1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -x86-experimental-vector-shuffle-lowering | 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 -x86-experimental-vector-shuffle-lowering | 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 -x86-experimental-vector-shuffle-lowering | 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 -x86-experimental-vector-shuffle-lowering | 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 -x86-experimental-vector-shuffle-lowering | 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 -x86-experimental-vector-shuffle-lowering | 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: unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
111 ; SSSE3-LABEL: shuffle_v2f64_00:
113 ; SSSE3-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
116 ; SSE41-LABEL: shuffle_v2f64_00:
118 ; SSE41-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
121 ; AVX-LABEL: shuffle_v2f64_00:
123 ; AVX-NEXT: vunpcklpd {{.*#+}} 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: unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
164 ; SSE3-NEXT: movapd %xmm1, %xmm0
167 ; SSSE3-LABEL: shuffle_v2f64_22:
169 ; SSSE3-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
170 ; SSSE3-NEXT: movapd %xmm1, %xmm0
173 ; SSE41-LABEL: shuffle_v2f64_22:
175 ; SSE41-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
176 ; SSE41-NEXT: movapd %xmm1, %xmm0
179 ; AVX-LABEL: shuffle_v2f64_22:
181 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0,0]
183 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 2, i32 2>
184 ret <2 x double> %shuffle
186 define <2 x double> @shuffle_v2f64_32(<2 x double> %a, <2 x double> %b) {
187 ; SSE-LABEL: shuffle_v2f64_32:
189 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
192 ; AVX-LABEL: shuffle_v2f64_32:
194 ; AVX-NEXT: vpermilpd {{.*#+}} xmm0 = xmm1[1,0]
196 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 3, i32 2>
197 ret <2 x double> %shuffle
199 define <2 x double> @shuffle_v2f64_33(<2 x double> %a, <2 x double> %b) {
200 ; SSE-LABEL: shuffle_v2f64_33:
202 ; SSE-NEXT: movhlps {{.*#+}} xmm1 = xmm1[1,1]
203 ; SSE-NEXT: movaps %xmm1, %xmm0
206 ; AVX-LABEL: shuffle_v2f64_33:
208 ; AVX-NEXT: vmovhlps {{.*#+}} xmm0 = xmm1[1,1]
210 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 3, i32 3>
211 ret <2 x double> %shuffle
213 define <2 x double> @shuffle_v2f64_03(<2 x double> %a, <2 x double> %b) {
214 ; SSE2-LABEL: shuffle_v2f64_03:
216 ; SSE2-NEXT: movsd %xmm0, %xmm1
217 ; SSE2-NEXT: movaps %xmm1, %xmm0
220 ; SSE3-LABEL: shuffle_v2f64_03:
222 ; SSE3-NEXT: movsd %xmm0, %xmm1
223 ; SSE3-NEXT: movaps %xmm1, %xmm0
226 ; SSSE3-LABEL: shuffle_v2f64_03:
228 ; SSSE3-NEXT: movsd %xmm0, %xmm1
229 ; SSSE3-NEXT: movaps %xmm1, %xmm0
232 ; SSE41-LABEL: shuffle_v2f64_03:
234 ; SSE41-NEXT: blendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
237 ; AVX-LABEL: shuffle_v2f64_03:
239 ; AVX-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
241 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 0, i32 3>
242 ret <2 x double> %shuffle
244 define <2 x double> @shuffle_v2f64_21(<2 x double> %a, <2 x double> %b) {
245 ; SSE2-LABEL: shuffle_v2f64_21:
247 ; SSE2-NEXT: movsd %xmm1, %xmm0
