1; RUN: llc < %s -mtriple=aarch64-none-eabi | FileCheck %s
2
3; float16x4_t select_64(float16x4_t a, float16x4_t b, uint16x4_t c) { return vbsl_u16(c, a, b); }
4define <4 x half> @select_64(<4 x half> %a, <4 x half> %b, <4 x i16> %c) #0 {
5; CHECK-LABEL: select_64:
6; CHECK: bsl
7entry:
8  %0 = bitcast <4 x half> %a to <4 x i16>
9  %1 = bitcast <4 x half> %b to <4 x i16>
10  %vbsl3.i = and <4 x i16> %0, %c
11  %2 = xor <4 x i16> %c, <i16 -1, i16 -1, i16 -1, i16 -1>
12  %vbsl4.i = and <4 x i16> %1, %2
13  %vbsl5.i = or <4 x i16> %vbsl3.i, %vbsl4.i
14  %3 = bitcast <4 x i16> %vbsl5.i to <4 x half>
15  ret <4 x half> %3
16}
17
18; float16x8_t select_128(float16x8_t a, float16x8_t b, uint16x8_t c) { return vbslq_u16(c, a, b); }
19define <8 x half> @select_128(<8 x half> %a, <8 x half> %b, <8 x i16> %c) #0 {
20; CHECK-LABEL: select_128:
21; CHECK: bsl
22entry:
23  %0 = bitcast <8 x half> %a to <8 x i16>
24  %1 = bitcast <8 x half> %b to <8 x i16>
25  %vbsl3.i = and <8 x i16> %0, %c
26  %2 = xor <8 x i16> %c, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
27  %vbsl4.i = and <8 x i16> %1, %2
28  %vbsl5.i = or <8 x i16> %vbsl3.i, %vbsl4.i
29  %3 = bitcast <8 x i16> %vbsl5.i to <8 x half>
30  ret <8 x half> %3
31}
32
33; float16x4_t lane_64_64(float16x4_t a, float16x4_t b) {
34;  return vcopy_lane_s16(a, 1, b, 2);
35; }
36define <4 x half> @lane_64_64(<4 x half> %a, <4 x half> %b) #0 {
37; CHECK-LABEL: lane_64_64:
38; CHECK: mov v{{[0-9]+}}.h
39entry:
40  %0 = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 6, i32 2, i32 3>
41  ret <4 x half> %0
42}
43
44; float16x8_t lane_128_64(float16x8_t a, float16x4_t b) {
45;   return vcopyq_lane_s16(a, 1, b, 2);
46; }
47define <8 x half> @lane_128_64(<8 x half> %a, <4 x half> %b) #0 {
48; CHECK-LABEL: lane_128_64:
49; CHECK: mov v{{[0-9]+}}.h
50entry:
51  %0 = bitcast <4 x half> %b to <4 x i16>
52  %vget_lane = extractelement <4 x i16> %0, i32 2
53  %1 = bitcast <8 x half> %a to <8 x i16>
54  %vset_lane = insertelement <8 x i16> %1, i16 %vget_lane, i32 1
55  %2 = bitcast <8 x i16> %vset_lane to <8 x half>
56  ret <8 x half> %2
57}
58
59; float16x4_t lane_64_128(float16x4_t a, float16x8_t b) {
60;   return vcopy_laneq_s16(a, 3, b, 5);
61; }
62define <4 x half> @lane_64_128(<4 x half> %a, <8 x half> %b) #0 {
63; CHECK-LABEL: lane_64_128:
64; CHECK: mov v{{[0-9]+}}.h
65entry:
66  %0 = bitcast <8 x half> %b to <8 x i16>
67  %vgetq_lane = extractelement <8 x i16> %0, i32 5
68  %1 = bitcast <4 x half> %a to <4 x i16>
69  %vset_lane = insertelement <4 x i16> %1, i16 %vgetq_lane, i32 3
70  %2 = bitcast <4 x i16> %vset_lane to <4 x half>
71  ret <4 x half> %2
72}
73
74; float16x8_t lane_128_128(float16x8_t a, float16x8_t b) {
75;   return vcopyq_laneq_s16(a, 3, b, 5);
76; }
77define <8 x half> @lane_128_128(<8 x half> %a, <8 x half> %b) #0 {
78; CHECK-LABEL: lane_128_128:
79; CHECK: mov v{{[0-9]+}}.h
80entry:
81  %0 = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 13, i32 4, i32 5, i32 6, i32 7>
82  ret <8 x half> %0
83}
84
85; float16x4_t ext_64(float16x4_t a, float16x4_t b) {
86;   return vext_s16(a, b, 3);
87; }
88define <4 x half> @ext_64(<4 x half> %a, <4 x half> %b) #0 {
89; CHECK-LABEL: ext_64:
90; CHECK: ext
91entry:
92  %0 = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
93  ret <4 x half> %0
94}
95
96; float16x8_t ext_128(float16x8_t a, float16x8_t b) {
97;   return vextq_s16(a, b, 3);
98; }
99define <8 x half> @ext_128(<8 x half> %a, <8 x half> %b) #0 {
