1 ; RUN: llc -mtriple=amdgcn-amd- -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck --check-prefixes=GCN,GFX8,GFX89 %s
2 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck --check-prefixes=GCN,GFX8,GFX89 %s
3 ; RUN: llc -mtriple=amdgcn-amd- -mcpu=gfx900 -verify-machineinstrs -amdgpu-enable-global-sgpr-addr < %s | FileCheck --check-prefixes=GCN,GFX9,GFX89 %s
4 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs -amdgpu-enable-global-sgpr-addr < %s | FileCheck --check-prefixes=GCN,GFX9,GFX89 %s
6 declare i32 @llvm.amdgcn.workitem.id.x()
8 ; GCN-LABEL: {{^}}system_unordered:
9 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
10 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
11 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
12 ; GFX89-NOT: buffer_wbinvl1_vol
13 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
14 define amdgpu_kernel void @system_unordered(
15 i32* %in, i32* %out) {
17 %val = load atomic i32, i32* %in unordered, align 4
18 store i32 %val, i32* %out
22 ; GCN-LABEL: {{^}}system_monotonic:
23 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
24 ; GFX89: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
25 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
26 ; GFX89-NOT: buffer_wbinvl1_vol
27 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
28 define amdgpu_kernel void @system_monotonic(
29 i32* %in, i32* %out) {
31 %val = load atomic i32, i32* %in monotonic, align 4
32 store i32 %val, i32* %out
36 ; GCN-LABEL: {{^}}system_acquire:
37 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
38 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
39 ; GCN-NEXT: s_waitcnt vmcnt(0){{$}}
40 ; GFX89-NEXT: buffer_wbinvl1_vol
41 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
42 define amdgpu_kernel void @system_acquire(
43 i32* %in, i32* %out) {
45 %val = load atomic i32, i32* %in acquire, align 4
46 store i32 %val, i32* %out
50 ; GCN-LABEL: {{^}}system_seq_cst:
51 ; GCN: s_waitcnt vmcnt(0){{$}}
52 ; GCN-NEXT: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
53 ; GCN-NEXT: s_waitcnt vmcnt(0){{$}}
54 ; GFX89-NEXT: buffer_wbinvl1_vol
55 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
56 define amdgpu_kernel void @system_seq_cst(
57 i32* %in, i32* %out) {
59 %val = load atomic i32, i32* %in seq_cst, align 4
60 store i32 %val, i32* %out
64 ; GCN-LABEL: {{^}}singlethread_unordered:
65 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
66 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
67 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
68 ; GFX89-NOT: buffer_wbinvl1_vol
69 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
70 define amdgpu_kernel void @singlethread_unordered(
71 i32* %in, i32* %out) {
73 %val = load atomic i32, i32* %in syncscope("singlethread") unordered, align 4
74 store i32 %val, i32* %out
78 ; GCN-LABEL: {{^}}singlethread_monotonic:
79 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
80 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
81 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
82 ; GFX89-NOT: buffer_wbinvl1_vol
83 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
84 define amdgpu_kernel void @singlethread_monotonic(
85 i32* %in, i32* %out) {
87 %val = load atomic i32, i32* %in syncscope("singlethread") monotonic, align 4
88 store i32 %val, i32* %out
92 ; GCN-LABEL: {{^}}singlethread_acquire:
93 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
94 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
95 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
96 ; GFX89-NOT: buffer_wbinvl1_vol
97 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
98 define amdgpu_kernel void @singlethread_acquire(
99 i32* %in, i32* %out) {
101 %val = load atomic i32, i32* %in syncscope("singlethread") acquire, align 4
102 store i32 %val, i32* %out
106 ; GCN-LABEL: {{^}}singlethread_seq_cst:
107 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
108 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
109 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
110 ; GFX89-NOT: buffer_wbinvl1_vol
111 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
112 define amdgpu_kernel void @singlethread_seq_cst(
113 i32* %in, i32* %out) {
115 %val = load atomic i32, i32* %in syncscope("singlethread") seq_cst, align 4
116 store i32 %val, i32* %out
120 ; GCN-LABEL: {{^}}agent_unordered:
121 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
