1 /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" { target { ! ilp32 } } } } */
2
3 #include "test_sve_acle.h"
4
5 /*
6 ** st4_bf16_base:
7 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
8 ** ret
9 */
10 TEST_STORE (st4_bf16_base, svbfloat16x4_t, bfloat16_t,
11 svst4_bf16 (p0, x0, z0),
12 svst4 (p0, x0, z0))
13
14 /*
15 ** st4_bf16_index:
16 ** st4h {z0\.h - z3\.h}, p0, \[x0, x1, lsl 1\]
17 ** ret
18 */
19 TEST_STORE (st4_bf16_index, svbfloat16x4_t, bfloat16_t,
20 svst4_bf16 (p0, x0 + x1, z0),
21 svst4 (p0, x0 + x1, z0))
22
23 /* Moving the constant into a register would also be OK. */
24 /*
25 ** st4_bf16_1:
26 ** incb x0
27 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
28 ** ret
29 */
30 TEST_STORE (st4_bf16_1, svbfloat16x4_t, bfloat16_t,
31 svst4_bf16 (p0, x0 + svcnth (), z0),
32 svst4 (p0, x0 + svcnth (), z0))
33
34 /* Moving the constant into a register would also be OK. */
35 /*
36 ** st4_bf16_2:
37 ** incb x0, all, mul #2
38 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
39 ** ret
40 */
41 TEST_STORE (st4_bf16_2, svbfloat16x4_t, bfloat16_t,
42 svst4_bf16 (p0, x0 + svcnth () * 2, z0),
43 svst4 (p0, x0 + svcnth () * 2, z0))
44
45 /* Moving the constant into a register would also be OK. */
46 /*
47 ** st4_bf16_3:
48 ** incb x0, all, mul #3
49 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
50 ** ret
51 */
52 TEST_STORE (st4_bf16_3, svbfloat16x4_t, bfloat16_t,
53 svst4_bf16 (p0, x0 + svcnth () * 3, z0),
54 svst4 (p0, x0 + svcnth () * 3, z0))
55
56 /*
57 ** st4_bf16_4:
58 ** st4h {z0\.h - z3\.h}, p0, \[x0, #4, mul vl\]
59 ** ret
60 */
61 TEST_STORE (st4_bf16_4, svbfloat16x4_t, bfloat16_t,
62 svst4_bf16 (p0, x0 + svcnth () * 4, z0),
63 svst4 (p0, x0 + svcnth () * 4, z0))
64
65 /*
66 ** st4_bf16_28:
67 ** st4h {z0\.h - z3\.h}, p0, \[x0, #28, mul vl\]
68 ** ret
69 */
70 TEST_STORE (st4_bf16_28, svbfloat16x4_t, bfloat16_t,
71 svst4_bf16 (p0, x0 + svcnth () * 28, z0),
72 svst4 (p0, x0 + svcnth () * 28, z0))
73
74 /*
75 ** st4_bf16_32:
76 ** [^{]*
77 ** st4h {z0\.h - z3\.h}, p0, \[x[0-9]+\]
78 ** ret
79 */
80 TEST_STORE (st4_bf16_32, svbfloat16x4_t, bfloat16_t,
81 svst4_bf16 (p0, x0 + svcnth () * 32, z0),
82 svst4 (p0, x0 + svcnth () * 32, z0))
83
84 /* Moving the constant into a register would also be OK. */
85 /*
86 ** st4_bf16_m1:
87 ** decb x0
88 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
89 ** ret
90 */
91 TEST_STORE (st4_bf16_m1, svbfloat16x4_t, bfloat16_t,
92 svst4_bf16 (p0, x0 - svcnth (), z0),
93 svst4 (p0, x0 - svcnth (), z0))
94
95 /* Moving the constant into a register would also be OK. */
96 /*
97 ** st4_bf16_m2:
98 ** decb x0, all, mul #2
