• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* ----------------------------------------------------------------------
2  * Project:      CMSIS DSP Library
3  * Title:        arm_rms_q15.c
4  * Description:  Root Mean Square of the elements of a Q15 vector
5  *
6  * $Date:        23 April 2021
7  * $Revision:    V1.9.0
8  *
9  * Target Processor: Cortex-M and Cortex-A cores
10  * -------------------------------------------------------------------- */
11 /*
12  * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved.
13  *
14  * SPDX-License-Identifier: Apache-2.0
15  *
16  * Licensed under the Apache License, Version 2.0 (the License); you may
17  * not use this file except in compliance with the License.
18  * You may obtain a copy of the License at
19  *
20  * www.apache.org/licenses/LICENSE-2.0
21  *
22  * Unless required by applicable law or agreed to in writing, software
23  * distributed under the License is distributed on an AS IS BASIS, WITHOUT
24  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
25  * See the License for the specific language governing permissions and
26  * limitations under the License.
27  */
28 
29 #include "dsp/statistics_functions.h"
30 
31 /**
32   @ingroup groupStats
33  */
34 
35 /**
36   @addtogroup RMS
37   @{
38  */
39 
40 /**
41   @brief         Root Mean Square of the elements of a Q15 vector.
42   @param[in]     pSrc       points to the input vector
43   @param[in]     blockSize  number of samples in input vector
44   @param[out]    pResult    root mean square value returned here
45   @return        none
46 
47   @par           Scaling and Overflow Behavior
48                    The function is implemented using a 64-bit internal accumulator.
49                    The input is represented in 1.15 format.
50                    Intermediate multiplication yields a 2.30 format, and this
51                    result is added without saturation to a 64-bit accumulator in 34.30 format.
52                    With 33 guard bits in the accumulator, there is no risk of overflow, and the
53                    full precision of the intermediate multiplication is preserved.
54                    Finally, the 34.30 result is truncated to 34.15 format by discarding the lower
55                    15 bits, and then saturated to yield a result in 1.15 format.
56  */
57 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
arm_rms_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult)58 void arm_rms_q15(
59   const q15_t * pSrc,
60         uint32_t blockSize,
61         q15_t * pResult)
62 {
63     q63_t pow = 0.0f;
64     q15_t normalizedPower;
65 
66     arm_power_q15(pSrc, blockSize, &pow);
67 
68     normalizedPower=__SSAT((pow / (q63_t) blockSize) >> 15,16);
69     arm_sqrt_q15(normalizedPower, pResult);
70 }
71 #else
arm_rms_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult)72 void arm_rms_q15(
73   const q15_t * pSrc,
74         uint32_t blockSize,
75         q15_t * pResult)
76 {
77         uint32_t blkCnt;                               /* Loop counter */
78         q63_t sum = 0;                                 /* Temporary result storage */
79         q15_t in;                                      /* Temporary variable to store input value */
80 
81 #if defined (ARM_MATH_LOOPUNROLL) && defined (ARM_MATH_DSP)
82         q31_t in32;                                    /* Temporary variable to store input value */
83 #endif
84 
85 #if defined (ARM_MATH_LOOPUNROLL)
86 
87   /* Loop unrolling: Compute 4 outputs at a time */
88   blkCnt = blockSize >> 2U;
89 
90   while (blkCnt > 0U)
91   {
92     /* C = A[0] * A[0] + A[1] * A[1] + ... + A[blockSize-1] * A[blockSize-1] */
93 
94     /* Compute sum of squares and store result in a temporary variable. */
95 #if defined (ARM_MATH_DSP)
96     in32 = read_q15x2_ia ((q15_t **) &pSrc);
97     sum = __SMLALD(in32, in32, sum);
98 
99     in32 = read_q15x2_ia ((q15_t **) &pSrc);
100     sum = __SMLALD(in32, in32, sum);
101 #else
102     in = *pSrc++;
103     sum += ((q31_t) in * in);
104 
105     in = *pSrc++;
106     sum += ((q31_t) in * in);
107 
108     in = *pSrc++;
109     sum += ((q31_t) in * in);
110 
111     in = *pSrc++;
112     sum += ((q31_t) in * in);
113 #endif /* #if defined (ARM_MATH_DSP) */
114 
115     /* Decrement loop counter */
116     blkCnt--;
117   }
118 
119   /* Loop unrolling: Compute remaining outputs */
120   blkCnt = blockSize % 0x4U;
121 
122 #else
123 
124   /* Initialize blkCnt with number of samples */
125   blkCnt = blockSize;
126 
127 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
128 
129   while (blkCnt > 0U)
130   {
131     /* C = A[0] * A[0] + A[1] * A[1] + ... + A[blockSize-1] * A[blockSize-1] */
132 
133     in = *pSrc++;
134     /* Compute sum of squares and store result in a temporary variable. */
135     sum += ((q31_t) in * in);
136 
137     /* Decrement loop counter */
138     blkCnt--;
139   }
140 
141   /* Truncating and saturating the accumulator to 1.15 format */
142   /* Store result in destination */
143   arm_sqrt_q15(__SSAT((sum / (q63_t)blockSize) >> 15, 16), pResult);
144 }
145 #endif /* defined(ARM_MATH_MVEI) */
146 
147 /**
148   @} end of RMS group
149  */
150