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
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
|
/* ----------------------------------------------------------------------
* Project: CMSIS DSP Library
* Title: arm_cmplx_mag_f16.c
* Description: Floating-point complex magnitude
*
* $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/complex_math_functions_f16.h"
#if defined(ARM_FLOAT16_SUPPORTED)
/**
@ingroup groupCmplxMath
*/
/**
@defgroup cmplx_mag Complex Magnitude
Computes the magnitude of the elements of a complex data vector.
The <code>pSrc</code> points to the source data and
<code>pDst</code> points to the where the result should be written.
<code>numSamples</code> specifies the number of complex samples
in the input array and the data is stored in an interleaved fashion
(real, imag, real, imag, ...).
The input array has a total of <code>2*numSamples</code> values;
the output array has a total of <code>numSamples</code> values.
The underlying algorithm is used:
<pre>
for (n = 0; n < numSamples; n++) {
pDst[n] = sqrt(pSrc[(2*n)+0]^2 + pSrc[(2*n)+1]^2);
}
</pre>
There are separate functions for floating-point, Q15, and Q31 data types.
*/
/**
@addtogroup cmplx_mag
@{
*/
/**
@brief Floating-point complex magnitude.
@param[in] pSrc points to input vector
@param[out] pDst points to output vector
@param[in] numSamples number of samples in each vector
@return none
*/
#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
#include "arm_helium_utils.h"
void arm_cmplx_mag_f16(
const float16_t * pSrc,
float16_t * pDst,
uint32_t numSamples)
{
int32_t blockSize = numSamples; /* loop counters */
uint32_t blkCnt; /* loop counters */
f16x8x2_t vecSrc;
f16x8_t sum;
/* Compute 4 complex samples at a time */
blkCnt = blockSize >> 3;
while (blkCnt > 0U)
{
q15x8_t newtonStartVec;
f16x8_t sumHalf, invSqrt;
vecSrc = vld2q(pSrc);
pSrc += 16;
sum = vmulq(vecSrc.val[0], vecSrc.val[0]);
sum = vfmaq(sum, vecSrc.val[1], vecSrc.val[1]);
/*
* inlined Fast SQRT using inverse SQRT newton-raphson method
*/
/* compute initial value */
newtonStartVec = vdupq_n_s16(INVSQRT_MAGIC_F16) - vshrq((q15x8_t) sum, 1);
sumHalf = sum * 0.5f;
/*
* compute 3 x iterations
*
* The more iterations, the more accuracy.
* If you need to trade a bit of accuracy for more performance,
* you can comment out the 3rd use of the macro.
*/
INVSQRT_NEWTON_MVE_F16(invSqrt, sumHalf, (f16x8_t) newtonStartVec);
INVSQRT_NEWTON_MVE_F16(invSqrt, sumHalf, invSqrt);
INVSQRT_NEWTON_MVE_F16(invSqrt, sumHalf, invSqrt);
/*
* set negative values to 0
*/
invSqrt = vdupq_m(invSqrt, (float16_t)0.0f, vcmpltq(invSqrt, (float16_t)0.0f));
/*
* sqrt(x) = x * invSqrt(x)
*/
sum = vmulq(sum, invSqrt);
vstrhq_f16(pDst, sum);
pDst += 8;
/*
* Decrement the blockSize loop counter
*/
blkCnt--;
}
/*
* tail
*/
blkCnt = blockSize & 7;
if (blkCnt > 0U)
{
mve_pred16_t p0 = vctp16q(blkCnt);
q15x8_t newtonStartVec;
f16x8_t sumHalf, invSqrt;
vecSrc = vld2q((float16_t const *)pSrc);
sum = vmulq(vecSrc.val[0], vecSrc.val[0]);
sum = vfmaq(sum, vecSrc.val[1], vecSrc.val[1]);
/*
* inlined Fast SQRT using inverse SQRT newton-raphson method
*/
/* compute initial value */
newtonStartVec = vdupq_n_s16(INVSQRT_MAGIC_F16) - vshrq((q15x8_t) sum, 1);
sumHalf = vmulq(sum, (float16_t)0.5);
/*
* compute 2 x iterations
*/
INVSQRT_NEWTON_MVE_F16(invSqrt, sumHalf, (f16x8_t) newtonStartVec);
INVSQRT_NEWTON_MVE_F16(invSqrt, sumHalf, invSqrt);
/*
* set negative values to 0
*/
invSqrt = vdupq_m(invSqrt, (float16_t)0.0, vcmpltq(invSqrt, (float16_t)0.0));
/*
* sqrt(x) = x * invSqrt(x)
*/
sum = vmulq(sum, invSqrt);
vstrhq_p_f16(pDst, sum, p0);
}
}
#else
void arm_cmplx_mag_f16(
const float16_t * pSrc,
float16_t * pDst,
uint32_t numSamples)
{
uint32_t blkCnt; /* loop counter */
_Float16 real, imag; /* Temporary variables to hold input values */
#if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE)
/* Loop unrolling: Compute 4 outputs at a time */
blkCnt = numSamples >> 2U;
while (blkCnt > 0U)
{
/* C[0] = sqrt(A[0] * A[0] + A[1] * A[1]) */
real = *pSrc++;
imag = *pSrc++;
/* store result in destination buffer. */
arm_sqrt_f16((real * real) + (imag * imag), pDst++);
real = *pSrc++;
imag = *pSrc++;
arm_sqrt_f16((real * real) + (imag * imag), pDst++);
real = *pSrc++;
imag = *pSrc++;
arm_sqrt_f16((real * real) + (imag * imag), pDst++);
real = *pSrc++;
imag = *pSrc++;
arm_sqrt_f16((real * real) + (imag * imag), pDst++);
/* Decrement loop counter */
blkCnt--;
}
/* Loop unrolling: Compute remaining outputs */
blkCnt = numSamples % 0x4U;
#else
/* Initialize blkCnt with number of samples */
blkCnt = numSamples;
#endif /* #if defined (ARM_MATH_LOOPUNROLL) */
while (blkCnt > 0U)
{
/* C[0] = sqrt(A[0] * A[0] + A[1] * A[1]) */
real = *pSrc++;
imag = *pSrc++;
/* store result in destination buffer. */
arm_sqrt_f16((real * real) + (imag * imag), pDst++);
/* Decrement loop counter */
blkCnt--;
}
}
#endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
/**
@} end of cmplx_mag group
*/
#endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
|
