Debian Codegraph @

gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/ld4q_bf16.c

raw
1 /* { dg-do assemble { target aarch64_asm_sve2p1_ok } } */
2 /* { dg-do compile { target { ! aarch64_asm_sve2p1_ok } } } */
3 /* { dg-skip-if "" { *-*-* } { "-DSTREAMING_COMPATIBLE" } { "" } } */
4 /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */
5
6 #include "test_sve_acle.h"
7
8 #pragma GCC target "+sve2p1"
9
10 /*
11 ** ld4q_bf16_base:
12 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
13 ** ret
14 */
15 TEST_LOAD (ld4q_bf16_base, svbfloat16x4_t, bfloat16_t,
16 z0 = svld4q_bf16 (p0, x0),
17 z0 = svld4q (p0, x0))
18
19 /*
20 ** ld4q_bf16_index:
21 ** add (x[0-9]), x0, x1, lsl #?1
22 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\]
23 ** ret
24 */
25 TEST_LOAD (ld4q_bf16_index, svbfloat16x4_t, bfloat16_t,
26 z0 = svld4q_bf16 (p0, x0 + x1),
27 z0 = svld4q (p0, x0 + x1))
28
29 /*
30 ** ld4q_bf16_index2:
31 ** add (x[0-9]), x0, x1, lsl #?2
32 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\]
33 ** ret
34 */
35 TEST_LOAD (ld4q_bf16_index2, svbfloat16x4_t, bfloat16_t,
36 z0 = svld4q_bf16 (p0, x0 + x1 * 2),
37 z0 = svld4q (p0, x0 + x1 * 2))
38
39 /*
40 ** ld4q_bf16_index4:
41 ** add (x[0-9]), x0, x1, lsl #?3
42 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\]
43 ** ret
44 */
45 TEST_LOAD (ld4q_bf16_index4, svbfloat16x4_t, bfloat16_t,
46 z0 = svld4q_bf16 (p0, x0 + x1 * 4),
47 z0 = svld4q (p0, x0 + x1 * 4))
48
49 /*
50 ** ld4q_bf16_index8:
51 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, x1, lsl #?4\]
52 ** ret
53 */
54 TEST_LOAD (ld4q_bf16_index8, svbfloat16x4_t, bfloat16_t,
55 z0 = svld4q_bf16 (p0, x0 + x1 * 8),
56 z0 = svld4q (p0, x0 + x1 * 8))
57
58 /* Moving the constant into a register would also be OK. */
59 /*
60 ** ld4q_bf16_1:
61 ** incb x0
62 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
63 ** ret
64 */
65 TEST_LOAD (ld4q_bf16_1, svbfloat16x4_t, bfloat16_t,
66 z0 = svld4q_bf16 (p0, x0 + svcnth ()),
67 z0 = svld4q (p0, x0 + svcnth ()))
68
69 /* Moving the constant into a register would also be OK. */
70 /*
71 ** ld4q_bf16_2:
72 ** incb x0, all, mul #2
73 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
74 ** ret
75 */
76 TEST_LOAD (ld4q_bf16_2, svbfloat16x4_t, bfloat16_t,
77 z0 = svld4q_bf16 (p0, x0 + svcnth () * 2),
78 z0 = svld4q (p0, x0 + svcnth () * 2))
79
80 /* Moving the constant into a register would also be OK. */
81 /*
82 ** ld4q_bf16_3:
83 ** incb x0, all, mul #3
84 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
85 ** ret
86 */
87 TEST_LOAD (ld4q_bf16_3, svbfloat16x4_t, bfloat16_t,
88 z0 = svld4q_bf16 (p0, x0 + svcnth () * 3),
89 z0 = svld4q (p0, x0 + svcnth () * 3))
90
91 /*
92 ** ld4q_bf16_4:
93 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #4, mul vl\]
94 ** ret
95 */
96 TEST_LOAD (ld4q_bf16_4, svbfloat16x4_t, bfloat16_t,
97 z0 = svld4q_bf16 (p0, x0 + svcnth () * 4),
98 z0 = svld4q (p0, x0 + svcnth () * 4))
99
100 /*
101 ** ld4q_bf16_28:
102 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #28, mul vl\]
103 ** ret
104 */
105 TEST_LOAD (ld4q_bf16_28, svbfloat16x4_t, bfloat16_t,
106 z0 = svld4q_bf16 (p0, x0 + svcnth () * 28),
107 z0 = svld4q (p0, x0 + svcnth () * 28))
