| blob | 2feb9742c60952dd885e2d3559b64b5254dc362c |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX --check-prefix=AVX512F
5 ;
6 ; Just one 32-bit run to make sure we do reasonable things.
7 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32-AVX
9 define <4 x double> @merge_4f64_2f64_23(<2 x double>* %ptr) nounwind uwtable noinline ssp {
10 ; AVX-LABEL: merge_4f64_2f64_23:
11 ; AVX: # %bb.0:
12 ; AVX-NEXT: vmovups 32(%rdi), %ymm0
13 ; AVX-NEXT: retq
14 ;
15 ; X32-AVX-LABEL: merge_4f64_2f64_23:
16 ; X32-AVX: # %bb.0:
17 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
18 ; X32-AVX-NEXT: vmovups 32(%eax), %ymm0
19 ; X32-AVX-NEXT: retl
20 %ptr0 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 2
21 %ptr1 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 3
22 %val0 = load <2 x double>, <2 x double>* %ptr0
23 %val1 = load <2 x double>, <2 x double>* %ptr1
24 %res = shufflevector <2 x double> %val0, <2 x double> %val1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
25 ret <4 x double> %res
26 }
28 define <4 x double> @merge_4f64_2f64_2z(<2 x double>* %ptr) nounwind uwtable noinline ssp {
29 ; AVX-LABEL: merge_4f64_2f64_2z:
30 ; AVX: # %bb.0:
31 ; AVX-NEXT: vmovaps 32(%rdi), %xmm0
32 ; AVX-NEXT: retq
33 ;
34 ; X32-AVX-LABEL: merge_4f64_2f64_2z:
35 ; X32-AVX: # %bb.0:
36 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
37 ; X32-AVX-NEXT: vmovaps 32(%eax), %xmm0
38 ; X32-AVX-NEXT: retl
39 %ptr0 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 2
40 %val0 = load <2 x double>, <2 x double>* %ptr0
41 %res = shufflevector <2 x double> %val0, <2 x double> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
42 ret <4 x double> %res
43 }
45 define <4 x double> @merge_4f64_f64_2345(double* %ptr) nounwind uwtable noinline ssp {
46 ; AVX-LABEL: merge_4f64_f64_2345:
47 ; AVX: # %bb.0:
48 ; AVX-NEXT: vmovups 16(%rdi), %ymm0
49 ; AVX-NEXT: retq
50 ;
51 ; X32-AVX-LABEL: merge_4f64_f64_2345:
52 ; X32-AVX: # %bb.0:
53 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
54 ; X32-AVX-NEXT: vmovups 16(%eax), %ymm0
55 ; X32-AVX-NEXT: retl
56 %ptr0 = getelementptr inbounds double, double* %ptr, i64 2
57 %ptr1 = getelementptr inbounds double, double* %ptr, i64 3
58 %ptr2 = getelementptr inbounds double, double* %ptr, i64 4
59 %ptr3 = getelementptr inbounds double, double* %ptr, i64 5
60 %val0 = load double, double* %ptr0
61 %val1 = load double, double* %ptr1
62 %val2 = load double, double* %ptr2
63 %val3 = load double, double* %ptr3
64 %res0 = insertelement <4 x double> undef, double %val0, i32 0
65 %res1 = insertelement <4 x double> %res0, double %val1, i32 1
66 %res2 = insertelement <4 x double> %res1, double %val2, i32 2
67 %res3 = insertelement <4 x double> %res2, double %val3, i32 3
68 ret <4 x double> %res3
69 }
71 define <4 x double> @merge_4f64_f64_3zuu(double* %ptr) nounwind uwtable noinline ssp {
72 ; AVX-LABEL: merge_4f64_f64_3zuu:
73 ; AVX: # %bb.0:
74 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
75 ; AVX-NEXT: retq
76 ;
77 ; X32-AVX-LABEL: merge_4f64_f64_3zuu:
78 ; X32-AVX: # %bb.0:
79 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
80 ; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
81 ; X32-AVX-NEXT: retl
82 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3
83 %val0 = load double, double* %ptr0
84 %res0 = insertelement <4 x double> undef, double %val0, i32 0
85 %res1 = insertelement <4 x double> %res0, double 0.0, i32 1
86 ret <4 x double> %res1
87 }
89 define <4 x double> @merge_4f64_f64_34uu(double* %ptr) nounwind uwtable noinline ssp {
90 ; AVX-LABEL: merge_4f64_f64_34uu:
91 ; AVX: # %bb.0:
92 ; AVX-NEXT: vmovups 24(%rdi), %xmm0
93 ; AVX-NEXT: retq
94 ;
95 ; X32-AVX-LABEL: merge_4f64_f64_34uu:
