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 <4 x i32> @shuffle_v4i32_0001(<4 x i32> %a, <4 x i32> %b) {
12 ; SSE-LABEL: shuffle_v4i32_0001:
14 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,0,0,1]
17 ; AVX-LABEL: shuffle_v4i32_0001:
19 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,1]
21 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 0, i32 0, i32 1>
22 ret <4 x i32> %shuffle
24 define <4 x i32> @shuffle_v4i32_0020(<4 x i32> %a, <4 x i32> %b) {
25 ; SSE-LABEL: shuffle_v4i32_0020:
27 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,0,2,0]
30 ; AVX-LABEL: shuffle_v4i32_0020:
32 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,2,0]
34 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 0, i32 2, i32 0>
35 ret <4 x i32> %shuffle
37 define <4 x i32> @shuffle_v4i32_0112(<4 x i32> %a, <4 x i32> %b) {
38 ; SSE-LABEL: shuffle_v4i32_0112:
40 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,1,2]
43 ; AVX-LABEL: shuffle_v4i32_0112:
45 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,1,2]
47 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 1, i32 2>
48 ret <4 x i32> %shuffle
50 define <4 x i32> @shuffle_v4i32_0300(<4 x i32> %a, <4 x i32> %b) {
51 ; SSE-LABEL: shuffle_v4i32_0300:
53 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,3,0,0]
56 ; AVX-LABEL: shuffle_v4i32_0300:
58 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,3,0,0]
60 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 3, i32 0, i32 0>
61 ret <4 x i32> %shuffle
63 define <4 x i32> @shuffle_v4i32_1000(<4 x i32> %a, <4 x i32> %b) {
64 ; SSE-LABEL: shuffle_v4i32_1000:
66 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,0,0,0]
69 ; AVX-LABEL: shuffle_v4i32_1000:
71 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,0,0,0]
73 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 1, i32 0, i32 0, i32 0>
74 ret <4 x i32> %shuffle
76 define <4 x i32> @shuffle_v4i32_2200(<4 x i32> %a, <4 x i32> %b) {
77 ; SSE-LABEL: shuffle_v4i32_2200:
79 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,2,0,0]
82 ; AVX-LABEL: shuffle_v4i32_2200:
84 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,2,0,0]
86 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 2, i32 0, i32 0>
87 ret <4 x i32> %shuffle
89 define <4 x i32> @shuffle_v4i32_3330(<4 x i32> %a, <4 x i32> %b) {
90 ; SSE-LABEL: shuffle_v4i32_3330:
92 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,0]
95 ; AVX-LABEL: shuffle_v4i32_3330:
97 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,0]
99 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 0>
100 ret <4 x i32> %shuffle
102 define <4 x i32> @shuffle_v4i32_3210(<4 x i32> %a, <4 x i32> %b) {
103 ; SSE-LABEL: shuffle_v4i32_3210:
105 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,2,1,0]
108 ; AVX-LABEL: shuffle_v4i32_3210:
110 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,2,1,0]
112 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
113 ret <4 x i32> %shuffle
116 define <4 x i32> @shuffle_v4i32_2121(<4 x i32> %a, <4 x i32> %b) {
117 ; SSE-LABEL: shuffle_v4i32_2121:
119 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,1,2,1]
122 ; AVX-LABEL: shuffle_v4i32_2121:
124 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,1,2,1]
126 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 1, i32 2, i32 1>
127 ret <4 x i32> %shuffle
130 define <4 x float> @shuffle_v4f32_0001(<4 x float> %a, <4 x float> %b) {
131 ; SSE-LABEL: shuffle_v4f32_0001:
133 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,0,1]
136 ; AVX-LABEL: shuffle_v4f32_0001:
138 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,1]
140 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 0, i32 1>
141 ret <4 x float> %shuffle
