• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -disable-peephole | FileCheck %s --check-prefix=ALL --check-prefix=AVX1
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 -disable-peephole | FileCheck %s --check-prefix=ALL --check-prefix=AVX2
4
5define <8 x float> @shuffle_v8f32_45670123(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
6; AVX1-LABEL: shuffle_v8f32_45670123:
7; AVX1:       # %bb.0: # %entry
8; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
9; AVX1-NEXT:    retq
10;
11; AVX2-LABEL: shuffle_v8f32_45670123:
12; AVX2:       # %bb.0: # %entry
13; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
14; AVX2-NEXT:    retq
15entry:
16  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
17  ret <8 x float> %shuffle
18}
19
20define <8 x float> @shuffle_v8f32_45670123_mem(<8 x float>* %pa, <8 x float>* %pb) nounwind uwtable readnone ssp {
21; AVX1-LABEL: shuffle_v8f32_45670123_mem:
22; AVX1:       # %bb.0: # %entry
23; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3,0,1]
24; AVX1-NEXT:    retq
25;
26; AVX2-LABEL: shuffle_v8f32_45670123_mem:
27; AVX2:       # %bb.0: # %entry
28; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = mem[2,3,0,1]
29; AVX2-NEXT:    retq
30entry:
31  %a = load <8 x float>, <8 x float>* %pa
32  %b = load <8 x float>, <8 x float>* %pb
33  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
34  ret <8 x float> %shuffle
35}
36
37define <8 x float> @shuffle_v8f32_0123cdef(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
38; ALL-LABEL: shuffle_v8f32_0123cdef:
39; ALL:       # %bb.0: # %entry
40; ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
41; ALL-NEXT:    retq
42entry:
43  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 13, i32 14, i32 15>
44  ret <8 x float> %shuffle
45}
46
47define <8 x float> @shuffle_v8f32_01230123(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
48; AVX1-LABEL: shuffle_v8f32_01230123:
49; AVX1:       # %bb.0: # %entry
50; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
51; AVX1-NEXT:    retq
52;
53; AVX2-LABEL: shuffle_v8f32_01230123:
54; AVX2:       # %bb.0: # %entry
55; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
56; AVX2-NEXT:    retq
57entry:
58  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
59  ret <8 x float> %shuffle
60}
61
62define <8 x float> @shuffle_v8f32_01230123_mem(<8 x float>* %pa, <8 x float>* %pb) nounwind uwtable readnone ssp {
63; AVX1-LABEL: shuffle_v8f32_01230123_mem:
64; AVX1:       # %bb.0: # %entry
65; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[0,1,0,1]
66; AVX1-NEXT:    retq
67;
68; AVX2-LABEL: shuffle_v8f32_01230123_mem:
69; AVX2:       # %bb.0: # %entry
70; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = mem[0,1,0,1]
71; AVX2-NEXT:    retq
72entry:
73  %a = load <8 x float>, <8 x float>* %pa
74  %b = load <8 x float>, <8 x float>* %pb
75  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
76  ret <8 x float> %shuffle
77}
78
79define <8 x float> @shuffle_v8f32_45674567(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
80; AVX1-LABEL: shuffle_v8f32_45674567:
81; AVX1:       # %bb.0: # %entry
82; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
83; AVX1-NEXT:    retq
84;
85; AVX2-LABEL: shuffle_v8f32_45674567:
86; AVX2:       # %bb.0: # %entry
87; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
88; AVX2-NEXT:    retq
89entry:
90  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
91  ret <8 x float> %shuffle
92}
93
94define <8 x float> @shuffle_v8f32_45674567_mem(<8 x float>* %pa, <8 x float>* %pb) nounwind uwtable readnone ssp {
95; AVX1-LABEL: shuffle_v8f32_45674567_mem:
96; AVX1:       # %bb.0: # %entry
97; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3,2,3]
98; AVX1-NEXT:    retq
99;
100; AVX2-LABEL: shuffle_v8f32_45674567_mem:
101; AVX2:       # %bb.0: # %entry
102; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = mem[2,3,2,3]
103; AVX2-NEXT:    retq
104entry:
105  %a = load <8 x float>, <8 x float>* %pa