96 ; X32-AVX: # %bb.0:
97 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
98 ; X32-AVX-NEXT: vmovups 24(%eax), %xmm0
99 ; X32-AVX-NEXT: retl
100 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3
101 %ptr1 = getelementptr inbounds double, double* %ptr, i64 4
102 %val0 = load double, double* %ptr0
103 %val1 = load double, double* %ptr1
104 %res0 = insertelement <4 x double> undef, double %val0, i32 0
105 %res1 = insertelement <4 x double> %res0, double %val1, i32 1
106 ret <4 x double> %res1
107 }
109 define <4 x double> @merge_4f64_f64_45zz(double* %ptr) nounwind uwtable noinline ssp {
110 ; AVX-LABEL: merge_4f64_f64_45zz:
111 ; AVX: # %bb.0:
112 ; AVX-NEXT: vmovups 32(%rdi), %xmm0
113 ; AVX-NEXT: retq
114 ;
115 ; X32-AVX-LABEL: merge_4f64_f64_45zz:
116 ; X32-AVX: # %bb.0:
117 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
118 ; X32-AVX-NEXT: vmovups 32(%eax), %xmm0
119 ; X32-AVX-NEXT: retl
120 %ptr0 = getelementptr inbounds double, double* %ptr, i64 4
121 %ptr1 = getelementptr inbounds double, double* %ptr, i64 5
122 %val0 = load double, double* %ptr0
123 %val1 = load double, double* %ptr1
124 %res0 = insertelement <4 x double> zeroinitializer, double %val0, i32 0
125 %res1 = insertelement <4 x double> %res0, double %val1, i32 1
126 ret <4 x double> %res1
127 }
129 define <4 x double> @merge_4f64_f64_34z6(double* %ptr) nounwind uwtable noinline ssp {
130 ; AVX-LABEL: merge_4f64_f64_34z6:
131 ; AVX: # %bb.0:
132 ; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
133 ; AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
134 ; AVX-NEXT: retq
135 ;
136 ; X32-AVX-LABEL: merge_4f64_f64_34z6:
137 ; X32-AVX: # %bb.0:
138 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
139 ; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
140 ; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
141 ; X32-AVX-NEXT: retl
142 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3
143 %ptr1 = getelementptr inbounds double, double* %ptr, i64 4
144 %ptr3 = getelementptr inbounds double, double* %ptr, i64 6
145 %val0 = load double, double* %ptr0
146 %val1 = load double, double* %ptr1
147 %val3 = load double, double* %ptr3
148 %res0 = insertelement <4 x double> undef, double %val0, i32 0
149 %res1 = insertelement <4 x double> %res0, double %val1, i32 1
150 %res2 = insertelement <4 x double> %res1, double 0.0, i32 2
151 %res3 = insertelement <4 x double> %res2, double %val3, i32 3
152 ret <4 x double> %res3
153 }
155 define <4 x i64> @merge_4i64_2i64_3z(<2 x i64>* %ptr) nounwind uwtable noinline ssp {
156 ; AVX-LABEL: merge_4i64_2i64_3z:
157 ; AVX: # %bb.0:
158 ; AVX-NEXT: vmovaps 48(%rdi), %xmm0
159 ; AVX-NEXT: retq
160 ;
161 ; X32-AVX-LABEL: merge_4i64_2i64_3z:
162 ; X32-AVX: # %bb.0:
163 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
164 ; X32-AVX-NEXT: vmovaps 48(%eax), %xmm0
165 ; X32-AVX-NEXT: retl
166 %ptr0 = getelementptr inbounds <2 x i64>, <2 x i64>* %ptr, i64 3
167 %val0 = load <2 x i64>, <2 x i64>* %ptr0
168 %res = shufflevector <2 x i64> %val0, <2 x i64> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
169 ret <4 x i64> %res
170 }
172 define <4 x i64> @merge_4i64_i64_1234(i64* %ptr) nounwind uwtable noinline ssp {
173 ; AVX-LABEL: merge_4i64_i64_1234:
174 ; AVX: # %bb.0:
175 ; AVX-NEXT: vmovups 8(%rdi), %ymm0
176 ; AVX-NEXT: retq
177 ;
178 ; X32-AVX-LABEL: merge_4i64_i64_1234:
179 ; X32-AVX: # %bb.0:
180 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
181 ; X32-AVX-NEXT: vmovups 8(%eax), %ymm0
182 ; X32-AVX-NEXT: retl
183 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1
184 %ptr1 = getelementptr inbounds i64, i64* %ptr, i64 2
185 %ptr2 = getelementptr inbounds i64, i64* %ptr, i64 3
