1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_var_f32.c
4 * Description: Variance of the elements of a floating-point 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 @defgroup variance Variance
37
38 Calculates the variance of the elements in the input vector.
39 The underlying algorithm used is the direct method sometimes referred to as the two-pass method:
40
41 <pre>
42 Result = sum(element - meanOfElements)^2) / numElement - 1
43
44 meanOfElements = ( pSrc[0] * pSrc[0] + pSrc[1] * pSrc[1] + ... + pSrc[blockSize-1] ) / blockSize
45 </pre>
46
47 There are separate functions for floating point, Q31, and Q15 data types.
48 */
49
50 /**
51 @addtogroup variance
52 @{
53 */
54
55 /**
56 @brief Variance of the elements of a floating-point vector.
57 @param[in] pSrc points to the input vector
58 @param[in] blockSize number of samples in input vector
59 @param[out] pResult variance value returned here
60 */
61 #if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
62
63 #include "arm_helium_utils.h"
64
arm_var_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)65 ARM_DSP_ATTRIBUTE void arm_var_f32(
66 const float32_t * pSrc,
67 uint32_t blockSize,
68 float32_t * pResult)
69 {
70 uint32_t blkCnt; /* loop counters */
71 f32x4_t vecSrc;
72 f32x4_t sumVec = vdupq_n_f32(0.0f);
73 float32_t fMean;
74 float32_t sum = 0.0f; /* accumulator */
75 float32_t in; /* Temporary variable to store input value */
76
77 if (blockSize <= 1U) {
78 *pResult = 0;
79 return;
80 }
81
82 arm_mean_f32(pSrc, blockSize, &fMean);
83
84 /* Compute 4 outputs at a time */
85 blkCnt = blockSize >> 2U;
86 while (blkCnt > 0U)
87 {
88
89 vecSrc = vldrwq_f32(pSrc);
90 /*
91 * sum lanes
92 */
93 vecSrc = vsubq(vecSrc, fMean);
94 sumVec = vfmaq(sumVec, vecSrc, vecSrc);
95
96 blkCnt --;
97 pSrc += 4;
98 }
99
100 sum = vecAddAcrossF32Mve(sumVec);
101
102 /*
103 * tail
104 */
105 blkCnt = blockSize & 0x3;
106 while (blkCnt > 0U)
107 {
108 in = *pSrc++ - fMean;
109 sum += in * in;
110
111 /* Decrement loop counter */
112 blkCnt--;
113 }
114
115 /* Variance */
116 *pResult = sum / (float32_t) (blockSize - 1);
117 }
118 #else
119 #if defined(ARM_MATH_NEON_EXPERIMENTAL) && !defined(ARM_MATH_AUTOVECTORIZE)
arm_var_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)120 ARM_DSP_ATTRIBUTE void arm_var_f32(
121 const float32_t * pSrc,
122 uint32_t blockSize,
123 float32_t * pResult)
124 {
125 float32_t mean;
126
127 float32_t sum = 0.0f; /* accumulator */
128 float32_t in; /* Temporary variable to store input value */
129 uint32_t blkCnt; /* loop counter */
130
131 float32x4_t sumV = vdupq_n_f32(0.0f); /* Temporary result storage */
132 float32x2_t sumV2;
133 float32x4_t inV;
134 float32x4_t avg;
135
136 arm_mean_f32(pSrc,blockSize,&mean);
137 avg = vdupq_n_f32(mean);
138
139 blkCnt = blockSize >> 2U;
140
141 /* Compute 4 outputs at a time.