106  %b = load <8 x float>, <8 x float>* %pb
107  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
108  ret <8 x float> %shuffle
109}
110
111define <32 x i8> @shuffle_v32i8_2323(<32 x i8> %a, <32 x i8> %b) nounwind uwtable readnone ssp {
112; AVX1-LABEL: shuffle_v32i8_2323:
113; AVX1:       # %bb.0: # %entry
114; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
115; AVX1-NEXT:    retq
116;
117; AVX2-LABEL: shuffle_v32i8_2323:
118; AVX2:       # %bb.0: # %entry
119; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
120; AVX2-NEXT:    retq
121entry:
122  %shuffle = shufflevector <32 x i8> %a, <32 x i8> %b, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
123  ret <32 x i8> %shuffle
124}
125
126define <32 x i8> @shuffle_v32i8_2323_domain(<32 x i8> %a, <32 x i8> %b) nounwind uwtable readnone ssp {
127; AVX1-LABEL: shuffle_v32i8_2323_domain:
128; AVX1:       # %bb.0: # %entry
129; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
130; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
131; AVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
132; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
133; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
134; AVX1-NEXT:    retq
135;
136; AVX2-LABEL: shuffle_v32i8_2323_domain:
137; AVX2:       # %bb.0: # %entry
138; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
139; AVX2-NEXT:    vpsubb %ymm1, %ymm0, %ymm0
140; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
141; AVX2-NEXT:    retq
142entry:
143  ; add forces execution domain
144  %a2 = add <32 x i8> %a, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
145  %shuffle = shufflevector <32 x i8> %a2, <32 x i8> %b, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
146  ret <32 x i8> %shuffle
147}
148
149define <4 x i64> @shuffle_v4i64_6701(<4 x i64> %a, <4 x i64> %b) nounwind uwtable readnone ssp {
150; ALL-LABEL: shuffle_v4i64_6701:
151; ALL:       # %bb.0: # %entry
152; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[0,1]
153; ALL-NEXT:    retq
154entry:
155  %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
156  ret <4 x i64> %shuffle
157}
158
159define <4 x i64> @shuffle_v4i64_6701_domain(<4 x i64> %a, <4 x i64> %b) nounwind uwtable readnone ssp {
160; AVX1-LABEL: shuffle_v4i64_6701_domain:
161; AVX1:       # %bb.0: # %entry
162; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
163; AVX1-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
164; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[0,1]
165; AVX1-NEXT:    retq
166;
167; AVX2-LABEL: shuffle_v4i64_6701_domain:
168; AVX2:       # %bb.0: # %entry
169; AVX2-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
170; AVX2-NEXT:    vpsubq %ymm2, %ymm0, %ymm0
171; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[0,1]
172; AVX2-NEXT:    retq
173entry:
174  ; add forces execution domain
175  %a2 = add <4 x i64> %a, <i64 1, i64 1, i64 1, i64 1>
176  %shuffle = shufflevector <4 x i64> %a2, <4 x i64> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
177  ret <4 x i64> %shuffle
178}
179
180define <8 x i32> @shuffle_v8i32_u5u7cdef(<8 x i32> %a, <8 x i32> %b) nounwind uwtable readnone ssp {
181; AVX1-LABEL: shuffle_v8i32_u5u7cdef:
182; AVX1:       # %bb.0: # %entry
183; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
184; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
185; AVX1-NEXT:    vpsubd %xmm2, %xmm0, %xmm0
186; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
187; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
188; AVX1-NEXT:    retq
189;
190; AVX2-LABEL: shuffle_v8i32_u5u7cdef:
191; AVX2:       # %bb.0: # %entry
192; AVX2-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
193; AVX2-NEXT:    vpsubd %ymm2, %ymm0, %ymm0
194; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
195; AVX2-NEXT:    retq
196entry:
197  ; add forces execution domain
198  %a2 = add <8 x i32> %a, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
