1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE2
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+sse3 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE3
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+ssse3 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSSE3
4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE41
5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
6 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
8 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
9 target triple = "x86_64-unknown-unknown"
11 define <2 x i64> @shuffle_v2i64_00(<2 x i64> %a, <2 x i64> %b) {
12 ; SSE-LABEL: shuffle_v2i64_00:
14 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
17 ; AVX1-LABEL: shuffle_v2i64_00:
19 ; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
22 ; AVX2-LABEL: shuffle_v2i64_00:
24 ; AVX2-NEXT: vpbroadcastq %xmm0, %xmm0
26 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 0>
27 ret <2 x i64> %shuffle
29 define <2 x i64> @shuffle_v2i64_10(<2 x i64> %a, <2 x i64> %b) {
30 ; SSE-LABEL: shuffle_v2i64_10:
32 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
35 ; AVX-LABEL: shuffle_v2i64_10:
37 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
39 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 0>
40 ret <2 x i64> %shuffle
42 define <2 x i64> @shuffle_v2i64_11(<2 x i64> %a, <2 x i64> %b) {
43 ; SSE-LABEL: shuffle_v2i64_11:
45 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
48 ; AVX-LABEL: shuffle_v2i64_11:
50 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
52 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 1>
53 ret <2 x i64> %shuffle
55 define <2 x i64> @shuffle_v2i64_22(<2 x i64> %a, <2 x i64> %b) {
56 ; SSE-LABEL: shuffle_v2i64_22:
58 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
61 ; AVX1-LABEL: shuffle_v2i64_22:
63 ; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
66 ; AVX2-LABEL: shuffle_v2i64_22:
68 ; AVX2-NEXT: vpbroadcastq %xmm1, %xmm0
70 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 2>
71 ret <2 x i64> %shuffle
73 define <2 x i64> @shuffle_v2i64_32(<2 x i64> %a, <2 x i64> %b) {
74 ; SSE-LABEL: shuffle_v2i64_32:
76 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
79 ; AVX-LABEL: shuffle_v2i64_32:
81 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
83 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 2>
84 ret <2 x i64> %shuffle
86 define <2 x i64> @shuffle_v2i64_33(<2 x i64> %a, <2 x i64> %b) {
87 ; SSE-LABEL: shuffle_v2i64_33:
89 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
92 ; AVX-LABEL: shuffle_v2i64_33:
94 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
96 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 3>
97 ret <2 x i64> %shuffle
100 define <2 x double> @shuffle_v2f64_00(<2 x double> %a, <2 x double> %b) {
101 ; SSE2-LABEL: shuffle_v2f64_00:
103 ; SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0,0]
106 ; SSE3-LABEL: shuffle_v2f64_00:
108 ; SSE3-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0]
111 ; SSSE3-LABEL: shuffle_v2f64_00:
113 ; SSSE3-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0]
116 ; SSE41-LABEL: shuffle_v2f64_00:
118 ; SSE41-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0]
121 ; AVX-LABEL: shuffle_v2f64_00:
123 ; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
125 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 0, i32 0>
126 ret <2 x double> %shuffle
128 define <2 x double> @shuffle_v2f64_10(<2 x double> %a, <2 x double> %b) {
129 ; SSE-LABEL: shuffle_v2f64_10:
131 ; SSE-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1,0]
134 ; AVX-LABEL: shuffle_v2f64_10:
136 ; AVX-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
138 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 1, i32 0>
139 ret <2 x double> %shuffle
141 define <2 x double> @shuffle_v2f64_11(<2 x double> %a, <2 x double> %b) {
