1 | // RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a\ |
2 | // RUN: -fallow-half-arguments-and-returns -S -disable-O0-optnone -emit-llvm -o - %s \ |
3 | // RUN: | opt -S -mem2reg \ |
4 | // RUN: | FileCheck %s |
5 | |
6 | // REQUIRES: aarch64-registered-target |
7 | |
8 | #include <arm_neon.h> |
9 | |
10 | // CHECK-LABEL: test_vabs_f16 |
11 | // CHECK: [[ABS:%.*]] = call <4 x half> @llvm.fabs.v4f16(<4 x half> %a) |
12 | // CHECK: ret <4 x half> [[ABS]] |
13 | float16x4_t test_vabs_f16(float16x4_t a) { |
14 | return vabs_f16(a); |
15 | } |
16 | |
17 | // CHECK-LABEL: test_vabsq_f16 |
18 | // CHECK: [[ABS:%.*]] = call <8 x half> @llvm.fabs.v8f16(<8 x half> %a) |
19 | // CHECK: ret <8 x half> [[ABS]] |
20 | float16x8_t test_vabsq_f16(float16x8_t a) { |
21 | return vabsq_f16(a); |
22 | } |
23 | |
24 | // CHECK-LABEL: test_vceqz_f16 |
25 | // CHECK: [[TMP1:%.*]] = fcmp oeq <4 x half> %a, zeroinitializer |
26 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
27 | // CHECK: ret <4 x i16> [[TMP2]] |
28 | uint16x4_t test_vceqz_f16(float16x4_t a) { |
29 | return vceqz_f16(a); |
30 | } |
31 | |
32 | // CHECK-LABEL: test_vceqzq_f16 |
33 | // CHECK: [[TMP1:%.*]] = fcmp oeq <8 x half> %a, zeroinitializer |
34 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
35 | // CHECK: ret <8 x i16> [[TMP2]] |
36 | uint16x8_t test_vceqzq_f16(float16x8_t a) { |
37 | return vceqzq_f16(a); |
38 | } |
39 | |
40 | // CHECK-LABEL: test_vcgez_f16 |
41 | // CHECK: [[TMP1:%.*]] = fcmp oge <4 x half> %a, zeroinitializer |
42 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
43 | // CHECK: ret <4 x i16> [[TMP2]] |
44 | uint16x4_t test_vcgez_f16(float16x4_t a) { |
45 | return vcgez_f16(a); |
46 | } |
47 | |
48 | // CHECK-LABEL: test_vcgezq_f16 |
49 | // CHECK: [[TMP1:%.*]] = fcmp oge <8 x half> %a, zeroinitializer |
50 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
51 | // CHECK: ret <8 x i16> [[TMP2]] |
52 | uint16x8_t test_vcgezq_f16(float16x8_t a) { |
53 | return vcgezq_f16(a); |
54 | } |
55 | |
56 | // CHECK-LABEL: test_vcgtz_f16 |
57 | // CHECK: [[TMP1:%.*]] = fcmp ogt <4 x half> %a, zeroinitializer |
58 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
59 | // CHECK: ret <4 x i16> [[TMP2]] |
60 | uint16x4_t test_vcgtz_f16(float16x4_t a) { |
61 | return vcgtz_f16(a); |
62 | } |
63 | |
64 | // CHECK-LABEL: test_vcgtzq_f16 |
65 | // CHECK: [[TMP1:%.*]] = fcmp ogt <8 x half> %a, zeroinitializer |
66 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
67 | // CHECK: ret <8 x i16> [[TMP2]] |
68 | uint16x8_t test_vcgtzq_f16(float16x8_t a) { |
69 | return vcgtzq_f16(a); |
70 | } |
71 | |
72 | // CHECK-LABEL: test_vclez_f16 |
73 | // CHECK: [[TMP1:%.*]] = fcmp ole <4 x half> %a, zeroinitializer |
74 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
75 | // CHECK: ret <4 x i16> [[TMP2]] |
76 | uint16x4_t test_vclez_f16(float16x4_t a) { |
77 | return vclez_f16(a); |
78 | } |
79 | |
80 | // CHECK-LABEL: test_vclezq_f16 |
81 | // CHECK: [[TMP1:%.*]] = fcmp ole <8 x half> %a, zeroinitializer |
82 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
83 | // CHECK: ret <8 x i16> [[TMP2]] |
84 | uint16x8_t test_vclezq_f16(float16x8_t a) { |
85 | return vclezq_f16(a); |
86 | } |
87 | |
88 | // CHECK-LABEL: test_vcltz_f16 |
89 | // CHECK: [[TMP1:%.*]] = fcmp olt <4 x half> %a, zeroinitializer |
90 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
91 | // CHECK: ret <4 x i16> [[TMP2]] |
92 | uint16x4_t test_vcltz_f16(float16x4_t a) { |
93 | return vcltz_f16(a); |
94 | } |
95 | |
96 | // CHECK-LABEL: test_vcltzq_f16 |
97 | // CHECK: [[TMP1:%.*]] = fcmp olt <8 x half> %a, zeroinitializer |
98 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
99 | // CHECK: ret <8 x i16> [[TMP2]] |
100 | uint16x8_t test_vcltzq_f16(float16x8_t a) { |
101 | return vcltzq_f16(a); |
102 | } |
103 | |
104 | // CHECK-LABEL: test_vcvt_f16_s16 |
105 | // CHECK: [[VCVT:%.*]] = sitofp <4 x i16> %a to <4 x half> |
106 | // CHECK: ret <4 x half> [[VCVT]] |
107 | float16x4_t test_vcvt_f16_s16 (int16x4_t a) { |
108 | return vcvt_f16_s16(a); |
109 | } |
110 | |
111 | // CHECK-LABEL: test_vcvtq_f16_s16 |
112 | // CHECK: [[VCVT:%.*]] = sitofp <8 x i16> %a to <8 x half> |
113 | // CHECK: ret <8 x half> [[VCVT]] |
114 | float16x8_t test_vcvtq_f16_s16 (int16x8_t a) { |
115 | return vcvtq_f16_s16(a); |
116 | } |
117 | |
118 | // CHECK-LABEL: test_vcvt_f16_u16 |
119 | // CHECK: [[VCVT:%.*]] = uitofp <4 x i16> %a to <4 x half> |
120 | // CHECK: ret <4 x half> [[VCVT]] |
121 | float16x4_t test_vcvt_f16_u16 (uint16x4_t a) { |
122 | return vcvt_f16_u16(a); |
123 | } |
124 | |
125 | // CHECK-LABEL: test_vcvtq_f16_u16 |
126 | // CHECK: [[VCVT:%.*]] = uitofp <8 x i16> %a to <8 x half> |
127 | // CHECK: ret <8 x half> [[VCVT]] |
128 | float16x8_t test_vcvtq_f16_u16 (uint16x8_t a) { |
129 | return vcvtq_f16_u16(a); |
130 | } |
131 | |
132 | // CHECK-LABEL: test_vcvt_s16_f16 |
133 | // CHECK: [[VCVT:%.*]] = fptosi <4 x half> %a to <4 x i16> |
134 | // CHECK: ret <4 x i16> [[VCVT]] |
135 | int16x4_t test_vcvt_s16_f16 (float16x4_t a) { |
136 | return vcvt_s16_f16(a); |
137 | } |
138 | |
139 | // CHECK-LABEL: test_vcvtq_s16_f16 |
140 | // CHECK: [[VCVT:%.*]] = fptosi <8 x half> %a to <8 x i16> |
141 | // CHECK: ret <8 x i16> [[VCVT]] |
142 | int16x8_t test_vcvtq_s16_f16 (float16x8_t a) { |
143 | return vcvtq_s16_f16(a); |
144 | } |
145 | |
146 | // CHECK-LABEL: test_vcvt_u16_f16 |
147 | // CHECK: [[VCVT:%.*]] = fptoui <4 x half> %a to <4 x i16> |
148 | // CHECK: ret <4 x i16> [[VCVT]] |
149 | int16x4_t test_vcvt_u16_f16 (float16x4_t a) { |
150 | return vcvt_u16_f16(a); |
151 | } |
152 | |
153 | // CHECK-LABEL: test_vcvtq_u16_f16 |
154 | // CHECK: [[VCVT:%.*]] = fptoui <8 x half> %a to <8 x i16> |
155 | // CHECK: ret <8 x i16> [[VCVT]] |
156 | int16x8_t test_vcvtq_u16_f16 (float16x8_t a) { |
157 | return vcvtq_u16_f16(a); |
158 | } |
159 | |
160 | // CHECK-LABEL: test_vcvta_s16_f16 |
161 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtas.v4i16.v4f16(<4 x half> %a) |
162 | // CHECK: ret <4 x i16> [[VCVT]] |
163 | int16x4_t test_vcvta_s16_f16 (float16x4_t a) { |
164 | return vcvta_s16_f16(a); |
165 | } |
166 | |
167 | // CHECK-LABEL: test_vcvta_u16_f16 |
168 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtau.v4i16.v4f16(<4 x half> %a) |
169 | // CHECK: ret <4 x i16> [[VCVT]] |
170 | int16x4_t test_vcvta_u16_f16 (float16x4_t a) { |
171 | return vcvta_u16_f16(a); |
172 | } |
173 | |
174 | // CHECK-LABEL: test_vcvtaq_s16_f16 |
175 | // CHECK: [[VCVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.fcvtas.v8i16.v8f16(<8 x half> %a) |
176 | // CHECK: ret <8 x i16> [[VCVT]] |
177 | int16x8_t test_vcvtaq_s16_f16 (float16x8_t a) { |
178 | return vcvtaq_s16_f16(a); |
179 | } |
180 | |
181 | // CHECK-LABEL: test_vcvtm_s16_f16 |
182 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtms.v4i16.v4f16(<4 x half> %a) |
183 | // CHECK: ret <4 x i16> [[VCVT]] |
184 | int16x4_t test_vcvtm_s16_f16 (float16x4_t a) { |
185 | return vcvtm_s16_f16(a); |
186 | } |
187 | |
188 | // CHECK-LABEL: test_vcvtmq_s16_f16 |
189 | // CHECK: [[VCVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.fcvtms.v8i16.v8f16(<8 x half> %a) |
190 | // CHECK: ret <8 x i16> [[VCVT]] |
191 | int16x8_t test_vcvtmq_s16_f16 (float16x8_t a) { |
192 | return vcvtmq_s16_f16(a); |
193 | } |
194 | |
195 | // CHECK-LABEL: test_vcvtm_u16_f16 |
196 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtmu.v4i16.v4f16(<4 x half> %a) |
197 | // CHECK: ret <4 x i16> [[VCVT]] |
198 | uint16x4_t test_vcvtm_u16_f16 (float16x4_t a) { |
199 | return vcvtm_u16_f16(a); |
200 | } |
201 | |
202 | // CHECK-LABEL: test_vcvtmq_u16_f16 |
203 | // CHECK: [[VCVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.fcvtmu.v8i16.v8f16(<8 x half> %a) |
204 | // CHECK: ret <8 x i16> [[VCVT]] |
205 | uint16x8_t test_vcvtmq_u16_f16 (float16x8_t a) { |
206 | return vcvtmq_u16_f16(a); |
207 | } |
208 | |
209 | // CHECK-LABEL: test_vcvtn_s16_f16 |
210 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtns.v4i16.v4f16(<4 x half> %a) |
211 | // CHECK: ret <4 x i16> [[VCVT]] |
212 | int16x4_t test_vcvtn_s16_f16 (float16x4_t a) { |
213 | return vcvtn_s16_f16(a); |
214 | } |
215 | |
216 | // CHECK-LABEL: test_vcvtnq_s16_f16 |
217 | // CHECK: [[VCVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.fcvtns.v8i16.v8f16(<8 x half> %a) |
218 | // CHECK: ret <8 x i16> [[VCVT]] |
219 | int16x8_t test_vcvtnq_s16_f16 (float16x8_t a) { |
220 | return vcvtnq_s16_f16(a); |
221 | } |
222 | |
223 | // CHECK-LABEL: test_vcvtn_u16_f16 |
224 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtnu.v4i16.v4f16(<4 x half> %a) |
225 | // CHECK: ret <4 x i16> [[VCVT]] |
226 | uint16x4_t test_vcvtn_u16_f16 (float16x4_t a) { |
227 | return vcvtn_u16_f16(a); |
228 | } |
229 | |
230 | // CHECK-LABEL: test_vcvtnq_u16_f16 |
231 | // CHECK: [[VCVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.fcvtnu.v8i16.v8f16(<8 x half> %a) |
232 | // CHECK: ret <8 x i16> [[VCVT]] |
233 | uint16x8_t test_vcvtnq_u16_f16 (float16x8_t a) { |
234 | return vcvtnq_u16_f16(a); |
235 | } |
236 | |
237 | // CHECK-LABEL: test_vcvtp_s16_f16 |
238 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtps.v4i16.v4f16(<4 x half> %a) |
239 | // CHECK: ret <4 x i16> [[VCVT]] |
240 | int16x4_t test_vcvtp_s16_f16 (float16x4_t a) { |
241 | return vcvtp_s16_f16(a); |
242 | } |
243 | |
244 | // CHECK-LABEL: test_vcvtpq_s16_f16 |
245 | // CHECK: [[VCVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.fcvtps.v8i16.v8f16(<8 x half> %a) |
246 | // CHECK: ret <8 x i16> [[VCVT]] |
