1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_accumulate_f32.c
4 * Description: Sum value of a floating-point vector
5 *
6 * $Date: 14 July 2022
7 * $Revision: V1.0.0
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 "dsp/statistics_functions.h"
30
31 /**
32 @ingroup groupStats
33 */
34
35
36 /**
37 @addtogroup Accumulation
38 @{
39 */
40
41 /**
42 @brief Accumulation value of a floating-point vector.
43 @param[in] pSrc points to the input vector.
44 @param[in] blockSize number of samples in input vector.
45 @param[out] pResult sum of values in input vector.
46 */
47
48 #if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
49
50 #include "arm_helium_utils.h"
51
arm_accumulate_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)52 ARM_DSP_ATTRIBUTE void arm_accumulate_f32(
53 const float32_t * pSrc,
54 uint32_t blockSize,
55 float32_t * pResult)
56 {
57 f32x4_t vecA;
58 f32x4_t vecSum;
59 uint32_t blkCnt;
60 float32_t sum = 0.0f;
61 vecSum = vdupq_n_f32(0.0f);
62
63 /* Compute 4 outputs at a time */
64 blkCnt = blockSize >> 2U;
65 while (blkCnt > 0U)
66 {
67 /*
68 * C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1]
69 * Calculate dot product and then store the result in a temporary buffer.
70 * and advance vector source and destination pointers
71 */
72 vecA = vld1q_f32(pSrc);
73 pSrc += 4;
74
75 vecSum = vaddq_f32(vecSum, vecA);
76 /*
77 * Decrement the blockSize loop counter
78 */
79 blkCnt --;
80 }
81
82
83 blkCnt = blockSize & 3;
84 if (blkCnt > 0U)
85 {
86 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
87
88 mve_pred16_t p0 = vctp32q(blkCnt);
89 vecA = vld1q(pSrc);
90 vecSum = vaddq_m(vecSum,vecSum, vecA, p0);
91 }
92
93 sum = vecAddAcrossF32Mve(vecSum);
94
95 /* Store result in destination buffer */
96 *pResult = sum;
97 }
98
99 #else
100
101 #if defined(ARM_MATH_NEON) && !defined(ARM_MATH_AUTOVECTORIZE)
arm_accumulate_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)102 ARM_DSP_ATTRIBUTE void arm_accumulate_f32(
103 const float32_t * pSrc,
104 uint32_t blockSize,
105 float32_t * pResult)
106 {
107 float32_t sum = 0.0f; /* Temporary result storage */
108 float32x4_t sumV = vdupq_n_f32(0.0f); /* Temporary result storage */
109 float32x2_t sumV2;
110
111 uint32_t blkCnt; /* Loop counter */
112
113 float32x4_t inV;
114
115 blkCnt = blockSize >> 2U;
116
117 /* Compute 4 outputs at a time.
118 ** a second loop below computes the remaining 1 to 3 samples. */
119 while (blkCnt > 0U)
120 {
121 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
122 inV = vld1q_f32(pSrc);
123 sumV = vaddq_f32(sumV, inV);
124
125 pSrc += 4;
126 /* Decrement the loop counter */
127 blkCnt--;
128 }
129
130 sumV2 = vpadd_f32(vget_low_f32(sumV),vget_high_f32(sumV));
131 sum = vget_lane_f32(sumV2, 0) + vget_lane_f32(sumV2, 1);
132
133 /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
134 ** No loop unrolling is used. */
135 blkCnt = blockSize & 3;
136
137 while (blkCnt > 0U)
138 {
139 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
140 sum += *pSrc++;
141
142 /* Decrement the loop counter */
143 blkCnt--;
144 }
145
146 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
147 /* Store the result to the destination */
148 *pResult = sum;
149 }
150
151 #else
arm_accumulate_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)152 ARM_DSP_ATTRIBUTE void arm_accumulate_f32(
153 const float32_t * pSrc,
154 uint32_t blockSize,
155 float32_t * pResult)
156 {
157 uint32_t blkCnt; /* Loop counter */
158 float32_t sum = 0.0f; /* Temporary result storage */
159
160 #if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE)
161
162 /* Loop unrolling: Compute 4 outputs at a time */
163 blkCnt = blockSize >> 2U;
164
165 while (blkCnt > 0U)
166 {
167 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
168 sum += *pSrc++;
169
170 sum += *pSrc++;
171
172 sum += *pSrc++;
173
174 sum += *pSrc++;
175
176 /* Decrement the loop counter */
177 blkCnt--;
178 }
179
180 /* Loop unrolling: Compute remaining outputs */
181 blkCnt = blockSize % 0x4U;
182
183 #else
184
185 /* Initialize blkCnt with number of samples */
186 blkCnt = blockSize;
187
188 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
189
190 while (blkCnt > 0U)
191 {
192 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
193 sum += *pSrc++;
194
195 /* Decrement loop counter */
196 blkCnt--;
197 }
198
199 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
200 /* Store result to destination */
201 *pResult = sum ;
202 }
203 #endif /* #if defined(ARM_MATH_NEON) */
204
205 #endif /* #if defined(ARM_MATH_MVEF) */
206 /**
207 @} end of Accumulation group
208 */
209
