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: opt -mtriple=amdgcn-- -S -structurizecfg -si-annotate-control-flow %s | FileCheck -check-prefix=IR %s
4 ; RUN: llc -march=amdgcn -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
6 ; After structurizing, there are 3 levels of loops. The i1 phi
7 ; conditions mutually depend on each other, so it isn't safe to delete
8 ; the condition that appears to have no uses until the loop is
9 ; completely processed.
11 define amdgpu_kernel void @reduced_nested_loop_conditions(i64 addrspace(3)* nocapture %arg) #0 {
12 ; GCN-LABEL: reduced_nested_loop_conditions:
14 ; GCN-NEXT: s_load_dword s0, s[0:1], 0x9
15 ; GCN-NEXT: v_lshlrev_b32_e32 v0, 3, v0
16 ; GCN-NEXT: s_mov_b32 m0, -1
17 ; GCN-NEXT: s_and_b64 s[2:3], exec, -1
18 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
19 ; GCN-NEXT: v_add_i32_e32 v0, vcc, s0, v0
20 ; GCN-NEXT: ds_read_b64 v[0:1], v0
21 ; GCN-NEXT: s_and_b64 s[0:1], exec, 0
22 ; GCN-NEXT: s_branch .LBB0_2
23 ; GCN-NEXT: .LBB0_1: ; %Flow
24 ; GCN-NEXT: ; in Loop: Header=BB0_2 Depth=1
25 ; GCN-NEXT: s_mov_b64 vcc, s[0:1]
26 ; GCN-NEXT: s_cbranch_vccz .LBB0_4
27 ; GCN-NEXT: .LBB0_2: ; %bb5
28 ; GCN-NEXT: ; =>This Inner Loop Header: Depth=1
29 ; GCN-NEXT: s_mov_b64 s[4:5], -1
30 ; GCN-NEXT: s_mov_b64 vcc, s[2:3]
31 ; GCN-NEXT: s_cbranch_vccz .LBB0_1
32 ; GCN-NEXT: ; %bb.3: ; %bb10
33 ; GCN-NEXT: ; in Loop: Header=BB0_2 Depth=1
34 ; GCN-NEXT: s_mov_b64 s[4:5], 0
35 ; GCN-NEXT: s_branch .LBB0_1
36 ; GCN-NEXT: .LBB0_4: ; %loop.exit.guard
37 ; GCN-NEXT: s_and_b64 vcc, exec, s[4:5]
38 ; GCN-NEXT: s_cbranch_vccz .LBB0_7
39 ; GCN-NEXT: ; %bb.5: ; %bb8
40 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
41 ; GCN-NEXT: ds_read_b32 v0, v0
42 ; GCN-NEXT: s_and_b64 vcc, exec, 0
43 ; GCN-NEXT: .LBB0_6: ; %bb9
44 ; GCN-NEXT: ; =>This Inner Loop Header: Depth=1
45 ; GCN-NEXT: s_mov_b64 vcc, vcc
46 ; GCN-NEXT: s_cbranch_vccz .LBB0_6
47 ; GCN-NEXT: .LBB0_7: ; %DummyReturnBlock
49 ; IR-LABEL: @reduced_nested_loop_conditions(
51 ; IR-NEXT: [[MY_TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() #4
52 ; IR-NEXT: [[MY_TMP1:%.*]] = getelementptr inbounds i64, i64 addrspace(3)* [[ARG:%.*]], i32 [[MY_TMP]]
53 ; IR-NEXT: [[MY_TMP2:%.*]] = load volatile i64, i64 addrspace(3)* [[MY_TMP1]]
54 ; IR-NEXT: br label [[BB5:%.*]]
56 ; IR-NEXT: br i1 true, label [[BB4:%.*]], label [[BB13:%.*]]
58 ; IR-NEXT: br label [[FLOW:%.*]]
60 ; IR-NEXT: [[PHI_BROKEN:%.*]] = phi i64 [ [[TMP6:%.*]], [[BB10:%.*]] ], [ 0, [[BB:%.*]] ]
61 ; IR-NEXT: [[MY_TMP6:%.*]] = phi i32 [ 0, [[BB]] ], [ [[TMP5:%.*]], [[BB10]] ]
62 ; IR-NEXT: [[MY_TMP7:%.*]] = icmp eq i32 [[MY_TMP6]], 1
63 ; IR-NEXT: [[TMP0:%.*]] = call { i1, i64 } @llvm.amdgcn.if.i64(i1 [[MY_TMP7]])
64 ; IR-NEXT: [[TMP1:%.*]] = extractvalue { i1, i64 } [[TMP0]], 0