247 | int16x8_t test_vcvtpq_s16_f16 (float16x8_t a) { |
248 | return vcvtpq_s16_f16(a); |
249 | } |
250 | |
251 | // CHECK-LABEL: test_vcvtp_u16_f16 |
252 | // CHECK: [[VCVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.fcvtpu.v4i16.v4f16(<4 x half> %a) |
253 | // CHECK: ret <4 x i16> [[VCVT]] |
254 | uint16x4_t test_vcvtp_u16_f16 (float16x4_t a) { |
255 | return vcvtp_u16_f16(a); |
256 | } |
257 | |
258 | // CHECK-LABEL: test_vcvtpq_u16_f16 |
259 | // CHECK: [[VCVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.fcvtpu.v8i16.v8f16(<8 x half> %a) |
260 | // CHECK: ret <8 x i16> [[VCVT]] |
261 | uint16x8_t test_vcvtpq_u16_f16 (float16x8_t a) { |
262 | return vcvtpq_u16_f16(a); |
263 | } |
264 | |
265 | // FIXME: Fix the zero constant when fp16 non-storage-only type becomes available. |
266 | // CHECK-LABEL: test_vneg_f16 |
267 | // CHECK: [[NEG:%.*]] = fsub <4 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %a |
268 | // CHECK: ret <4 x half> [[NEG]] |
269 | float16x4_t test_vneg_f16(float16x4_t a) { |
270 | return vneg_f16(a); |
271 | } |
272 | |
273 | // CHECK-LABEL: test_vnegq_f16 |
274 | // CHECK: [[NEG:%.*]] = fsub <8 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %a |
275 | // CHECK: ret <8 x half> [[NEG]] |
276 | float16x8_t test_vnegq_f16(float16x8_t a) { |
277 | return vnegq_f16(a); |
278 | } |
279 | |
280 | // CHECK-LABEL: test_vrecpe_f16 |
281 | // CHECK: [[RCP:%.*]] = call <4 x half> @llvm.aarch64.neon.frecpe.v4f16(<4 x half> %a) |
282 | // CHECK: ret <4 x half> [[RCP]] |
283 | float16x4_t test_vrecpe_f16(float16x4_t a) { |
284 | return vrecpe_f16(a); |
285 | } |
286 | |
287 | // CHECK-LABEL: test_vrecpeq_f16 |
288 | // CHECK: [[RCP:%.*]] = call <8 x half> @llvm.aarch64.neon.frecpe.v8f16(<8 x half> %a) |
289 | // CHECK: ret <8 x half> [[RCP]] |
290 | float16x8_t test_vrecpeq_f16(float16x8_t a) { |
291 | return vrecpeq_f16(a); |
292 | } |
293 | |
294 | // CHECK-LABEL: test_vrnd_f16 |
295 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.trunc.v4f16(<4 x half> %a) |
296 | // CHECK: ret <4 x half> [[RND]] |
297 | float16x4_t test_vrnd_f16(float16x4_t a) { |
298 | return vrnd_f16(a); |
299 | } |
300 | |
301 | // CHECK-LABEL: test_vrndq_f16 |
302 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.trunc.v8f16(<8 x half> %a) |
303 | // CHECK: ret <8 x half> [[RND]] |
304 | float16x8_t test_vrndq_f16(float16x8_t a) { |
305 | return vrndq_f16(a); |
306 | } |
307 | |
308 | // CHECK-LABEL: test_vrnda_f16 |
309 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.round.v4f16(<4 x half> %a) |
310 | // CHECK: ret <4 x half> [[RND]] |
311 | float16x4_t test_vrnda_f16(float16x4_t a) { |
312 | return vrnda_f16(a); |
313 | } |
314 | |
315 | // CHECK-LABEL: test_vrndaq_f16 |
316 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.round.v8f16(<8 x half> %a) |
317 | // CHECK: ret <8 x half> [[RND]] |
318 | float16x8_t test_vrndaq_f16(float16x8_t a) { |
319 | return vrndaq_f16(a); |
320 | } |
321 | |
322 | // CHECK-LABEL: test_vrndi_f16 |
323 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.nearbyint.v4f16(<4 x half> %a) |
324 | // CHECK: ret <4 x half> [[RND]] |
325 | float16x4_t test_vrndi_f16(float16x4_t a) { |
326 | return vrndi_f16(a); |
327 | } |
328 | |
329 | // CHECK-LABEL: test_vrndiq_f16 |
330 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.nearbyint.v8f16(<8 x half> %a) |
331 | // CHECK: ret <8 x half> [[RND]] |
332 | float16x8_t test_vrndiq_f16(float16x8_t a) { |
333 | return vrndiq_f16(a); |
334 | } |
335 | |
336 | // CHECK-LABEL: test_vrndm_f16 |
337 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.floor.v4f16(<4 x half> %a) |
338 | // CHECK: ret <4 x half> [[RND]] |
339 | float16x4_t test_vrndm_f16(float16x4_t a) { |
340 | return vrndm_f16(a); |
341 | } |
342 | |
343 | // CHECK-LABEL: test_vrndmq_f16 |
344 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.floor.v8f16(<8 x half> %a) |
345 | // CHECK: ret <8 x half> [[RND]] |
346 | float16x8_t test_vrndmq_f16(float16x8_t a) { |
347 | return vrndmq_f16(a); |
348 | } |
349 | |
350 | // CHECK-LABEL: test_vrndn_f16 |
351 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.aarch64.neon.frintn.v4f16(<4 x half> %a) |
352 | // CHECK: ret <4 x half> [[RND]] |
353 | float16x4_t test_vrndn_f16(float16x4_t a) { |
354 | return vrndn_f16(a); |
355 | } |
356 | |
357 | // CHECK-LABEL: test_vrndnq_f16 |
358 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.aarch64.neon.frintn.v8f16(<8 x half> %a) |
359 | // CHECK: ret <8 x half> [[RND]] |
360 | float16x8_t test_vrndnq_f16(float16x8_t a) { |
361 | return vrndnq_f16(a); |
362 | } |
363 | |
364 | // CHECK-LABEL: test_vrndp_f16 |
365 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.ceil.v4f16(<4 x half> %a) |
366 | // CHECK: ret <4 x half> [[RND]] |
367 | float16x4_t test_vrndp_f16(float16x4_t a) { |
368 | return vrndp_f16(a); |
369 | } |
370 | |
371 | // CHECK-LABEL: test_vrndpq_f16 |
372 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.ceil.v8f16(<8 x half> %a) |
373 | // CHECK: ret <8 x half> [[RND]] |
374 | float16x8_t test_vrndpq_f16(float16x8_t a) { |
375 | return vrndpq_f16(a); |
376 | } |
377 | |
378 | // CHECK-LABEL: test_vrndx_f16 |
379 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.rint.v4f16(<4 x half> %a) |
380 | // CHECK: ret <4 x half> [[RND]] |
381 | float16x4_t test_vrndx_f16(float16x4_t a) { |
382 | return vrndx_f16(a); |
383 | } |
384 | |
385 | // CHECK-LABEL: test_vrndxq_f16 |
386 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.rint.v8f16(<8 x half> %a) |
387 | // CHECK: ret <8 x half> [[RND]] |
388 | float16x8_t test_vrndxq_f16(float16x8_t a) { |
389 | return vrndxq_f16(a); |
390 | } |
391 | |
392 | // CHECK-LABEL: test_vrsqrte_f16 |
393 | // CHECK: [[RND:%.*]] = call <4 x half> @llvm.aarch64.neon.frsqrte.v4f16(<4 x half> %a) |
394 | // CHECK: ret <4 x half> [[RND]] |
395 | float16x4_t test_vrsqrte_f16(float16x4_t a) { |
396 | return vrsqrte_f16(a); |
397 | } |
398 | |
399 | // CHECK-LABEL: test_vrsqrteq_f16 |
400 | // CHECK: [[RND:%.*]] = call <8 x half> @llvm.aarch64.neon.frsqrte.v8f16(<8 x half> %a) |
401 | // CHECK: ret <8 x half> [[RND]] |
402 | float16x8_t test_vrsqrteq_f16(float16x8_t a) { |
403 | return vrsqrteq_f16(a); |
404 | } |
405 | |
406 | // CHECK-LABEL: test_vsqrt_f16 |
407 | // CHECK: [[SQR:%.*]] = call <4 x half> @llvm.sqrt.v4f16(<4 x half> %a) |
408 | // CHECK: ret <4 x half> [[SQR]] |
409 | float16x4_t test_vsqrt_f16(float16x4_t a) { |
410 | return vsqrt_f16(a); |
411 | } |
412 | |
413 | // CHECK-LABEL: test_vsqrtq_f16 |
414 | // CHECK: [[SQR:%.*]] = call <8 x half> @llvm.sqrt.v8f16(<8 x half> %a) |
415 | // CHECK: ret <8 x half> [[SQR]] |
416 | float16x8_t test_vsqrtq_f16(float16x8_t a) { |
417 | return vsqrtq_f16(a); |
418 | } |
419 | |
420 | // CHECK-LABEL: test_vadd_f16 |
421 | // CHECK: [[ADD:%.*]] = fadd <4 x half> %a, %b |
422 | // CHECK: ret <4 x half> [[ADD]] |
423 | float16x4_t test_vadd_f16(float16x4_t a, float16x4_t b) { |
424 | return vadd_f16(a, b); |
425 | } |
426 | |
427 | // CHECK-LABEL: test_vaddq_f16 |
428 | // CHECK: [[ADD:%.*]] = fadd <8 x half> %a, %b |
429 | // CHECK: ret <8 x half> [[ADD]] |
430 | float16x8_t test_vaddq_f16(float16x8_t a, float16x8_t b) { |
431 | return vaddq_f16(a, b); |
432 | } |
433 | |
434 | // CHECK-LABEL: test_vabd_f16 |
435 | // CHECK: [[ABD:%.*]] = call <4 x half> @llvm.aarch64.neon.fabd.v4f16(<4 x half> %a, <4 x half> %b) |
436 | // CHECK: ret <4 x half> [[ABD]] |
437 | float16x4_t test_vabd_f16(float16x4_t a, float16x4_t b) { |
438 | return vabd_f16(a, b); |
439 | } |
440 | |
441 | // CHECK-LABEL: test_vabdq_f16 |
442 | // CHECK: [[ABD:%.*]] = call <8 x half> @llvm.aarch64.neon.fabd.v8f16(<8 x half> %a, <8 x half> %b) |
443 | // CHECK: ret <8 x half> [[ABD]] |
444 | float16x8_t test_vabdq_f16(float16x8_t a, float16x8_t b) { |
445 | return vabdq_f16(a, b); |
446 | } |
447 | |
448 | // CHECK-LABEL: test_vcage_f16 |
449 | // CHECK: [[ABS:%.*]] = call <4 x i16> @llvm.aarch64.neon.facge.v4i16.v4f16(<4 x half> %a, <4 x half> %b) |
450 | // CHECK: ret <4 x i16> [[ABS]] |
451 | uint16x4_t test_vcage_f16(float16x4_t a, float16x4_t b) { |
452 | return vcage_f16(a, b); |
453 | } |
454 | |
455 | // CHECK-LABEL: test_vcageq_f16 |
456 | // CHECK: [[ABS:%.*]] = call <8 x i16> @llvm.aarch64.neon.facge.v8i16.v8f16(<8 x half> %a, <8 x half> %b) |
457 | // CHECK: ret <8 x i16> [[ABS]] |
458 | uint16x8_t test_vcageq_f16(float16x8_t a, float16x8_t b) { |
459 | return vcageq_f16(a, b); |
460 | } |
461 | |
462 | // CHECK-LABEL: test_vcagt_f16 |
463 | // CHECK: [[ABS:%.*]] = call <4 x i16> @llvm.aarch64.neon.facgt.v4i16.v4f16(<4 x half> %a, <4 x half> %b) |
464 | // CHECK: ret <4 x i16> [[ABS]] |
465 | uint16x4_t test_vcagt_f16(float16x4_t a, float16x4_t b) { |
466 | return vcagt_f16(a, b); |
467 | } |
468 | |
469 | // CHECK-LABEL: test_vcagtq_f16 |
470 | // CHECK: [[ABS:%.*]] = call <8 x i16> @llvm.aarch64.neon.facgt.v8i16.v8f16(<8 x half> %a, <8 x half> %b) |
471 | // CHECK: ret <8 x i16> [[ABS]] |
472 | uint16x8_t test_vcagtq_f16(float16x8_t a, float16x8_t b) { |
473 | return vcagtq_f16(a, b); |
474 | } |
475 | |
476 | // CHECK-LABEL: test_vcale_f16 |
477 | // CHECK: [[ABS:%.*]] = call <4 x i16> @llvm.aarch64.neon.facge.v4i16.v4f16(<4 x half> %b, <4 x half> %a) |
478 | // CHECK: ret <4 x i16> [[ABS]] |
479 | uint16x4_t test_vcale_f16(float16x4_t a, float16x4_t b) { |
480 | return vcale_f16(a, b); |
481 | } |
482 | |
483 | // CHECK-LABEL: test_vcaleq_f16 |
484 | // CHECK: [[ABS:%.*]] = call <8 x i16> @llvm.aarch64.neon.facge.v8i16.v8f16(<8 x half> %b, <8 x half> %a) |
485 | // CHECK: ret <8 x i16> [[ABS]] |
