1  
  2  /* ----------------------------------------------------------------------
  3   * Project:      CMSIS DSP Library
  4   * Title:        arm_chebyshev_distance_f16.c
  5   * Description:  Chebyshev 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 Chebyshev Chebyshev distance
 43  
 44    Chebyshev distance
 45   */
 46  
 47  /**
 48    @addtogroup Chebyshev
 49    @{
 50   */
 51  
 52  
 53  /**
 54   * @brief        Chebyshev distance between two vectors
 55   * @param[in]    pA         First vector
 56   * @param[in]    pB         Second vector
 57   * @param[in]    blockSize  vector length
 58   * @return distance
 59   *
 60   */
 61  
 62  #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
 63  
 64  #include "arm_helium_utils.h"
 65  #include "arm_vec_math.h"
 66  
 67  float16_t arm_chebyshev_distance_f16(const float16_t *pA,const float16_t *pB, uint32_t blockSize)
 68  {
 69      uint32_t        blkCnt;     /* loop counters */
 70      f16x8_t         vecA, vecB;
 71      f16x8_t         vecDiff = vdupq_n_f16(0.0);
 72      float16_t       maxValue = 0.0f16;
 73  
 74  
 75      blkCnt = blockSize >> 3;
 76      while (blkCnt > 0U) {
 77          vecA = vld1q(pA);
 78          pA += 8;
 79          vecB = vld1q(pB);
 80          pB += 8;
 81          /*
 82           * update per-lane max.
 83           */
 84          vecDiff = vmaxnmaq(vsubq(vecA, vecB), vecDiff);
 85          /*
 86           * Decrement the blockSize loop counter
 87           */
 88          blkCnt--;
 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          vecA = vldrhq_z_f16(pA, p0);
 99          vecB = vldrhq_z_f16(pB, p0);
100  
101          /*
102           * Get current max per lane and current index per lane
103           * when a max is selected
104           */
105          vecDiff = vmaxnmaq_m(vecDiff, vsubq(vecA, vecB), p0);
106      }
107      /*
108       * Get max value across the vector
109       */
110      return vmaxnmavq(maxValue, vecDiff);
111  }
112  
113  #else
114  float16_t arm_chebyshev_distance_f16(const float16_t *pA,const float16_t *pB, uint32_t blockSize)
115  {
116     _Float16 diff=0.0f,  maxVal,tmpA, tmpB;
117  
118     tmpA = *pA++;
119     tmpB = *pB++;
120     diff = (_Float16)fabsf((float32_t)((_Float16)tmpA - (_Float16)tmpB));
121     maxVal = diff;
122     blockSize--;
123  
124     while(blockSize > 0)
125     {
126        tmpA = *pA++;
127        tmpB = *pB++;
128        diff = (_Float16)fabsf((float32_t)((_Float16)tmpA - (_Float16)tmpB));
129        if ((_Float16)diff > (_Float16)maxVal)
130        {
131          maxVal = diff;
132        }
133        blockSize --;
134     }
135    
136     return(maxVal);
137  }
138  #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
139  
140  
141  /**
142   * @} end of Chebyshev group
143   */
144  
145  #endif /* #if defined(ARM_FLOAT16_SUPPORTED) */ 
146