1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE41
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
5
6define <4 x float> @shuffle_v4f32_0z27(<4 x float> %x, <4 x float> %a) {
7; SSE-LABEL: shuffle_v4f32_0z27:
8; SSE:       # BB#0:
9; SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0],zero,xmm0[2],xmm1[2]
10; SSE-NEXT:    retq
11;
12; AVX-LABEL: shuffle_v4f32_0z27:
13; AVX:       # BB#0:
14; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],zero,xmm0[2],xmm1[2]
15; AVX-NEXT:    retq
16  %vecext = extractelement <4 x float> %x, i32 0
17  %vecinit = insertelement <4 x float> undef, float %vecext, i32 0
18  %vecinit1 = insertelement <4 x float> %vecinit, float 0.0, i32 1
19  %vecinit3 = shufflevector <4 x float> %vecinit1, <4 x float> %x, <4 x i32> <i32 0, i32 1, i32 6, i32 undef>
20  %vecinit5 = shufflevector <4 x float> %vecinit3, <4 x float> %a, <4 x i32> <i32 0, i32 1, i32 2, i32 6>
21  ret <4 x float> %vecinit5
22}
23
24define <4 x float> @shuffle_v4f32_0zz4(<4 x float> %xyzw, <4 x float> %abcd) {
25; SSE-LABEL: shuffle_v4f32_0zz4:
26; SSE:       # BB#0:
27; SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0],zero,zero,xmm1[0]
28; SSE-NEXT:    retq
29;
30; AVX-LABEL: shuffle_v4f32_0zz4:
31; AVX:       # BB#0:
32; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],zero,zero,xmm1[0]
33; AVX-NEXT:    retq
34  %vecext = extractelement <4 x float> %xyzw, i32 0
35  %vecinit = insertelement <4 x float> undef, float %vecext, i32 0
36  %vecinit1 = insertelement <4 x float> %vecinit, float 0.000000e+00, i32 1
37  %vecinit2 = insertelement <4 x float> %vecinit1, float 0.000000e+00, i32 2
38  %vecinit4 = shufflevector <4 x float> %vecinit2, <4 x float> %abcd, <4 x i32> <i32 0, i32 1, i32 2, i32 4>
39  ret <4 x float> %vecinit4
40}
41
42define <4 x float> @shuffle_v4f32_0z24(<4 x float> %xyzw, <4 x float> %abcd) {
43; SSE-LABEL: shuffle_v4f32_0z24:
44; SSE:       # BB#0:
45; SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0],zero,xmm0[2],xmm1[0]
46; SSE-NEXT:    retq
47;
48; AVX-LABEL: shuffle_v4f32_0z24:
49; AVX:       # BB#0:
50; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],zero,xmm0[2],xmm1[0]
51; AVX-NEXT:    retq
52  %vecext = extractelement <4 x float> %xyzw, i32 0
53  %vecinit = insertelement <4 x float> undef, float %vecext, i32 0
54  %vecinit1 = insertelement <4 x float> %vecinit, float 0.000000e+00, i32 1
55  %vecinit3 = shufflevector <4 x float> %vecinit1, <4 x float> %xyzw, <4 x i32> <i32 0, i32 1, i32 6, i32 undef>
56  %vecinit5 = shufflevector <4 x float> %vecinit3, <4 x float> %abcd, <4 x i32> <i32 0, i32 1, i32 2, i32 4>
57  ret <4 x float> %vecinit5
58}
59
60define <4 x float> @shuffle_v4f32_0zz0(float %a) {
61; SSE-LABEL: shuffle_v4f32_0zz0:
62; SSE:       # BB#0:
63; SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0],zero,zero,xmm0[0]
64; SSE-NEXT:    retq
65;
66; AVX-LABEL: shuffle_v4f32_0zz0:
67; AVX:       # BB#0:
68; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],zero,zero,xmm0[0]
69; AVX-NEXT:    retq
70  %vecinit = insertelement <4 x float> undef, float %a, i32 0