142 ; SSE-LABEL: shuffle_v2f64_11:
144 ; SSE-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
147 ; AVX-LABEL: shuffle_v2f64_11:
149 ; AVX-NEXT: vmovhlps {{.*#+}} xmm0 = xmm0[1,1]
151 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 1, i32 1>
152 ret <2 x double> %shuffle
154 define <2 x double> @shuffle_v2f64_22(<2 x double> %a, <2 x double> %b) {
155 ; SSE2-LABEL: shuffle_v2f64_22:
157 ; SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0,0]
158 ; SSE2-NEXT: movaps %xmm1, %xmm0
161 ; SSE3-LABEL: shuffle_v2f64_22:
163 ; SSE3-NEXT: movddup {{.*#+}} xmm0 = xmm1[0,0]
166 ; SSSE3-LABEL: shuffle_v2f64_22:
168 ; SSSE3-NEXT: movddup {{.*#+}} xmm0 = xmm1[0,0]
171 ; SSE41-LABEL: shuffle_v2f64_22:
173 ; SSE41-NEXT: movddup {{.*#+}} xmm0 = xmm1[0,0]
176 ; AVX-LABEL: shuffle_v2f64_22:
178 ; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm1[0,0]
180 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 2, i32 2>
181 ret <2 x double> %shuffle
183 define <2 x double> @shuffle_v2f64_32(<2 x double> %a, <2 x double> %b) {
184 ; SSE-LABEL: shuffle_v2f64_32:
186 ; SSE-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1,0]
187 ; SSE-NEXT: movapd %xmm1, %xmm0
190 ; AVX-LABEL: shuffle_v2f64_32:
192 ; AVX-NEXT: vpermilpd {{.*#+}} xmm0 = xmm1[1,0]
194 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 3, i32 2>
195 ret <2 x double> %shuffle
197 define <2 x double> @shuffle_v2f64_33(<2 x double> %a, <2 x double> %b) {
198 ; SSE-LABEL: shuffle_v2f64_33:
200 ; SSE-NEXT: movhlps {{.*#+}} xmm1 = xmm1[1,1]
201 ; SSE-NEXT: movaps %xmm1, %xmm0
204 ; AVX-LABEL: shuffle_v2f64_33:
206 ; AVX-NEXT: vmovhlps {{.*#+}} xmm0 = xmm1[1,1]
208 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 3, i32 3>
209 ret <2 x double> %shuffle
211 define <2 x double> @shuffle_v2f64_03(<2 x double> %a, <2 x double> %b) {
212 ; SSE2-LABEL: shuffle_v2f64_03:
214 ; SSE2-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
215 ; SSE2-NEXT: movapd %xmm1, %xmm0
218 ; SSE3-LABEL: shuffle_v2f64_03:
220 ; SSE3-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
221 ; SSE3-NEXT: movapd %xmm1, %xmm0
224 ; SSSE3-LABEL: shuffle_v2f64_03:
226 ; SSSE3-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
227 ; SSSE3-NEXT: movapd %xmm1, %xmm0
230 ; SSE41-LABEL: shuffle_v2f64_03:
232 ; SSE41-NEXT: blendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
235 ; AVX-LABEL: shuffle_v2f64_03:
237 ; AVX-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
239 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 0, i32 3>
240 ret <2 x double> %shuffle
242 define <2 x double> @shuffle_v2f64_21(<2 x double> %a, <2 x double> %b) {
243 ; SSE2-LABEL: shuffle_v2f64_21:
245 ; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
248 ; SSE3-LABEL: shuffle_v2f64_21:
250 ; SSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
253 ; SSSE3-LABEL: shuffle_v2f64_21:
255 ; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
258 ; SSE41-LABEL: shuffle_v2f64_21:
260 ; SSE41-NEXT: blendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
263 ; AVX-LABEL: shuffle_v2f64_21:
265 ; AVX-NEXT: vblendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
267 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 2, i32 1>
268 ret <2 x double> %shuffle
272 define <2 x i64> @shuffle_v2i64_02(<2 x i64> %a, <2 x i64> %b) {
273 ; SSE-LABEL: shuffle_v2i64_02:
275 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
278 ; AVX-LABEL: shuffle_v2i64_02:
280 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
282 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 2>
283 ret <2 x i64> %shuffle
285 define <2 x i64> @shuffle_v2i64_02_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
286 ; SSE-LABEL: shuffle_v2i64_02_copy:
288 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
289 ; SSE-NEXT: movdqa %xmm1, %xmm0
