• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F
3; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
4
5define   <16 x i32> @_inreg16xi32(i32 %a) {
6; ALL-LABEL: _inreg16xi32:
7; ALL:       # BB#0:
8; ALL-NEXT:    vpbroadcastd %edi, %zmm0
9; ALL-NEXT:    retq
10  %b = insertelement <16 x i32> undef, i32 %a, i32 0
11  %c = shufflevector <16 x i32> %b, <16 x i32> undef, <16 x i32> zeroinitializer
12  ret <16 x i32> %c
13}
14
15define   <8 x i64> @_inreg8xi64(i64 %a) {
16; ALL-LABEL: _inreg8xi64:
17; ALL:       # BB#0:
18; ALL-NEXT:    vpbroadcastq %rdi, %zmm0
19; ALL-NEXT:    retq
20  %b = insertelement <8 x i64> undef, i64 %a, i32 0
21  %c = shufflevector <8 x i64> %b, <8 x i64> undef, <8 x i32> zeroinitializer
22  ret <8 x i64> %c
23}
24
25define   <16 x float> @_ss16xfloat_v4(<4 x float> %a) {
26; ALL-LABEL: _ss16xfloat_v4:
27; ALL:       # BB#0:
28; ALL-NEXT:    vbroadcastss %xmm0, %zmm0
29; ALL-NEXT:    retq
30  %b = shufflevector <4 x float> %a, <4 x float> undef, <16 x i32> zeroinitializer
31  ret <16 x float> %b
32}
33
34define   <16 x float> @_inreg16xfloat(float %a) {
35; ALL-LABEL: _inreg16xfloat:
36; ALL:       # BB#0:
37; ALL-NEXT:    vbroadcastss %xmm0, %zmm0
38; ALL-NEXT:    retq
39  %b = insertelement <16 x float> undef, float %a, i32 0
40  %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
41  ret <16 x float> %c
42}
43
44define   <16 x float> @_ss16xfloat_mask(float %a, <16 x float> %i, <16 x i32> %mask1) {
45; ALL-LABEL: _ss16xfloat_mask:
46; ALL:       # BB#0:
47; ALL-NEXT:    vpxord %zmm3, %zmm3, %zmm3
48; ALL-NEXT:    vpcmpneqd %zmm3, %zmm2, %k1
49; ALL-NEXT:    vbroadcastss %xmm0, %zmm1 {%k1}
50; ALL-NEXT:    vmovaps %zmm1, %zmm0
51; ALL-NEXT:    retq
52  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
53  %b = insertelement <16 x float> undef, float %a, i32 0
54  %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
55  %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> %i
56  ret <16 x float> %r
57}
58
59define   <16 x float> @_ss16xfloat_maskz(float %a, <16 x i32> %mask1) {
60; ALL-LABEL: _ss16xfloat_maskz:
61; ALL:       # BB#0:
62; ALL-NEXT:    vpxord %zmm2, %zmm2, %zmm2
63; ALL-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
64; ALL-NEXT:    vbroadcastss %xmm0, %zmm0 {%k1} {z}
65; ALL-NEXT:    retq
66  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
67  %b = insertelement <16 x float> undef, float %a, i32 0
68  %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
69  %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> zeroinitializer
70  ret <16 x float> %r
71}
72
73define   <16 x float> @_ss16xfloat_load(float* %a.ptr) {
74; ALL-LABEL: _ss16xfloat_load:
75; ALL:       # BB#0:
76; ALL-NEXT:    vbroadcastss (%rdi), %zmm0
77; ALL-NEXT:    retq
78  %a = load float, float* %a.ptr
79  %b = insertelement <16 x float> undef, float %a, i32 0
80  %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
81  ret <16 x float> %c
82}
83
84define   <16 x float> @_ss16xfloat_mask_load(float* %a.ptr, <16 x float> %i, <16 x i32> %mask1) {
85; ALL-LABEL: _ss16xfloat_mask_load:
86; ALL:       # BB#0:
87; ALL-NEXT:    vpxord %zmm2, %zmm2, %zmm2
88; ALL-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
89; ALL-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1}
90; ALL-NEXT:    retq
91  %a = load float, float* %a.ptr
92  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
93  %b = insertelement <16 x float> undef, float %a, i32 0
94  %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
95  %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> %i
96  ret <16 x float> %r
97}
98
99define   <16 x float> @_ss16xfloat_maskz_load(float* %a.ptr, <16 x i32> %mask1) {
100; ALL-LABEL: _ss16xfloat_maskz_load:
101; ALL:       # BB#0:
102; ALL-NEXT:    vpxord %zmm1, %zmm1, %zmm1
103; ALL-NEXT:    vpcmpneqd %zmm1, %zmm0, %k1
104; ALL-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1} {z}
105; ALL-NEXT:    retq
106  %a = load float, float* %a.ptr
107  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
108  %b = insertelement <16 x float> undef, float %a, i32 0
109  %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
110  %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> zeroinitializer
111  ret <16 x float> %r
112}
113
114define   <8 x double> @_inreg8xdouble(double %a) {
115; ALL-LABEL: _inreg8xdouble:
116; ALL:       # BB#0:
117; ALL-NEXT:    vbroadcastsd %xmm0, %zmm0
118; ALL-NEXT:    retq
119  %b = insertelement <8 x double> undef, double %a, i32 0
120  %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
121  ret <8 x double> %c
122}
123
124define   <8 x double> @_sd8xdouble_mask(double %a, <8 x double> %i, <8 x i32> %mask1) {
125; ALL-LABEL: _sd8xdouble_mask:
126; ALL:       # BB#0:
127; ALL-NEXT:    # kill: %YMM2<def> %YMM2<kill> %ZMM2<def>
128; ALL-NEXT:    vpxor %ymm3, %ymm3, %ymm3
129; ALL-NEXT:    vpcmpneqd %zmm3, %zmm2, %k1
130; ALL-NEXT:    vbroadcastsd %xmm0, %zmm1 {%k1}
131; ALL-NEXT:    vmovaps %zmm1, %zmm0
132; ALL-NEXT:    retq
133  %mask = icmp ne <8 x i32> %mask1, zeroinitializer
134  %b = insertelement <8 x double> undef, double %a, i32 0
135  %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
136  %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> %i
137  ret <8 x double> %r
138}
139
140define   <8 x double> @_sd8xdouble_maskz(double %a, <8 x i32> %mask1) {
141; ALL-LABEL: _sd8xdouble_maskz:
142; ALL:       # BB#0:
143; ALL-NEXT:    # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
144; ALL-NEXT:    vpxor %ymm2, %ymm2, %ymm2
145; ALL-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
146; ALL-NEXT:    vbroadcastsd %xmm0, %zmm0 {%k1} {z}
147; ALL-NEXT:    retq
148  %mask = icmp ne <8 x i32> %mask1, zeroinitializer
149  %b = insertelement <8 x double> undef, double %a, i32 0
150  %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
151  %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> zeroinitializer
152  ret <8 x double> %r
153}
154
155define   <8 x double> @_sd8xdouble_load(double* %a.ptr) {
156; ALL-LABEL: _sd8xdouble_load:
157; ALL:       # BB#0:
158; ALL-NEXT:    vbroadcastsd (%rdi), %zmm0
159; ALL-NEXT:    retq