199  %shuffle = shufflevector <8 x i32> %a2, <8 x i32> %b, <8 x i32> <i32 undef, i32 5, i32 undef, i32 7, i32 12, i32 13, i32 14, i32 15>
200  ret <8 x i32> %shuffle
201}
202
203define <16 x i16> @shuffle_v16i16_4501(<16 x i16> %a, <16 x i16> %b) nounwind uwtable readnone ssp {
204; AVX1-LABEL: shuffle_v16i16_4501:
205; AVX1:       # %bb.0: # %entry
206; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
207; AVX1-NEXT:    vpsubw %xmm2, %xmm0, %xmm0
208; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
209; AVX1-NEXT:    retq
210;
211; AVX2-LABEL: shuffle_v16i16_4501:
212; AVX2:       # %bb.0: # %entry
213; AVX2-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
214; AVX2-NEXT:    vpsubw %ymm2, %ymm0, %ymm0
215; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
216; AVX2-NEXT:    retq
217entry:
218  ; add forces execution domain
219  %a2 = add <16 x i16> %a, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
220  %shuffle = shufflevector <16 x i16> %a2, <16 x i16> %b, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
221  ret <16 x i16> %shuffle
222}
223
224define <16 x i16> @shuffle_v16i16_4501_mem(<16 x i16>* %a, <16 x i16>* %b) nounwind uwtable readnone ssp {
225; AVX1-LABEL: shuffle_v16i16_4501_mem:
226; AVX1:       # %bb.0: # %entry
227; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
228; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
229; AVX1-NEXT:    vpsubw %xmm1, %xmm0, %xmm0
230; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[0,1],ymm0[0,1]
231; AVX1-NEXT:    retq
232;
233; AVX2-LABEL: shuffle_v16i16_4501_mem:
234; AVX2:       # %bb.0: # %entry
235; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
236; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
237; AVX2-NEXT:    vpsubw %ymm1, %ymm0, %ymm0
238; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = mem[0,1],ymm0[0,1]
239; AVX2-NEXT:    retq
240entry:
241  %c = load <16 x i16>, <16 x i16>* %a
242  %d = load <16 x i16>, <16 x i16>* %b
243  %c2 = add <16 x i16> %c, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
244  %shuffle = shufflevector <16 x i16> %c2, <16 x i16> %d, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
245  ret <16 x i16> %shuffle
246}
247
248;;;; Cases with undef indicies mixed in the mask
249
250define <8 x float> @shuffle_v8f32_uu67u9ub(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
251; ALL-LABEL: shuffle_v8f32_uu67u9ub:
252; ALL:       # %bb.0: # %entry
253; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[0,1]
254; ALL-NEXT:    retq
255entry:
256  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 9, i32 undef, i32 11>
257  ret <8 x float> %shuffle
258}
259
260define <8 x float> @shuffle_v8f32_uu67uu67(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
261; AVX1-LABEL: shuffle_v8f32_uu67uu67:
262; AVX1:       # %bb.0: # %entry
263; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
264; AVX1-NEXT:    retq
265;
266; AVX2-LABEL: shuffle_v8f32_uu67uu67:
267; AVX2:       # %bb.0: # %entry
268; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
269; AVX2-NEXT:    retq
270entry:
271  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 undef, i32 6, i32 7>
272  ret <8 x float> %shuffle
273}
274
275define <8 x float> @shuffle_v8f32_uu67uuab(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
276; ALL-LABEL: shuffle_v8f32_uu67uuab:
277; ALL:       # %bb.0: # %entry
278; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[0,1]
279; ALL-NEXT:    retq
280entry:
281  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 undef, i32 10, i32 11>
282  ret <8 x float> %shuffle
283}
284
285define <8 x float> @shuffle_v8f32_uu67uuef(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
286; ALL-LABEL: shuffle_v8f32_uu67uuef:
287; ALL:       # %bb.0: # %entry
288; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
289; ALL-NEXT:    retq
290entry:
291  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 undef, i32 14, i32 15>
292  ret <8 x float> %shuffle
293}
294
