1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
|
/* ----------------------------------------------------------------------
* Project: CMSIS DSP Library
* Title: arm_vlog_f16.c
* Description: Fast vectorized log
*
* $Date: 23 April 2021
* $Revision: V1.9.0
*
* Target Processor: Cortex-M and Cortex-A cores
* -------------------------------------------------------------------- */
/*
* Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved.
*
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the License); you may
* not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an AS IS BASIS, WITHOUT
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "dsp/fast_math_functions_f16.h"
#include "dsp/support_functions_f16.h"
#if defined(ARM_FLOAT16_SUPPORTED)
/* Degree of the polynomial approximation */
#define NB_DEG_LOGF16 3
/*
Related to the Log2 of the number of approximations.
For instance, with 3 there are 1 + 2^3 polynomials
*/
#define NB_DIV_LOGF16 3
/* Length of the LUT table */
#define NB_LUT_LOGF16 (NB_DEG_LOGF16+1)*(1 + (1<<NB_DIV_LOGF16))
/*
LUT of polynomial approximations.
Could be generated with:
ClearAll[lut, coefs, nb, deg];
nb = 3;
deg = 3;
lut = Table[
MiniMaxApproximation[
Log[x/2^nb + i], {x, {10^-6, 1.0/2^nb}, deg, 0},
MaxIterations -> 1000][[2, 1]], {i, 1, 2, (1.0/2^nb)}];
coefs = Chop@Flatten[CoefficientList[lut, x]];
*/
static float16_t lut_logf16[NB_LUT_LOGF16]={
0,0.125,-0.00781197,0.00063974,0.117783,
0.111111,-0.00617212,0.000447935,0.223144,
0.1,-0.00499952,0.000327193,0.318454,0.0909091,
-0.00413191,0.000246234,0.405465,0.0833333,
-0.00347199,0.000189928,0.485508,0.0769231,
-0.00295841,0.00014956,0.559616,0.0714286,
-0.0025509,0.000119868,0.628609,0.0666667,
-0.00222213,0.0000975436,0.693147,
0.0625,-0.00195305,0.0000804357};
float16_t logf16_scalar(float16_t x)
{
int16_t i = arm_typecast_s16_f16(x);
int32_t vecExpUnBiased = (i >> 10) - 15;
i = i - (vecExpUnBiased << 10);
float16_t vecTmpFlt1 = arm_typecast_f16_s16(i);
float16_t *lut;
int n;
float16_t tmp,v;
tmp = ((_Float16)vecTmpFlt1 - 1.0f16) * (1 << NB_DIV_LOGF16);
n = (int)floor((double)tmp);
v = (_Float16)tmp - (_Float16)n;
lut = lut_logf16 + n * (1+NB_DEG_LOGF16);
float16_t res = lut[NB_DEG_LOGF16-1];
for(int j=NB_DEG_LOGF16-2; j >=0 ; j--)
{
res = (_Float16)lut[j] + (_Float16)v * (_Float16)res;
}
res = (_Float16)res + 0.693147f16 * (_Float16)vecExpUnBiased;
return(res);
}
#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
#include "arm_common_tables.h"
#include "arm_vec_math_f16.h"
float16x8_t vlogq_lut_f16(float16x8_t vecIn)
{
int16x8_t i = vreinterpretq_s16_f16(vecIn);
int16x8_t vecExpUnBiased = vsubq_n_s16(vshrq_n_s16(i,10), 15);
i = vsubq_s16(i,vshlq_n_s16(vecExpUnBiased,10));
float16x8_t vecTmpFlt1 = vreinterpretq_f16_s16(i);
float16x8_t lutV;
int16x8_t n;
int16x8_t offset;
float16x8_t tmp,v,res;
tmp = vmulq_n_f16(vsubq_n_f16(vecTmpFlt1,1.0f16),(_Float16)(1 << NB_DIV_LOGF16));
n = vcvtq_s16_f16(tmp);
v = vsubq_f16(tmp,vcvtq_f16_s16(n));
offset = vmulq_n_s16(n,(1+NB_DEG_LOGF16));
offset = vaddq_n_s16(offset,NB_DEG_LOGF16-1);
res = vldrhq_gather_shifted_offset_f16(lut_logf16,(uint16x8_t)offset);
offset = vsubq_n_s16(offset,1);
for(int j=NB_DEG_LOGF16-2; j >=0 ; j--)
{
lutV = vldrhq_gather_shifted_offset_f16(lut_logf16,(uint16x8_t)offset);
res = vfmaq_f16(lutV,v,res);
offset = vsubq_n_s16(offset,1);
}
res = vfmaq_n_f16(res,vcvtq_f16_s16(vecExpUnBiased),0.693147f16);
return(res);
}
#endif
/**
@ingroup groupFastMath
*/
/**
@addtogroup vlog
@{
*/
/**
@brief Floating-point vector of log values.
@param[in] pSrc points to the input vector
@param[out] pDst points to the output vector
@param[in] blockSize number of samples in each vector
@return none
*/
void arm_vlog_f16(
const float16_t * pSrc,
float16_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt;
#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
f16x8_t src;
f16x8_t dst;
blkCnt = blockSize >> 3;
while (blkCnt > 0U)
{
src = vld1q(pSrc);
dst = vlogq_lut_f16(src);
vst1q(pDst, dst);
pSrc += 8;
pDst += 8;
/* Decrement loop counter */
blkCnt--;
}
blkCnt = blockSize & 7;
#else
blkCnt = blockSize;
#endif
while (blkCnt > 0U)
{
/* C = log(A) */
/* Calculate log and store result in destination buffer. */
*pDst++ = logf16_scalar(*pSrc++);
/* Decrement loop counter */
blkCnt--;
}
}
/**
@} end of vlog group
*/
#endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
|
