Debian Codegraph @

gcc/testsuite/gcc.target/aarch64/sve2/acle/asm/add_bf16.c

raw
1 /* { dg-do assemble { target aarch64_asm_sve-b16b16_ok } } */
2 /* { dg-do compile { target { ! aarch64_asm_sve-b16b16_ok } } } */
3 /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */
4
5 #include "test_sve_acle.h"
6
7 #pragma GCC target "+sve-b16b16"
8 #ifdef STREAMING_COMPATIBLE
9 #pragma GCC target "+sme2"
10 #endif
11
12 /*
13 ** add_bf16_m_tied1:
14 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
15 ** ret
16 */
17 TEST_UNIFORM_Z (add_bf16_m_tied1, svbfloat16_t,
18 z0 = svadd_bf16_m (p0, z0, z1),
19 z0 = svadd_m (p0, z0, z1))
20
21 /*
22 ** add_bf16_m_tied2:
23 ** mov (z[0-9]+)\.d, z0\.d
24 ** movprfx z0, z1
25 ** bfadd z0\.h, p0/m, z0\.h, \1\.h
26 ** ret
27 */
28 TEST_UNIFORM_Z (add_bf16_m_tied2, svbfloat16_t,
29 z0 = svadd_bf16_m (p0, z1, z0),
30 z0 = svadd_m (p0, z1, z0))
31
32 /*
33 ** add_bf16_m_untied:
34 ** movprfx z0, z1
35 ** bfadd z0\.h, p0/m, z0\.h, z2\.h
36 ** ret
37 */
38 TEST_UNIFORM_Z (add_bf16_m_untied, svbfloat16_t,
39 z0 = svadd_bf16_m (p0, z1, z2),
40 z0 = svadd_m (p0, z1, z2))
41
42 /*
43 ** add_h4_bf16_m_tied1:
44 ** mov (z[0-9]+\.h), h4
45 ** bfadd z0\.h, p0/m, z0\.h, \1
46 ** ret
47 */
48 TEST_UNIFORM_ZD (add_h4_bf16_m_tied1, svbfloat16_t, __bf16,
49 z0 = svadd_n_bf16_m (p0, z0, d4),
50 z0 = svadd_m (p0, z0, d4))
51
52 /*
53 ** add_h4_bf16_m_untied:
54 ** mov (z[0-9]+\.h), h4
55 ** movprfx z0, z1
56 ** bfadd z0\.h, p0/m, z0\.h, \1
57 ** ret
58 */
59 TEST_UNIFORM_ZD (add_h4_bf16_m_untied, svbfloat16_t, __bf16,
60 z0 = svadd_n_bf16_m (p0, z1, d4),
61 z0 = svadd_m (p0, z1, d4))
62
63 /*
64 ** add_1_bf16_m:
65 ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)?
66 ** bfadd z0\.h, p0/m, z0\.h, \1
67 ** ret
68 */
69 TEST_UNIFORM_Z (add_1_bf16_m, svbfloat16_t,
70 z0 = svadd_n_bf16_m (p0, z0, 1),
71 z0 = svadd_m (p0, z0, 1))
72
73 /*
74 ** add_0p5_bf16_m:
75 ** fmov (z[0-9]+\.h), #1\.75(?:e\+0)?
76 ** bfadd z0\.h, p0/m, z0\.h, \1
77 ** ret
78 */
79 TEST_UNIFORM_Z (add_0p5_bf16_m, svbfloat16_t,
80 z0 = svadd_n_bf16_m (p0, z0, (bfloat16_t) (0.5)),
81 z0 = svadd_m (p0, z0, (bfloat16_t) (0.5)))
82
83 /*
84 ** add_m1_bf16_m:
85 ** fmov (z[0-9]+\.h), #-1\.875(?:e\+0)?
86 ** bfadd z0\.h, p0/m, z0\.h, \1
87 ** ret
88 */
89 TEST_UNIFORM_Z (add_m1_bf16_m, svbfloat16_t,
90 z0 = svadd_n_bf16_m (p0, z0, -1),
91 z0 = svadd_m (p0, z0, -1))
92
93 /*
94 ** add_m0p5_bf16_m:
95 ** fmov (z[0-9]+\.h), #-1\.75(?:e\+0)?
96 ** bfadd z0\.h, p0/m, z0\.h, \1
97 ** ret
98 */
99 TEST_UNIFORM_Z (add_m0p5_bf16_m, svbfloat16_t,
100 z0 = svadd_n_bf16_m (p0, z0, (bfloat16_t) (-0.5)),
101 z0 = svadd_m (p0, z0, (bfloat16_t) (-0.5)))
102
103 /*
104 ** add_m2_bf16_m_tied1:
105 ** fmov (z[0-9]+\.h), #-2\.0(?:e\+0)?