65 ; IR-NEXT: [[TMP2:%.*]] = extractvalue { i1, i64 } [[TMP0]], 1
66 ; IR-NEXT: br i1 [[TMP1]], label [[BB8:%.*]], label [[FLOW]]
68 ; IR-NEXT: br label [[BB13]]
70 ; IR-NEXT: br i1 false, label [[BB3:%.*]], label [[BB9:%.*]]
72 ; IR-NEXT: [[TMP3:%.*]] = call i1 @llvm.amdgcn.loop.i64(i64 [[TMP6]])
73 ; IR-NEXT: br i1 [[TMP3]], label [[BB23:%.*]], label [[BB5]]
75 ; IR-NEXT: [[TMP4:%.*]] = phi i1 [ [[MY_TMP22:%.*]], [[BB4]] ], [ true, [[BB5]] ]
76 ; IR-NEXT: [[TMP5]] = phi i32 [ [[MY_TMP21:%.*]], [[BB4]] ], [ undef, [[BB5]] ]
77 ; IR-NEXT: call void @llvm.amdgcn.end.cf.i64(i64 [[TMP2]])
78 ; IR-NEXT: [[TMP6]] = call i64 @llvm.amdgcn.if.break.i64(i1 [[TMP4]], i64 [[PHI_BROKEN]])
79 ; IR-NEXT: br label [[BB10]]
81 ; IR-NEXT: [[MY_TMP14:%.*]] = phi i1 [ [[MY_TMP22]], [[BB3]] ], [ true, [[BB8]] ]
82 ; IR-NEXT: [[MY_TMP15:%.*]] = bitcast i64 [[MY_TMP2]] to <2 x i32>
83 ; IR-NEXT: br i1 [[MY_TMP14]], label [[BB16:%.*]], label [[BB20:%.*]]
85 ; IR-NEXT: [[MY_TMP17:%.*]] = extractelement <2 x i32> [[MY_TMP15]], i64 1
86 ; IR-NEXT: [[MY_TMP18:%.*]] = getelementptr inbounds i32, i32 addrspace(3)* undef, i32 [[MY_TMP17]]
87 ; IR-NEXT: [[MY_TMP19:%.*]] = load volatile i32, i32 addrspace(3)* [[MY_TMP18]]
88 ; IR-NEXT: br label [[BB20]]
90 ; IR-NEXT: [[MY_TMP21]] = phi i32 [ [[MY_TMP19]], [[BB16]] ], [ 0, [[BB13]] ]
91 ; IR-NEXT: [[MY_TMP22]] = phi i1 [ false, [[BB16]] ], [ [[MY_TMP14]], [[BB13]] ]
92 ; IR-NEXT: br label [[BB9]]
94 ; IR-NEXT: call void @llvm.amdgcn.end.cf.i64(i64 [[TMP6]])
97 %my.tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
98 %my.tmp1 = getelementptr inbounds i64, i64 addrspace(3)* %arg, i32 %my.tmp
99 %my.tmp2 = load volatile i64, i64 addrspace(3)* %my.tmp1
103 br i1 true, label %bb4, label %bb13
108 bb5: ; preds = %bb10, %bb
109 %my.tmp6 = phi i32 [ 0, %bb ], [ %my.tmp11, %bb10 ]
110 %my.tmp7 = icmp eq i32 %my.tmp6, 1
111 br i1 %my.tmp7, label %bb8, label %bb10
116 bb9: ; preds = %bb20, %bb9
117 br i1 false, label %bb3, label %bb9
119 bb10: ; preds = %bb5, %bb4
120 %my.tmp11 = phi i32 [ %my.tmp21, %bb4 ], [ undef, %bb5 ]
121 %my.tmp12 = phi i1 [ %my.tmp22, %bb4 ], [ true, %bb5 ]
122 br i1 %my.tmp12, label %bb23, label %bb5
124 bb13: ; preds = %bb8, %bb3
125 %my.tmp14 = phi i1 [ %my.tmp22, %bb3 ], [ true, %bb8 ]
126 %my.tmp15 = bitcast i64 %my.tmp2 to <2 x i32>
127 br i1 %my.tmp14, label %bb16, label %bb20
129 bb16: ; preds = %bb13
130 %my.tmp17 = extractelement <2 x i32> %my.tmp15, i64 1
131 %my.tmp18 = getelementptr inbounds i32, i32 addrspace(3)* undef, i32 %my.tmp17
132 %my.tmp19 = load volatile i32, i32 addrspace(3)* %my.tmp18
135 bb20: ; preds = %bb16, %bb13
136 %my.tmp21 = phi i32 [ %my.tmp19, %bb16 ], [ 0, %bb13 ]
137 %my.tmp22 = phi i1 [ false, %bb16 ], [ %my.tmp14, %bb13 ]
140 bb23: ; preds = %bb10
144 ; Earlier version of above, before a run of the structurizer.
