1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmin_no_idx_q15.c
4 * Description: Minimum value of absolute values of a Q15 vector
5 *
6 * $Date: 16 November 2021
7 * $Revision: V1.10.0
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 */
47 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
48
49 #include "arm_helium_utils.h"
50
arm_absmin_no_idx_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult)51 void arm_absmin_no_idx_q15(
52 const q15_t * pSrc,
53 uint32_t blockSize,
54 q15_t * pResult)
55 {
56 uint16_t blkCnt; /* loop counters */
57 q15x8_t vecSrc;
58 q15_t const *pSrcVec;
59 uint16x8_t curExtremValVec = vdupq_n_s16(Q15_ABSMAX);
60 q15_t minValue = Q15_ABSMAX;
61 mve_pred16_t p0;
62
63
64 pSrcVec = (q15_t const *) pSrc;
65 blkCnt = blockSize >> 3;
66 while (blkCnt > 0)
67 {
68 vecSrc = vld1q(pSrcVec);
69 pSrcVec += 8;
70 /*
71 * update per-lane min.
72 */
73 curExtremValVec = vminaq(curExtremValVec, vecSrc);
74 /*
75 * Decrement the blockSize loop counter
76 */
77 blkCnt--;
78 }
79 /*
80 * tail
81 * (will be merged thru tail predication)
82 */
83 blkCnt = blockSize & 7;
84 if (blkCnt > 0)
85 {
86 vecSrc = vld1q(pSrcVec);
87 pSrcVec += 8;
88 p0 = vctp16q(blkCnt);
89 /*
90 * Get current min per lane and current index per lane
91 * when a min is selected
92 */
93 curExtremValVec = vminaq_m(curExtremValVec, vecSrc, p0);
94 }
95 /*
96 * Get min value across the vector
97 */
98 minValue = vminavq(minValue, (q15x8_t)curExtremValVec);
99 *pResult = minValue;
100 }
101
102 #else
103 #if defined(ARM_MATH_DSP)
arm_absmin_no_idx_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult)104 void arm_absmin_no_idx_q15(
105 const q15_t * pSrc,
106 uint32_t blockSize,
107 q15_t * pResult)
108 {
109 q15_t cur_absmin, out; /* Temporary variables to store the output value. */\
110 uint32_t blkCnt; /* Loop counter */ \
111 \
112 \
113 /* Load first input value that act as reference value for comparision */ \
114 out = *pSrc++; \
115 out = (out > 0) ? out : (q15_t)__QSUB16(0, out); \
116 \
117 \
118 /* Loop unrolling: Compute 4 outputs at a time */ \
119 blkCnt = (blockSize - 1U) >> 2U; \
120 \
121 while (blkCnt > 0U) \
122 { \
123 /* Initialize cur_absmin to next consecutive values one by one */ \
124 cur_absmin = *pSrc++; \
125 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
126 /* compare for the extrema value */ \
127 if (cur_absmin < out) \
128 { \
129 /* Update the extrema value and it's index */ \
130 out = cur_absmin; \
131 } \
132 \
133 cur_absmin = *pSrc++; \
134 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
135 if (cur_absmin < out) \
136 { \
137 out = cur_absmin; \
138 } \
139 \
140 cur_absmin = *pSrc++; \
141 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
142 if (cur_absmin < out) \
143 { \
144 out = cur_absmin; \
145 } \
146 \
147 cur_absmin = *pSrc++; \
148 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
149 if (cur_absmin < out) \
150 { \
151 out = cur_absmin; \
152 } \
153 \
154 \
155 /* Decrement loop counter */ \
156 blkCnt--; \
157 } \
158 \
159 /* Loop unrolling: Compute remaining outputs */ \
160 blkCnt = (blockSize - 1U) % 4U; \
161 \
162 \
163 while (blkCnt > 0U) \
164 { \
165 cur_absmin = *pSrc++; \
166 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q15_t)__QSUB16(0, cur_absmin); \
167 if (cur_absmin < out) \
168 { \
169 out = cur_absmin; \
170 } \
171 \
172 /* Decrement loop counter */ \
173 blkCnt--; \
174 } \
175 \
176 /* Store the extrema value and it's index into destination pointers */ \
177 *pResult = out; \
178 }
179 #else
arm_absmin_no_idx_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult)180 void arm_absmin_no_idx_q15(
181 const q15_t * pSrc,
182 uint32_t blockSize,
183 q15_t * pResult)
184 {
185 q15_t minVal, out; /* Temporary variables to store the output value. */
186 uint32_t blkCnt; /* Loop counter */
187
188
189
190 /* Load first input value that act as reference value for comparision */
191 out = (*pSrc > 0) ? *pSrc : ((*pSrc == (q15_t) 0x8000) ? 0x7fff : -*pSrc);
192 pSrc++;
193
194 /* Initialize blkCnt with number of samples */
195 blkCnt = (blockSize - 1U);
196
197 while (blkCnt > 0U)
198 {
199 /* Initialize minVal to the next consecutive values one by one */
200 minVal = (*pSrc > 0) ? *pSrc : ((*pSrc == (q15_t) 0x8000) ? 0x7fff : -*pSrc);
201 pSrc++;
202
203 /* compare for the minimum value */
204 if (out > minVal)
205 {
206 /* Update the minimum value and it's index */
207 out = minVal;
208 }
209
210 /* Decrement loop counter */
211 blkCnt--;
212 }
213
214 /* Store the minimum value and it's index into destination pointers */
215 *pResult = out;
216 }
217 #endif /* defined(ARM_MATH_DSP) */
218 #endif /* defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) */
219 /**
220 @} end of AbsMin group
221 */
222
