150
|
1 ; RUN: llc -march=amdgcn -mtriple=amdgcn-- -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=GCN -check-prefix=FUNC %s
|
|
2 ; RUN: llc -march=r600 -mtriple=r600-- -mcpu=redwood -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=R600 -check-prefix=FUNC %s
|
|
3
|
|
4 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
|
|
5
|
|
6 ; FUNC-LABEL: {{^}}setcc_v2i32:
|
|
7 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW]}}, KC0[3].X, KC0[3].Z
|
|
8 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW]}}, KC0[2].W, KC0[3].Y
|
|
9
|
|
10 ; GCN: v_cmp_eq_u32_e32
|
|
11 ; GCN: v_cmp_eq_u32_e32
|
|
12 define amdgpu_kernel void @setcc_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32> %b) #0 {
|
|
13 %result = icmp eq <2 x i32> %a, %b
|
|
14 %sext = sext <2 x i1> %result to <2 x i32>
|
|
15 store <2 x i32> %sext, <2 x i32> addrspace(1)* %out
|
|
16 ret void
|
|
17 }
|
|
18
|
|
19 ; FUNC-LABEL: {{^}}setcc_v4i32:
|
|
20 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
|
21 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
|
22 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
|
23 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
|
24
|
|
25 ; GCN: v_cmp_eq_u32_e32
|
|
26 ; GCN: v_cmp_eq_u32_e32
|
|
27 ; GCN: v_cmp_eq_u32_e32
|
|
28 ; GCN: v_cmp_eq_u32_e32
|
|
29 define amdgpu_kernel void @setcc_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) #0 {
|
|
30 %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
|
|
31 %a = load <4 x i32>, <4 x i32> addrspace(1)* %in
|
|
32 %b = load <4 x i32>, <4 x i32> addrspace(1)* %b_ptr
|
|
33 %result = icmp eq <4 x i32> %a, %b
|
|
34 %sext = sext <4 x i1> %result to <4 x i32>
|
|
35 store <4 x i32> %sext, <4 x i32> addrspace(1)* %out
|
|
36 ret void
|
|
37 }
|
|
38
|
|
39 ;;;==========================================================================;;;
|
|
40 ;; Float comparisons
|
|
41 ;;;==========================================================================;;;
|
|
42
|
|
43 ; FUNC-LABEL: {{^}}f32_oeq:
|
|
44 ; R600: SETE_DX10
|
|
45 ; GCN: v_cmp_eq_f32
|
|
46 define amdgpu_kernel void @f32_oeq(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
47 entry:
|
|
48 %0 = fcmp oeq float %a, %b
|
|
49 %1 = sext i1 %0 to i32
|
|
50 store i32 %1, i32 addrspace(1)* %out
|
|
51 ret void
|
|
52 }
|
|
53
|
|
54 ; FUNC-LABEL: {{^}}f32_ogt:
|
|
55 ; R600: SETGT_DX10
|
|
56 ; GCN: v_cmp_gt_f32
|
|
57 define amdgpu_kernel void @f32_ogt(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
58 entry:
|
|
59 %0 = fcmp ogt float %a, %b
|
|
60 %1 = sext i1 %0 to i32
|
|
61 store i32 %1, i32 addrspace(1)* %out
|
|
62 ret void
|
|
63 }
|
|
64
|
|
65 ; FUNC-LABEL: {{^}}f32_oge:
|
|
66 ; R600: SETGE_DX10
|
|
67 ; GCN: v_cmp_ge_f32
|
|
68 define amdgpu_kernel void @f32_oge(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
69 entry:
|
|
70 %0 = fcmp oge float %a, %b
|
|