160  %a = load double, double* %a.ptr
161  %b = insertelement <8 x double> undef, double %a, i32 0
162  %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
163  ret <8 x double> %c
164}
165
166define   <8 x double> @_sd8xdouble_mask_load(double* %a.ptr, <8 x double> %i, <8 x i32> %mask1) {
167; ALL-LABEL: _sd8xdouble_mask_load:
168; ALL:       # BB#0:
169; ALL-NEXT:    # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
170; ALL-NEXT:    vpxor %ymm2, %ymm2, %ymm2
171; ALL-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
172; ALL-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1}
173; ALL-NEXT:    retq
174  %a = load double, double* %a.ptr
175  %mask = icmp ne <8 x i32> %mask1, zeroinitializer
176  %b = insertelement <8 x double> undef, double %a, i32 0
177  %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
178  %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> %i
179  ret <8 x double> %r
180}
181
182define   <8 x double> @_sd8xdouble_maskz_load(double* %a.ptr, <8 x i32> %mask1) {
183; ALL-LABEL: _sd8xdouble_maskz_load:
184; ALL:       # BB#0:
185; ALL-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
186; ALL-NEXT:    vpxor %ymm1, %ymm1, %ymm1
187; ALL-NEXT:    vpcmpneqd %zmm1, %zmm0, %k1
188; ALL-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
189; ALL-NEXT:    retq
190  %a = load double, double* %a.ptr
191  %mask = icmp ne <8 x i32> %mask1, zeroinitializer
192  %b = insertelement <8 x double> undef, double %a, i32 0
193  %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
194  %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> zeroinitializer
195  ret <8 x double> %r
196}
197
198define   <16 x i32> @_xmm16xi32(<16 x i32> %a) {
199; ALL-LABEL: _xmm16xi32:
200; ALL:       # BB#0:
201; ALL-NEXT:    vpbroadcastd %xmm0, %zmm0
202; ALL-NEXT:    retq
203  %b = shufflevector <16 x i32> %a, <16 x i32> undef, <16 x i32> zeroinitializer
204  ret <16 x i32> %b
205}
206
207define   <16 x float> @_xmm16xfloat(<16 x float> %a) {
208; ALL-LABEL: _xmm16xfloat:
209; ALL:       # BB#0:
210; ALL-NEXT:    vbroadcastss %xmm0, %zmm0
211; ALL-NEXT:    retq
212  %b = shufflevector <16 x float> %a, <16 x float> undef, <16 x i32> zeroinitializer
213  ret <16 x float> %b
214}
215
216define <16 x i32> @test_vbroadcast() {
217; ALL-LABEL: test_vbroadcast:
218; ALL:       # BB#0: # %entry
219; ALL-NEXT:    vpxord %zmm0, %zmm0, %zmm0
220; ALL-NEXT:    vcmpunordps %zmm0, %zmm0, %k1
221; ALL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
222; ALL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
223; ALL-NEXT:    knotw %k1, %k1
224; ALL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
225; ALL-NEXT:    retq
226entry:
227  %0 = sext <16 x i1> zeroinitializer to <16 x i32>
228  %1 = fcmp uno <16 x float> undef, zeroinitializer
229  %2 = sext <16 x i1> %1 to <16 x i32>
230  %3 = select <16 x i1> %1, <16 x i32> %0, <16 x i32> %2
231  ret <16 x i32> %3
232}
233
234; We implement the set1 intrinsics with vector initializers.  Verify that the
235; IR generated will produce broadcasts at the end.
