1 ; RUN: llc -march=amdgcn -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=CI %s
2 ; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=VI %s
3 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=VI %s
5 ; GCN-LABEL: {{^}}void_func_i1:
6 ; GCN: v_and_b32_e32 v0, 1, v0
7 ; GCN: buffer_store_byte v0, off
8 define void @void_func_i1(i1 %arg0) #0 {
9 store i1 %arg0, i1 addrspace(1)* undef
13 ; GCN-LABEL: {{^}}void_func_i1_zeroext:
15 ; GCN-NEXT: v_or_b32_e32 v0, 12, v0
17 ; GCN: buffer_store_dword v0, off
18 define void @void_func_i1_zeroext(i1 zeroext %arg0) #0 {
19 %ext = zext i1 %arg0 to i32
20 %add = add i32 %ext, 12
21 store i32 %add, i32 addrspace(1)* undef
25 ; GCN-LABEL: {{^}}void_func_i1_signext:
27 ; GCN-NEXT: v_add_i32_e32 v0, vcc, 12, v0
29 ; GCN: buffer_store_dword v0, off
30 define void @void_func_i1_signext(i1 signext %arg0) #0 {
31 %ext = sext i1 %arg0 to i32
32 %add = add i32 %ext, 12
33 store i32 %add, i32 addrspace(1)* undef
37 ; GCN-LABEL: {{^}}i1_arg_i1_use:
38 ; GCN: v_and_b32_e32 v0, 1, v0
39 ; GCN: v_cmp_eq_u32_e32 vcc, 1, v0
40 ; GCN: s_xor_b64 s{{\[[0-9]+:[0-9]+\]}}, vcc, -1
41 define void @i1_arg_i1_use(i1 %arg) #0 {
43 br i1 %arg, label %bb2, label %bb1
46 store volatile i32 0, i32 addrspace(1)* undef
53 ; GCN-LABEL: {{^}}void_func_i8:
55 ; GCN: buffer_store_byte v0, off
56 define void @void_func_i8(i8 %arg0) #0 {
57 store i8 %arg0, i8 addrspace(1)* undef
61 ; GCN-LABEL: {{^}}void_func_i8_zeroext:
63 ; GCN: v_add_i32_e32 v0, vcc, 12, v0
64 define void @void_func_i8_zeroext(i8 zeroext %arg0) #0 {
65 %ext = zext i8 %arg0 to i32
66 %add = add i32 %ext, 12
67 store i32 %add, i32 addrspace(1)* undef
71 ; GCN-LABEL: {{^}}void_func_i8_signext:
73 ; GCN: v_add_i32_e32 v0, vcc, 12, v0
74 define void @void_func_i8_signext(i8 signext %arg0) #0 {
75 %ext = sext i8 %arg0 to i32
76 %add = add i32 %ext, 12
77 store i32 %add, i32 addrspace(1)* undef
81 ; GCN-LABEL: {{^}}void_func_i16:
82 ; GCN: buffer_store_short v0, off
83 define void @void_func_i16(i16 %arg0) #0 {
84 store i16 %arg0, i16 addrspace(1)* undef
88 ; GCN-LABEL: {{^}}void_func_i16_zeroext:
90 ; GCN: v_add_i32_e32 v0, vcc, 12, v0
91 define void @void_func_i16_zeroext(i16 zeroext %arg0) #0 {
92 %ext = zext i16 %arg0 to i32
93 %add = add i32 %ext, 12
94 store i32 %add, i32 addrspace(1)* undef
98 ; GCN-LABEL: {{^}}void_func_i16_signext:
100 ; GCN: v_add_i32_e32 v0, vcc, 12, v0
101 define void @void_func_i16_signext(i16 signext %arg0) #0 {
102 %ext = sext i16 %arg0 to i32
103 %add = add i32 %ext, 12
104 store i32 %add, i32 addrspace(1)* undef
108 ; GCN-LABEL: {{^}}void_func_i32:
110 ; GCN: buffer_store_dword v0, off
111 define void @void_func_i32(i32 %arg0) #0 {
112 store i32 %arg0, i32 addrspace(1)* undef
116 ; GCN-LABEL: {{^}}void_func_i64:
120 ; GCN: buffer_store_dwordx2 v[0:1], off
121 define void @void_func_i64(i64 %arg0) #0 {
122 store i64 %arg0, i64 addrspace(1)* undef
126 ; GCN-LABEL: {{^}}void_func_f16:
128 ; CI: v_cvt_f16_f32_e32 v0, v0
129 ; GCN: buffer_store_short v0, off
130 define void @void_func_f16(half %arg0) #0 {
131 store half %arg0, half addrspace(1)* undef
135 ; GCN-LABEL: {{^}}void_func_f32
137 ; GCN: buffer_store_dword v0, off
138 define void @void_func_f32(float %arg0) #0 {
139 store float %arg0, float addrspace(1)* undef
143 ; GCN-LABEL: {{^}}void_func_f64:
147 ; GCN: buffer_store_dwordx2 v[0:1], off
148 define void @void_func_f64(double %arg0) #0 {
149 store double %arg0, double addrspace(1)* undef