486 | uint16x8_t test_vcaleq_f16(float16x8_t a, float16x8_t b) { |
487 | return vcaleq_f16(a, b); |
488 | } |
489 | |
490 | // CHECK-LABEL: test_vcalt_f16 |
491 | // CHECK: [[ABS:%.*]] = call <4 x i16> @llvm.aarch64.neon.facgt.v4i16.v4f16(<4 x half> %b, <4 x half> %a) |
492 | // CHECK: ret <4 x i16> [[ABS]] |
493 | uint16x4_t test_vcalt_f16(float16x4_t a, float16x4_t b) { |
494 | return vcalt_f16(a, b); |
495 | } |
496 | |
497 | // CHECK-LABEL: test_vcaltq_f16 |
498 | // CHECK: [[ABS:%.*]] = call <8 x i16> @llvm.aarch64.neon.facgt.v8i16.v8f16(<8 x half> %b, <8 x half> %a) |
499 | // CHECK: ret <8 x i16> [[ABS]] |
500 | uint16x8_t test_vcaltq_f16(float16x8_t a, float16x8_t b) { |
501 | return vcaltq_f16(a, b); |
502 | } |
503 | |
504 | // CHECK-LABEL: test_vceq_f16 |
505 | // CHECK: [[TMP1:%.*]] = fcmp oeq <4 x half> %a, %b |
506 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
507 | // CHECK: ret <4 x i16> [[TMP2]] |
508 | uint16x4_t test_vceq_f16(float16x4_t a, float16x4_t b) { |
509 | return vceq_f16(a, b); |
510 | } |
511 | |
512 | // CHECK-LABEL: test_vceqq_f16 |
513 | // CHECK: [[TMP1:%.*]] = fcmp oeq <8 x half> %a, %b |
514 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
515 | // CHECK: ret <8 x i16> [[TMP2]] |
516 | uint16x8_t test_vceqq_f16(float16x8_t a, float16x8_t b) { |
517 | return vceqq_f16(a, b); |
518 | } |
519 | |
520 | // CHECK-LABEL: test_vcge_f16 |
521 | // CHECK: [[TMP1:%.*]] = fcmp oge <4 x half> %a, %b |
522 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
523 | // CHECK: ret <4 x i16> [[TMP2]] |
524 | uint16x4_t test_vcge_f16(float16x4_t a, float16x4_t b) { |
525 | return vcge_f16(a, b); |
526 | } |
527 | |
528 | // CHECK-LABEL: test_vcgeq_f16 |
529 | // CHECK: [[TMP1:%.*]] = fcmp oge <8 x half> %a, %b |
530 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
531 | // CHECK: ret <8 x i16> [[TMP2]] |
532 | uint16x8_t test_vcgeq_f16(float16x8_t a, float16x8_t b) { |
533 | return vcgeq_f16(a, b); |
534 | } |
535 | |
536 | // CHECK-LABEL: test_vcgt_f16 |
537 | // CHECK: [[TMP1:%.*]] = fcmp ogt <4 x half> %a, %b |
538 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
539 | // CHECK: ret <4 x i16> [[TMP2]] |
540 | uint16x4_t test_vcgt_f16(float16x4_t a, float16x4_t b) { |
541 | return vcgt_f16(a, b); |
542 | } |
543 | |
544 | // CHECK-LABEL: test_vcgtq_f16 |
545 | // CHECK: [[TMP1:%.*]] = fcmp ogt <8 x half> %a, %b |
546 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
547 | // CHECK: ret <8 x i16> [[TMP2]] |
548 | uint16x8_t test_vcgtq_f16(float16x8_t a, float16x8_t b) { |
549 | return vcgtq_f16(a, b); |
550 | } |
551 | |
552 | // CHECK-LABEL: test_vcle_f16 |
553 | // CHECK: [[TMP1:%.*]] = fcmp ole <4 x half> %a, %b |
554 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
555 | // CHECK: ret <4 x i16> [[TMP2]] |
556 | uint16x4_t test_vcle_f16(float16x4_t a, float16x4_t b) { |
557 | return vcle_f16(a, b); |
558 | } |
559 | |
560 | // CHECK-LABEL: test_vcleq_f16 |
561 | // CHECK: [[TMP1:%.*]] = fcmp ole <8 x half> %a, %b |
562 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
563 | // CHECK: ret <8 x i16> [[TMP2]] |
564 | uint16x8_t test_vcleq_f16(float16x8_t a, float16x8_t b) { |
565 | return vcleq_f16(a, b); |
566 | } |
567 | |
568 | // CHECK-LABEL: test_vclt_f16 |
569 | // CHECK: [[TMP1:%.*]] = fcmp olt <4 x half> %a, %b |
570 | // CHECK: [[TMP2:%.*]] = sext <4 x i1> [[TMP1]] to <4 x i16> |
571 | // CHECK: ret <4 x i16> [[TMP2]] |
572 | uint16x4_t test_vclt_f16(float16x4_t a, float16x4_t b) { |
573 | return vclt_f16(a, b); |
574 | } |
575 | |
576 | // CHECK-LABEL: test_vcltq_f16 |
577 | // CHECK: [[TMP1:%.*]] = fcmp olt <8 x half> %a, %b |
578 | // CHECK: [[TMP2:%.*]] = sext <8 x i1> [[TMP1:%.*]] to <8 x i16> |
579 | // CHECK: ret <8 x i16> [[TMP2]] |
580 | uint16x8_t test_vcltq_f16(float16x8_t a, float16x8_t b) { |
581 | return vcltq_f16(a, b); |
582 | } |
583 | |
584 | // CHECK-LABEL: test_vcvt_n_f16_s16 |
585 | // CHECK: [[CVT:%.*]] = call <4 x half> @llvm.aarch64.neon.vcvtfxs2fp.v4f16.v4i16(<4 x i16> %vcvt_n, i32 2) |
586 | // CHECK: ret <4 x half> [[CVT]] |
587 | float16x4_t test_vcvt_n_f16_s16(int16x4_t a) { |
588 | return vcvt_n_f16_s16(a, 2); |
589 | } |
590 | |
591 | // CHECK-LABEL: test_vcvtq_n_f16_s16 |
592 | // CHECK: [[CVT:%.*]] = call <8 x half> @llvm.aarch64.neon.vcvtfxs2fp.v8f16.v8i16(<8 x i16> %vcvt_n, i32 2) |
593 | // CHECK: ret <8 x half> [[CVT]] |
594 | float16x8_t test_vcvtq_n_f16_s16(int16x8_t a) { |
595 | return vcvtq_n_f16_s16(a, 2); |
596 | } |
597 | |
598 | // CHECK-LABEL: test_vcvt_n_f16_u16 |
599 | // CHECK: [[CVT:%.*]] = call <4 x half> @llvm.aarch64.neon.vcvtfxu2fp.v4f16.v4i16(<4 x i16> %vcvt_n, i32 2) |
600 | // CHECK: ret <4 x half> [[CVT]] |
601 | float16x4_t test_vcvt_n_f16_u16(uint16x4_t a) { |
602 | return vcvt_n_f16_u16(a, 2); |
603 | } |
604 | |
605 | // CHECK-LABEL: test_vcvtq_n_f16_u16 |
606 | // CHECK: [[CVT:%.*]] = call <8 x half> @llvm.aarch64.neon.vcvtfxu2fp.v8f16.v8i16(<8 x i16> %vcvt_n, i32 2) |
607 | // CHECK: ret <8 x half> [[CVT]] |
608 | float16x8_t test_vcvtq_n_f16_u16(uint16x8_t a) { |
609 | return vcvtq_n_f16_u16(a, 2); |
610 | } |
611 | |
612 | // CHECK-LABEL: test_vcvt_n_s16_f16 |
613 | // CHECK: [[CVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.vcvtfp2fxs.v4i16.v4f16(<4 x half> %vcvt_n, i32 2) |
614 | // CHECK: ret <4 x i16> [[CVT]] |
615 | int16x4_t test_vcvt_n_s16_f16(float16x4_t a) { |
616 | return vcvt_n_s16_f16(a, 2); |
617 | } |
618 | |
619 | // CHECK-LABEL: test_vcvtq_n_s16_f16 |
620 | // CHECK: [[CVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.vcvtfp2fxs.v8i16.v8f16(<8 x half> %vcvt_n, i32 2) |
621 | // CHECK: ret <8 x i16> [[CVT]] |
622 | int16x8_t test_vcvtq_n_s16_f16(float16x8_t a) { |
623 | return vcvtq_n_s16_f16(a, 2); |
624 | } |
625 | |
626 | // CHECK-LABEL: test_vcvt_n_u16_f16 |
627 | // CHECK: [[CVT:%.*]] = call <4 x i16> @llvm.aarch64.neon.vcvtfp2fxu.v4i16.v4f16(<4 x half> %vcvt_n, i32 2) |
628 | // CHECK: ret <4 x i16> [[CVT]] |
629 | uint16x4_t test_vcvt_n_u16_f16(float16x4_t a) { |
630 | return vcvt_n_u16_f16(a, 2); |
631 | } |
632 | |
633 | // CHECK-LABEL: test_vcvtq_n_u16_f16 |
634 | // CHECK: [[CVT:%.*]] = call <8 x i16> @llvm.aarch64.neon.vcvtfp2fxu.v8i16.v8f16(<8 x half> %vcvt_n, i32 2) |
635 | // CHECK: ret <8 x i16> [[CVT]] |
636 | uint16x8_t test_vcvtq_n_u16_f16(float16x8_t a) { |
637 | return vcvtq_n_u16_f16(a, 2); |
638 | } |
639 | |
640 | // CHECK-LABEL: test_vdiv_f16 |
641 | // CHECK: [[DIV:%.*]] = fdiv <4 x half> %a, %b |
642 | // CHECK: ret <4 x half> [[DIV]] |
643 | float16x4_t test_vdiv_f16(float16x4_t a, float16x4_t b) { |
644 | return vdiv_f16(a, b); |
645 | } |
646 | |
647 | // CHECK-LABEL: test_vdivq_f16 |
648 | // CHECK: [[DIV:%.*]] = fdiv <8 x half> %a, %b |
649 | // CHECK: ret <8 x half> [[DIV]] |
650 | float16x8_t test_vdivq_f16(float16x8_t a, float16x8_t b) { |
651 | return vdivq_f16(a, b); |
652 | } |
653 | |
654 | // CHECK-LABEL: test_vmax_f16 |
655 | // CHECK: [[MAX:%.*]] = call <4 x half> @llvm.aarch64.neon.fmax.v4f16(<4 x half> %a, <4 x half> %b) |
656 | // CHECK: ret <4 x half> [[MAX]] |
657 | float16x4_t test_vmax_f16(float16x4_t a, float16x4_t b) { |
658 | return vmax_f16(a, b); |
659 | } |
660 | |
661 | // CHECK-LABEL: test_vmaxq_f16 |
662 | // CHECK: [[MAX:%.*]] = call <8 x half> @llvm.aarch64.neon.fmax.v8f16(<8 x half> %a, <8 x half> %b) |
663 | // CHECK: ret <8 x half> [[MAX]] |
664 | float16x8_t test_vmaxq_f16(float16x8_t a, float16x8_t b) { |
665 | return vmaxq_f16(a, b); |
666 | } |
667 | |
668 | // CHECK-LABEL: test_vmaxnm_f16 |
669 | // CHECK: [[MAX:%.*]] = call <4 x half> @llvm.aarch64.neon.fmaxnm.v4f16(<4 x half> %a, <4 x half> %b) |
670 | // CHECK: ret <4 x half> [[MAX]] |
671 | float16x4_t test_vmaxnm_f16(float16x4_t a, float16x4_t b) { |
672 | return vmaxnm_f16(a, b); |
673 | } |
674 | |
675 | // CHECK-LABEL: test_vmaxnmq_f16 |
676 | // CHECK: [[MAX:%.*]] = call <8 x half> @llvm.aarch64.neon.fmaxnm.v8f16(<8 x half> %a, <8 x half> %b) |
677 | // CHECK: ret <8 x half> [[MAX]] |
678 | float16x8_t test_vmaxnmq_f16(float16x8_t a, float16x8_t b) { |
679 | return vmaxnmq_f16(a, b); |
680 | } |
681 | |
682 | // CHECK-LABEL: test_vmin_f16 |
683 | // CHECK: [[MIN:%.*]] = call <4 x half> @llvm.aarch64.neon.fmin.v4f16(<4 x half> %a, <4 x half> %b) |
684 | // CHECK: ret <4 x half> [[MIN]] |
685 | float16x4_t test_vmin_f16(float16x4_t a, float16x4_t b) { |
686 | return vmin_f16(a, b); |
687 | } |
688 | |
689 | // CHECK-LABEL: test_vminq_f16 |
690 | // CHECK: [[MIN:%.*]] = call <8 x half> @llvm.aarch64.neon.fmin.v8f16(<8 x half> %a, <8 x half> %b) |
691 | // CHECK: ret <8 x half> [[MIN]] |
692 | float16x8_t test_vminq_f16(float16x8_t a, float16x8_t b) { |
693 | return vminq_f16(a, b); |
694 | } |
695 | |
696 | // CHECK-LABEL: test_vminnm_f16 |
697 | // CHECK: [[MIN:%.*]] = call <4 x half> @llvm.aarch64.neon.fminnm.v4f16(<4 x half> %a, <4 x half> %b) |
698 | // CHECK: ret <4 x half> [[MIN]] |
699 | float16x4_t test_vminnm_f16(float16x4_t a, float16x4_t b) { |
700 | return vminnm_f16(a, b); |
701 | } |
702 | |
703 | // CHECK-LABEL: test_vminnmq_f16 |
704 | // CHECK: [[MIN:%.*]] = call <8 x half> @llvm.aarch64.neon.fminnm.v8f16(<8 x half> %a, <8 x half> %b) |
705 | // CHECK: ret <8 x half> [[MIN]] |
706 | float16x8_t test_vminnmq_f16(float16x8_t a, float16x8_t b) { |
707 | return vminnmq_f16(a, b); |
708 | } |
709 | |
710 | // CHECK-LABEL: test_vmul_f16 |
711 | // CHECK: [[MUL:%.*]] = fmul <4 x half> %a, %b |
712 | // CHECK: ret <4 x half> [[MUL]] |
713 | float16x4_t test_vmul_f16(float16x4_t a, float16x4_t b) { |