236define <8 x double> @test_set1_pd(double %d) #2 {
237; ALL-LABEL: test_set1_pd:
238; ALL:       # BB#0: # %entry
239; ALL-NEXT:    vbroadcastsd %xmm0, %zmm0
240; ALL-NEXT:    retq
241entry:
242  %vecinit.i = insertelement <8 x double> undef, double %d, i32 0
243  %vecinit1.i = insertelement <8 x double> %vecinit.i, double %d, i32 1
244  %vecinit2.i = insertelement <8 x double> %vecinit1.i, double %d, i32 2
245  %vecinit3.i = insertelement <8 x double> %vecinit2.i, double %d, i32 3
246  %vecinit4.i = insertelement <8 x double> %vecinit3.i, double %d, i32 4
247  %vecinit5.i = insertelement <8 x double> %vecinit4.i, double %d, i32 5
248  %vecinit6.i = insertelement <8 x double> %vecinit5.i, double %d, i32 6
249  %vecinit7.i = insertelement <8 x double> %vecinit6.i, double %d, i32 7
250  ret <8 x double> %vecinit7.i
251}
252
253define <8 x i64> @test_set1_epi64(i64 %d) #2 {
254; ALL-LABEL: test_set1_epi64:
255; ALL:       # BB#0: # %entry
256; ALL-NEXT:    vpbroadcastq %rdi, %zmm0
257; ALL-NEXT:    retq
258entry:
259  %vecinit.i = insertelement <8 x i64> undef, i64 %d, i32 0
260  %vecinit1.i = insertelement <8 x i64> %vecinit.i, i64 %d, i32 1
261  %vecinit2.i = insertelement <8 x i64> %vecinit1.i, i64 %d, i32 2
262  %vecinit3.i = insertelement <8 x i64> %vecinit2.i, i64 %d, i32 3
263  %vecinit4.i = insertelement <8 x i64> %vecinit3.i, i64 %d, i32 4
264  %vecinit5.i = insertelement <8 x i64> %vecinit4.i, i64 %d, i32 5
265  %vecinit6.i = insertelement <8 x i64> %vecinit5.i, i64 %d, i32 6
266  %vecinit7.i = insertelement <8 x i64> %vecinit6.i, i64 %d, i32 7
267  ret <8 x i64> %vecinit7.i
268}
269
270define <16 x float> @test_set1_ps(float %f) #2 {
271; ALL-LABEL: test_set1_ps:
272; ALL:       # BB#0: # %entry
273; ALL-NEXT:    vbroadcastss %xmm0, %zmm0
274; ALL-NEXT:    retq
275entry:
276  %vecinit.i = insertelement <16 x float> undef, float %f, i32 0
277  %vecinit1.i = insertelement <16 x float> %vecinit.i, float %f, i32 1
278  %vecinit2.i = insertelement <16 x float> %vecinit1.i, float %f, i32 2
279  %vecinit3.i = insertelement <16 x float> %vecinit2.i, float %f, i32 3
280  %vecinit4.i = insertelement <16 x float> %vecinit3.i, float %f, i32 4
281  %vecinit5.i = insertelement <16 x float> %vecinit4.i, float %f, i32 5
282  %vecinit6.i = insertelement <16 x float> %vecinit5.i, float %f, i32 6
283  %vecinit7.i = insertelement <16 x float> %vecinit6.i, float %f, i32 7
284  %vecinit8.i = insertelement <16 x float> %vecinit7.i, float %f, i32 8
285  %vecinit9.i = insertelement <16 x float> %vecinit8.i, float %f, i32 9
286  %vecinit10.i = insertelement <16 x float> %vecinit9.i, float %f, i32 10
287  %vecinit11.i = insertelement <16 x float> %vecinit10.i, float %f, i32 11
288  %vecinit12.i = insertelement <16 x float> %vecinit11.i, float %f, i32 12
289  %vecinit13.i = insertelement <16 x float> %vecinit12.i, float %f, i32 13
290  %vecinit14.i = insertelement <16 x float> %vecinit13.i, float %f, i32 14
291  %vecinit15.i = insertelement <16 x float> %vecinit14.i, float %f, i32 15