146 define amdgpu_kernel void @nested_loop_conditions(i64 addrspace(1)* nocapture %arg) #0 {
147 ; GCN-LABEL: nested_loop_conditions:
148 ; GCN: ; %bb.0: ; %bb
149 ; GCN-NEXT: s_mov_b32 s3, 0xf000
150 ; GCN-NEXT: s_mov_b32 s2, -1
151 ; GCN-NEXT: buffer_load_dword v0, off, s[0:3], 0 glc
152 ; GCN-NEXT: s_waitcnt vmcnt(0)
153 ; GCN-NEXT: v_cmp_lt_i32_e32 vcc, 8, v0
154 ; GCN-NEXT: s_cbranch_vccnz .LBB1_6
155 ; GCN-NEXT: ; %bb.1: ; %bb14.lr.ph
156 ; GCN-NEXT: s_load_dword s4, s[0:1], 0x0
157 ; GCN-NEXT: s_branch .LBB1_3
158 ; GCN-NEXT: .LBB1_2: ; %Flow
159 ; GCN-NEXT: ; in Loop: Header=BB1_3 Depth=1
160 ; GCN-NEXT: s_and_b64 vcc, exec, s[0:1]
161 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
162 ; GCN-NEXT: s_mov_b64 vcc, vcc
163 ; GCN-NEXT: s_cbranch_vccnz .LBB1_6
164 ; GCN-NEXT: .LBB1_3: ; %bb14
165 ; GCN-NEXT: ; =>This Loop Header: Depth=1
166 ; GCN-NEXT: ; Child Loop BB1_4 Depth 2
167 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
168 ; GCN-NEXT: s_cmp_lg_u32 s4, 1
169 ; GCN-NEXT: s_mov_b64 s[0:1], -1
170 ; GCN-NEXT: ; implicit-def: $sgpr4
171 ; GCN-NEXT: s_cbranch_scc1 .LBB1_2
172 ; GCN-NEXT: .LBB1_4: ; %bb18
173 ; GCN-NEXT: ; Parent Loop BB1_3 Depth=1
174 ; GCN-NEXT: ; => This Inner Loop Header: Depth=2
175 ; GCN-NEXT: buffer_load_dword v0, off, s[0:3], 0 glc
176 ; GCN-NEXT: s_waitcnt vmcnt(0)
177 ; GCN-NEXT: v_cmp_lt_i32_e32 vcc, 8, v0
178 ; GCN-NEXT: s_cbranch_vccnz .LBB1_4
179 ; GCN-NEXT: ; %bb.5: ; %bb21
180 ; GCN-NEXT: ; in Loop: Header=BB1_3 Depth=1
181 ; GCN-NEXT: s_load_dword s4, s[0:1], 0x0
182 ; GCN-NEXT: buffer_load_dword v0, off, s[0:3], 0 glc
183 ; GCN-NEXT: s_waitcnt vmcnt(0)
184 ; GCN-NEXT: v_cmp_lt_i32_e64 s[0:1], 8, v0
185 ; GCN-NEXT: s_branch .LBB1_2
186 ; GCN-NEXT: .LBB1_6: ; %bb31
187 ; GCN-NEXT: v_mov_b32_e32 v0, 0
188 ; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0
189 ; GCN-NEXT: s_waitcnt vmcnt(0)
191 ; IR-LABEL: @nested_loop_conditions(
193 ; IR-NEXT: [[MY_TMP1134:%.*]] = load volatile i32, i32 addrspace(1)* undef
194 ; IR-NEXT: [[MY_TMP1235:%.*]] = icmp slt i32 [[MY_TMP1134]], 9
195 ; IR-NEXT: br i1 [[MY_TMP1235]], label [[BB14_LR_PH:%.*]], label [[FLOW:%.*]]
197 ; IR-NEXT: [[MY_TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() #4
198 ; IR-NEXT: [[MY_TMP1:%.*]] = zext i32 [[MY_TMP]] to i64
199 ; IR-NEXT: [[MY_TMP2:%.*]] = getelementptr inbounds i64, i64 addrspace(1)* [[ARG:%.*]], i64 [[MY_TMP1]]