71 %1 = sext i1 %0 to i32
|
|
72 store i32 %1, i32 addrspace(1)* %out
|
|
73 ret void
|
|
74 }
|
|
75
|
|
76 ; FUNC-LABEL: {{^}}f32_olt:
|
|
77 ; R600: SETGT_DX10
|
|
78 ; GCN: v_cmp_lt_f32
|
|
79 define amdgpu_kernel void @f32_olt(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
80 entry:
|
|
81 %0 = fcmp olt float %a, %b
|
|
82 %1 = sext i1 %0 to i32
|
|
83 store i32 %1, i32 addrspace(1)* %out
|
|
84 ret void
|
|
85 }
|
|
86
|
|
87 ; FUNC-LABEL: {{^}}f32_ole:
|
|
88 ; R600: SETGE_DX10
|
|
89 ; GCN: v_cmp_le_f32
|
|
90 define amdgpu_kernel void @f32_ole(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
91 entry:
|
|
92 %0 = fcmp ole float %a, %b
|
|
93 %1 = sext i1 %0 to i32
|
|
94 store i32 %1, i32 addrspace(1)* %out
|
|
95 ret void
|
|
96 }
|
|
97
|
|
98 ; FUNC-LABEL: {{^}}f32_one:
|
|
99 ; R600-DAG: SETE_DX10
|
|
100 ; R600-DAG: SETE_DX10
|
|
101 ; R600-DAG: AND_INT
|
|
102 ; R600-DAG: SETNE_DX10
|
|
103 ; R600-DAG: AND_INT
|
|
104 ; R600-DAG: SETNE_INT
|
|
105
|
|
106 ; GCN: v_cmp_lg_f32_e32 vcc
|
|
107 ; GCN-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
|
|
108 define amdgpu_kernel void @f32_one(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
109 entry:
|
|
110 %0 = fcmp one float %a, %b
|
|
111 %1 = sext i1 %0 to i32
|
|
112 store i32 %1, i32 addrspace(1)* %out
|
|
113 ret void
|
|
114 }
|
|
115
|
|
116 ; FUNC-LABEL: {{^}}f32_ord:
|
|
117 ; R600-DAG: SETE_DX10
|
|
118 ; R600-DAG: SETE_DX10
|
|
119 ; R600-DAG: AND_INT
|
|
120 ; R600-DAG: SETNE_INT
|
|
121 ; GCN: v_cmp_o_f32
|
|
122 define amdgpu_kernel void @f32_ord(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
123 entry:
|
|
124 %0 = fcmp ord float %a, %b
|
|
125 %1 = sext i1 %0 to i32
|
|
126 store i32 %1, i32 addrspace(1)* %out
|
|
127 ret void
|
|
128 }
|
|
129
|
|
130 ; FUNC-LABEL: {{^}}f32_ueq:
|
|
131 ; R600-DAG: SETNE_DX10
|
|
132 ; R600-DAG: SETNE_DX10
|
|
133 ; R600-DAG: OR_INT
|
|
134 ; R600-DAG: SETE_DX10
|
|
135 ; R600-DAG: OR_INT
|
|
136 ; R600-DAG: SETNE_INT
|
|
137
|
|
138 ; GCN: v_cmp_nlg_f32_e32 vcc
|
|
139 ; GCN-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
|
|
140 define amdgpu_kernel void @f32_ueq(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
141 entry:
|
|
142 %0 = fcmp ueq float %a, %b
|
|
143 %1 = sext i1 %0 to i32
|
|
144 store i32 %1, i32 addrspace(1)* %out
|
|
145 ret void
|
|
146 }
|
|
147
|
|
148 ; FUNC-LABEL: {{^}}f32_ugt:
|
|
149 ; R600: SETGE
|
|
150 ; R600: SETE_DX10
|
|
151 ; GCN: v_cmp_nle_f32_e32 vcc
|
|
152 ; GCN-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
|
|
153 define amdgpu_kernel void @f32_ugt(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
154 entry:
|
|
155 %0 = fcmp ugt float %a, %b
|
|
156 %1 = sext i1 %0 to i32
|
|
157 store i32 %1, i32 addrspace(1)* %out
|
|
158 ret void
|
|
159 }
|
|
160
|
|
161 ; FUNC-LABEL: {{^}}f32_uge:
|
|
162 ; R600: SETGT
|
|
163 ; R600: SETE_DX10