714 | return vmul_f16(a, b); |
715 | } |
716 | |
717 | // CHECK-LABEL: test_vmulq_f16 |
718 | // CHECK: [[MUL:%.*]] = fmul <8 x half> %a, %b |
719 | // CHECK: ret <8 x half> [[MUL]] |
720 | float16x8_t test_vmulq_f16(float16x8_t a, float16x8_t b) { |
721 | return vmulq_f16(a, b); |
722 | } |
723 | |
724 | // CHECK-LABEL: test_vmulx_f16 |
725 | // CHECK: [[MUL:%.*]] = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> %a, <4 x half> %b) |
726 | // CHECK: ret <4 x half> [[MUL]] |
727 | float16x4_t test_vmulx_f16(float16x4_t a, float16x4_t b) { |
728 | return vmulx_f16(a, b); |
729 | } |
730 | |
731 | // CHECK-LABEL: test_vmulxq_f16 |
732 | // CHECK: [[MUL:%.*]] = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> %a, <8 x half> %b) |
733 | // CHECK: ret <8 x half> [[MUL]] |
734 | float16x8_t test_vmulxq_f16(float16x8_t a, float16x8_t b) { |
735 | return vmulxq_f16(a, b); |
736 | } |
737 | |
738 | // CHECK-LABEL: test_vpadd_f16 |
739 | // CHECK: [[ADD:%.*]] = call <4 x half> @llvm.aarch64.neon.faddp.v4f16(<4 x half> %a, <4 x half> %b) |
740 | // CHECK: ret <4 x half> [[ADD]] |
741 | float16x4_t test_vpadd_f16(float16x4_t a, float16x4_t b) { |
742 | return vpadd_f16(a, b); |
743 | } |
744 | |
745 | // CHECK-LABEL: test_vpaddq_f16 |
746 | // CHECK: [[ADD:%.*]] = call <8 x half> @llvm.aarch64.neon.faddp.v8f16(<8 x half> %a, <8 x half> %b) |
747 | // CHECK: ret <8 x half> [[ADD]] |
748 | float16x8_t test_vpaddq_f16(float16x8_t a, float16x8_t b) { |
749 | return vpaddq_f16(a, b); |
750 | } |
751 | |
752 | // CHECK-LABEL: test_vpmax_f16 |
753 | // CHECK: [[MAX:%.*]] = call <4 x half> @llvm.aarch64.neon.fmaxp.v4f16(<4 x half> %a, <4 x half> %b) |
754 | // CHECK: ret <4 x half> [[MAX]] |
755 | float16x4_t test_vpmax_f16(float16x4_t a, float16x4_t b) { |
756 | return vpmax_f16(a, b); |
757 | } |
758 | |
759 | // CHECK-LABEL: test_vpmaxq_f16 |
760 | // CHECK: [[MAX:%.*]] = call <8 x half> @llvm.aarch64.neon.fmaxp.v8f16(<8 x half> %a, <8 x half> %b) |
761 | // CHECK: ret <8 x half> [[MAX]] |
762 | float16x8_t test_vpmaxq_f16(float16x8_t a, float16x8_t b) { |
763 | return vpmaxq_f16(a, b); |
764 | } |
765 | |
766 | // CHECK-LABEL: test_vpmaxnm_f16 |
767 | // CHECK: [[MAX:%.*]] = call <4 x half> @llvm.aarch64.neon.fmaxnmp.v4f16(<4 x half> %a, <4 x half> %b) |
768 | // CHECK: ret <4 x half> [[MAX]] |
769 | float16x4_t test_vpmaxnm_f16(float16x4_t a, float16x4_t b) { |
770 | return vpmaxnm_f16(a, b); |
771 | } |
772 | |
773 | // CHECK-LABEL: test_vpmaxnmq_f16 |
774 | // CHECK: [[MAX:%.*]] = call <8 x half> @llvm.aarch64.neon.fmaxnmp.v8f16(<8 x half> %a, <8 x half> %b) |
775 | // CHECK: ret <8 x half> [[MAX]] |
776 | float16x8_t test_vpmaxnmq_f16(float16x8_t a, float16x8_t b) { |
777 | return vpmaxnmq_f16(a, b); |
778 | } |
779 | |
780 | // CHECK-LABEL: test_vpmin_f16 |
781 | // CHECK: [[MIN:%.*]] = call <4 x half> @llvm.aarch64.neon.fminp.v4f16(<4 x half> %a, <4 x half> %b) |
782 | // CHECK: ret <4 x half> [[MIN]] |
783 | float16x4_t test_vpmin_f16(float16x4_t a, float16x4_t b) { |
784 | return vpmin_f16(a, b); |
785 | } |
786 | |
787 | // CHECK-LABEL: test_vpminq_f16 |
788 | // CHECK: [[MIN:%.*]] = call <8 x half> @llvm.aarch64.neon.fminp.v8f16(<8 x half> %a, <8 x half> %b) |
789 | // CHECK: ret <8 x half> [[MIN]] |
790 | float16x8_t test_vpminq_f16(float16x8_t a, float16x8_t b) { |
791 | return vpminq_f16(a, b); |
792 | } |
793 | |
794 | // CHECK-LABEL: test_vpminnm_f16 |
795 | // CHECK: [[MIN:%.*]] = call <4 x half> @llvm.aarch64.neon.fminnmp.v4f16(<4 x half> %a, <4 x half> %b) |
796 | // CHECK: ret <4 x half> [[MIN]] |
797 | float16x4_t test_vpminnm_f16(float16x4_t a, float16x4_t b) { |
798 | return vpminnm_f16(a, b); |
799 | } |
800 | |
801 | // CHECK-LABEL: test_vpminnmq_f16 |
802 | // CHECK: [[MIN:%.*]] = call <8 x half> @llvm.aarch64.neon.fminnmp.v8f16(<8 x half> %a, <8 x half> %b) |
803 | // CHECK: ret <8 x half> [[MIN]] |
804 | float16x8_t test_vpminnmq_f16(float16x8_t a, float16x8_t b) { |
805 | return vpminnmq_f16(a, b); |
806 | } |
807 | |
808 | // CHECK-LABEL: test_vrecps_f16 |
809 | // CHECK: [[MIN:%.*]] = call <4 x half> @llvm.aarch64.neon.frecps.v4f16(<4 x half> %a, <4 x half> %b) |
810 | // CHECK: ret <4 x half> [[MIN]] |
811 | float16x4_t test_vrecps_f16(float16x4_t a, float16x4_t b) { |
812 | return vrecps_f16(a, b); |
813 | } |
814 | |
815 | // CHECK-LABEL: test_vrecpsq_f16 |
816 | // CHECK: [[MIN:%.*]] = call <8 x half> @llvm.aarch64.neon.frecps.v8f16(<8 x half> %a, <8 x half> %b) |
817 | // CHECK: ret <8 x half> [[MIN]] |
818 | float16x8_t test_vrecpsq_f16(float16x8_t a, float16x8_t b) { |
819 | return vrecpsq_f16(a, b); |
820 | } |
821 | |
822 | // CHECK-LABEL: test_vrsqrts_f16 |
823 | // CHECK: [[MIN:%.*]] = call <4 x half> @llvm.aarch64.neon.frsqrts.v4f16(<4 x half> %a, <4 x half> %b) |
824 | // CHECK: ret <4 x half> [[MIN]] |
825 | float16x4_t test_vrsqrts_f16(float16x4_t a, float16x4_t b) { |
826 | return vrsqrts_f16(a, b); |
827 | } |
828 | |
829 | // CHECK-LABEL: test_vrsqrtsq_f16 |
830 | // CHECK: [[MIN:%.*]] = call <8 x half> @llvm.aarch64.neon.frsqrts.v8f16(<8 x half> %a, <8 x half> %b) |
831 | // CHECK: ret <8 x half> [[MIN]] |
832 | float16x8_t test_vrsqrtsq_f16(float16x8_t a, float16x8_t b) { |
833 | return vrsqrtsq_f16(a, b); |
834 | } |
835 | |
836 | // CHECK-LABEL: test_vsub_f16 |
837 | // CHECK: [[ADD:%.*]] = fsub <4 x half> %a, %b |
838 | // CHECK: ret <4 x half> [[ADD]] |
839 | float16x4_t test_vsub_f16(float16x4_t a, float16x4_t b) { |
840 | return vsub_f16(a, b); |
841 | } |
842 | |
843 | // CHECK-LABEL: test_vsubq_f16 |
844 | // CHECK: [[ADD:%.*]] = fsub <8 x half> %a, %b |
845 | // CHECK: ret <8 x half> [[ADD]] |
846 | float16x8_t test_vsubq_f16(float16x8_t a, float16x8_t b) { |
847 | return vsubq_f16(a, b); |
848 | } |
849 | |
850 | // CHECK-LABEL: test_vfma_f16 |
851 | // CHECK: [[ADD:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> %b, <4 x half> %c, <4 x half> %a) |
852 | // CHECK: ret <4 x half> [[ADD]] |
853 | float16x4_t test_vfma_f16(float16x4_t a, float16x4_t b, float16x4_t c) { |
854 | return vfma_f16(a, b, c); |
855 | } |
856 | |
857 | // CHECK-LABEL: test_vfmaq_f16 |
858 | // CHECK: [[ADD:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> %b, <8 x half> %c, <8 x half> %a) |
859 | // CHECK: ret <8 x half> [[ADD]] |
860 | float16x8_t test_vfmaq_f16(float16x8_t a, float16x8_t b, float16x8_t c) { |
861 | return vfmaq_f16(a, b, c); |
862 | } |
863 | |
864 | // CHECK-LABEL: test_vfms_f16 |
865 | // CHECK: [[SUB:%.*]] = fsub <4 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
866 | // CHECK: [[ADD:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[SUB]], <4 x half> %c, <4 x half> %a) |
867 | // CHECK: ret <4 x half> [[ADD]] |
868 | float16x4_t test_vfms_f16(float16x4_t a, float16x4_t b, float16x4_t c) { |
869 | return vfms_f16(a, b, c); |
870 | } |
871 | |
872 | // CHECK-LABEL: test_vfmsq_f16 |
873 | // CHECK: [[SUB:%.*]] = fsub <8 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
874 | // CHECK: [[ADD:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[SUB]], <8 x half> %c, <8 x half> %a) |
875 | // CHECK: ret <8 x half> [[ADD]] |
876 | float16x8_t test_vfmsq_f16(float16x8_t a, float16x8_t b, float16x8_t c) { |
877 | return vfmsq_f16(a, b, c); |
878 | } |
879 | |
880 | // CHECK-LABEL: test_vfma_lane_f16 |
881 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
882 | // CHECK: [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8> |
883 | // CHECK: [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8> |
884 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half> |
885 | // CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <4 x i32> <i32 3, i32 3, i32 3, i32 3> |
886 | // CHECK: [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> |
887 | // CHECK: [[TMP5:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
888 | // CHECK: [[FMLA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]]) |
889 | // CHECK: ret <4 x half> [[FMLA]] |
890 | float16x4_t test_vfma_lane_f16(float16x4_t a, float16x4_t b, float16x4_t c) { |
891 | return vfma_lane_f16(a, b, c, 3); |
892 | } |
893 | |
894 | // CHECK-LABEL: test_vfmaq_lane_f16 |
895 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
896 | // CHECK: [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8> |
897 | // CHECK: [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8> |
898 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half> |
899 | // CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3> |
900 | // CHECK: [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> |
901 | // CHECK: [[TMP5:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
902 | // CHECK: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]]) |
903 | // CHECK: ret <8 x half> [[FMLA]] |
904 | float16x8_t test_vfmaq_lane_f16(float16x8_t a, float16x8_t b, float16x4_t c) { |
905 | return vfmaq_lane_f16(a, b, c, 3); |
906 | } |
907 | |
908 | // CHECK-LABEL: test_vfma_laneq_f16 |
909 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
910 | // CHECK: [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8> |
911 | // CHECK: [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8> |
912 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
913 | // CHECK: [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> |
914 | // CHECK: [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half> |
915 | // CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <4 x i32> <i32 7, i32 7, i32 7, i32 7> |
916 | // CHECK: [[FMLA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]]) |
917 | // CHECK: ret <4 x half> [[FMLA]] |