200 ; IR-NEXT: [[MY_TMP3:%.*]] = load i64, i64 addrspace(1)* [[MY_TMP2]], align 16
201 ; IR-NEXT: [[MY_TMP932:%.*]] = load <4 x i32>, <4 x i32> addrspace(1)* undef, align 16
202 ; IR-NEXT: [[MY_TMP1033:%.*]] = extractelement <4 x i32> [[MY_TMP932]], i64 0
203 ; IR-NEXT: br label [[BB14:%.*]]
205 ; IR-NEXT: call void @llvm.amdgcn.end.cf.i64(i64 [[TMP21:%.*]])
206 ; IR-NEXT: [[TMP0:%.*]] = call { i1, i64 } @llvm.amdgcn.if.i64(i1 [[TMP14:%.*]])
207 ; IR-NEXT: [[TMP1:%.*]] = extractvalue { i1, i64 } [[TMP0]], 0
208 ; IR-NEXT: [[TMP2:%.*]] = extractvalue { i1, i64 } [[TMP0]], 1
209 ; IR-NEXT: br i1 [[TMP1]], label [[BB4_BB13_CRIT_EDGE:%.*]], label [[FLOW4:%.*]]
210 ; IR: bb4.bb13_crit_edge:
211 ; IR-NEXT: br label [[FLOW4]]
213 ; IR-NEXT: [[TMP3:%.*]] = phi i1 [ true, [[BB4_BB13_CRIT_EDGE]] ], [ false, [[FLOW3:%.*]] ]
214 ; IR-NEXT: call void @llvm.amdgcn.end.cf.i64(i64 [[TMP2]])
215 ; IR-NEXT: br label [[FLOW]]
217 ; IR-NEXT: br label [[BB31:%.*]]
219 ; IR-NEXT: [[TMP4:%.*]] = phi i1 [ [[TMP3]], [[FLOW4]] ], [ true, [[BB:%.*]] ]
220 ; IR-NEXT: [[TMP5:%.*]] = call { i1, i64 } @llvm.amdgcn.if.i64(i1 [[TMP4]])
221 ; IR-NEXT: [[TMP6:%.*]] = extractvalue { i1, i64 } [[TMP5]], 0
222 ; IR-NEXT: [[TMP7:%.*]] = extractvalue { i1, i64 } [[TMP5]], 1
223 ; IR-NEXT: br i1 [[TMP6]], label [[BB13:%.*]], label [[BB31]]
225 ; IR-NEXT: [[PHI_BROKEN:%.*]] = phi i64 [ [[TMP16:%.*]], [[FLOW1:%.*]] ], [ 0, [[BB14_LR_PH]] ]
226 ; IR-NEXT: [[MY_TMP1037:%.*]] = phi i32 [ [[MY_TMP1033]], [[BB14_LR_PH]] ], [ [[TMP12:%.*]], [[FLOW1]] ]
227 ; IR-NEXT: [[MY_TMP936:%.*]] = phi <4 x i32> [ [[MY_TMP932]], [[BB14_LR_PH]] ], [ [[TMP11:%.*]], [[FLOW1]] ]
228 ; IR-NEXT: [[MY_TMP15:%.*]] = icmp eq i32 [[MY_TMP1037]], 1
229 ; IR-NEXT: [[TMP8:%.*]] = call { i1, i64 } @llvm.amdgcn.if.i64(i1 [[MY_TMP15]])
230 ; IR-NEXT: [[TMP9:%.*]] = extractvalue { i1, i64 } [[TMP8]], 0
231 ; IR-NEXT: [[TMP10:%.*]] = extractvalue { i1, i64 } [[TMP8]], 1
232 ; IR-NEXT: br i1 [[TMP9]], label [[BB16:%.*]], label [[FLOW1]]
234 ; IR-NEXT: [[MY_TMP17:%.*]] = bitcast i64 [[MY_TMP3]] to <2 x i32>
235 ; IR-NEXT: br label [[BB18:%.*]]
237 ; IR-NEXT: [[TMP11]] = phi <4 x i32> [ [[MY_TMP9:%.*]], [[BB21:%.*]] ], [ undef, [[BB14]] ]
238 ; IR-NEXT: [[TMP12]] = phi i32 [ [[MY_TMP10:%.*]], [[BB21]] ], [ undef, [[BB14]] ]
239 ; IR-NEXT: [[TMP13:%.*]] = phi i1 [ [[MY_TMP12:%.*]], [[BB21]] ], [ true, [[BB14]] ]