153 ; GCN-LABEL: {{^}}void_func_v2i32:
157 ; GCN: buffer_store_dwordx2 v[0:1], off
158 define void @void_func_v2i32(<2 x i32> %arg0) #0 {
159 store <2 x i32> %arg0, <2 x i32> addrspace(1)* undef
163 ; GCN-LABEL: {{^}}void_func_v3i32:
164 ; GCN-DAG: buffer_store_dword v2, off
165 ; GCN-DAG: buffer_store_dwordx2 v[0:1], off
166 define void @void_func_v3i32(<3 x i32> %arg0) #0 {
167 store <3 x i32> %arg0, <3 x i32> addrspace(1)* undef
171 ; GCN-LABEL: {{^}}void_func_v4i32:
172 ; GCN: buffer_store_dwordx4 v[0:3], off
173 define void @void_func_v4i32(<4 x i32> %arg0) #0 {
174 store <4 x i32> %arg0, <4 x i32> addrspace(1)* undef
178 ; GCN-LABEL: {{^}}void_func_v5i32:
179 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
180 ; GCN-DAG: buffer_store_dword v4, off
181 define void @void_func_v5i32(<5 x i32> %arg0) #0 {
182 store <5 x i32> %arg0, <5 x i32> addrspace(1)* undef
186 ; GCN-LABEL: {{^}}void_func_v8i32:
187 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
188 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
189 define void @void_func_v8i32(<8 x i32> %arg0) #0 {
190 store <8 x i32> %arg0, <8 x i32> addrspace(1)* undef
194 ; GCN-LABEL: {{^}}void_func_v16i32:
195 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
196 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
197 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
198 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
199 define void @void_func_v16i32(<16 x i32> %arg0) #0 {
200 store <16 x i32> %arg0, <16 x i32> addrspace(1)* undef
204 ; GCN-LABEL: {{^}}void_func_v32i32:
205 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
206 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
207 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
208 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
209 ; GCN-DAG: buffer_store_dwordx4 v[16:19], off
210 ; GCN-DAG: buffer_store_dwordx4 v[20:23], off
211 ; GCN-DAG: buffer_store_dwordx4 v[24:27], off
212 ; GCN-DAG: buffer_store_dwordx4 v[28:31], off
213 define void @void_func_v32i32(<32 x i32> %arg0) #0 {
214 store <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
218 ; 1 over register limit
219 ; GCN-LABEL: {{^}}void_func_v33i32:
220 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
221 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
222 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
223 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
224 ; GCN-DAG: buffer_load_dword [[STACKLOAD:v[0-9]+]], off, s[0:3], s5
225 ; GCN-DAG: buffer_store_dwordx4 v[16:19], off
226 ; GCN-DAG: buffer_store_dwordx4 v[20:23], off
227 ; GCN-DAG: buffer_store_dwordx4 v[24:27], off
228 ; GCN-DAG: buffer_store_dwordx4 v[28:31], off
229 ; GCN: buffer_store_dword [[STACKLOAD]], off
230 define void @void_func_v33i32(<33 x i32> %arg0) #0 {
231 store <33 x i32> %arg0, <33 x i32> addrspace(1)* undef
235 ; GCN-LABEL: {{^}}void_func_v2i64:
236 ; GCN: buffer_store_dwordx4 v[0:3], off
237 define void @void_func_v2i64(<2 x i64> %arg0) #0 {
238 store <2 x i64> %arg0, <2 x i64> addrspace(1)* undef
242 ; GCN-LABEL: {{^}}void_func_v3i64:
243 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
244 ; GCN-DAG: buffer_store_dwordx2 v[4:5], off
245 define void @void_func_v3i64(<3 x i64> %arg0) #0 {
246 store <3 x i64> %arg0, <3 x i64> addrspace(1)* undef
250 ; GCN-LABEL: {{^}}void_func_v4i64:
251 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
252 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
