| blob | e2e5cadf9d465f071403b7e4dc4521f6e7198fa0 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X32
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64
5 define <4 x i64> @test_vpaddq(<4 x i64> %i, <4 x i64> %j) nounwind readnone {
6 ; X32-LABEL: test_vpaddq:
7 ; X32: # %bb.0:
8 ; X32-NEXT: vpaddq %ymm1, %ymm0, %ymm0
9 ; X32-NEXT: retl
10 ;
11 ; X64-LABEL: test_vpaddq:
12 ; X64: # %bb.0:
13 ; X64-NEXT: vpaddq %ymm1, %ymm0, %ymm0
14 ; X64-NEXT: retq
15 %x = add <4 x i64> %i, %j
16 ret <4 x i64> %x
17 }
19 define <8 x i32> @test_vpaddd(<8 x i32> %i, <8 x i32> %j) nounwind readnone {
20 ; X32-LABEL: test_vpaddd:
21 ; X32: # %bb.0:
22 ; X32-NEXT: vpaddd %ymm1, %ymm0, %ymm0
23 ; X32-NEXT: retl
24 ;
25 ; X64-LABEL: test_vpaddd:
26 ; X64: # %bb.0:
27 ; X64-NEXT: vpaddd %ymm1, %ymm0, %ymm0
28 ; X64-NEXT: retq
29 %x = add <8 x i32> %i, %j
30 ret <8 x i32> %x
31 }
33 define <16 x i16> @test_vpaddw(<16 x i16> %i, <16 x i16> %j) nounwind readnone {
34 ; X32-LABEL: test_vpaddw:
35 ; X32: # %bb.0:
36 ; X32-NEXT: vpaddw %ymm1, %ymm0, %ymm0
37 ; X32-NEXT: retl
38 ;
39 ; X64-LABEL: test_vpaddw:
40 ; X64: # %bb.0:
41 ; X64-NEXT: vpaddw %ymm1, %ymm0, %ymm0
42 ; X64-NEXT: retq
43 %x = add <16 x i16> %i, %j
44 ret <16 x i16> %x
45 }
47 define <32 x i8> @test_vpaddb(<32 x i8> %i, <32 x i8> %j) nounwind readnone {
48 ; X32-LABEL: test_vpaddb:
49 ; X32: # %bb.0:
50 ; X32-NEXT: vpaddb %ymm1, %ymm0, %ymm0
51 ; X32-NEXT: retl
52 ;
53 ; X64-LABEL: test_vpaddb:
54 ; X64: # %bb.0:
55 ; X64-NEXT: vpaddb %ymm1, %ymm0, %ymm0
56 ; X64-NEXT: retq
57 %x = add <32 x i8> %i, %j
58 ret <32 x i8> %x
59 }
61 define <4 x i64> @test_vpsubq(<4 x i64> %i, <4 x i64> %j) nounwind readnone {
62 ; X32-LABEL: test_vpsubq:
63 ; X32: # %bb.0:
64 ; X32-NEXT: vpsubq %ymm1, %ymm0, %ymm0
65 ; X32-NEXT: retl
66 ;
67 ; X64-LABEL: test_vpsubq:
68 ; X64: # %bb.0:
69 ; X64-NEXT: vpsubq %ymm1, %ymm0, %ymm0
70 ; X64-NEXT: retq
71 %x = sub <4 x i64> %i, %j
72 ret <4 x i64> %x
73 }
75 define <8 x i32> @test_vpsubd(<8 x i32> %i, <8 x i32> %j) nounwind readnone {
76 ; X32-LABEL: test_vpsubd:
77 ; X32: # %bb.0:
78 ; X32-NEXT: vpsubd %ymm1, %ymm0, %ymm0
79 ; X32-NEXT: retl
80 ;
81 ; X64-LABEL: test_vpsubd:
82 ; X64: # %bb.0:
83 ; X64-NEXT: vpsubd %ymm1, %ymm0, %ymm0
84 ; X64-NEXT: retq
85 %x = sub <8 x i32> %i, %j
86 ret <8 x i32> %x
87 }
89 define <16 x i16> @test_vpsubw(<16 x i16> %i, <16 x i16> %j) nounwind readnone {
90 ; X32-LABEL: test_vpsubw:
91 ; X32: # %bb.0:
92 ; X32-NEXT: vpsubw %ymm1, %ymm0, %ymm0
93 ; X32-NEXT: retl
94 ;
95 ; X64-LABEL: test_vpsubw:
96 ; X64: # %bb.0:
97 ; X64-NEXT: vpsubw %ymm1, %ymm0, %ymm0
98 ; X64-NEXT: retq
99 %x = sub <16 x i16> %i, %j
100 ret <16 x i16> %x
101 }
103 define <32 x i8> @test_vpsubb(<32 x i8> %i, <32 x i8> %j) nounwind readnone {
104 ; X32-LABEL: test_vpsubb:
105 ; X32: # %bb.0:
106 ; X32-NEXT: vpsubb %ymm1, %ymm0, %ymm0
107 ; X32-NEXT: retl
108 ;
109 ; X64-LABEL: test_vpsubb:
110 ; X64: # %bb.0:
111 ; X64-NEXT: vpsubb %ymm1, %ymm0, %ymm0
112 ; X64-NEXT: retq
113 %x = sub <32 x i8> %i, %j
114 ret <32 x i8> %x
115 }
