1 /* { dg-do assemble { target aarch64_asm_sve_ok } } */
2 /* { dg-options "-O -msve-vector-bits=2048 -mlittle-endian --save-temps" } */
3 /* { dg-final { check-function-bodies "**" "" } } */
4
5 typedef unsigned char v128qi __attribute__((vector_size(128)));
6 typedef unsigned char v64qi __attribute__((vector_size(64)));
7 typedef unsigned char v32qi __attribute__((vector_size(32)));
8 typedef unsigned short v64hi __attribute__((vector_size(128)));
9 typedef unsigned short v32hi __attribute__((vector_size(64)));
10 typedef _Float16 v64hf __attribute__((vector_size(128)));
11 typedef _Float16 v32hf __attribute__((vector_size(64)));
12 typedef __bf16 v64bf __attribute__((vector_size(128)));
13 typedef __bf16 v32bf __attribute__((vector_size(64)));
14 typedef unsigned int v32si __attribute__((vector_size(128)));
15 typedef float v32sf __attribute__((vector_size(128)));
16
17 #define PERM0(B, C) B, B + C
18 #define PERM1(B, C) PERM0 (B, C), PERM0 (B + 2, C)
19 #define PERM2(B, C) PERM1 (B, C), PERM1 (B + 4, C)
20 #define PERM3(B, C) PERM2 (B, C), PERM2 (B + 8, C)
21 #define PERM4(B, C) PERM3 (B, C), PERM3 (B + 16, C)
22 #define PERM5(B, C) PERM4 (B, C), PERM4 (B + 32, C)
23 #define PERM6(B, C) PERM5 (B, C), PERM5 (B + 64, C)
24
25 /*
26 ** qi_trn1_h_a:
27 ** ptrue (p[0-7])\.b, vl256
28 ** ld1b (z[0-9]+)\.h, \1/z, \[x0\]
29 ** trn1 (z[0-9]+)\.h, \2\.h, \2\.h
30 ** st1b \3\.h, \1, \[x8\]
31 ** ret
32 */
33 v128qi
34 qi_trn1_h_a (v128qi x)
35 {
36 return __builtin_shuffle (x, x, (v128qi) { PERM6 (0, 0) });
37 }
38
39 /*
40 ** qi_trn1_h_b:
41 ** ptrue (p[0-7])\.b, vl256
42 ** ld1b (z[0-9]+)\.h, \1/z, \[x0\]
43 ** trn1 (z[0-9]+)\.h, \2\.h, \2\.h
44 ** st1b \3\.h, \1, \[x8\]
45 ** ret
46 */
47 v128qi
48 qi_trn1_h_b (v128qi x)
49 {
50 return __builtin_shuffle (x, x, (v128qi) { PERM6 (0, 128) });
51 }
52
53 /*
54 ** qi_trn1_h_c:
55 ** ptrue (p[0-7])\.b, vl256
56 ** ld1b (z[0-9]+)\.h, \1/z, \[x0\]
57 ** trn1 (z[0-9]+)\.h, \2\.h, \2\.h
58 ** st1b \3\.h, \1, \[x8\]
59 ** ret
60 */
61 v128qi
62 qi_trn1_h_c (v128qi x)
63 {
64 return __builtin_shuffle (x, x, (v128qi) { PERM6 (128, 0) });
65 }
66
67 /*
68 ** qi_trn1_h_two_op:
69 ** ptrue (p[0-7])\.b, vl256
70 ** (
71 ** ld1b (z[0-9]+)\.h, \1/z, \[x1\]
72 ** ld1b (z[0-9]+)\.h, \1/z, \[x0\]
73 ** trn1 \3\.h, \3\.h, \2\.h
74 ** st1b \3\.h, \1, \[x8\]
75 ** |
76 ** ld1b (z[0-9]+)\.h, \1/z, \[x0\]
77 ** ld1b (z[0-9]+)\.h, \1/z, \[x1\]
78 ** trn1 \4\.h, \4\.h, \5\.h
79 ** st1b \4\.h, \1, \[x8\]
80 ** )
81 ** ret
82 */
83 v128qi