106 ** bfadd z0\.h, p0/m, z0\.h, \1
107 ** ret
108 */
109 TEST_UNIFORM_Z (add_m2_bf16_m_tied1, svbfloat16_t,
110 z0 = svadd_n_bf16_m (p0, z0, -2),
111 z0 = svadd_m (p0, z0, -2))
112
113 /*
114 ** add_m2_bf16_m_untied:
115 ** fmov (z[0-9]+\.h), #-2\.0(?:e\+0)?
116 ** movprfx z0, z1
117 ** bfadd z0\.h, p0/m, z0\.h, \1
118 ** ret
119 */
120 TEST_UNIFORM_Z (add_m2_bf16_m_untied, svbfloat16_t,
121 z0 = svadd_n_bf16_m (p0, z1, -2),
122 z0 = svadd_m (p0, z1, -2))
123
124 /*
125 ** add_bf16_z_tied1:
126 ** movprfx z0\.h, p0/z, z0\.h
127 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
128 ** ret
129 */
130 TEST_UNIFORM_Z (add_bf16_z_tied1, svbfloat16_t,
131 z0 = svadd_bf16_z (p0, z0, z1),
132 z0 = svadd_z (p0, z0, z1))
133
134 /*
135 ** add_bf16_z_tied2:
136 ** movprfx z0\.h, p0/z, z0\.h
137 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
138 ** ret
139 */
140 TEST_UNIFORM_Z (add_bf16_z_tied2, svbfloat16_t,
141 z0 = svadd_bf16_z (p0, z1, z0),
142 z0 = svadd_z (p0, z1, z0))
143
144 /*
145 ** add_bf16_z_untied:
146 ** (
147 ** movprfx z0\.h, p0/z, z1\.h
148 ** bfadd z0\.h, p0/m, z0\.h, z2\.h
149 ** |
150 ** movprfx z0\.h, p0/z, z2\.h
151 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
152 ** )
153 ** ret
154 */
155 TEST_UNIFORM_Z (add_bf16_z_untied, svbfloat16_t,
156 z0 = svadd_bf16_z (p0, z1, z2),
157 z0 = svadd_z (p0, z1, z2))
158
159 /*
160 ** add_h4_bf16_z_tied1:
161 ** mov (z[0-9]+\.h), h4
162 ** movprfx z0\.h, p0/z, z0\.h
163 ** bfadd z0\.h, p0/m, z0\.h, \1
164 ** ret
165 */
166 TEST_UNIFORM_ZD (add_h4_bf16_z_tied1, svbfloat16_t, __bf16,
167 z0 = svadd_n_bf16_z (p0, z0, d4),
168 z0 = svadd_z (p0, z0, d4))
169
170 /*
171 ** add_h4_bf16_z_untied:
172 ** mov (z[0-9]+\.h), h4
173 ** (
174 ** movprfx z0\.h, p0/z, z1\.h
175 ** bfadd z0\.h, p0/m, z0\.h, \1
176 ** |
177 ** movprfx z0\.h, p0/z, \1
178 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
179 ** )
180 ** ret
181 */
182 TEST_UNIFORM_ZD (add_h4_bf16_z_untied, svbfloat16_t, __bf16,
183 z0 = svadd_n_bf16_z (p0, z1, d4),
184 z0 = svadd_z (p0, z1, d4))
185
186 /*
187 ** add_1_bf16_z:
188 ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)?