253 define void @void_func_v4i64(<4 x i64> %arg0) #0 {
254 store <4 x i64> %arg0, <4 x i64> addrspace(1)* undef
258 ; GCN-LABEL: {{^}}void_func_v5i64:
259 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
260 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
261 ; GCN-DAG: buffer_store_dwordx2 v[8:9], off
262 define void @void_func_v5i64(<5 x i64> %arg0) #0 {
263 store <5 x i64> %arg0, <5 x i64> addrspace(1)* undef
267 ; GCN-LABEL: {{^}}void_func_v8i64:
268 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
269 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
270 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
271 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
272 define void @void_func_v8i64(<8 x i64> %arg0) #0 {
273 store <8 x i64> %arg0, <8 x i64> addrspace(1)* undef
277 ; GCN-LABEL: {{^}}void_func_v16i64:
278 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
279 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
280 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
281 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
282 ; GCN-DAG: buffer_store_dwordx4 v[16:19], off
283 ; GCN-DAG: buffer_store_dwordx4 v[20:23], off
284 ; GCN-DAG: buffer_store_dwordx4 v[24:27], off
285 ; GCN-DAG: buffer_store_dwordx4 v[28:31], off
286 define void @void_func_v16i64(<16 x i64> %arg0) #0 {
287 store <16 x i64> %arg0, <16 x i64> addrspace(1)* undef
291 ; GCN-LABEL: {{^}}void_func_v2i16:
293 ; GFX9: buffer_store_dword v0, off
294 define void @void_func_v2i16(<2 x i16> %arg0) #0 {
295 store <2 x i16> %arg0, <2 x i16> addrspace(1)* undef
299 ; GCN-LABEL: {{^}}void_func_v3i16:
300 ; GCN-DAG: buffer_store_dword v0, off
301 ; GCN-DAG: buffer_store_short v2, off
302 define void @void_func_v3i16(<3 x i16> %arg0) #0 {
303 store <3 x i16> %arg0, <3 x i16> addrspace(1)* undef
307 ; GCN-LABEL: {{^}}void_func_v4i16:
310 ; GFX9: buffer_store_dwordx2 v[0:1], off
311 define void @void_func_v4i16(<4 x i16> %arg0) #0 {
312 store <4 x i16> %arg0, <4 x i16> addrspace(1)* undef
316 ; GCN-LABEL: {{^}}void_func_v5i16:
317 ; GCN-DAG: buffer_store_short v4, off,
318 ; GCN-DAG: buffer_store_dwordx2 v[1:2], off
319 define void @void_func_v5i16(<5 x i16> %arg0) #0 {
320 store <5 x i16> %arg0, <5 x i16> addrspace(1)* undef
324 ; GCN-LABEL: {{^}}void_func_v8i16:
325 ; GFX9-DAG: buffer_store_dwordx4 v[0:3], off
326 define void @void_func_v8i16(<8 x i16> %arg0) #0 {
327 store <8 x i16> %arg0, <8 x i16> addrspace(1)* undef
331 ; GCN-LABEL: {{^}}void_func_v16i16:
332 ; GFX9-DAG: buffer_store_dwordx4 v[0:3], off
333 ; GFX9-DAG: buffer_store_dwordx4 v[4:7], off
334 define void @void_func_v16i16(<16 x i16> %arg0) #0 {
335 store <16 x i16> %arg0, <16 x i16> addrspace(1)* undef
339 ; GCN-LABEL: {{^}}void_func_v2f32:
343 ; GCN: buffer_store_dwordx2 v[0:1], off
344 define void @void_func_v2f32(<2 x float> %arg0) #0 {
345 store <2 x float> %arg0, <2 x float> addrspace(1)* undef
349 ; GCN-LABEL: {{^}}void_func_v3f32:
350 ; GCN-DAG: buffer_store_dword v2, off
351 ; GCN-DAG: buffer_store_dwordx2 v[0:1], off
352 define void @void_func_v3f32(<3 x float> %arg0) #0 {
353 store <3 x float> %arg0, <3 x float> addrspace(1)* undef
357 ; GCN-LABEL: {{^}}void_func_v4f32:
358 ; GCN: buffer_store_dwordx4 v[0:3], off
359 define void @void_func_v4f32(<4 x float> %arg0) #0 {
360 store <4 x float> %arg0, <4 x float> addrspace(1)* undef
364 ; GCN-LABEL: {{^}}void_func_v8f32:
365 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