295define <8 x float> @shuffle_v8f32_uu674567(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
296; AVX1-LABEL: shuffle_v8f32_uu674567:
297; AVX1:       # %bb.0: # %entry
298; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
299; AVX1-NEXT:    retq
300;
301; AVX2-LABEL: shuffle_v8f32_uu674567:
302; AVX2:       # %bb.0: # %entry
303; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
304; AVX2-NEXT:    retq
305entry:
306  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
307  ret <8 x float> %shuffle
308}
309
310define <8 x float> @shuffle_v8f32_uu6789ab(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
311; ALL-LABEL: shuffle_v8f32_uu6789ab:
312; ALL:       # %bb.0: # %entry
313; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[0,1]
314; ALL-NEXT:    retq
315entry:
316  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>
317  ret <8 x float> %shuffle
318}
319
320define <8 x float> @shuffle_v8f32_4567uu67(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
321; AVX1-LABEL: shuffle_v8f32_4567uu67:
322; AVX1:       # %bb.0: # %entry
323; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]
324; AVX1-NEXT:    retq
325;
326; AVX2-LABEL: shuffle_v8f32_4567uu67:
327; AVX2:       # %bb.0: # %entry
328; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
329; AVX2-NEXT:    retq
330entry:
331  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 6, i32 7>
332  ret <8 x float> %shuffle
333}
334
335define <8 x float> @shuffle_v8f32_4567uuef(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
336; ALL-LABEL: shuffle_v8f32_4567uuef:
337; ALL:       # %bb.0: # %entry
338; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
339; ALL-NEXT:    retq
340entry:
341  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 14, i32 15>
342  ret <8 x float> %shuffle
343}
344
345;;;; Cases we must not select vperm2f128
346
347define <8 x float> @shuffle_v8f32_uu67ucuf(<8 x float> %a, <8 x float> %b) nounwind uwtable readnone ssp {
348; ALL-LABEL: shuffle_v8f32_uu67ucuf:
349; ALL:       # %bb.0: # %entry
350; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
351; ALL-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,3,4,4,6,7]
352; ALL-NEXT:    retq
353entry:
354  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 12, i32 undef, i32 15>
355  ret <8 x float> %shuffle
356}
357
358;; Test zero mask generation.
359;; PR22984: https://llvm.org/bugs/show_bug.cgi?id=22984
360;; Prefer xor+vblendpd over vperm2f128 because that has better performance.
361;; TODO: When building for optsize we should use vperm2f128.
362
363define <4 x double> @shuffle_v4f64_zz01(<4 x double> %a) {
364; ALL-LABEL: shuffle_v4f64_zz01:
365; ALL:       # %bb.0:
366; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]
367; ALL-NEXT:    retq
368  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
369  ret <4 x double> %s
370}
371define <4 x double> @shuffle_v4f64_zz01_optsize(<4 x double> %a) optsize {
372; ALL-LABEL: shuffle_v4f64_zz01_optsize:
373; ALL:       # %bb.0:
374; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]
375; ALL-NEXT:    retq
376  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
377  ret <4 x double> %s
378}
379
380define <4 x double> @shuffle_v4f64_zz23(<4 x double> %a) {
381; ALL-LABEL: shuffle_v4f64_zz23:
382; ALL:       # %bb.0:
383; ALL-NEXT:    vxorps %xmm1, %xmm1, %xmm1
384; ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
385; ALL-NEXT:    retq
386  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
387  ret <4 x double> %s
388}
389define <4 x double> @shuffle_v4f64_zz23_optsize(<4 x double> %a) optsize {
390; ALL-LABEL: shuffle_v4f64_zz23_optsize:
391; ALL:       # %bb.0:
392; ALL-NEXT:    vxorps %xmm1, %xmm1, %xmm1
393; ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
394; ALL-NEXT:    retq
395  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
396  ret <4 x double> %s
397}
398
399define <4 x double> @shuffle_v4f64_zz45(<4 x double> %a) {