186 %ptr3 = getelementptr inbounds i64, i64* %ptr, i64 4
187 %val0 = load i64, i64* %ptr0
188 %val1 = load i64, i64* %ptr1
189 %val2 = load i64, i64* %ptr2
190 %val3 = load i64, i64* %ptr3
191 %res0 = insertelement <4 x i64> undef, i64 %val0, i32 0
192 %res1 = insertelement <4 x i64> %res0, i64 %val1, i32 1
193 %res2 = insertelement <4 x i64> %res1, i64 %val2, i32 2
194 %res3 = insertelement <4 x i64> %res2, i64 %val3, i32 3
195 ret <4 x i64> %res3
196 }
198 define <4 x i64> @merge_4i64_i64_1zzu(i64* %ptr) nounwind uwtable noinline ssp {
199 ; AVX-LABEL: merge_4i64_i64_1zzu:
200 ; AVX: # %bb.0:
201 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
202 ; AVX-NEXT: retq
203 ;
204 ; X32-AVX-LABEL: merge_4i64_i64_1zzu:
205 ; X32-AVX: # %bb.0:
206 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
207 ; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
208 ; X32-AVX-NEXT: retl
209 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1
210 %val0 = load i64, i64* %ptr0
211 %res0 = insertelement <4 x i64> undef, i64 %val0, i32 0
212 %res1 = insertelement <4 x i64> %res0, i64 0, i32 1
213 %res2 = insertelement <4 x i64> %res1, i64 0, i32 2
214 ret <4 x i64> %res2
215 }
217 define <4 x i64> @merge_4i64_i64_23zz(i64* %ptr) nounwind uwtable noinline ssp {
218 ; AVX-LABEL: merge_4i64_i64_23zz:
219 ; AVX: # %bb.0:
220 ; AVX-NEXT: vmovups 16(%rdi), %xmm0
221 ; AVX-NEXT: retq
222 ;
223 ; X32-AVX-LABEL: merge_4i64_i64_23zz:
224 ; X32-AVX: # %bb.0:
225 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
226 ; X32-AVX-NEXT: vmovups 16(%eax), %xmm0
227 ; X32-AVX-NEXT: retl
228 %ptr0 = getelementptr inbounds i64, i64* %ptr, i64 2
229 %ptr1 = getelementptr inbounds i64, i64* %ptr, i64 3
230 %val0 = load i64, i64* %ptr0
231 %val1 = load i64, i64* %ptr1
232 %res0 = insertelement <4 x i64> zeroinitializer, i64 %val0, i32 0
233 %res1 = insertelement <4 x i64> %res0, i64 %val1, i32 1
234 ret <4 x i64> %res1
235 }
237 define <8 x float> @merge_8f32_2f32_23z5(<2 x float>* %ptr) nounwind uwtable noinline ssp {
238 ; AVX1-LABEL: merge_8f32_2f32_23z5:
239 ; AVX1: # %bb.0:
240 ; AVX1-NEXT: vmovups 16(%rdi), %xmm0
241 ; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1
242 ; AVX1-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
243 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
244 ; AVX1-NEXT: retq
245 ;
246 ; AVX2-LABEL: merge_8f32_2f32_23z5:
247 ; AVX2: # %bb.0:
248 ; AVX2-NEXT: vmovupd 16(%rdi), %xmm0
249 ; AVX2-NEXT: vxorpd %xmm1, %xmm1, %xmm1
250 ; AVX2-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
251 ; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
252 ; AVX2-NEXT: retq
253 ;
254 ; AVX512F-LABEL: merge_8f32_2f32_23z5:
255 ; AVX512F: # %bb.0:
256 ; AVX512F-NEXT: vmovupd 16(%rdi), %xmm0
257 ; AVX512F-NEXT: vxorpd %xmm1, %xmm1, %xmm1
258 ; AVX512F-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
259 ; AVX512F-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
260 ; AVX512F-NEXT: retq
261 ;
262 ; X32-AVX-LABEL: merge_8f32_2f32_23z5:
263 ; X32-AVX: # %bb.0:
264 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
265 ; X32-AVX-NEXT: vmovups 16(%eax), %xmm0
266 ; X32-AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
267 ; X32-AVX-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
268 ; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
269 ; X32-AVX-NEXT: retl
270 %ptr0 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 2
271 %ptr1 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 3
272 %ptr3 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 5
273 %val0 = load <2 x float>, <2 x float>* %ptr0
274 %val1 = load <2 x float>, <2 x float>* %ptr1
275 %val3 = load <2 x float>, <2 x float>* %ptr3