117 define <8 x i32> @test_vpmulld(<8 x i32> %i, <8 x i32> %j) nounwind readnone {
118 ; X32-LABEL: test_vpmulld:
119 ; X32: # %bb.0:
120 ; X32-NEXT: vpmulld %ymm1, %ymm0, %ymm0
121 ; X32-NEXT: retl
122 ;
123 ; X64-LABEL: test_vpmulld:
124 ; X64: # %bb.0:
125 ; X64-NEXT: vpmulld %ymm1, %ymm0, %ymm0
126 ; X64-NEXT: retq
127 %x = mul <8 x i32> %i, %j
128 ret <8 x i32> %x
129 }
131 define <16 x i16> @test_vpmullw(<16 x i16> %i, <16 x i16> %j) nounwind readnone {
132 ; X32-LABEL: test_vpmullw:
133 ; X32: # %bb.0:
134 ; X32-NEXT: vpmullw %ymm1, %ymm0, %ymm0
135 ; X32-NEXT: retl
136 ;
137 ; X64-LABEL: test_vpmullw:
138 ; X64: # %bb.0:
139 ; X64-NEXT: vpmullw %ymm1, %ymm0, %ymm0
140 ; X64-NEXT: retq
141 %x = mul <16 x i16> %i, %j
142 ret <16 x i16> %x
143 }
145 define <16 x i8> @mul_v16i8(<16 x i8> %i, <16 x i8> %j) nounwind readnone {
146 ; X32-LABEL: mul_v16i8:
147 ; X32: # %bb.0:
148 ; X32-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
149 ; X32-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
150 ; X32-NEXT: vpmullw %ymm1, %ymm0, %ymm0
151 ; X32-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
152 ; X32-NEXT: vextracti128 $1, %ymm0, %xmm1
153 ; X32-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
154 ; X32-NEXT: vzeroupper
155 ; X32-NEXT: retl
156 ;
157 ; X64-LABEL: mul_v16i8:
158 ; X64: # %bb.0:
159 ; X64-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
160 ; X64-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
161 ; X64-NEXT: vpmullw %ymm1, %ymm0, %ymm0
162 ; X64-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
163 ; X64-NEXT: vextracti128 $1, %ymm0, %xmm1
164 ; X64-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
165 ; X64-NEXT: vzeroupper
166 ; X64-NEXT: retq
167 %x = mul <16 x i8> %i, %j
168 ret <16 x i8> %x
169 }
171 define <32 x i8> @mul_v32i8(<32 x i8> %i, <32 x i8> %j) nounwind readnone {
172 ; X32-LABEL: mul_v32i8:
173 ; X32: # %bb.0:
174 ; X32-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
175 ; X32-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
176 ; X32-NEXT: vpmullw %ymm2, %ymm3, %ymm2
177 ; X32-NEXT: vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
178 ; X32-NEXT: vpand %ymm3, %ymm2, %ymm2
179 ; X32-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
180 ; X32-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
181 ; X32-NEXT: vpmullw %ymm1, %ymm0, %ymm0
182 ; X32-NEXT: vpand %ymm3, %ymm0, %ymm0
183 ; X32-NEXT: vpackuswb %ymm2, %ymm0, %ymm0
184 ; X32-NEXT: retl
185 ;
186 ; X64-LABEL: mul_v32i8:
187 ; X64: # %bb.0:
188 ; X64-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
189 ; X64-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
190 ; X64-NEXT: vpmullw %ymm2, %ymm3, %ymm2
191 ; X64-NEXT: vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
192 ; X64-NEXT: vpand %ymm3, %ymm2, %ymm2
193 ; X64-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
194 ; X64-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
195 ; X64-NEXT: vpmullw %ymm1, %ymm0, %ymm0
196 ; X64-NEXT: vpand %ymm3, %ymm0, %ymm0
197 ; X64-NEXT: vpackuswb %ymm2, %ymm0, %ymm0