250 ; SSE3-LABEL: shuffle_v2f64_21:
252 ; SSE3-NEXT: movsd %xmm1, %xmm0
255 ; SSSE3-LABEL: shuffle_v2f64_21:
257 ; SSSE3-NEXT: movsd %xmm1, %xmm0
260 ; SSE41-LABEL: shuffle_v2f64_21:
262 ; SSE41-NEXT: blendpd {{.*#+}} xmm1 = xmm1[0],xmm0[1]
263 ; SSE41-NEXT: movapd %xmm1, %xmm0
266 ; AVX-LABEL: shuffle_v2f64_21:
268 ; AVX-NEXT: vblendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
270 %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 2, i32 1>
271 ret <2 x double> %shuffle
275 define <2 x i64> @shuffle_v2i64_02(<2 x i64> %a, <2 x i64> %b) {
276 ; SSE-LABEL: shuffle_v2i64_02:
278 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
281 ; AVX-LABEL: shuffle_v2i64_02:
283 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
285 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 2>
286 ret <2 x i64> %shuffle
288 define <2 x i64> @shuffle_v2i64_02_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
289 ; SSE-LABEL: shuffle_v2i64_02_copy:
291 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
292 ; SSE-NEXT: movdqa %xmm1, %xmm0
295 ; AVX-LABEL: shuffle_v2i64_02_copy:
297 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm2[0]
299 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 2>
300 ret <2 x i64> %shuffle
302 define <2 x i64> @shuffle_v2i64_03(<2 x i64> %a, <2 x i64> %b) {
303 ; SSE2-LABEL: shuffle_v2i64_03:
305 ; SSE2-NEXT: movsd %xmm0, %xmm1
306 ; SSE2-NEXT: movaps %xmm1, %xmm0
309 ; SSE3-LABEL: shuffle_v2i64_03:
311 ; SSE3-NEXT: movsd %xmm0, %xmm1
312 ; SSE3-NEXT: movaps %xmm1, %xmm0
315 ; SSSE3-LABEL: shuffle_v2i64_03:
317 ; SSSE3-NEXT: movsd %xmm0, %xmm1
318 ; SSSE3-NEXT: movaps %xmm1, %xmm0
321 ; SSE41-LABEL: shuffle_v2i64_03:
323 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
326 ; AVX1-LABEL: shuffle_v2i64_03:
328 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
331 ; AVX2-LABEL: shuffle_v2i64_03:
333 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
335 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
336 ret <2 x i64> %shuffle
338 define <2 x i64> @shuffle_v2i64_03_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
339 ; SSE2-LABEL: shuffle_v2i64_03_copy:
341 ; SSE2-NEXT: movsd %xmm1, %xmm2
342 ; SSE2-NEXT: movaps %xmm2, %xmm0
345 ; SSE3-LABEL: shuffle_v2i64_03_copy:
347 ; SSE3-NEXT: movsd %xmm1, %xmm2
348 ; SSE3-NEXT: movaps %xmm2, %xmm0
351 ; SSSE3-LABEL: shuffle_v2i64_03_copy:
353 ; SSSE3-NEXT: movsd %xmm1, %xmm2
354 ; SSSE3-NEXT: movaps %xmm2, %xmm0
357 ; SSE41-LABEL: shuffle_v2i64_03_copy:
359 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7]
360 ; SSE41-NEXT: movdqa %xmm1, %xmm0
363 ; AVX1-LABEL: shuffle_v2i64_03_copy:
365 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm2[4,5,6,7]
368 ; AVX2-LABEL: shuffle_v2i64_03_copy:
370 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm2[2,3]
372 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
373 ret <2 x i64> %shuffle
375 define <2 x i64> @shuffle_v2i64_12(<2 x i64> %a, <2 x i64> %b) {
376 ; SSE2-LABEL: shuffle_v2i64_12:
378 ; SSE2-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
381 ; SSE3-LABEL: shuffle_v2i64_12:
383 ; SSE3-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
386 ; SSSE3-LABEL: shuffle_v2i64_12:
388 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
389 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
392 ; SSE41-LABEL: shuffle_v2i64_12:
394 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
395 ; SSE41-NEXT: movdqa %xmm1, %xmm0
398 ; AVX-LABEL: shuffle_v2i64_12:
400 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
402 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 2>
403 ret <2 x i64> %shuffle
405 define <2 x i64> @shuffle_v2i64_12_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
406 ; SSE2-LABEL: shuffle_v2i64_12_copy:
408 ; SSE2-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm2[0]
409 ; SSE2-NEXT: movapd %xmm1, %xmm0
412 ; SSE3-LABEL: shuffle_v2i64_12_copy:
414 ; SSE3-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm2[0]
415 ; SSE3-NEXT: movapd %xmm1, %xmm0
418 ; SSSE3-LABEL: shuffle_v2i64_12_copy:
420 ; SSSE3-NEXT: palignr {{.*#+}} xmm2 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
421 ; SSSE3-NEXT: movdqa %xmm2, %xmm0
424 ; SSE41-LABEL: shuffle_v2i64_12_copy:
426 ; SSE41-NEXT: palignr {{.*#+}} xmm2 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
427 ; SSE41-NEXT: movdqa %xmm2, %xmm0
430 ; AVX-LABEL: shuffle_v2i64_12_copy:
432 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
434 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 2>
435 ret <2 x i64> %shuffle
437 define <2 x i64> @shuffle_v2i64_13(<2 x i64> %a, <2 x i64> %b) {
438 ; SSE-LABEL: shuffle_v2i64_13:
440 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
443 ; AVX-LABEL: shuffle_v2i64_13:
445 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
447 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 3>
448 ret <2 x i64> %shuffle
450 define <2 x i64> @shuffle_v2i64_13_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
451 ; SSE-LABEL: shuffle_v2i64_13_copy:
453 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm2[1]
454 ; SSE-NEXT: movdqa %xmm1, %xmm0
457 ; AVX-LABEL: shuffle_v2i64_13_copy:
459 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm1[1],xmm2[1]
461 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 3>
462 ret <2 x i64> %shuffle
464 define <2 x i64> @shuffle_v2i64_20(<2 x i64> %a, <2 x i64> %b) {
465 ; SSE-LABEL: shuffle_v2i64_20:
467 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
468 ; SSE-NEXT: movdqa %xmm1, %xmm0
471 ; AVX-LABEL: shuffle_v2i64_20:
473 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
475 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
476 ret <2 x i64> %shuffle
478 define <2 x i64> @shuffle_v2i64_20_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
479 ; SSE-LABEL: shuffle_v2i64_20_copy:
481 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0]
482 ; SSE-NEXT: movdqa %xmm2, %xmm0
485 ; AVX-LABEL: shuffle_v2i64_20_copy:
487 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm1[0]
489 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
490 ret <2 x i64> %shuffle
492 define <2 x i64> @shuffle_v2i64_21(<2 x i64> %a, <2 x i64> %b) {
493 ; SSE2-LABEL: shuffle_v2i64_21:
495 ; SSE2-NEXT: movsd %xmm1, %xmm0
498 ; SSE3-LABEL: shuffle_v2i64_21:
500 ; SSE3-NEXT: movsd %xmm1, %xmm0
503 ; SSSE3-LABEL: shuffle_v2i64_21:
505 ; SSSE3-NEXT: movsd %xmm1, %xmm0
508 ; SSE41-LABEL: shuffle_v2i64_21:
510 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
511 ; SSE41-NEXT: movdqa %xmm1, %xmm0
514 ; AVX1-LABEL: shuffle_v2i64_21:
516 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
519 ; AVX2-LABEL: shuffle_v2i64_21:
521 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
523 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 1>
524 ret <2 x i64> %shuffle
526 define <2 x i64> @shuffle_v2i64_21_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
527 ; SSE2-LABEL: shuffle_v2i64_21_copy:
529 ; SSE2-NEXT: movsd %xmm2, %xmm1
530 ; SSE2-NEXT: movaps %xmm1, %xmm0
533 ; SSE3-LABEL: shuffle_v2i64_21_copy:
535 ; SSE3-NEXT: movsd %xmm2, %xmm1
536 ; SSE3-NEXT: movaps %xmm1, %xmm0
539 ; SSSE3-LABEL: shuffle_v2i64_21_copy:
541 ; SSSE3-NEXT: movsd %xmm2, %xmm1
542 ; SSSE3-NEXT: movaps %xmm1, %xmm0
545 ; SSE41-LABEL: shuffle_v2i64_21_copy:
547 ; SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5,6,7]
548 ; SSE41-NEXT: movdqa %xmm2, %xmm0
551 ; AVX1-LABEL: shuffle_v2i64_21_copy:
553 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm1[4,5,6,7]
556 ; AVX2-LABEL: shuffle_v2i64_21_copy:
558 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0,1],xmm1[2,3]
560 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 1>
561 ret <2 x i64> %shuffle
563 define <2 x i64> @shuffle_v2i64_30(<2 x i64> %a, <2 x i64> %b) {
564 ; SSE2-LABEL: shuffle_v2i64_30:
566 ; SSE2-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
567 ; SSE2-NEXT: movapd %xmm1, %xmm0
570 ; SSE3-LABEL: shuffle_v2i64_30:
572 ; SSE3-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
573 ; SSE3-NEXT: movapd %xmm1, %xmm0
576 ; SSSE3-LABEL: shuffle_v2i64_30:
578 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
581 ; SSE41-LABEL: shuffle_v2i64_30:
583 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
586 ; AVX-LABEL: shuffle_v2i64_30:
588 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
590 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 0>
591 ret <2 x i64> %shuffle
593 define <2 x i64> @shuffle_v2i64_30_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
594 ; SSE2-LABEL: shuffle_v2i64_30_copy:
596 ; SSE2-NEXT: shufpd {{.*#+}} xmm2 = xmm2[1],xmm1[0]
597 ; SSE2-NEXT: movapd %xmm2, %xmm0
600 ; SSE3-LABEL: shuffle_v2i64_30_copy:
602 ; SSE3-NEXT: shufpd {{.*#+}} xmm2 = xmm2[1],xmm1[0]
603 ; SSE3-NEXT: movapd %xmm2, %xmm0
606 ; SSSE3-LABEL: shuffle_v2i64_30_copy:
608 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
609 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
612 ; SSE41-LABEL: shuffle_v2i64_30_copy:
614 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
615 ; SSE41-NEXT: movdqa %xmm1, %xmm0
618 ; AVX-LABEL: shuffle_v2i64_30_copy:
620 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
622 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 0>
623 ret <2 x i64> %shuffle
625 define <2 x i64> @shuffle_v2i64_31(<2 x i64> %a, <2 x i64> %b) {
626 ; SSE-LABEL: shuffle_v2i64_31:
628 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm0[1]
629 ; SSE-NEXT: movdqa %xmm1, %xmm0
632 ; AVX-LABEL: shuffle_v2i64_31:
634 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm1[1],xmm0[1]
636 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 1>
637 ret <2 x i64> %shuffle
