_branch .LBB1_2
182entry:
183  br label %for.body
184
185for.exit:
186  ret void
187
188for.body:
189  %indvar = phi i32 [ %inc, %for.body ], [ 0, %entry ]
190  %tmp = add i32 %indvar, 32
191  %arrayidx = getelementptr float, ptr addrspace(3) %ptr, i32 %tmp
192  %vecload = load float, ptr addrspace(3) %arrayidx, align 4
193  %add = fadd float %vecload, 1.0
194  store float %add, ptr addrspace(3) %arrayidx, align 8
195  %inc = add i32 %indvar, 1
196  br i1 true, label %for.body, label %for.exit
197}
198
199define amdgpu_kernel void @loop_const_false(ptr addrspace(3) %ptr, i32 %n) nounwind {
200; GCN-LABEL: loop_const_false:
201; GCN:       ; %bb.0: ; %entry
202; GCN-NEXT:    s_load_dword s0, s[4:5], 0x9
203; GCN-NEXT:    s_waitcnt lgkmcnt(0)
204; GCN-NEXT:    v_mov_b32_e32 v0, s0
205; GCN-NEXT:    s_mov_b32 m0, -1
206; GCN-NEXT:    ds_read_b32 v1, v0 offset:128
207; GCN-NEXT:    s_waitcnt lgkmcnt(0)
208; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
209; GCN-NEXT:    ds_write_b32 v0, v1 offset:128
210; GCN-NEXT:    s_endpgm
211;
212; GCN_DBG-LABEL: loop_const_false:
213; GCN_DBG:       ; %bb.0: ; %entry
214; GCN_DBG-NEXT:    s_mov_b32 s12, SCRATCH_RSRC_DWORD0
215; GCN_DBG-NEXT:    s_mov_b32 s13, SCRATCH_RSRC_DWORD1
216; GCN_DBG-NEXT:    s_mov_b32 s14, -1
217; GCN_DBG-NEXT:    s_mov_b32 s15, 0xe8f000
218; GCN_DBG-NEXT:    s_add_u32 s12, s12, s11
219; GCN_DBG-NEXT:    s_addc_u32 s13, s13, 0
220; GCN_DBG-NEXT:    s_load_dword s0, s[4:5], 0x9
221; GCN_DBG-NEXT:    ; implicit-def: $vgpr2 : SGPR spill to VGPR lane
222; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
223; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
224; GCN_DBG-NEXT:    s_mov_b32 s0, 0
225; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
226; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
227; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
228; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
229; GCN_DBG-NEXT:    s_branch .LBB2_2
230; GCN_DBG-NEXT:  .LBB2_1: ; %for.exit
231; GCN_DBG-NEXT:    s_endpgm
232; GCN_DBG-NEXT:  .LBB2_2: ; %for.body
233; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
234; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
235; GCN_DBG-NEXT:    s_waitcnt expcnt(0)
236; GCN_DBG-NEXT:    buffer_load_dword v2, off, s[12:15], 0 ; 4-byte Folded Reload
237; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
238; GCN_DBG-NEXT:    s_waitcnt vmcnt(0)
239; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
240; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
241; GCN_DBG-NEXT:    s_mov_b32 s1, 2
242; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
243; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
244; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
245; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
246; GCN_DBG-NEXT:    s_mov_b32 m0, -1
247; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
248; GCN_DBG-NEXT:    ds_read_b32 v0, v0
249; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
250; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
251; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
252; GCN_DBG-NEXT:    s_mov_b32 m0, -1
253; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
254; GCN_DBG-NEXT:    ds_write_b32 v0, v1
255; GCN_DBG-NEXT:    s_mov_b32 s1, 1
256; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
257; GCN_DBG-NEXT:    s_mov_b64 s[2:3], -1
258; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
259; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
260; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
261; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
262; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
263; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB2_1
264; GCN_DBG-NEXT:    s_branch .LBB2_2
265entry:
266  br label %for.body
267
268for.exit:
269  ret void
270
271; XXX - Should there be an S_ENDPGM?