71  %vecinit1 = insertelement <4 x float> %vecinit, float 0.000000e+00, i32 1
72  %vecinit2 = insertelement <4 x float> %vecinit1, float 0.000000e+00, i32 2
73  %vecinit3 = insertelement <4 x float> %vecinit2, float %a, i32 3
74  ret <4 x float> %vecinit3
75}
76
77define <4 x float> @shuffle_v4f32_0z6z(<4 x float> %A, <4 x float> %B) {
78; SSE-LABEL: shuffle_v4f32_0z6z:
79; SSE:       # BB#0:
80; SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0],zero,xmm1[2],zero
81; SSE-NEXT:    retq
82;
83; AVX-LABEL: shuffle_v4f32_0z6z:
84; AVX:       # BB#0:
85; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],zero,xmm1[2],zero
86; AVX-NEXT:    retq
87  %vecext = extractelement <4 x float> %A, i32 0
88  %vecinit = insertelement <4 x float> undef, float %vecext, i32 0
89  %vecinit1 = insertelement <4 x float> %vecinit, float 0.000000e+00, i32 1
90  %vecext2 = extractelement <4 x float> %B, i32 2
91  %vecinit3 = insertelement <4 x float> %vecinit1, float %vecext2, i32 2
92  %vecinit4 = insertelement <4 x float> %vecinit3, float 0.000000e+00, i32 3
93  ret <4 x float> %vecinit4
94}
95
96define <4 x float> @shuffle_v4f32_z06z(<4 x float> %a, <4 x float> %b) {
97; SSE-LABEL: shuffle_v4f32_z06z:
98; SSE:       # BB#0:
99; SSE-NEXT:    insertps {{.*#+}} xmm1 = zero,xmm0[0],xmm1[2],zero
100; SSE-NEXT:    movaps %xmm1, %xmm0
101; SSE-NEXT:    retq
102;
103; AVX-LABEL: shuffle_v4f32_z06z:
104; AVX:       # BB#0:
105; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = zero,xmm0[0],xmm1[2],zero
106; AVX-NEXT:    retq
107  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 undef, i32 0, i32 6, i32 undef>
108  %shuffle1 = shufflevector <4 x float> %shuffle, <4 x float> <float 0.000000e+00, float undef, float undef, float 0.000000e+00>, <4 x i32> <i32 4, i32 1, i32 2, i32 7>
109  ret <4 x float> %shuffle1
110}
111
112define <4 x float> @shuffle_v4f32_05zz(<4 x float> %a, <4 x float> %b) {
113; SSE-LABEL: shuffle_v4f32_05zz:
114; SSE:       # BB#0:
115; SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0],xmm1[1],zero,zero
116; SSE-NEXT:    retq
117;
118; AVX-LABEL: shuffle_v4f32_05zz:
119; AVX:       # BB#0:
120; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[1],zero,zero
121; AVX-NEXT:    retq
122  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 undef, i32 undef>
123  %shuffle1 = shufflevector <4 x float> %shuffle, <4 x float> <float undef, float undef, float 0.000000e+00, float 0.000000e+00>, <4 x i32> <i32 0, i32 1, i32 6, i32 7>
124  ret <4 x float> %shuffle1
125}
126
127define <4 x float> @insertps_undef_input0(<4 x float> %a0, <4 x float> %a1) {
128; SSE-LABEL: insertps_undef_input0:
129; SSE:       # BB#0:
130; SSE-NEXT:    insertps {{.*#+}} xmm0 = zero,xmm1[0],zero,zero
131; SSE-NEXT:    retq
132;
133; AVX-LABEL: insertps_undef_input0:
134; AVX:       # BB#0:
135; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = zero,xmm1[0],zero,zero
136; AVX-NEXT:    retq
137  %res0 = fadd <4 x float> %a0, <float 1.0, float 1.0, float 1.0, float 1.0>
