| blob | 65e3c1f0633d7967b15b77f5ac22294751c68c24 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE
3 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
4 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX,AVX512
6 declare void @use_v8i1(<8 x i1>)
7 declare void @use_v8i8(<8 x i8>)
9 define <8 x i16> @cmp_ne_load_const(ptr %x) nounwind {
10 ; SSE-LABEL: cmp_ne_load_const:
11 ; SSE: # %bb.0:
12 ; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
13 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
14 ; SSE-NEXT: pxor %xmm1, %xmm1
15 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
16 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
17 ; SSE-NEXT: pxor %xmm1, %xmm0
18 ; SSE-NEXT: retq
19 ;
20 ; AVX-LABEL: cmp_ne_load_const:
21 ; AVX: # %bb.0:
22 ; AVX-NEXT: vpmovzxbw {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
23 ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
24 ; AVX-NEXT: vpcmpgtw %xmm1, %xmm0, %xmm0
25 ; AVX-NEXT: retq
26 %loadx = load <8 x i8>, ptr %x
27 %icmp = icmp ne <8 x i8> %loadx, zeroinitializer
28 %sext = sext <8 x i1> %icmp to <8 x i16>
29 ret <8 x i16> %sext
30 }
32 ; negative test - simple loads only
34 define <8 x i16> @cmp_ne_load_const_volatile(ptr %x) nounwind {
35 ; SSE-LABEL: cmp_ne_load_const_volatile:
36 ; SSE: # %bb.0:
37 ; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
38 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
39 ; SSE-NEXT: pxor %xmm1, %xmm1
40 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
41 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
42 ; SSE-NEXT: pxor %xmm1, %xmm0
43 ; SSE-NEXT: retq
44 ;
45 ; AVX2-LABEL: cmp_ne_load_const_volatile:
46 ; AVX2: # %bb.0:
47 ; AVX2-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
48 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
49 ; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
50 ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
51 ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
52 ; AVX2-NEXT: vpmovsxbw %xmm0, %xmm0
53 ; AVX2-NEXT: retq
54 ;
55 ; AVX512-LABEL: cmp_ne_load_const_volatile:
56 ; AVX512: # %bb.0:
57 ; AVX512-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
58 ; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
59 ; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
60 ; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
61 ; AVX512-NEXT: vpmovsxbw %xmm0, %xmm0
62 ; AVX512-NEXT: vzeroupper
63 ; AVX512-NEXT: retq
64 %loadx = load volatile <8 x i8>, ptr %x
65 %icmp = icmp ne <8 x i8> %loadx, zeroinitializer
66 %sext = sext <8 x i1> %icmp to <8 x i16>
67 ret <8 x i16> %sext
68 }
70 ; negative test - don't create extra load
72 define <8 x i16> @cmp_ne_load_const_extra_use1(ptr %x) nounwind {
73 ; SSE-LABEL: cmp_ne_load_const_extra_use1:
74 ; SSE: # %bb.0:
75 ; SSE-NEXT: subq $24, %rsp
76 ; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
77 ; SSE-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
78 ; SSE-NEXT: callq use_v8i8@PLT
79 ; SSE-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload
80 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
81 ; SSE-NEXT: pxor %xmm1, %xmm1
82 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
83 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
84 ; SSE-NEXT: pxor %xmm1, %xmm0
85 ; SSE-NEXT: addq $24, %rsp
86 ; SSE-NEXT: retq
87 ;
88 ; AVX2-LABEL: cmp_ne_load_const_extra_use1:
89 ; AVX2: # %bb.0:
90 ; AVX2-NEXT: subq $24, %rsp
91 ; AVX2-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
92 ; AVX2-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
93 ; AVX2-NEXT: callq use_v8i8@PLT
94 ; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0
95 ; AVX2-NEXT: vpcmpeqb (%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
96 ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
97 ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
98 ; AVX2-NEXT: vpmovsxbw %xmm0, %xmm0
99 ; AVX2-NEXT: addq $24, %rsp
100 ; AVX2-NEXT: retq
101 ;
102 ; AVX512-LABEL: cmp_ne_load_const_extra_use1:
103 ; AVX512: # %bb.0:
104 ; AVX512-NEXT: subq $24, %rsp
105 ; AVX512-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
106 ; AVX512-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
107 ; AVX512-NEXT: callq use_v8i8@PLT