292 ; AVX-LABEL: shuffle_v2i64_02_copy:
294 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm2[0]
296 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 2>
297 ret <2 x i64> %shuffle
299 define <2 x i64> @shuffle_v2i64_03(<2 x i64> %a, <2 x i64> %b) {
300 ; SSE2-LABEL: shuffle_v2i64_03:
302 ; SSE2-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
303 ; SSE2-NEXT: movapd %xmm1, %xmm0
306 ; SSE3-LABEL: shuffle_v2i64_03:
308 ; SSE3-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
309 ; SSE3-NEXT: movapd %xmm1, %xmm0
312 ; SSSE3-LABEL: shuffle_v2i64_03:
314 ; SSSE3-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
315 ; SSSE3-NEXT: movapd %xmm1, %xmm0
318 ; SSE41-LABEL: shuffle_v2i64_03:
320 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
323 ; AVX1-LABEL: shuffle_v2i64_03:
325 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
328 ; AVX2-LABEL: shuffle_v2i64_03:
330 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
332 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
333 ret <2 x i64> %shuffle
335 define <2 x i64> @shuffle_v2i64_03_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
336 ; SSE2-LABEL: shuffle_v2i64_03_copy:
338 ; SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
339 ; SSE2-NEXT: movapd %xmm2, %xmm0
342 ; SSE3-LABEL: shuffle_v2i64_03_copy:
344 ; SSE3-NEXT: movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
345 ; SSE3-NEXT: movapd %xmm2, %xmm0
348 ; SSSE3-LABEL: shuffle_v2i64_03_copy:
350 ; SSSE3-NEXT: movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
351 ; SSSE3-NEXT: movapd %xmm2, %xmm0
354 ; SSE41-LABEL: shuffle_v2i64_03_copy:
356 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7]
357 ; SSE41-NEXT: movdqa %xmm1, %xmm0
360 ; AVX1-LABEL: shuffle_v2i64_03_copy:
362 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm2[4,5,6,7]
365 ; AVX2-LABEL: shuffle_v2i64_03_copy:
367 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm2[2,3]
369 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
370 ret <2 x i64> %shuffle
372 define <2 x i64> @shuffle_v2i64_12(<2 x i64> %a, <2 x i64> %b) {
373 ; SSE2-LABEL: shuffle_v2i64_12:
375 ; SSE2-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
378 ; SSE3-LABEL: shuffle_v2i64_12:
380 ; SSE3-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
383 ; SSSE3-LABEL: shuffle_v2i64_12:
385 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
386 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
389 ; SSE41-LABEL: shuffle_v2i64_12:
391 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
392 ; SSE41-NEXT: movdqa %xmm1, %xmm0
395 ; AVX-LABEL: shuffle_v2i64_12:
397 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
399 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 2>
400 ret <2 x i64> %shuffle
402 define <2 x i64> @shuffle_v2i64_12_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
403 ; SSE2-LABEL: shuffle_v2i64_12_copy:
405 ; SSE2-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm2[0]
406 ; SSE2-NEXT: movapd %xmm1, %xmm0
409 ; SSE3-LABEL: shuffle_v2i64_12_copy:
411 ; SSE3-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm2[0]
412 ; SSE3-NEXT: movapd %xmm1, %xmm0
415 ; SSSE3-LABEL: shuffle_v2i64_12_copy:
417 ; SSSE3-NEXT: palignr {{.*#+}} xmm2 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
418 ; SSSE3-NEXT: movdqa %xmm2, %xmm0
421 ; SSE41-LABEL: shuffle_v2i64_12_copy:
423 ; SSE41-NEXT: palignr {{.*#+}} xmm2 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
424 ; SSE41-NEXT: movdqa %xmm2, %xmm0
427 ; AVX-LABEL: shuffle_v2i64_12_copy:
429 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
431 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 2>
432 ret <2 x i64> %shuffle
434 define <2 x i64> @shuffle_v2i64_13(<2 x i64> %a, <2 x i64> %b) {
435 ; SSE-LABEL: shuffle_v2i64_13:
437 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