138  %res1 = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %res0, <4 x float> %a1, i8 21)
139  %res2 = shufflevector <4 x float> %res1, <4 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
140  ret <4 x float> %res2
141}
142
143define <4 x float> @insertps_undef_input1(<4 x float> %a0, <4 x float> %a1) {
144; SSE-LABEL: insertps_undef_input1:
145; SSE:       # BB#0:
146; SSE-NEXT:    xorps %xmm1, %xmm1
147; SSE-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
148; SSE-NEXT:    retq
149;
150; AVX-LABEL: insertps_undef_input1:
151; AVX:       # BB#0:
152; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
153; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
154; AVX-NEXT:    retq
155  %res0 = fadd <4 x float> %a1, <float 1.0, float 1.0, float 1.0, float 1.0>
156  %res1 = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a0, <4 x float> %res0, i8 21)
157  %res2 = shufflevector <4 x float> %res1, <4 x float> zeroinitializer, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
158  ret <4 x float> %res2
159}
160
161define <4 x float> @insertps_zero_from_v2f64(<4 x float> %a0, <2 x double>* %a1) nounwind {
162; SSE-LABEL: insertps_zero_from_v2f64:
163; SSE:       # BB#0:
164; SSE-NEXT:    movapd (%rdi), %xmm1
165; SSE-NEXT:    addpd {{.*}}(%rip), %xmm1
166; SSE-NEXT:    insertps {{.*#+}} xmm0 = zero,xmm0[2,2,3]
167; SSE-NEXT:    movapd %xmm1, (%rdi)
168; SSE-NEXT:    retq
169;
170; AVX-LABEL: insertps_zero_from_v2f64:
171; AVX:       # BB#0:
172; AVX-NEXT:    vmovapd (%rdi), %xmm1
173; AVX-NEXT:    vaddpd {{.*}}(%rip), %xmm1, %xmm1
174; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = zero,xmm0[2,2,3]
175; AVX-NEXT:    vmovapd %xmm1, (%rdi)
176; AVX-NEXT:    retq
177  %1 = load <2 x double>, <2 x double>* %a1
178  %2 = bitcast <2 x double> <double 1.0, double 2.0> to <4 x float>
179  %3 = fadd <2 x double> %1, <double 1.0, double 2.0>
180  %4 = shufflevector <4 x float> %a0, <4 x float> %2, <4 x i32> <i32 6, i32 2, i32 2, i32 3>
181  store <2 x double> %3, <2 x double> *%a1
182  ret <4 x float> %4
183}
184
185define <4 x float> @insertps_zero_from_v2i64(<4 x float> %a0, <2 x i64>* %a1) nounwind {
186; SSE-LABEL: insertps_zero_from_v2i64:
187; SSE:       # BB#0:
188; SSE-NEXT:    movdqa (%rdi), %xmm1
189; SSE-NEXT:    paddq {{.*}}(%rip), %xmm1
190; SSE-NEXT:    insertps {{.*#+}} xmm0 = zero,xmm0[2,2,3]
191; SSE-NEXT:    movdqa %xmm1, (%rdi)
192; SSE-NEXT:    retq
193;
194; AVX-LABEL: insertps_zero_from_v2i64:
195; AVX:       # BB#0:
196; AVX-NEXT:    vmovdqa (%rdi), %xmm1
197; AVX-NEXT:    vpaddq {{.*}}(%rip), %xmm1, %xmm1
198; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = zero,xmm0[2,2,3]
199; AVX-NEXT:    vmovdqa %xmm1, (%rdi)
200; AVX-NEXT:    retq
201  %1 = load <2 x i64>, <2 x i64>* %a1
202  %2 = bitcast <2 x i64> <i64 1, i64 -2> to <4 x float>
203  %3 = add <2 x i64> %1, <i64 1, i64 -2>
204  %4 = shufflevector <4 x float> %a0, <4 x float> %2, <4 x i32> <i32 5, i32 2, i32 2, i32 3>
205  store <2 x i64> %3, <2 x i64> *%a1
206  ret <4 x float> %4
207}
208
209define <4 x float> @insertps_zero_from_v8i16(<4 x float> %a0, <8 x i16>* %a1) nounwind {
