Robert Lopez / CMSIS5

Aded CMSIS5 DSP and NN folder. Needs some work

DSP/source/StatisticsFunctions/arm_var_f32.c@0:eedb7d567a5d, 2018-04-12 (annotated)

Committer:
robert_lp
Date:
Thu Apr 12 01:31:58 2018 +0000
Revision:
0:eedb7d567a5d
CMSIS5 Library

Who changed what in which revision?

UserRevisionLine numberNew contents of line
robert_lp 0:eedb7d567a5d 1 /* ----------------------------------------------------------------------
robert_lp 0:eedb7d567a5d 2 * Project: CMSIS DSP Library
robert_lp 0:eedb7d567a5d 3 * Title: arm_var_f32.c
robert_lp 0:eedb7d567a5d 4 * Description: Variance of the elements of a floating-point vector
robert_lp 0:eedb7d567a5d 5 *
robert_lp 0:eedb7d567a5d 6 * $Date: 27. January 2017
robert_lp 0:eedb7d567a5d 7 * $Revision: V.1.5.1
robert_lp 0:eedb7d567a5d 8 *
robert_lp 0:eedb7d567a5d 9 * Target Processor: Cortex-M cores
robert_lp 0:eedb7d567a5d 10 * -------------------------------------------------------------------- */
robert_lp 0:eedb7d567a5d 11 /*
robert_lp 0:eedb7d567a5d 12 * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved.
robert_lp 0:eedb7d567a5d 13 *
robert_lp 0:eedb7d567a5d 14 * SPDX-License-Identifier: Apache-2.0
robert_lp 0:eedb7d567a5d 15 *
robert_lp 0:eedb7d567a5d 16 * Licensed under the Apache License, Version 2.0 (the License); you may
robert_lp 0:eedb7d567a5d 17 * not use this file except in compliance with the License.
robert_lp 0:eedb7d567a5d 18 * You may obtain a copy of the License at
robert_lp 0:eedb7d567a5d 19 *
robert_lp 0:eedb7d567a5d 20 * www.apache.org/licenses/LICENSE-2.0
robert_lp 0:eedb7d567a5d 21 *
robert_lp 0:eedb7d567a5d 22 * Unless required by applicable law or agreed to in writing, software
robert_lp 0:eedb7d567a5d 23 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
robert_lp 0:eedb7d567a5d 24 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
robert_lp 0:eedb7d567a5d 25 * See the License for the specific language governing permissions and
robert_lp 0:eedb7d567a5d 26 * limitations under the License.
robert_lp 0:eedb7d567a5d 27 */
robert_lp 0:eedb7d567a5d 28
robert_lp 0:eedb7d567a5d 29 #include "arm_math.h"
robert_lp 0:eedb7d567a5d 30
robert_lp 0:eedb7d567a5d 31 /**
robert_lp 0:eedb7d567a5d 32 * @ingroup groupStats
robert_lp 0:eedb7d567a5d 33 */
robert_lp 0:eedb7d567a5d 34
robert_lp 0:eedb7d567a5d 35 /**
robert_lp 0:eedb7d567a5d 36 * @defgroup variance Variance
robert_lp 0:eedb7d567a5d 37 *
robert_lp 0:eedb7d567a5d 38 * Calculates the variance of the elements in the input vector.
robert_lp 0:eedb7d567a5d 39 * The underlying algorithm used is the direct method sometimes referred to as the two-pass method:
robert_lp 0:eedb7d567a5d 40 *
robert_lp 0:eedb7d567a5d 41 * <pre>
robert_lp 0:eedb7d567a5d 42 * Result = sum(element - meanOfElements)^2) / numElement - 1
robert_lp 0:eedb7d567a5d 43 *
robert_lp 0:eedb7d567a5d 44 * where, meanOfElements = ( pSrc[0] * pSrc[0] + pSrc[1] * pSrc[1] + ... + pSrc[blockSize-1] ) / blockSize
robert_lp 0:eedb7d567a5d 45 *
robert_lp 0:eedb7d567a5d 46 * </pre>
robert_lp 0:eedb7d567a5d 47 *
robert_lp 0:eedb7d567a5d 48 * There are separate functions for floating point, Q31, and Q15 data types.