276 %res01 = shufflevector <2 x float> %val0, <2 x float> %val1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
277 %res23 = shufflevector <2 x float> zeroinitializer, <2 x float> %val3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
278 %res = shufflevector <4 x float> %res01, <4 x float> %res23, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
279 ret <8 x float> %res
280 }
282 define <8 x float> @merge_8f32_4f32_z2(<4 x float>* %ptr) nounwind uwtable noinline ssp {
283 ; AVX-LABEL: merge_8f32_4f32_z2:
284 ; AVX: # %bb.0:
285 ; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
286 ; AVX-NEXT: vinsertf128 $1, 32(%rdi), %ymm0, %ymm0
287 ; AVX-NEXT: retq
288 ;
289 ; X32-AVX-LABEL: merge_8f32_4f32_z2:
290 ; X32-AVX: # %bb.0:
291 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
292 ; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
293 ; X32-AVX-NEXT: vinsertf128 $1, 32(%eax), %ymm0, %ymm0
294 ; X32-AVX-NEXT: retl
295 %ptr1 = getelementptr inbounds <4 x float>, <4 x float>* %ptr, i64 2
296 %val1 = load <4 x float>, <4 x float>* %ptr1
297 %res = shufflevector <4 x float> zeroinitializer, <4 x float> %val1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
298 ret <8 x float> %res
299 }
301 define <8 x float> @merge_8f32_f32_12zzuuzz(float* %ptr) nounwind uwtable noinline ssp {
302 ; AVX-LABEL: merge_8f32_f32_12zzuuzz:
303 ; AVX: # %bb.0:
304 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
305 ; AVX-NEXT: retq
306 ;
307 ; X32-AVX-LABEL: merge_8f32_f32_12zzuuzz:
308 ; X32-AVX: # %bb.0:
309 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
310 ; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
311 ; X32-AVX-NEXT: retl
312 %ptr0 = getelementptr inbounds float, float* %ptr, i64 1
313 %ptr1 = getelementptr inbounds float, float* %ptr, i64 2
314 %val0 = load float, float* %ptr0
315 %val1 = load float, float* %ptr1
316 %res0 = insertelement <8 x float> undef, float %val0, i32 0
317 %res1 = insertelement <8 x float> %res0, float %val1, i32 1
318 %res2 = insertelement <8 x float> %res1, float 0.0, i32 2
319 %res3 = insertelement <8 x float> %res2, float 0.0, i32 3
320 %res6 = insertelement <8 x float> %res3, float 0.0, i32 6
321 %res7 = insertelement <8 x float> %res6, float 0.0, i32 7
322 ret <8 x float> %res7
323 }
325 define <8 x float> @merge_8f32_f32_1u3u5zu8(float* %ptr) nounwind uwtable noinline ssp {
326 ; AVX-LABEL: merge_8f32_f32_1u3u5zu8:
327 ; AVX: # %bb.0:
328 ; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
329 ; AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
330 ; AVX-NEXT: retq
331 ;
332 ; X32-AVX-LABEL: merge_8f32_f32_1u3u5zu8:
333 ; X32-AVX: # %bb.0:
334 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
335 ; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
336 ; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
337 ; X32-AVX-NEXT: retl
338 %ptr0 = getelementptr inbounds float, float* %ptr, i64 1
339 %ptr2 = getelementptr inbounds float, float* %ptr, i64 3
340 %ptr4 = getelementptr inbounds float, float* %ptr, i64 5
341 %ptr7 = getelementptr inbounds float, float* %ptr, i64 8
342 %val0 = load float, float* %ptr0
343 %val2 = load float, float* %ptr2
344 %val4 = load float, float* %ptr4
345 %val7 = load float, float* %ptr7
346 %res0 = insertelement <8 x float> undef, float %val0, i32 0
347 %res2 = insertelement <8 x float> %res0, float %val2, i32 2
348 %res4 = insertelement <8 x float> %res2, float %val4, i32 4
349 %res5 = insertelement <8 x float> %res4, float 0.0, i32 5
350 %res7 = insertelement <8 x float> %res5, float %val7, i32 7
351 ret <8 x float> %res7
352 }
354 define <8 x i32> @merge_8i32_4i32_z3(<4 x i32>* %ptr) nounwind uwtable noinline ssp {
355 ; AVX-LABEL: merge_8i32_4i32_z3:
356 ; AVX: # %bb.0:
357 ; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
358 ; AVX-NEXT: vinsertf128 $1, 48(%rdi), %ymm0, %ymm0
359 ; AVX-NEXT: retq
360 ;
361 ; X32-AVX-LABEL: merge_8i32_4i32_z3:
362 ; X32-AVX: # %bb.0:
363 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
364 ; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
365 ; X32-AVX-NEXT: vinsertf128 $1, 48(%eax), %ymm0, %ymm0
366 ; X32-AVX-NEXT: retl
367 %ptr1 = getelementptr inbounds <4 x i32>, <4 x i32>* %ptr, i64 3
368 %val1 = load <4 x i32>, <4 x i32>* %ptr1
369 %res = shufflevector <4 x i32> zeroinitializer, <4 x i32> %val1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
370 ret <8 x i32> %res
371 }
373 define <8 x i32> @merge_8i32_i32_56zz9uzz(i32* %ptr) nounwind uwtable noinline ssp {
374 ; AVX-LABEL: merge_8i32_i32_56zz9uzz:
375 ; AVX: # %bb.0:
376 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
377 ; AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
378 ; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
379 ; AVX-NEXT: retq
380 ;
381 ; X32-AVX-LABEL: merge_8i32_i32_56zz9uzz:
382 ; X32-AVX: # %bb.0:
383 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
384 ; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
385 ; X32-AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
386 ; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
387 ; X32-AVX-NEXT: retl
388 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 5
389 %ptr1 = getelementptr inbounds i32, i32* %ptr, i64 6
390 %ptr4 = getelementptr inbounds i32, i32* %ptr, i64 9
391 %val0 = load i32, i32* %ptr0
392 %val1 = load i32, i32* %ptr1
393 %val4 = load i32, i32* %ptr4
394 %res0 = insertelement <8 x i32> undef, i32 %val0, i32 0
395 %res1 = insertelement <8 x i32> %res0, i32 %val1, i32 1
396 %res2 = insertelement <8 x i32> %res1, i32 0, i32 2
397 %res3 = insertelement <8 x i32> %res2, i32 0, i32 3
398 %res4 = insertelement <8 x i32> %res3, i32 %val4, i32 4
399 %res6 = insertelement <8 x i32> %res4, i32 0, i32 6
400 %res7 = insertelement <8 x i32> %res6, i32 0, i32 7
401 ret <8 x i32> %res7
402 }
404 define <8 x i32> @merge_8i32_i32_1u3u5zu8(i32* %ptr) nounwind uwtable noinline ssp {
405 ; AVX-LABEL: merge_8i32_i32_1u3u5zu8:
406 ; AVX: # %bb.0:
407 ; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
408 ; AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
409 ; AVX-NEXT: retq
410 ;
411 ; X32-AVX-LABEL: merge_8i32_i32_1u3u5zu8:
412 ; X32-AVX: # %bb.0:
413 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
414 ; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
415 ; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
416 ; X32-AVX-NEXT: retl
417 %ptr0 = getelementptr inbounds i32, i32* %ptr, i64 1
418 %ptr2 = getelementptr inbounds i32, i32* %ptr, i64 3
419 %ptr4 = getelementptr inbounds i32, i32* %ptr, i64 5
420 %ptr7 = getelementptr inbounds i32, i32* %ptr, i64 8
421 %val0 = load i32, i32* %ptr0
422 %val2 = load i32, i32* %ptr2
423 %val4 = load i32, i32* %ptr4
424 %val7 = load i32, i32* %ptr7
425 %res0 = insertelement <8 x i32> undef, i32 %val0, i32 0
426 %res2 = insertelement <8 x i32> %res0, i32 %val2, i32 2
427 %res4 = insertelement <8 x i32> %res2, i32 %val4, i32 4
428 %res5 = insertelement <8 x i32> %res4, i32 0, i32 5
429 %res7 = insertelement <8 x i32> %res5, i32 %val7, i32 7
430 ret <8 x i32> %res7
431 }
433 define <16 x i16> @merge_16i16_i16_89zzzuuuuuuuuuuuz(i16* %ptr) nounwind uwtable noinline ssp {
434 ; AVX-LABEL: merge_16i16_i16_89zzzuuuuuuuuuuuz:
435 ; AVX: # %bb.0:
436 ; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
437 ; AVX-NEXT: retq
438 ;
439 ; X32-AVX-LABEL: merge_16i16_i16_89zzzuuuuuuuuuuuz:
440 ; X32-AVX: # %bb.0:
441 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