|
|
164
|
|
165 ; GCN: v_cmp_nlt_f32_e32 vcc
|
|
166 ; GCN-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
|
|
167 define amdgpu_kernel void @f32_uge(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
168 entry:
|
|
169 %0 = fcmp uge float %a, %b
|
|
170 %1 = sext i1 %0 to i32
|
|
171 store i32 %1, i32 addrspace(1)* %out
|
|
172 ret void
|
|
173 }
|
|
174
|
|
175 ; FUNC-LABEL: {{^}}f32_ult:
|
|
176 ; R600: SETGE
|
|
177 ; R600: SETE_DX10
|
|
178
|
|
179 ; GCN: v_cmp_nge_f32_e32 vcc
|
|
180 ; GCN-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
|
|
181 define amdgpu_kernel void @f32_ult(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
182 entry:
|
|
183 %0 = fcmp ult float %a, %b
|
|
184 %1 = sext i1 %0 to i32
|
|
185 store i32 %1, i32 addrspace(1)* %out
|
|
186 ret void
|
|
187 }
|
|
188
|
|
189 ; FUNC-LABEL: {{^}}f32_ule:
|
|
190 ; R600: SETGT
|
|
191 ; R600: SETE_DX10
|
|
192
|
|
193 ; GCN: v_cmp_ngt_f32_e32 vcc
|
|
194 ; GCN-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
|
|
195 define amdgpu_kernel void @f32_ule(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
196 entry:
|
|
197 %0 = fcmp ule float %a, %b
|
|
198 %1 = sext i1 %0 to i32
|
|
199 store i32 %1, i32 addrspace(1)* %out
|
|
200 ret void
|
|
201 }
|
|
202
|
|
203 ; FUNC-LABEL: {{^}}f32_une:
|
|
204 ; R600: SETNE_DX10
|
|
205 ; GCN: v_cmp_neq_f32
|
|
206 define amdgpu_kernel void @f32_une(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
207 entry:
|
|
208 %0 = fcmp une float %a, %b
|
|
209 %1 = sext i1 %0 to i32
|
|
210 store i32 %1, i32 addrspace(1)* %out
|
|
211 ret void
|
|
212 }
|
|
213
|
|
214 ; FUNC-LABEL: {{^}}f32_uno:
|
|
215 ; R600: SETNE_DX10
|
|
216 ; R600: SETNE_DX10
|
|
217 ; R600: OR_INT
|
|
218 ; R600: SETNE_INT
|
|
219 ; GCN: v_cmp_u_f32
|
|
220 define amdgpu_kernel void @f32_uno(i32 addrspace(1)* %out, float %a, float %b) #0 {
|
|
221 entry:
|
|
222 %0 = fcmp uno float %a, %b
|
|
223 %1 = sext i1 %0 to i32
|
|
224 store i32 %1, i32 addrspace(1)* %out
|
|
225 ret void
|
|
226 }
|
|
227
|
|
228 ;;;==========================================================================;;;
|
|
229 ;; 32-bit integer comparisons
|
|
230 ;;;==========================================================================;;;
|
|
231
|
|
232 ; FUNC-LABEL: {{^}}i32_eq:
|
|
233 ; R600: SETE_INT
|
|
234 ; GCN: v_cmp_eq_u32
|
|
235 define amdgpu_kernel void @i32_eq(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
236 entry:
|
|
237 %0 = icmp eq i32 %a, %b
|
|
238 %1 = sext i1 %0 to i32
|
|
239 store i32 %1, i32 addrspace(1)* %out
|
|
240 ret void
|
|
241 }
|
|
242
|
|
243 ; FUNC-LABEL: {{^}}i32_ne:
|
|
244 ; R600: SETNE_INT
|
|
245 ; GCN: v_cmp_ne_u32
|
|
246 define amdgpu_kernel void @i32_ne(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
247 entry:
|
|
248 %0 = icmp ne i32 %a, %b
|
|
249 %1 = sext i1 %0 to i32
|
|
250 store i32 %1, i32 addrspace(1)* %out