400; ALL-LABEL: shuffle_v4f64_zz45:
401; ALL:       # %bb.0:
402; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]
403; ALL-NEXT:    retq
404  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
405  ret <4 x double> %s
406}
407define <4 x double> @shuffle_v4f64_zz45_optsize(<4 x double> %a) optsize {
408; ALL-LABEL: shuffle_v4f64_zz45_optsize:
409; ALL:       # %bb.0:
410; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]
411; ALL-NEXT:    retq
412  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
413  ret <4 x double> %s
414}
415
416define <4 x double> @shuffle_v4f64_zz67(<4 x double> %a) {
417; ALL-LABEL: shuffle_v4f64_zz67:
418; ALL:       # %bb.0:
419; ALL-NEXT:    vxorps %xmm1, %xmm1, %xmm1
420; ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
421; ALL-NEXT:    retq
422  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 0, i32 1, i32 6, i32 7>
423  ret <4 x double> %s
424}
425define <4 x double> @shuffle_v4f64_zz67_optsize(<4 x double> %a) optsize {
426; ALL-LABEL: shuffle_v4f64_zz67_optsize:
427; ALL:       # %bb.0:
428; ALL-NEXT:    vxorps %xmm1, %xmm1, %xmm1
429; ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
430; ALL-NEXT:    retq
431  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 0, i32 1, i32 6, i32 7>
432  ret <4 x double> %s
433}
434
435define <4 x double> @shuffle_v4f64_01zz(<4 x double> %a) {
436; ALL-LABEL: shuffle_v4f64_01zz:
437; ALL:       # %bb.0:
438; ALL-NEXT:    vmovaps %xmm0, %xmm0
439; ALL-NEXT:    retq
440  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
441  ret <4 x double> %s
442}
443define <4 x double> @shuffle_v4f64_01zz_optsize(<4 x double> %a) optsize {
444; ALL-LABEL: shuffle_v4f64_01zz_optsize:
445; ALL:       # %bb.0:
446; ALL-NEXT:    vmovaps %xmm0, %xmm0
447; ALL-NEXT:    retq
448  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
449  ret <4 x double> %s
450}
451
452define <4 x double> @shuffle_v4f64_23zz(<4 x double> %a) {
453; ALL-LABEL: shuffle_v4f64_23zz:
454; ALL:       # %bb.0:
455; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],zero,zero
456; ALL-NEXT:    retq
457  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
458  ret <4 x double> %s
459}
460define <4 x double> @shuffle_v4f64_23zz_optsize(<4 x double> %a) optsize {
461; ALL-LABEL: shuffle_v4f64_23zz_optsize:
462; ALL:       # %bb.0:
463; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],zero,zero
464; ALL-NEXT:    retq
465  %s = shufflevector <4 x double> %a, <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
466  ret <4 x double> %s
467}
468
469define <4 x double> @shuffle_v4f64_45zz(<4 x double> %a) {
470; ALL-LABEL: shuffle_v4f64_45zz:
471; ALL:       # %bb.0:
472; ALL-NEXT:    vmovaps %xmm0, %xmm0
473; ALL-NEXT:    retq
474  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
475  ret <4 x double> %s
476}
477define <4 x double> @shuffle_v4f64_45zz_optsize(<4 x double> %a) optsize {
478; ALL-LABEL: shuffle_v4f64_45zz_optsize:
479; ALL:       # %bb.0:
480; ALL-NEXT:    vmovaps %xmm0, %xmm0
481; ALL-NEXT:    retq
482  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
483  ret <4 x double> %s
484}
485
486define <4 x double> @shuffle_v4f64_67zz(<4 x double> %a) {
487; ALL-LABEL: shuffle_v4f64_67zz:
488; ALL:       # %bb.0:
489; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],zero,zero
490; ALL-NEXT:    retq
491  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
492  ret <4 x double> %s
493}
494define <4 x double> @shuffle_v4f64_67zz_optsize(<4 x double> %a) optsize {
495; ALL-LABEL: shuffle_v4f64_67zz_optsize:
496; ALL:       # %bb.0:
497; ALL-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],zero,zero
498; ALL-NEXT:    retq
499  %s = shufflevector <4 x double> <double 0.0, double 0.0, double undef, double undef>, <4 x double> %a, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
500  ret <4 x double> %s
501}
502
503;; With AVX2 select the integer version of the instruction. Use an add to force the domain selection.