918 | float16x4_t test_vfma_laneq_f16(float16x4_t a, float16x4_t b, float16x8_t c) { |
919 | return vfma_laneq_f16(a, b, c, 7); |
920 | } |
921 | |
922 | // CHECK-LABEL: test_vfmaq_laneq_f16 |
923 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
924 | // CHECK: [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8> |
925 | // CHECK: [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8> |
926 | // CHECK: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
927 | // CHECK: [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> |
928 | // CHECK: [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half> |
929 | // CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7> |
930 | // CHECK: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]]) |
931 | // CHECK: ret <8 x half> [[FMLA]] |
932 | float16x8_t test_vfmaq_laneq_f16(float16x8_t a, float16x8_t b, float16x8_t c) { |
933 | return vfmaq_laneq_f16(a, b, c, 7); |
934 | } |
935 | |
936 | // CHECK-LABEL: test_vfma_n_f16 |
937 | // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %c, i32 0 |
938 | // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %c, i32 1 |
939 | // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %c, i32 2 |
940 | // CHECK: [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %c, i32 3 |
941 | // CHECK: [[FMA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> %b, <4 x half> [[TMP3]], <4 x half> %a) |
942 | // CHECK: ret <4 x half> [[FMA]] |
943 | float16x4_t test_vfma_n_f16(float16x4_t a, float16x4_t b, float16_t c) { |
944 | return vfma_n_f16(a, b, c); |
945 | } |
946 | |
947 | // CHECK-LABEL: test_vfmaq_n_f16 |
948 | // CHECK: [[TMP0:%.*]] = insertelement <8 x half> undef, half %c, i32 0 |
949 | // CHECK: [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %c, i32 1 |
950 | // CHECK: [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %c, i32 2 |
951 | // CHECK: [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %c, i32 3 |
952 | // CHECK: [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %c, i32 4 |
953 | // CHECK: [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %c, i32 5 |
954 | // CHECK: [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %c, i32 6 |
955 | // CHECK: [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %c, i32 7 |
956 | // CHECK: [[FMA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> %b, <8 x half> [[TMP7]], <8 x half> %a) |
957 | // CHECK: ret <8 x half> [[FMA]] |
958 | float16x8_t test_vfmaq_n_f16(float16x8_t a, float16x8_t b, float16_t c) { |
959 | return vfmaq_n_f16(a, b, c); |
960 | } |
961 | |
962 | // CHECK-LABEL: test_vfmah_lane_f16 |
963 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %c to <8 x i8> |
964 | // CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
965 | // CHECK: [[EXTR:%.*]] = extractelement <4 x half> [[TMP1]], i32 3 |
966 | // CHECK: [[FMA:%.*]] = call half @llvm.fma.f16(half %b, half [[EXTR]], half %a) |
967 | // CHECK: ret half [[FMA]] |
968 | float16_t test_vfmah_lane_f16(float16_t a, float16_t b, float16x4_t c) { |
969 | return vfmah_lane_f16(a, b, c, 3); |
970 | } |
971 | |
972 | // CHECK-LABEL: test_vfmah_laneq_f16 |
973 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %c to <16 x i8> |
974 | // CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
975 | // CHECK: [[EXTR:%.*]] = extractelement <8 x half> [[TMP1]], i32 7 |
976 | // CHECK: [[FMA:%.*]] = call half @llvm.fma.f16(half %b, half [[EXTR]], half %a) |
977 | // CHECK: ret half [[FMA]] |
978 | float16_t test_vfmah_laneq_f16(float16_t a, float16_t b, float16x8_t c) { |
979 | return vfmah_laneq_f16(a, b, c, 7); |
980 | } |
981 | |
982 | // CHECK-LABEL: test_vfms_lane_f16 |
983 | // CHECK: [[SUB:%.*]] = fsub <4 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
984 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
985 | // CHECK: [[TMP1:%.*]] = bitcast <4 x half> [[SUB]] to <8 x i8> |
986 | // CHECK: [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8> |
987 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half> |
988 | // CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <4 x i32> <i32 3, i32 3, i32 3, i32 3> |
989 | // CHECK: [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> |
990 | // CHECK: [[TMP5:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
991 | // CHECK: [[FMA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[TMP4]], <4 x half> [[LANE]], <4 x half> [[TMP5]]) |
992 | // CHECK: ret <4 x half> [[FMA]] |
993 | float16x4_t test_vfms_lane_f16(float16x4_t a, float16x4_t b, float16x4_t c) { |
994 | return vfms_lane_f16(a, b, c, 3); |
995 | } |
996 | |
997 | // CHECK-LABEL: test_vfmsq_lane_f16 |
998 | // CHECK: [[SUB:%.*]] = fsub <8 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
999 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
1000 | // CHECK: [[TMP1:%.*]] = bitcast <8 x half> [[SUB]] to <16 x i8> |
1001 | // CHECK: [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8> |
1002 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half> |
1003 | // CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3> |
1004 | // CHECK: [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> |
1005 | // CHECK: [[TMP5:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1006 | // CHECK: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[TMP4]], <8 x half> [[LANE]], <8 x half> [[TMP5]]) |
1007 | // CHECK: ret <8 x half> [[FMLA]] |
1008 | float16x8_t test_vfmsq_lane_f16(float16x8_t a, float16x8_t b, float16x4_t c) { |
1009 | return vfmsq_lane_f16(a, b, c, 3); |
1010 | } |
1011 | |
1012 | // CHECK-LABEL: test_vfms_laneq_f16 |
1013 | // CHECK: [[SUB:%.*]] = fsub <4 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
1014 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
1015 | // CHECK: [[TMP1:%.*]] = bitcast <4 x half> [[SUB]] to <8 x i8> |
1016 | // CHECK: [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8> |
1017 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
1018 | // CHECK: [[TMP4:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> |
1019 | // CHECK: [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half> |
1020 | // CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <4 x i32> <i32 7, i32 7, i32 7, i32 7> |
1021 | // CHECK: [[FMLA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[LANE]], <4 x half> [[TMP4]], <4 x half> [[TMP3]]) |
1022 | // CHECK: ret <4 x half> [[FMLA]] |
1023 | float16x4_t test_vfms_laneq_f16(float16x4_t a, float16x4_t b, float16x8_t c) { |
1024 | return vfms_laneq_f16(a, b, c, 7); |
1025 | } |
1026 | |
1027 | // CHECK-LABEL: test_vfmsq_laneq_f16 |
1028 | // CHECK: [[SUB:%.*]] = fsub <8 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
1029 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
1030 | // CHECK: [[TMP1:%.*]] = bitcast <8 x half> [[SUB]] to <16 x i8> |
1031 | // CHECK: [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8> |
1032 | // CHECK: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1033 | // CHECK: [[TMP4:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> |
1034 | // CHECK: [[TMP5:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half> |
1035 | // CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP5]], <8 x half> [[TMP5]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7> |
1036 | // CHECK: [[FMLA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[LANE]], <8 x half> [[TMP4]], <8 x half> [[TMP3]]) |
1037 | // CHECK: ret <8 x half> [[FMLA]] |
1038 | float16x8_t test_vfmsq_laneq_f16(float16x8_t a, float16x8_t b, float16x8_t c) { |
1039 | return vfmsq_laneq_f16(a, b, c, 7); |
1040 | } |
1041 | |
1042 | // CHECK-LABEL: test_vfms_n_f16 |
1043 | // CHECK: [[SUB:%.*]] = fsub <4 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
1044 | // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %c, i32 0 |
1045 | // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %c, i32 1 |
1046 | // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %c, i32 2 |
1047 | // CHECK: [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %c, i32 3 |
1048 | // CHECK: [[FMA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half> [[SUB]], <4 x half> [[TMP3]], <4 x half> %a) |
1049 | // CHECK: ret <4 x half> [[FMA]] |
1050 | float16x4_t test_vfms_n_f16(float16x4_t a, float16x4_t b, float16_t c) { |
1051 | return vfms_n_f16(a, b, c); |
1052 | } |
1053 | |
1054 | // CHECK-LABEL: test_vfmsq_n_f16 |
1055 | // CHECK: [[SUB:%.*]] = fsub <8 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, %b |
1056 | // CHECK: [[TMP0:%.*]] = insertelement <8 x half> undef, half %c, i32 0 |
1057 | // CHECK: [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %c, i32 1 |
1058 | // CHECK: [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %c, i32 2 |
1059 | // CHECK: [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %c, i32 3 |
1060 | // CHECK: [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %c, i32 4 |
1061 | // CHECK: [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %c, i32 5 |
1062 | // CHECK: [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %c, i32 6 |
1063 | // CHECK: [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %c, i32 7 |
1064 | // CHECK: [[FMA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[SUB]], <8 x half> [[TMP7]], <8 x half> %a) |
1065 | // CHECK: ret <8 x half> [[FMA]] |
1066 | float16x8_t test_vfmsq_n_f16(float16x8_t a, float16x8_t b, float16_t c) { |
1067 | return vfmsq_n_f16(a, b, c); |
1068 | } |
1069 | |
1070 | // CHECK-LABEL: test_vfmsh_lane_f16 |
1071 | // CHECK: [[TMP0:%.*]] = fpext half %b to float |