639 define <2 x i64> @shuffle_v2i64_31_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
640 ; SSE-LABEL: shuffle_v2i64_31_copy:
642 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm1[1]
643 ; SSE-NEXT: movdqa %xmm2, %xmm0
646 ; AVX-LABEL: shuffle_v2i64_31_copy:
648 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm2[1],xmm1[1]
650 %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 1>
651 ret <2 x i64> %shuffle
654 define <2 x i64> @shuffle_v2i64_0z(<2 x i64> %a) {
655 ; SSE-LABEL: shuffle_v2i64_0z:
657 ; SSE-NEXT: movq %xmm0, %xmm0
660 ; AVX-LABEL: shuffle_v2i64_0z:
662 ; AVX-NEXT: vmovq %xmm0, %xmm0
664 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 0, i32 3>
665 ret <2 x i64> %shuffle
668 define <2 x i64> @shuffle_v2i64_1z(<2 x i64> %a) {
669 ; SSE-LABEL: shuffle_v2i64_1z:
671 ; SSE-NEXT: pxor %xmm1, %xmm1
672 ; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
675 ; AVX-LABEL: shuffle_v2i64_1z:
677 ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
678 ; AVX-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
680 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 1, i32 3>
681 ret <2 x i64> %shuffle
684 define <2 x i64> @shuffle_v2i64_z0(<2 x i64> %a) {
685 ; SSE-LABEL: shuffle_v2i64_z0:
687 ; SSE-NEXT: movq %xmm0, %xmm0
688 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
691 ; AVX-LABEL: shuffle_v2i64_z0:
693 ; AVX-NEXT: vmovq %xmm0, %xmm0
694 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
696 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 2, i32 0>
697 ret <2 x i64> %shuffle
700 define <2 x i64> @shuffle_v2i64_z1(<2 x i64> %a) {
701 ; SSE2-LABEL: shuffle_v2i64_z1:
703 ; SSE2-NEXT: xorps %xmm1, %xmm1
704 ; SSE2-NEXT: movsd %xmm1, %xmm0
707 ; SSE3-LABEL: shuffle_v2i64_z1:
709 ; SSE3-NEXT: xorps %xmm1, %xmm1
710 ; SSE3-NEXT: movsd %xmm1, %xmm0
713 ; SSSE3-LABEL: shuffle_v2i64_z1:
715 ; SSSE3-NEXT: xorps %xmm1, %xmm1
716 ; SSSE3-NEXT: movsd %xmm1, %xmm0
719 ; SSE41-LABEL: shuffle_v2i64_z1:
721 ; SSE41-NEXT: pxor %xmm1, %xmm1
722 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
723 ; SSE41-NEXT: movdqa %xmm1, %xmm0
726 ; AVX1-LABEL: shuffle_v2i64_z1:
728 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
729 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
732 ; AVX2-LABEL: shuffle_v2i64_z1:
734 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
735 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
737 %shuffle = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32> <i32 2, i32 1>
738 ret <2 x i64> %shuffle
741 define <2 x double> @shuffle_v2f64_0z(<2 x double> %a) {
742 ; SSE-LABEL: shuffle_v2f64_0z:
744 ; SSE-NEXT: movq %xmm0, %xmm0
747 ; AVX-LABEL: shuffle_v2f64_0z:
749 ; AVX-NEXT: vmovq %xmm0, %xmm0
751 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 0, i32 3>
752 ret <2 x double> %shuffle
755 define <2 x double> @shuffle_v2f64_1z(<2 x double> %a) {
756 ; SSE-LABEL: shuffle_v2f64_1z:
758 ; SSE-NEXT: xorpd %xmm1, %xmm1
759 ; SSE-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
762 ; AVX-LABEL: shuffle_v2f64_1z:
764 ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
765 ; AVX-NEXT: vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