84 qi_trn1_h_two_op (v128qi x, v128qi y)
85 {
86 return __builtin_shuffle (x, y, (v128qi) { PERM6 (0, 128) });
87 }
88
89 /*
90 ** qi_trn1_s:
91 ** ptrue (p[0-7])\.b, vl256
92 ** ld1b (z[0-9]+)\.s, \1/z, \[x0\]
93 ** trn1 (z[0-9]+)\.s, \2\.s, \2\.s
94 ** st1b \3\.s, \1, \[x8\]
95 ** ret
96 */
97 v64qi
98 qi_trn1_s (v64qi x)
99 {
100 return __builtin_shuffle (x, x, (v64qi) { PERM5 (0, 64) });
101 }
102
103 /*
104 ** qi_trn1_s_two_op:
105 ** ptrue (p[0-7])\.b, vl256
106 ** (
107 ** ld1b (z[0-9]+)\.s, \1/z, \[x1\]
108 ** ld1b (z[0-9]+)\.s, \1/z, \[x0\]
109 ** trn1 \3\.s, \3\.s, \2\.s
110 ** st1b \3\.s, \1, \[x8\]
111 ** |
112 ** ld1b (z[0-9]+)\.s, \1/z, \[x0\]
113 ** ld1b (z[0-9]+)\.s, \1/z, \[x1\]
114 ** trn1 \4\.s, \4\.s, \5\.s
115 ** st1b \4\.s, \1, \[x8\]
116 ** )
117 ** ret
118 */
119 v64qi
120 qi_trn1_s_two_op (v64qi x, v64qi y)
121 {
122 return __builtin_shuffle (x, y, (v64qi) { PERM5 (0, 64) });
123 }
124
125 /*
126 ** qi_trn1_d:
127 ** ptrue (p[0-7])\.b, vl256
128 ** ld1b (z[0-9]+)\.d, \1/z, \[x0\]
129 ** trn1 (z[0-9]+)\.d, \2\.d, \2\.d
130 ** st1b \3\.d, \1, \[x8\]
131 ** ret
132 */
133 v32qi
134 qi_trn1_d (v32qi x)
135 {
136 return __builtin_shuffle (x, x, (v32qi) { PERM4 (0, 32) });
137 }
138
139 /*
140 ** qi_trn1_d_two_op:
141 ** ptrue (p[0-7])\.b, vl256
142 ** (
143 ** ld1b (z[0-9]+)\.d, \1/z, \[x1\]
144 ** ld1b (z[0-9]+)\.d, \1/z, \[x0\]
145 ** trn1 \3\.d, \3\.d, \2\.d
146 ** st1b \3\.d, \1, \[x8\]
147 ** |
148 ** ld1b (z[0-9]+)\.d, \1/z, \[x0\]
149 ** ld1b (z[0-9]+)\.d, \1/z, \[x1\]
150 ** trn1 \4\.d, \4\.d, \5\.d
151 ** st1b \4\.d, \1, \[x8\]
152 ** )
153 ** ret
154 */
155 v32qi
156 qi_trn1_d_two_op (v32qi x, v32qi y)
157 {
158 return __builtin_shuffle (x, y, (v32qi) { PERM4 (0, 32) });
159 }
160
161 /*
162 ** hi_trn1_s:
163 ** ptrue (p[0-7])\.b, vl256
164 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
165 ** trn1 (z[0-9]+)\.s, \2\.s, \2\.s
166 ** st1h \3\.s, \1, \[x8\]
167 ** ret
168 */
169 v64hi
170 hi_trn1_s (v64hi x)
171 {
172 return __builtin_shuffle (x, x, (v64hi) { PERM5 (0, 64) });
173 }
174
175 /*
176 ** hi_trn1_s_two_op:
177 ** ptrue (p[0-7])\.b, vl256
178 ** (
179 ** ld1h (z[0-9]+)\.s, \1/z, \[x1\]
180 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
181 ** trn1 \3\.s, \3\.s, \2\.s
182 ** st1h \3\.s, \1, \[x8\]
183 ** |
184 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
185 ** ld1h (z[0-9]+)\.s, \1/z, \[x1\]
186 ** trn1 \4\.s, \4\.s, \5\.s
187 ** st1h \4\.s, \1, \[x8\]
188 ** )
189 ** ret
190 */
191 v64hi
192 hi_trn1_s_two_op (v64hi x, v64hi y)