99 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
100 ** ret
101 */
102 TEST_STORE (st4_bf16_m2, svbfloat16x4_t, bfloat16_t,
103 svst4_bf16 (p0, x0 - svcnth () * 2, z0),
104 svst4 (p0, x0 - svcnth () * 2, z0))
105
106 /* Moving the constant into a register would also be OK. */
107 /*
108 ** st4_bf16_m3:
109 ** decb x0, all, mul #3
110 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
111 ** ret
112 */
113 TEST_STORE (st4_bf16_m3, svbfloat16x4_t, bfloat16_t,
114 svst4_bf16 (p0, x0 - svcnth () * 3, z0),
115 svst4 (p0, x0 - svcnth () * 3, z0))
116
117 /*
118 ** st4_bf16_m4:
119 ** st4h {z0\.h - z3\.h}, p0, \[x0, #-4, mul vl\]
120 ** ret
121 */
122 TEST_STORE (st4_bf16_m4, svbfloat16x4_t, bfloat16_t,
123 svst4_bf16 (p0, x0 - svcnth () * 4, z0),
124 svst4 (p0, x0 - svcnth () * 4, z0))
125
126 /*
127 ** st4_bf16_m32:
128 ** st4h {z0\.h - z3\.h}, p0, \[x0, #-32, mul vl\]
129 ** ret
130 */
131 TEST_STORE (st4_bf16_m32, svbfloat16x4_t, bfloat16_t,
132 svst4_bf16 (p0, x0 - svcnth () * 32, z0),
133 svst4 (p0, x0 - svcnth () * 32, z0))
134
135 /*
136 ** st4_bf16_m36:
137 ** [^{]*
138 ** st4h {z0\.h - z3\.h}, p0, \[x[0-9]+\]
139 ** ret
140 */
141 TEST_STORE (st4_bf16_m36, svbfloat16x4_t, bfloat16_t,
142 svst4_bf16 (p0, x0 - svcnth () * 36, z0),
143 svst4 (p0, x0 - svcnth () * 36, z0))
144
145 /*
146 ** st4_vnum_bf16_0:
147 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
148 ** ret
149 */
150 TEST_STORE (st4_vnum_bf16_0, svbfloat16x4_t, bfloat16_t,
151 svst4_vnum_bf16 (p0, x0, 0, z0),
152 svst4_vnum (p0, x0, 0, z0))
153
154 /* Moving the constant into a register would also be OK. */
155 /*
156 ** st4_vnum_bf16_1:
157 ** incb x0
158 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
159 ** ret
160 */
161 TEST_STORE (st4_vnum_bf16_1, svbfloat16x4_t, bfloat16_t,
162 svst4_vnum_bf16 (p0, x0, 1, z0),
163 svst4_vnum (p0, x0, 1, z0))
164
165 /* Moving the constant into a register would also be OK. */
166 /*
167 ** st4_vnum_bf16_2:
168 ** incb x0, all, mul #2
169 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
170 ** ret
171 */
172 TEST_STORE (st4_vnum_bf16_2, svbfloat16x4_t, bfloat16_t,
173 svst4_vnum_bf16 (p0, x0, 2, z0),
174 svst4_vnum (p0, x0, 2, z0))
175
176 /* Moving the constant into a register would also be OK. */
177 /*
178 ** st4_vnum_bf16_3:
179 ** incb x0, all, mul #3
180 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
181 ** ret
182 */
183 TEST_STORE (st4_vnum_bf16_3, svbfloat16x4_t, bfloat16_t,
184 svst4_vnum_bf16 (p0, x0, 3, z0),
185 svst4_vnum (p0, x0, 3, z0))
186
187 /*
188 ** st4_vnum_bf16_4:
189 ** st4h {z0\.h - z3\.h}, p0, \[x0, #4, mul vl\]
190 ** ret
191 */
