1 /*
2 * Copyright (c) 2020-2021 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24 #ifndef ARM_COMPUTE_SVEMATH_H
25 #define ARM_COMPUTE_SVEMATH_H
26
27 #if defined(ARM_COMPUTE_ENABLE_SVE)
28 #include "src/core/NEON/wrapper/intrinsics/svcvt.h"
29 #include "src/core/NEON/wrapper/intrinsics/svdup_n.h"
30 #include "src/core/NEON/wrapper/intrinsics/svreinterpret.h"
31 #include <arm_sve.h>
32 #include <array>
33
34 namespace arm_compute
35 {
36 /** Calculate exponent.
37 *
38 * @param[in] pg Input predicate.
39 * @param[in] val Input vector value in F32 format.
40 *
41 * @return The calculated exponent.
42 */
43 svfloat32_t svexp_f32_z(svbool_t pg, svfloat32_t val);
44
45 /** Calculate reciprocal.
46 *
47 * @param[in] pg Input predicate.
48 * @param[in] x Input value.
49 *
50 * @return The calculated reciprocal.
51 */
52 svfloat32_t svinv_f32_z(svbool_t pg, svfloat32_t x);
53
54 /** Calculate logarithm
55 *
56 * @param[in] pg Input predicate.
57 * @param[in] x Input vector value in F32 format.
58 *
59 * @return The calculated logarithm.
60 */
61 svfloat32_t svlog_f32_z(svbool_t pg, svfloat32_t x);
62
63 /** Calculate hyperbolic tangent.
64 *
65 * tanh(x) = (e^2x - 1)/(e^2x + 1)
66 *
67 * @note We clamp x to [-5,5] to avoid overflowing issues.
68 *
69 * @param[in] pg Input predicate.
70 * @param[in] val Input vector value in F32 format.
71 *
72 * @return The calculated Hyperbolic Tangent.
73 */
74 svfloat32_t svtanh_f32_z(svbool_t pg, svfloat32_t val);
75
76 /** Calculate hyperbolic tangent.
77 *
78 * tanh(x) = (e^2x - 1)/(e^2x + 1)
79 *
80 * @note We clamp x to [-5,5] to avoid overflowing issues.
81 *
82 * @param[in] pg Input predicate.
83 * @param[in] val Input vector value in F16 format.
84 *
85 * @return The calculated Hyperbolic Tangent.
86 */
87 svfloat16_t svtanh_f16_z(svbool_t pg, svfloat16_t val);
88
89 /** Calculate exponential
90 *
91 * @param[in] pg Input predicate.
92 * @param[in] x Input vector value in F16 format.
93 *
94 * @return The calculated exponent.
95 */
96 svfloat16_t svexp_f16_z(svbool_t pg, svfloat16_t x);
97
98 /** Calculate reciprocal.
99 *
100 * @param[in] pg Input predicate.
101 * @param[in] x Input value.
102 *
103 * @return The calculated reciprocal.
104 */
105 svfloat16_t svinv_f16_z(svbool_t pg, svfloat16_t x);
106
107 /** Calculate logarithm
108 *
109 * @param[in] pg Input predicate.
110 * @param[in] x Input vector value in F32 format.
111 *
112 * @return The calculated logarithm.
113 */
114 svfloat16_t svlog_f16_z(svbool_t pg, svfloat16_t x);
115
116 /** Calculate inverse square root.
117 *
118 * @param[in] pg Input predicate.
119 * @param[in] val Input value.
120 *
121 * @return The calculated inverse square root.
122 */
123 template <typename VectorType>
svinvsqrt(svbool_t pg,VectorType val)124 inline VectorType svinvsqrt(svbool_t pg, VectorType val)
125 {
126 auto sqrt_reciprocal = svrsqrte(val);
127 sqrt_reciprocal = svmul_z(pg, svrsqrts(svmul_z(pg, val, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);
128 sqrt_reciprocal = svmul_z(pg, svrsqrts(svmul_z(pg, val, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);
129 return sqrt_reciprocal;
130 }
131
132 /** Calculate sine.
133 *
134 * @param[in] pg Input predicate.
135 * @param[in] val Input vector value in radians, F32 format.
136 *
137 * @return The calculated sine.
138 */
139 svfloat32_t svsin_f32_z(svbool_t pg, svfloat32_t val);
140
141 /** Calculate sine.
142 *
143 * @param[in] pg Input predicate.
144 * @param[in] val Input vector value in radians, F16 format.
145 *
146 * @return The calculated sine.
147 */
148 svfloat16_t svsin_f16_z(svbool_t pg, svfloat16_t val);
149
150 /** Calculate n power of a number.
151 *
152 * pow(x,n) = e^(n*log(x))
153 *
154 * @param[in] pg Input predicate.
155 * @param[in] a Input vector value in F32 format.
156 * @param[in] b Powers to raise the input to.
157 *
158 * @return The calculated power.
159 */
160 svfloat32_t svpow_f32_z(svbool_t pg, svfloat32_t a, svfloat32_t b);
161
162 /** Calculate n power of a number.
163 *
164 * pow(x,n) = e^(n*log(x))
165 *
166 * @param[in] pg Input predicate.
167 * @param[in] a Input vector value in F16 format.
168 * @param[in] b Powers to raise the input to.
169 *
170 * @return The calculated power.
171 */
172 svfloat16_t svpow_f16_z(svbool_t pg, svfloat16_t a, svfloat16_t b);
173
174 /** Convert and pack four 32-bit float vectors into an 8-bit integer vector
175 *
176 * @param[in] in_0 The first float vector
177 * @param[in] in_1 The second float vector
178 * @param[in] in_2 The third float vector
179 * @param[in] in_3 The fourth float vector
180 *
181 * @return The converted integer vector
182 */
183 template <typename int_vec_type>
184 int_vec_type convert_float_to_int(const svfloat32_t &in_0, const svfloat32_t &in_1, const svfloat32_t &in_2, const svfloat32_t &in_3);
185
186 } // namespace arm_compute
187 #include "src/core/NEON/SVEMath.inl"
188 #endif /* defined(ARM_COMPUTE_ENABLE_SVE) */
189 #endif /* ARM_COMPUTE_SVEMATH_H */