1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmax_q31.c
4 * Description: Maximum value of absolute values of a Q31 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 @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 @param[out] pIndex index of maximum value returned here
46 */
47
48 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
49
50 #include "arm_helium_utils.h"
arm_absmax_q31(const q31_t * pSrc,uint32_t blockSize,q31_t * pResult,uint32_t * pIndex)51 void arm_absmax_q31(
52 const q31_t * pSrc,
53 uint32_t blockSize,
54 q31_t * pResult,
55 uint32_t * pIndex)
56 {
57 int32_t blkCnt; /* loop counters */
58 q31x4_t extremValVec = vdupq_n_s32(Q31_ABSMIN);
59 q31_t maxValue = Q31_ABSMIN;
60 uint32x4_t indexVec;
61 uint32x4_t extremIdxVec;
62 mve_pred16_t p0;
63 uint32_t extremIdxArr[4];
64
65 indexVec = vidupq_u32(0U, 1);
66
67 blkCnt = blockSize;
68 do {
69 mve_pred16_t p = vctp32q(blkCnt);
70 q31x4_t extremIdxVal = vld1q_z_s32(pSrc, p);
71
72 extremIdxVal = vqabsq(extremIdxVal);
73 /*
74 * Get current max per lane and current index per lane
75 * when a max is selected
76 */
77 p0 = vcmpgtq_m(extremIdxVal, extremValVec, p);
78
79 extremValVec = vorrq_m(extremValVec, extremIdxVal, extremIdxVal, p0);
80 /* store per-lane extrema indexes */
81 vst1q_p_u32(extremIdxArr, indexVec, p0);
82
83 indexVec += 4;
84 pSrc += 4;
85 blkCnt -= 4;
86 }
87 while (blkCnt > 0);
88
89
90 /* Get max value across the vector */
91 maxValue = vmaxvq(maxValue, extremValVec);
92
93 /* set index for lower values to max possible index */
94 p0 = vcmpgeq(extremValVec, maxValue);
95 extremIdxVec = vld1q_u32(extremIdxArr);
96
97 indexVec = vpselq(extremIdxVec, vdupq_n_u32(blockSize - 1), p0);
98 *pIndex = vminvq(blockSize - 1, indexVec);
99 *pResult = maxValue;
100 }
101 #else
102 #if defined(ARM_MATH_DSP)
arm_absmax_q31(const q31_t * pSrc,uint32_t blockSize,q31_t * pResult,uint32_t * pIndex)103 void arm_absmax_q31(
104 const q31_t * pSrc,
105 uint32_t blockSize,
106 q31_t * pResult,
107 uint32_t * pIndex)
108 {
109 q31_t cur_absmax, out; /* Temporary variables to store the output value. */\
110 uint32_t blkCnt, outIndex; /* Loop counter */ \
111 uint32_t index; /* index of maximum value */ \
112 \
113 /* Initialize index value to zero. */ \
114 outIndex = 0U; \
115 /* Load first input value that act as reference value for comparision */ \
116 out = *pSrc++; \
117 out = (out > 0) ? out : (q31_t)__QSUB(0, out); \
118 /* Initialize index of extrema value. */ \
119 index = 0U; \
120 \
121 /* Loop unrolling: Compute 4 outputs at a time */ \
122 blkCnt = (blockSize - 1U) >> 2U; \
123 \
124 while (blkCnt > 0U) \
125 { \
126 /* Initialize cur_absmax to next consecutive values one by one */ \
127 cur_absmax = *pSrc++; \
128 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
129 /* compare for the extrema value */ \
130 if (cur_absmax > out) \
131 { \
132 /* Update the extrema value and it's index */ \
133 out = cur_absmax; \
134 outIndex = index + 1U; \
135 } \
136 \
137 cur_absmax = *pSrc++; \
138 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
139 if (cur_absmax > out) \
140 { \
141 out = cur_absmax; \
142 outIndex = index + 2U; \
143 } \
144 \
145 cur_absmax = *pSrc++; \
146 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
147 if (cur_absmax > out) \
148 { \
149 out = cur_absmax; \
150 outIndex = index + 3U; \
151 } \
152 \
153 cur_absmax = *pSrc++; \
154 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
155 if (cur_absmax > out) \
156 { \
157 out = cur_absmax; \
158 outIndex = index + 4U; \
159 } \
160 \
161 index += 4U; \
162 \
163 /* Decrement loop counter */ \
164 blkCnt--; \
165 } \
166 \
167 /* Loop unrolling: Compute remaining outputs */ \
168 blkCnt = (blockSize - 1U) % 4U; \
169 \
170 \
171 while (blkCnt > 0U) \
172 { \
173 cur_absmax = *pSrc++; \
174 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q31_t)__QSUB(0, cur_absmax); \
175 if (cur_absmax > out) \
176 { \
177 out = cur_absmax; \
178 outIndex = blockSize - blkCnt; \
179 } \
180 \
181 /* Decrement loop counter */ \
182 blkCnt--; \
183 } \
184 \
185 /* Store the extrema value and it's index into destination pointers */ \
186 *pResult = out; \
187 *pIndex = outIndex;
188 }
189 #else
arm_absmax_q31(const q31_t * pSrc,uint32_t blockSize,q31_t * pResult,uint32_t * pIndex)190 void arm_absmax_q31(
191 const q31_t * pSrc,
192 uint32_t blockSize,
193 q31_t * pResult,
194 uint32_t * pIndex)
195 {
196 q31_t maxVal, out; /* Temporary variables to store the output value. */
197 uint32_t blkCnt, outIndex; /* Loop counter */
198
199
200 /* Initialise index value to zero. */
201 outIndex = 0U;
202 /* Load first input value that act as reference value for comparision */
203 out = (*pSrc > 0) ? *pSrc : ((*pSrc == INT32_MIN) ? INT32_MAX : -*pSrc);
204 pSrc++;
205
206 /* Initialize blkCnt with number of samples */
207 blkCnt = (blockSize - 1U);
208
209 while (blkCnt > 0U)
210 {
211 /* Initialize maxVal to the next consecutive values one by one */
212 maxVal = (*pSrc > 0) ? *pSrc : ((*pSrc == INT32_MIN) ? INT32_MAX : -*pSrc);
213 pSrc++;
214
215 /* compare for the maximum value */
216 if (out < maxVal)
217 {
218 /* Update the maximum value and it's index */
219 out = maxVal;
220 outIndex = blockSize - blkCnt;
221 }
222
223 /* Decrement loop counter */
224 blkCnt--;
225 }
226
227 /* Store the maximum value and it's index into destination pointers */
228 *pResult = out;
229 *pIndex = outIndex;
230 }
231 #endif /* defined(ARM_MATH_DSP) */
232 #endif /* defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) */
233 /**
234 @} end of AbsMax group
235 */
236