108 ; AVX512-NEXT: vpxor %xmm0, %xmm0, %xmm0
109 ; AVX512-NEXT: vpcmpeqb (%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
110 ; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
111 ; AVX512-NEXT: vpmovsxbw %xmm0, %xmm0
112 ; AVX512-NEXT: addq $24, %rsp
113 ; AVX512-NEXT: vzeroupper
114 ; AVX512-NEXT: retq
115 %loadx = load <8 x i8>, ptr %x
116 call void @use_v8i8(<8 x i8> %loadx)
117 %icmp = icmp ne <8 x i8> %loadx, zeroinitializer
118 %sext = sext <8 x i1> %icmp to <8 x i16>
119 ret <8 x i16> %sext
120 }
122 ; negative test - don't create extra compare
124 define <8 x i16> @cmp_ne_load_const_extra_use2(ptr %x) nounwind {
125 ; SSE-LABEL: cmp_ne_load_const_extra_use2:
126 ; SSE: # %bb.0:
127 ; SSE-NEXT: subq $24, %rsp
128 ; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
129 ; SSE-NEXT: pxor %xmm1, %xmm1
130 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
131 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
132 ; SSE-NEXT: pxor %xmm1, %xmm0
133 ; SSE-NEXT: movdqa %xmm0, (%rsp) # 16-byte Spill
134 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
135 ; SSE-NEXT: callq use_v8i1@PLT
136 ; SSE-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload
137 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
138 ; SSE-NEXT: addq $24, %rsp
139 ; SSE-NEXT: retq
140 ;
141 ; AVX2-LABEL: cmp_ne_load_const_extra_use2:
142 ; AVX2: # %bb.0:
143 ; AVX2-NEXT: subq $24, %rsp
144 ; AVX2-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
145 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
146 ; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
147 ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
148 ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
149 ; AVX2-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
150 ; AVX2-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
151 ; AVX2-NEXT: callq use_v8i1@PLT
152 ; AVX2-NEXT: vpmovsxbw (%rsp), %xmm0 # 16-byte Folded Reload
153 ; AVX2-NEXT: addq $24, %rsp
154 ; AVX2-NEXT: retq
155 ;
156 ; AVX512-LABEL: cmp_ne_load_const_extra_use2:
157 ; AVX512: # %bb.0:
158 ; AVX512-NEXT: subq $72, %rsp
159 ; AVX512-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
160 ; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
161 ; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
162 ; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
163 ; AVX512-NEXT: vmovdqu64 %zmm0, (%rsp) # 64-byte Spill
164 ; AVX512-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
165 ; AVX512-NEXT: vzeroupper
166 ; AVX512-NEXT: callq use_v8i1@PLT
167 ; AVX512-NEXT: vpmovsxbw (%rsp), %xmm0 # 16-byte Folded Reload
168 ; AVX512-NEXT: addq $72, %rsp
169 ; AVX512-NEXT: retq
170 %loadx = load <8 x i8>, ptr %x
171 %icmp = icmp ne <8 x i8> %loadx, zeroinitializer
172 call void @use_v8i1(<8 x i1> %icmp)
173 %sext = sext <8 x i1> %icmp to <8 x i16>
174 ret <8 x i16> %sext
175 }
177 ; negative test - not free extend
179 define <8 x i16> @cmp_ne_no_load_const(i64 %x) nounwind {
180 ; SSE-LABEL: cmp_ne_no_load_const:
181 ; SSE: # %bb.0:
182 ; SSE-NEXT: movq %rdi, %xmm0
183 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
184 ; SSE-NEXT: pxor %xmm1, %xmm1
185 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
186 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
187 ; SSE-NEXT: pxor %xmm1, %xmm0
188 ; SSE-NEXT: retq
189 ;
190 ; AVX2-LABEL: cmp_ne_no_load_const:
191 ; AVX2: # %bb.0:
192 ; AVX2-NEXT: vmovq %rdi, %xmm0
193 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
194 ; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
195 ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
196 ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
197 ; AVX2-NEXT: vpmovsxbw %xmm0, %xmm0
198 ; AVX2-NEXT: retq
199 ;
200 ; AVX512-LABEL: cmp_ne_no_load_const:
201 ; AVX512: # %bb.0:
202 ; AVX512-NEXT: vmovq %rdi, %xmm0
203 ; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
204 ; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
205 ; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
206 ; AVX512-NEXT: vpmovsxbw %xmm0, %xmm0
207 ; AVX512-NEXT: vzeroupper
208 ; AVX512-NEXT: retq
209 %t = bitcast i64 %x to <8 x i8>
210 %icmp = icmp ne <8 x i8> %t, zeroinitializer
211 %sext = sext <8 x i1> %icmp to <8 x i16>