193 {
194 return __builtin_shuffle (x, y, (v64hi) { PERM5 (0, 64) });
195 }
196
197 /*
198 ** hf_trn1_s:
199 ** ptrue (p[0-7])\.b, vl256
200 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
201 ** trn1 (z[0-9]+)\.s, \2\.s, \2\.s
202 ** st1h \3\.s, \1, \[x8\]
203 ** ret
204 */
205 v64hf
206 hf_trn1_s (v64hf x)
207 {
208 return __builtin_shuffle (x, x, (v64hi) { PERM5 (0, 64) });
209 }
210
211 /*
212 ** hf_trn1_s_two_op:
213 ** ptrue (p[0-7])\.b, vl256
214 ** (
215 ** ld1h (z[0-9]+)\.s, \1/z, \[x1\]
216 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
217 ** trn1 \3\.s, \3\.s, \2\.s
218 ** st1h \3\.s, \1, \[x8\]
219 ** |
220 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
221 ** ld1h (z[0-9]+)\.s, \1/z, \[x1\]
222 ** trn1 \4\.s, \4\.s, \5\.s
223 ** st1h \4\.s, \1, \[x8\]
224 ** )
225 ** ret
226 */
227 v64hf
228 hf_trn1_s_two_op (v64hf x, v64hf y)
229 {
230 return __builtin_shuffle (x, y, (v64hi) { PERM5 (0, 64) });
231 }
232
233 /*
234 ** bf_trn1_s:
235 ** ptrue (p[0-7])\.b, vl256
236 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
237 ** trn1 (z[0-9]+)\.s, \2\.s, \2\.s
238 ** st1h \3\.s, \1, \[x8\]
239 ** ret
240 */
241 v64bf
242 bf_trn1_s (v64bf x)
243 {
244 return __builtin_shuffle (x, x, (v64hi) { PERM5 (0, 64) });
245 }
246
247 /*
248 ** bf_trn1_s_two_op:
249 ** ptrue (p[0-7])\.b, vl256
250 ** (
251 ** ld1h (z[0-9]+)\.s, \1/z, \[x1\]
252 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
253 ** trn1 \3\.s, \3\.s, \2\.s
254 ** st1h \3\.s, \1, \[x8\]
255 ** |
256 ** ld1h (z[0-9]+)\.s, \1/z, \[x0\]
257 ** ld1h (z[0-9]+)\.s, \1/z, \[x1\]
258 ** trn1 \4\.s, \4\.s, \5\.s
259 ** st1h \4\.s, \1, \[x8\]
260 ** )
261 ** ret
262 */
263 v64bf
264 bf_trn1_s_two_op (v64bf x, v64bf y)
265 {
266 return __builtin_shuffle (x, y, (v64hi) { PERM5 (0, 64) });
267 }
268
269 /*
270 ** hi_trn1_d:
271 ** ptrue (p[0-7])\.b, vl256
272 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
273 ** trn1 (z[0-9]+)\.d, \2\.d, \2\.d
274 ** st1h \3\.d, \1, \[x8\]
275 ** ret
276 */
277 v32hi
278 hi_trn1_d (v32hi x)
279 {
280 return __builtin_shuffle (x, x, (v32hi) { PERM4 (0, 32) });
281 }
282
283 /*
284 ** hi_trn1_d_two_op:
285 ** ptrue (p[0-7])\.b, vl256
286 ** (
287 ** ld1h (z[0-9]+)\.d, \1/z, \[x1\]
288 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
289 ** trn1 \3\.d, \3\.d, \2\.d
290 ** st1h \3\.d, \1, \[x8\]
291 ** |
292 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
293 ** ld1h (z[0-9]+)\.d, \1/z, \[x1\]
294 ** trn1 \4\.d, \4\.d, \5\.d
