1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_weighted_sum_f32.c
4 * Description: Weighted Sum
5 *
6 * $Date: 23 April 2021
7 * $Revision: V1.9.0
8 *
9 * Target Processor: Cortex-M and Cortex-A cores
10 * -------------------------------------------------------------------- */
11 /*
12 * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved.
13 *
14 * SPDX-License-Identifier: Apache-2.0
15 *
16 * Licensed under the Apache License, Version 2.0 (the License); you may
17 * not use this file except in compliance with the License.
18 * You may obtain a copy of the License at
19 *
20 * www.apache.org/licenses/LICENSE-2.0
21 *
22 * Unless required by applicable law or agreed to in writing, software
23 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
24 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
25 * See the License for the specific language governing permissions and
26 * limitations under the License.
27 */
28
29 #include <limits.h>
30 #include <math.h>
31
32 #include "dsp/support_functions.h"
33
34 /**
35 * @addtogroup weightedsum
36 * @{
37 */
38
39
40 /**
41 * @brief Weighted sum
42 *
43 *
44 * @param[in] *in Array of input values.
45 * @param[in] *weigths Weights
46 * @param[in] blockSize Number of samples in the input array.
47 * @return Weighted sum
48 *
49 */
50
51 #if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
52
53 #include "arm_helium_utils.h"
54
arm_weighted_sum_f32(const float32_t * in,const float32_t * weigths,uint32_t blockSize)55 float32_t arm_weighted_sum_f32(const float32_t *in,const float32_t *weigths, uint32_t blockSize)
56 {
57 float32_t accum1, accum2;
58 f32x4_t accum1V, accum2V;
59 f32x4_t inV, wV;
60 const float32_t *pIn, *pW;
61 uint32_t blkCnt;
62
63
64 pIn = in;
65 pW = weigths;
66
67
68 accum1V = vdupq_n_f32(0.0);
69 accum2V = vdupq_n_f32(0.0);
70
71 blkCnt = blockSize >> 2;
72 while (blkCnt > 0)
73 {
74 inV = vld1q(pIn);
75 wV = vld1q(pW);
76
77 pIn += 4;
78 pW += 4;
79
80 accum1V = vfmaq(accum1V, inV, wV);
81 accum2V = vaddq(accum2V, wV);
82 blkCnt--;
83 }
84
85 accum1 = vecAddAcrossF32Mve(accum1V);
86 accum2 = vecAddAcrossF32Mve(accum2V);
87
88 blkCnt = blockSize & 3;
89 while(blkCnt > 0)
90 {
91 accum1 += *pIn++ * *pW;
92 accum2 += *pW++;
93 blkCnt--;
94 }
95
96
97 return (accum1 / accum2);
98 }
99
100 #else
101 #if defined(ARM_MATH_NEON)
102
103 #include "NEMath.h"
arm_weighted_sum_f32(const float32_t * in,const float32_t * weigths,uint32_t blockSize)104 float32_t arm_weighted_sum_f32(const float32_t *in,const float32_t *weigths, uint32_t blockSize)
105 {
106
107 float32_t accum1, accum2;
108 float32x4_t accum1V, accum2V;
109 float32x2_t tempV;
110
111 float32x4_t inV,wV;
112
113 const float32_t *pIn, *pW;
114 uint32_t blkCnt;
115
116
117 pIn = in;
118 pW = weigths;
119
120 accum1=0.0f;
121 accum2=0.0f;
122
123 accum1V = vdupq_n_f32(0.0f);
124 accum2V = vdupq_n_f32(0.0f);
125
126 blkCnt = blockSize >> 2;
127 while(blkCnt > 0)
128 {
129 inV = vld1q_f32(pIn);
130 wV = vld1q_f32(pW);
131
132 pIn += 4;
133 pW += 4;
134
135 accum1V = vmlaq_f32(accum1V,inV,wV);
136 accum2V = vaddq_f32(accum2V,wV);
137 blkCnt--;
138 }
139
140 tempV = vpadd_f32(vget_low_f32(accum1V),vget_high_f32(accum1V));
141 accum1 = vget_lane_f32(tempV, 0) + vget_lane_f32(tempV, 1);
142
143 tempV = vpadd_f32(vget_low_f32(accum2V),vget_high_f32(accum2V));
144 accum2 = vget_lane_f32(tempV, 0) + vget_lane_f32(tempV, 1);
145
146 blkCnt = blockSize & 3;
147 while(blkCnt > 0)
148 {
149 accum1 += *pIn++ * *pW;
150 accum2 += *pW++;
151 blkCnt--;
152 }
153
154
155 return(accum1 / accum2);
156 }
157 #else
arm_weighted_sum_f32(const float32_t * in,const float32_t * weigths,uint32_t blockSize)158 float32_t arm_weighted_sum_f32(const float32_t *in, const float32_t *weigths, uint32_t blockSize)
159 {
160
161 float32_t accum1, accum2;
162 const float32_t *pIn, *pW;
163 uint32_t blkCnt;
164
165
166 pIn = in;
167 pW = weigths;
168
169 accum1=0.0f;
170 accum2=0.0f;
171
172 blkCnt = blockSize;
173 while(blkCnt > 0)
174 {
175 accum1 += *pIn++ * *pW;
176 accum2 += *pW++;
177 blkCnt--;
178 }
179
180 return(accum1 / accum2);
181 }
182 #endif
183 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
184
185 /**
186 * @} end of weightedsum group
187 */
188