100; CHECK-LABEL: ext_128:
101; CHECK: ext
102entry:
103  %0 = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>
104  ret <8 x half> %0
105}
106
107; float16x4_t rev32_64(float16x4_t a) {
108;   return vrev32_s16(a);
109; }
110define <4 x half> @rev32_64(<4 x half> %a) #0 {
111entry:
112; CHECK-LABEL: rev32_64:
113; CHECK: rev32
114  %0 = shufflevector <4 x half> %a, <4 x half> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
115  ret <4 x half> %0
116}
117
118; float16x4_t rev64_64(float16x4_t a) {
119;   return vrev64_s16(a);
120; }
121define <4 x half> @rev64_64(<4 x half> %a) #0 {
122entry:
123; CHECK-LABEL: rev64_64:
124; CHECK: rev64
125  %0 = shufflevector <4 x half> %a, <4 x half> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
126  ret <4 x half> %0
127}
128
129; float16x8_t rev32_128(float16x8_t a) {
130;   return vrev32q_s16(a);
131; }
132define <8 x half> @rev32_128(<8 x half> %a) #0 {
133entry:
134; CHECK-LABEL: rev32_128:
135; CHECK: rev32
136  %0 = shufflevector <8 x half> %a, <8 x half> undef, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
137  ret <8 x half> %0
138}
139
140; float16x8_t rev64_128(float16x8_t a) {
141;   return vrev64q_s16(a);
142; }
143define <8 x half> @rev64_128(<8 x half> %a) #0 {
144entry:
145; CHECK-LABEL: rev64_128:
146; CHECK: rev64
147  %0 = shufflevector <8 x half> %a, <8 x half> undef, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
148  ret <8 x half> %0
149}
150
151; float16x4_t create_64(long long a) { return vcreate_f16(a); }
152define <4 x half> @create_64(i64 %a) #0 {
153; CHECK-LABEL: create_64:
154; CHECK: fmov
155entry:
156  %0 = bitcast i64 %a to <4 x half>
157  ret <4 x half> %0
158}
159
160; float16x4_t dup_64(__fp16 a) { return vdup_n_f16(a); }
161define <4 x half> @dup_64(half %a) #0 {
162; CHECK-LABEL: dup_64:
163; CHECK: dup
164entry:
165  %vecinit = insertelement <4 x half> undef, half %a, i32 0
166  %vecinit1 = insertelement <4 x half> %vecinit, half %a, i32 1
167  %vecinit2 = insertelement <4 x half> %vecinit1, half %a, i32 2
168  %vecinit3 = insertelement <4 x half> %vecinit2, half %a, i32 3
169  ret <4 x half> %vecinit3
170}
171
172; float16x8_t dup_128(__fp16 a) { return vdupq_n_f16(a); }
173define <8 x half> @dup_128(half %a) #0 {
174entry:
175; CHECK-LABEL: dup_128:
176; CHECK: dup
177  %vecinit = insertelement <8 x half> undef, half %a, i32 0
178  %vecinit1 = insertelement <8 x half> %vecinit, half %a, i32 1
179  %vecinit2 = insertelement <8 x half> %vecinit1, half %a, i32 2
180  %vecinit3 = insertelement <8 x half> %vecinit2, half %a, i32 3
181  %vecinit4 = insertelement <8 x half> %vecinit3, half %a, i32 4
182  %vecinit5 = insertelement <8 x half> %vecinit4, half %a, i32 5
183  %vecinit6 = insertelement <8 x half> %vecinit5, half %a, i32 6
184  %vecinit7 = insertelement <8 x half> %vecinit6, half %a, i32 7
185  ret <8 x half> %vecinit7
186}
187
188; float16x4_t dup_lane_64(float16x4_t a) { return vdup_lane_f16(a, 2); }
189define <4 x half> @dup_lane_64(<4 x half> %a) #0 {
190entry:
191; CHECK-LABEL: dup_lane_64:
192; CHECK: dup
193  %shuffle = shufflevector <4 x half> %a, <4 x half> undef, <4 x i32> <i32 2, i32 2, i32 2, i32 2>
194  ret <4 x half> %shuffle
195}
196
197; float16x8_t dup_lane_128(float16x4_t a) { return vdupq_lane_f16(a, 2); }
198define <8 x half> @dup_lane_128(<4 x half> %a) #0 {
199entry:
200; CHECK-LABEL: dup_lane_128:
201; CHECK: dup