366 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
367 define void @void_func_v8f32(<8 x float> %arg0) #0 {
368 store <8 x float> %arg0, <8 x float> addrspace(1)* undef
372 ; GCN-LABEL: {{^}}void_func_v16f32:
373 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
374 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
375 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
376 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
377 define void @void_func_v16f32(<16 x float> %arg0) #0 {
378 store <16 x float> %arg0, <16 x float> addrspace(1)* undef
382 ; GCN-LABEL: {{^}}void_func_v2f64:
383 ; GCN: buffer_store_dwordx4 v[0:3], off
384 define void @void_func_v2f64(<2 x double> %arg0) #0 {
385 store <2 x double> %arg0, <2 x double> addrspace(1)* undef
389 ; GCN-LABEL: {{^}}void_func_v3f64:
390 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
391 ; GCN-DAG: buffer_store_dwordx2 v[4:5], off
392 define void @void_func_v3f64(<3 x double> %arg0) #0 {
393 store <3 x double> %arg0, <3 x double> addrspace(1)* undef
397 ; GCN-LABEL: {{^}}void_func_v4f64:
398 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
399 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
400 define void @void_func_v4f64(<4 x double> %arg0) #0 {
401 store <4 x double> %arg0, <4 x double> addrspace(1)* undef
405 ; GCN-LABEL: {{^}}void_func_v8f64:
406 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
407 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
408 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
409 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
410 define void @void_func_v8f64(<8 x double> %arg0) #0 {
411 store <8 x double> %arg0, <8 x double> addrspace(1)* undef
415 ; GCN-LABEL: {{^}}void_func_v16f64:
416 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
417 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
418 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
419 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
420 ; GCN-DAG: buffer_store_dwordx4 v[16:19], off
421 ; GCN-DAG: buffer_store_dwordx4 v[20:23], off
422 ; GCN-DAG: buffer_store_dwordx4 v[24:27], off
423 ; GCN-DAG: buffer_store_dwordx4 v[28:31], off
424 define void @void_func_v16f64(<16 x double> %arg0) #0 {
425 store <16 x double> %arg0, <16 x double> addrspace(1)* undef
429 ; GCN-LABEL: {{^}}void_func_v2f16:
431 ; GFX9: buffer_store_dword v0, off
432 define void @void_func_v2f16(<2 x half> %arg0) #0 {
433 store <2 x half> %arg0, <2 x half> addrspace(1)* undef
437 ; GCN-LABEL: {{^}}void_func_v3f16:
439 ; GCN-DAG: buffer_store_dword v0, off
440 ; GCN-DAG: buffer_store_short v2, off
441 define void @void_func_v3f16(<3 x half> %arg0) #0 {
442 store <3 x half> %arg0, <3 x half> addrspace(1)* undef
446 ; GCN-LABEL: {{^}}void_func_v4f16:
450 ; GFX9: buffer_store_dwordx2 v[0:1], off
451 define void @void_func_v4f16(<4 x half> %arg0) #0 {
452 store <4 x half> %arg0, <4 x half> addrspace(1)* undef
456 ; GCN-LABEL: {{^}}void_func_v8f16:
459 ; GFX9: buffer_store_dwordx4 v[0:3], off
460 define void @void_func_v8f16(<8 x half> %arg0) #0 {
461 store <8 x half> %arg0, <8 x half> addrspace(1)* undef
465 ; GCN-LABEL: {{^}}void_func_v16f16:
468 ; GFX9-DAG: buffer_store_dwordx4 v[0:3], off
469 ; GFX9-DAG: buffer_store_dwordx4 v[4:7], off
470 define void @void_func_v16f16(<16 x half> %arg0) #0 {
471 store <16 x half> %arg0, <16 x half> addrspace(1)* undef
475 ; Make sure there is no alignment requirement for passed vgprs.
