1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmax_no_idx_q31.c
4 * Description: Maximum value of absolute values of a Q31 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 @addtogroup AbsMax
37 @{
38 */
39
40 /**
41 @brief Maximum value of absolute values of a Q31 vector.
42 @param[in] pSrc points to the input vector
43 @param[in] blockSize number of samples in input vector
44 @param[out] pResult maximum value returned here
45 */
46
47 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
48
49 #include "arm_helium_utils.h"
arm_absmax_no_idx_q31(const q31_t * pSrc,uint32_t blockSize,q31_t * pResult)50 ARM_DSP_ATTRIBUTE void arm_absmax_no_idx_q31(
51 const q31_t * pSrc,
52 uint32_t blockSize,
53 q31_t * pResult)
54 {
55 int32_t blkCnt; /* loop counters */
56 q31x4_t vecSrc;
57 q31_t const *pSrcVec;
58 uint32x4_t curExtremValVec = vdupq_n_u32(Q31_ABSMIN);
59 uint32_t maxValue = Q31_ABSMIN;
60
61
62 pSrcVec = (q31_t const *) pSrc;
63 blkCnt = blockSize ;
64 while (blkCnt > 0)
65 {
66 mve_pred16_t p = vctp32q(blkCnt);
67 vecSrc = vldrwq_z_s32(pSrcVec,p);
68 pSrcVec += 4;
69 /*
70 * update per-lane max.
71 */
72 curExtremValVec = vmaxaq_m(curExtremValVec, vecSrc,p);
73 /*
74 * Decrement the blockSize loop counter
75 */
76 blkCnt -= 4;
77 }
78 /*
79 * Get max value across the vector
80 */
81 maxValue = vmaxvq(maxValue, curExtremValVec);
82 *pResult = clip_q63_to_q31((q63_t)maxValue);
83 }
84 #else
85 #if defined(ARM_MATH_DSP)
arm_absmax_no_idx_q31(const q31_t * pSrc,uint32_t blockSize,q31_t * pResult)86 ARM_DSP_ATTRIBUTE void arm_absmax_no_idx_q31(
87 const q31_t * pSrc,
88 uint32_t blockSize,
89 q31_t * pResult)
90 {
91 q31_t cur_absmax, out; /* Temporary variables to store the output value. */\
92 uint32_t blkCnt; /* Loop counter */ \
93 \
94 \
95 /* Load first input value that act as reference value for comparision */ \
96 out = *pSrc++; \
97 out = (out > 0) ? out : (q31_t)__QSUB(0, out); \
98 \
99 \
100 /* Loop unrolling: Compute 4 outputs at a time */ \
101 blkCnt = (blockSize - 1U) >> 2U; \
102 \
103 while (blkCnt > 0U) \
104 { \
105 /* Initialize cur_absmax to next consecutive values one by one */ \
106 cur_absmax = *pSrc++; \
107 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
108 /* compare for the extrema value */ \
109 if (cur_absmax > out) \
110 { \
111 /* Update the extrema value and it's index */ \
112 out = cur_absmax; \
113 } \
114 \
115 cur_absmax = *pSrc++; \
116 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
117 if (cur_absmax > out) \
118 { \
119 out = cur_absmax; \
120 } \
121 \
122 cur_absmax = *pSrc++; \
123 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
124 if (cur_absmax > out) \
125 { \
126 out = cur_absmax; \
127 } \
128 \
129 cur_absmax = *pSrc++; \
130 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
131 if (cur_absmax > out) \
132 { \
133 out = cur_absmax; \
134 } \
135 \
136 \
137 /* Decrement loop counter */ \
138 blkCnt--; \
139 } \
140 \
141 /* Loop unrolling: Compute remaining outputs */ \
142 blkCnt = (blockSize - 1U) % 4U; \
143 \
144 \
145 while (blkCnt > 0U) \
146 { \
147 cur_absmax = *pSrc++; \
148 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
149 if (cur_absmax > out) \
150 { \
151 out = cur_absmax; \
152 } \
153 \
154 /* Decrement loop counter */ \
155 blkCnt--; \
156 } \
157 \
158 /* Store the extrema value and it's index into destination pointers */ \
159 *pResult = out; \
160 }
161 #else
arm_absmax_no_idx_q31(const q31_t * pSrc,uint32_t blockSize,q31_t * pResult)162 ARM_DSP_ATTRIBUTE void arm_absmax_no_idx_q31(
163 const q31_t * pSrc,
164 uint32_t blockSize,
165 q31_t * pResult)
166 {
167 q31_t maxVal, out; /* Temporary variables to store the output value. */
168 uint32_t blkCnt; /* Loop counter */
169
170
171
172 /* Load first input value that act as reference value for comparision */
173 out = (*pSrc > 0) ? *pSrc : ((*pSrc == INT32_MIN) ? INT32_MAX : -*pSrc);
174 pSrc++;
175
176 /* Initialize blkCnt with number of samples */
177 blkCnt = (blockSize - 1U);
178
179 while (blkCnt > 0U)
180 {
181 /* Initialize maxVal to the next consecutive values one by one */
182 maxVal = (*pSrc > 0) ? *pSrc : ((*pSrc == INT32_MIN) ? INT32_MAX : -*pSrc);
183 pSrc++;
184
185 /* compare for the maximum value */
186 if (out < maxVal)
187 {
188 /* Update the maximum value and it's index */
189 out = maxVal;
190 }
191
192 /* Decrement loop counter */
193 blkCnt--;
194 }
195
196 /* Store the maximum value and it's index into destination pointers */
197 *pResult = out;
198 }
199 #endif /* defined(ARM_MATH_DSP) */
200 #endif /* defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) */
201 /**
202 @} end of AbsMax group
203 */
204