240 ; IR-NEXT: [[TMP14]] = phi i1 [ [[MY_TMP12]], [[BB21]] ], [ false, [[BB14]] ]
241 ; IR-NEXT: [[TMP15:%.*]] = phi i1 [ false, [[BB21]] ], [ true, [[BB14]] ]
242 ; IR-NEXT: call void @llvm.amdgcn.end.cf.i64(i64 [[TMP10]])
243 ; IR-NEXT: [[TMP16]] = call i64 @llvm.amdgcn.if.break.i64(i1 [[TMP13]], i64 [[PHI_BROKEN]])
244 ; IR-NEXT: [[TMP17:%.*]] = call i1 @llvm.amdgcn.loop.i64(i64 [[TMP16]])
245 ; IR-NEXT: br i1 [[TMP17]], label [[FLOW2:%.*]], label [[BB14]]
247 ; IR-NEXT: [[MY_TMP19:%.*]] = load volatile i32, i32 addrspace(1)* undef
248 ; IR-NEXT: [[MY_TMP20:%.*]] = icmp slt i32 [[MY_TMP19]], 9
249 ; IR-NEXT: br i1 [[MY_TMP20]], label [[BB21]], label [[BB18]]
251 ; IR-NEXT: [[MY_TMP22:%.*]] = extractelement <2 x i32> [[MY_TMP17]], i64 1
252 ; IR-NEXT: [[MY_TMP23:%.*]] = lshr i32 [[MY_TMP22]], 16
253 ; IR-NEXT: [[MY_TMP24:%.*]] = select i1 undef, i32 undef, i32 [[MY_TMP23]]
254 ; IR-NEXT: [[MY_TMP25:%.*]] = uitofp i32 [[MY_TMP24]] to float
255 ; IR-NEXT: [[MY_TMP26:%.*]] = fmul float [[MY_TMP25]], 0x3EF0001000000000
256 ; IR-NEXT: [[MY_TMP27:%.*]] = fsub float [[MY_TMP26]], undef
257 ; IR-NEXT: [[MY_TMP28:%.*]] = fcmp olt float [[MY_TMP27]], 5.000000e-01
258 ; IR-NEXT: [[MY_TMP29:%.*]] = select i1 [[MY_TMP28]], i64 1, i64 2
259 ; IR-NEXT: [[MY_TMP30:%.*]] = extractelement <4 x i32> [[MY_TMP936]], i64 [[MY_TMP29]]
260 ; IR-NEXT: [[MY_TMP7:%.*]] = zext i32 [[MY_TMP30]] to i64
261 ; IR-NEXT: [[MY_TMP8:%.*]] = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* undef, i64 [[MY_TMP7]]
262 ; IR-NEXT: [[MY_TMP9]] = load <4 x i32>, <4 x i32> addrspace(1)* [[MY_TMP8]], align 16
263 ; IR-NEXT: [[MY_TMP10]] = extractelement <4 x i32> [[MY_TMP9]], i64 0
264 ; IR-NEXT: [[MY_TMP11:%.*]] = load volatile i32, i32 addrspace(1)* undef
265 ; IR-NEXT: [[MY_TMP12]] = icmp sge i32 [[MY_TMP11]], 9
266 ; IR-NEXT: br label [[FLOW1]]
268 ; IR-NEXT: call void @llvm.amdgcn.end.cf.i64(i64 [[TMP16]])
269 ; IR-NEXT: [[TMP19:%.*]] = call { i1, i64 } @llvm.amdgcn.if.i64(i1 [[TMP15]])
270 ; IR-NEXT: [[TMP20:%.*]] = extractvalue { i1, i64 } [[TMP19]], 0
271 ; IR-NEXT: [[TMP21]] = extractvalue { i1, i64 } [[TMP19]], 1
272 ; IR-NEXT: br i1 [[TMP20]], label [[BB31_LOOPEXIT:%.*]], label [[FLOW3]]
274 ; IR-NEXT: br label [[FLOW3]]
276 ; IR-NEXT: call void @llvm.amdgcn.end.cf.i64(i64 [[TMP7]])
277 ; IR-NEXT: store volatile i32 0, i32 addrspace(1)* undef