|
|
251 ret void
|
|
252 }
|
|
253
|
|
254 ; FUNC-LABEL: {{^}}i32_ugt:
|
|
255 ; R600: SETGT_UINT
|
|
256 ; GCN: v_cmp_gt_u32
|
|
257 define amdgpu_kernel void @i32_ugt(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
258 entry:
|
|
259 %0 = icmp ugt i32 %a, %b
|
|
260 %1 = sext i1 %0 to i32
|
|
261 store i32 %1, i32 addrspace(1)* %out
|
|
262 ret void
|
|
263 }
|
|
264
|
|
265 ; FUNC-LABEL: {{^}}i32_uge:
|
|
266 ; R600: SETGE_UINT
|
|
267 ; GCN: v_cmp_ge_u32
|
|
268 define amdgpu_kernel void @i32_uge(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
269 entry:
|
|
270 %0 = icmp uge i32 %a, %b
|
|
271 %1 = sext i1 %0 to i32
|
|
272 store i32 %1, i32 addrspace(1)* %out
|
|
273 ret void
|
|
274 }
|
|
275
|
|
276 ; FUNC-LABEL: {{^}}i32_ult:
|
|
277 ; R600: SETGT_UINT
|
|
278 ; GCN: v_cmp_lt_u32
|
|
279 define amdgpu_kernel void @i32_ult(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
280 entry:
|
|
281 %0 = icmp ult i32 %a, %b
|
|
282 %1 = sext i1 %0 to i32
|
|
283 store i32 %1, i32 addrspace(1)* %out
|
|
284 ret void
|
|
285 }
|
|
286
|
|
287 ; FUNC-LABEL: {{^}}i32_ule:
|
|
288 ; R600: SETGE_UINT
|
|
289 ; GCN: v_cmp_le_u32
|
|
290 define amdgpu_kernel void @i32_ule(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
291 entry:
|
|
292 %0 = icmp ule i32 %a, %b
|
|
293 %1 = sext i1 %0 to i32
|
|
294 store i32 %1, i32 addrspace(1)* %out
|
|
295 ret void
|
|
296 }
|
|
297
|
|
298 ; FUNC-LABEL: {{^}}i32_sgt:
|
|
299 ; R600: SETGT_INT
|
|
300 ; GCN: v_cmp_gt_i32
|
|
301 define amdgpu_kernel void @i32_sgt(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
302 entry:
|
|
303 %0 = icmp sgt i32 %a, %b
|
|
304 %1 = sext i1 %0 to i32
|
|
305 store i32 %1, i32 addrspace(1)* %out
|
|
306 ret void
|
|
307 }
|
|
308
|
|
309 ; FUNC-LABEL: {{^}}i32_sge:
|
|
310 ; R600: SETGE_INT
|
|
311 ; GCN: v_cmp_ge_i32
|
|
312 define amdgpu_kernel void @i32_sge(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
313 entry:
|
|
314 %0 = icmp sge i32 %a, %b
|
|
315 %1 = sext i1 %0 to i32
|
|
316 store i32 %1, i32 addrspace(1)* %out
|
|
317 ret void
|
|
318 }
|
|
319
|
|
320 ; FUNC-LABEL: {{^}}i32_slt:
|
|
321 ; R600: SETGT_INT
|
|
322 ; GCN: v_cmp_lt_i32
|
|
323 define amdgpu_kernel void @i32_slt(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
324 entry:
|
|
325 %0 = icmp slt i32 %a, %b
|
|
326 %1 = sext i1 %0 to i32
|
|
327 store i32 %1, i32 addrspace(1)* %out
|
|
328 ret void
|
|
329 }
|
|
330
|
|
331 ; FUNC-LABEL: {{^}}i32_sle:
|
|
332 ; R600: SETGE_INT
|
|
333 ; GCN: v_cmp_le_i32
|
|
334 define amdgpu_kernel void @i32_sle(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
|
|
335 entry:
|
|
336 %0 = icmp sle i32 %a, %b
|
|
337 %1 = sext i1 %0 to i32
|
|
338 store i32 %1, i32 addrspace(1)* %out
|
|
339 ret void
|
|
340 }
|
|
341
|
|
342 ; FIXME: This does 4 compares
|
|
343 ; FUNC-LABEL: {{^}}v3i32_eq:
|
|