1072 | // CHECK: [[TMP1:%.*]] = fsub float -0.000000e+00, [[TMP0]] |
1073 | // CHECK: [[SUB:%.*]] = fptrunc float [[TMP1]] to half |
1074 | // CHECK: [[TMP2:%.*]] = bitcast <4 x half> %c to <8 x i8> |
1075 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <4 x half> |
1076 | // CHECK: [[EXTR:%.*]] = extractelement <4 x half> [[TMP3]], i32 3 |
1077 | // CHECK: [[FMA:%.*]] = call half @llvm.fma.f16(half [[SUB]], half [[EXTR]], half %a) |
1078 | // CHECK: ret half [[FMA]] |
1079 | float16_t test_vfmsh_lane_f16(float16_t a, float16_t b, float16x4_t c) { |
1080 | return vfmsh_lane_f16(a, b, c, 3); |
1081 | } |
1082 | |
1083 | // CHECK-LABEL: test_vfmsh_laneq_f16 |
1084 | // CHECK: [[TMP0:%.*]] = fpext half %b to float |
1085 | // CHECK: [[TMP1:%.*]] = fsub float -0.000000e+00, [[TMP0]] |
1086 | // CHECK: [[SUB:%.*]] = fptrunc float [[TMP1]] to half |
1087 | // CHECK: [[TMP2:%.*]] = bitcast <8 x half> %c to <16 x i8> |
1088 | // CHECK: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP2]] to <8 x half> |
1089 | // CHECK: [[EXTR:%.*]] = extractelement <8 x half> [[TMP3]], i32 7 |
1090 | // CHECK: [[FMA:%.*]] = call half @llvm.fma.f16(half [[SUB]], half [[EXTR]], half %a) |
1091 | // CHECK: ret half [[FMA]] |
1092 | float16_t test_vfmsh_laneq_f16(float16_t a, float16_t b, float16x8_t c) { |
1093 | return vfmsh_laneq_f16(a, b, c, 7); |
1094 | } |
1095 | |
1096 | // CHECK-LABEL: test_vmul_lane_f16 |
1097 | // CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3> |
1098 | // CHECK: [[MUL:%.*]] = fmul <4 x half> %a, [[TMP0]] |
1099 | // CHECK: ret <4 x half> [[MUL]] |
1100 | float16x4_t test_vmul_lane_f16(float16x4_t a, float16x4_t b) { |
1101 | return vmul_lane_f16(a, b, 3); |
1102 | } |
1103 | |
1104 | // CHECK-LABEL: test_vmulq_lane_f16 |
1105 | // CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7> |
1106 | // CHECK: [[MUL:%.*]] = fmul <8 x half> %a, [[TMP0]] |
1107 | // CHECK: ret <8 x half> [[MUL]] |
1108 | float16x8_t test_vmulq_lane_f16(float16x8_t a, float16x4_t b) { |
1109 | return vmulq_lane_f16(a, b, 7); |
1110 | } |
1111 | |
1112 | // CHECK-LABEL: test_vmul_laneq_f16 |
1113 | // CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <4 x i32> <i32 7, i32 7, i32 7, i32 7> |
1114 | // CHECK: [[MUL:%.*]] = fmul <4 x half> %a, [[TMP0]] |
1115 | // CHECK: ret <4 x half> [[MUL]] |
1116 | float16x4_t test_vmul_laneq_f16(float16x4_t a, float16x8_t b) { |
1117 | return vmul_laneq_f16(a, b, 7); |
1118 | } |
1119 | |
1120 | // CHECK-LABEL: test_vmulq_laneq_f16 |
1121 | // CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7> |
1122 | // CHECK: [[MUL:%.*]] = fmul <8 x half> %a, [[TMP0]] |
1123 | // CHECK: ret <8 x half> [[MUL]] |
1124 | float16x8_t test_vmulq_laneq_f16(float16x8_t a, float16x8_t b) { |
1125 | return vmulq_laneq_f16(a, b, 7); |
1126 | } |
1127 | |
1128 | // CHECK-LABEL: test_vmul_n_f16 |
1129 | // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %b, i32 0 |
1130 | // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %b, i32 1 |
1131 | // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %b, i32 2 |
1132 | // CHECK: [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %b, i32 3 |
1133 | // CHECK: [[MUL:%.*]] = fmul <4 x half> %a, [[TMP3]] |
1134 | // CHECK: ret <4 x half> [[MUL]] |
1135 | float16x4_t test_vmul_n_f16(float16x4_t a, float16_t b) { |
1136 | return vmul_n_f16(a, b); |
1137 | } |
1138 | |
1139 | // CHECK-LABEL: test_vmulq_n_f16 |
1140 | // CHECK: [[TMP0:%.*]] = insertelement <8 x half> undef, half %b, i32 0 |
1141 | // CHECK: [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %b, i32 1 |
1142 | // CHECK: [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %b, i32 2 |
1143 | // CHECK: [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %b, i32 3 |
1144 | // CHECK: [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %b, i32 4 |
1145 | // CHECK: [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %b, i32 5 |
1146 | // CHECK: [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %b, i32 6 |
1147 | // CHECK: [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %b, i32 7 |
1148 | // CHECK: [[MUL:%.*]] = fmul <8 x half> %a, [[TMP7]] |
1149 | // CHECK: ret <8 x half> [[MUL]] |
1150 | float16x8_t test_vmulq_n_f16(float16x8_t a, float16_t b) { |
1151 | return vmulq_n_f16(a, b); |
1152 | } |
1153 | |
1154 | // FIXME: Fix it when fp16 non-storage-only type becomes available. |
1155 | // CHECK-LABEL: test_vmulh_lane_f16 |
1156 | // CHECK: [[CONV0:%.*]] = fpext half %a to float |
1157 | // CHECK: [[CONV1:%.*]] = fpext half %{{.*}} to float |
1158 | // CHECK: [[MUL:%.*]] = fmul float [[CONV0:%.*]], [[CONV0:%.*]] |
1159 | // CHECK: [[CONV3:%.*]] = fptrunc float %mul to half |
1160 | // CHECK: ret half [[CONV3:%.*]] |
1161 | float16_t test_vmulh_lane_f16(float16_t a, float16x4_t b) { |
1162 | return vmulh_lane_f16(a, b, 3); |
1163 | } |
1164 | |
1165 | // CHECK-LABEL: test_vmulh_laneq_f16 |
1166 | // CHECK: [[CONV0:%.*]] = fpext half %a to float |
1167 | // CHECK: [[CONV1:%.*]] = fpext half %{{.*}} to float |
1168 | // CHECK: [[MUL:%.*]] = fmul float [[CONV0:%.*]], [[CONV0:%.*]] |
1169 | // CHECK: [[CONV3:%.*]] = fptrunc float %mul to half |
1170 | // CHECK: ret half [[CONV3:%.*]] |
1171 | float16_t test_vmulh_laneq_f16(float16_t a, float16x8_t b) { |
1172 | return vmulh_laneq_f16(a, b, 7); |
1173 | } |
1174 | |
1175 | // CHECK-LABEL: test_vmulx_lane_f16 |
1176 | // CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3> |
1177 | // CHECK: [[MUL:%.*]] = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> %a, <4 x half> [[TMP0]]) |
1178 | // CHECK: ret <4 x half> [[MUL]] |
1179 | float16x4_t test_vmulx_lane_f16(float16x4_t a, float16x4_t b) { |
1180 | return vmulx_lane_f16(a, b, 3); |
1181 | } |
1182 | |
1183 | // CHECK-LABEL: test_vmulxq_lane_f16 |
1184 | // CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7> |
1185 | // CHECK: [[MUL:%.*]] = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> %a, <8 x half> [[TMP0]]) |
1186 | // CHECK: ret <8 x half> [[MUL]] |
1187 | float16x8_t test_vmulxq_lane_f16(float16x8_t a, float16x4_t b) { |
1188 | return vmulxq_lane_f16(a, b, 7); |
1189 | } |
1190 | |
1191 | // CHECK-LABEL: test_vmulx_laneq_f16 |
1192 | // CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <4 x i32> <i32 7, i32 7, i32 7, i32 7> |
1193 | // CHECK: [[MUL:%.*]] = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> %a, <4 x half> [[TMP0]]) |
1194 | // CHECK: ret <4 x half> [[MUL]] |
1195 | float16x4_t test_vmulx_laneq_f16(float16x4_t a, float16x8_t b) { |
1196 | return vmulx_laneq_f16(a, b, 7); |
1197 | } |
1198 | |
1199 | // CHECK-LABEL: test_vmulxq_laneq_f16 |
1200 | // CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7> |
1201 | // CHECK: [[MUL:%.*]] = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> %a, <8 x half> [[TMP0]]) |
1202 | // CHECK: ret <8 x half> [[MUL]] |
1203 | float16x8_t test_vmulxq_laneq_f16(float16x8_t a, float16x8_t b) { |
1204 | return vmulxq_laneq_f16(a, b, 7); |
1205 | } |
1206 | |
1207 | // CHECK-LABEL: test_vmulx_n_f16 |
1208 | // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %b, i32 0 |
1209 | // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %b, i32 1 |
1210 | // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %b, i32 2 |
1211 | // CHECK: [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %b, i32 3 |
1212 | // CHECK: [[MUL:%.*]] = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> %a, <4 x half> [[TMP3]]) |
1213 | // CHECK: ret <4 x half> [[MUL]] |
1214 | float16x4_t test_vmulx_n_f16(float16x4_t a, float16_t b) { |
1215 | return vmulx_n_f16(a, b); |
1216 | } |
1217 | |
1218 | // CHECK-LABEL: test_vmulxq_n_f16 |
1219 | // CHECK: [[TMP0:%.*]] = insertelement <8 x half> undef, half %b, i32 0 |
1220 | // CHECK: [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %b, i32 1 |
1221 | // CHECK: [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %b, i32 2 |
1222 | // CHECK: [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %b, i32 3 |
1223 | // CHECK: [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %b, i32 4 |
1224 | // CHECK: [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %b, i32 5 |
1225 | // CHECK: [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %b, i32 6 |
1226 | // CHECK: [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %b, i32 7 |
1227 | // CHECK: [[MUL:%.*]] = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> %a, <8 x half> [[TMP7]]) |
1228 | // CHECK: ret <8 x half> [[MUL]] |
1229 | float16x8_t test_vmulxq_n_f16(float16x8_t a, float16_t b) { |
1230 | return vmulxq_n_f16(a, b); |
1231 | } |
1232 | |
1233 | // CHECK-LABEL: test_vmulxh_lane_f16 |
1234 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %b to <8 x i8> |
1235 | // CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
1236 | // CHECK: [[EXTR:%.*]] = extractelement <4 x half> [[TMP1]], i32 3 |
1237 | // CHECK: [[MULX:%.*]] = call half @llvm.aarch64.neon.fmulx.f16(half %a, half [[EXTR]] |
1238 | // CHECK: ret half [[MULX]] |
1239 | float16_t test_vmulxh_lane_f16(float16_t a, float16x4_t b) { |
1240 | return vmulxh_lane_f16(a, b, 3); |
1241 | } |
1242 | |
1243 | // CHECK-LABEL: test_vmulxh_laneq_f16 |
1244 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %b to <16 x i8> |
1245 | // CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1246 | // CHECK: [[EXTR:%.*]] = extractelement <8 x half> [[TMP1]], i32 7 |
1247 | // CHECK: [[MULX:%.*]] = call half @llvm.aarch64.neon.fmulx.f16(half %a, half [[EXTR]]) |
1248 | // CHECK: ret half [[MULX]] |
1249 | float16_t test_vmulxh_laneq_f16(float16_t a, float16x8_t b) { |