212 ret <8 x i16> %sext
213 }
215 define <4 x i32> @cmp_ult_load_const(ptr %x) nounwind {
216 ; SSE-LABEL: cmp_ult_load_const:
217 ; SSE: # %bb.0:
218 ; SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
219 ; SSE-NEXT: movdqa {{.*#+}} xmm1 = <42,214,0,255,u,u,u,u,u,u,u,u,u,u,u,u>
220 ; SSE-NEXT: pmaxub %xmm0, %xmm1
221 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
222 ; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
223 ; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3]
224 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
225 ; SSE-NEXT: pxor %xmm1, %xmm0
226 ; SSE-NEXT: retq
227 ;
228 ; AVX-LABEL: cmp_ult_load_const:
229 ; AVX: # %bb.0:
230 ; AVX-NEXT: vpmovzxbd {{.*#+}} xmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero
231 ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [42,214,0,255]
232 ; AVX-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm0
233 ; AVX-NEXT: retq
234 %loadx = load <4 x i8>, ptr %x
235 %icmp = icmp ult <4 x i8> %loadx, <i8 42, i8 -42, i8 0, i8 -1>
236 %sext = sext <4 x i1> %icmp to <4 x i32>
237 ret <4 x i32> %sext
238 }
240 ; negative test - type must be legal
242 define <3 x i32> @cmp_ult_load_const_bad_type(ptr %x) nounwind {
243 ; SSE-LABEL: cmp_ult_load_const_bad_type:
244 ; SSE: # %bb.0:
245 ; SSE-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
246 ; SSE-NEXT: movdqa {{.*#+}} xmm1 = <42,214,0,u,u,u,u,u,u,u,u,u,u,u,u,u>
247 ; SSE-NEXT: pmaxub %xmm0, %xmm1
248 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
249 ; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
250 ; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3]
251 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
252 ; SSE-NEXT: pxor %xmm1, %xmm0
253 ; SSE-NEXT: retq
254 ;
255 ; AVX2-LABEL: cmp_ult_load_const_bad_type:
256 ; AVX2: # %bb.0:
257 ; AVX2-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
258 ; AVX2-NEXT: vpmaxub {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
259 ; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
260 ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
261 ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
262 ; AVX2-NEXT: vpmovsxbd %xmm0, %xmm0
263 ; AVX2-NEXT: retq
264 ;
265 ; AVX512-LABEL: cmp_ult_load_const_bad_type:
266 ; AVX512: # %bb.0:
267 ; AVX512-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
268 ; AVX512-NEXT: vpmaxub {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
269 ; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
270 ; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
271 ; AVX512-NEXT: vpmovsxbd %xmm0, %xmm0
272 ; AVX512-NEXT: vzeroupper
273 ; AVX512-NEXT: retq
274 %loadx = load <3 x i8>, ptr %x
275 %icmp = icmp ult <3 x i8> %loadx, <i8 42, i8 -42, i8 0>
276 %sext = sext <3 x i1> %icmp to <3 x i32>
277 ret <3 x i32> %sext
278 }
280 ; Signed compare needs signed extend.
282 define <4 x i32> @cmp_slt_load_const(ptr %x) nounwind {
283 ; SSE-LABEL: cmp_slt_load_const:
284 ; SSE: # %bb.0:
285 ; SSE-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
286 ; SSE-NEXT: movdqa {{.*#+}} xmm0 = <42,214,0,255,u,u,u,u,u,u,u,u,u,u,u,u>
287 ; SSE-NEXT: pcmpgtb %xmm1, %xmm0
288 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
289 ; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
290 ; SSE-NEXT: retq
291 ;
292 ; AVX-LABEL: cmp_slt_load_const:
293 ; AVX: # %bb.0:
294 ; AVX-NEXT: vpmovsxbd (%rdi), %xmm0
295 ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [42,4294967254,0,4294967295]
296 ; AVX-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm0
297 ; AVX-NEXT: retq
298 %loadx = load <4 x i8>, ptr %x
299 %icmp = icmp slt <4 x i8> %loadx, <i8 42, i8 -42, i8 0, i8 -1>
300 %sext = sext <4 x i1> %icmp to <4 x i32>
301 ret <4 x i32> %sext
302 }
304 define <2 x i64> @cmp_ne_zextload(ptr %x, ptr %y) nounwind {
305 ; SSE-LABEL: cmp_ne_zextload:
306 ; SSE: # %bb.0:
307 ; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
308 ; SSE-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
309 ; SSE-NEXT: pcmpeqd %xmm0, %xmm1
310 ; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
311 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
312 ; SSE-NEXT: pxor %xmm1, %xmm0
313 ; SSE-NEXT: retq
314 ;
315 ; AVX2-LABEL: cmp_ne_zextload:
