Debian Codegraph @

gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/ldnt1_bf16_x2.c

raw
1 /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" { target { ! ilp32 } } } } */
2
3 #include "test_sme2_acle.h"
4
5 /*
6 ** ldnt1_bf16_base:
7 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
8 ** ret
9 */
10 TEST_LOAD_COUNT (ldnt1_bf16_base, svbfloat16x2_t, bfloat16_t,
11 z0 = svldnt1_bf16_x2 (pn8, x0),
12 z0 = svldnt1_x2 (pn8, x0))
13
14 /*
15 ** ldnt1_bf16_index:
16 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, x1, lsl #?1\]
17 ** ret
18 */
19 TEST_LOAD_COUNT (ldnt1_bf16_index, svbfloat16x2_t, bfloat16_t,
20 z0 = svldnt1_bf16_x2 (pn8, x0 + x1),
21 z0 = svldnt1_x2 (pn8, x0 + x1))
22
23 /* Moving the constant into a register would also be OK. */
24 /*
25 ** ldnt1_bf16_1:
26 ** incb x0
27 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
28 ** ret
29 */
30 TEST_LOAD_COUNT (ldnt1_bf16_1, svbfloat16x2_t, bfloat16_t,
31 z0 = svldnt1_bf16_x2 (pn8, x0 + svcnth ()),
32 z0 = svldnt1_x2 (pn8, x0 + svcnth ()))
33
34 /*
35 ** ldnt1_bf16_2:
36 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #2, mul vl\]
37 ** ret
38 */
39 TEST_LOAD_COUNT (ldnt1_bf16_2, svbfloat16x2_t, bfloat16_t,
40 z0 = svldnt1_bf16_x2 (pn8, x0 + svcnth () * 2),
41 z0 = svldnt1_x2 (pn8, x0 + svcnth () * 2))
42
43 /*
44 ** ldnt1_bf16_14:
45 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #14, mul vl\]
46 ** ret
47 */
48 TEST_LOAD_COUNT (ldnt1_bf16_14, svbfloat16x2_t, bfloat16_t,
49 z0 = svldnt1_bf16_x2 (pn8, x0 + svcnth () * 14),
50 z0 = svldnt1_x2 (pn8, x0 + svcnth () * 14))
51
52 /* Moving the constant into a register would also be OK. */
53 /*
54 ** ldnt1_bf16_16:
55 ** incb x0, all, mul #16
56 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
57 ** ret
58 */
59 TEST_LOAD_COUNT (ldnt1_bf16_16, svbfloat16x2_t, bfloat16_t,
60 z0 = svldnt1_bf16_x2 (pn8, x0 + svcnth () * 16),
61 z0 = svldnt1_x2 (pn8, x0 + svcnth () * 16))
62
63 /* Moving the constant into a register would also be OK. */
64 /*
65 ** ldnt1_bf16_m1:
66 ** decb x0
67 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
68 ** ret
69 */
70 TEST_LOAD_COUNT (ldnt1_bf16_m1, svbfloat16x2_t, bfloat16_t,
71 z0 = svldnt1_bf16_x2 (pn8, x0 - svcnth ()),
72 z0 = svldnt1_x2 (pn8, x0 - svcnth ()))
73
74 /*
75 ** ldnt1_bf16_m2:
76 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #-2, mul vl\]
77 ** ret
78 */
79 TEST_LOAD_COUNT (ldnt1_bf16_m2, svbfloat16x2_t, bfloat16_t,
80 z0 = svldnt1_bf16_x2 (pn8, x0 - svcnth () * 2),
81 z0 = svldnt1_x2 (pn8, x0 - svcnth () * 2))
82
83 /*
84 ** ldnt1_bf16_m16:
85 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #-16, mul vl\]
86 ** ret
87 */
88 TEST_LOAD_COUNT (ldnt1_bf16_m16, svbfloat16x2_t, bfloat16_t,
89 z0 = svldnt1_bf16_x2 (pn8, x0 - svcnth () * 16),
90 z0 = svldnt1_x2 (pn8, x0 - svcnth () * 16))
91
92 /*
93 ** ldnt1_bf16_m18:
94 ** addvl (x[0-9]+), x0, #-18
95 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[\1\]
96 ** ret
97 */
98 TEST_LOAD_COUNT (ldnt1_bf16_m18, svbfloat16x2_t, bfloat16_t,
99 z0 = svldnt1_bf16_x2 (pn8, x0 - svcnth () * 18),
100 z0 = svldnt1_x2 (pn8, x0 - svcnth () * 18))
101
102 /*
103 ** ldnt1_bf16_z17:
104 ** ldnt1h {z[^\n]+}, pn8/z, \[x0\]
105 ** mov [^\n]+
106 ** mov [^\n]+
107 ** ret
108 */
109 TEST_LOAD_COUNT (ldnt1_bf16_z17, svbfloat16x2_t, bfloat16_t,
110 z17 = svldnt1_bf16_x2 (pn8, x0),
111 z17 = svldnt1_x2 (pn8, x0))
112
113 /*
114 ** ldnt1_bf16_z22:
115 ** ldnt1h {z22\.h(?: - |, )z23\.h}, pn8/z, \[x0\]
116 ** ret
117 */
118 TEST_LOAD_COUNT (ldnt1_bf16_z22, svbfloat16x2_t, bfloat16_t,
119 z22 = svldnt1_bf16_x2 (pn8, x0),
120 z22 = svldnt1_x2 (pn8, x0))
121
122 /*
123 ** ldnt1_bf16_z28:
124 ** ldnt1h {z28\.h(?: - |, )z29\.h}, pn8/z, \[x0\]
125 ** ret
126 */
127 TEST_LOAD_COUNT (ldnt1_bf16_z28, svbfloat16x2_t, bfloat16_t,
128 z28 = svldnt1_bf16_x2 (pn8, x0),
129 z28 = svldnt1_x2 (pn8, x0))
130
131 /*
132 ** ldnt1_bf16_pn0:
133 ** mov p([89]|1[0-5])\.b, p0\.b
134 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn\1/z, \[x0\]
135 ** ret
136 */