442 ; X32-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
443 ; X32-AVX-NEXT: retl
444 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 8
445 %ptr1 = getelementptr inbounds i16, i16* %ptr, i64 9
446 %val0 = load i16, i16* %ptr0
447 %val1 = load i16, i16* %ptr1
448 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
449 %res1 = insertelement <16 x i16> %res0, i16 %val1, i16 1
450 %res2 = insertelement <16 x i16> %res1, i16 0, i16 2
451 %res3 = insertelement <16 x i16> %res2, i16 0, i16 3
452 %res4 = insertelement <16 x i16> %res3, i16 0, i16 4
453 %resF = insertelement <16 x i16> %res4, i16 0, i16 15
454 ret <16 x i16> %resF
455 }
457 define <16 x i16> @merge_16i16_i16_45u7uuuuuuuuuuuu(i16* %ptr) nounwind uwtable noinline ssp {
458 ; AVX-LABEL: merge_16i16_i16_45u7uuuuuuuuuuuu:
459 ; AVX: # %bb.0:
460 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
461 ; AVX-NEXT: retq
462 ;
463 ; X32-AVX-LABEL: merge_16i16_i16_45u7uuuuuuuuuuuu:
464 ; X32-AVX: # %bb.0:
465 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
466 ; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
467 ; X32-AVX-NEXT: retl
468 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 4
469 %ptr1 = getelementptr inbounds i16, i16* %ptr, i64 5
470 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 7
471 %val0 = load i16, i16* %ptr0
472 %val1 = load i16, i16* %ptr1
473 %val3 = load i16, i16* %ptr3
474 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
475 %res1 = insertelement <16 x i16> %res0, i16 %val1, i16 1
476 %res3 = insertelement <16 x i16> %res1, i16 %val3, i16 3
477 ret <16 x i16> %res3
478 }
480 define <16 x i16> @merge_16i16_i16_0uu3uuuuuuuuCuEF(i16* %ptr) nounwind uwtable noinline ssp {
481 ; AVX-LABEL: merge_16i16_i16_0uu3uuuuuuuuCuEF:
482 ; AVX: # %bb.0:
483 ; AVX-NEXT: vmovups (%rdi), %ymm0
484 ; AVX-NEXT: retq
485 ;
486 ; X32-AVX-LABEL: merge_16i16_i16_0uu3uuuuuuuuCuEF:
487 ; X32-AVX: # %bb.0:
488 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
489 ; X32-AVX-NEXT: vmovups (%eax), %ymm0
490 ; X32-AVX-NEXT: retl
491 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0
492 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3
493 %ptrC = getelementptr inbounds i16, i16* %ptr, i64 12
494 %ptrE = getelementptr inbounds i16, i16* %ptr, i64 14
495 %ptrF = getelementptr inbounds i16, i16* %ptr, i64 15
496 %val0 = load i16, i16* %ptr0
497 %val3 = load i16, i16* %ptr3
498 %valC = load i16, i16* %ptrC
499 %valE = load i16, i16* %ptrE
500 %valF = load i16, i16* %ptrF
501 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
502 %res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3
503 %resC = insertelement <16 x i16> %res3, i16 %valC, i16 12
504 %resE = insertelement <16 x i16> %resC, i16 %valE, i16 14
505 %resF = insertelement <16 x i16> %resE, i16 %valF, i16 15
506 ret <16 x i16> %resF
507 }
509 define <16 x i16> @merge_16i16_i16_0uu3zzuuuuuzCuEF(i16* %ptr) nounwind uwtable noinline ssp {
510 ; AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF:
511 ; AVX: # %bb.0:
512 ; AVX-NEXT: vmovups (%rdi), %ymm0
513 ; AVX-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0
514 ; AVX-NEXT: retq
515 ;
516 ; X32-AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF:
517 ; X32-AVX: # %bb.0:
518 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
519 ; X32-AVX-NEXT: vmovups (%eax), %ymm0
520 ; X32-AVX-NEXT: vandps {{\.LCPI.*}}, %ymm0, %ymm0
521 ; X32-AVX-NEXT: retl
522 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0
523 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3
524 %ptrC = getelementptr inbounds i16, i16* %ptr, i64 12
525 %ptrE = getelementptr inbounds i16, i16* %ptr, i64 14
526 %ptrF = getelementptr inbounds i16, i16* %ptr, i64 15
