1
2 /* ----------------------------------------------------------------------
3 * Project: CMSIS DSP Library
4 * Title: arm_minkowski_distance_f16.c
5 * Description: Minkowski 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 @ingroup FloatDist
39 */
40
41 /**
42 @defgroup Minkowski Minkowski distance
43
44 Minkowski distance
45 */
46
47 /**
48 @addtogroup Minkowski
49 @{
50 */
51
52
53 /**
54 * @brief Minkowski distance between two vectors
55 *
56 * @param[in] pA First vector
57 * @param[in] pB Second vector
58 * @param[in] order Distance order
59 * @param[in] blockSize Number of samples
60 * @return distance
61 *
62 */
63
64 #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
65
66 #include "arm_helium_utils.h"
67 #include "arm_vec_math_f16.h"
68
arm_minkowski_distance_f16(const float16_t * pA,const float16_t * pB,int32_t order,uint32_t blockSize)69 float16_t arm_minkowski_distance_f16(const float16_t *pA,const float16_t *pB, int32_t order, uint32_t blockSize)
70 {
71 uint32_t blkCnt;
72 f16x8_t a, b, tmpV, sumV;
73
74 sumV = vdupq_n_f16(0.0f);
75
76 blkCnt = blockSize >> 3;
77 while (blkCnt > 0U) {
78 a = vld1q(pA);
79 b = vld1q(pB);
80
81 tmpV = vabdq(a, b);
82 tmpV = vpowq_f16(tmpV, vdupq_n_f16(order));
83 sumV = vaddq(sumV, tmpV);
84
85 pA += 8;
86 pB += 8;
87 blkCnt--;
88 }
89
90 /*
91 * tail
92 * (will be merged thru tail predication)
93 */
94 blkCnt = blockSize & 7;
95 if (blkCnt > 0U) {
96 mve_pred16_t p0 = vctp16q(blkCnt);
97
98 a = vldrhq_z_f16(pA, p0);
99 b = vldrhq_z_f16(pB, p0);
100
101 tmpV = vabdq(a, b);
102 tmpV = vpowq_f16(tmpV, vdupq_n_f16(order));
103 sumV = vaddq_m(sumV, sumV, tmpV, p0);
104 }
105
106 return (powf(vecAddAcrossF16Mve(sumV), (1.0f / (float16_t) order)));
107 }
108
109
110 #else
111
112
arm_minkowski_distance_f16(const float16_t * pA,const float16_t * pB,int32_t order,uint32_t blockSize)113 float16_t arm_minkowski_distance_f16(const float16_t *pA,const float16_t *pB, int32_t order, uint32_t blockSize)
114 {
115 _Float16 sum;
116 uint32_t i;
117
118 sum = 0.0f;
119 for(i=0; i < blockSize; i++)
120 {
121 sum += (_Float16)powf(fabsf(pA[i] - pB[i]),order);
122 }
123
124
125 return(powf(sum,(1.0f/order)));
126
127 }
128
129 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
130
131
132 /**
133 * @} end of Minkowski group
134 */
135
136 #endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
137
138
