Aded CMSIS5 DSP and NN folder. Needs some work
DSP/source/StatisticsFunctions/arm_rms_q31.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?
User | Revision | Line number | New contents of line |
---|---|---|---|
robert_lp | 0:eedb7d567a5d | 1 | /* ---------------------------------------------------------------------- |
robert_lp | 0:eedb7d567a5d | 2 | * Project: CMSIS DSP Library |
robert_lp | 0:eedb7d567a5d | 3 | * Title: arm_rms_q31.c |
robert_lp | 0:eedb7d567a5d | 4 | * Description: Root Mean Square of the elements of a Q31 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 | * @addtogroup RMS |
robert_lp | 0:eedb7d567a5d | 33 | * @{ |
robert_lp | 0:eedb7d567a5d | 34 | */ |
robert_lp | 0:eedb7d567a5d | 35 | |
robert_lp | 0:eedb7d567a5d | 36 | |
robert_lp | 0:eedb7d567a5d | 37 | /** |
robert_lp | 0:eedb7d567a5d | 38 | * @brief Root Mean Square of the elements of a Q31 vector. |
robert_lp | 0:eedb7d567a5d | 39 | * @param[in] *pSrc points to the input vector |
robert_lp | 0:eedb7d567a5d | 40 | * @param[in] blockSize length of the input vector |
robert_lp | 0:eedb7d567a5d | 41 | * @param[out] *pResult rms value returned here |
robert_lp | 0:eedb7d567a5d | 42 | * @return none. |
robert_lp | 0:eedb7d567a5d | 43 | * |
robert_lp | 0:eedb7d567a5d | 44 | * @details |
robert_lp | 0:eedb7d567a5d | 45 | * <b>Scaling and Overflow Behavior:</b> |
robert_lp | 0:eedb7d567a5d | 46 | * |
robert_lp | 0:eedb7d567a5d | 47 | *\par |
robert_lp | 0:eedb7d567a5d | 48 | * The function is implemented using an internal 64-bit accumulator. |
robert_lp | 0:eedb7d567a5d | 49 | * The input is represented in 1.31 format, and intermediate multiplication |
robert_lp | 0:eedb7d567a5d | 50 | * yields a 2.62 format. |
robert_lp | 0:eedb7d567a5d | 51 | * The accumulator maintains full precision of the intermediate multiplication results, |
robert_lp | 0:eedb7d567a5d | 52 | * but provides only a single guard bit. |
robert_lp | 0:eedb7d567a5d | 53 | * There is no saturation on intermediate additions. |
robert_lp | 0:eedb7d567a5d | 54 | * If the accumulator overflows, it wraps around and distorts the result. |
robert_lp | 0:eedb7d567a5d | 55 | * In order to avoid overflows completely, the input signal must be scaled down by |
robert_lp | 0:eedb7d567a5d | 56 | * log2(blockSize) bits, as a total of blockSize additions are performed internally. |
robert_lp | 0:eedb7d567a5d | 57 | * Finally, the 2.62 accumulator is right shifted by 31 bits to yield a 1.31 format value. |
robert_lp | 0:eedb7d567a5d | 58 | * |
robert_lp | 0:eedb7d567a5d | 59 | */ |
robert_lp | 0:eedb7d567a5d | 60 | |
robert_lp | 0:eedb7d567a5d | 61 | void arm_rms_q31( |
robert_lp | 0:eedb7d567a5d | 62 | q31_t * pSrc, |
robert_lp | 0:eedb7d567a5d | 63 | uint32_t blockSize, |
robert_lp | 0:eedb7d567a5d | 64 | q31_t * pResult) |
robert_lp | 0:eedb7d567a5d | 65 | { |
robert_lp | 0:eedb7d567a5d | 66 | q63_t sum = 0; /* accumulator */ |
robert_lp | 0:eedb7d567a5d | 67 | q31_t in; /* Temporary variable to store the input */ |
robert_lp | 0:eedb7d567a5d | 68 | uint32_t blkCnt; /* loop counter */ |
robert_lp | 0:eedb7d567a5d | 69 | |
robert_lp | 0:eedb7d567a5d | 70 | #if defined (ARM_MATH_DSP) |
robert_lp | 0:eedb7d567a5d | 71 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
robert_lp | 0:eedb7d567a5d | 72 | |
robert_lp | 0:eedb7d567a5d | 73 | q31_t in1, in2, in3, in4; /* Temporary input variables */ |
robert_lp | 0:eedb7d567a5d | 74 | |
robert_lp | 0:eedb7d567a5d | 75 | /*loop Unrolling */ |
robert_lp | 0:eedb7d567a5d | 76 | blkCnt = blockSize >> 2U; |
robert_lp | 0:eedb7d567a5d | 77 | |
robert_lp | 0:eedb7d567a5d | 78 | /* First part of the processing with loop unrolling. Compute 8 outputs at a time. |