143 define <4 x float> @shuffle_v4f32_0020(<4 x float> %a, <4 x float> %b) {
144 ; SSE-LABEL: shuffle_v4f32_0020:
146 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,2,0]
149 ; AVX-LABEL: shuffle_v4f32_0020:
151 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,2,0]
153 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 2, i32 0>
154 ret <4 x float> %shuffle
156 define <4 x float> @shuffle_v4f32_0300(<4 x float> %a, <4 x float> %b) {
157 ; SSE-LABEL: shuffle_v4f32_0300:
159 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3,0,0]
162 ; AVX-LABEL: shuffle_v4f32_0300:
164 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,3,0,0]
166 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 3, i32 0, i32 0>
167 ret <4 x float> %shuffle
169 define <4 x float> @shuffle_v4f32_1000(<4 x float> %a, <4 x float> %b) {
170 ; SSE-LABEL: shuffle_v4f32_1000:
172 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0,0,0]
175 ; AVX-LABEL: shuffle_v4f32_1000:
177 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[1,0,0,0]
179 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 1, i32 0, i32 0, i32 0>
180 ret <4 x float> %shuffle
182 define <4 x float> @shuffle_v4f32_2200(<4 x float> %a, <4 x float> %b) {
183 ; SSE-LABEL: shuffle_v4f32_2200:
185 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,2,0,0]
188 ; AVX-LABEL: shuffle_v4f32_2200:
190 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[2,2,0,0]
192 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 2, i32 2, i32 0, i32 0>
193 ret <4 x float> %shuffle
195 define <4 x float> @shuffle_v4f32_3330(<4 x float> %a, <4 x float> %b) {
196 ; SSE-LABEL: shuffle_v4f32_3330:
198 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,0]
201 ; AVX-LABEL: shuffle_v4f32_3330:
203 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,3,3,0]
205 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 0>
206 ret <4 x float> %shuffle
208 define <4 x float> @shuffle_v4f32_3210(<4 x float> %a, <4 x float> %b) {
209 ; SSE-LABEL: shuffle_v4f32_3210:
211 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,2,1,0]
214 ; AVX-LABEL: shuffle_v4f32_3210:
216 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
218 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
219 ret <4 x float> %shuffle
221 define <4 x float> @shuffle_v4f32_0011(<4 x float> %a, <4 x float> %b) {
222 ; SSE-LABEL: shuffle_v4f32_0011:
224 ; SSE-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0,0,1,1]
227 ; AVX-LABEL: shuffle_v4f32_0011:
229 ; AVX-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0,0,1,1]
231 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 1, i32 1>
232 ret <4 x float> %shuffle
234 define <4 x float> @shuffle_v4f32_2233(<4 x float> %a, <4 x float> %b) {
235 ; SSE-LABEL: shuffle_v4f32_2233:
237 ; SSE-NEXT: unpckhps {{.*#+}} xmm0 = xmm0[2,2,3,3]
240 ; AVX-LABEL: shuffle_v4f32_2233:
242 ; AVX-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2,2,3,3]
244 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 2, i32 2, i32 3, i32 3>
245 ret <4 x float> %shuffle
247 define <4 x float> @shuffle_v4f32_0022(<4 x float> %a, <4 x float> %b) {
248 ; SSE2-LABEL: shuffle_v4f32_0022:
250 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,2,2]
253 ; SSE3-LABEL: shuffle_v4f32_0022:
255 ; SSE3-NEXT: movsldup {{.*#+}} xmm0 = xmm0[0,0,2,2]
258 ; SSSE3-LABEL: shuffle_v4f32_0022:
260 ; SSSE3-NEXT: movsldup {{.*#+}} xmm0 = xmm0[0,0,2,2]
263 ; SSE41-LABEL: shuffle_v4f32_0022:
265 ; SSE41-NEXT: movsldup {{.*#+}} xmm0 = xmm0[0,0,2,2]
268 ; AVX-LABEL: shuffle_v4f32_0022:
270 ; AVX-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2]
272 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
273 ret <4 x float> %shuffle