108
109 /*
110 ** ld4q_bf16_32:
111 ** [^{]*
112 ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\]
113 ** ret
114 */
115 TEST_LOAD (ld4q_bf16_32, svbfloat16x4_t, bfloat16_t,
116 z0 = svld4q_bf16 (p0, x0 + svcnth () * 32),
117 z0 = svld4q (p0, x0 + svcnth () * 32))
118
119 /* Moving the constant into a register would also be OK. */
120 /*
121 ** ld4q_bf16_m1:
122 ** decb x0
123 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
124 ** ret
125 */
126 TEST_LOAD (ld4q_bf16_m1, svbfloat16x4_t, bfloat16_t,
127 z0 = svld4q_bf16 (p0, x0 - svcnth ()),
128 z0 = svld4q (p0, x0 - svcnth ()))
129
130 /* Moving the constant into a register would also be OK. */
131 /*
132 ** ld4q_bf16_m2:
133 ** decb x0, all, mul #2
134 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
135 ** ret
136 */
137 TEST_LOAD (ld4q_bf16_m2, svbfloat16x4_t, bfloat16_t,
138 z0 = svld4q_bf16 (p0, x0 - svcnth () * 2),
139 z0 = svld4q (p0, x0 - svcnth () * 2))
140
141 /* Moving the constant into a register would also be OK. */
142 /*
143 ** ld4q_bf16_m3:
144 ** decb x0, all, mul #3
145 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
146 ** ret
147 */
148 TEST_LOAD (ld4q_bf16_m3, svbfloat16x4_t, bfloat16_t,
149 z0 = svld4q_bf16 (p0, x0 - svcnth () * 3),
150 z0 = svld4q (p0, x0 - svcnth () * 3))
151
152 /*
153 ** ld4q_bf16_m4:
154 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-4, mul vl\]
155 ** ret
156 */
157 TEST_LOAD (ld4q_bf16_m4, svbfloat16x4_t, bfloat16_t,
158 z0 = svld4q_bf16 (p0, x0 - svcnth () * 4),
159 z0 = svld4q (p0, x0 - svcnth () * 4))
160
161 /*
162 ** ld4q_bf16_m32:
163 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-32, mul vl\]
164 ** ret
165 */
166 TEST_LOAD (ld4q_bf16_m32, svbfloat16x4_t, bfloat16_t,
167 z0 = svld4q_bf16 (p0, x0 - svcnth () * 32),
168 z0 = svld4q (p0, x0 - svcnth () * 32))
169
170 /*
171 ** ld4q_bf16_m36:
172 ** [^{]*
173 ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\]
174 ** ret
175 */
176 TEST_LOAD (ld4q_bf16_m36, svbfloat16x4_t, bfloat16_t,
177 z0 = svld4q_bf16 (p0, x0 - svcnth () * 36),
178 z0 = svld4q (p0, x0 - svcnth () * 36))
179
180 /*
181 ** ld4q_vnum_bf16_0:
182 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
183 ** ret
184 */
185 TEST_LOAD (ld4q_vnum_bf16_0, svbfloat16x4_t, bfloat16_t,
186 z0 = svld4q_vnum_bf16 (p0, x0, 0),
187 z0 = svld4q_vnum (p0, x0, 0))
188
189 /* Moving the constant into a register would also be OK. */
190 /*
191 ** ld4q_vnum_bf16_1:
192 ** incb x0
193 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
194 ** ret
195 */
196 TEST_LOAD (ld4q_vnum_bf16_1, svbfloat16x4_t, bfloat16_t,
197 z0 = svld4q_vnum_bf16 (p0, x0, 1),
198 z0 = svld4q_vnum (p0, x0, 1))
199
200 /* Moving the constant into a register would also be OK. */
201 /*
202 ** ld4q_vnum_bf16_2:
203 ** incb x0, all, mul #2
204 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
205 ** ret
206 */
207 TEST_LOAD (ld4q_vnum_bf16_2, svbfloat16x4_t, bfloat16_t,
208 z0 = svld4q_vnum_bf16 (p0, x0, 2),
209 z0 = svld4q_vnum (p0, x0, 2))
210
211 /* Moving the constant into a register would also be OK. */
212 /*
213 ** ld4q_vnum_bf16_3:
214 ** incb x0, all, mul #3
215 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