316 ; AVX2: # %bb.0:
317 ; AVX2-NEXT: vpmovzxdq {{.*#+}} xmm0 = mem[0],zero,mem[1],zero
318 ; AVX2-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
319 ; AVX2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
320 ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
321 ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
322 ; AVX2-NEXT: retq
323 ;
324 ; AVX512-LABEL: cmp_ne_zextload:
325 ; AVX512: # %bb.0:
326 ; AVX512-NEXT: vpmovzxdq {{.*#+}} xmm0 = mem[0],zero,mem[1],zero
327 ; AVX512-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
328 ; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
329 ; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
330 ; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
331 ; AVX512-NEXT: vzeroupper
332 ; AVX512-NEXT: retq
333 %loadx = load <2 x i32>, ptr %x
334 %loady = load <2 x i32>, ptr %y
335 %icmp = icmp ne <2 x i32> %loadx, %loady
336 %sext = sext <2 x i1> %icmp to <2 x i64>
337 ret <2 x i64> %sext
338 }
340 define <8 x i16> @cmp_ugt_zextload(ptr %x, ptr %y) nounwind {
341 ; SSE-LABEL: cmp_ugt_zextload:
342 ; SSE: # %bb.0:
343 ; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
344 ; SSE-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
345 ; SSE-NEXT: pminub %xmm0, %xmm1
346 ; SSE-NEXT: pcmpeqb %xmm0, %xmm1
347 ; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
348 ; SSE-NEXT: pcmpeqd %xmm0, %xmm0
349 ; SSE-NEXT: pxor %xmm1, %xmm0
350 ; SSE-NEXT: retq
351 ;
352 ; AVX-LABEL: cmp_ugt_zextload:
353 ; AVX: # %bb.0:
354 ; AVX-NEXT: vpmovzxbw {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
355 ; AVX-NEXT: vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
356 ; AVX-NEXT: vpcmpgtw %xmm1, %xmm0, %xmm0
357 ; AVX-NEXT: retq
358 %loadx = load <8 x i8>, ptr %x
359 %loady = load <8 x i8>, ptr %y
360 %icmp = icmp ugt <8 x i8> %loadx, %loady
361 %sext = sext <8 x i1> %icmp to <8 x i16>
362 ret <8 x i16> %sext
363 }
365 ; Signed compare needs signed extends.
367 define <8 x i16> @cmp_sgt_zextload(ptr %x, ptr %y) nounwind {
368 ; SSE-LABEL: cmp_sgt_zextload:
369 ; SSE: # %bb.0:
370 ; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
371 ; SSE-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
372 ; SSE-NEXT: pcmpgtb %xmm1, %xmm0
373 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
374 ; SSE-NEXT: retq
375 ;
376 ; AVX-LABEL: cmp_sgt_zextload:
377 ; AVX: # %bb.0:
378 ; AVX-NEXT: vpmovsxbw (%rdi), %xmm0
379 ; AVX-NEXT: vpmovsxbw (%rsi), %xmm1
380 ; AVX-NEXT: vpcmpgtw %xmm1, %xmm0, %xmm0
381 ; AVX-NEXT: retq
382 %loadx = load <8 x i8>, ptr %x
383 %loady = load <8 x i8>, ptr %y
384 %icmp = icmp sgt <8 x i8> %loadx, %loady
385 %sext = sext <8 x i1> %icmp to <8 x i16>
386 ret <8 x i16> %sext
387 }
389 ; negative test - don't change a legal op
390 ; TODO: Or should we? We can eliminate the vpmovsxwd at the cost of a 256-bit ymm vpcmpeqw.
392 define <8 x i32> @cmp_ne_zextload_from_legal_op(ptr %x, ptr %y) {
393 ; SSE-LABEL: cmp_ne_zextload_from_legal_op:
394 ; SSE: # %bb.0:
395 ; SSE-NEXT: movdqa (%rdi), %xmm0
396 ; SSE-NEXT: pcmpeqw (%rsi), %xmm0
397 ; SSE-NEXT: pcmpeqd %xmm1, %xmm1
398 ; SSE-NEXT: pxor %xmm0, %xmm1
399 ; SSE-NEXT: movdqa %xmm1, %xmm0
400 ; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
401 ; SSE-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
402 ; SSE-NEXT: retq
403 ;
404 ; AVX2-LABEL: cmp_ne_zextload_from_legal_op:
405 ; AVX2: # %bb.0:
406 ; AVX2-NEXT: vmovdqa (%rdi), %xmm0
407 ; AVX2-NEXT: vpcmpeqw (%rsi), %xmm0, %xmm0
408 ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
409 ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
410 ; AVX2-NEXT: vpmovsxwd %xmm0, %ymm0
411 ; AVX2-NEXT: retq
412 ;
413 ; AVX512-LABEL: cmp_ne_zextload_from_legal_op:
414 ; AVX512: # %bb.0:
415 ; AVX512-NEXT: vmovdqa (%rdi), %xmm0
416 ; AVX512-NEXT: vpcmpeqw (%rsi), %xmm0, %xmm0
417 ; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
418 ; AVX512-NEXT: vpmovsxwd %xmm0, %ymm0
419 ; AVX512-NEXT: retq
420 %loadx = load <8 x i16>, ptr %x
421 %loady = load <8 x i16>, ptr %y
422 %icmp = icmp ne <8 x i16> %loadx, %loady
423 %sext = sext <8 x i1> %icmp to <8 x i32>
424 ret <8 x i32> %sext
425 }
427 ; Both uses of the load can be absorbed by the zext-load, so we eliminate the explicit casts.