440 ; AVX-LABEL: shuffle_v2i64_13:
442 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
444 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 3>
445 ret <2 x i64> %shuffle
447 define <2 x i64> @shuffle_v2i64_13_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
448 ; SSE-LABEL: shuffle_v2i64_13_copy:
450 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm2[1]
451 ; SSE-NEXT: movdqa %xmm1, %xmm0
454 ; AVX-LABEL: shuffle_v2i64_13_copy:
456 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm1[1],xmm2[1]
458 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 3>
459 ret <2 x i64> %shuffle
461 define <2 x i64> @shuffle_v2i64_20(<2 x i64> %a, <2 x i64> %b) {
462 ; SSE-LABEL: shuffle_v2i64_20:
464 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
465 ; SSE-NEXT: movdqa %xmm1, %xmm0
468 ; AVX-LABEL: shuffle_v2i64_20:
470 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
472 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
473 ret <2 x i64> %shuffle
475 define <2 x i64> @shuffle_v2i64_20_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
476 ; SSE-LABEL: shuffle_v2i64_20_copy:
478 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0]
479 ; SSE-NEXT: movdqa %xmm2, %xmm0
482 ; AVX-LABEL: shuffle_v2i64_20_copy:
484 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm1[0]
486 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
487 ret <2 x i64> %shuffle
489 define <2 x i64> @shuffle_v2i64_21(<2 x i64> %a, <2 x i64> %b) {
490 ; SSE2-LABEL: shuffle_v2i64_21:
492 ; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
495 ; SSE3-LABEL: shuffle_v2i64_21:
497 ; SSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
500 ; SSSE3-LABEL: shuffle_v2i64_21:
502 ; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
505 ; SSE41-LABEL: shuffle_v2i64_21:
507 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
510 ; AVX1-LABEL: shuffle_v2i64_21:
512 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
515 ; AVX2-LABEL: shuffle_v2i64_21:
517 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
519 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 1>
520 ret <2 x i64> %shuffle
522 define <2 x i64> @shuffle_v2i64_21_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
523 ; SSE2-LABEL: shuffle_v2i64_21_copy:
525 ; SSE2-NEXT: movsd {{.*#+}} xmm1 = xmm2[0],xmm1[1]
526 ; SSE2-NEXT: movapd %xmm1, %xmm0
529 ; SSE3-LABEL: shuffle_v2i64_21_copy:
531 ; SSE3-NEXT: movsd {{.*#+}} xmm1 = xmm2[0],xmm1[1]
532 ; SSE3-NEXT: movapd %xmm1, %xmm0
535 ; SSSE3-LABEL: shuffle_v2i64_21_copy:
537 ; SSSE3-NEXT: movsd {{.*#+}} xmm1 = xmm2[0],xmm1[1]
538 ; SSSE3-NEXT: movapd %xmm1, %xmm0
541 ; SSE41-LABEL: shuffle_v2i64_21_copy:
543 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
544 ; SSE41-NEXT: movdqa %xmm1, %xmm0
547 ; AVX1-LABEL: shuffle_v2i64_21_copy:
549 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm1[4,5,6,7]
552 ; AVX2-LABEL: shuffle_v2i64_21_copy:
554 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0,1],xmm1[2,3]
556 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 1>
557 ret <2 x i64> %shuffle
559 define <2 x i64> @shuffle_v2i64_30(<2 x i64> %a, <2 x i64> %b) {
560 ; SSE2-LABEL: shuffle_v2i64_30:
562 ; SSE2-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
563 ; SSE2-NEXT: movapd %xmm1, %xmm0
566 ; SSE3-LABEL: shuffle_v2i64_30:
568 ; SSE3-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
569 ; SSE3-NEXT: movapd %xmm1, %xmm0
572 ; SSSE3-LABEL: shuffle_v2i64_30:
574 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
577 ; SSE41-LABEL: shuffle_v2i64_30:
579 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
582 ; AVX-LABEL: shuffle_v2i64_30:
584 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
586 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 0>
587 ret <2 x i64> %shuffle