198 ; X64-NEXT: retq
199 %x = mul <32 x i8> %i, %j
200 ret <32 x i8> %x
201 }
203 define <4 x i64> @mul_v4i64(<4 x i64> %i, <4 x i64> %j) nounwind readnone {
204 ; X32-LABEL: mul_v4i64:
205 ; X32: # %bb.0:
206 ; X32-NEXT: vpsrlq $32, %ymm0, %ymm2
207 ; X32-NEXT: vpmuludq %ymm1, %ymm2, %ymm2
208 ; X32-NEXT: vpsrlq $32, %ymm1, %ymm3
209 ; X32-NEXT: vpmuludq %ymm3, %ymm0, %ymm3
210 ; X32-NEXT: vpaddq %ymm2, %ymm3, %ymm2
211 ; X32-NEXT: vpsllq $32, %ymm2, %ymm2
212 ; X32-NEXT: vpmuludq %ymm1, %ymm0, %ymm0
213 ; X32-NEXT: vpaddq %ymm2, %ymm0, %ymm0
214 ; X32-NEXT: retl
215 ;
216 ; X64-LABEL: mul_v4i64:
217 ; X64: # %bb.0:
218 ; X64-NEXT: vpsrlq $32, %ymm0, %ymm2
219 ; X64-NEXT: vpmuludq %ymm1, %ymm2, %ymm2
220 ; X64-NEXT: vpsrlq $32, %ymm1, %ymm3
221 ; X64-NEXT: vpmuludq %ymm3, %ymm0, %ymm3
222 ; X64-NEXT: vpaddq %ymm2, %ymm3, %ymm2
223 ; X64-NEXT: vpsllq $32, %ymm2, %ymm2
224 ; X64-NEXT: vpmuludq %ymm1, %ymm0, %ymm0
225 ; X64-NEXT: vpaddq %ymm2, %ymm0, %ymm0
226 ; X64-NEXT: retq
227 %x = mul <4 x i64> %i, %j
228 ret <4 x i64> %x
229 }
231 define <8 x i32> @mul_const1(<8 x i32> %x) {
232 ; X32-LABEL: mul_const1:
233 ; X32: # %bb.0:
234 ; X32-NEXT: vpaddd %ymm0, %ymm0, %ymm0
235 ; X32-NEXT: retl
236 ;
237 ; X64-LABEL: mul_const1:
238 ; X64: # %bb.0:
239 ; X64-NEXT: vpaddd %ymm0, %ymm0, %ymm0
240 ; X64-NEXT: retq
241 %y = mul <8 x i32> %x, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
242 ret <8 x i32> %y
243 }
245 define <4 x i64> @mul_const2(<4 x i64> %x) {
246 ; X32-LABEL: mul_const2:
247 ; X32: # %bb.0:
248 ; X32-NEXT: vpsllq $2, %ymm0, %ymm0
249 ; X32-NEXT: retl
250 ;
251 ; X64-LABEL: mul_const2:
252 ; X64: # %bb.0:
253 ; X64-NEXT: vpsllq $2, %ymm0, %ymm0
254 ; X64-NEXT: retq
255 %y = mul <4 x i64> %x, <i64 4, i64 4, i64 4, i64 4>
256 ret <4 x i64> %y
257 }
259 define <16 x i16> @mul_const3(<16 x i16> %x) {
260 ; X32-LABEL: mul_const3:
261 ; X32: # %bb.0:
262 ; X32-NEXT: vpsllw $3, %ymm0, %ymm0
263 ; X32-NEXT: retl
264 ;
265 ; X64-LABEL: mul_const3:
266 ; X64: # %bb.0:
267 ; X64-NEXT: vpsllw $3, %ymm0, %ymm0
268 ; X64-NEXT: retq
269 %y = mul <16 x i16> %x, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8>
270 ret <16 x i16> %y
271 }
273 define <4 x i64> @mul_const4(<4 x i64> %x) {
274 ; X32-LABEL: mul_const4:
275 ; X32: # %bb.0:
276 ; X32-NEXT: vpxor %xmm1, %xmm1, %xmm1
277 ; X32-NEXT: vpsubq %ymm0, %ymm1, %ymm0
278 ; X32-NEXT: retl
279 ;
280 ; X64-LABEL: mul_const4:
281 ; X64: # %bb.0:
282 ; X64-NEXT: vpxor %xmm1, %xmm1, %xmm1
283 ; X64-NEXT: vpsubq %ymm0, %ymm1, %ymm0
284 ; X64-NEXT: retq
285 %y = mul <4 x i64> %x, <i64 -1, i64 -1, i64 -1, i64 -1>
286 ret <4 x i64> %y
287 }
289 define <8 x i32> @mul_const5(<8 x i32> %x) {
290 ; X32-LABEL: mul_const5:
291 ; X32: # %bb.0:
292 ; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0
293 ; X32-NEXT: retl
294 ;
295 ; X64-LABEL: mul_const5:
296 ; X64: # %bb.0:
297 ; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0
298 ; X64-NEXT: retq
299 %y = mul <8 x i32> %x, <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
300 ret <8 x i32> %y