137 TEST_LOAD_COUNT (ldnt1_bf16_pn0, svbfloat16x2_t, bfloat16_t,
138 z0 = svldnt1_bf16_x2 (pn0, x0),
139 z0 = svldnt1_x2 (pn0, x0))
140
141 /*
142 ** ldnt1_bf16_pn7:
143 ** mov p([89]|1[0-5])\.b, p7\.b
144 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn\1/z, \[x0\]
145 ** ret
146 */
147 TEST_LOAD_COUNT (ldnt1_bf16_pn7, svbfloat16x2_t, bfloat16_t,
148 z0 = svldnt1_bf16_x2 (pn7, x0),
149 z0 = svldnt1_x2 (pn7, x0))
150
151 /*
152 ** ldnt1_bf16_pn15:
153 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn15/z, \[x0\]
154 ** ret
155 */
156 TEST_LOAD_COUNT (ldnt1_bf16_pn15, svbfloat16x2_t, bfloat16_t,
157 z0 = svldnt1_bf16_x2 (pn15, x0),
158 z0 = svldnt1_x2 (pn15, x0))
159
160 /*
161 ** ldnt1_vnum_bf16_0:
162 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
163 ** ret
164 */
165 TEST_LOAD_COUNT (ldnt1_vnum_bf16_0, svbfloat16x2_t, bfloat16_t,
166 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, 0),
167 z0 = svldnt1_vnum_x2 (pn8, x0, 0))
168
169 /* Moving the constant into a register would also be OK. */
170 /*
171 ** ldnt1_vnum_bf16_1:
172 ** incb x0
173 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
174 ** ret
175 */
176 TEST_LOAD_COUNT (ldnt1_vnum_bf16_1, svbfloat16x2_t, bfloat16_t,
177 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, 1),
178 z0 = svldnt1_vnum_x2 (pn8, x0, 1))
179
180 /*
181 ** ldnt1_vnum_bf16_2:
182 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #2, mul vl\]
183 ** ret
184 */
185 TEST_LOAD_COUNT (ldnt1_vnum_bf16_2, svbfloat16x2_t, bfloat16_t,
186 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, 2),
187 z0 = svldnt1_vnum_x2 (pn8, x0, 2))
188
189 /*
190 ** ldnt1_vnum_bf16_14:
191 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #14, mul vl\]
192 ** ret
193 */
194 TEST_LOAD_COUNT (ldnt1_vnum_bf16_14, svbfloat16x2_t, bfloat16_t,
195 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, 14),
196 z0 = svldnt1_vnum_x2 (pn8, x0, 14))
197
198 /* Moving the constant into a register would also be OK. */
199 /*
200 ** ldnt1_vnum_bf16_16:
201 ** incb x0, all, mul #16
202 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
203 ** ret
204 */
205 TEST_LOAD_COUNT (ldnt1_vnum_bf16_16, svbfloat16x2_t, bfloat16_t,
206 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, 16),
207 z0 = svldnt1_vnum_x2 (pn8, x0, 16))
208
209 /* Moving the constant into a register would also be OK. */
210 /*
211 ** ldnt1_vnum_bf16_m1:
212 ** decb x0
213 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0\]
214 ** ret
215 */
216 TEST_LOAD_COUNT (ldnt1_vnum_bf16_m1, svbfloat16x2_t, bfloat16_t,
217 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, -1),
218 z0 = svldnt1_vnum_x2 (pn8, x0, -1))
219
220 /*
221 ** ldnt1_vnum_bf16_m2:
222 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #-2, mul vl\]
223 ** ret
224 */
225 TEST_LOAD_COUNT (ldnt1_vnum_bf16_m2, svbfloat16x2_t, bfloat16_t,
226 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, -2),
227 z0 = svldnt1_vnum_x2 (pn8, x0, -2))
228
229 /*
230 ** ldnt1_vnum_bf16_m16:
231 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, #-16, mul vl\]
232 ** ret
233 */
234 TEST_LOAD_COUNT (ldnt1_vnum_bf16_m16, svbfloat16x2_t, bfloat16_t,
235 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, -16),
236 z0 = svldnt1_vnum_x2 (pn8, x0, -16))
237
238 /*
239 ** ldnt1_vnum_bf16_m18:
240 ** addvl (x[0-9]+), x0, #-18
241 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[\1\]
242 ** ret
243 */
244 TEST_LOAD_COUNT (ldnt1_vnum_bf16_m18, svbfloat16x2_t, bfloat16_t,
245 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, -18),
246 z0 = svldnt1_vnum_x2 (pn8, x0, -18))
247
248 /*
249 ** ldnt1_vnum_bf16_x1:
250 ** cntb (x[0-9]+)
251 ** (
252 ** madd (x[0-9]+), (?:x1, \1|\1, x1), x0
253 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[\2\]
254 ** |
255 ** mul (x[0-9]+), (?:x1, \1|\1, x1)
256 ** ldnt1h {z0\.h(?: - |, )z1\.h}, pn8/z, \[x0, \3\]
257 ** )
258 ** ret
259 */
260 TEST_LOAD_COUNT (ldnt1_vnum_bf16_x1, svbfloat16x2_t, bfloat16_t,
261 z0 = svldnt1_vnum_bf16_x2 (pn8, x0, x1),
262 z0 = svldnt1_vnum_x2 (pn8, x0, x1))