589 define <2 x i64> @shuffle_v2i64_30_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
590 ; SSE2-LABEL: shuffle_v2i64_30_copy:
592 ; SSE2-NEXT: shufpd {{.*#+}} xmm2 = xmm2[1],xmm1[0]
593 ; SSE2-NEXT: movapd %xmm2, %xmm0
596 ; SSE3-LABEL: shuffle_v2i64_30_copy:
598 ; SSE3-NEXT: shufpd {{.*#+}} xmm2 = xmm2[1],xmm1[0]
599 ; SSE3-NEXT: movapd %xmm2, %xmm0
602 ; SSSE3-LABEL: shuffle_v2i64_30_copy:
604 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
605 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
608 ; SSE41-LABEL: shuffle_v2i64_30_copy:
610 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
611 ; SSE41-NEXT: movdqa %xmm1, %xmm0
614 ; AVX-LABEL: shuffle_v2i64_30_copy:
616 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
618 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 0>
619 ret <2 x i64> %shuffle
621 define <2 x i64> @shuffle_v2i64_31(<2 x i64> %a, <2 x i64> %b) {
622 ; SSE-LABEL: shuffle_v2i64_31:
624 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm0[1]
625 ; SSE-NEXT: movdqa %xmm1, %xmm0
628 ; AVX-LABEL: shuffle_v2i64_31:
630 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm1[1],xmm0[1]
632 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 1>
633 ret <2 x i64> %shuffle
635 define <2 x i64> @shuffle_v2i64_31_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
636 ; SSE-LABEL: shuffle_v2i64_31_copy:
638 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm1[1]
639 ; SSE-NEXT: movdqa %xmm2, %xmm0
642 ; AVX-LABEL: shuffle_v2i64_31_copy:
644 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm2[1],xmm1[1]
646 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 1>
647 ret <2 x i64> %shuffle
650 define <2 x i64> @shuffle_v2i64_0z(<2 x i64> %a) {
651 ; SSE-LABEL: shuffle_v2i64_0z:
653 ; SSE-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
656 ; AVX-LABEL: shuffle_v2i64_0z:
658 ; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
660 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 0, i32 3>
661 ret <2 x i64> %shuffle
664 define <2 x i64> @shuffle_v2i64_1z(<2 x i64> %a) {
665 ; SSE-LABEL: shuffle_v2i64_1z:
667 ; SSE-NEXT: psrldq {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero
670 ; AVX-LABEL: shuffle_v2i64_1z:
672 ; AVX-NEXT: vpsrldq {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero
674 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 1, i32 3>
675 ret <2 x i64> %shuffle
678 define <2 x i64> @shuffle_v2i64_z0(<2 x i64> %a) {
679 ; SSE-LABEL: shuffle_v2i64_z0:
681 ; SSE-NEXT: pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7]
684 ; AVX-LABEL: shuffle_v2i64_z0:
686 ; AVX-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7]
688 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 2, i32 0>
689 ret <2 x i64> %shuffle
692 define <2 x i64> @shuffle_v2i64_z1(<2 x i64> %a) {
693 ; SSE2-LABEL: shuffle_v2i64_z1:
695 ; SSE2-NEXT: xorpd %xmm1, %xmm1
696 ; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
699 ; SSE3-LABEL: shuffle_v2i64_z1:
701 ; SSE3-NEXT: xorpd %xmm1, %xmm1
702 ; SSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
705 ; SSSE3-LABEL: shuffle_v2i64_z1:
707 ; SSSE3-NEXT: xorpd %xmm1, %xmm1
708 ; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
711 ; SSE41-LABEL: shuffle_v2i64_z1:
713 ; SSE41-NEXT: pxor %xmm1, %xmm1
714 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
717 ; AVX1-LABEL: shuffle_v2i64_z1:
719 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
720 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
723 ; AVX2-LABEL: shuffle_v2i64_z1:
725 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
726 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
728 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 2, i32 1>
729 ret <2 x i64> %shuffle
732 define <2 x double> @shuffle_v2f64_0z(<2 x double> %a) {
733 ; SSE-LABEL: shuffle_v2f64_0z:
735 ; SSE-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