429 define <8 x i32> @PR50055(ptr %src, ptr %dst) nounwind {
430 ; SSE-LABEL: PR50055:
431 ; SSE: # %bb.0:
432 ; SSE-NEXT: movq {{.*#+}} xmm2 = mem[0],zero
433 ; SSE-NEXT: pxor %xmm3, %xmm3
434 ; SSE-NEXT: movdqa %xmm2, %xmm1
435 ; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
436 ; SSE-NEXT: movdqa %xmm1, %xmm0
437 ; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
438 ; SSE-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
439 ; SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
440 ; SSE-NEXT: pcmpeqb %xmm3, %xmm2
441 ; SSE-NEXT: pcmpeqd %xmm3, %xmm3
442 ; SSE-NEXT: pxor %xmm2, %xmm3
443 ; SSE-NEXT: movdqa %xmm3, %xmm2
444 ; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
445 ; SSE-NEXT: punpckhwd {{.*#+}} xmm3 = xmm3[4,4,5,5,6,6,7,7]
446 ; SSE-NEXT: movdqa %xmm3, 16(%rsi)
447 ; SSE-NEXT: movdqa %xmm2, (%rsi)
448 ; SSE-NEXT: retq
449 ;
450 ; AVX-LABEL: PR50055:
451 ; AVX: # %bb.0:
452 ; AVX-NEXT: vpmovzxbd {{.*#+}} ymm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero
453 ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
454 ; AVX-NEXT: vpcmpgtd %ymm1, %ymm0, %ymm1
455 ; AVX-NEXT: vmovdqa %ymm1, (%rsi)
456 ; AVX-NEXT: retq
457 %load = load <8 x i8>, ptr %src
458 %zext = zext <8 x i8> %load to <8 x i32>
459 %icmp = icmp ne <8 x i8> %load, zeroinitializer
460 %sext = sext <8 x i1> %icmp to <8 x i32>
461 store <8 x i32> %sext, ptr %dst
462 ret <8 x i32> %zext
463 }
465 ; negative test - extra uses must be absorbable by a zext-load.
467 define <8 x i16> @multi_use_narrower_size(ptr %src, ptr %dst) nounwind {
468 ; SSE-LABEL: multi_use_narrower_size:
469 ; SSE: # %bb.0:
470 ; SSE-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
471 ; SSE-NEXT: pxor %xmm2, %xmm2
472 ; SSE-NEXT: movdqa %xmm1, %xmm0
473 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
474 ; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
475 ; SSE-NEXT: pcmpeqb %xmm2, %xmm1
476 ; SSE-NEXT: movdqa %xmm1, %xmm2
477 ; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
478 ; SSE-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
479 ; SSE-NEXT: movdqa %xmm1, 16(%rsi)
480 ; SSE-NEXT: movdqa %xmm2, (%rsi)
481 ; SSE-NEXT: retq
482 ;
483 ; AVX-LABEL: multi_use_narrower_size:
484 ; AVX: # %bb.0:
485 ; AVX-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
486 ; AVX-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
487 ; AVX-NEXT: vpxor %xmm2, %xmm2, %xmm2
488 ; AVX-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1
489 ; AVX-NEXT: vpmovsxbd %xmm1, %ymm1
490 ; AVX-NEXT: vmovdqa %ymm1, (%rsi)
491 ; AVX-NEXT: vzeroupper
492 ; AVX-NEXT: retq
493 %load = load <8 x i8>, ptr %src
494 %zext = zext <8 x i8> %load to <8 x i16>
495 %icmp = icmp eq <8 x i8> %load, zeroinitializer
496 %sext = sext <8 x i1> %icmp to <8 x i32>
497 store <8 x i32> %sext, ptr %dst
498 ret <8 x i16> %zext
499 }
501 ; negative test - extra uses must be absorbable by a zext-load.