1250 | return vmulxh_laneq_f16(a, b, 7); |
1251 | } |
1252 | |
1253 | // CHECK-LABEL: test_vmaxv_f16 |
1254 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
1255 | // CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
1256 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fmaxv.f16.v4f16(<4 x half> [[TMP1]]) |
1257 | // CHECK: ret half [[MAX]] |
1258 | float16_t test_vmaxv_f16(float16x4_t a) { |
1259 | return vmaxv_f16(a); |
1260 | } |
1261 | |
1262 | // CHECK-LABEL: test_vmaxvq_f16 |
1263 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
1264 | // CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1265 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fmaxv.f16.v8f16(<8 x half> [[TMP1]]) |
1266 | // CHECK: ret half [[MAX]] |
1267 | float16_t test_vmaxvq_f16(float16x8_t a) { |
1268 | return vmaxvq_f16(a); |
1269 | } |
1270 | |
1271 | // CHECK-LABEL: test_vminv_f16 |
1272 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
1273 | // CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
1274 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fminv.f16.v4f16(<4 x half> [[TMP1]]) |
1275 | // CHECK: ret half [[MAX]] |
1276 | float16_t test_vminv_f16(float16x4_t a) { |
1277 | return vminv_f16(a); |
1278 | } |
1279 | |
1280 | // CHECK-LABEL: test_vminvq_f16 |
1281 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
1282 | // CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1283 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fminv.f16.v8f16(<8 x half> [[TMP1]]) |
1284 | // CHECK: ret half [[MAX]] |
1285 | float16_t test_vminvq_f16(float16x8_t a) { |
1286 | return vminvq_f16(a); |
1287 | } |
1288 | |
1289 | // CHECK-LABEL: test_vmaxnmv_f16 |
1290 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
1291 | // CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
1292 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fmaxnmv.f16.v4f16(<4 x half> [[TMP1]]) |
1293 | // CHECK: ret half [[MAX]] |
1294 | float16_t test_vmaxnmv_f16(float16x4_t a) { |
1295 | return vmaxnmv_f16(a); |
1296 | } |
1297 | |
1298 | // CHECK-LABEL: test_vmaxnmvq_f16 |
1299 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
1300 | // CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1301 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fmaxnmv.f16.v8f16(<8 x half> [[TMP1]]) |
1302 | // CHECK: ret half [[MAX]] |
1303 | float16_t test_vmaxnmvq_f16(float16x8_t a) { |
1304 | return vmaxnmvq_f16(a); |
1305 | } |
1306 | |
1307 | // CHECK-LABEL: test_vminnmv_f16 |
1308 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
1309 | // CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
1310 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fminnmv.f16.v4f16(<4 x half> [[TMP1]]) |
1311 | // CHECK: ret half [[MAX]] |
1312 | float16_t test_vminnmv_f16(float16x4_t a) { |
1313 | return vminnmv_f16(a); |
1314 | } |
1315 | |
1316 | // CHECK-LABEL: test_vminnmvq_f16 |
1317 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
1318 | // CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1319 | // CHECK: [[MAX:%.*]] = call half @llvm.aarch64.neon.fminnmv.f16.v8f16(<8 x half> [[TMP1]]) |
1320 | // CHECK: ret half [[MAX]] |
1321 | float16_t test_vminnmvq_f16(float16x8_t a) { |
1322 | return vminnmvq_f16(a); |
1323 | } |
1324 | |
1325 | // CHECK-LABEL: test_vbsl_f16 |
1326 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %b to <8 x i8> |
1327 | // CHECK: [[TMP1:%.*]] = bitcast <4 x half> %c to <8 x i8> |
1328 | // CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16> |
1329 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16> |
1330 | // CHECK: [[TMP4:%.*]] = and <4 x i16> %a, [[TMP2]] |
1331 | // CHECK: [[TMP5:%.*]] = xor <4 x i16> %a, <i16 -1, i16 -1, i16 -1, i16 -1> |
1332 | // CHECK: [[TMP6:%.*]] = and <4 x i16> [[TMP5]], [[TMP3]] |
1333 | // CHECK: [[TMP7:%.*]] = or <4 x i16> [[TMP4]], [[TMP6]] |
1334 | // CHECK: [[TMP8:%.*]] = bitcast <4 x i16> [[TMP7]] to <4 x half> |
1335 | // CHECK: ret <4 x half> [[TMP8]] |
1336 | float16x4_t test_vbsl_f16(uint16x4_t a, float16x4_t b, float16x4_t c) { |
1337 | return vbsl_f16(a, b, c); |
1338 | } |
1339 | |
1340 | // CHECK-LABEL: test_vbslq_f16 |
1341 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %b to <16 x i8> |
1342 | // CHECK: [[TMP1:%.*]] = bitcast <8 x half> %c to <16 x i8> |
1343 | // CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16> |
1344 | // CHECK: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16> |
1345 | // CHECK: [[TMP4:%.*]] = and <8 x i16> %a, [[TMP2]] |
1346 | // CHECK: [[TMP5:%.*]] = xor <8 x i16> %a, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1> |
1347 | // CHECK: [[TMP6:%.*]] = and <8 x i16> [[TMP5]], [[TMP3]] |
1348 | // CHECK: [[TMP7:%.*]] = or <8 x i16> [[TMP4]], [[TMP6]] |
1349 | // CHECK: [[TMP8:%.*]] = bitcast <8 x i16> [[TMP7]] to <8 x half> |
1350 | // CHECK: ret <8 x half> [[TMP8]] |
1351 | float16x8_t test_vbslq_f16(uint16x8_t a, float16x8_t b, float16x8_t c) { |
1352 | return vbslq_f16(a, b, c); |
1353 | } |
1354 | |
1355 | // CHECK-LABEL: test_vzip_f16 |
1356 | // CHECK: [[RETVAL:%.*]] = alloca %struct.float16x4x2_t, align 8 |
1357 | // CHECK: [[__RET_I:%.*]] = alloca %struct.float16x4x2_t, align 8 |
1358 | // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x2_t* [[RETVAL]] to i8* |
1359 | // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x half>* |
1360 | // CHECK: [[VZIP0_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 4, i32 1, i32 5> |
1361 | // CHECK: store <4 x half> [[VZIP0_I]], <4 x half>* [[TMP1]] |
1362 | // CHECK: [[TMP2:%.*]] = getelementptr inbounds <4 x half>, <4 x half>* [[TMP1]], i32 1 |
1363 | // CHECK: [[VZIP1_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 2, i32 6, i32 3, i32 7> |
1364 | // CHECK: store <4 x half> [[VZIP1_I]], <4 x half>* [[TMP2]] |
1365 | float16x4x2_t test_vzip_f16(float16x4_t a, float16x4_t b) { |
1366 | return vzip_f16(a, b); |
1367 | } |
1368 | |
1369 | // CHECK-LABEL: test_vzipq_f16 |
1370 | // CHECK: [[RETVAL:%.*]] = alloca %struct.float16x8x2_t, align 16 |
1371 | // CHECK: [[__RET_I:%.*]] = alloca %struct.float16x8x2_t, align 16 |
1372 | // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x8x2_t* [[RETVAL]] to i8* |
1373 | // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <8 x half>* |
1374 | // CHECK: [[VZIP0_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11> |
1375 | // CHECK: store <8 x half> [[VZIP0_I]], <8 x half>* [[TMP1]] |
1376 | // CHECK: [[TMP2:%.*]] = getelementptr inbounds <8 x half>, <8 x half>* [[TMP1]], i32 1 |
1377 | // CHECK: [[VZIP1_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15> |
1378 | // CHECK: store <8 x half> [[VZIP1_I]], <8 x half>* [[TMP2]] |
1379 | float16x8x2_t test_vzipq_f16(float16x8_t a, float16x8_t b) { |
1380 | return vzipq_f16(a, b); |
1381 | } |
1382 | |
1383 | // CHECK-LABEL: test_vuzp_f16 |
1384 | // CHECK: [[RETVAL:%.*]] = alloca %struct.float16x4x2_t, align 8 |
1385 | // CHECK: [[__RET_I:%.*]] = alloca %struct.float16x4x2_t, align 8 |
1386 | // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x2_t* [[RETVAL]] to i8* |
1387 | // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x half>* |
1388 | // CHECK: [[VZIP0_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 6> |
1389 | // CHECK: store <4 x half> [[VZIP0_I]], <4 x half>* [[TMP1]] |
1390 | // CHECK: [[TMP2:%.*]] = getelementptr inbounds <4 x half>, <4 x half>* [[TMP1]], i32 1 |
1391 | // CHECK: [[VZIP1_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 1, i32 3, i32 5, i32 7> |
1392 | // CHECK: store <4 x half> [[VZIP1_I]], <4 x half>* [[TMP2]] |
1393 | float16x4x2_t test_vuzp_f16(float16x4_t a, float16x4_t b) { |
1394 | return vuzp_f16(a, b); |
1395 | } |
1396 | |
1397 | // CHECK-LABEL: test_vuzpq_f16 |
1398 | // CHECK: [[RETVAL:%.*]] = alloca %struct.float16x8x2_t, align 16 |
1399 | // CHECK: [[__RET_I:%.*]] = alloca %struct.float16x8x2_t, align 16 |
1400 | // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x8x2_t* [[RETVAL]] to i8* |
1401 | // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <8 x half>* |
1402 | // CHECK: [[VZIP0_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> |
1403 | // CHECK: store <8 x half> [[VZIP0_I]], <8 x half>* [[TMP1]] |
1404 | // CHECK: [[TMP2:%.*]] = getelementptr inbounds <8 x half>, <8 x half>* [[TMP1]], i32 1 |
1405 | // CHECK: [[VZIP1_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> |
1406 | // CHECK: store <8 x half> [[VZIP1_I]], <8 x half>* [[TMP2]] |
1407 | float16x8x2_t test_vuzpq_f16(float16x8_t a, float16x8_t b) { |
1408 | return vuzpq_f16(a, b); |
1409 | } |
1410 | |
1411 | // CHECK-LABEL: test_vtrn_f16 |
1412 | // CHECK: [[RETVAL:%.*]] = alloca %struct.float16x4x2_t, align 8 |
1413 | // CHECK: [[__RET_I:%.*]] = alloca %struct.float16x4x2_t, align 8 |
1414 | // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x2_t* [[RETVAL]] to i8* |
1415 | // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x half>* |
1416 | // CHECK: [[VZIP0_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 4, i32 2, i32 6> |
1417 | // CHECK: store <4 x half> [[VZIP0_I]], <4 x half>* [[TMP1]] |
1418 | // CHECK: [[TMP2:%.*]] = getelementptr inbounds <4 x half>, <4 x half>* [[TMP1]], i32 1 |
1419 | // CHECK: [[VZIP1_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 1, i32 5, i32 3, i32 7> |
1420 | // CHECK: store <4 x half> [[VZIP1_I]], <4 x half>* [[TMP2]] |
1421 | float16x4x2_t test_vtrn_f16(float16x4_t a, float16x4_t b) { |
1422 | return vtrn_f16(a, b); |
1423 | } |
1424 | |
1425 | // CHECK-LABEL: test_vtrnq_f16 |
1426 | // CHECK: [[RETVAL:%.*]] = alloca %struct.float16x8x2_t, align 16 |
