|
111
|
1 /* ARM FP16 intrinsics include file.
|
|
|
2
|
|
|
3 Copyright (C) 2016-2017 Free Software Foundation, Inc.
|
|
|
4 Contributed by ARM Ltd.
|
|
|
5
|
|
|
6 This file is part of GCC.
|
|
|
7
|
|
|
8 GCC is free software; you can redistribute it and/or modify it
|
|
|
9 under the terms of the GNU General Public License as published
|
|
|
10 by the Free Software Foundation; either version 3, or (at your
|
|
|
11 option) any later version.
|
|
|
12
|
|
|
13 GCC is distributed in the hope that it will be useful, but WITHOUT
|
|
|
14 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
|
15 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
|
16 License for more details.
|
|
|
17
|
|
|
18 Under Section 7 of GPL version 3, you are granted additional
|
|
|
19 permissions described in the GCC Runtime Library Exception, version
|
|
|
20 3.1, as published by the Free Software Foundation.
|
|
|
21
|
|
|
22 You should have received a copy of the GNU General Public License and
|
|
|
23 a copy of the GCC Runtime Library Exception along with this program;
|
|
|
24 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
|
25 <http://www.gnu.org/licenses/>. */
|
|
|
26
|
|
|
27 #ifndef _GCC_ARM_FP16_H
|
|
|
28 #define _GCC_ARM_FP16_H 1
|
|
|
29
|
|
|
30 #ifdef __cplusplus
|
|
|
31 extern "C" {
|
|
|
32 #endif
|
|
|
33
|
|
|
34 #include <stdint.h>
|
|
|
35
|
|
|
36 /* Intrinsics for FP16 instructions. */
|
|
|
37 #pragma GCC push_options
|
|
|
38 #pragma GCC target ("fpu=fp-armv8")
|
|
|
39
|
|
|
40 #if defined (__ARM_FEATURE_FP16_SCALAR_ARITHMETIC)
|
|
|
41
|
|
|
42 typedef __fp16 float16_t;
|
|
|
43
|
|
|
44 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
45 vabsh_f16 (float16_t __a)
|
|
|
46 {
|
|
|
47 return __builtin_neon_vabshf (__a);
|
|
|
48 }
|
|
|
49
|
|
|
50 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
51 vaddh_f16 (float16_t __a, float16_t __b)
|
|
|
52 {
|
|
|
53 return __a + __b;
|
|
|
54 }
|
|
|
55
|
|
|
56 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
|
|
|
57 vcvtah_s32_f16 (float16_t __a)
|
|
|
58 {
|
|
|
59 return __builtin_neon_vcvtahssi (__a);
|
|
|
60 }
|
|
|
61
|
|
|
62 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
|
|
|
63 vcvtah_u32_f16 (float16_t __a)
|
|
|
64 {
|
|
|
65 return __builtin_neon_vcvtahusi (__a);
|
|
|
66 }
|
|
|
67
|
|
|
68 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
69 vcvth_f16_s32 (int32_t __a)
|
|
|
70 {
|
|
|
71 return __builtin_neon_vcvthshf (__a);
|
|
|
72 }
|
|
|
73
|
|
|
74 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
75 vcvth_f16_u32 (uint32_t __a)
|
|
|
76 {
|
|
|
77 return __builtin_neon_vcvthuhf (__a);
|
|
|
78 }
|
|
|
79
|
|
|
80 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
81 vcvth_n_f16_s32 (int32_t __a, const int __b)
|
|
|
82 {
|
|
|
83 return __builtin_neon_vcvths_nhf (__a, __b);
|
|
|
84 }
|
|
|
85
|
|
|
86 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
87 vcvth_n_f16_u32 (uint32_t __a, const int __b)
|
|
|
88 {
|
|
|
89 return __builtin_neon_vcvthu_nhf ((int32_t)__a, __b);
|
|
|
90 }
|
|
|
91
|
|
|
92 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
|
|
|
93 vcvth_n_s32_f16 (float16_t __a, const int __b)
|
|
|
94 {
|
|
|
95 return __builtin_neon_vcvths_nsi (__a, __b);
|
|
|
96 }
|
|
|
97
|
|
|
98 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
|
|
|
99 vcvth_n_u32_f16 (float16_t __a, const int __b)
|
|
|
100 {
|
|
|
101 return (uint32_t)__builtin_neon_vcvthu_nsi (__a, __b);
|
|
|
102 }
|
|
|
103
|
|
|
104 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
|
|
|
105 vcvth_s32_f16 (float16_t __a)
|
|
|
106 {
|
|
|
107 return __builtin_neon_vcvthssi (__a);
|
|
|
108 }
|
|
|
109
|
|
|
110 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
|
|
|
111 vcvth_u32_f16 (float16_t __a)
|
|
|
112 {
|
|
|
113 return __builtin_neon_vcvthusi (__a);
|
|
|
114 }
|
|
|
115
|
|
|
116 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
|
|
|
117 vcvtmh_s32_f16 (float16_t __a)
|
|
|
118 {
|
|
|
119 return __builtin_neon_vcvtmhssi (__a);
|
|
|
120 }
|