503 define <8 x i32> @multi_use_wider_size(ptr %src, ptr %dst) nounwind {
504 ; SSE-LABEL: multi_use_wider_size:
505 ; SSE: # %bb.0:
506 ; SSE-NEXT: movq {{.*#+}} xmm2 = mem[0],zero
507 ; SSE-NEXT: pxor %xmm3, %xmm3
508 ; SSE-NEXT: movdqa %xmm2, %xmm1
509 ; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
510 ; SSE-NEXT: movdqa %xmm1, %xmm0
511 ; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
512 ; SSE-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
513 ; SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
514 ; SSE-NEXT: pcmpeqb %xmm3, %xmm2
515 ; SSE-NEXT: movdqa %xmm2, (%rsi)
516 ; SSE-NEXT: retq
517 ;
518 ; AVX-LABEL: multi_use_wider_size:
519 ; AVX: # %bb.0:
520 ; AVX-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
521 ; AVX-NEXT: vpmovzxbd {{.*#+}} ymm0 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero
522 ; AVX-NEXT: vpxor %xmm2, %xmm2, %xmm2
523 ; AVX-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1
524 ; AVX-NEXT: vpmovsxbw %xmm1, %xmm1
525 ; AVX-NEXT: vmovdqa %xmm1, (%rsi)
526 ; AVX-NEXT: retq
527 %load = load <8 x i8>, ptr %src
528 %zext = zext <8 x i8> %load to <8 x i32>
529 %icmp = icmp eq <8 x i8> %load, zeroinitializer
530 %sext = sext <8 x i1> %icmp to <8 x i16>
531 store <8 x i16> %sext, ptr %dst
532 ret <8 x i32> %zext
533 }
535 define <4 x i64> @PR50055_signed(ptr %src, ptr %dst) {
536 ; SSE-LABEL: PR50055_signed:
537 ; SSE: # %bb.0:
538 ; SSE-NEXT: movq {{.*#+}} xmm2 = mem[0],zero
539 ; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
540 ; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
541 ; SSE-NEXT: psrad $24, %xmm0
542 ; SSE-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
543 ; SSE-NEXT: psrad $24, %xmm1
544 ; SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
545 ; SSE-NEXT: pxor %xmm3, %xmm3
546 ; SSE-NEXT: pcmpgtb %xmm3, %xmm2
547 ; SSE-NEXT: movdqa %xmm2, %xmm3
548 ; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
549 ; SSE-NEXT: punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
550 ; SSE-NEXT: movdqa %xmm2, 16(%rsi)
551 ; SSE-NEXT: movdqa %xmm3, (%rsi)
552 ; SSE-NEXT: retq
553 ;
554 ; AVX-LABEL: PR50055_signed:
555 ; AVX: # %bb.0:
556 ; AVX-NEXT: vpmovsxbd (%rdi), %ymm0
557 ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
558 ; AVX-NEXT: vpcmpgtd %ymm1, %ymm0, %ymm1
559 ; AVX-NEXT: vmovdqa %ymm1, (%rsi)
560 ; AVX-NEXT: retq
561 %t1 = load <8 x i8>, ptr %src, align 1
562 %conv = sext <8 x i8> %t1 to <8 x i32>
563 %t2 = bitcast <8 x i32> %conv to <4 x i64>
564 %cmp = icmp sgt <8 x i8> %t1, zeroinitializer
565 %sext = sext <8 x i1> %cmp to <8 x i32>
566 store <8 x i32> %sext, ptr %dst, align 32
567 ret <4 x i64> %t2
568 }
570 define <8 x i32> @PR63946(<8 x i32> %a0, <8 x i32> %b0) nounwind {
571 ; SSE-LABEL: PR63946:
572 ; SSE: # %bb.0: # %entry
573 ; SSE-NEXT: movdqa %xmm1, %xmm4
574 ; SSE-NEXT: movdqa %xmm0, %xmm13
575 ; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm2[1,2,3,0]
576 ; SSE-NEXT: pshufd {{.*#+}} xmm9 = xmm3[1,2,3,0]
577 ; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm3[2,3,0,1]
578 ; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm2[2,3,0,1]