504
505define <4 x i64> @shuffle_v4i64_67zz(<4 x i64> %a, <4 x i64> %b) {
506; AVX1-LABEL: shuffle_v4i64_67zz:
507; AVX1:       # %bb.0:
508; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
509; AVX1-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
510; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
511; AVX1-NEXT:    retq
512;
513; AVX2-LABEL: shuffle_v4i64_67zz:
514; AVX2:       # %bb.0:
515; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],zero,zero
516; AVX2-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
517; AVX2-NEXT:    retq
518  %s = shufflevector <4 x i64> <i64 0, i64 0, i64 undef, i64 undef>, <4 x i64> %a, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
519  %c = add <4 x i64> %b, %s
520  ret <4 x i64> %c
521}
522
523;;; Memory folding cases
524
525define <4 x double> @ld0_hi0_lo1_4f64(<4 x double> * %pa, <4 x double> %b) nounwind uwtable readnone ssp {
526; AVX1-LABEL: ld0_hi0_lo1_4f64:
527; AVX1:       # %bb.0: # %entry
528; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
529; AVX1-NEXT:    vaddpd {{.*}}(%rip), %ymm0, %ymm0
530; AVX1-NEXT:    retq
531;
532; AVX2-LABEL: ld0_hi0_lo1_4f64:
533; AVX2:       # %bb.0: # %entry
534; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
535; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [1,1,1,1]
536; AVX2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0
537; AVX2-NEXT:    retq
538entry:
539  %a = load <4 x double>, <4 x double> * %pa
540  %shuffle = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
541  %res = fadd <4 x double> %shuffle, <double 1.0, double 1.0, double 1.0, double 1.0>
542  ret <4 x double> %res
543}
544
545define <4 x double> @ld1_hi0_hi1_4f64(<4 x double> %a, <4 x double> * %pb) nounwind uwtable readnone ssp {
546; AVX1-LABEL: ld1_hi0_hi1_4f64:
547; AVX1:       # %bb.0: # %entry
548; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
549; AVX1-NEXT:    vaddpd {{.*}}(%rip), %ymm0, %ymm0
550; AVX1-NEXT:    retq
551;
552; AVX2-LABEL: ld1_hi0_hi1_4f64:
553; AVX2:       # %bb.0: # %entry
554; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
555; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [1,1,1,1]
556; AVX2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0
557; AVX2-NEXT:    retq
558entry:
559  %b = load <4 x double>, <4 x double> * %pb
560  %shuffle = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
561  %res = fadd <4 x double> %shuffle, <double 1.0, double 1.0, double 1.0, double 1.0>
562  ret <4 x double> %res
563}
564
565define <8 x float> @ld0_hi0_lo1_8f32(<8 x float> * %pa, <8 x float> %b) nounwind uwtable readnone ssp {
566; AVX1-LABEL: ld0_hi0_lo1_8f32:
567; AVX1:       # %bb.0: # %entry
568; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
569; AVX1-NEXT:    vaddps {{.*}}(%rip), %ymm0, %ymm0
570; AVX1-NEXT:    retq
571;
572; AVX2-LABEL: ld0_hi0_lo1_8f32:
573; AVX2:       # %bb.0: # %entry
574; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
575; AVX2-NEXT:    vbroadcastss {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1]
576; AVX2-NEXT:    vaddps %ymm1, %ymm0, %ymm0
577; AVX2-NEXT:    retq
578entry:
579  %a = load <8 x float>, <8 x float> * %pa
580  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>
581  %res = fadd <8 x float> %shuffle, <float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0>
582  ret <8 x float> %res
583}
584
585define <8 x float> @ld1_hi0_hi1_8f32(<8 x float> %a, <8 x float> * %pb) nounwind uwtable readnone ssp {
586; AVX1-LABEL: ld1_hi0_hi1_8f32:
587; AVX1:       # %bb.0: # %entry
588; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
589; AVX1-NEXT:    vaddps {{.*}}(%rip), %ymm0, %ymm0
590; AVX1-NEXT:    retq
591;
592; AVX2-LABEL: ld1_hi0_hi1_8f32:
593; AVX2:       # %bb.0: # %entry
594; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
595; AVX2-NEXT:    vbroadcastss {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1]
596; AVX2-NEXT:    vaddps %ymm1, %ymm0, %ymm0
597; AVX2-NEXT:    retq
598entry:
599  %b = load <8 x float>, <8 x float> * %pb