344 ; GCN-DAG: v_cmp_eq_u32
|
|
345 ; GCN-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
|
|
346 ; GCN-DAG: v_cmp_eq_u32
|
|
347 ; GCN-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
|
|
348 ; GCN-DAG: v_cmp_eq_u32
|
|
349 ; GCN-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
|
|
350 ; GCN: s_endpgm
|
|
351 define amdgpu_kernel void @v3i32_eq(<3 x i32> addrspace(1)* %out, <3 x i32> addrspace(1)* %ptra, <3 x i32> addrspace(1)* %ptrb) #0 {
|
|
352 %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
|
|
353 %gep.a = getelementptr <3 x i32>, <3 x i32> addrspace(1)* %ptra, i32 %tid
|
|
354 %gep.b = getelementptr <3 x i32>, <3 x i32> addrspace(1)* %ptrb, i32 %tid
|
|
355 %gep.out = getelementptr <3 x i32>, <3 x i32> addrspace(1)* %out, i32 %tid
|
|
356 %a = load <3 x i32>, <3 x i32> addrspace(1)* %gep.a
|
|
357 %b = load <3 x i32>, <3 x i32> addrspace(1)* %gep.b
|
|
358 %cmp = icmp eq <3 x i32> %a, %b
|
|
359 %ext = sext <3 x i1> %cmp to <3 x i32>
|
|
360 store <3 x i32> %ext, <3 x i32> addrspace(1)* %gep.out
|
|
361 ret void
|
|
362 }
|
|
363
|
|
364 ; FUNC-LABEL: {{^}}v3i8_eq:
|
|
365 ; GCN-DAG: v_cmp_eq_u32
|
|
366 ; GCN-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
|
|
367 ; GCN-DAG: v_cmp_eq_u32
|
|
368 ; GCN-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
|
|
369 ; GCN-DAG: v_cmp_eq_u32
|
|
370 ; GCN-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
|
|
371 ; GCN: s_endpgm
|
|
372 define amdgpu_kernel void @v3i8_eq(<3 x i8> addrspace(1)* %out, <3 x i8> addrspace(1)* %ptra, <3 x i8> addrspace(1)* %ptrb) #0 {
|
|
373 %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
|
|
374 %gep.a = getelementptr <3 x i8>, <3 x i8> addrspace(1)* %ptra, i32 %tid
|
|
375 %gep.b = getelementptr <3 x i8>, <3 x i8> addrspace(1)* %ptrb, i32 %tid
|
|
376 %gep.out = getelementptr <3 x i8>, <3 x i8> addrspace(1)* %out, i32 %tid
|
|
377 %a = load <3 x i8>, <3 x i8> addrspace(1)* %gep.a
|
|
378 %b = load <3 x i8>, <3 x i8> addrspace(1)* %gep.b
|
|
379 %cmp = icmp eq <3 x i8> %a, %b
|
|
380 %ext = sext <3 x i1> %cmp to <3 x i8>
|
|
381 store <3 x i8> %ext, <3 x i8> addrspace(1)* %gep.out
|
|
382 ret void
|
|
383 }
|
|
384
|
|
385 ; Make sure we don't try to emit i1 setcc ops
|
|
386 ; FUNC-LABEL: setcc-i1
|
|
387 ; GCN: s_and_b32 [[AND:s[0-9]+]], s{{[0-9]+}}, 1
|
|
388 ; GCN: s_cmp_eq_u32 [[AND]], 0
|
|
389 define amdgpu_kernel void @setcc-i1(i32 %in) #0 {
|
|
390 %and = and i32 %in, 1
|
|
391 %cmp = icmp eq i32 %and, 0
|
|
392 br i1 %cmp, label %endif, label %if
|
|
393 if:
|
|
394 unreachable
|
|
395 endif:
|
|
396 ret void
|
|
397 }
|
|
398
|
|
399 ; FUNC-LABEL: setcc-i1-and-xor
|
173
|
400 ; GCN-DAG: v_cmp_ge_f32_e64 [[A:s\[[0-9]+:[0-9]+\]]], s{{[0-9]+}}, 0{{$}}
|
|
401 ; GCN-DAG: v_cmp_le_f32_e64 [[B:s\[[0-9]+:[0-9]+\]]], s{{[0-9]+}}, 1.0
|
|
402 ; GCN: s_and_b64 s[2:3], [[A]], [[B]]
|
150
|