301 }
303 define <8 x i32> @mul_const6(<8 x i32> %x) {
304 ; X32-LABEL: mul_const6:
305 ; X32: # %bb.0:
306 ; X32-NEXT: vpmulld {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
307 ; X32-NEXT: retl
308 ;
309 ; X64-LABEL: mul_const6:
310 ; X64: # %bb.0:
311 ; X64-NEXT: vpmulld {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
312 ; X64-NEXT: retq
313 %y = mul <8 x i32> %x, <i32 0, i32 0, i32 0, i32 2, i32 0, i32 2, i32 0, i32 0>
314 ret <8 x i32> %y
315 }
317 define <8 x i64> @mul_const7(<8 x i64> %x) {
318 ; X32-LABEL: mul_const7:
319 ; X32: # %bb.0:
320 ; X32-NEXT: vpaddq %ymm0, %ymm0, %ymm0
321 ; X32-NEXT: vpaddq %ymm1, %ymm1, %ymm1
322 ; X32-NEXT: retl
323 ;
324 ; X64-LABEL: mul_const7:
325 ; X64: # %bb.0:
326 ; X64-NEXT: vpaddq %ymm0, %ymm0, %ymm0
327 ; X64-NEXT: vpaddq %ymm1, %ymm1, %ymm1
328 ; X64-NEXT: retq
329 %y = mul <8 x i64> %x, <i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2>
330 ret <8 x i64> %y
331 }
333 define <8 x i16> @mul_const8(<8 x i16> %x) {
334 ; X32-LABEL: mul_const8:
335 ; X32: # %bb.0:
336 ; X32-NEXT: vpsllw $3, %xmm0, %xmm0
337 ; X32-NEXT: retl
338 ;
339 ; X64-LABEL: mul_const8:
340 ; X64: # %bb.0:
341 ; X64-NEXT: vpsllw $3, %xmm0, %xmm0
342 ; X64-NEXT: retq
343 %y = mul <8 x i16> %x, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8>
344 ret <8 x i16> %y
345 }
347 define <8 x i32> @mul_const9(<8 x i32> %x) {
348 ; X32-LABEL: mul_const9:
349 ; X32: # %bb.0:
350 ; X32-NEXT: vmovdqa {{.*#+}} xmm1 = [2,0,0,0]
351 ; X32-NEXT: vpmulld %ymm1, %ymm0, %ymm0
352 ; X32-NEXT: retl
353 ;
354 ; X64-LABEL: mul_const9:
355 ; X64: # %bb.0:
356 ; X64-NEXT: vmovdqa {{.*#+}} xmm1 = [2,0,0,0]
357 ; X64-NEXT: vpmulld %ymm1, %ymm0, %ymm0
358 ; X64-NEXT: retq
359 %y = mul <8 x i32> %x, <i32 2, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
360 ret <8 x i32> %y
361 }
363 ; ptr 0x01010101
364 define <4 x i32> @mul_const10(<4 x i32> %x) {
365 ; X32-LABEL: mul_const10:
366 ; X32: # %bb.0:
367 ; X32-NEXT: vpbroadcastd {{.*#+}} xmm1 = [16843009,16843009,16843009,16843009]
368 ; X32-NEXT: vpmulld %xmm1, %xmm0, %xmm0
369 ; X32-NEXT: retl
370 ;
371 ; X64-LABEL: mul_const10:
372 ; X64: # %bb.0:
373 ; X64-NEXT: vpbroadcastd {{.*#+}} xmm1 = [16843009,16843009,16843009,16843009]
374 ; X64-NEXT: vpmulld %xmm1, %xmm0, %xmm0
375 ; X64-NEXT: retq
376 %m = mul <4 x i32> %x, <i32 16843009, i32 16843009, i32 16843009, i32 16843009>
377 ret <4 x i32> %m
378 }
380 ; ptr 0x80808080
381 define <4 x i32> @mul_const11(<4 x i32> %x) {
382 ; X32-LABEL: mul_const11:
383 ; X32: # %bb.0:
384 ; X32-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2155905152,2155905152,2155905152,2155905152]
385 ; X32-NEXT: vpmulld %xmm1, %xmm0, %xmm0
386 ; X32-NEXT: retl
387 ;
388 ; X64-LABEL: mul_const11:
389 ; X64: # %bb.0:
390 ; X64-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2155905152,2155905152,2155905152,2155905152]
391 ; X64-NEXT: vpmulld %xmm1, %xmm0, %xmm0
392 ; X64-NEXT: retq
393 %m = mul <4 x i32> %x, <i32 2155905152, i32 2155905152, i32 2155905152, i32 2155905152>
394 ret <4 x i32> %m
395 }