1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmax_no_idx_q7.c
4 * Description: Maximum 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 @addtogroup AbsMax
37 @{
38 */
39
40 /**
41 @brief Maximum value of absolute values of a Q7 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 @return none
46 */
47
48
49 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
50
51 #include <stdint.h>
52 #include "arm_helium_utils.h"
53
54
55
arm_absmax_no_idx_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult)56 void arm_absmax_no_idx_q7(
57 const q7_t * pSrc,
58 uint32_t blockSize,
59 q7_t * pResult)
60 {
61 int32_t blkCnt; /* loop counters */
62 q7x16_t vecSrc;
63 q7_t const *pSrcVec;
64 uint8x16_t curExtremValVec = vdupq_n_u8(Q7_ABSMIN);
65 uint8_t maxValue = Q7_ABSMIN;
66
67
68 pSrcVec = (q7_t const *) pSrc;
69 blkCnt = blockSize;
70 while (blkCnt > 0)
71 {
72 mve_pred16_t p = vctp8q(blkCnt);
73 vecSrc = vld1q_z_s8(pSrcVec,p);
74 pSrcVec += 16;
75 /*
76 * update per-lane max.
77 */
78 curExtremValVec = vmaxaq_m(curExtremValVec, vecSrc,p);
79 /*
80 * Decrement the blockSize loop counter
81 */
82 blkCnt -= 16;
83 }
84 /*
85 * Get max value across the vector
86 */
87 maxValue = vmaxvq(maxValue, curExtremValVec);
88 *pResult = __USAT(maxValue,7);
89 }
90 #else
91 #if defined(ARM_MATH_DSP)
arm_absmax_no_idx_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult)92 void arm_absmax_no_idx_q7(
93 const q7_t * pSrc,
94 uint32_t blockSize,
95 q7_t * pResult)
96 {
97 q7_t cur_absmax, out; /* Temporary variables to store the output value. */\
98 uint32_t blkCnt; /* Loop counter */ \
99 \
100 \
101 /* Load first input value that act as reference value for comparision */ \
102 out = *pSrc++; \
103 out = (out > 0) ? out : (q7_t)__QSUB8(0, out); \
104 \
105 \
106 /* Loop unrolling: Compute 4 outputs at a time */ \
107 blkCnt = (blockSize - 1U) >> 2U; \
108 \
109 while (blkCnt > 0U) \
110 { \
111 /* Initialize cur_absmax to next consecutive values one by one */ \
112 cur_absmax = *pSrc++; \
113 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q7_t)__QSUB8(0, cur_absmax); \
114 /* compare for the extrema value */ \
115 if (cur_absmax > out) \
116 { \
117 /* Update the extrema value and it's index */ \
118 out = cur_absmax; \
119 } \
120 \
121 cur_absmax = *pSrc++; \
122 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q7_t)__QSUB8(0, cur_absmax); \
123 if (cur_absmax > out) \
124 { \
125 out = cur_absmax; \
126 } \
127 \
128 cur_absmax = *pSrc++; \
129 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q7_t)__QSUB8(0, cur_absmax); \
130 if (cur_absmax > out) \
131 { \
132 out = cur_absmax; \
133 } \
134 \
135 cur_absmax = *pSrc++; \
136 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q7_t)__QSUB8(0, cur_absmax); \
137 if (cur_absmax > out) \
138 { \
139 out = cur_absmax; \
140 } \
141 \
142 \
143 /* Decrement loop counter */ \
144 blkCnt--; \
145 } \
146 \
147 /* Loop unrolling: Compute remaining outputs */ \
148 blkCnt = (blockSize - 1U) % 4U; \
149 \
150 \
151 while (blkCnt > 0U) \
152 { \
153 cur_absmax = *pSrc++; \
154 cur_absmax = (cur_absmax > 0) ? cur_absmax : (q7_t)__QSUB8(0, cur_absmax); \
155 if (cur_absmax > out) \
156 { \
157 out = cur_absmax; \
158 } \
159 \
160 /* Decrement loop counter */ \
161 blkCnt--; \
162 } \
163 \
164 /* Store the extrema value and it's index into destination pointers */ \
165 *pResult = out; \
166 }
167 #else
arm_absmax_no_idx_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult)168 void arm_absmax_no_idx_q7(
169 const q7_t * pSrc,
170 uint32_t blockSize,
171 q7_t * pResult)
172 {
173 q7_t maxVal, out; /* Temporary variables to store the output value. */
174 uint32_t blkCnt; /* Loop counter */
175
176
177
178 /* Load first input value that act as reference value for comparision */
179 out = (*pSrc > 0) ? *pSrc : ((*pSrc == (q7_t) 0x80) ? (q7_t) 0x7f : -*pSrc);
180 pSrc++;
181
182 /* Initialize blkCnt with number of samples */
183 blkCnt = (blockSize - 1U);
184
185 while (blkCnt > 0U)
186 {
187 /* Initialize maxVal to the next consecutive values one by one */
188 maxVal = (*pSrc > 0) ? *pSrc : ((*pSrc == (q7_t) 0x80) ? (q7_t) 0x7f : -*pSrc);
189 pSrc++;
190
191 /* compare for the maximum value */
192 if (out < maxVal)
193 {
194 /* Update the maximum value and it's index */
195 out = maxVal;
196 }
197
198 /* Decrement loop counter */
199 blkCnt--;
200 }
201
202 /* Store the maximum value and it's index into destination pointers */
203 *pResult = out;
204 }
205 #endif /* defined(ARM_MATH_DSP) */
206 #endif /* defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) */
207 /**
208 @} end of AbsMax group
209 */
210