272for.body:
273  %indvar = phi i32 [ %inc, %for.body ], [ 0, %entry ]
274  %tmp = add i32 %indvar, 32
275  %arrayidx = getelementptr float, ptr addrspace(3) %ptr, i32 %tmp
276  %vecload = load float, ptr addrspace(3) %arrayidx, align 4
277  %add = fadd float %vecload, 1.0
278  store float %add, ptr addrspace(3) %arrayidx, align 8
279  %inc = add i32 %indvar, 1
280  br i1 false, label %for.body, label %for.exit
281}
282
283define amdgpu_kernel void @loop_const_undef(ptr addrspace(3) %ptr, i32 %n) nounwind {
284; GCN-LABEL: loop_const_undef:
285; GCN:       ; %bb.0: ; %entry
286; GCN-NEXT:    s_load_dword s0, s[4:5], 0x9
287; GCN-NEXT:    s_waitcnt lgkmcnt(0)
288; GCN-NEXT:    v_mov_b32_e32 v0, s0
289; GCN-NEXT:    s_mov_b32 m0, -1
290; GCN-NEXT:    ds_read_b32 v1, v0 offset:128
291; GCN-NEXT:    s_waitcnt lgkmcnt(0)
292; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
293; GCN-NEXT:    ds_write_b32 v0, v1 offset:128
294; GCN-NEXT:    s_endpgm
295;
296; GCN_DBG-LABEL: loop_const_undef:
297; GCN_DBG:       ; %bb.0: ; %entry
298; GCN_DBG-NEXT:    s_mov_b32 s12, SCRATCH_RSRC_DWORD0
299; GCN_DBG-NEXT:    s_mov_b32 s13, SCRATCH_RSRC_DWORD1
300; GCN_DBG-NEXT:    s_mov_b32 s14, -1
301; GCN_DBG-NEXT:    s_mov_b32 s15, 0xe8f000
302; GCN_DBG-NEXT:    s_add_u32 s12, s12, s11
303; GCN_DBG-NEXT:    s_addc_u32 s13, s13, 0
304; GCN_DBG-NEXT:    s_load_dword s0, s[4:5], 0x9
305; GCN_DBG-NEXT:    ; implicit-def: $vgpr2 : SGPR spill to VGPR lane
306; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
307; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
308; GCN_DBG-NEXT:    s_mov_b32 s0, 0
309; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
310; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
311; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
312; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
313; GCN_DBG-NEXT:    s_branch .LBB3_2
314; GCN_DBG-NEXT:  .LBB3_1: ; %for.exit
315; GCN_DBG-NEXT:    s_endpgm
316; GCN_DBG-NEXT:  .LBB3_2: ; %for.body
317; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
318; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
319; GCN_DBG-NEXT:    s_waitcnt expcnt(0)
320; GCN_DBG-NEXT:    buffer_load_dword v2, off, s[12:15], 0 ; 4-byte Folded Reload
321; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
322; GCN_DBG-NEXT:    s_waitcnt vmcnt(0)
323; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
324; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
325; GCN_DBG-NEXT:    s_mov_b32 s1, 2
326; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
327; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
328; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
329; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
330; GCN_DBG-NEXT:    s_mov_b32 m0, -1
331; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
332; GCN_DBG-NEXT:    ds_read_b32 v0, v0
333; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
334; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
335; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
336; GCN_DBG-NEXT:    s_mov_b32 m0, -1
337; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
338; GCN_DBG-NEXT:    ds_write_b32 v0, v1
339; GCN_DBG-NEXT:    s_mov_b32 s1, 1
340; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
341; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
342; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
343; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
344; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
345; GCN_DBG-NEXT:    s_cbranch_scc1 .LBB3_1
346; GCN_DBG-NEXT:    s_branch .LBB3_2
347entry:
348  br label %for.body
349
350for.exit:
351  ret void
352
353; XXX - Should there be an s_endpgm?
354for.body:
355  %indvar = phi i32 [ %inc, %for.body ], [ 0, %entry ]
356  %tmp = add i32 %indvar, 32
357  %arrayidx = getelementptr float, ptr addrspace(3) %ptr, i32 %tmp
358  %vecload = load float, ptr addrspace(3) %arrayidx, align 4
359  %add = fadd float %vecload, 1.0
360  store float %add, ptr addrspace(3) %arrayidx, align 8
361  %inc = add i32 %indvar, 1
362  br i1 undef, label %for.body, label %for.exit
363}
364
365define amdgpu_kernel void @loop_arg_0(ptr addrspace(3) %ptr, i32 %n) nounwind {
366; GCN-LABEL: loop_arg_0:
367; GCN:       ; %bb.0: ; %entry
368; GCN-NEXT:    v_mov_b32_e32 v0, 0
369; GCN-NEXT:    s_mov_b32 m0, -1
370; GCN-NEXT:    ds_read_u8 v0, v0
371; GCN-NEXT:    s_load_dword s4, s[4:5], 0x9
372; GCN-NEXT:    s_waitcnt lgkmcnt(0)
373; GCN-NEXT:    v_readfirstlane_b32 s0, v0
374; GCN-NEXT:    s_bitcmp1_b32 s0, 0
375; GCN-NEXT:    s_cselect_b64 s[0:1], -1, 0