275 define <4 x float> @shuffle_v4f32_1133(<4 x float> %a, <4 x float> %b) {
276 ; SSE2-LABEL: shuffle_v4f32_1133:
278 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,3,3]
281 ; SSE3-LABEL: shuffle_v4f32_1133:
283 ; SSE3-NEXT: movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
286 ; SSSE3-LABEL: shuffle_v4f32_1133:
288 ; SSSE3-NEXT: movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
291 ; SSE41-LABEL: shuffle_v4f32_1133:
293 ; SSE41-NEXT: movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
296 ; AVX-LABEL: shuffle_v4f32_1133:
298 ; AVX-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
300 %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 1, i32 1, i32 3, i32 3>
301 ret <4 x float> %shuffle
304 define <4 x i32> @shuffle_v4i32_0124(<4 x i32> %a, <4 x i32> %b) {
305 ; SSE2-LABEL: shuffle_v4i32_0124:
307 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
308 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
311 ; SSE3-LABEL: shuffle_v4i32_0124:
313 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
314 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
317 ; SSSE3-LABEL: shuffle_v4i32_0124:
319 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
320 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
323 ; SSE41-LABEL: shuffle_v4i32_0124:
325 ; SSE41-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[0]
328 ; AVX-LABEL: shuffle_v4i32_0124:
330 ; AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[0]
332 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 4>
333 ret <4 x i32> %shuffle
335 define <4 x i32> @shuffle_v4i32_0142(<4 x i32> %a, <4 x i32> %b) {
336 ; SSE-LABEL: shuffle_v4i32_0142:
338 ; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
339 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
342 ; AVX-LABEL: shuffle_v4i32_0142:
344 ; AVX-NEXT: vshufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
345 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
347 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 4, i32 2>
348 ret <4 x i32> %shuffle
350 define <4 x i32> @shuffle_v4i32_0412(<4 x i32> %a, <4 x i32> %b) {
351 ; SSE-LABEL: shuffle_v4i32_0412:
353 ; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
354 ; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,2]
355 ; SSE-NEXT: movaps %xmm1, %xmm0
358 ; AVX-LABEL: shuffle_v4i32_0412:
360 ; AVX-NEXT: vshufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
361 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm1[2,0],xmm0[1,2]
363 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 4, i32 1, i32 2>
364 ret <4 x i32> %shuffle
366 define <4 x i32> @shuffle_v4i32_4012(<4 x i32> %a, <4 x i32> %b) {
367 ; SSE-LABEL: shuffle_v4i32_4012:
369 ; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
370 ; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
371 ; SSE-NEXT: movaps %xmm1, %xmm0
374 ; AVX-LABEL: shuffle_v4i32_4012:
376 ; AVX-NEXT: vshufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
377 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm1[0,2],xmm0[1,2]
379 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 0, i32 1, i32 2>
380 ret <4 x i32> %shuffle
382 define <4 x i32> @shuffle_v4i32_0145(<4 x i32> %a, <4 x i32> %b) {
383 ; SSE-LABEL: shuffle_v4i32_0145:
385 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
388 ; AVX-LABEL: shuffle_v4i32_0145:
390 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
392 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
393 ret <4 x i32> %shuffle
395 define <4 x i32> @shuffle_v4i32_0451(<4 x i32> %a, <4 x i32> %b) {
396 ; SSE-LABEL: shuffle_v4i32_0451:
398 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
399 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2,3,1]
402 ; AVX-LABEL: shuffle_v4i32_0451:
404 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
405 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2,3,1]
407 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 4, i32 5, i32 1>
408 ret <4 x i32> %shuffle
410 define <4 x i32> @shuffle_v4i32_4501(<4 x i32> %a, <4 x i32> %b) {
411 ; SSE-LABEL: shuffle_v4i32_4501:
413 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
414 ; SSE-NEXT: movdqa %xmm1, %xmm0
417 ; AVX-LABEL: shuffle_v4i32_4501:
419 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
421 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
422 ret <4 x i32> %shuffle
424 define <4 x i32> @shuffle_v4i32_4015(<4 x i32> %a, <4 x i32> %b) {
425 ; SSE-LABEL: shuffle_v4i32_4015:
427 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
428 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0,1,3]
431 ; AVX-LABEL: shuffle_v4i32_4015:
433 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
434 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm0[2,0,1,3]
436 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 0, i32 1, i32 5>
437 ret <4 x i32> %shuffle
440 define <4 x float> @shuffle_v4f32_4zzz(<4 x float> %a) {
441 ; SSE2-LABEL: shuffle_v4f32_4zzz:
443 ; SSE2-NEXT: xorps %xmm1, %xmm1
444 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,0]
445 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[2,3]
448 ; SSE3-LABEL: shuffle_v4f32_4zzz:
450 ; SSE3-NEXT: xorps %xmm1, %xmm1
451 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,0]
452 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[2,3]
455 ; SSSE3-LABEL: shuffle_v4f32_4zzz:
457 ; SSSE3-NEXT: xorps %xmm1, %xmm1
458 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,0]
459 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[2,3]
462 ; SSE41-LABEL: shuffle_v4f32_4zzz:
464 ; SSE41-NEXT: xorps %xmm1, %xmm1
465 ; SSE41-NEXT: blendps {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
466 ; SSE41-NEXT: movaps %xmm1, %xmm0
469 ; AVX-LABEL: shuffle_v4f32_4zzz:
471 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
472 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
474 %shuffle = shufflevector <4 x float> zeroinitializer, <4 x float> %a, <4 x i32> <i32 4, i32 1, i32 2, i32 3>
475 ret <4 x float> %shuffle
478 define <4 x float> @shuffle_v4f32_z4zz(<4 x float> %a) {
479 ; SSE2-LABEL: shuffle_v4f32_z4zz:
481 ; SSE2-NEXT: xorps %xmm1, %xmm1
482 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[2,0]
483 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[3,0]
486 ; SSE3-LABEL: shuffle_v4f32_z4zz:
488 ; SSE3-NEXT: xorps %xmm1, %xmm1
489 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[2,0]
490 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[3,0]
493 ; SSSE3-LABEL: shuffle_v4f32_z4zz:
495 ; SSSE3-NEXT: xorps %xmm1, %xmm1
496 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[2,0]
497 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[3,0]
500 ; SSE41-LABEL: shuffle_v4f32_z4zz:
502 ; SSE41-NEXT: insertps {{.*#+}} xmm0 = zero,xmm0[0],zero,zero
505 ; AVX-LABEL: shuffle_v4f32_z4zz:
507 ; AVX-NEXT: vinsertps {{.*#+}} xmm0 = zero,xmm0[0],zero,zero
509 %shuffle = shufflevector <4 x float> zeroinitializer, <4 x float> %a, <4 x i32> <i32 2, i32 4, i32 3, i32 0>
510 ret <4 x float> %shuffle
513 define <4 x float> @shuffle_v4f32_zz4z(<4 x float> %a) {
514 ; SSE2-LABEL: shuffle_v4f32_zz4z:
516 ; SSE2-NEXT: xorps %xmm1, %xmm1
517 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
518 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,2]
519 ; SSE2-NEXT: movaps %xmm1, %xmm0
522 ; SSE3-LABEL: shuffle_v4f32_zz4z:
524 ; SSE3-NEXT: xorps %xmm1, %xmm1
525 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