476 ; GCN-LABEL: {{^}}void_func_i32_i64_i32:
478 ; GCN: buffer_store_dword v0, off
479 ; GCN: buffer_store_dwordx2 v[1:2]
480 ; GCN: buffer_store_dword v3
481 define void @void_func_i32_i64_i32(i32 %arg0, i64 %arg1, i32 %arg2) #0 {
482 store volatile i32 %arg0, i32 addrspace(1)* undef
483 store volatile i64 %arg1, i64 addrspace(1)* undef
484 store volatile i32 %arg2, i32 addrspace(1)* undef
488 ; GCN-LABEL: {{^}}void_func_struct_i32:
490 ; GCN: buffer_store_dword v0, off
491 define void @void_func_struct_i32({ i32 } %arg0) #0 {
492 store { i32 } %arg0, { i32 } addrspace(1)* undef
496 ; GCN-LABEL: {{^}}void_func_struct_i8_i32:
497 ; GCN-DAG: buffer_store_byte v0, off
498 ; GCN-DAG: buffer_store_dword v1, off
499 define void @void_func_struct_i8_i32({ i8, i32 } %arg0) #0 {
500 store { i8, i32 } %arg0, { i8, i32 } addrspace(1)* undef
504 ; GCN-LABEL: {{^}}void_func_byval_struct_i8_i32:
505 ; GCN-DAG: buffer_load_ubyte v[[ELT0:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
506 ; GCN-DAG: buffer_load_dword v[[ELT1:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
507 ; GCN-DAG: buffer_store_dword v[[ELT1]]
508 ; GCN-DAG: buffer_store_byte v[[ELT0]]
509 define void @void_func_byval_struct_i8_i32({ i8, i32 }* byval %arg0) #0 {
510 %arg0.load = load { i8, i32 }, { i8, i32 }* %arg0
511 store { i8, i32 } %arg0.load, { i8, i32 } addrspace(1)* undef
515 ; GCN-LABEL: {{^}}void_func_byval_struct_i8_i32_x2:
516 ; GCN: buffer_load_ubyte v[[ELT0_0:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
517 ; GCN: buffer_load_dword v[[ELT1_0:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
518 ; GCN: buffer_load_ubyte v[[ELT0_1:[0-9]+]], off, s[0:3], s5 offset:12{{$}}
519 ; GCN: buffer_load_dword v[[ELT1_1:[0-9]+]], off, s[0:3], s5 offset:16{{$}}
521 ; GCN: ds_write_b32 v0, v0
523 define void @void_func_byval_struct_i8_i32_x2({ i8, i32 }* byval %arg0, { i8, i32 }* byval %arg1, i32 %arg2) #0 {
524 %arg0.load = load volatile { i8, i32 }, { i8, i32 }* %arg0
525 %arg1.load = load volatile { i8, i32 }, { i8, i32 }* %arg1
526 store volatile { i8, i32 } %arg0.load, { i8, i32 } addrspace(1)* undef
527 store volatile { i8, i32 } %arg1.load, { i8, i32 } addrspace(1)* undef
528 store volatile i32 %arg2, i32 addrspace(3)* undef
532 ; GCN-LABEL: {{^}}void_func_byval_i32_byval_i64:
533 ; GCN-DAG: buffer_load_dword v[[ARG0_LOAD:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
534 ; GCN-DAG: buffer_load_dword v[[ARG1_LOAD0:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
535 ; GCN-DAG: buffer_load_dword v[[ARG1_LOAD1:[0-9]+]], off, s[0:3], s5 offset:12{{$}}
536 ; GCN-DAG: buffer_store_dword v[[ARG0_LOAD]], off
537 ; GCN-DAG: buffer_store_dwordx2 v{{\[}}[[ARG1_LOAD0]]:[[ARG1_LOAD1]]{{\]}}, off
538 define void @void_func_byval_i32_byval_i64(i32* byval %arg0, i64* byval %arg1) #0 {
539 %arg0.load = load i32, i32* %arg0
540 %arg1.load = load i64, i64* %arg1
541 store i32 %arg0.load, i32 addrspace(1)* undef
542 store i64 %arg1.load, i64 addrspace(1)* undef
546 ; GCN-LABEL: {{^}}void_func_v32i32_i32_i64:
547 ; GCN-DAG: buffer_store_dwordx4 v[0:3], off
548 ; GCN-DAG: buffer_store_dwordx4 v[4:7], off
549 ; GCN-DAG: buffer_store_dwordx4 v[8:11], off
550 ; GCN-DAG: buffer_store_dwordx4 v[12:15], off
551 ; GCN-DAG: buffer_store_dwordx4 v[16:19], off
552 ; GCN-DAG: buffer_store_dwordx4 v[20:23], off
553 ; GCN-DAG: buffer_store_dwordx4 v[24:27], off
554 ; GCN-DAG: buffer_store_dwordx4 v[28:31], off
555 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
556 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_0:[0-9]+]], off, s[0:3], s5 offset:8
557 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_1:[0-9]+]], off, s[0:3], s5 offset:12
559 ; GCN: buffer_store_dword v[[LOAD_ARG1]]
560 ; GCN: buffer_store_dwordx2 v{{\[}}[[LOAD_ARG2_0]]:[[LOAD_ARG2_1]]{{\]}}, off
561 define void @void_func_v32i32_i32_i64(<32 x i32> %arg0, i32 %arg1, i64 %arg2) #0 {
562 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
563 store volatile i32 %arg1, i32 addrspace(1)* undef
564 store volatile i64 %arg2, i64 addrspace(1)* undef