527 %val0 = load i16, i16* %ptr0
528 %val3 = load i16, i16* %ptr3
529 %valC = load i16, i16* %ptrC
530 %valE = load i16, i16* %ptrE
531 %valF = load i16, i16* %ptrF
532 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
533 %res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3
534 %res4 = insertelement <16 x i16> %res3, i16 0, i16 4
535 %res5 = insertelement <16 x i16> %res4, i16 0, i16 5
536 %resC = insertelement <16 x i16> %res5, i16 %valC, i16 12
537 %resD = insertelement <16 x i16> %resC, i16 0, i16 13
538 %resE = insertelement <16 x i16> %resD, i16 %valE, i16 14
539 %resF = insertelement <16 x i16> %resE, i16 %valF, i16 15
540 ret <16 x i16> %resF
541 }
543 define <32 x i8> @merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu(i8* %ptr) nounwind uwtable noinline ssp {
544 ; AVX-LABEL: merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu:
545 ; AVX: # %bb.0:
546 ; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
547 ; AVX-NEXT: retq
548 ;
549 ; X32-AVX-LABEL: merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu:
550 ; X32-AVX: # %bb.0:
551 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
552 ; X32-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
553 ; X32-AVX-NEXT: retl
554 %ptr0 = getelementptr inbounds i8, i8* %ptr, i64 4
555 %ptr1 = getelementptr inbounds i8, i8* %ptr, i64 5
556 %ptr3 = getelementptr inbounds i8, i8* %ptr, i64 7
557 %val0 = load i8, i8* %ptr0
558 %val1 = load i8, i8* %ptr1
559 %val3 = load i8, i8* %ptr3
560 %res0 = insertelement <32 x i8> undef, i8 %val0, i8 0
561 %res1 = insertelement <32 x i8> %res0, i8 %val1, i8 1
562 %res3 = insertelement <32 x i8> %res1, i8 %val3, i8 3
563 ret <32 x i8> %res3
564 }
566 define <32 x i8> @merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu(i8* %ptr) nounwind uwtable noinline ssp {
567 ; AVX-LABEL: merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu:
568 ; AVX: # %bb.0:
569 ; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
570 ; AVX-NEXT: retq
571 ;
572 ; X32-AVX-LABEL: merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu:
573 ; X32-AVX: # %bb.0:
574 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
575 ; X32-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
576 ; X32-AVX-NEXT: retl
577 %ptr0 = getelementptr inbounds i8, i8* %ptr, i64 2
578 %ptr1 = getelementptr inbounds i8, i8* %ptr, i64 3
579 %ptr3 = getelementptr inbounds i8, i8* %ptr, i64 5
580 %val0 = load i8, i8* %ptr0
581 %val1 = load i8, i8* %ptr1
582 %val3 = load i8, i8* %ptr3
583 %res0 = insertelement <32 x i8> undef, i8 %val0, i8 0
584 %res1 = insertelement <32 x i8> %res0, i8 %val1, i8 1
585 %res3 = insertelement <32 x i8> %res1, i8 %val3, i8 3
586 %resE = insertelement <32 x i8> %res3, i8 0, i8 14
587 %resF = insertelement <32 x i8> %resE, i8 0, i8 15
588 %resG = insertelement <32 x i8> %resF, i8 0, i8 16
589 %resH = insertelement <32 x i8> %resG, i8 0, i8 17
590 ret <32 x i8> %resH
591 }
593 ;
594 ; consecutive loads including any/all volatiles may not be combined
595 ;
597 define <4 x double> @merge_4f64_f64_34uz_volatile(double* %ptr) nounwind uwtable noinline ssp {
598 ; AVX-LABEL: merge_4f64_f64_34uz_volatile:
599 ; AVX: # %bb.0:
600 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
601 ; AVX-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
602 ; AVX-NEXT: retq
603 ;
604 ; X32-AVX-LABEL: merge_4f64_f64_34uz_volatile:
605 ; X32-AVX: # %bb.0:
606 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
607 ; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
608 ; X32-AVX-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
609 ; X32-AVX-NEXT: retl
610 %ptr0 = getelementptr inbounds double, double* %ptr, i64 3
611 %ptr1 = getelementptr inbounds double, double* %ptr, i64 4