767 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 1, i32 3>
768 ret <2 x double> %shuffle
771 define <2 x double> @shuffle_v2f64_z0(<2 x double> %a) {
772 ; SSE-LABEL: shuffle_v2f64_z0:
774 ; SSE-NEXT: xorpd %xmm1, %xmm1
775 ; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
776 ; SSE-NEXT: movapd %xmm1, %xmm0
779 ; AVX-LABEL: shuffle_v2f64_z0:
781 ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
782 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
784 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 2, i32 0>
785 ret <2 x double> %shuffle
788 define <2 x double> @shuffle_v2f64_z1(<2 x double> %a) {
789 ; SSE2-LABEL: shuffle_v2f64_z1:
791 ; SSE2-NEXT: xorps %xmm1, %xmm1
792 ; SSE2-NEXT: movsd %xmm1, %xmm0
795 ; SSE3-LABEL: shuffle_v2f64_z1:
797 ; SSE3-NEXT: xorps %xmm1, %xmm1
798 ; SSE3-NEXT: movsd %xmm1, %xmm0
801 ; SSSE3-LABEL: shuffle_v2f64_z1:
803 ; SSSE3-NEXT: xorps %xmm1, %xmm1
804 ; SSSE3-NEXT: movsd %xmm1, %xmm0
807 ; SSE41-LABEL: shuffle_v2f64_z1:
809 ; SSE41-NEXT: xorpd %xmm1, %xmm1
810 ; SSE41-NEXT: blendpd {{.*#+}} xmm1 = xmm1[0],xmm0[1]
811 ; SSE41-NEXT: movapd %xmm1, %xmm0
814 ; AVX-LABEL: shuffle_v2f64_z1:
816 ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
817 ; AVX-NEXT: vblendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
819 %shuffle = shufflevector <2 x double> %a, <2 x double> zeroinitializer, <2 x i32> <i32 2, i32 1>
820 ret <2 x double> %shuffle
823 define <2 x i64> @insert_reg_and_zero_v2i64(i64 %a) {
824 ; SSE-LABEL: insert_reg_and_zero_v2i64:
826 ; SSE-NEXT: movd %rdi, %xmm0
829 ; AVX-LABEL: insert_reg_and_zero_v2i64:
831 ; AVX-NEXT: vmovq %rdi, %xmm0
833 %v = insertelement <2 x i64> undef, i64 %a, i32 0
834 %shuffle = shufflevector <2 x i64> %v, <2 x i64> zeroinitializer, <2 x i32> <i32 0, i32 3>
835 ret <2 x i64> %shuffle
838 define <2 x i64> @insert_mem_and_zero_v2i64(i64* %ptr) {
839 ; SSE-LABEL: insert_mem_and_zero_v2i64:
841 ; SSE-NEXT: movq (%rdi), %xmm0
844 ; AVX-LABEL: insert_mem_and_zero_v2i64:
846 ; AVX-NEXT: vmovq (%rdi), %xmm0
849 %v = insertelement <2 x i64> undef, i64 %a, i32 0
850 %shuffle = shufflevector <2 x i64> %v, <2 x i64> zeroinitializer, <2 x i32> <i32 0, i32 3>
851 ret <2 x i64> %shuffle
854 define <2 x double> @insert_reg_and_zero_v2f64(double %a) {
855 ; SSE-LABEL: insert_reg_and_zero_v2f64:
857 ; SSE-NEXT: movq %xmm0, %xmm0
860 ; AVX-LABEL: insert_reg_and_zero_v2f64:
862 ; AVX-NEXT: vmovq %xmm0, %xmm0
864 %v = insertelement <2 x double> undef, double %a, i32 0
865 %shuffle = shufflevector <2 x double> %v, <2 x double> zeroinitializer, <2 x i32> <i32 0, i32 3>
866 ret <2 x double> %shuffle
869 define <2 x double> @insert_mem_and_zero_v2f64(double* %ptr) {
870 ; SSE-LABEL: insert_mem_and_zero_v2f64:
872 ; SSE-NEXT: movsd (%rdi), %xmm0
875 ; AVX-LABEL: insert_mem_and_zero_v2f64:
877 ; AVX-NEXT: vmovsd (%rdi), %xmm0
879 %a = load double* %ptr
880 %v = insertelement <2 x double> undef, double %a, i32 0
881 %shuffle = shufflevector <2 x double> %v, <2 x double> zeroinitializer, <2 x i32> <i32 0, i32 3>
882 ret <2 x double> %shuffle
885 define <2 x i64> @insert_reg_lo_v2i64(i64 %a, <2 x i64> %b) {
886 ; SSE2-LABEL: insert_reg_lo_v2i64:
888 ; SSE2-NEXT: movd %rdi, %xmm1
889 ; SSE2-NEXT: movsd %xmm1, %xmm0