568 ; FIXME: Different ext load types on CI vs. VI
569 ; GCN-LABEL: {{^}}void_func_v32i32_i1_i8_i16:
570 ; GCN-DAG: buffer_load_ubyte [[LOAD_ARG1:v[0-9]+]], off, s[0:3], s5 offset:4{{$}}
571 ; VI-DAG: buffer_load_ushort [[LOAD_ARG2:v[0-9]+]], off, s[0:3], s5 offset:8{{$}}
572 ; VI-DAG: buffer_load_ushort [[LOAD_ARG3:v[0-9]+]], off, s[0:3], s5 offset:12{{$}}
573 ; VI-DAG: buffer_load_ushort [[LOAD_ARG4:v[0-9]+]], off, s[0:3], s5 offset:16{{$}}
575 ; CI-DAG: buffer_load_dword [[LOAD_ARG2:v[0-9]+]], off, s[0:3], s5 offset:8{{$}}
576 ; CI-DAG: buffer_load_dword [[LOAD_ARG3:v[0-9]+]], off, s[0:3], s5 offset:12{{$}}
577 ; CI-DAG: buffer_load_dword [[LOAD_ARG4:v[0-9]+]], off, s[0:3], s5 offset:16{{$}}
579 ; GCN-DAG: v_and_b32_e32 [[TRUNC_ARG1_I1:v[0-9]+]], 1, [[LOAD_ARG1]]
580 ; CI-DAG: v_cvt_f16_f32_e32 [[CVT_ARG4:v[0-9]+]], [[LOAD_ARG4]]
582 ; GCN: buffer_store_byte [[TRUNC_ARG1_I1]], off
583 ; GCN: buffer_store_byte [[LOAD_ARG2]], off
584 ; GCN: buffer_store_short [[LOAD_ARG3]], off
585 ; VI: buffer_store_short [[LOAD_ARG4]], off
587 ; CI: buffer_store_short [[CVT_ARG4]], off
588 define void @void_func_v32i32_i1_i8_i16(<32 x i32> %arg0, i1 %arg1, i8 %arg2, i16 %arg3, half %arg4) #0 {
589 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
590 store volatile i1 %arg1, i1 addrspace(1)* undef
591 store volatile i8 %arg2, i8 addrspace(1)* undef
592 store volatile i16 %arg3, i16 addrspace(1)* undef
593 store volatile half %arg4, half addrspace(1)* undef
597 ; GCN-LABEL: {{^}}void_func_v32i32_v2i32_v2f32:
598 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_0:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
599 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_1:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
600 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_0:[0-9]+]], off, s[0:3], s5 offset:12{{$}}
601 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_1:[0-9]+]], off, s[0:3], s5 offset:16{{$}}
603 ; GCN: buffer_store_dwordx2 v{{\[}}[[LOAD_ARG1_0]]:[[LOAD_ARG1_1]]{{\]}}, off
604 ; GCN: buffer_store_dwordx2 v{{\[}}[[LOAD_ARG2_0]]:[[LOAD_ARG2_1]]{{\]}}, off
605 define void @void_func_v32i32_v2i32_v2f32(<32 x i32> %arg0, <2 x i32> %arg1, <2 x float> %arg2) #0 {
606 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
607 store volatile <2 x i32> %arg1, <2 x i32> addrspace(1)* undef
608 store volatile <2 x float> %arg2, <2 x float> addrspace(1)* undef
612 ; GCN-LABEL: {{^}}void_func_v32i32_v2i16_v2f16:
613 ; GFX9-DAG: buffer_load_dword [[LOAD_ARG1:v[0-9]+]], off, s[0:3], s5 offset:4{{$}}
614 ; GFX9-DAG: buffer_load_dword [[LOAD_ARG2:v[0-9]+]], off, s[0:3], s5 offset:8{{$}}
615 ; GFX9: buffer_store_dword [[LOAD_ARG1]], off
616 ; GFX9: buffer_store_short [[LOAD_ARG2]], off
617 define void @void_func_v32i32_v2i16_v2f16(<32 x i32> %arg0, <2 x i16> %arg1, <2 x half> %arg2) #0 {
618 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
619 store volatile <2 x i16> %arg1, <2 x i16> addrspace(1)* undef
620 store volatile <2 x half> %arg2, <2 x half> addrspace(1)* undef
624 ; GCN-LABEL: {{^}}void_func_v32i32_v2i64_v2f64:
625 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_0:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
626 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_1:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
627 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_2:[0-9]+]], off, s[0:3], s5 offset:12{{$}}
628 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_3:[0-9]+]], off, s[0:3], s5 offset:16{{$}}
630 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_0:[0-9]+]], off, s[0:3], s5 offset:20{{$}}
631 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_1:[0-9]+]], off, s[0:3], s5 offset:24{{$}}
632 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_2:[0-9]+]], off, s[0:3], s5 offset:28{{$}}
633 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_3:[0-9]+]], off, s[0:3], s5 offset:32{{$}}
635 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG1_0]]:[[LOAD_ARG1_3]]{{\]}}, off
636 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG2_0]]:[[LOAD_ARG2_3]]{{\]}}, off