526 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,2]
527 ; SSE3-NEXT: movaps %xmm1, %xmm0
530 ; SSSE3-LABEL: shuffle_v4f32_zz4z:
532 ; SSSE3-NEXT: xorps %xmm1, %xmm1
533 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
534 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,2]
535 ; SSSE3-NEXT: movaps %xmm1, %xmm0
538 ; SSE41-LABEL: shuffle_v4f32_zz4z:
540 ; SSE41-NEXT: insertps {{.*#+}} xmm0 = zero,zero,xmm0[0],zero
543 ; AVX-LABEL: shuffle_v4f32_zz4z:
545 ; AVX-NEXT: vinsertps {{.*#+}} xmm0 = zero,zero,xmm0[0],zero
547 %shuffle = shufflevector <4 x float> zeroinitializer, <4 x float> %a, <4 x i32> <i32 0, i32 0, i32 4, i32 0>
548 ret <4 x float> %shuffle
551 define <4 x float> @shuffle_v4f32_zuu4(<4 x float> %a) {
552 ; SSE2-LABEL: shuffle_v4f32_zuu4:
554 ; SSE2-NEXT: xorps %xmm1, %xmm1
555 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
556 ; SSE2-NEXT: movaps %xmm1, %xmm0
559 ; SSE3-LABEL: shuffle_v4f32_zuu4:
561 ; SSE3-NEXT: xorps %xmm1, %xmm1
562 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
563 ; SSE3-NEXT: movaps %xmm1, %xmm0
566 ; SSSE3-LABEL: shuffle_v4f32_zuu4:
568 ; SSSE3-NEXT: xorps %xmm1, %xmm1
569 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
570 ; SSSE3-NEXT: movaps %xmm1, %xmm0
573 ; SSE41-LABEL: shuffle_v4f32_zuu4:
575 ; SSE41-NEXT: insertps {{.*#+}} xmm0 = zero,zero,zero,xmm0[0]
578 ; AVX-LABEL: shuffle_v4f32_zuu4:
580 ; AVX-NEXT: vinsertps {{.*#+}} xmm0 = zero,zero,zero,xmm0[0]
582 %shuffle = shufflevector <4 x float> zeroinitializer, <4 x float> %a, <4 x i32> <i32 0, i32 undef, i32 undef, i32 4>
583 ret <4 x float> %shuffle
586 define <4 x float> @shuffle_v4f32_zzz7(<4 x float> %a) {
587 ; SSE2-LABEL: shuffle_v4f32_zzz7:
589 ; SSE2-NEXT: xorps %xmm1, %xmm1
590 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
591 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
592 ; SSE2-NEXT: movaps %xmm1, %xmm0
595 ; SSE3-LABEL: shuffle_v4f32_zzz7:
597 ; SSE3-NEXT: xorps %xmm1, %xmm1
598 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
599 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
600 ; SSE3-NEXT: movaps %xmm1, %xmm0
603 ; SSSE3-LABEL: shuffle_v4f32_zzz7:
605 ; SSSE3-NEXT: xorps %xmm1, %xmm1
606 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
607 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
608 ; SSSE3-NEXT: movaps %xmm1, %xmm0
611 ; SSE41-LABEL: shuffle_v4f32_zzz7:
613 ; SSE41-NEXT: xorps %xmm1, %xmm1
614 ; SSE41-NEXT: blendps {{.*#+}} xmm1 = xmm1[0,1,2],xmm0[3]
615 ; SSE41-NEXT: movaps %xmm1, %xmm0
618 ; AVX-LABEL: shuffle_v4f32_zzz7:
620 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
621 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
623 %shuffle = shufflevector <4 x float> zeroinitializer, <4 x float> %a, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
624 ret <4 x float> %shuffle
627 define <4 x float> @shuffle_v4f32_z6zz(<4 x float> %a) {
628 ; SSE2-LABEL: shuffle_v4f32_z6zz:
630 ; SSE2-NEXT: xorps %xmm1, %xmm1
631 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
632 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
635 ; SSE3-LABEL: shuffle_v4f32_z6zz:
637 ; SSE3-NEXT: xorps %xmm1, %xmm1
638 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
639 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
642 ; SSSE3-LABEL: shuffle_v4f32_z6zz:
644 ; SSSE3-NEXT: xorps %xmm1, %xmm1
645 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
646 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
649 ; SSE41-LABEL: shuffle_v4f32_z6zz:
651 ; SSE41-NEXT: insertps {{.*#+}} xmm0 = zero,xmm0[2],zero,zero
654 ; AVX-LABEL: shuffle_v4f32_z6zz:
656 ; AVX-NEXT: vinsertps {{.*#+}} xmm0 = zero,xmm0[2],zero,zero
658 %shuffle = shufflevector <4 x float> zeroinitializer, <4 x float> %a, <4 x i32> <i32 0, i32 6, i32 2, i32 3>