robert_lp 0:eedb7d567a5d 49 */
robert_lp 0:eedb7d567a5d 50
robert_lp 0:eedb7d567a5d 51 /**
robert_lp 0:eedb7d567a5d 52 * @addtogroup variance
robert_lp 0:eedb7d567a5d 53 * @{
robert_lp 0:eedb7d567a5d 54 */
robert_lp 0:eedb7d567a5d 55
robert_lp 0:eedb7d567a5d 56
robert_lp 0:eedb7d567a5d 57 /**
robert_lp 0:eedb7d567a5d 58 * @brief Variance of the elements of a floating-point vector.
robert_lp 0:eedb7d567a5d 59 * @param[in] *pSrc points to the input vector
robert_lp 0:eedb7d567a5d 60 * @param[in] blockSize length of the input vector
robert_lp 0:eedb7d567a5d 61 * @param[out] *pResult variance value returned here
robert_lp 0:eedb7d567a5d 62 * @return none.
robert_lp 0:eedb7d567a5d 63 */
robert_lp 0:eedb7d567a5d 64
robert_lp 0:eedb7d567a5d 65 void arm_var_f32(
robert_lp 0:eedb7d567a5d 66 float32_t * pSrc,
robert_lp 0:eedb7d567a5d 67 uint32_t blockSize,
robert_lp 0:eedb7d567a5d 68 float32_t * pResult)
robert_lp 0:eedb7d567a5d 69 {
robert_lp 0:eedb7d567a5d 70 float32_t fMean, fValue;
robert_lp 0:eedb7d567a5d 71 uint32_t blkCnt; /* loop counter */
robert_lp 0:eedb7d567a5d 72 float32_t * pInput = pSrc;
robert_lp 0:eedb7d567a5d 73 float32_t sum = 0.0f;
robert_lp 0:eedb7d567a5d 74 float32_t fSum = 0.0f;
robert_lp 0:eedb7d567a5d 75 #if defined(ARM_MATH_DSP)
robert_lp 0:eedb7d567a5d 76 float32_t in1, in2, in3, in4;
robert_lp 0:eedb7d567a5d 77 #endif
robert_lp 0:eedb7d567a5d 78
robert_lp 0:eedb7d567a5d 79 if (blockSize <= 1U)
robert_lp 0:eedb7d567a5d 80 {
robert_lp 0:eedb7d567a5d 81 *pResult = 0;
robert_lp 0:eedb7d567a5d 82 return;
robert_lp 0:eedb7d567a5d 83 }
robert_lp 0:eedb7d567a5d 84
robert_lp 0:eedb7d567a5d 85 #if defined(ARM_MATH_DSP)
robert_lp 0:eedb7d567a5d 86 /* Run the below code for Cortex-M4 and Cortex-M7 */
robert_lp 0:eedb7d567a5d 87
robert_lp 0:eedb7d567a5d 88 /*loop Unrolling */
robert_lp 0:eedb7d567a5d 89 blkCnt = blockSize >> 2U;
robert_lp 0:eedb7d567a5d 90
robert_lp 0:eedb7d567a5d 91 /* First part of the processing with loop unrolling. Compute 4 outputs at a time.
robert_lp 0:eedb7d567a5d 92 ** a second loop below computes the remaining 1 to 3 samples. */
robert_lp 0:eedb7d567a5d 93 while (blkCnt > 0U)
robert_lp 0:eedb7d567a5d 94 {
robert_lp 0:eedb7d567a5d 95 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
robert_lp 0:eedb7d567a5d 96 in1 = *pInput++;
robert_lp 0:eedb7d567a5d 97 in2 = *pInput++;
robert_lp 0:eedb7d567a5d 98 in3 = *pInput++;
robert_lp 0:eedb7d567a5d 99 in4 = *pInput++;
robert_lp 0:eedb7d567a5d 100
robert_lp 0:eedb7d567a5d 101 sum += in1;
robert_lp 0:eedb7d567a5d 102 sum += in2;
robert_lp 0:eedb7d567a5d 103 sum += in3;
robert_lp 0:eedb7d567a5d 104 sum += in4;
robert_lp 0:eedb7d567a5d 105
robert_lp 0:eedb7d567a5d 106 /* Decrement the loop counter */
robert_lp 0:eedb7d567a5d 107 blkCnt--;
robert_lp 0:eedb7d567a5d 108 }
robert_lp 0:eedb7d567a5d 109
robert_lp 0:eedb7d567a5d 110 /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
robert_lp 0:eedb7d567a5d 111 ** No loop unrolling is used. */
robert_lp 0:eedb7d567a5d 112 blkCnt = blockSize % 0x4U;
robert_lp 0:eedb7d567a5d 113
robert_lp 0:eedb7d567a5d 114 #else
robert_lp 0:eedb7d567a5d 115 /* Run the below code for Cortex-M0 or Cortex-M3 */