280 %my.tmp1134 = load volatile i32, i32 addrspace(1)* undef
281 %my.tmp1235 = icmp slt i32 %my.tmp1134, 9
282 br i1 %my.tmp1235, label %bb14.lr.ph, label %bb13
284 bb14.lr.ph: ; preds = %bb
285 %my.tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
286 %my.tmp1 = zext i32 %my.tmp to i64
287 %my.tmp2 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i64 %my.tmp1
288 %my.tmp3 = load i64, i64 addrspace(1)* %my.tmp2, align 16
289 %my.tmp932 = load <4 x i32>, <4 x i32> addrspace(1)* undef, align 16
290 %my.tmp1033 = extractelement <4 x i32> %my.tmp932, i64 0
293 bb4.bb13_crit_edge: ; preds = %bb21
296 bb13: ; preds = %bb4.bb13_crit_edge, %bb
299 bb14: ; preds = %bb21, %bb14.lr.ph
300 %my.tmp1037 = phi i32 [ %my.tmp1033, %bb14.lr.ph ], [ %my.tmp10, %bb21 ]
301 %my.tmp936 = phi <4 x i32> [ %my.tmp932, %bb14.lr.ph ], [ %my.tmp9, %bb21 ]
302 %my.tmp15 = icmp eq i32 %my.tmp1037, 1
303 br i1 %my.tmp15, label %bb16, label %bb31.loopexit
305 bb16: ; preds = %bb14
306 %my.tmp17 = bitcast i64 %my.tmp3 to <2 x i32>
309 bb18: ; preds = %bb18, %bb16
310 %my.tmp19 = load volatile i32, i32 addrspace(1)* undef
311 %my.tmp20 = icmp slt i32 %my.tmp19, 9
312 br i1 %my.tmp20, label %bb21, label %bb18
314 bb21: ; preds = %bb18
315 %my.tmp22 = extractelement <2 x i32> %my.tmp17, i64 1
316 %my.tmp23 = lshr i32 %my.tmp22, 16
317 %my.tmp24 = select i1 undef, i32 undef, i32 %my.tmp23
318 %my.tmp25 = uitofp i32 %my.tmp24 to float
319 %my.tmp26 = fmul float %my.tmp25, 0x3EF0001000000000
320 %my.tmp27 = fsub float %my.tmp26, undef
321 %my.tmp28 = fcmp olt float %my.tmp27, 5.000000e-01
322 %my.tmp29 = select i1 %my.tmp28, i64 1, i64 2
323 %my.tmp30 = extractelement <4 x i32> %my.tmp936, i64 %my.tmp29
324 %my.tmp7 = zext i32 %my.tmp30 to i64
325 %my.tmp8 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* undef, i64 %my.tmp7
326 %my.tmp9 = load <4 x i32>, <4 x i32> addrspace(1)* %my.tmp8, align 16
327 %my.tmp10 = extractelement <4 x i32> %my.tmp9, i64 0
328 %my.tmp11 = load volatile i32, i32 addrspace(1)* undef
329 %my.tmp12 = icmp slt i32 %my.tmp11, 9
330 br i1 %my.tmp12, label %bb14, label %bb4.bb13_crit_edge
332 bb31.loopexit: ; preds = %bb14
335 bb31: ; preds = %bb31.loopexit, %bb13
336 store volatile i32 0, i32 addrspace(1)* undef
340 declare i32 @llvm.amdgcn.workitem.id.x() #1
342 attributes #0 = { nounwind }
343 attributes #1 = { nounwind readnone }