189 ** movprfx z0\.h, p0/z, z0\.h
190 ** bfadd z0\.h, p0/m, z0\.h, \1
191 ** ret
192 */
193 TEST_UNIFORM_Z (add_1_bf16_z, svbfloat16_t,
194 z0 = svadd_n_bf16_z (p0, z0, 1),
195 z0 = svadd_z (p0, z0, 1))
196
197 /*
198 ** add_bf16_x_tied1:
199 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
200 ** ret
201 */
202 TEST_UNIFORM_Z (add_bf16_x_tied1, svbfloat16_t,
203 z0 = svadd_bf16_x (p0, z0, z1),
204 z0 = svadd_x (p0, z0, z1))
205
206 /*
207 ** add_bf16_x_tied2:
208 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
209 ** ret
210 */
211 TEST_UNIFORM_Z (add_bf16_x_tied2, svbfloat16_t,
212 z0 = svadd_bf16_x (p0, z1, z0),
213 z0 = svadd_x (p0, z1, z0))
214
215 /*
216 ** add_bf16_x_untied:
217 ** (
218 ** movprfx z0, z1
219 ** bfadd z0\.h, p0/m, z0\.h, z2\.h
220 ** |
221 ** movprfx z0, z2
222 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
223 ** )
224 ** ret
225 */
226 TEST_UNIFORM_Z (add_bf16_x_untied, svbfloat16_t,
227 z0 = svadd_bf16_x (p0, z1, z2),
228 z0 = svadd_x (p0, z1, z2))
229
230 /*
231 ** add_h4_bf16_x_tied1:
232 ** mov (z[0-9]+\.h), h4
233 ** bfadd z0\.h, p0/m, z0\.h, \1
234 ** ret
235 */
236 TEST_UNIFORM_ZD (add_h4_bf16_x_tied1, svbfloat16_t, __bf16,
237 z0 = svadd_n_bf16_x (p0, z0, d4),
238 z0 = svadd_x (p0, z0, d4))
239
240 /*
241 ** add_h4_bf16_x_untied:
242 ** mov z0\.h, h4
243 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
244 ** ret
245 */
246 TEST_UNIFORM_ZD (add_h4_bf16_x_untied, svbfloat16_t, __bf16,
247 z0 = svadd_n_bf16_x (p0, z1, d4),
248 z0 = svadd_x (p0, z1, d4))
249
250 /*
251 ** add_1_bf16_x_tied1:
252 ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)?
253 ** bfadd z0\.h, p0/m, z0\.h, \1
254 ** ret
255 */
256 TEST_UNIFORM_Z (add_1_bf16_x_tied1, svbfloat16_t,
257 z0 = svadd_n_bf16_x (p0, z0, 1),
258 z0 = svadd_x (p0, z0, 1))
259
260 /*
261 ** add_1_bf16_x_untied:
262 ** fmov z0\.h, #1\.875(?:e\+0)?
263 ** bfadd z0\.h, p0/m, z0\.h, z1\.h
264 ** ret
265 */
266 TEST_UNIFORM_Z (add_1_bf16_x_untied, svbfloat16_t,
267 z0 = svadd_n_bf16_x (p0, z1, 1),
268 z0 = svadd_x (p0, z1, 1))
269
270 /*
271 ** ptrue_add_bf16_x_tied1:
272 ** bfadd z0\.h, (z0\.h, z1\.h|z1\.h, z0\.h)
273 ** ret
274 */
275 TEST_UNIFORM_Z (ptrue_add_bf16_x_tied1, svbfloat16_t,
276 z0 = svadd_bf16_x (svptrue_b16 (), z0, z1),
277 z0 = svadd_x (svptrue_b16 (), z0, z1))
278
279 /*
280 ** ptrue_add_bf16_x_tied2:
281 ** bfadd z0\.h, (z0\.h, z1\.h|z1\.h, z0\.h)
282 ** ret
283 */
284 TEST_UNIFORM_Z (ptrue_add_bf16_x_tied2, svbfloat16_t,
285 z0 = svadd_bf16_x (svptrue_b16 (), z1, z0),
286 z0 = svadd_x (svptrue_b16 (), z1, z0))
287
288 /*
289 ** ptrue_add_bf16_x_untied:
290 ** bfadd z0\.h, (z1\.h, z2\.h|z2\.h, z1\.h)
291 ** ret
292 */
293 TEST_UNIFORM_Z (ptrue_add_bf16_x_untied, svbfloat16_t,
294 z0 = svadd_bf16_x (svptrue_b16 (), z1, z2),
295 z0 = svadd_x (svptrue_b16 (), z1, z2))
296
297 /*
298 ** ptrue_add_1_bf16_x_tied1:
299 ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)?
300 ** bfadd z0\.h, (z0\.h, \1|\1, z0\.h)
301 ** ret
302 */
303 TEST_UNIFORM_Z (ptrue_add_1_bf16_x_tied1, svbfloat16_t,
304 z0 = svadd_n_bf16_x (svptrue_b16 (), z0, 1),
305 z0 = svadd_x (svptrue_b16 (), z0, 1))
306
307 /*
308 ** ptrue_add_1_bf16_x_untied:
309 ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)?
310 ** bfadd z0\.h, (z1\.h, \1|\1, z1\.h)
311 ** ret
312 */
313 TEST_UNIFORM_Z (ptrue_add_1_bf16_x_untied, svbfloat16_t,
314 z0 = svadd_n_bf16_x (svptrue_b16 (), z1, 1),
315 z0 = svadd_x (svptrue_b16 (), z1, 1))