579 ; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm3[3,0,1,2]
580 ; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm2[3,0,1,2]
581 ; SSE-NEXT: pcmpeqd %xmm2, %xmm0
582 ; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
583 ; SSE-NEXT: movdqa %xmm9, %xmm11
584 ; SSE-NEXT: pcmpeqd %xmm4, %xmm11
585 ; SSE-NEXT: movdqa %xmm7, %xmm12
586 ; SSE-NEXT: movdqa %xmm8, %xmm10
587 ; SSE-NEXT: movdqa %xmm5, %xmm15
588 ; SSE-NEXT: pcmpeqd %xmm4, %xmm15
589 ; SSE-NEXT: movdqa %xmm1, %xmm14
590 ; SSE-NEXT: pcmpeqd %xmm4, %xmm14
591 ; SSE-NEXT: pcmpeqd %xmm4, %xmm2
592 ; SSE-NEXT: pcmpeqd %xmm4, %xmm7
593 ; SSE-NEXT: pcmpeqd %xmm4, %xmm8
594 ; SSE-NEXT: movdqa %xmm6, %xmm0
595 ; SSE-NEXT: pcmpeqd %xmm4, %xmm6
596 ; SSE-NEXT: pcmpeqd %xmm3, %xmm4
597 ; SSE-NEXT: por %xmm4, %xmm11
598 ; SSE-NEXT: pcmpeqd %xmm13, %xmm12
599 ; SSE-NEXT: por {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
600 ; SSE-NEXT: pcmpeqd %xmm13, %xmm10
601 ; SSE-NEXT: pcmpeqd %xmm13, %xmm0
602 ; SSE-NEXT: por %xmm15, %xmm2
603 ; SSE-NEXT: por %xmm11, %xmm2
604 ; SSE-NEXT: pcmpeqd %xmm13, %xmm3
605 ; SSE-NEXT: por %xmm3, %xmm10
606 ; SSE-NEXT: por %xmm12, %xmm10
607 ; SSE-NEXT: por %xmm14, %xmm7
608 ; SSE-NEXT: pcmpeqd %xmm13, %xmm9
609 ; SSE-NEXT: por %xmm0, %xmm9
610 ; SSE-NEXT: pcmpeqd %xmm13, %xmm5
611 ; SSE-NEXT: por %xmm9, %xmm5
612 ; SSE-NEXT: por %xmm10, %xmm5
613 ; SSE-NEXT: por %xmm7, %xmm8
614 ; SSE-NEXT: por %xmm2, %xmm8
615 ; SSE-NEXT: packssdw %xmm8, %xmm5
616 ; SSE-NEXT: pcmpeqd %xmm13, %xmm1
617 ; SSE-NEXT: packssdw %xmm6, %xmm1
618 ; SSE-NEXT: por %xmm5, %xmm1
619 ; SSE-NEXT: movdqa %xmm1, %xmm0
620 ; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
621 ; SSE-NEXT: pslld $31, %xmm0
622 ; SSE-NEXT: psrad $31, %xmm0
623 ; SSE-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
624 ; SSE-NEXT: pslld $31, %xmm1
625 ; SSE-NEXT: psrad $31, %xmm1
626 ; SSE-NEXT: retq
627 ;
628 ; AVX2-LABEL: PR63946:
629 ; AVX2: # %bb.0: # %entry
630 ; AVX2-NEXT: vpshufd {{.*#+}} ymm2 = ymm1[1,2,3,0,5,6,7,4]
631 ; AVX2-NEXT: vpshufd {{.*#+}} ymm3 = ymm1[2,3,0,1,6,7,4,5]
632 ; AVX2-NEXT: vpshufd {{.*#+}} ymm4 = ymm1[3,0,1,2,7,4,5,6]
633 ; AVX2-NEXT: vpermq {{.*#+}} ymm5 = ymm1[2,3,0,1]
634 ; AVX2-NEXT: vpermq {{.*#+}} ymm6 = ymm2[2,3,0,1]
635 ; AVX2-NEXT: vpermq {{.*#+}} ymm7 = ymm3[2,3,0,1]
636 ; AVX2-NEXT: vpermq {{.*#+}} ymm8 = ymm4[2,3,0,1]
637 ; AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm1
638 ; AVX2-NEXT: vpcmpeqd %ymm0, %ymm2, %ymm2
639 ; AVX2-NEXT: vpor %ymm1, %ymm2, %ymm1
640 ; AVX2-NEXT: vpcmpeqd %ymm0, %ymm3, %ymm2
641 ; AVX2-NEXT: vpcmpeqd %ymm0, %ymm4, %ymm3
642 ; AVX2-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm4
643 ; AVX2-NEXT: vpor %ymm4, %ymm2, %ymm2
644 ; AVX2-NEXT: vpor %ymm2, %ymm1, %ymm1
645 ; AVX2-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm2
646 ; AVX2-NEXT: vpor %ymm2, %ymm3, %ymm2
647 ; AVX2-NEXT: vpor %ymm2, %ymm1, %ymm1
648 ; AVX2-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm2
649 ; AVX2-NEXT: vpcmpeqd %ymm0, %ymm8, %ymm0
650 ; AVX2-NEXT: vpor %ymm0, %ymm2, %ymm0