122 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
123 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
124 ; GFX89-NOT: buffer_wbinvl1_vol
125 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
126 define amdgpu_kernel void @agent_unordered(
127 i32* %in, i32* %out) {
129 %val = load atomic i32, i32* %in syncscope("agent") unordered, align 4
130 store i32 %val, i32* %out
134 ; GCN-LABEL: {{^}}agent_monotonic:
135 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
136 ; GFX89: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
137 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
138 ; GFX89-NOT: buffer_wbinvl1_vol
139 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
140 define amdgpu_kernel void @agent_monotonic(
141 i32* %in, i32* %out) {
143 %val = load atomic i32, i32* %in syncscope("agent") monotonic, align 4
144 store i32 %val, i32* %out
148 ; GCN-LABEL: {{^}}agent_acquire:
149 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
150 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
151 ; GCN-NEXT: s_waitcnt vmcnt(0){{$}}
152 ; GFX89-NEXT: buffer_wbinvl1_vol
153 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
154 define amdgpu_kernel void @agent_acquire(
155 i32* %in, i32* %out) {
157 %val = load atomic i32, i32* %in syncscope("agent") acquire, align 4
158 store i32 %val, i32* %out
162 ; GCN-LABEL: {{^}}agent_seq_cst:
163 ; GCN: s_waitcnt vmcnt(0){{$}}
164 ; GCN-NEXT: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
165 ; GCN-NEXT: s_waitcnt vmcnt(0){{$}}
166 ; GFX89-NEXT: buffer_wbinvl1_vol
167 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
168 define amdgpu_kernel void @agent_seq_cst(
169 i32* %in, i32* %out) {
171 %val = load atomic i32, i32* %in syncscope("agent") seq_cst, align 4
172 store i32 %val, i32* %out
176 ; GCN-LABEL: {{^}}workgroup_unordered:
177 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
178 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
179 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
180 ; GFX89-NOT: buffer_wbinvl1_vol
181 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
182 define amdgpu_kernel void @workgroup_unordered(
183 i32* %in, i32* %out) {
185 %val = load atomic i32, i32* %in syncscope("workgroup") unordered, align 4
186 store i32 %val, i32* %out
190 ; GCN-LABEL: {{^}}workgroup_monotonic:
191 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
192 ; GFX89: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
193 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
194 ; GFX89-NOT: buffer_wbinvl1_vol
195 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
196 define amdgpu_kernel void @workgroup_monotonic(
197 i32* %in, i32* %out) {
199 %val = load atomic i32, i32* %in syncscope("workgroup") monotonic, align 4
200 store i32 %val, i32* %out
204 ; GCN-LABEL: {{^}}workgroup_acquire:
205 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
206 ; GFX89: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
207 ; GFX89-NOT: s_waitcnt vmcnt(0){{$}}
208 ; GFX89-NOT: buffer_wbinvl1_vol
209 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
210 define amdgpu_kernel void @workgroup_acquire(
211 i32* %in, i32* %out) {
213 %val = load atomic i32, i32* %in syncscope("workgroup") acquire, align 4
214 store i32 %val, i32* %out
218 ; GCN-LABEL: {{^}}workgroup_seq_cst:
219 ; GFX89-NOT: s_waitcnt vmcnt(0){{$}}
220 ; GFX89: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
221 ; GFX89-NOT: s_waitcnt vmcnt(0){{$}}
222 ; GFX89-NOT: buffer_wbinvl1_vol
223 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
224 define amdgpu_kernel void @workgroup_seq_cst(
225 i32* %in, i32* %out) {
227 %val = load atomic i32, i32* %in syncscope("workgroup") seq_cst, align 4
228 store i32 %val, i32* %out
232 ; GCN-LABEL: {{^}}wavefront_unordered:
233 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
234 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
235 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
236 ; GFX89-NOT: buffer_wbinvl1_vol
237 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
238 define amdgpu_kernel void @wavefront_unordered(
239 i32* %in, i32* %out) {
241 %val = load atomic i32, i32* %in syncscope("wavefront") unordered, align 4
242 store i32 %val, i32* %out
246 ; GCN-LABEL: {{^}}wavefront_monotonic:
247 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
248 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