192 TEST_STORE (st4_vnum_bf16_4, svbfloat16x4_t, bfloat16_t,
193 svst4_vnum_bf16 (p0, x0, 4, z0),
194 svst4_vnum (p0, x0, 4, z0))
195
196 /*
197 ** st4_vnum_bf16_28:
198 ** st4h {z0\.h - z3\.h}, p0, \[x0, #28, mul vl\]
199 ** ret
200 */
201 TEST_STORE (st4_vnum_bf16_28, svbfloat16x4_t, bfloat16_t,
202 svst4_vnum_bf16 (p0, x0, 28, z0),
203 svst4_vnum (p0, x0, 28, z0))
204
205 /*
206 ** st4_vnum_bf16_32:
207 ** [^{]*
208 ** st4h {z0\.h - z3\.h}, p0, \[x[0-9]+\]
209 ** ret
210 */
211 TEST_STORE (st4_vnum_bf16_32, svbfloat16x4_t, bfloat16_t,
212 svst4_vnum_bf16 (p0, x0, 32, z0),
213 svst4_vnum (p0, x0, 32, z0))
214
215 /* Moving the constant into a register would also be OK. */
216 /*
217 ** st4_vnum_bf16_m1:
218 ** decb x0
219 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
220 ** ret
221 */
222 TEST_STORE (st4_vnum_bf16_m1, svbfloat16x4_t, bfloat16_t,
223 svst4_vnum_bf16 (p0, x0, -1, z0),
224 svst4_vnum (p0, x0, -1, z0))
225
226 /* Moving the constant into a register would also be OK. */
227 /*
228 ** st4_vnum_bf16_m2:
229 ** decb x0, all, mul #2
230 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
231 ** ret
232 */
233 TEST_STORE (st4_vnum_bf16_m2, svbfloat16x4_t, bfloat16_t,
234 svst4_vnum_bf16 (p0, x0, -2, z0),
235 svst4_vnum (p0, x0, -2, z0))
236
237 /* Moving the constant into a register would also be OK. */
238 /*
239 ** st4_vnum_bf16_m3:
240 ** decb x0, all, mul #3
241 ** st4h {z0\.h - z3\.h}, p0, \[x0\]
242 ** ret
243 */
244 TEST_STORE (st4_vnum_bf16_m3, svbfloat16x4_t, bfloat16_t,
245 svst4_vnum_bf16 (p0, x0, -3, z0),
246 svst4_vnum (p0, x0, -3, z0))
247
248 /*
249 ** st4_vnum_bf16_m4:
250 ** st4h {z0\.h - z3\.h}, p0, \[x0, #-4, mul vl\]
251 ** ret
252 */
253 TEST_STORE (st4_vnum_bf16_m4, svbfloat16x4_t, bfloat16_t,
254 svst4_vnum_bf16 (p0, x0, -4, z0),
255 svst4_vnum (p0, x0, -4, z0))
256
257 /*
258 ** st4_vnum_bf16_m32:
259 ** st4h {z0\.h - z3\.h}, p0, \[x0, #-32, mul vl\]
260 ** ret
261 */
262 TEST_STORE (st4_vnum_bf16_m32, svbfloat16x4_t, bfloat16_t,
263 svst4_vnum_bf16 (p0, x0, -32, z0),
264 svst4_vnum (p0, x0, -32, z0))
265
266 /*
267 ** st4_vnum_bf16_m36:
268 ** [^{]*
269 ** st4h {z0\.h - z3\.h}, p0, \[x[0-9]+\]
270 ** ret
271 */
272 TEST_STORE (st4_vnum_bf16_m36, svbfloat16x4_t, bfloat16_t,
273 svst4_vnum_bf16 (p0, x0, -36, z0),
274 svst4_vnum (p0, x0, -36, z0))
275
276 /* Using MUL to calculate an index would also be OK. */
277 /*
278 ** st4_vnum_bf16_x1:
279 ** cntb (x[0-9]+)
280 ** madd (x[0-9]+), (x1, \1|\1, x1), x0
281 ** st4h {z0\.h - z3\.h}, p0, \[\2\]
282 ** ret
283 */
284 TEST_STORE (st4_vnum_bf16_x1, svbfloat16x4_t, bfloat16_t,
285 svst4_vnum_bf16 (p0, x0, x1, z0),
286 svst4_vnum (p0, x0, x1, z0))