651 ; AVX2-NEXT: vpor %ymm0, %ymm1, %ymm0
652 ; AVX2-NEXT: vpslld $31, %ymm0, %ymm0
653 ; AVX2-NEXT: vpsrad $31, %ymm0, %ymm0
654 ; AVX2-NEXT: retq
655 ;
656 ; AVX512-LABEL: PR63946:
657 ; AVX512: # %bb.0: # %entry
658 ; AVX512-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
659 ; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
660 ; AVX512-NEXT: vpshufd {{.*#+}} ymm2 = ymm1[1,2,3,0,5,6,7,4]
661 ; AVX512-NEXT: vpshufd {{.*#+}} ymm3 = ymm1[2,3,0,1,6,7,4,5]
662 ; AVX512-NEXT: vpshufd {{.*#+}} ymm4 = ymm1[3,0,1,2,7,4,5,6]
663 ; AVX512-NEXT: vpermq {{.*#+}} ymm5 = ymm1[2,3,0,1]
664 ; AVX512-NEXT: vpermq {{.*#+}} ymm6 = ymm2[2,3,0,1]
665 ; AVX512-NEXT: vpermq {{.*#+}} ymm7 = ymm3[2,3,0,1]
666 ; AVX512-NEXT: vpermq {{.*#+}} ymm8 = ymm4[2,3,0,1]
667 ; AVX512-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
668 ; AVX512-NEXT: vpcmpeqd %zmm0, %zmm2, %k1
669 ; AVX512-NEXT: vpcmpeqd %zmm0, %zmm3, %k3
670 ; AVX512-NEXT: vpcmpeqd %zmm0, %zmm4, %k2
671 ; AVX512-NEXT: vpcmpeqd %zmm0, %zmm5, %k4
672 ; AVX512-NEXT: vpcmpeqd %zmm0, %zmm6, %k5
673 ; AVX512-NEXT: vpcmpeqd %zmm0, %zmm7, %k6
674 ; AVX512-NEXT: vpcmpeqd %zmm0, %zmm8, %k7
675 ; AVX512-NEXT: korw %k0, %k1, %k0
676 ; AVX512-NEXT: korw %k3, %k0, %k0
677 ; AVX512-NEXT: korw %k4, %k0, %k0
678 ; AVX512-NEXT: korw %k2, %k0, %k0
679 ; AVX512-NEXT: korw %k5, %k0, %k0
680 ; AVX512-NEXT: korw %k6, %k0, %k0
681 ; AVX512-NEXT: korw %k7, %k0, %k1
682 ; AVX512-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
683 ; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
684 ; AVX512-NEXT: retq
685 entry:
686 %shuffle = shufflevector <8 x i32> %b0, <8 x i32> poison, <8 x i32> <i32 1, i32 2, i32 3, i32 0, i32 5, i32 6, i32 7, i32 4>
687 %shuffle1 = shufflevector <8 x i32> %b0, <8 x i32> poison, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 6, i32 7, i32 4, i32 5>
688 %shuffle2 = shufflevector <8 x i32> %shuffle, <8 x i32> poison, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 6, i32 7, i32 4, i32 5>
689 %shuffle3 = shufflevector <8 x i32> %b0, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
690 %shuffle4 = shufflevector <8 x i32> %shuffle, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
691 %shuffle5 = shufflevector <8 x i32> %shuffle1, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
692 %shuffle6 = shufflevector <8 x i32> %shuffle2, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
693 %cmp = icmp eq <8 x i32> %a0, %b0
694 %cmp7 = icmp eq <8 x i32> %shuffle, %a0
695 %cmp9 = icmp eq <8 x i32> %shuffle1, %a0
696 %cmp11 = icmp eq <8 x i32> %shuffle2, %a0
697 %cmp13 = icmp eq <8 x i32> %shuffle3, %a0
698 %cmp15 = icmp eq <8 x i32> %shuffle4, %a0
699 %cmp17 = icmp eq <8 x i32> %shuffle5, %a0
700 %cmp19 = icmp eq <8 x i32> %shuffle6, %a0
701 %or2365 = or <8 x i1> %cmp7, %cmp
702 %or2264 = or <8 x i1> %or2365, %cmp9
703 %or2567 = or <8 x i1> %or2264, %cmp13
704 %or2163 = or <8 x i1> %or2567, %cmp11
705 %or62 = or <8 x i1> %or2163, %cmp15
706 %or2466 = or <8 x i1> %or62, %cmp17
707 %or2668 = or <8 x i1> %or2466, %cmp19
708 %or26 = sext <8 x i1> %or2668 to <8 x i32>
709 ret <8 x i32> %or26
710 }