892 ; SSE3-LABEL: insert_reg_lo_v2i64:
894 ; SSE3-NEXT: movd %rdi, %xmm1
895 ; SSE3-NEXT: movsd %xmm1, %xmm0
898 ; SSSE3-LABEL: insert_reg_lo_v2i64:
900 ; SSSE3-NEXT: movd %rdi, %xmm1
901 ; SSSE3-NEXT: movsd %xmm1, %xmm0
904 ; SSE41-LABEL: insert_reg_lo_v2i64:
906 ; SSE41-NEXT: movd %rdi, %xmm1
907 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
908 ; SSE41-NEXT: movdqa %xmm1, %xmm0
911 ; AVX1-LABEL: insert_reg_lo_v2i64:
913 ; AVX1-NEXT: vmovq %rdi, %xmm1
914 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
917 ; AVX2-LABEL: insert_reg_lo_v2i64:
919 ; AVX2-NEXT: vmovq %rdi, %xmm1
920 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
922 %v = insertelement <2 x i64> undef, i64 %a, i32 0
923 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
924 ret <2 x i64> %shuffle
927 define <2 x i64> @insert_mem_lo_v2i64(i64* %ptr, <2 x i64> %b) {
928 ; SSE2-LABEL: insert_mem_lo_v2i64:
930 ; SSE2-NEXT: movlpd (%rdi), %xmm0
933 ; SSE3-LABEL: insert_mem_lo_v2i64:
935 ; SSE3-NEXT: movlpd (%rdi), %xmm0
938 ; SSSE3-LABEL: insert_mem_lo_v2i64:
940 ; SSSE3-NEXT: movlpd (%rdi), %xmm0
943 ; SSE41-LABEL: insert_mem_lo_v2i64:
945 ; SSE41-NEXT: movq (%rdi), %xmm1
946 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
947 ; SSE41-NEXT: movdqa %xmm1, %xmm0
950 ; AVX1-LABEL: insert_mem_lo_v2i64:
952 ; AVX1-NEXT: vmovq (%rdi), %xmm1
953 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
956 ; AVX2-LABEL: insert_mem_lo_v2i64:
958 ; AVX2-NEXT: vmovq (%rdi), %xmm1
959 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
962 %v = insertelement <2 x i64> undef, i64 %a, i32 0
963 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
964 ret <2 x i64> %shuffle
967 define <2 x i64> @insert_reg_hi_v2i64(i64 %a, <2 x i64> %b) {
968 ; SSE-LABEL: insert_reg_hi_v2i64:
970 ; SSE-NEXT: movd %rdi, %xmm1
971 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
974 ; AVX-LABEL: insert_reg_hi_v2i64:
976 ; AVX-NEXT: vmovq %rdi, %xmm1
977 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
979 %v = insertelement <2 x i64> undef, i64 %a, i32 0
980 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
981 ret <2 x i64> %shuffle
984 define <2 x i64> @insert_mem_hi_v2i64(i64* %ptr, <2 x i64> %b) {
985 ; SSE-LABEL: insert_mem_hi_v2i64:
987 ; SSE-NEXT: movq (%rdi), %xmm1
988 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
991 ; AVX-LABEL: insert_mem_hi_v2i64:
993 ; AVX-NEXT: vmovq (%rdi), %xmm1
994 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
997 %v = insertelement <2 x i64> undef, i64 %a, i32 0
998 %shuffle = shufflevector <2 x i64> %v, <2 x i64> %b, <2 x i32> <i32 2, i32 0>
999 ret <2 x i64> %shuffle
1002 define <2 x double> @insert_reg_lo_v2f64(double %a, <2 x double> %b) {
1003 ; SSE-LABEL: insert_reg_lo_v2f64:
1005 ; SSE-NEXT: movsd %xmm0, %xmm1
1006 ; SSE-NEXT: movaps %xmm1, %xmm0
1009 ; AVX-LABEL: insert_reg_lo_v2f64:
1011 ; AVX-NEXT: vmovsd %xmm0, %xmm1, %xmm0
1013 %v = insertelement <2 x double> undef, double %a, i32 0
1014 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 0, i32 3>
1015 ret <2 x double> %shuffle
1018 define <2 x double> @insert_mem_lo_v2f64(double* %ptr, <2 x double> %b) {
1019 ; SSE-LABEL: insert_mem_lo_v2f64:
1021 ; SSE-NEXT: movlpd (%rdi), %xmm0