1427 | // CHECK: [[__RET_I:%.*]] = alloca %struct.float16x8x2_t, align 16 |
1428 | // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x8x2_t* [[RETVAL]] to i8* |
1429 | // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <8 x half>* |
1430 | // CHECK: [[VZIP0_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14> |
1431 | // CHECK: store <8 x half> [[VZIP0_I]], <8 x half>* [[TMP1]] |
1432 | // CHECK: [[TMP2:%.*]] = getelementptr inbounds <8 x half>, <8 x half>* [[TMP1]], i32 1 |
1433 | // CHECK: [[VZIP1_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15> |
1434 | // CHECK: store <8 x half> [[VZIP1_I]], <8 x half>* [[TMP2]] |
1435 | float16x8x2_t test_vtrnq_f16(float16x8_t a, float16x8_t b) { |
1436 | return vtrnq_f16(a, b); |
1437 | } |
1438 | |
1439 | // CHECK-LABEL: test_vmov_n_f16 |
1440 | // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %a, i32 0 |
1441 | // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %a, i32 1 |
1442 | // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %a, i32 2 |
1443 | // CHECK: [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %a, i32 3 |
1444 | // CHECK: ret <4 x half> [[TMP3]] |
1445 | float16x4_t test_vmov_n_f16(float16_t a) { |
1446 | return vmov_n_f16(a); |
1447 | } |
1448 | |
1449 | // CHECK-LABEL: test_vmovq_n_f16 |
1450 | // CHECK: [[TMP0:%.*]] = insertelement <8 x half> undef, half %a, i32 0 |
1451 | // CHECK: [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %a, i32 1 |
1452 | // CHECK: [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %a, i32 2 |
1453 | // CHECK: [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %a, i32 3 |
1454 | // CHECK: [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %a, i32 4 |
1455 | // CHECK: [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %a, i32 5 |
1456 | // CHECK: [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %a, i32 6 |
1457 | // CHECK: [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %a, i32 7 |
1458 | // CHECK: ret <8 x half> [[TMP7]] |
1459 | float16x8_t test_vmovq_n_f16(float16_t a) { |
1460 | return vmovq_n_f16(a); |
1461 | } |
1462 | |
1463 | // CHECK-LABEL: test_vdup_n_f16 |
1464 | // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %a, i32 0 |
1465 | // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %a, i32 1 |
1466 | // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %a, i32 2 |
1467 | // CHECK: [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half %a, i32 3 |
1468 | // CHECK: ret <4 x half> [[TMP3]] |
1469 | float16x4_t test_vdup_n_f16(float16_t a) { |
1470 | return vdup_n_f16(a); |
1471 | } |
1472 | |
1473 | // CHECK-LABEL: test_vdupq_n_f16 |
1474 | // CHECK: [[TMP0:%.*]] = insertelement <8 x half> undef, half %a, i32 0 |
1475 | // CHECK: [[TMP1:%.*]] = insertelement <8 x half> [[TMP0]], half %a, i32 1 |
1476 | // CHECK: [[TMP2:%.*]] = insertelement <8 x half> [[TMP1]], half %a, i32 2 |
1477 | // CHECK: [[TMP3:%.*]] = insertelement <8 x half> [[TMP2]], half %a, i32 3 |
1478 | // CHECK: [[TMP4:%.*]] = insertelement <8 x half> [[TMP3]], half %a, i32 4 |
1479 | // CHECK: [[TMP5:%.*]] = insertelement <8 x half> [[TMP4]], half %a, i32 5 |
1480 | // CHECK: [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %a, i32 6 |
1481 | // CHECK: [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %a, i32 7 |
1482 | // CHECK: ret <8 x half> [[TMP7]] |
1483 | float16x8_t test_vdupq_n_f16(float16_t a) { |
1484 | return vdupq_n_f16(a); |
1485 | } |
1486 | |
1487 | // CHECK-LABEL: test_vdup_lane_f16 |
1488 | // CHECK: [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %a, <4 x i32> <i32 3, i32 3, i32 3, i32 3> |
1489 | // CHECK: ret <4 x half> [[SHFL]] |
1490 | float16x4_t test_vdup_lane_f16(float16x4_t a) { |
1491 | return vdup_lane_f16(a, 3); |
1492 | } |
1493 | |
1494 | // CHECK-LABEL: test_vdupq_lane_f16 |
1495 | // CHECK: [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %a, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7> |
1496 | // CHECK: ret <8 x half> [[SHFL]] |
1497 | float16x8_t test_vdupq_lane_f16(float16x4_t a) { |
1498 | return vdupq_lane_f16(a, 7); |
1499 | } |
1500 | |
1501 | // CHECK-LABEL: @test_vext_f16( |
1502 | // CHECK: [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8> |
1503 | // CHECK: [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8> |
1504 | // CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half> |
1505 | // CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> |
1506 | // CHECK: [[VEXT:%.*]] = shufflevector <4 x half> [[TMP2]], <4 x half> [[TMP3]], <4 x i32> <i32 2, i32 3, i32 4, i32 5> |
1507 | // CHECK: ret <4 x half> [[VEXT]] |
1508 | float16x4_t test_vext_f16(float16x4_t a, float16x4_t b) { |
1509 | return vext_f16(a, b, 2); |
1510 | } |
1511 | |
1512 | // CHECK-LABEL: @test_vextq_f16( |
1513 | // CHECK: [[TMP0:%.*]] = bitcast <8 x half> %a to <16 x i8> |
1514 | // CHECK: [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8> |
1515 | // CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half> |
1516 | // CHECK: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> |
1517 | // CHECK: [[VEXT:%.*]] = shufflevector <8 x half> [[TMP2]], <8 x half> [[TMP3]], <8 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12> |
1518 | // CHECK: ret <8 x half> [[VEXT]] |
1519 | float16x8_t test_vextq_f16(float16x8_t a, float16x8_t b) { |
1520 | return vextq_f16(a, b, 5); |
1521 | } |
1522 | |
1523 | // CHECK-LABEL: @test_vrev64_f16( |
1524 | // CHECK: [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %a, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
1525 | // CHECK: ret <4 x half> [[SHFL]] |
1526 | float16x4_t test_vrev64_f16(float16x4_t a) { |
1527 | return vrev64_f16(a); |
1528 | } |
1529 | |
1530 | // CHECK-LABEL: @test_vrev64q_f16( |
1531 | // CHECK: [[SHFL:%.*]] = shufflevector <8 x half> %a, <8 x half> %a, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4> |
1532 | // CHECK: ret <8 x half> [[SHFL]] |
1533 | float16x8_t test_vrev64q_f16(float16x8_t a) { |
1534 | return vrev64q_f16(a); |
1535 | } |
1536 | |
1537 | // CHECK-LABEL: @test_vzip1_f16( |
1538 | // CHECK: [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 4, i32 1, i32 5> |
1539 | // CHECK: ret <4 x half> [[SHFL]] |
1540 | float16x4_t test_vzip1_f16(float16x4_t a, float16x4_t b) { |
1541 | return vzip1_f16(a, b); |
1542 | } |
1543 | |
1544 | // CHECK-LABEL: @test_vzip1q_f16( |
1545 | // CHECK: [[SHFL:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11> |
1546 | // CHECK: ret <8 x half> [[SHFL]] |
1547 | float16x8_t test_vzip1q_f16(float16x8_t a, float16x8_t b) { |
1548 | return vzip1q_f16(a, b); |
1549 | } |
1550 | |
1551 | // CHECK-LABEL: @test_vzip2_f16( |
1552 | // CHECK: [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 2, i32 6, i32 3, i32 7> |
1553 | // CHECK: ret <4 x half> [[SHFL]] |
1554 | float16x4_t test_vzip2_f16(float16x4_t a, float16x4_t b) { |
1555 | return vzip2_f16(a, b); |
1556 | } |
1557 | |
1558 | // CHECK-LABEL: @test_vzip2q_f16( |
1559 | // CHECK: [[SHFL:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15> |
1560 | // CHECK: ret <8 x half> [[SHFL]] |
1561 | float16x8_t test_vzip2q_f16(float16x8_t a, float16x8_t b) { |
1562 | return vzip2q_f16(a, b); |
1563 | } |
1564 | |
1565 | // CHECK-LABEL: @test_vuzp1_f16( |
1566 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 6> |
1567 | // CHECK: ret <4 x half> [[SHUFFLE_I]] |
1568 | float16x4_t test_vuzp1_f16(float16x4_t a, float16x4_t b) { |
1569 | return vuzp1_f16(a, b); |
1570 | } |
1571 | |
1572 | // CHECK-LABEL: @test_vuzp1q_f16( |
1573 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> |
1574 | // CHECK: ret <8 x half> [[SHUFFLE_I]] |
1575 | float16x8_t test_vuzp1q_f16(float16x8_t a, float16x8_t b) { |
1576 | return vuzp1q_f16(a, b); |
1577 | } |
1578 | |
1579 | // CHECK-LABEL: @test_vuzp2_f16( |
1580 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 1, i32 3, i32 5, i32 7> |
1581 | // CHECK: ret <4 x half> [[SHUFFLE_I]] |
1582 | float16x4_t test_vuzp2_f16(float16x4_t a, float16x4_t b) { |
1583 | return vuzp2_f16(a, b); |
1584 | } |
1585 | |
1586 | // CHECK-LABEL: @test_vuzp2q_f16( |
1587 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> |
1588 | // CHECK: ret <8 x half> [[SHUFFLE_I]] |
1589 | float16x8_t test_vuzp2q_f16(float16x8_t a, float16x8_t b) { |
1590 | return vuzp2q_f16(a, b); |
1591 | } |
1592 | |
1593 | // CHECK-LABEL: @test_vtrn1_f16( |
1594 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 4, i32 2, i32 6> |
1595 | // CHECK: ret <4 x half> [[SHUFFLE_I]] |
1596 | float16x4_t test_vtrn1_f16(float16x4_t a, float16x4_t b) { |
1597 | return vtrn1_f16(a, b); |
1598 | } |
1599 | |
1600 | // CHECK-LABEL: @test_vtrn1q_f16( |
1601 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14> |
1602 | // CHECK: ret <8 x half> [[SHUFFLE_I]] |
1603 | float16x8_t test_vtrn1q_f16(float16x8_t a, float16x8_t b) { |
1604 | return vtrn1q_f16(a, b); |
1605 | } |
1606 | |
1607 | // CHECK-LABEL: @test_vtrn2_f16( |
1608 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 1, i32 5, i32 3, i32 7> |
1609 | // CHECK: ret <4 x half> [[SHUFFLE_I]] |
1610 | float16x4_t test_vtrn2_f16(float16x4_t a, float16x4_t b) { |
1611 | return vtrn2_f16(a, b); |
1612 | } |
1613 | |
1614 | // CHECK-LABEL: @test_vtrn2q_f16( |
1615 | // CHECK: [[SHUFFLE_I:%.*]] = shufflevector <8 x half> %a, <8 x half> %b, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15> |
1616 | // CHECK: ret <8 x half> [[SHUFFLE_I]] |
1617 | float16x8_t test_vtrn2q_f16(float16x8_t a, float16x8_t b) { |
1618 | return vtrn2q_f16(a, b); |
1619 | } |
1620 | |
1621 | |