659 ret <4 x float> %shuffle
662 define <4 x i32> @shuffle_v4i32_4zzz(i32 %i) {
663 ; SSE-LABEL: shuffle_v4i32_4zzz:
665 ; SSE-NEXT: movd %edi, %xmm0
668 ; AVX-LABEL: shuffle_v4i32_4zzz:
670 ; AVX-NEXT: vmovd %edi, %xmm0
672 %a = insertelement <4 x i32> undef, i32 %i, i32 0
673 %shuffle = shufflevector <4 x i32> zeroinitializer, <4 x i32> %a, <4 x i32> <i32 4, i32 1, i32 2, i32 3>
674 ret <4 x i32> %shuffle
677 define <4 x i32> @shuffle_v4i32_z4zz(i32 %i) {
678 ; SSE-LABEL: shuffle_v4i32_z4zz:
680 ; SSE-NEXT: movd %edi, %xmm0
681 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,0,1,1]
684 ; AVX-LABEL: shuffle_v4i32_z4zz:
686 ; AVX-NEXT: vmovd %edi, %xmm0
687 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,0,1,1]
689 %a = insertelement <4 x i32> undef, i32 %i, i32 0
690 %shuffle = shufflevector <4 x i32> zeroinitializer, <4 x i32> %a, <4 x i32> <i32 2, i32 4, i32 3, i32 0>
691 ret <4 x i32> %shuffle
694 define <4 x i32> @shuffle_v4i32_zz4z(i32 %i) {
695 ; SSE-LABEL: shuffle_v4i32_zz4z:
697 ; SSE-NEXT: movd %edi, %xmm0
698 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,0,1]
701 ; AVX-LABEL: shuffle_v4i32_zz4z:
703 ; AVX-NEXT: vmovd %edi, %xmm0
704 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,0,1]
706 %a = insertelement <4 x i32> undef, i32 %i, i32 0
707 %shuffle = shufflevector <4 x i32> zeroinitializer, <4 x i32> %a, <4 x i32> <i32 0, i32 0, i32 4, i32 0>
708 ret <4 x i32> %shuffle
711 define <4 x i32> @shuffle_v4i32_zuu4(i32 %i) {
712 ; SSE-LABEL: shuffle_v4i32_zuu4:
714 ; SSE-NEXT: movd %edi, %xmm0
715 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,0]
718 ; AVX-LABEL: shuffle_v4i32_zuu4:
720 ; AVX-NEXT: vmovd %edi, %xmm0
721 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,1,0]
723 %a = insertelement <4 x i32> undef, i32 %i, i32 0
724 %shuffle = shufflevector <4 x i32> zeroinitializer, <4 x i32> %a, <4 x i32> <i32 0, i32 undef, i32 undef, i32 4>
725 ret <4 x i32> %shuffle
728 define <4 x i32> @shuffle_v4i32_z6zz(i32 %i) {
729 ; SSE-LABEL: shuffle_v4i32_z6zz:
731 ; SSE-NEXT: movd %edi, %xmm0
732 ; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,0,1,1]
735 ; AVX-LABEL: shuffle_v4i32_z6zz:
737 ; AVX-NEXT: vmovd %edi, %xmm0
738 ; AVX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,0,1,1]
740 %a = insertelement <4 x i32> undef, i32 %i, i32 2
741 %shuffle = shufflevector <4 x i32> zeroinitializer, <4 x i32> %a, <4 x i32> <i32 0, i32 6, i32 2, i32 3>
742 ret <4 x i32> %shuffle
745 define <4 x i32> @shuffle_v4i32_7012(<4 x i32> %a, <4 x i32> %b) {
746 ; SSE2-LABEL: shuffle_v4i32_7012:
748 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[0,0]
749 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
750 ; SSE2-NEXT: movaps %xmm1, %xmm0
753 ; SSE3-LABEL: shuffle_v4i32_7012:
755 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[0,0]
756 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
757 ; SSE3-NEXT: movaps %xmm1, %xmm0
760 ; SSSE3-LABEL: shuffle_v4i32_7012:
762 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
765 ; SSE41-LABEL: shuffle_v4i32_7012:
767 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
770 ; AVX-LABEL: shuffle_v4i32_7012:
772 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
774 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 7, i32 0, i32 1, i32 2>
775 ret <4 x i32> %shuffle
778 define <4 x i32> @shuffle_v4i32_6701(<4 x i32> %a, <4 x i32> %b) {
779 ; SSE2-LABEL: shuffle_v4i32_6701:
781 ; SSE2-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
782 ; SSE2-NEXT: movapd %xmm1, %xmm0
785 ; SSE3-LABEL: shuffle_v4i32_6701:
787 ; SSE3-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
788 ; SSE3-NEXT: movapd %xmm1, %xmm0
