1
2 /* ----------------------------------------------------------------------
3 * Project: CMSIS DSP Library
4 * Title: arm_euclidean_distance_f32.c
5 * Description: Euclidean distance between two vectors
6 *
7 * $Date: 23 April 2021
8 * $Revision: V1.9.0
9 *
10 * Target Processor: Cortex-M and Cortex-A cores
11 * -------------------------------------------------------------------- */
12 /*
13 * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved.
14 *
15 * SPDX-License-Identifier: Apache-2.0
16 *
17 * Licensed under the Apache License, Version 2.0 (the License); you may
18 * not use this file except in compliance with the License.
19 * You may obtain a copy of the License at
20 *
21 * www.apache.org/licenses/LICENSE-2.0
22 *
23 * Unless required by applicable law or agreed to in writing, software
24 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
25 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
26 * See the License for the specific language governing permissions and
27 * limitations under the License.
28 */
29
30 #include "dsp/distance_functions.h"
31 #include <limits.h>
32 #include <math.h>
33
34
35
36 /**
37 @addtogroup Euclidean
38 @{
39 */
40
41
42 /**
43 * @brief Euclidean distance between two vectors
44 * @param[in] pA First vector
45 * @param[in] pB Second vector
46 * @param[in] blockSize vector length
47 * @return distance
48 *
49 */
50 #if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
51
52 #include "arm_helium_utils.h"
53 #include "arm_vec_math.h"
arm_euclidean_distance_f32(const float32_t * pA,const float32_t * pB,uint32_t blockSize)54 float32_t arm_euclidean_distance_f32(const float32_t *pA,const float32_t *pB, uint32_t blockSize)
55 {
56 uint32_t blkCnt;
57 float32_t tmp;
58 f32x4_t a, b, accumV, tempV;
59
60 accumV = vdupq_n_f32(0.0f);
61
62 blkCnt = blockSize >> 2;
63 while (blkCnt > 0U) {
64 a = vld1q(pA);
65 b = vld1q(pB);
66
67 tempV = vsubq(a, b);
68 accumV = vfmaq(accumV, tempV, tempV);
69
70 pA += 4;
71 pB += 4;
72 blkCnt--;
73 }
74
75 /*
76 * tail
77 * (will be merged thru tail predication)
78 */
79 blkCnt = blockSize & 3;
80 if (blkCnt > 0U) {
81 mve_pred16_t p0 = vctp32q(blkCnt);
82
83 a = vldrwq_z_f32(pA, p0);
84 b = vldrwq_z_f32(pB, p0);
85
86 tempV = vsubq(a, b);
87 accumV = vfmaq_m(accumV, tempV, tempV, p0);
88 }
89
90 arm_sqrt_f32(vecAddAcrossF32Mve(accumV), &tmp);
91 return (tmp);
92 }
93 #else
94 #if defined(ARM_MATH_NEON)
95
96 #include "NEMath.h"
97
arm_euclidean_distance_f32(const float32_t * pA,const float32_t * pB,uint32_t blockSize)98 float32_t arm_euclidean_distance_f32(const float32_t *pA,const float32_t *pB, uint32_t blockSize)
99 {
100 float32_t accum=0.0f,tmp;
101 uint32_t blkCnt;
102 float32x4_t a,b,accumV;
103 float32x2_t accumV2;
104
105 accumV = vdupq_n_f32(0.0f);
106 blkCnt = blockSize >> 2;
107 while(blkCnt > 0)
108 {
109 a = vld1q_f32(pA);
110 b = vld1q_f32(pB);
111
112 a = vsubq_f32(a,b);
113 accumV = vmlaq_f32(accumV,a,a);
114 pA += 4;
115 pB += 4;
116 blkCnt --;
117 }
118 accumV2 = vpadd_f32(vget_low_f32(accumV),vget_high_f32(accumV));
119 accum = vget_lane_f32(accumV2, 0) + vget_lane_f32(accumV2, 1);
120
121 blkCnt = blockSize & 3;
122 while(blkCnt > 0)
123 {
124 tmp = *pA++ - *pB++;
125 accum += ARM_SQ(tmp);
126 blkCnt --;
127 }
128 arm_sqrt_f32(accum,&tmp);
129 return(tmp);
130 }
131
132 #else
arm_euclidean_distance_f32(const float32_t * pA,const float32_t * pB,uint32_t blockSize)133 float32_t arm_euclidean_distance_f32(const float32_t *pA,const float32_t *pB, uint32_t blockSize)
134 {
135 float32_t accum=0.0f,tmp;
136
137 while(blockSize > 0)
138 {
139 tmp = *pA++ - *pB++;
140 accum += ARM_SQ(tmp);
141 blockSize --;
142 }
143 arm_sqrt_f32(accum,&tmp);
144 return(tmp);
145 }
146 #endif
147 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
148
149
150 /**
151 * @} end of Euclidean group
152 */
153