600  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 12, i32 13, i32 14, i32 15>
601  %res = fadd <8 x float> %shuffle, <float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0>
602  ret <8 x float> %res
603}
604
605define <4 x i64> @ld0_hi0_lo1_4i64(<4 x i64> * %pa, <4 x i64> %b) nounwind uwtable readnone ssp {
606; AVX1-LABEL: ld0_hi0_lo1_4i64:
607; AVX1:       # %bb.0: # %entry
608; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
609; AVX1-NEXT:    vpaddq {{.*}}(%rip), %xmm0, %xmm1
610; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
611; AVX1-NEXT:    vpaddq {{.*}}(%rip), %xmm0, %xmm0
612; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
613; AVX1-NEXT:    retq
614;
615; AVX2-LABEL: ld0_hi0_lo1_4i64:
616; AVX2:       # %bb.0: # %entry
617; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
618; AVX2-NEXT:    vpaddq {{.*}}(%rip), %ymm0, %ymm0
619; AVX2-NEXT:    retq
620entry:
621  %a = load <4 x i64>, <4 x i64> * %pa
622  %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
623  %res = add <4 x i64> %shuffle, <i64 1, i64 2, i64 3, i64 4>
624  ret <4 x i64> %res
625}
626
627define <4 x i64> @ld1_hi0_hi1_4i64(<4 x i64> %a, <4 x i64> * %pb) nounwind uwtable readnone ssp {
628; AVX1-LABEL: ld1_hi0_hi1_4i64:
629; AVX1:       # %bb.0: # %entry
630; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
631; AVX1-NEXT:    vpaddq {{.*}}(%rip), %xmm0, %xmm1
632; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
633; AVX1-NEXT:    vpaddq {{.*}}(%rip), %xmm0, %xmm0
634; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
635; AVX1-NEXT:    retq
636;
637; AVX2-LABEL: ld1_hi0_hi1_4i64:
638; AVX2:       # %bb.0: # %entry
639; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
640; AVX2-NEXT:    vpaddq {{.*}}(%rip), %ymm0, %ymm0
641; AVX2-NEXT:    retq
642entry:
643  %b = load <4 x i64>, <4 x i64> * %pb
644  %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
645  %res = add <4 x i64> %shuffle, <i64 1, i64 2, i64 3, i64 4>
646  ret <4 x i64> %res
647}
648
649define <8 x i32> @ld0_hi0_lo1_8i32(<8 x i32> * %pa, <8 x i32> %b) nounwind uwtable readnone ssp {
650; AVX1-LABEL: ld0_hi0_lo1_8i32:
651; AVX1:       # %bb.0: # %entry
652; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
653; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
654; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,2,3,4]
655; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
656; AVX1-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
657; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
658; AVX1-NEXT:    retq
659;
660; AVX2-LABEL: ld0_hi0_lo1_8i32:
661; AVX2:       # %bb.0: # %entry
662; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
663; AVX2-NEXT:    vpaddd {{.*}}(%rip), %ymm0, %ymm0
664; AVX2-NEXT:    retq
665entry:
666  %a = load <8 x i32>, <8 x i32> * %pa
667  %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>
668  %res = add <8 x i32> %shuffle, <i32 1, i32 2, i32 3, i32 4, i32 1, i32 2, i32 3, i32 4>
669  ret <8 x i32> %res
670}
671
672define <8 x i32> @ld1_hi0_hi1_8i32(<8 x i32> %a, <8 x i32> * %pb) nounwind uwtable readnone ssp {
673; AVX1-LABEL: ld1_hi0_hi1_8i32:
674; AVX1:       # %bb.0: # %entry
675; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
676; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
677; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,2,3,4]
678; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
679; AVX1-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
680; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
681; AVX1-NEXT:    retq
682;
683; AVX2-LABEL: ld1_hi0_hi1_8i32:
684; AVX2:       # %bb.0: # %entry
685; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],mem[2,3]
686; AVX2-NEXT:    vpaddd {{.*}}(%rip), %ymm0, %ymm0
687; AVX2-NEXT:    retq
688entry:
689  %b = load <8 x i32>, <8 x i32> * %pb
690  %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 12, i32 13, i32 14, i32 15>
691  %res = add <8 x i32> %shuffle, <i32 1, i32 2, i32 3, i32 4, i32 1, i32 2, i32 3, i32 4>
692  ret <8 x i32> %res
693}
694