791 ; SSSE3-LABEL: shuffle_v4i32_6701:
793 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
796 ; SSE41-LABEL: shuffle_v4i32_6701:
798 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
801 ; AVX-LABEL: shuffle_v4i32_6701:
803 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
805 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
806 ret <4 x i32> %shuffle
809 define <4 x i32> @shuffle_v4i32_5670(<4 x i32> %a, <4 x i32> %b) {
810 ; SSE2-LABEL: shuffle_v4i32_5670:
812 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
813 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,2],xmm0[2,0]
814 ; SSE2-NEXT: movaps %xmm1, %xmm0
817 ; SSE3-LABEL: shuffle_v4i32_5670:
819 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
820 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,2],xmm0[2,0]
821 ; SSE3-NEXT: movaps %xmm1, %xmm0
824 ; SSSE3-LABEL: shuffle_v4i32_5670:
826 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
829 ; SSE41-LABEL: shuffle_v4i32_5670:
831 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
834 ; AVX-LABEL: shuffle_v4i32_5670:
836 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
838 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 5, i32 6, i32 7, i32 0>
839 ret <4 x i32> %shuffle
842 define <4 x i32> @shuffle_v4i32_1234(<4 x i32> %a, <4 x i32> %b) {
843 ; SSE2-LABEL: shuffle_v4i32_1234:
845 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[3,0]
846 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,2],xmm1[2,0]
849 ; SSE3-LABEL: shuffle_v4i32_1234:
851 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[3,0]
852 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,2],xmm1[2,0]
855 ; SSSE3-LABEL: shuffle_v4i32_1234:
857 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
858 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
861 ; SSE41-LABEL: shuffle_v4i32_1234:
863 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
864 ; SSE41-NEXT: movdqa %xmm1, %xmm0
867 ; AVX-LABEL: shuffle_v4i32_1234:
869 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
871 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
872 ret <4 x i32> %shuffle
875 define <4 x i32> @shuffle_v4i32_2345(<4 x i32> %a, <4 x i32> %b) {
876 ; SSE2-LABEL: shuffle_v4i32_2345:
878 ; SSE2-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
881 ; SSE3-LABEL: shuffle_v4i32_2345:
883 ; SSE3-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
886 ; SSSE3-LABEL: shuffle_v4i32_2345:
888 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
889 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
892 ; SSE41-LABEL: shuffle_v4i32_2345:
894 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
895 ; SSE41-NEXT: movdqa %xmm1, %xmm0
898 ; AVX-LABEL: shuffle_v4i32_2345:
900 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
902 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
903 ret <4 x i32> %shuffle
906 define <4 x i32> @shuffle_v4i32_3456(<4 x i32> %a, <4 x i32> %b) {
907 ; SSE2-LABEL: shuffle_v4i32_3456:
909 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[0,0]
910 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
913 ; SSE3-LABEL: shuffle_v4i32_3456:
915 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[0,0]
916 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
919 ; SSSE3-LABEL: shuffle_v4i32_3456:
921 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
922 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
925 ; SSE41-LABEL: shuffle_v4i32_3456:
927 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
928 ; SSE41-NEXT: movdqa %xmm1, %xmm0
931 ; AVX-LABEL: shuffle_v4i32_3456:
933 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
935 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
936 ret <4 x i32> %shuffle