295 ** st1h \4\.d, \1, \[x8\]
296 ** )
297 ** ret
298 */
299 v32hi
300 hi_trn1_d_two_op (v32hi x, v32hi y)
301 {
302 return __builtin_shuffle (x, y, (v32hi) { PERM4 (0, 32) });
303 }
304
305 /*
306 ** hf_trn1_d:
307 ** ptrue (p[0-7])\.b, vl256
308 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
309 ** trn1 (z[0-9]+)\.d, \2\.d, \2\.d
310 ** st1h \3\.d, \1, \[x8\]
311 ** ret
312 */
313 v32hf
314 hf_trn1_d (v32hf x)
315 {
316 return __builtin_shuffle (x, x, (v32hi) { PERM4 (0, 32) });
317 }
318
319 /*
320 ** hf_trn1_d_two_op:
321 ** ptrue (p[0-7])\.b, vl256
322 ** (
323 ** ld1h (z[0-9]+)\.d, \1/z, \[x1\]
324 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
325 ** trn1 \3\.d, \3\.d, \2\.d
326 ** st1h \3\.d, \1, \[x8\]
327 ** |
328 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
329 ** ld1h (z[0-9]+)\.d, \1/z, \[x1\]
330 ** trn1 \4\.d, \4\.d, \5\.d
331 ** st1h \4\.d, \1, \[x8\]
332 ** )
333 ** ret
334 */
335 v32hf
336 hf_trn1_d_two_op (v32hf x, v32hf y)
337 {
338 return __builtin_shuffle (x, y, (v32hi) { PERM4 (0, 32) });
339 }
340
341 /*
342 ** bf_trn1_d:
343 ** ptrue (p[0-7])\.b, vl256
344 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
345 ** trn1 (z[0-9]+)\.d, \2\.d, \2\.d
346 ** st1h \3\.d, \1, \[x8\]
347 ** ret
348 */
349 v32bf
350 bf_trn1_d (v32bf x)
351 {
352 return __builtin_shuffle (x, x, (v32hi) { PERM4 (0, 32) });
353 }
354
355 /*
356 ** bf_trn1_d_two_op:
357 ** ptrue (p[0-7])\.b, vl256
358 ** (
359 ** ld1h (z[0-9]+)\.d, \1/z, \[x1\]
360 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
361 ** trn1 \3\.d, \3\.d, \2\.d
362 ** st1h \3\.d, \1, \[x8\]
363 ** |
364 ** ld1h (z[0-9]+)\.d, \1/z, \[x0\]
365 ** ld1h (z[0-9]+)\.d, \1/z, \[x1\]
366 ** trn1 \4\.d, \4\.d, \5\.d
367 ** st1h \4\.d, \1, \[x8\]
368 ** )
369 ** ret
370 */
371 v32bf
372 bf_trn1_d_two_op (v32bf x, v32bf y)
373 {
374 return __builtin_shuffle (x, y, (v32hi) { PERM4 (0, 32) });
375 }
376
377 /*
378 ** si_trn1_d:
379 ** ptrue (p[0-7])\.b, vl256
380 ** ld1w (z[0-9]+)\.d, \1/z, \[x0\]
381 ** trn1 (z[0-9]+)\.d, \2\.d, \2\.d
382 ** st1w \3\.d, \1, \[x8\]
383 ** ret
384 */
385 v32si
386 si_trn1_d (v32si x)
387 {
388 return __builtin_shuffle (x, x, (v32si) { PERM4 (0, 32) });
389 }
390
391 /*
392 ** sf_trn1_d:
393 ** ptrue (p[0-7])\.b, vl256
394 ** ld1w (z[0-9]+)\.d, \1/z, \[x0\]
395 ** trn1 (z[0-9]+)\.d, \2\.d, \2\.d
396 ** st1w \3\.d, \1, \[x8\]
397 ** ret
398 */
399 v32sf
400 sf_trn1_d (v32sf x)
401 {
402 return __builtin_shuffle (x, x, (v32si) { PERM4 (0, 32) });
403 }