376; GCN-NEXT:    s_xor_b64 s[2:3], s[0:1], -1
377; GCN-NEXT:    s_add_i32 s0, s4, 0x80
378; GCN-NEXT:    s_and_b64 vcc, exec, s[2:3]
379; GCN-NEXT:  .LBB4_1: ; %for.body
380; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
381; GCN-NEXT:    v_mov_b32_e32 v0, s0
382; GCN-NEXT:    ds_read_b32 v1, v0
383; GCN-NEXT:    s_waitcnt lgkmcnt(0)
384; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
385; GCN-NEXT:    ds_write_b32 v0, v1
386; GCN-NEXT:    s_add_i32 s0, s0, 4
387; GCN-NEXT:    s_mov_b64 vcc, vcc
388; GCN-NEXT:    s_cbranch_vccz .LBB4_1
389; GCN-NEXT:  ; %bb.2: ; %for.exit
390; GCN-NEXT:    s_endpgm
391;
392; GCN_DBG-LABEL: loop_arg_0:
393; GCN_DBG:       ; %bb.0: ; %entry
394; GCN_DBG-NEXT:    s_mov_b32 s12, SCRATCH_RSRC_DWORD0
395; GCN_DBG-NEXT:    s_mov_b32 s13, SCRATCH_RSRC_DWORD1
396; GCN_DBG-NEXT:    s_mov_b32 s14, -1
397; GCN_DBG-NEXT:    s_mov_b32 s15, 0xe8f000
398; GCN_DBG-NEXT:    s_add_u32 s12, s12, s11
399; GCN_DBG-NEXT:    s_addc_u32 s13, s13, 0
400; GCN_DBG-NEXT:    s_load_dword s0, s[4:5], 0x9
401; GCN_DBG-NEXT:    ; implicit-def: $vgpr2 : SGPR spill to VGPR lane
402; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
403; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
404; GCN_DBG-NEXT:    v_mov_b32_e32 v0, 0
405; GCN_DBG-NEXT:    s_mov_b32 m0, -1
406; GCN_DBG-NEXT:    ds_read_u8 v0, v0
407; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
408; GCN_DBG-NEXT:    v_readfirstlane_b32 s0, v0
409; GCN_DBG-NEXT:    s_and_b32 s0, 1, s0
410; GCN_DBG-NEXT:    s_cmp_eq_u32 s0, 1
411; GCN_DBG-NEXT:    s_cselect_b64 s[0:1], -1, 0
412; GCN_DBG-NEXT:    s_mov_b64 s[2:3], -1
413; GCN_DBG-NEXT:    s_xor_b64 s[0:1], s[0:1], s[2:3]
414; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
415; GCN_DBG-NEXT:    v_writelane_b32 v2, s1, 2
416; GCN_DBG-NEXT:    s_mov_b32 s0, 0
417; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 3
418; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
419; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
420; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
421; GCN_DBG-NEXT:    s_branch .LBB4_2
422; GCN_DBG-NEXT:  .LBB4_1: ; %for.exit
423; GCN_DBG-NEXT:    s_endpgm
424; GCN_DBG-NEXT:  .LBB4_2: ; %for.body
425; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
426; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
427; GCN_DBG-NEXT:    s_waitcnt expcnt(0)
428; GCN_DBG-NEXT:    buffer_load_dword v2, off, s[12:15], 0 ; 4-byte Folded Reload
429; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
430; GCN_DBG-NEXT:    s_waitcnt vmcnt(0)
431; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 3
432; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 1
433; GCN_DBG-NEXT:    v_readlane_b32 s3, v2, 2
434; GCN_DBG-NEXT:    v_readlane_b32 s4, v2, 0
435; GCN_DBG-NEXT:    s_mov_b32 s1, 2
436; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
437; GCN_DBG-NEXT:    s_add_i32 s1, s1, s4
438; GCN_DBG-NEXT:    s_mov_b32 s4, 0x80
439; GCN_DBG-NEXT:    s_add_i32 s1, s1, s4
440; GCN_DBG-NEXT:    s_mov_b32 m0, -1
441; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
442; GCN_DBG-NEXT:    ds_read_b32 v0, v0
443; GCN_DBG-NEXT:    s_mov_b32 s4, 1.0
444; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
445; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s4
446; GCN_DBG-NEXT:    s_mov_b32 m0, -1
447; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
448; GCN_DBG-NEXT:    ds_write_b32 v0, v1
449; GCN_DBG-NEXT:    s_mov_b32 s1, 1
450; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
451; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
452; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 3
453; GCN_DBG-NEXT:    s_or_saveexec_b64 s[6:7], -1
454; GCN_DBG-NEXT:    buffer_store_dword v2, off, s[12:15], 0 ; 4-byte Folded Spill
455; GCN_DBG-NEXT:    s_mov_b64 exec, s[6:7]
456; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB4_1
457; GCN_DBG-NEXT:    s_branch .LBB4_2
458entry:
459  %cond = load volatile i1, ptr addrspace(3) null
460  br label %for.body
461
462for.exit:
463  ret void
464
465for.body:
466  %indvar = phi i32 [ %inc, %for.body ], [ 0, %entry ]
467  %tmp = add i32 %indvar, 32
468  %arrayidx = getelementptr float, ptr addrspace(3) %ptr, i32 %tmp
469  %vecload = load float, ptr addrspace(3) %arrayidx, align 4
470  %add = fadd float %vecload, 1.0
471  store float %add, ptr addrspace(3) %arrayidx, align 8
472  %inc = add i32 %indvar, 1
473  br i1 %cond, label %for.body, label %for.exit
474}
475
476!llvm.module.flags = !{!0}
477!0 = !{i32 1, !"amdhsa_code_object_version", i32 500}
478