210; SSE-LABEL: insertps_zero_from_v8i16:
211; SSE:       # BB#0:
212; SSE-NEXT:    movdqa (%rdi), %xmm1
213; SSE-NEXT:    paddw {{.*}}(%rip), %xmm1
214; SSE-NEXT:    insertps {{.*#+}} xmm0 = zero,xmm0[2,2,3]
215; SSE-NEXT:    movdqa %xmm1, (%rdi)
216; SSE-NEXT:    retq
217;
218; AVX-LABEL: insertps_zero_from_v8i16:
219; AVX:       # BB#0:
220; AVX-NEXT:    vmovdqa (%rdi), %xmm1
221; AVX-NEXT:    vpaddw {{.*}}(%rip), %xmm1, %xmm1
222; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = zero,xmm0[2,2,3]
223; AVX-NEXT:    vmovdqa %xmm1, (%rdi)
224; AVX-NEXT:    retq
225  %1 = load <8 x i16>, <8 x i16>* %a1
226  %2 = bitcast <8 x i16> <i16 0, i16 0, i16 1, i16 1, i16 2, i16 2, i16 3, i16 3> to <4 x float>
227  %3 = add <8 x i16> %1, <i16 0, i16 0, i16 1, i16 1, i16 2, i16 2, i16 3, i16 3>
228  %4 = shufflevector <4 x float> %a0, <4 x float> %2, <4 x i32> <i32 4, i32 2, i32 2, i32 3>
229  store <8 x i16> %3, <8 x i16> *%a1
230  ret <4 x float> %4
231}
232
233define <4 x float> @consecutive_load_insertps_04zz(float* %p) {
234; SSE-LABEL: consecutive_load_insertps_04zz:
235; SSE:       # BB#0:
236; SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
237; SSE-NEXT:    retq
238;
239; AVX-LABEL: consecutive_load_insertps_04zz:
240; AVX:       # BB#0:
241; AVX-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
242; AVX-NEXT:    retq
243  %p0 = getelementptr inbounds float, float* %p, i64 1
244  %p1 = getelementptr inbounds float, float* %p, i64 2
245  %s0 = load float, float* %p0
246  %s1 = load float, float* %p1
247  %v0 = insertelement <4 x float> undef, float %s0, i32 0
248  %v1 = insertelement <4 x float> undef, float %s1, i32 0
249  %res = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %v0, <4 x float> %v1, i8 28)
250  ret <4 x float> %res
251}
252
253define float @extract_zero_insertps_z0z7(<4 x float> %a0, <4 x float> %a1) {
254; SSE-LABEL: extract_zero_insertps_z0z7:
255; SSE:       # BB#0:
256; SSE-NEXT:    xorps %xmm0, %xmm0
257; SSE-NEXT:    retq
258;
259; AVX-LABEL: extract_zero_insertps_z0z7:
260; AVX:       # BB#0:
261; AVX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
262; AVX-NEXT:    retq
263  %res = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a0, <4 x float> %a1, i8 21)
264  %ext = extractelement <4 x float> %res, i32 0
265  ret float %ext
266}
267
268define float @extract_lane_insertps_5123(<4 x float> %a0, <4 x float> *%p1) {
269; SSE-LABEL: extract_lane_insertps_5123:
270; SSE:       # BB#0:
271; SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
272; SSE-NEXT:    retq
273;
274; AVX-LABEL: extract_lane_insertps_5123:
275; AVX:       # BB#0:
276; AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
277; AVX-NEXT:    retq
278  %a1 = load <4 x float>, <4 x float> *%p1
279  %res = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a0, <4 x float> %a1, i8 128)
280  %ext = extractelement <4 x float> %res, i32 0
281  ret float %ext
282}
283
284declare <4 x float> @llvm.x86.sse41.insertps(<4 x float>, <4 x float>, i8) nounwind readnone
285