939 define <4 x i32> @shuffle_v4i32_0u1u(<4 x i32> %a, <4 x i32> %b) {
940 ; SSE2-LABEL: shuffle_v4i32_0u1u:
942 ; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
945 ; SSE3-LABEL: shuffle_v4i32_0u1u:
947 ; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
950 ; SSSE3-LABEL: shuffle_v4i32_0u1u:
952 ; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
955 ; SSE41-LABEL: shuffle_v4i32_0u1u:
957 ; SSE41-NEXT: pmovzxdq %xmm0, %xmm0
960 ; AVX-LABEL: shuffle_v4i32_0u1u:
962 ; AVX-NEXT: vpmovzxdq %xmm0, %xmm0
964 %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 undef, i32 1, i32 undef>
965 ret <4 x i32> %shuffle
968 define <4 x i32> @shuffle_v4i32_0z1z(<4 x i32> %a) {
969 ; SSE2-LABEL: shuffle_v4i32_0z1z:
971 ; SSE2-NEXT: pxor %xmm1, %xmm1
972 ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
975 ; SSE3-LABEL: shuffle_v4i32_0z1z:
977 ; SSE3-NEXT: pxor %xmm1, %xmm1
978 ; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
981 ; SSSE3-LABEL: shuffle_v4i32_0z1z:
983 ; SSSE3-NEXT: pxor %xmm1, %xmm1
984 ; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
987 ; SSE41-LABEL: shuffle_v4i32_0z1z:
989 ; SSE41-NEXT: pmovzxdq %xmm0, %xmm0
992 ; AVX-LABEL: shuffle_v4i32_0z1z:
994 ; AVX-NEXT: vpmovzxdq %xmm0, %xmm0
996 %shuffle = shufflevector <4 x i32> %a, <4 x i32> zeroinitializer, <4 x i32> <i32 0, i32 5, i32 1, i32 7>
997 ret <4 x i32> %shuffle
1000 define <4 x i32> @insert_reg_and_zero_v4i32(i32 %a) {
1001 ; SSE-LABEL: insert_reg_and_zero_v4i32:
1003 ; SSE-NEXT: movd %edi, %xmm0
1006 ; AVX-LABEL: insert_reg_and_zero_v4i32:
1008 ; AVX-NEXT: vmovd %edi, %xmm0
1010 %v = insertelement <4 x i32> undef, i32 %a, i32 0
1011 %shuffle = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
1012 ret <4 x i32> %shuffle
1015 define <4 x i32> @insert_mem_and_zero_v4i32(i32* %ptr) {
1016 ; SSE-LABEL: insert_mem_and_zero_v4i32:
1018 ; SSE-NEXT: movd (%rdi), %xmm0
1021 ; AVX-LABEL: insert_mem_and_zero_v4i32:
1023 ; AVX-NEXT: vmovd (%rdi), %xmm0
1026 %v = insertelement <4 x i32> undef, i32 %a, i32 0
1027 %shuffle = shufflevector <4 x i32> %v, <4 x i32> zeroinitializer, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
1028 ret <4 x i32> %shuffle
1031 define <4 x float> @insert_reg_and_zero_v4f32(float %a) {
1032 ; SSE-LABEL: insert_reg_and_zero_v4f32:
1034 ; SSE-NEXT: xorps %xmm1, %xmm1
1035 ; SSE-NEXT: movss %xmm0, %xmm1
1036 ; SSE-NEXT: movaps %xmm1, %xmm0
1039 ; AVX-LABEL: insert_reg_and_zero_v4f32:
1041 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
1042 ; AVX-NEXT: vmovss %xmm0, %xmm1, %xmm0
1044 %v = insertelement <4 x float> undef, float %a, i32 0
1045 %shuffle = shufflevector <4 x float> %v, <4 x float> zeroinitializer, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
1046 ret <4 x float> %shuffle
1049 define <4 x float> @insert_mem_and_zero_v4f32(float* %ptr) {
1050 ; SSE-LABEL: insert_mem_and_zero_v4f32:
1052 ; SSE-NEXT: movss (%rdi), %xmm0
1055 ; AVX-LABEL: insert_mem_and_zero_v4f32:
1057 ; AVX-NEXT: vmovss (%rdi), %xmm0
1059 %a = load float* %ptr
1060 %v = insertelement <4 x float> undef, float %a, i32 0
1061 %shuffle = shufflevector <4 x float> %v, <4 x float> zeroinitializer, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
1062 ret <4 x float> %shuffle
1065 define <4 x float> @shuffle_mem_v4f32_3210(<4 x float>* %ptr) {
1066 ; SSE-LABEL: shuffle_mem_v4f32_3210:
1068 ; SSE-NEXT: movaps (%rdi), %xmm0
1069 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,2,1,0]
1072 ; AVX-LABEL: shuffle_mem_v4f32_3210:
1074 ; AVX-NEXT: vpermilps {{.*#+}} xmm0 = mem[3,2,1,0]
1076 %a = load <4 x float>* %ptr
1077 %shuffle = shufflevector <4 x float> %a, <4 x float> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
1078 ret <4 x float> %shuffle