|
|
121
|
|
|
122 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
|
|
|
123 vcvtmh_u32_f16 (float16_t __a)
|
|
|
124 {
|
|
|
125 return __builtin_neon_vcvtmhusi (__a);
|
|
|
126 }
|
|
|
127
|
|
|
128 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
|
|
|
129 vcvtnh_s32_f16 (float16_t __a)
|
|
|
130 {
|
|
|
131 return __builtin_neon_vcvtnhssi (__a);
|
|
|
132 }
|
|
|
133
|
|
|
134 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
|
|
|
135 vcvtnh_u32_f16 (float16_t __a)
|
|
|
136 {
|
|
|
137 return __builtin_neon_vcvtnhusi (__a);
|
|
|
138 }
|
|
|
139
|
|
|
140 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
|
|
|
141 vcvtph_s32_f16 (float16_t __a)
|
|
|
142 {
|
|
|
143 return __builtin_neon_vcvtphssi (__a);
|
|
|
144 }
|
|
|
145
|
|
|
146 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
|
|
|
147 vcvtph_u32_f16 (float16_t __a)
|
|
|
148 {
|
|
|
149 return __builtin_neon_vcvtphusi (__a);
|
|
|
150 }
|
|
|
151
|
|
|
152 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
153 vdivh_f16 (float16_t __a, float16_t __b)
|
|
|
154 {
|
|
|
155 return __a / __b;
|
|
|
156 }
|
|
|
157
|
|
|
158 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
159 vfmah_f16 (float16_t __a, float16_t __b, float16_t __c)
|
|
|
160 {
|
|
|
161 return __builtin_neon_vfmahf (__a, __b, __c);
|
|
|
162 }
|
|
|
163
|
|
|
164 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
165 vfmsh_f16 (float16_t __a, float16_t __b, float16_t __c)
|
|
|
166 {
|
|
|
167 return __builtin_neon_vfmshf (__a, __b, __c);
|
|
|
168 }
|
|
|
169
|
|
|
170 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
171 vmaxnmh_f16 (float16_t __a, float16_t __b)
|
|
|
172 {
|
|
|
173 return __builtin_neon_vmaxnmhf (__a, __b);
|
|
|
174 }
|
|
|
175
|
|
|
176 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
177 vminnmh_f16 (float16_t __a, float16_t __b)
|
|
|
178 {
|
|
|
179 return __builtin_neon_vminnmhf (__a, __b);
|
|
|
180 }
|
|
|
181
|
|
|
182 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
183 vmulh_f16 (float16_t __a, float16_t __b)
|
|
|
184 {
|
|
|
185 return __a * __b;
|
|
|
186 }
|
|
|
187
|
|
|
188 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
189 vnegh_f16 (float16_t __a)
|
|
|
190 {
|
|
|
191 return - __a;
|
|
|
192 }
|
|
|
193
|
|
|
194 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
195 vrndah_f16 (float16_t __a)
|
|
|
196 {
|
|
|
197 return __builtin_neon_vrndahf (__a);
|
|
|
198 }
|
|
|
199
|
|
|
200 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
201 vrndh_f16 (float16_t __a)
|
|
|
202 {
|
|
|
203 return __builtin_neon_vrndhf (__a);
|
|
|
204 }
|
|
|
205
|
|
|
206 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
207 vrndih_f16 (float16_t __a)
|
|
|
208 {
|
|
|
209 return __builtin_neon_vrndihf (__a);
|
|
|
210 }
|
|
|
211
|
|
|
212 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
213 vrndmh_f16 (float16_t __a)
|
|
|
214 {
|
|
|
215 return __builtin_neon_vrndmhf (__a);
|
|
|
216 }
|
|
|
217
|
|
|
218 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
219 vrndnh_f16 (float16_t __a)
|
|
|
220 {
|
|
|
221 return __builtin_neon_vrndnhf (__a);
|
|
|
222 }
|
|
|
223
|
|
|
224 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
225 vrndph_f16 (float16_t __a)
|
|
|
226 {
|
|
|
227 return __builtin_neon_vrndphf (__a);
|
|
|
228 }
|
|
|
229
|
|
|
230 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
231 vrndxh_f16 (float16_t __a)
|
|
|
232 {
|
|
|
233 return __builtin_neon_vrndxhf (__a);
|
|
|
234 }
|
|
|
235
|
|
|
236 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
237 vsqrth_f16 (float16_t __a)
|
|
|
238 {
|
|
|
239 return __builtin_neon_vsqrthf (__a);
|
|
|
240 }
|
|
|
241
|
|
|
242 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
|
|
|
243 vsubh_f16 (float16_t __a, float16_t __b)
|
|
|
244 {
|
|
|
245 return __a - __b;
|
|
|
246 }
|
|
|
247
|
|
|
248 #endif /* __ARM_FEATURE_FP16_SCALAR_ARITHMETIC */
|
|
|
249 #pragma GCC pop_options
|
|
|
250
|
|
|
251 #ifdef __cplusplus
|
|
|
252 }
|
|
|
253 #endif
|
|
|
254
|
|
|
255 #endif
|