292  ret <16 x float> %vecinit15.i
293}
294
295define <16 x i32> @test_set1_epi32(i32 %f) #2 {
296; ALL-LABEL: test_set1_epi32:
297; ALL:       # BB#0: # %entry
298; ALL-NEXT:    vpbroadcastd %edi, %zmm0
299; ALL-NEXT:    retq
300entry:
301  %vecinit.i = insertelement <16 x i32> undef, i32 %f, i32 0
302  %vecinit1.i = insertelement <16 x i32> %vecinit.i, i32 %f, i32 1
303  %vecinit2.i = insertelement <16 x i32> %vecinit1.i, i32 %f, i32 2
304  %vecinit3.i = insertelement <16 x i32> %vecinit2.i, i32 %f, i32 3
305  %vecinit4.i = insertelement <16 x i32> %vecinit3.i, i32 %f, i32 4
306  %vecinit5.i = insertelement <16 x i32> %vecinit4.i, i32 %f, i32 5
307  %vecinit6.i = insertelement <16 x i32> %vecinit5.i, i32 %f, i32 6
308  %vecinit7.i = insertelement <16 x i32> %vecinit6.i, i32 %f, i32 7
309  %vecinit8.i = insertelement <16 x i32> %vecinit7.i, i32 %f, i32 8
310  %vecinit9.i = insertelement <16 x i32> %vecinit8.i, i32 %f, i32 9
311  %vecinit10.i = insertelement <16 x i32> %vecinit9.i, i32 %f, i32 10
312  %vecinit11.i = insertelement <16 x i32> %vecinit10.i, i32 %f, i32 11
313  %vecinit12.i = insertelement <16 x i32> %vecinit11.i, i32 %f, i32 12
314  %vecinit13.i = insertelement <16 x i32> %vecinit12.i, i32 %f, i32 13
315  %vecinit14.i = insertelement <16 x i32> %vecinit13.i, i32 %f, i32 14
316  %vecinit15.i = insertelement <16 x i32> %vecinit14.i, i32 %f, i32 15
317  ret <16 x i32> %vecinit15.i
318}
319
320; We implement the scalar broadcast intrinsics with vector initializers.
321; Verify that the IR generated will produce the broadcast at the end.
322define <8 x double> @test_mm512_broadcastsd_pd(<2 x double> %a) {
323; ALL-LABEL: test_mm512_broadcastsd_pd:
324; ALL:       # BB#0: # %entry
325; ALL-NEXT:    vbroadcastsd %xmm0, %zmm0
326; ALL-NEXT:    retq
327entry:
328  %0 = extractelement <2 x double> %a, i32 0
329  %vecinit.i = insertelement <8 x double> undef, double %0, i32 0
330  %vecinit1.i = insertelement <8 x double> %vecinit.i, double %0, i32 1
331  %vecinit2.i = insertelement <8 x double> %vecinit1.i, double %0, i32 2
332  %vecinit3.i = insertelement <8 x double> %vecinit2.i, double %0, i32 3
333  %vecinit4.i = insertelement <8 x double> %vecinit3.i, double %0, i32 4
334  %vecinit5.i = insertelement <8 x double> %vecinit4.i, double %0, i32 5
335  %vecinit6.i = insertelement <8 x double> %vecinit5.i, double %0, i32 6
336  %vecinit7.i = insertelement <8 x double> %vecinit6.i, double %0, i32 7
337  ret <8 x double> %vecinit7.i
338}
339
340define <16 x float> @test1(<8 x float>%a)  {
341; ALL-LABEL: test1:
342; ALL:       # BB#0:
343; ALL-NEXT:    vbroadcastss %xmm0, %zmm0
344; ALL-NEXT:    retq
345  %res = shufflevector <8 x float> %a, <8 x float> undef, <16 x i32> zeroinitializer
346  ret <16 x float>%res
347}
348
349define <8 x double> @test2(<4 x double>%a)  {
350; ALL-LABEL: test2:
351; ALL:       # BB#0:
352; ALL-NEXT:    vbroadcastsd %xmm0, %zmm0
353; ALL-NEXT:    retq
354  %res = shufflevector <4 x double> %a, <4 x double> undef, <8 x i32> zeroinitializer
355  ret <8 x double>%res