216 ** ret
217 */
218 TEST_LOAD (ld4q_vnum_bf16_3, svbfloat16x4_t, bfloat16_t,
219 z0 = svld4q_vnum_bf16 (p0, x0, 3),
220 z0 = svld4q_vnum (p0, x0, 3))
221
222 /*
223 ** ld4q_vnum_bf16_4:
224 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #4, mul vl\]
225 ** ret
226 */
227 TEST_LOAD (ld4q_vnum_bf16_4, svbfloat16x4_t, bfloat16_t,
228 z0 = svld4q_vnum_bf16 (p0, x0, 4),
229 z0 = svld4q_vnum (p0, x0, 4))
230
231 /*
232 ** ld4q_vnum_bf16_28:
233 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #28, mul vl\]
234 ** ret
235 */
236 TEST_LOAD (ld4q_vnum_bf16_28, svbfloat16x4_t, bfloat16_t,
237 z0 = svld4q_vnum_bf16 (p0, x0, 28),
238 z0 = svld4q_vnum (p0, x0, 28))
239
240 /*
241 ** ld4q_vnum_bf16_32:
242 ** [^{]*
243 ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\]
244 ** ret
245 */
246 TEST_LOAD (ld4q_vnum_bf16_32, svbfloat16x4_t, bfloat16_t,
247 z0 = svld4q_vnum_bf16 (p0, x0, 32),
248 z0 = svld4q_vnum (p0, x0, 32))
249
250 /* Moving the constant into a register would also be OK. */
251 /*
252 ** ld4q_vnum_bf16_m1:
253 ** decb x0
254 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
255 ** ret
256 */
257 TEST_LOAD (ld4q_vnum_bf16_m1, svbfloat16x4_t, bfloat16_t,
258 z0 = svld4q_vnum_bf16 (p0, x0, -1),
259 z0 = svld4q_vnum (p0, x0, -1))
260
261 /* Moving the constant into a register would also be OK. */
262 /*
263 ** ld4q_vnum_bf16_m2:
264 ** decb x0, all, mul #2
265 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
266 ** ret
267 */
268 TEST_LOAD (ld4q_vnum_bf16_m2, svbfloat16x4_t, bfloat16_t,
269 z0 = svld4q_vnum_bf16 (p0, x0, -2),
270 z0 = svld4q_vnum (p0, x0, -2))
271
272 /* Moving the constant into a register would also be OK. */
273 /*
274 ** ld4q_vnum_bf16_m3:
275 ** decb x0, all, mul #3
276 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\]
277 ** ret
278 */
279 TEST_LOAD (ld4q_vnum_bf16_m3, svbfloat16x4_t, bfloat16_t,
280 z0 = svld4q_vnum_bf16 (p0, x0, -3),
281 z0 = svld4q_vnum (p0, x0, -3))
282
283 /*
284 ** ld4q_vnum_bf16_m4:
285 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-4, mul vl\]
286 ** ret
287 */
288 TEST_LOAD (ld4q_vnum_bf16_m4, svbfloat16x4_t, bfloat16_t,
289 z0 = svld4q_vnum_bf16 (p0, x0, -4),
290 z0 = svld4q_vnum (p0, x0, -4))
291
292 /*
293 ** ld4q_vnum_bf16_m32:
294 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-32, mul vl\]
295 ** ret
296 */
297 TEST_LOAD (ld4q_vnum_bf16_m32, svbfloat16x4_t, bfloat16_t,
298 z0 = svld4q_vnum_bf16 (p0, x0, -32),
299 z0 = svld4q_vnum (p0, x0, -32))
300
301 /*
302 ** ld4q_vnum_bf16_m36:
303 ** [^{]*
304 ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\]
305 ** ret
306 */
307 TEST_LOAD (ld4q_vnum_bf16_m36, svbfloat16x4_t, bfloat16_t,
308 z0 = svld4q_vnum_bf16 (p0, x0, -36),
309 z0 = svld4q_vnum (p0, x0, -36))
310
311 /*
312 ** ld4q_vnum_bf16_x1:
313 ** cntb (x[0-9]+)
314 ** (
315 ** madd (x[0-9]+), (?:x1, \1|\1, x1), x0
316 ** ld4q {z0\.q - z3\.q}, p0/z, \[\2\]
317 ** |
318 ** mul (x[0-9]+), (?:x1, \1|\1, x1)
319 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, \3\]
320 ** )
321 ** ret
322 */
323 TEST_LOAD (ld4q_vnum_bf16_x1, svbfloat16x4_t, bfloat16_t,
324 z0 = svld4q_vnum_bf16 (p0, x0, x1),
325 z0 = svld4q_vnum (p0, x0, x1))