1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmin_q15.c
4 * Description: Minimum value of absolute values 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 /**
37 @addtogroup AbsMin
38 @{
39 */
40
41 /**
42 @brief Minimum value of absolute values of a Q15 vector.
43 @param[in] pSrc points to the input vector
44 @param[in] blockSize number of samples in input vector
45 @param[out] pResult minimum value returned here
46 @param[out] pIndex index of minimum value returned here
47 @return none
48 */
49 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
50
51 #include "arm_helium_utils.h"
52
arm_absmin_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult,uint32_t * pIndex)53 void arm_absmin_q15(
54 const q15_t * pSrc,
55 uint32_t blockSize,
56 q15_t * pResult,
57 uint32_t * pIndex)
58 {
59 uint16_t blkCnt; /* loop counters */
60 q15x8_t vecSrc;
61 q15_t const *pSrcVec;
62 q15x8_t curExtremValVec = vdupq_n_s16(Q15_ABSMAX);
63 q15_t minValue = Q15_ABSMAX;
64 uint16_t idx = blockSize;
65 uint16x8_t indexVec;
66 uint16x8_t curExtremIdxVec;
67 uint32_t startIdx = 0;
68 mve_pred16_t p0;
69
70
71 indexVec = vidupq_wb_u16(&startIdx, 1);
72 curExtremIdxVec = vdupq_n_u16(0);
73
74 pSrcVec = (q15_t const *) pSrc;
75 blkCnt = blockSize >> 3;
76 while (blkCnt > 0U)
77 {
78 vecSrc = vld1q(pSrcVec);
79 pSrcVec += 8;
80 vecSrc = vabsq(vecSrc);
81 /*
82 * Get current min per lane and current index per lane
83 * when a min is selected
84 */
85 p0 = vcmpleq(vecSrc, curExtremValVec);
86 curExtremValVec = vpselq(vecSrc, curExtremValVec, p0);
87 curExtremIdxVec = vpselq(indexVec, curExtremIdxVec, p0);
88
89 indexVec = vidupq_wb_u16(&startIdx, 1);
90 /*
91 * Decrement the blockSize loop counter
92 */
93 blkCnt--;
94 }
95 /*
96 * tail
97 * (will be merged thru tail predication)
98 */
99 blkCnt = blockSize & 7;
100 if (blkCnt > 0U)
101 {
102 vecSrc = vld1q(pSrcVec);
103 pSrcVec += 8;
104 vecSrc = vabsq(vecSrc);
105
106 p0 = vctp16q(blkCnt);
107 /*
108 * Get current min per lane and current index per lane
109 * when a min is selected
110 */
111 p0 = vcmpleq_m(vecSrc, curExtremValVec, p0);
112 curExtremValVec = vpselq(vecSrc, curExtremValVec, p0);
113 curExtremIdxVec = vpselq(indexVec, curExtremIdxVec, p0);
114 }
115 /*
116 * Get min value across the vector
117 */
118 minValue = vminvq(minValue, curExtremValVec);
119 /*
120 * set index for lower values to min possible index
121 */
122 p0 = vcmpleq(curExtremValVec, minValue);
123 indexVec = vpselq(curExtremIdxVec, vdupq_n_u16(blockSize), p0);
124 /*
125 * Get min index which is thus for a min value
126 */
127 idx = vminvq(idx, indexVec);
128 /*
129 * Save result
130 */
131 *pIndex = idx;
132 *pResult = minValue;
133 }
134
135 #else
136 #if defined(ARM_MATH_DSP)
arm_absmin_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult,uint32_t * pIndex)137 void arm_absmin_q15(
138 const q15_t * pSrc,
139 uint32_t blockSize,
140 q15_t * pResult,
141 uint32_t * pIndex)
142 {
143 q15_t cur_absmin, out; /* Temporary variables to store the output value. */\
144 uint32_t blkCnt, outIndex; /* Loop counter */ \
145 uint32_t index; /* index of maximum value */ \
146 \
147 /* Initialize index value to zero. */ \
148 outIndex = 0U; \
149 /* Load first input value that act as reference value for comparision */ \