142 ** a second loop below computes the remaining 1 to 3 samples. */
143 while (blkCnt > 0U)
144 {
145 /* C = A[0] * A[0] + A[1] * A[1] + A[2] * A[2] + ... + A[blockSize-1] * A[blockSize-1] */
146 /* Compute Power and then store the result in a temporary variable, sum. */
147 inV = vld1q_f32(pSrc);
148 inV = vsubq_f32(inV, avg);
149 sumV = vmlaq_f32(sumV, inV, inV);
150 pSrc += 4;
151
152 /* Decrement the loop counter */
153 blkCnt--;
154 }
155
156 sumV2 = vpadd_f32(vget_low_f32(sumV),vget_high_f32(sumV));
157 sum = vget_lane_f32(sumV2, 0) + vget_lane_f32(sumV2, 1);
158
159 /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
160 ** No loop unrolling is used. */
161 blkCnt = blockSize % 0x4U;
162
163 while (blkCnt > 0U)
164 {
165 /* C = A[0] * A[0] + A[1] * A[1] + A[2] * A[2] + ... + A[blockSize-1] * A[blockSize-1] */
166 /* compute power and then store the result in a temporary variable, sum. */
167 in = *pSrc++;
168 in = in - mean;
169 sum += in * in;
170
171 /* Decrement the loop counter */
172 blkCnt--;
173 }
174
175 /* Variance */
176 *pResult = sum / (float32_t)(blockSize - 1.0f);
177
178 }
179
180 #else
arm_var_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)181 ARM_DSP_ATTRIBUTE void arm_var_f32(
182 const float32_t * pSrc,
183 uint32_t blockSize,
184 float32_t * pResult)
185 {
186 uint32_t blkCnt; /* Loop counter */
187 float32_t sum = 0.0f; /* Temporary result storage */
188 float32_t fSum = 0.0f;
189 float32_t fMean, fValue;
190 const float32_t * pInput = pSrc;
191
192 if (blockSize <= 1U)
193 {
194 *pResult = 0;
195 return;
196 }
197
198 #if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE)
199
200 /* Loop unrolling: Compute 4 outputs at a time */
201 blkCnt = blockSize >> 2U;
202
203 while (blkCnt > 0U)
204 {
205 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
206
207 sum += *pInput++;
208 sum += *pInput++;
209 sum += *pInput++;
210 sum += *pInput++;
211
212
213 /* Decrement loop counter */
214 blkCnt--;
215 }
216
217 /* Loop unrolling: Compute remaining outputs */
218 blkCnt = blockSize % 0x4U;
219
220 #else
221
222 /* Initialize blkCnt with number of samples */
223 blkCnt = blockSize;
224
225 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
226
227 while (blkCnt > 0U)
228 {
229 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */
230
231 sum += *pInput++;
232
233 /* Decrement loop counter */
234 blkCnt--;
235 }
236
237 /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) / blockSize */
238 fMean = sum / (float32_t) blockSize;
239
240 pInput = pSrc;
241
242 #if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE)
243
244 /* Loop unrolling: Compute 4 outputs at a time */
245 blkCnt = blockSize >> 2U;
246
247 while (blkCnt > 0U)
248 {
249 fValue = *pInput++ - fMean;
250 fSum += fValue * fValue;
251
252 fValue = *pInput++ - fMean;
253 fSum += fValue * fValue;
254
255 fValue = *pInput++ - fMean;
256 fSum += fValue * fValue;
257
258 fValue = *pInput++ - fMean;
259 fSum += fValue * fValue;
260
261 /* Decrement loop counter */
262 blkCnt--;
263 }
264
265 /* Loop unrolling: Compute remaining outputs */
266 blkCnt = blockSize % 0x4U;
267
268 #else
269
270 /* Initialize blkCnt with number of samples */
271 blkCnt = blockSize;
272
273 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
274
275 while (blkCnt > 0U)
276 {
277 fValue = *pInput++ - fMean;
278 fSum += fValue * fValue;
279
280 /* Decrement loop counter */
281 blkCnt--;
282 }
283
284 /* Variance */
285 *pResult = fSum / (float32_t)(blockSize - 1.0f);
286 }
287 #endif /* #if defined(ARM_MATH_NEON) */
288 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
289
290 /**
291 @} end of variance group
292 */
293
