1
2 /* ----------------------------------------------------------------------
3 * Project: CMSIS DSP Library
4 * Title: arm_euclidean_distance_f16.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_f16.h"
31
32 #if defined(ARM_FLOAT16_SUPPORTED)
33
34 #include <limits.h>
35 #include <math.h>
36
37
38 /**
39 @ingroup FloatDist
40 */
41
42 /**
43 @defgroup Euclidean Euclidean distance
44
45 Euclidean distance
46 */
47
48
49 /**
50 @addtogroup Euclidean
51 @{
52 */
53
54
55 /**
56 * @brief Euclidean distance between two vectors
57 * @param[in] pA First vector
58 * @param[in] pB Second vector
59 * @param[in] blockSize vector length
60 * @return distance
61 *
62 */
63 #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
64
65 #include "arm_helium_utils.h"
66 #include "arm_vec_math.h"
arm_euclidean_distance_f16(const float16_t * pA,const float16_t * pB,uint32_t blockSize)67 float16_t arm_euclidean_distance_f16(const float16_t *pA,const float16_t *pB, uint32_t blockSize)
68 {
69 uint32_t blkCnt;
70 float16_t tmp;
71 f16x8_t a, b, accumV, tempV;
72
73 accumV = vdupq_n_f16(0.0f);
74
75 blkCnt = blockSize >> 3;
76 while (blkCnt > 0U) {
77 a = vld1q(pA);
78 b = vld1q(pB);
79
80 tempV = vsubq(a, b);
81 accumV = vfmaq(accumV, tempV, tempV);
82
83 pA += 8;
84 pB += 8;
85 blkCnt--;
86 }
87
88 /*
89 * tail
90 * (will be merged thru tail predication)
91 */
92 blkCnt = blockSize & 7;
93 if (blkCnt > 0U) {
94 mve_pred16_t p0 = vctp16q(blkCnt);
95
96 a = vldrhq_z_f16(pA, p0);
97 b = vldrhq_z_f16(pB, p0);
98
99 tempV = vsubq(a, b);
100 accumV = vfmaq_m(accumV, tempV, tempV, p0);
101 }
102
103 arm_sqrt_f16(vecAddAcrossF16Mve(accumV), &tmp);
104 return (tmp);
105 }
106
107 #else
arm_euclidean_distance_f16(const float16_t * pA,const float16_t * pB,uint32_t blockSize)108 float16_t arm_euclidean_distance_f16(const float16_t *pA,const float16_t *pB, uint32_t blockSize)
109 {
110 _Float16 accum=0.0f,tmp;
111 float16_t result;
112
113 while(blockSize > 0)
114 {
115 tmp = (_Float16)*pA++ - (_Float16)*pB++;
116 accum += SQ(tmp);
117 blockSize --;
118 }
119 arm_sqrt_f16(accum,&result);
120 return(result);
121 }
122
123 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
124
125
126 /**
127 * @} end of Euclidean group
128 */
129
130 #endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
131
132