150 out = *pSrc++; \
151 out = (out > 0) ? out : (q15_t)__QSUB16(0, out); \
152 /* Initialize index of extrema value. */ \
153 index = 0U; \
154 \
155 /* Loop unrolling: Compute 4 outputs at a time */ \
156 blkCnt = (blockSize - 1U) >> 2U; \
157 \
158 while (blkCnt > 0U) \
159 { \
160 /* Initialize cur_absmin to next consecutive values one by one */ \
161 cur_absmin = *pSrc++; \
162 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
163 /* compare for the extrema value */ \
164 if (cur_absmin < out) \
165 { \
166 /* Update the extrema value and it's index */ \
167 out = cur_absmin; \
168 outIndex = index + 1U; \
169 } \
170 \
171 cur_absmin = *pSrc++; \
172 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
173 if (cur_absmin < out) \
174 { \
175 out = cur_absmin; \
176 outIndex = index + 2U; \
177 } \
178 \
179 cur_absmin = *pSrc++; \
180 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
181 if (cur_absmin < out) \
182 { \
183 out = cur_absmin; \
184 outIndex = index + 3U; \
185 } \
186 \
187 cur_absmin = *pSrc++; \
188 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
189 if (cur_absmin < out) \
190 { \
191 out = cur_absmin; \
192 outIndex = index + 4U; \
193 } \
194 \
195 index += 4U; \
196 \
197 /* Decrement loop counter */ \
198 blkCnt--; \
199 } \
200 \
201 /* Loop unrolling: Compute remaining outputs */ \
202 blkCnt = (blockSize - 1U) % 4U; \
203 \
204 \
205 while (blkCnt > 0U) \
206 { \
207 cur_absmin = *pSrc++; \
208 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
209 if (cur_absmin < out) \
210 { \
211 out = cur_absmin; \
212 outIndex = blockSize - blkCnt; \
213 } \
214 \
215 /* Decrement loop counter */ \
216 blkCnt--; \
217 } \
218 \
219 /* Store the extrema value and it's index into destination pointers */ \
220 *pResult = out; \
221 *pIndex = outIndex;
222 }
223 #else
arm_absmin_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult,uint32_t * pIndex)224 void arm_absmin_q15(
225 const q15_t * pSrc,
226 uint32_t blockSize,
227 q15_t * pResult,
228 uint32_t * pIndex)
229 {
230 q15_t minVal, out; /* Temporary variables to store the output value. */
231 uint32_t blkCnt, outIndex; /* Loop counter */
232
233
234 /* Initialise index value to zero. */
235 outIndex = 0U;
236 /* Load first input value that act as reference value for comparision */
237 out = (*pSrc > 0) ? *pSrc : ((*pSrc == (q15_t) 0x8000) ? 0x7fff : -*pSrc);
238 pSrc++;
239
240 /* Initialize blkCnt with number of samples */
241 blkCnt = (blockSize - 1U);
242
243 while (blkCnt > 0U)
244 {
245 /* Initialize minVal to the next consecutive values one by one */
246 minVal = (*pSrc > 0) ? *pSrc : ((*pSrc == (q15_t) 0x8000) ? 0x7fff : -*pSrc);
247 pSrc++;
248
249 /* compare for the minimum value */
250 if (out > minVal)
251 {
252 /* Update the minimum value and it's index */
253 out = minVal;
254 outIndex = blockSize - blkCnt;
255 }
256
257 /* Decrement loop counter */
258 blkCnt--;
259 }
260
261 /* Store the minimum value and it's index into destination pointers */
262 *pResult = out;
263 *pIndex = outIndex;
264 }
265 #endif /* defined(ARM_MATH_DSP) */
266 #endif /* defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) */
267 /**
268 @} end of AbsMin group
269 */
270