637 define void @void_func_v32i32_v2i64_v2f64(<32 x i32> %arg0, <2 x i64> %arg1, <2 x double> %arg2) #0 {
638 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
639 store volatile <2 x i64> %arg1, <2 x i64> addrspace(1)* undef
640 store volatile <2 x double> %arg2, <2 x double> addrspace(1)* undef
644 ; GCN-LABEL: {{^}}void_func_v32i32_v4i32_v4f32:
645 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_0:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
646 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_1:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
647 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_2:[0-9]+]], off, s[0:3], s5 offset:12{{$}}
648 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_3:[0-9]+]], off, s[0:3], s5 offset:16{{$}}
650 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_0:[0-9]+]], off, s[0:3], s5 offset:20{{$}}
651 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_1:[0-9]+]], off, s[0:3], s5 offset:24{{$}}
652 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_2:[0-9]+]], off, s[0:3], s5 offset:28{{$}}
653 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_3:[0-9]+]], off, s[0:3], s5 offset:32{{$}}
655 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG1_0]]:[[LOAD_ARG1_3]]{{\]}}, off
656 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG2_0]]:[[LOAD_ARG2_3]]{{\]}}, off
657 define void @void_func_v32i32_v4i32_v4f32(<32 x i32> %arg0, <4 x i32> %arg1, <4 x float> %arg2) #0 {
658 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
659 store volatile <4 x i32> %arg1, <4 x i32> addrspace(1)* undef
660 store volatile <4 x float> %arg2, <4 x float> addrspace(1)* undef
664 ; GCN-LABEL: {{^}}void_func_v32i32_v8i32_v8f32:
665 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_0:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
666 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_1:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
667 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_2:[0-9]+]], off, s[0:3], s5 offset:12{{$}}
668 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_3:[0-9]+]], off, s[0:3], s5 offset:16{{$}}
669 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_4:[0-9]+]], off, s[0:3], s5 offset:20{{$}}
670 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_5:[0-9]+]], off, s[0:3], s5 offset:24{{$}}
671 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_6:[0-9]+]], off, s[0:3], s5 offset:28{{$}}
672 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_7:[0-9]+]], off, s[0:3], s5 offset:32{{$}}
674 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_0:[0-9]+]], off, s[0:3], s5 offset:36{{$}}
675 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_1:[0-9]+]], off, s[0:3], s5 offset:40{{$}}
676 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_2:[0-9]+]], off, s[0:3], s5 offset:44{{$}}
677 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_3:[0-9]+]], off, s[0:3], s5 offset:48{{$}}
678 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_4:[0-9]+]], off, s[0:3], s5 offset:52{{$}}
679 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_5:[0-9]+]], off, s[0:3], s5 offset:56{{$}}
680 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_6:[0-9]+]], off, s[0:3], s5 offset:60{{$}}
681 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_7:[0-9]+]], off, s[0:3], s5 offset:64{{$}}
683 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG1_4]]:[[LOAD_ARG1_7]]{{\]}}, off
684 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG1_0]]:[[LOAD_ARG1_3]]{{\]}}, off
685 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG2_4]]:[[LOAD_ARG2_7]]{{\]}}, off
686 ; GCN: buffer_store_dwordx4 v{{\[}}[[LOAD_ARG2_0]]:[[LOAD_ARG2_3]]{{\]}}, off
687 define void @void_func_v32i32_v8i32_v8f32(<32 x i32> %arg0, <8 x i32> %arg1, <8 x float> %arg2) #0 {
688 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
689 store volatile <8 x i32> %arg1, <8 x i32> addrspace(1)* undef
690 store volatile <8 x float> %arg2, <8 x float> addrspace(1)* undef
694 ; GCN-LABEL: {{^}}void_func_v32i32_v16i32_v16f32:
695 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_0:[0-9]+]], off, s[0:3], s5 offset:4{{$}}
696 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_1:[0-9]+]], off, s[0:3], s5 offset:8{{$}}
697 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_2:[0-9]+]], off, s[0:3], s5 offset:12{{$}}
698 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_:[0-9]+]], off, s[0:3], s5 offset:16{{$}}
699 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_4:[0-9]+]], off, s[0:3], s5 offset:20{{$}}
700 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_5:[0-9]+]], off, s[0:3], s5 offset:24{{$}}
701 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_6:[0-9]+]], off, s[0:3], s5 offset:28{{$}}
702 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_7:[0-9]+]], off, s[0:3], s5 offset:32{{$}}
703 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_8:[0-9]+]], off, s[0:3], s5 offset:36{{$}}
704 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_9:[0-9]+]], off, s[0:3], s5 offset:40{{$}}
705 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_10:[0-9]+]], off, s[0:3], s5 offset:44{{$}}
706 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_11:[0-9]+]], off, s[0:3], s5 offset:48{{$}}
707 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_12:[0-9]+]], off, s[0:3], s5 offset:52{{$}}
708 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_13:[0-9]+]], off, s[0:3], s5 offset:56{{$}}
709 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG1_14:[0-9]+]], off, s[0:3], s5 offset:60{{$}}
710 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_15:[0-9]+]], off, s[0:3], s5 offset:64{{$}}
712 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_0:[0-9]+]], off, s[0:3], s5 offset:68{{$}}
713 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_1:[0-9]+]], off, s[0:3], s5 offset:72{{$}}
714 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_2:[0-9]+]], off, s[0:3], s5 offset:76{{$}}
715 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_3:[0-9]+]], off, s[0:3], s5 offset:80{{$}}
716 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_4:[0-9]+]], off, s[0:3], s5 offset:84{{$}}
717 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_5:[0-9]+]], off, s[0:3], s5 offset:88{{$}}
718 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_6:[0-9]+]], off, s[0:3], s5 offset:92{{$}}
719 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_7:[0-9]+]], off, s[0:3], s5 offset:96{{$}}
720 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_8:[0-9]+]], off, s[0:3], s5 offset:100{{$}}
721 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_9:[0-9]+]], off, s[0:3], s5 offset:104{{$}}
722 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_10:[0-9]+]], off, s[0:3], s5 offset:108{{$}}
723 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_11:[0-9]+]], off, s[0:3], s5 offset:112{{$}}
724 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_12:[0-9]+]], off, s[0:3], s5 offset:116{{$}}
725 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_13:[0-9]+]], off, s[0:3], s5 offset:120{{$}}
726 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_14:[0-9]+]], off, s[0:3], s5 offset:124{{$}}
727 ; GCN-DAG: buffer_load_dword v[[LOAD_ARG2_15:[0-9]+]], off, s[0:3], s5 offset:128{{$}}
728 define void @void_func_v32i32_v16i32_v16f32(<32 x i32> %arg0, <16 x i32> %arg1, <16 x float> %arg2) #0 {
729 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
730 store volatile <16 x i32> %arg1, <16 x i32> addrspace(1)* undef
731 store volatile <16 x float> %arg2, <16 x float> addrspace(1)* undef
735 ; Check there is no crash.
736 ; GCN-LABEL: {{^}}void_func_v16i8:
737 define void @void_func_v16i8(<16 x i8> %arg0) #0 {
738 store volatile <16 x i8> %arg0, <16 x i8> addrspace(1)* undef
742 ; Check there is no crash.
743 ; GCN-LABEL: {{^}}void_func_v32i32_v16i8:
744 define void @void_func_v32i32_v16i8(<32 x i32> %arg0, <16 x i8> %arg1) #0 {
745 store volatile <32 x i32> %arg0, <32 x i32> addrspace(1)* undef
746 store volatile <16 x i8> %arg1, <16 x i8> addrspace(1)* undef
750 attributes #0 = { nounwind }