738 ; AVX-LABEL: shuffle_v2f64_0z:
740 ; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
742 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 0, i32 3>
743 ret <2 x double> %shuffle
746 define <2 x double> @shuffle_v2f64_1z(<2 x double> %a) {
747 ; SSE-LABEL: shuffle_v2f64_1z:
749 ; SSE-NEXT: xorpd %xmm1, %xmm1
750 ; SSE-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
753 ; AVX-LABEL: shuffle_v2f64_1z:
755 ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
756 ; AVX-NEXT: vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
758 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 1, i32 3>
759 ret <2 x double> %shuffle
762 define <2 x double> @shuffle_v2f64_z0(<2 x double> %a) {
763 ; SSE-LABEL: shuffle_v2f64_z0:
765 ; SSE-NEXT: xorpd %xmm1, %xmm1
766 ; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
767 ; SSE-NEXT: movapd %xmm1, %xmm0
770 ; AVX-LABEL: shuffle_v2f64_z0:
772 ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
773 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
775 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 2, i32 0>
776 ret <2 x double> %shuffle
779 define <2 x double> @shuffle_v2f64_z1(<2 x double> %a) {
780 ; SSE2-LABEL: shuffle_v2f64_z1:
782 ; SSE2-NEXT: xorpd %xmm1, %xmm1
783 ; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
786 ; SSE3-LABEL: shuffle_v2f64_z1:
788 ; SSE3-NEXT: xorpd %xmm1, %xmm1
789 ; SSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
792 ; SSSE3-LABEL: shuffle_v2f64_z1:
794 ; SSSE3-NEXT: xorpd %xmm1, %xmm1
795 ; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
798 ; SSE41-LABEL: shuffle_v2f64_z1:
800 ; SSE41-NEXT: xorpd %xmm1, %xmm1
801 ; SSE41-NEXT: blendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
804 ; AVX-LABEL: shuffle_v2f64_z1:
806 ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
807 ; AVX-NEXT: vblendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
809 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 2, i32 1>
810 ret <2 x double> %shuffle
813 define <2 x double> @shuffle_v2f64_bitcast_1z(<2 x double> %a) {
814 ; SSE-LABEL: shuffle_v2f64_bitcast_1z:
816 ; SSE-NEXT: xorpd %xmm1, %xmm1
817 ; SSE-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
820 ; AVX-LABEL: shuffle_v2f64_bitcast_1z:
822 ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
823 ; AVX-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
825 %shuffle64 = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 2, i32 1>
826 %bitcast32 = bitcast <2 x double> %shuffle64 to <4 x float>
827 %shuffle32 = shufflevector <4 x float> %bitcast32, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 0, i32 1>
828 %bitcast64 = bitcast <4 x float> %shuffle32 to <2 x double>
829 ret <2 x double> %bitcast64
832 define <2 x i64> @insert_reg_and_zero_v2i64(i64 %a) {
833 ; SSE-LABEL: insert_reg_and_zero_v2i64:
835 ; SSE-NEXT: movd %rdi, %xmm0
838 ; AVX-LABEL: insert_reg_and_zero_v2i64:
840 ; AVX-NEXT: vmovq %rdi, %xmm0
842 %v = insertelement <2 x i64> undef, i64 %a, i32 0
843 %shuffle = shufflevector <2 x i64> %v, <2 x i64> zeroinitializer, <2 x i32> <i32 0, i32 3>
844 ret <2 x i64> %shuffle
847 define <2 x i64> @insert_mem_and_zero_v2i64(i64* %ptr) {
848 ; SSE-LABEL: insert_mem_and_zero_v2i64:
850 ; SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
853 ; AVX-LABEL: insert_mem_and_zero_v2i64:
855 ; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
857 %a = load i64, i64* %ptr
858 %v = insertelement <2 x i64> undef, i64 %a, i32 0
859 %shuffle = shufflevector <2 x i64> %v, <2 x i64> zeroinitializer, <2 x i32> <i32 0, i32 3>
860 ret <2 x i64> %shuffle
863 define <2 x double> @insert_reg_and_zero_v2f64(double %a) {
864 ; SSE-LABEL: insert_reg_and_zero_v2f64:
866 ; SSE-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
869 ; AVX-LABEL: insert_reg_and_zero_v2f64:
871 ; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
873 %v = insertelement <2 x double> undef, double %a, i32 0
874 %shuffle = shufflevector <2 x double> %v, <2 x double> zeroinitializer, <2 x i32> <i32 0, i32 3>
875 ret <2 x double> %shuffle
878 define <2 x double> @insert_mem_and_zero_v2f64(double* %ptr) {
879 ; SSE-LABEL: insert_mem_and_zero_v2f64:
881 ; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
884 ; AVX-LABEL: insert_mem_and_zero_v2f64:
886 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
888 %a = load double, double* %ptr
889 %v = insertelement <2 x double> undef, double %a, i32 0
890 %shuffle = shufflevector <2 x double> %v, <2 x double> zeroinitializer, <2 x i32> <i32 0, i32 3>
891 ret <2 x double> %shuffle
894 define <2 x i64> @insert_reg_lo_v2i64(i64 %a, <2 x i64> %b) {
895 ; SSE2-LABEL: insert_reg_lo_v2i64:
897 ; SSE2-NEXT: movd %rdi, %xmm1
898 ; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
901 ; SSE3-LABEL: insert_reg_lo_v2i64:
903 ; SSE3-NEXT: movd %rdi, %xmm1
904 ; SSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
907 ; SSSE3-LABEL: insert_reg_lo_v2i64:
909 ; SSSE3-NEXT: movd %rdi, %xmm1
910 ; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
913 ; SSE41-LABEL: insert_reg_lo_v2i64:
915 ; SSE41-NEXT: movd %rdi, %xmm1
916 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
919 ; AVX1-LABEL: insert_reg_lo_v2i64:
921 ; AVX1-NEXT: vmovq %rdi, %xmm1
922 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
925 ; AVX2-LABEL: insert_reg_lo_v2i64:
927 ; AVX2-NEXT: vmovq %rdi, %xmm1
928 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
930 %v = insertelement <2 x i64> undef, i64 %a, i32 0
931 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
932 ret <2 x i64> %shuffle
935 define <2 x i64> @insert_mem_lo_v2i64(i64* %ptr, <2 x i64> %b) {
936 ; SSE2-LABEL: insert_mem_lo_v2i64:
938 ; SSE2-NEXT: movlpd (%rdi), %xmm0
941 ; SSE3-LABEL: insert_mem_lo_v2i64:
943 ; SSE3-NEXT: movlpd (%rdi), %xmm0
946 ; SSSE3-LABEL: insert_mem_lo_v2i64:
948 ; SSSE3-NEXT: movlpd (%rdi), %xmm0
951 ; SSE41-LABEL: insert_mem_lo_v2i64:
953 ; SSE41-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
954 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
957 ; AVX1-LABEL: insert_mem_lo_v2i64:
959 ; AVX1-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
960 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
963 ; AVX2-LABEL: insert_mem_lo_v2i64:
965 ; AVX2-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
966 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
968 %a = load i64, i64* %ptr
969 %v = insertelement <2 x i64> undef, i64 %a, i32 0
970 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
971 ret <2 x i64> %shuffle
974 define <2 x i64> @insert_reg_hi_v2i64(i64 %a, <2 x i64> %b) {
975 ; SSE-LABEL: insert_reg_hi_v2i64:
977 ; SSE-NEXT: movd %rdi, %xmm1
978 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
981 ; AVX-LABEL: insert_reg_hi_v2i64:
983 ; AVX-NEXT: vmovq %rdi, %xmm1
984 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
986 %v = insertelement <2 x i64> undef, i64 %a, i32 0
987 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
988 ret <2 x i64> %shuffle
991 define <2 x i64> @insert_mem_hi_v2i64(i64* %ptr, <2 x i64> %b) {
992 ; SSE-LABEL: insert_mem_hi_v2i64:
994 ; SSE-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
995 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
998 ; AVX-LABEL: insert_mem_hi_v2i64:
1000 ; AVX-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
1001 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1003 %a = load i64, i64* %ptr
1004 %v = insertelement <2 x i64> undef, i64 %a, i32 0
1005 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
1006 ret <2 x i64> %shuffle
1009 define <2 x double> @insert_reg_lo_v2f64(double %a, <2 x double> %b) {
1010 ; SSE-LABEL: insert_reg_lo_v2f64:
1012 ; SSE-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
1013 ; SSE-NEXT: movapd %xmm1, %xmm0
1016 ; AVX-LABEL: insert_reg_lo_v2f64:
1018 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
1020 %v = insertelement <2 x double> undef, double %a, i32 0
1021 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 0, i32 3>
1022 ret <2 x double> %shuffle
1025 define <2 x double> @insert_mem_lo_v2f64(double* %ptr, <2 x double> %b) {
1026 ; SSE-LABEL: insert_mem_lo_v2f64:
1028 ; SSE-NEXT: movlpd (%rdi), %xmm0
1031 ; AVX-LABEL: insert_mem_lo_v2f64:
1033 ; AVX-NEXT: vmovlpd (%rdi), %xmm0, %xmm0
1035 %a = load double, double* %ptr
1036 %v = insertelement <2 x double> undef, double %a, i32 0
1037 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 0, i32 3>
1038 ret <2 x double> %shuffle
1041 define <2 x double> @insert_reg_hi_v2f64(double %a, <2 x double> %b) {
1042 ; SSE-LABEL: insert_reg_hi_v2f64:
1044 ; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
1045 ; SSE-NEXT: movapd %xmm1, %xmm0
1048 ; AVX-LABEL: insert_reg_hi_v2f64:
1050 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
1052 %v = insertelement <2 x double> undef, double %a, i32 0
1053 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 2, i32 0>
1054 ret <2 x double> %shuffle
1057 define <2 x double> @insert_mem_hi_v2f64(double* %ptr, <2 x double> %b) {
1058 ; SSE-LABEL: insert_mem_hi_v2f64:
1060 ; SSE-NEXT: movhpd (%rdi), %xmm0
1063 ; AVX-LABEL: insert_mem_hi_v2f64:
1065 ; AVX-NEXT: vmovhpd (%rdi), %xmm0, %xmm0
1067 %a = load double, double* %ptr
1068 %v = insertelement <2 x double> undef, double %a, i32 0
1069 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 2, i32 0>
1070 ret <2 x double> %shuffle
1073 define <2 x double> @insert_dup_reg_v2f64(double %a) {
1074 ; FIXME: We should match movddup for SSE3 and higher here.
1076 ; SSE2-LABEL: insert_dup_reg_v2f64:
1078 ; SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0,0]
1081 ; SSE3-LABEL: insert_dup_reg_v2f64:
1083 ; SSE3-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0]
1086 ; SSSE3-LABEL: insert_dup_reg_v2f64:
1088 ; SSSE3-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0]
1091 ; SSE41-LABEL: insert_dup_reg_v2f64:
1093 ; SSE41-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0]
1096 ; AVX-LABEL: insert_dup_reg_v2f64:
1098 ; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
1100 %v = insertelement <2 x double> undef, double %a, i32 0
1101 %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> <i32 0, i32 0>
1102 ret <2 x double> %shuffle
1104 define <2 x double> @insert_dup_mem_v2f64(double* %ptr) {
1105 ; SSE2-LABEL: insert_dup_mem_v2f64:
1107 ; SSE2-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
1108 ; SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0,0]
1111 ; SSE3-LABEL: insert_dup_mem_v2f64:
1113 ; SSE3-NEXT: movddup {{.*#+}} xmm0 = mem[0,0]
1116 ; SSSE3-LABEL: insert_dup_mem_v2f64:
1118 ; SSSE3-NEXT: movddup {{.*#+}} xmm0 = mem[0,0]
1121 ; SSE41-LABEL: insert_dup_mem_v2f64:
1123 ; SSE41-NEXT: movddup {{.*#+}} xmm0 = mem[0,0]
1126 ; AVX-LABEL: insert_dup_mem_v2f64:
1128 ; AVX-NEXT: vmovddup {{.*#+}} xmm0 = mem[0,0]
1130 %a = load double, double* %ptr
1131 %v = insertelement <2 x double> undef, double %a, i32 0
1132 %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> <i32 0, i32 0>
1133 ret <2 x double> %shuffle
1136 define <2 x double> @shuffle_mem_v2f64_10(<2 x double>* %ptr) {
1137 ; SSE-LABEL: shuffle_mem_v2f64_10:
1139 ; SSE-NEXT: movapd (%rdi), %xmm0
1140 ; SSE-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1,0]
1143 ; AVX-LABEL: shuffle_mem_v2f64_10:
1145 ; AVX-NEXT: vpermilpd {{.*#+}} xmm0 = mem[1,0]
1147 %a = load <2 x double>, <2 x double>* %ptr
1148 %shuffle = shufflevector <2 x double> %a, <2 x double> undef, <2 x i32> <i32 1, i32 0>
1149 ret <2 x double> %shuffle