1024 ; AVX-LABEL: insert_mem_lo_v2f64:
1026 ; AVX-NEXT: vmovlpd (%rdi), %xmm0, %xmm0
1028 %a = load double* %ptr
1029 %v = insertelement <2 x double> undef, double %a, i32 0
1030 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 0, i32 3>
1031 ret <2 x double> %shuffle
1034 define <2 x double> @insert_reg_hi_v2f64(double %a, <2 x double> %b) {
1035 ; SSE-LABEL: insert_reg_hi_v2f64:
1037 ; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
1038 ; SSE-NEXT: movapd %xmm1, %xmm0
1041 ; AVX-LABEL: insert_reg_hi_v2f64:
1043 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
1045 %v = insertelement <2 x double> undef, double %a, i32 0
1046 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 2, i32 0>
1047 ret <2 x double> %shuffle
1050 define <2 x double> @insert_mem_hi_v2f64(double* %ptr, <2 x double> %b) {
1051 ; SSE-LABEL: insert_mem_hi_v2f64:
1053 ; SSE-NEXT: movhpd (%rdi), %xmm0
1056 ; AVX-LABEL: insert_mem_hi_v2f64:
1058 ; AVX-NEXT: vmovhpd (%rdi), %xmm0, %xmm0
1060 %a = load double* %ptr
1061 %v = insertelement <2 x double> undef, double %a, i32 0
1062 %shuffle = shufflevector <2 x double> %v, <2 x double> %b, <2 x i32> <i32 2, i32 0>
1063 ret <2 x double> %shuffle
1066 define <2 x double> @insert_dup_reg_v2f64(double %a) {
1067 ; FIXME: We should match movddup for SSE3 and higher here.
1069 ; SSE2-LABEL: insert_dup_reg_v2f64:
1071 ; SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0,0]
1074 ; SSE3-LABEL: insert_dup_reg_v2f64:
1076 ; SSE3-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
1079 ; SSSE3-LABEL: insert_dup_reg_v2f64:
1081 ; SSSE3-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
1084 ; SSE41-LABEL: insert_dup_reg_v2f64:
1086 ; SSE41-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
1089 ; AVX-LABEL: insert_dup_reg_v2f64:
1091 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0,0]
1093 %v = insertelement <2 x double> undef, double %a, i32 0
1094 %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> <i32 0, i32 0>
1095 ret <2 x double> %shuffle
1097 define <2 x double> @insert_dup_mem_v2f64(double* %ptr) {
1098 ; SSE2-LABEL: insert_dup_mem_v2f64:
1100 ; SSE2-NEXT: movsd (%rdi), %xmm0
1101 ; SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0,0]
1104 ; SSE3-LABEL: insert_dup_mem_v2f64:
1106 ; SSE3-NEXT: movddup (%rdi), %xmm0
1109 ; SSSE3-LABEL: insert_dup_mem_v2f64:
1111 ; SSSE3-NEXT: movddup (%rdi), %xmm0
1114 ; SSE41-LABEL: insert_dup_mem_v2f64:
1116 ; SSE41-NEXT: movddup (%rdi), %xmm0
1119 ; AVX-LABEL: insert_dup_mem_v2f64:
1121 ; AVX-NEXT: vmovddup (%rdi), %xmm0
1123 %a = load double* %ptr
1124 %v = insertelement <2 x double> undef, double %a, i32 0
1125 %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> <i32 0, i32 0>
1126 ret <2 x double> %shuffle
1129 define <2 x double> @shuffle_mem_v2f64_10(<2 x double>* %ptr) {
1130 ; SSE-LABEL: shuffle_mem_v2f64_10:
1132 ; SSE-NEXT: movapd (%rdi), %xmm0
1133 ; SSE-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1,0]
1136 ; AVX-LABEL: shuffle_mem_v2f64_10:
1138 ; AVX-NEXT: vpermilpd {{.*#+}} xmm0 = mem[1,0]
1140 %a = load <2 x double>* %ptr
1141 %shuffle = shufflevector <2 x double> %a, <2 x double> undef, <2 x i32> <i32 1, i32 0>
1142 ret <2 x double> %shuffle