1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmin_no_idx_q7.c
4 * Description: Minimum value of absolute values of a Q7 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 Q7 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 <stdint.h>
50 #include "arm_helium_utils.h"
51
52
53
arm_absmin_no_idx_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult)54 void arm_absmin_no_idx_q7(
55 const q7_t * pSrc,
56 uint32_t blockSize,
57 q7_t * pResult)
58 {
59 int32_t blkCnt; /* loop counters */
60 q7x16_t vecSrc;
61 q7_t const *pSrcVec;
62 uint8x16_t curExtremValVec = vdupq_n_s8(Q7_ABSMAX);
63 q7_t minValue = Q7_ABSMAX;
64 mve_pred16_t p0;
65
66
67 pSrcVec = (q7_t const *) pSrc;
68 blkCnt = blockSize >> 4;
69 while (blkCnt > 0)
70 {
71 vecSrc = vld1q(pSrcVec);
72 pSrcVec += 16;
73 /*
74 * update per-lane min.
75 */
76 curExtremValVec = vminaq(curExtremValVec, vecSrc);
77 /*
78 * Decrement the blockSize loop counter
79 */
80 blkCnt--;
81 }
82 /*
83 * tail
84 * (will be merged thru tail predication)
85 */
86 blkCnt = blockSize & 0xF;
87 if (blkCnt > 0)
88 {
89 vecSrc = vld1q(pSrcVec);
90 pSrcVec += 16;
91 p0 = vctp8q(blkCnt);
92 /*
93 * Get current min per lane and current index per lane
94 * when a min is selected
95 */
96 curExtremValVec = vminaq_m(curExtremValVec, vecSrc, p0);
97 }
98 /*
99 * Get min value across the vector
100 */
101 minValue = vminavq(minValue, (q7x16_t)curExtremValVec);
102 *pResult = minValue;
103 }
104
105 #else
106 #if defined(ARM_MATH_DSP)
arm_absmin_no_idx_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult)107 void arm_absmin_no_idx_q7(
108 const q7_t * pSrc,
109 uint32_t blockSize,
110 q7_t * pResult)
111 {
112 q7_t cur_absmin, out; /* Temporary variables to store the output value. */\
113 uint32_t blkCnt; /* Loop counter */ \
114 \
115 \
116 /* Load first input value that act as reference value for comparision */ \
117 out = *pSrc++; \
118 out = (out > 0) ? out : (q7_t)__QSUB8(0, out); \
119 \
120 /* Loop unrolling: Compute 4 outputs at a time */ \
121 blkCnt = (blockSize - 1U) >> 2U; \
122 \
123 while (blkCnt > 0U) \
124 { \
125 /* Initialize cur_absmin to next consecutive values one by one */ \
126 cur_absmin = *pSrc++; \
127 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
128 /* compare for the extrema value */ \
129 if (cur_absmin < out) \
130 { \
131 /* Update the extrema value and it's index */ \
132 out = cur_absmin; \
133 } \
134 \
135 cur_absmin = *pSrc++; \
136 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
137 if (cur_absmin < out) \
138 { \
139 out = cur_absmin; \
140 } \
141 \
142 cur_absmin = *pSrc++; \
143 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
144 if (cur_absmin < out) \
145 { \
146 out = cur_absmin; \
147 } \
148 \
149 cur_absmin = *pSrc++; \
150 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
151 if (cur_absmin < out) \
152 { \
153 out = cur_absmin; \
154 } \
155 \
156 \
157 /* Decrement loop counter */ \
158 blkCnt--; \
159 } \
160 \
161 /* Loop unrolling: Compute remaining outputs */ \
162 blkCnt = (blockSize - 1U) % 4U; \
163 \
164 \
165 while (blkCnt > 0U) \
166 { \
167 cur_absmin = *pSrc++; \
168 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
169 if (cur_absmin < out) \
170 { \
171 out = cur_absmin; \
172 } \
173 \
174 /* Decrement loop counter */ \
175 blkCnt--; \
176 } \
177 \
178 /* Store the extrema value and it's index into destination pointers */ \
179 *pResult = out; \
180 }
181 #else
arm_absmin_no_idx_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult)182 void arm_absmin_no_idx_q7(
183 const q7_t * pSrc,
184 uint32_t blockSize,
185 q7_t * pResult)
186 {
187 q7_t minVal, out; /* Temporary variables to store the output value. */
188 uint32_t blkCnt; /* Loop counter */
189
190
191 /* Load first input value that act as reference value for comparision */
192 out = (*pSrc > 0) ? *pSrc : ((*pSrc == (q7_t) 0x80) ? (q7_t) 0x7f : -*pSrc);
193 pSrc++;
194
195 /* Initialize blkCnt with number of samples */
196 blkCnt = (blockSize - 1U);
197
198 while (blkCnt > 0U)
199 {
200 /* Initialize minVal to the next consecutive values one by one */
201 minVal = (*pSrc > 0) ? *pSrc : ((*pSrc == (q7_t) 0x80) ? (q7_t) 0x7f : -*pSrc);
202 pSrc++;
203
204 /* compare for the minimum value */
205 if (out > minVal)
206 {
207 /* Update the minimum value and it's index */
208 out = minVal;
209 }
210
211 /* Decrement loop counter */
212 blkCnt--;
213 }
214
215 /* Store the minimum value and it's index into destination pointers */
216 *pResult = out;
217 }
218 #endif /* defined(ARM_MATH_DSP) */
219 #endif /* defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) */
220 /**
221 @} end of AbsMin group
222 */
223