robert_lp | 0:eedb7d567a5d | 79 | ** a second loop below computes the remaining 1 to 7 samples. */ |
robert_lp | 0:eedb7d567a5d | 80 | while (blkCnt > 0U) |
robert_lp | 0:eedb7d567a5d | 81 | { |
robert_lp | 0:eedb7d567a5d | 82 | /* C = A[0] * A[0] + A[1] * A[1] + A[2] * A[2] + ... + A[blockSize-1] * A[blockSize-1] */ |
robert_lp | 0:eedb7d567a5d | 83 | /* Compute sum of the squares and then store the result in a temporary variable, sum */ |
robert_lp | 0:eedb7d567a5d | 84 | /* read two samples from source buffer */ |
robert_lp | 0:eedb7d567a5d | 85 | in1 = pSrc[0]; |
robert_lp | 0:eedb7d567a5d | 86 | in2 = pSrc[1]; |
robert_lp | 0:eedb7d567a5d | 87 | |
robert_lp | 0:eedb7d567a5d | 88 | /* calculate power and accumulate to accumulator */ |
robert_lp | 0:eedb7d567a5d | 89 | sum += (q63_t) in1 *in1; |
robert_lp | 0:eedb7d567a5d | 90 | sum += (q63_t) in2 *in2; |
robert_lp | 0:eedb7d567a5d | 91 | |
robert_lp | 0:eedb7d567a5d | 92 | /* read two samples from source buffer */ |
robert_lp | 0:eedb7d567a5d | 93 | in3 = pSrc[2]; |
robert_lp | 0:eedb7d567a5d | 94 | in4 = pSrc[3]; |
robert_lp | 0:eedb7d567a5d | 95 | |
robert_lp | 0:eedb7d567a5d | 96 | /* calculate power and accumulate to accumulator */ |
robert_lp | 0:eedb7d567a5d | 97 | sum += (q63_t) in3 *in3; |
robert_lp | 0:eedb7d567a5d | 98 | sum += (q63_t) in4 *in4; |
robert_lp | 0:eedb7d567a5d | 99 | |
robert_lp | 0:eedb7d567a5d | 100 | |
robert_lp | 0:eedb7d567a5d | 101 | /* update source buffer to process next samples */ |
robert_lp | 0:eedb7d567a5d | 102 | pSrc += 4U; |
robert_lp | 0:eedb7d567a5d | 103 | |
robert_lp | 0:eedb7d567a5d | 104 | /* Decrement the loop counter */ |
robert_lp | 0:eedb7d567a5d | 105 | blkCnt--; |
robert_lp | 0:eedb7d567a5d | 106 | } |
robert_lp | 0:eedb7d567a5d | 107 | |
robert_lp | 0:eedb7d567a5d | 108 | /* If the blockSize is not a multiple of 8, compute any remaining output samples here. |
robert_lp | 0:eedb7d567a5d | 109 | ** No loop unrolling is used. */ |
robert_lp | 0:eedb7d567a5d | 110 | blkCnt = blockSize % 0x4U; |
robert_lp | 0:eedb7d567a5d | 111 | |
robert_lp | 0:eedb7d567a5d | 112 | #else |
robert_lp | 0:eedb7d567a5d | 113 | /* Run the below code for Cortex-M0 */ |
robert_lp | 0:eedb7d567a5d | 114 | |
robert_lp | 0:eedb7d567a5d | 115 | blkCnt = blockSize; |
robert_lp | 0:eedb7d567a5d | 116 | |
robert_lp | 0:eedb7d567a5d | 117 | #endif /* #if defined (ARM_MATH_DSP) */ |
robert_lp | 0:eedb7d567a5d | 118 | |
robert_lp | 0:eedb7d567a5d | 119 | while (blkCnt > 0U) |
robert_lp | 0:eedb7d567a5d | 120 | { |
robert_lp | 0:eedb7d567a5d | 121 | /* C = A[0] * A[0] + A[1] * A[1] + A[2] * A[2] + ... + A[blockSize-1] * A[blockSize-1] */ |
robert_lp | 0:eedb7d567a5d | 122 | /* Compute sum of the squares and then store the results in a temporary variable, sum */ |
robert_lp | 0:eedb7d567a5d | 123 | in = *pSrc++; |
robert_lp | 0:eedb7d567a5d | 124 | sum += (q63_t) in *in; |
robert_lp | 0:eedb7d567a5d | 125 | |
robert_lp | 0:eedb7d567a5d | 126 | /* Decrement the loop counter */ |
robert_lp | 0:eedb7d567a5d | 127 | blkCnt--; |
robert_lp | 0:eedb7d567a5d | 128 | } |
robert_lp | 0:eedb7d567a5d | 129 | |
robert_lp | 0:eedb7d567a5d | 130 | /* Convert data in 2.62 to 1.31 by 31 right shifts and saturate */ |
robert_lp | 0:eedb7d567a5d | 131 | /* Compute Rms and store the result in the destination vector */ |
robert_lp | 0:eedb7d567a5d | 132 | arm_sqrt_q31(clip_q63_to_q31((sum / (q63_t) blockSize) >> 31), pResult); |
robert_lp | 0:eedb7d567a5d | 133 | } |
robert_lp | 0:eedb7d567a5d | 134 | |
robert_lp | 0:eedb7d567a5d | 135 | /** |
robert_lp | 0:eedb7d567a5d | 136 | * @} end of RMS group |
robert_lp | 0:eedb7d567a5d | 137 | */ |
robert_lp | 0:eedb7d567a5d | 138 |