249 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
250 ; GFX89-NOT: buffer_wbinvl1_vol
251 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
252 define amdgpu_kernel void @wavefront_monotonic(
253 i32* %in, i32* %out) {
255 %val = load atomic i32, i32* %in syncscope("wavefront") monotonic, align 4
256 store i32 %val, i32* %out
260 ; GCN-LABEL: {{^}}wavefront_acquire:
261 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
262 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
263 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
264 ; GFX89-NOT: buffer_wbinvl1_vol
265 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
266 define amdgpu_kernel void @wavefront_acquire(
267 i32* %in, i32* %out) {
269 %val = load atomic i32, i32* %in syncscope("wavefront") acquire, align 4
270 store i32 %val, i32* %out
274 ; GCN-LABEL: {{^}}wavefront_seq_cst:
275 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
276 ; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
277 ; GCN-NOT: s_waitcnt vmcnt(0){{$}}
278 ; GFX89-NOT: buffer_wbinvl1_vol
279 ; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
280 define amdgpu_kernel void @wavefront_seq_cst(
281 i32* %in, i32* %out) {
283 %val = load atomic i32, i32* %in syncscope("wavefront") seq_cst, align 4
284 store i32 %val, i32* %out
288 ; GCN-LABEL: {{^}}nontemporal_private_0:
289 ; GFX89: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], s{{[0-9]+}} offen glc slc{{$}}
290 define amdgpu_kernel void @nontemporal_private_0(
291 i32 addrspace(5)* %in, i32* %out) {
293 %val = load i32, i32 addrspace(5)* %in, align 4, !nontemporal !0
294 store i32 %val, i32* %out
298 ; GCN-LABEL: {{^}}nontemporal_private_1:
299 ; GFX89: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], s{{[0-9]+}} offen glc slc{{$}}
300 define amdgpu_kernel void @nontemporal_private_1(
301 i32 addrspace(5)* %in, i32* %out) {
303 %tid = call i32 @llvm.amdgcn.workitem.id.x()
304 %val.gep = getelementptr inbounds i32, i32 addrspace(5)* %in, i32 %tid
305 %val = load i32, i32 addrspace(5)* %val.gep, align 4, !nontemporal !0
306 store i32 %val, i32* %out
310 ; GCN-LABEL: {{^}}nontemporal_global_0:
311 ; GCN: s_load_dword s{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0x0{{$}}
312 define amdgpu_kernel void @nontemporal_global_0(
313 i32 addrspace(1)* %in, i32* %out) {
315 %val = load i32, i32 addrspace(1)* %in, align 4, !nontemporal !0
316 store i32 %val, i32* %out
320 ; GCN-LABEL: {{^}}nontemporal_global_1:
321 ; GFX8: flat_load_dword v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}] glc slc{{$}}
322 ; GFX9: global_load_dword v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}] glc slc{{$}}
323 define amdgpu_kernel void @nontemporal_global_1(
324 i32 addrspace(1)* %in, i32* %out) {
326 %tid = call i32 @llvm.amdgcn.workitem.id.x()
327 %val.gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %tid
328 %val = load i32, i32 addrspace(1)* %val.gep, align 4, !nontemporal !0
329 store i32 %val, i32* %out
333 ; GCN-LABEL: {{^}}nontemporal_local_0:
334 ; GCN: ds_read_b32 v{{[0-9]+}}, v{{[0-9]+}}{{$}}
335 define amdgpu_kernel void @nontemporal_local_0(
336 i32 addrspace(3)* %in, i32* %out) {
338 %val = load i32, i32 addrspace(3)* %in, align 4, !nontemporal !0
339 store i32 %val, i32* %out
343 ; GCN-LABEL: {{^}}nontemporal_local_1:
344 ; GCN: ds_read_b32 v{{[0-9]+}}, v{{[0-9]+}}{{$}}
345 define amdgpu_kernel void @nontemporal_local_1(
346 i32 addrspace(3)* %in, i32* %out) {
348 %tid = call i32 @llvm.amdgcn.workitem.id.x()
349 %val.gep = getelementptr inbounds i32, i32 addrspace(3)* %in, i32 %tid
350 %val = load i32, i32 addrspace(3)* %val.gep, align 4, !nontemporal !0
351 store i32 %val, i32* %out
355 ; GCN-LABEL: {{^}}nontemporal_flat_0:
356 ; GFX89: flat_load_dword v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}] glc slc{{$}}
357 define amdgpu_kernel void @nontemporal_flat_0(
358 i32* %in, i32* %out) {
360 %val = load i32, i32* %in, align 4, !nontemporal !0
361 store i32 %val, i32* %out
365 ; GCN-LABEL: {{^}}nontemporal_flat_1:
366 ; GFX89: flat_load_dword v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}] glc slc{{$}}
367 define amdgpu_kernel void @nontemporal_flat_1(
368 i32* %in, i32* %out) {
370 %tid = call i32 @llvm.amdgcn.workitem.id.x()
371 %val.gep = getelementptr inbounds i32, i32* %in, i32 %tid
372 %val = load i32, i32* %val.gep, align 4, !nontemporal !0
373 store i32 %val, i32* %out