612 %val0 = load volatile double, double* %ptr0
613 %val1 = load volatile double, double* %ptr1
614 %res0 = insertelement <4 x double> undef, double %val0, i32 0
615 %res1 = insertelement <4 x double> %res0, double %val1, i32 1
616 %res3 = insertelement <4 x double> %res1, double 0.0, i32 3
617 ret <4 x double> %res3
618 }
620 define <16 x i16> @merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile(i16* %ptr) nounwind uwtable noinline ssp {
621 ; AVX1-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
622 ; AVX1: # %bb.0:
623 ; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0
624 ; AVX1-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1
625 ; AVX1-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0
626 ; AVX1-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0
627 ; AVX1-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0
628 ; AVX1-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1
629 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
630 ; AVX1-NEXT: retq
631 ;
632 ; AVX2-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
633 ; AVX2: # %bb.0:
634 ; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0
635 ; AVX2-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1
636 ; AVX2-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0
637 ; AVX2-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0
638 ; AVX2-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0
639 ; AVX2-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1
640 ; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
641 ; AVX2-NEXT: retq
642 ;
643 ; AVX512F-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
644 ; AVX512F: # %bb.0:
645 ; AVX512F-NEXT: vpxor %xmm0, %xmm0, %xmm0
646 ; AVX512F-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1
647 ; AVX512F-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0
648 ; AVX512F-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0
649 ; AVX512F-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0
650 ; AVX512F-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1
651 ; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
652 ; AVX512F-NEXT: retq
653 ;
654 ; X32-AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
655 ; X32-AVX: # %bb.0:
656 ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
657 ; X32-AVX-NEXT: vpxor %xmm0, %xmm0, %xmm0
658 ; X32-AVX-NEXT: vpinsrw $0, (%eax), %xmm0, %xmm1
659 ; X32-AVX-NEXT: vpinsrw $4, 24(%eax), %xmm0, %xmm0
660 ; X32-AVX-NEXT: vpinsrw $6, 28(%eax), %xmm0, %xmm0
661 ; X32-AVX-NEXT: vpinsrw $7, 30(%eax), %xmm0, %xmm0
662 ; X32-AVX-NEXT: vpinsrw $3, 6(%eax), %xmm1, %xmm1
663 ; X32-AVX-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
664 ; X32-AVX-NEXT: retl
665 %ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0
666 %ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3
667 %ptrC = getelementptr inbounds i16, i16* %ptr, i64 12
668 %ptrE = getelementptr inbounds i16, i16* %ptr, i64 14
669 %ptrF = getelementptr inbounds i16, i16* %ptr, i64 15
670 %val0 = load volatile i16, i16* %ptr0
671 %val3 = load i16, i16* %ptr3
672 %valC = load i16, i16* %ptrC
673 %valE = load i16, i16* %ptrE
674 %valF = load volatile i16, i16* %ptrF
675 %res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
676 %res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3
677 %res4 = insertelement <16 x i16> %res3, i16 0, i16 4
678 %res5 = insertelement <16 x i16> %res4, i16 0, i16 5
679 %resC = insertelement <16 x i16> %res5, i16 %valC, i16 12
680 %resD = insertelement <16 x i16> %resC, i16 0, i16 13
681 %resE = insertelement <16 x i16> %resD, i16 %valE, i16 14
682 %resF = insertelement <16 x i16> %resE, i16 %valF, i16 15
683 ret <16 x i16> %resF
684 }