403 define amdgpu_kernel void @setcc-i1-and-xor(i32 addrspace(1)* %out, float %cond) #0 {
|
|
404 bb0:
|
|
405 %tmp5 = fcmp oge float %cond, 0.000000e+00
|
|
406 %tmp7 = fcmp ole float %cond, 1.000000e+00
|
|
407 %tmp9 = and i1 %tmp5, %tmp7
|
|
408 %tmp11 = xor i1 %tmp9, 1
|
|
409 br i1 %tmp11, label %bb2, label %bb1
|
|
410
|
|
411 bb1:
|
|
412 store i32 0, i32 addrspace(1)* %out
|
|
413 br label %bb2
|
|
414
|
|
415 bb2:
|
|
416 ret void
|
|
417 }
|
|
418
|
|
419 ; FUNC-LABEL: setcc_v2i32_expand
|
|
420 ; GCN: v_cmp_gt_i32
|
|
421 ; GCN: v_cmp_gt_i32
|
|
422 define amdgpu_kernel void @setcc_v2i32_expand(
|
|
423 <2 x i32> addrspace(1)* %a,
|
|
424 <2 x i32> addrspace(1)* %b,
|
|
425 <2 x i32> addrspace(1)* %c,
|
|
426 <2 x float> addrspace(1)* %r) {
|
|
427 entry:
|
|
428 %a.val = load <2 x i32>, <2 x i32> addrspace(1)* %a
|
|
429 %b.val = load <2 x i32>, <2 x i32> addrspace(1)* %b
|
|
430 %c.val = load <2 x i32>, <2 x i32> addrspace(1)* %c
|
|
431
|
|
432 %icmp.val.1 = icmp sgt <2 x i32> %a.val, <i32 1, i32 1>
|
|
433 %zext.val.1 = zext <2 x i1> %icmp.val.1 to <2 x i32>
|
|
434 %shl.val.1 = shl nuw <2 x i32> %zext.val.1, <i32 31, i32 31>
|
|
435 %xor.val.1 = xor <2 x i32> %shl.val.1, %b.val
|
|
436 %bitcast.val.1 = bitcast <2 x i32> %xor.val.1 to <2 x float>
|
|
437 %icmp.val.2 = icmp sgt <2 x i32> %c.val, <i32 1199570944, i32 1199570944>
|
|
438 %select.val.1 = select <2 x i1> %icmp.val.2, <2 x float> <float 1.000000e+00, float 1.000000e+00>, <2 x float> %bitcast.val.1
|
|
439
|
|
440 store <2 x float> %select.val.1, <2 x float> addrspace(1)* %r
|
|
441 ret void
|
|
442 }
|
|
443
|
|
444 ; FUNC-LABEL: setcc_v4i32_expand
|
|
445 ; GCN: v_cmp_gt_i32
|
|
446 ; GCN: v_cmp_gt_i32
|
|
447 ; GCN: v_cmp_gt_i32
|
|
448 ; GCN: v_cmp_gt_i32
|
|
449 define amdgpu_kernel void @setcc_v4i32_expand(
|
|
450 <4 x i32> addrspace(1)* %a,
|
|
451 <4 x i32> addrspace(1)* %b,
|
|
452 <4 x i32> addrspace(1)* %c,
|
|
453 <4 x float> addrspace(1)* %r) {
|
|
454 entry:
|
|
455 %a.val = load <4 x i32>, <4 x i32> addrspace(1)* %a
|
|
456 %b.val = load <4 x i32>, <4 x i32> addrspace(1)* %b
|
|
457 %c.val = load <4 x i32>, <4 x i32> addrspace(1)* %c
|
|
458
|
|
459 %icmp.val.1 = icmp sgt <4 x i32> %a.val, <i32 1, i32 1, i32 1, i32 1>
|
|
460 %zext.val.1 = zext <4 x i1> %icmp.val.1 to <4 x i32>
|
|
461 %shl.val.1 = shl nuw <4 x i32> %zext.val.1, <i32 31, i32 31, i32 31, i32 31>
|
|
462 %xor.val.1 = xor <4 x i32> %shl.val.1, %b.val
|
|
463 %bitcast.val.1 = bitcast <4 x i32> %xor.val.1 to <4 x float>
|
|
464 %icmp.val.2 = icmp sgt <4 x i32> %c.val, <i32 1199570944, i32 1199570944, i32 1199570944, i32 1199570944>
|
|
465 %select.val.1 = select <4 x i1> %icmp.val.2, <4 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>, <4 x float> %bitcast.val.1
|
|
466
|
|
467 store <4 x float> %select.val.1, <4 x float> addrspace(1)* %r
|
|
468 ret void
|
|
469 }
|
|
470
|
|
471 attributes #0 = { nounwind }
|