robert_lp 0:eedb7d567a5d 116
robert_lp 0:eedb7d567a5d 117 /* Loop over blockSize number of values */
robert_lp 0:eedb7d567a5d 118 blkCnt = blockSize;
robert_lp 0:eedb7d567a5d 119
robert_lp 0:eedb7d567a5d 120 #endif
robert_lp 0:eedb7d567a5d 121
robert_lp 0:eedb7d567a5d 122 while (blkCnt > 0U)
robert_lp 0:eedb7d567a5d 123 {
robert_lp 0:eedb7d567a5d 124 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
robert_lp 0:eedb7d567a5d 125 sum += *pInput++;
robert_lp 0:eedb7d567a5d 126
robert_lp 0:eedb7d567a5d 127 /* Decrement the loop counter */
robert_lp 0:eedb7d567a5d 128 blkCnt--;
robert_lp 0:eedb7d567a5d 129 }
robert_lp 0:eedb7d567a5d 130
robert_lp 0:eedb7d567a5d 131 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) / blockSize */
robert_lp 0:eedb7d567a5d 132 fMean = sum / (float32_t) blockSize;
robert_lp 0:eedb7d567a5d 133
robert_lp 0:eedb7d567a5d 134 pInput = pSrc;
robert_lp 0:eedb7d567a5d 135
robert_lp 0:eedb7d567a5d 136 #if defined(ARM_MATH_DSP)
robert_lp 0:eedb7d567a5d 137
robert_lp 0:eedb7d567a5d 138 /*loop Unrolling */
robert_lp 0:eedb7d567a5d 139 blkCnt = blockSize >> 2U;
robert_lp 0:eedb7d567a5d 140
robert_lp 0:eedb7d567a5d 141 /* First part of the processing with loop unrolling. Compute 4 outputs at a time.
robert_lp 0:eedb7d567a5d 142 ** a second loop below computes the remaining 1 to 3 samples. */
robert_lp 0:eedb7d567a5d 143 while (blkCnt > 0U)
robert_lp 0:eedb7d567a5d 144 {
robert_lp 0:eedb7d567a5d 145 fValue = *pInput++ - fMean;
robert_lp 0:eedb7d567a5d 146 fSum += fValue * fValue;
robert_lp 0:eedb7d567a5d 147 fValue = *pInput++ - fMean;
robert_lp 0:eedb7d567a5d 148 fSum += fValue * fValue;
robert_lp 0:eedb7d567a5d 149 fValue = *pInput++ - fMean;
robert_lp 0:eedb7d567a5d 150 fSum += fValue * fValue;
robert_lp 0:eedb7d567a5d 151 fValue = *pInput++ - fMean;
robert_lp 0:eedb7d567a5d 152 fSum += fValue * fValue;
robert_lp 0:eedb7d567a5d 153
robert_lp 0:eedb7d567a5d 154 /* Decrement the loop counter */
robert_lp 0:eedb7d567a5d 155 blkCnt--;
robert_lp 0:eedb7d567a5d 156 }
robert_lp 0:eedb7d567a5d 157
robert_lp 0:eedb7d567a5d 158 blkCnt = blockSize % 0x4U;
robert_lp 0:eedb7d567a5d 159 #else
robert_lp 0:eedb7d567a5d 160 /* Run the below code for Cortex-M0 or Cortex-M3 */
robert_lp 0:eedb7d567a5d 161
robert_lp 0:eedb7d567a5d 162 /* Loop over blockSize number of values */
robert_lp 0:eedb7d567a5d 163 blkCnt = blockSize;
robert_lp 0:eedb7d567a5d 164 #endif
robert_lp 0:eedb7d567a5d 165
robert_lp 0:eedb7d567a5d 166 while (blkCnt > 0U)
robert_lp 0:eedb7d567a5d 167 {
robert_lp 0:eedb7d567a5d 168 fValue = *pInput++ - fMean;
robert_lp 0:eedb7d567a5d 169 fSum += fValue * fValue;
robert_lp 0:eedb7d567a5d 170
robert_lp 0:eedb7d567a5d 171 /* Decrement the loop counter */
robert_lp 0:eedb7d567a5d 172 blkCnt--;
robert_lp 0:eedb7d567a5d 173 }
robert_lp 0:eedb7d567a5d 174
robert_lp 0:eedb7d567a5d 175 /* Variance */
robert_lp 0:eedb7d567a5d 176 *pResult = fSum / (float32_t)(blockSize - 1.0f);
robert_lp 0:eedb7d567a5d 177 }
robert_lp 0:eedb7d567a5d 178
robert_lp 0:eedb7d567a5d 179 /**
robert_lp 0:eedb7d567a5d 180 * @} end of variance group
robert_lp 0:eedb7d567a5d 181 */
robert_lp 0:eedb7d567a5d 182