356}
357
358define <64 x i8> @_invec32xi8(<32 x i8>%a)  {
359; AVX512F-LABEL: _invec32xi8:
360; AVX512F:       # BB#0:
361; AVX512F-NEXT:    vpbroadcastb %xmm0, %ymm0
362; AVX512F-NEXT:    vmovaps %zmm0, %zmm1
363; AVX512F-NEXT:    retq
364;
365; AVX512BW-LABEL: _invec32xi8:
366; AVX512BW:       # BB#0:
367; AVX512BW-NEXT:    vpbroadcastb %xmm0, %zmm0
368; AVX512BW-NEXT:    retq
369  %res = shufflevector <32 x i8> %a, <32 x i8> undef, <64 x i32> zeroinitializer
370  ret <64 x i8>%res
371}
372
373define <32 x i16> @_invec16xi16(<16 x i16>%a)  {
374; AVX512F-LABEL: _invec16xi16:
375; AVX512F:       # BB#0:
376; AVX512F-NEXT:    vpbroadcastw %xmm0, %ymm0
377; AVX512F-NEXT:    vmovaps %zmm0, %zmm1
378; AVX512F-NEXT:    retq
379;
380; AVX512BW-LABEL: _invec16xi16:
381; AVX512BW:       # BB#0:
382; AVX512BW-NEXT:    vpbroadcastw %xmm0, %zmm0
383; AVX512BW-NEXT:    retq
384  %res = shufflevector <16 x i16> %a, <16 x i16> undef, <32 x i32> zeroinitializer
385  ret <32 x i16>%res
386}
387
388define <16 x i32> @_invec8xi32(<8 x i32>%a)  {
389; ALL-LABEL: _invec8xi32:
390; ALL:       # BB#0:
391; ALL-NEXT:    vpbroadcastd %xmm0, %zmm0
392; ALL-NEXT:    retq
393  %res = shufflevector <8 x i32> %a, <8 x i32> undef, <16 x i32> zeroinitializer
394  ret <16 x i32>%res
395}
396
397define <8 x i64> @_invec4xi64(<4 x i64>%a)  {
398; ALL-LABEL: _invec4xi64:
399; ALL:       # BB#0:
400; ALL-NEXT:    vpbroadcastq %xmm0, %zmm0
401; ALL-NEXT:    retq
402  %res = shufflevector <4 x i64> %a, <4 x i64> undef, <8 x i32> zeroinitializer
403  ret <8 x i64>%res
404}
405
406declare void @func_f32(float)
407define <16 x float> @broadcast_ss_spill(float %x) {
408; ALL-LABEL: broadcast_ss_spill:
409; ALL:       # BB#0:
410; ALL-NEXT:    pushq %rax
411; ALL-NEXT:  .Ltmp0:
412; ALL-NEXT:    .cfi_def_cfa_offset 16
413; ALL-NEXT:    vaddss %xmm0, %xmm0, %xmm0
414; ALL-NEXT:    vmovss %xmm0, {{[0-9]+}}(%rsp) # 4-byte Folded Spill
415; ALL-NEXT:    callq func_f32
416; ALL-NEXT:    vbroadcastss {{[0-9]+}}(%rsp), %zmm0 # 4-byte Folded Reload
417; ALL-NEXT:    popq %rax
418; ALL-NEXT:    retq
419  %a  = fadd float %x, %x
420  call void @func_f32(float %a)
421  %b = insertelement <16 x float> undef, float %a, i32 0
422  %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
423  ret <16 x float> %c
424}
425
426declare void @func_f64(double)
427define <8 x double> @broadcast_sd_spill(double %x) {
428; ALL-LABEL: broadcast_sd_spill:
429; ALL:       # BB#0:
430; ALL-NEXT:    pushq %rax
431; ALL-NEXT:  .Ltmp1:
432; ALL-NEXT:    .cfi_def_cfa_offset 16
433; ALL-NEXT:    vaddsd %xmm0, %xmm0, %xmm0
434; ALL-NEXT:    vmovsd %xmm0, (%rsp) # 8-byte Folded Spill
435; ALL-NEXT:    callq func_f64
436; ALL-NEXT:    vbroadcastsd (%rsp), %zmm0 # 8-byte Folded Reload
437; ALL-NEXT:    popq %rax
438; ALL-NEXT:    retq
439  %a  = fadd double %x, %x
440  call void @func_f64(double %a)
441  %b = insertelement <8 x double> undef, double %a, i32 0
442  %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
443  ret <8 x double> %c
444}
445