1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-- -mattr=avx | FileCheck %s
3
4; PR38527 - https://bugs.llvm.org/show_bug.cgi?id=38527
5
6; Use an AVX target to show that the potential problem
7; is not limited to 128-bit types/registers. Ie, widening
8; up to 256-bits may also result in bogus libcalls.
9
10; Use fsin as the representative test for various data types.
11
12declare <1 x float> @llvm.sin.v1f32(<1 x float>)
13declare <2 x float> @llvm.sin.v2f32(<2 x float>)
14declare <3 x float> @llvm.sin.v3f32(<3 x float>)
15declare <4 x float> @llvm.sin.v4f32(<4 x float>)
16declare <5 x float> @llvm.sin.v5f32(<5 x float>)
17declare <6 x float> @llvm.sin.v6f32(<6 x float>)
18declare <3 x double> @llvm.sin.v3f64(<3 x double>)
19
20; Verify that all of the potential libcall candidates are handled.
21; Some of these have custom lowering, so those cases won't have
22; libcalls.
23
24declare <2 x float> @llvm.fabs.v2f32(<2 x float>)
25declare <2 x float> @llvm.ceil.v2f32(<2 x float>)
26declare <2 x float> @llvm.cos.v2f32(<2 x float>)
27declare <2 x float> @llvm.exp.v2f32(<2 x float>)
28declare <2 x float> @llvm.exp2.v2f32(<2 x float>)
29declare <2 x float> @llvm.floor.v2f32(<2 x float>)
30declare <2 x float> @llvm.log.v2f32(<2 x float>)
31declare <2 x float> @llvm.log10.v2f32(<2 x float>)
32declare <2 x float> @llvm.log2.v2f32(<2 x float>)
33declare <2 x float> @llvm.nearbyint.v2f32(<2 x float>)
34declare <2 x float> @llvm.rint.v2f32(<2 x float>)
35declare <2 x float> @llvm.round.v2f32(<2 x float>)
36declare <2 x float> @llvm.sqrt.v2f32(<2 x float>)
37declare <2 x float> @llvm.trunc.v2f32(<2 x float>)
38
39define <1 x float> @sin_v1f32(<1 x float> %x) nounwind {
40; CHECK-LABEL: sin_v1f32:
41; CHECK:       # %bb.0:
42; CHECK-NEXT:    pushq %rax
43; CHECK-NEXT:    callq sinf@PLT
44; CHECK-NEXT:    popq %rax
45; CHECK-NEXT:    retq
46  %r = call <1 x float> @llvm.sin.v1f32(<1 x float> %x)
47  ret <1 x float> %r
48}
49
50define <2 x float> @sin_v2f32(<2 x float> %x) nounwind {
51; CHECK-LABEL: sin_v2f32:
52; CHECK:       # %bb.0:
53; CHECK-NEXT:    subq $40, %rsp
54; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
55; CHECK-NEXT:    callq sinf@PLT
56; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
57; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
58; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
59; CHECK-NEXT:    callq sinf@PLT
60; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
61; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
62; CHECK-NEXT:    addq $40, %rsp
63; CHECK-NEXT:    retq
64  %r = call <2 x float> @llvm.sin.v2f32(<2 x float> %x)
65  ret <2 x float> %r
66}
67
68define <3 x float> @sin_v3f32(<3 x float> %x) nounwind {
69; CHECK-LABEL: sin_v3f32:
70; CHECK:       # %bb.0:
71; CHECK-NEXT:    subq $40, %rsp
72; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
73; CHECK-NEXT:    callq sinf@PLT
74; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
75; CHECK-NEXT:    vmovshdup {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
76; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
77; CHECK-NEXT:    callq sinf@PLT
78; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
79; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
80; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
81; CHECK-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
82; CHECK-NEXT:    # xmm0 = mem[1,0]
83; CHECK-NEXT:    callq sinf@PLT
84; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
85; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1],xmm0[0],xmm1[3]
86; CHECK-NEXT:    addq $40, %rsp
87; CHECK-NEXT:    retq
88  %r = call <3 x float> @llvm.sin.v3f32(<3 x float> %x)
89  ret <3 x float> %r
90}
91
92define <4 x float> @sin_v4f32(<4 x float> %x) nounwind {
93; CHECK-LABEL: sin_v4f32:
94; CHECK:       # %bb.0:
95; CHECK-NEXT:    subq $40, %rsp
96; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
97; CHECK-NEXT:    callq sinf@PLT
98; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
99; CHECK-NEXT:    vmovshdup {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
100; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
101; CHECK-NEXT:    callq sinf@PLT
102; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
103; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
104; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
105; CHECK-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
106; CHECK-NEXT:    # xmm0 = mem[1,0]
107; CHECK-NEXT:    callq sinf@PLT
108; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
109; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1],xmm0[0],xmm1[3]
110; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
111; CHECK-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
112; CHECK-NEXT:    # xmm0 = mem[3,3,3,3]
113; CHECK-NEXT:    callq sinf@PLT
114; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
115; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
116; CHECK-NEXT:    addq $40, %rsp
117; CHECK-NEXT:    retq
118  %r = call <4 x float> @llvm.sin.v4f32(<4 x float> %x)
119  ret <4 x float> %r
120}
121
122define <5 x float> @sin_v5f32(<5 x float> %x) nounwind {
123; CHECK-LABEL: sin_v5f32:
124; CHECK:       # %bb.0:
125; CHECK-NEXT:    subq $72, %rsp
126; CHECK-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
127; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
128; CHECK-NEXT:    vzeroupper
129; CHECK-NEXT:    callq sinf@PLT
130; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
131; CHECK-NEXT:    vmovshdup {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
132; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
133; CHECK-NEXT:    callq sinf@PLT
134; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
135; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
136; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
137; CHECK-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
138; CHECK-NEXT:    # xmm0 = mem[1,0]
139; CHECK-NEXT:    callq sinf@PLT
140; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
141; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1],xmm0[0],xmm1[3]
142; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
143; CHECK-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
144; CHECK-NEXT:    # xmm0 = mem[3,3,3,3]
145; CHECK-NEXT:    callq sinf@PLT
146; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
147; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
148; CHECK-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
149; CHECK-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
150; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm0
151; CHECK-NEXT:    vzeroupper
152; CHECK-NEXT:    callq sinf@PLT
153; CHECK-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
154; CHECK-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
155; CHECK-NEXT:    addq $72, %rsp
156; CHECK-NEXT:    retq
157  %r = call <5 x float> @llvm.sin.v5f32(<5 x float> %x)
158  ret <5 x float> %r
159}
160
161define <6 x float> @sin_v6f32(<6 x float> %x) nounwind {
162; CHECK-LABEL: sin_v6f32:
163; CHECK:       # %bb.0:
164; CHECK-NEXT:    subq $72, %rsp
165; CHECK-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
166; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm0
167; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
168; CHECK-NEXT:    vzeroupper
169; CHECK-NEXT:    callq sinf@PLT
170; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
171; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
172; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
173; CHECK-NEXT:    callq sinf@PLT
174; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
175; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
176; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
177; CHECK-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
178; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
179; CHECK-NEXT:    vzeroupper
180; CHECK-NEXT:    callq sinf@PLT
181; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
182; CHECK-NEXT:    vmovshdup {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
183; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
184; CHECK-NEXT:    callq sinf@PLT
185; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
186; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
187; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
188; CHECK-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
189; CHECK-NEXT:    # xmm0 = mem[1,0]
190; CHECK-NEXT:    callq sinf@PLT
191; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
192; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1],xmm0[0],xmm1[3]
193; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
194; CHECK-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
195; CHECK-NEXT:    # xmm0 = mem[3,3,3,3]
196; CHECK-NEXT:    callq sinf@PLT
197; CHECK-NEXT:    vmovaps (%rsp), %xmm1 # 16-byte Reload
198; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
199; CHECK-NEXT:    vinsertf128 $1, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 16-byte Folded Reload
200; CHECK-NEXT:    addq $72, %rsp
201; CHECK-NEXT:    retq
202  %r = call <6 x float> @llvm.sin.v6f32(<6 x float> %x)
203  ret <6 x float> %r
204}
205
206define <3 x double> @sin_v3f64(<3 x double> %x) nounwind {
207; CHECK-LABEL: sin_v3f64:
208; CHECK:       # %bb.0:
209; CHECK-NEXT:    subq $72, %rsp
210; CHECK-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
211; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
212; CHECK-NEXT:    vzeroupper
213; CHECK-NEXT:    callq sin@PLT
214; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
215; CHECK-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
216; CHECK-NEXT:    # xmm0 = mem[1,0]
217; CHECK-NEXT:    callq sin@PLT
218; CHECK-NEXT:    vmovapd (%rsp), %xmm1 # 16-byte Reload
219; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
220; CHECK-NEXT:    vmovupd %ymm0, (%rsp) # 32-byte Spill
221; CHECK-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
222; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm0
223; CHECK-NEXT:    vzeroupper
224; CHECK-NEXT:    callq sin@PLT
225; CHECK-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
226; CHECK-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
227; CHECK-NEXT:    addq $72, %rsp
228; CHECK-NEXT:    retq
229  %r = call <3 x double> @llvm.sin.v3f64(<3 x double> %x)
230  ret <3 x double> %r
231}
232
233define <2 x float> @fabs_v2f32(<2 x float> %x) nounwind {
234; CHECK-LABEL: fabs_v2f32:
235; CHECK:       # %bb.0:
236; CHECK-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
237; CHECK-NEXT:    retq
238  %r = call <2 x float> @llvm.fabs.v2f32(<2 x float> %x)
239  ret <2 x float> %r
240}
241
242define <2 x float> @ceil_v2f32(<2 x float> %x) nounwind {
243; CHECK-LABEL: ceil_v2f32:
244; CHECK:       # %bb.0:
245; CHECK-NEXT:    vroundps $10, %xmm0, %xmm0
246; CHECK-NEXT:    retq
247  %r = call <2 x float> @llvm.ceil.v2f32(<2 x float> %x)
248  ret <2 x float> %r
249}
250
251define <2 x float> @cos_v2f32(<2 x float> %x) nounwind {
252; CHECK-LABEL: cos_v2f32:
253; CHECK:       # %bb.0:
254; CHECK-NEXT:    subq $40, %rsp
255; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
256; CHECK-NEXT:    callq cosf@PLT
257; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
258; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
259; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
260; CHECK-NEXT:    callq cosf@PLT
261; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
262; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
263; CHECK-NEXT:    addq $40, %rsp
264; CHECK-NEXT:    retq
265  %r = call <2 x float> @llvm.cos.v2f32(<2 x float> %x)
266  ret <2 x float> %r
267}
268
269define <2 x float> @exp_v2f32(<2 x float> %x) nounwind {
270; CHECK-LABEL: exp_v2f32:
271; CHECK:       # %bb.0:
272; CHECK-NEXT:    subq $40, %rsp
273; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
274; CHECK-NEXT:    callq expf@PLT
275; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
276; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
277; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
278; CHECK-NEXT:    callq expf@PLT
279; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
280; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
281; CHECK-NEXT:    addq $40, %rsp
282; CHECK-NEXT:    retq
283  %r = call <2 x float> @llvm.exp.v2f32(<2 x float> %x)
284  ret <2 x float> %r
285}
286
287define <2 x float> @exp2_v2f32(<2 x float> %x) nounwind {
288; CHECK-LABEL: exp2_v2f32:
289; CHECK:       # %bb.0:
290; CHECK-NEXT:    subq $40, %rsp
291; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
292; CHECK-NEXT:    callq exp2f@PLT
293; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
294; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
295; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
296; CHECK-NEXT:    callq exp2f@PLT
297; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
298; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
299; CHECK-NEXT:    addq $40, %rsp
300; CHECK-NEXT:    retq
301  %r = call <2 x float> @llvm.exp2.v2f32(<2 x float> %x)
302  ret <2 x float> %r
303}
304
305define <2 x float> @floor_v2f32(<2 x float> %x) nounwind {
306; CHECK-LABEL: floor_v2f32:
307; CHECK:       # %bb.0:
308; CHECK-NEXT:    vroundps $9, %xmm0, %xmm0
309; CHECK-NEXT:    retq
310  %r = call <2 x float> @llvm.floor.v2f32(<2 x float> %x)
311  ret <2 x float> %r
312}
313
314define <2 x float> @log_v2f32(<2 x float> %x) nounwind {
315; CHECK-LABEL: log_v2f32:
316; CHECK:       # %bb.0:
317; CHECK-NEXT:    subq $40, %rsp
318; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
319; CHECK-NEXT:    callq logf@PLT
320; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
321; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
322; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
323; CHECK-NEXT:    callq logf@PLT
324; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
325; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
326; CHECK-NEXT:    addq $40, %rsp
327; CHECK-NEXT:    retq
328  %r = call <2 x float> @llvm.log.v2f32(<2 x float> %x)
329  ret <2 x float> %r
330}
331
332define <2 x float> @log10_v2f32(<2 x float> %x) nounwind {
333; CHECK-LABEL: log10_v2f32:
334; CHECK:       # %bb.0:
335; CHECK-NEXT:    subq $40, %rsp
336; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
337; CHECK-NEXT:    callq log10f@PLT
338; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
339; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
340; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
341; CHECK-NEXT:    callq log10f@PLT
342; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
343; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
344; CHECK-NEXT:    addq $40, %rsp
345; CHECK-NEXT:    retq
346  %r = call <2 x float> @llvm.log10.v2f32(<2 x float> %x)
347  ret <2 x float> %r
348}
349
350define <2 x float> @log2_v2f32(<2 x float> %x) nounwind {
351; CHECK-LABEL: log2_v2f32:
352; CHECK:       # %bb.0:
353; CHECK-NEXT:    subq $40, %rsp
354; CHECK-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
355; CHECK-NEXT:    callq log2f@PLT
356; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
357; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
358; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
359; CHECK-NEXT:    callq log2f@PLT
360; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
361; CHECK-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
362; CHECK-NEXT:    addq $40, %rsp
363; CHECK-NEXT:    retq
364  %r = call <2 x float> @llvm.log2.v2f32(<2 x float> %x)
365  ret <2 x float> %r
366}
367
368define <2 x float> @nearbyint__v2f32(<2 x float> %x) nounwind {
369; CHECK-LABEL: nearbyint__v2f32:
370; CHECK:       # %bb.0:
371; CHECK-NEXT:    vroundps $12, %xmm0, %xmm0
372; CHECK-NEXT:    retq
373  %r = call <2 x float> @llvm.nearbyint.v2f32(<2 x float> %x)
374  ret <2 x float> %r
375}
376
377define <2 x float> @rint_v2f32(<2 x float> %x) nounwind {
378; CHECK-LABEL: rint_v2f32:
379; CHECK:       # %bb.0:
380; CHECK-NEXT:    vroundps $4, %xmm0, %xmm0
381; CHECK-NEXT:    retq
382  %r = call <2 x float> @llvm.rint.v2f32(<2 x float> %x)
383  ret <2 x float> %r
384}
385
386define <2 x float> @round_v2f32(<2 x float> %x) nounwind {
387; CHECK-LABEL: round_v2f32:
388; CHECK:       # %bb.0:
389; CHECK-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
390; CHECK-NEXT:    vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
391; CHECK-NEXT:    vaddps %xmm1, %xmm0, %xmm0
392; CHECK-NEXT:    vroundps $11, %xmm0, %xmm0
393; CHECK-NEXT:    retq
394  %r = call <2 x float> @llvm.round.v2f32(<2 x float> %x)
395  ret <2 x float> %r
396}
397
398define <2 x float> @sqrt_v2f32(<2 x float> %x) nounwind {
399; CHECK-LABEL: sqrt_v2f32:
400; CHECK:       # %bb.0:
401; CHECK-NEXT:    vsqrtps %xmm0, %xmm0
402; CHECK-NEXT:    retq
403  %r = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %x)
404  ret <2 x float> %r
405}
406
407define <2 x float> @trunc_v2f32(<2 x float> %x) nounwind {
408; CHECK-LABEL: trunc_v2f32:
409; CHECK:       # %bb.0:
410; CHECK-NEXT:    vroundps $11, %xmm0, %xmm0
411; CHECK-NEXT:    retq
412  %r = call <2 x float> @llvm.trunc.v2f32(<2 x float> %x)
413  ret <2 x float> %r
414}
415
416