202  %shuffle = shufflevector <4 x half> %a, <4 x half> undef, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
203  ret <8 x half> %shuffle
204}
205
206; float16x4_t dup_laneq_64(float16x8_t a) { return vdup_laneq_f16(a, 2); }
207define <4 x half> @dup_laneq_64(<8 x half> %a) #0 {
208entry:
209; CHECK-LABEL: dup_laneq_64:
210; CHECK: dup
211  %shuffle = shufflevector <8 x half> %a, <8 x half> undef, <4 x i32> <i32 2, i32 2, i32 2, i32 2>
212  ret <4 x half> %shuffle
213}
214
215; float16x8_t dup_laneq_128(float16x8_t a) { return vdupq_laneq_f16(a, 2); }
216define <8 x half> @dup_laneq_128(<8 x half> %a) #0 {
217entry:
218; CHECK-LABEL: dup_laneq_128:
219; CHECK: dup
220  %shuffle = shufflevector <8 x half> %a, <8 x half> undef, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
221  ret <8 x half> %shuffle
222}
223
224; float16x8_t vcombine(float16x4_t a, float16x4_t b) { return vcombine_f16(a, b); }
225define <8 x half> @vcombine(<4 x half> %a, <4 x half> %b) #0 {
226entry:
227; CHECK-LABEL: vcombine:
228; CHECK: mov v0.d[1], v1.d[0]
229  %shuffle.i = shufflevector <4 x half> %a, <4 x half> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
230  ret <8 x half> %shuffle.i
231}
232
233; float16x4_t get_high(float16x8_t a) { return vget_high_f16(a); }
234define <4 x half> @get_high(<8 x half> %a) #0 {
235; CHECK-LABEL: get_high:
236; CHECK: ext
237entry:
238  %shuffle.i = shufflevector <8 x half> %a, <8 x half> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
239  ret <4 x half> %shuffle.i
240}
241
242
243; float16x4_t get_low(float16x8_t a) { return vget_low_f16(a); }
244define <4 x half> @get_low(<8 x half> %a) #0 {
245; CHECK-LABEL: get_low:
246; CHECK-NOT: ext
247entry:
248  %shuffle.i = shufflevector <8 x half> %a, <8 x half> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
249  ret <4 x half> %shuffle.i
250}
251
252; float16x4_t set_lane_64(float16x4_t a, __fp16 b) { return vset_lane_f16(b, a, 2); }
253define <4 x half> @set_lane_64(<4 x half> %a, half %b) #0 {
254; CHECK-LABEL: set_lane_64:
255; CHECK: fmov
256; CHECK: mov v{{[0-9]+}}.h
257entry:
258  %0 = bitcast half %b to i16
259  %1 = bitcast <4 x half> %a to <4 x i16>
260  %vset_lane = insertelement <4 x i16> %1, i16 %0, i32 2
261  %2 = bitcast <4 x i16> %vset_lane to <4 x half>
262  ret <4 x half> %2
263}
264
265
266; float16x8_t set_lane_128(float16x8_t a, __fp16 b) { return vsetq_lane_f16(b, a, 2); }
267define <8 x half> @set_lane_128(<8 x half> %a, half %b) #0 {
268; CHECK-LABEL: set_lane_128:
269; CHECK: fmov
270; CHECK: mov v{{[0-9]+}}.h
271entry:
272  %0 = bitcast half %b to i16
273  %1 = bitcast <8 x half> %a to <8 x i16>
274  %vset_lane = insertelement <8 x i16> %1, i16 %0, i32 2
275  %2 = bitcast <8 x i16> %vset_lane to <8 x half>
276  ret <8 x half> %2
277}
278
279; __fp16 get_lane_64(float16x4_t a) { return vget_lane_f16(a, 2); }
280define half @get_lane_64(<4 x half> %a) #0 {
281; CHECK-LABEL: get_lane_64:
282; CHECK: umov
283; CHECK: fmov
284entry:
285  %0 = bitcast <4 x half> %a to <4 x i16>
286  %vget_lane = extractelement <4 x i16> %0, i32 2
287  %1 = bitcast i16 %vget_lane to half
288  ret half %1
289}
290
291; __fp16 get_lane_128(float16x8_t a) { return vgetq_lane_f16(a, 2); }
292define half @get_lane_128(<8 x half> %a) #0 {
293; CHECK-LABEL: get_lane_128:
294; CHECK: umov
295; CHECK: fmov
296entry:
297  %0 = bitcast <8 x half> %a to <8 x i16>
298  %vgetq_lane = extractelement <8 x i16> %0, i32 2
299  %1 = bitcast i16 %vgetq_lane to half
300  ret half %1
301}
302