1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_max_q15.c
4 * Description: Maximum value of a Q15 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 Max
37 @{
38 */
39
40 /**
41 @brief Maximum value of a Q15 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 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
48
49 #include "arm_helium_utils.h"
50
arm_max_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult,uint32_t * pIndex)51 void arm_max_q15(
52 const q15_t * pSrc,
53 uint32_t blockSize,
54 q15_t * pResult,
55 uint32_t * pIndex)
56 {
57 int32_t blkCnt; /* loop counters */
58 q15x8_t extremValVec = vdupq_n_s16(Q15_MIN);
59 q15_t maxValue = Q15_MIN;
60 uint16x8_t indexVec;
61 uint16x8_t extremIdxVec;
62 mve_pred16_t p0;
63 uint16_t extremIdxArr[8];
64
65 indexVec = vidupq_u16(0U, 1);
66
67 blkCnt = blockSize;
68 do {
69 mve_pred16_t p = vctp16q(blkCnt);
70 q15x8_t extremIdxVal = vld1q_z_s16(pSrc, p);
71 /*
72 * Get current max per lane and current index per lane
73 * when a max is selected
74 */
75 p0 = vcmpgeq_m(extremIdxVal, extremValVec, p);
76
77 extremValVec = vorrq_m(extremValVec, extremIdxVal, extremIdxVal, p0);
78 /* store per-lane extrema indexes */
79 vst1q_p_u16(extremIdxArr, indexVec, p0);
80
81 indexVec += 8;
82 pSrc += 8;
83 blkCnt -= 8;
84 }
85 while (blkCnt > 0);
86
87
88 /* Get max value across the vector */
89 maxValue = vmaxvq(maxValue, extremValVec);
90
91 /* set index for lower values to max possible index */
92 p0 = vcmpgeq(extremValVec, maxValue);
93 extremIdxVec = vld1q_u16(extremIdxArr);
94
95 indexVec = vpselq(extremIdxVec, vdupq_n_u16(blockSize - 1), p0);
96 *pIndex = vminvq(blockSize - 1, indexVec);
97 *pResult = maxValue;
98 }
99
100 #else
arm_max_q15(const q15_t * pSrc,uint32_t blockSize,q15_t * pResult,uint32_t * pIndex)101 void arm_max_q15(
102 const q15_t * pSrc,
103 uint32_t blockSize,
104 q15_t * pResult,
105 uint32_t * pIndex)
106 {
107 q15_t maxVal, out; /* Temporary variables to store the output value. */
108 uint32_t blkCnt, outIndex; /* Loop counter */
109
110 #if defined (ARM_MATH_LOOPUNROLL)
111 uint32_t index; /* index of maximum value */
112 #endif
113
114 /* Initialise index value to zero. */
115 outIndex = 0U;
116 /* Load first input value that act as reference value for comparision */
117 out = *pSrc++;
118
119 #if defined (ARM_MATH_LOOPUNROLL)
120 /* Initialise index of maximum value. */
121 index = 0U;
122
123 /* Loop unrolling: Compute 4 outputs at a time */
124 blkCnt = (blockSize - 1U) >> 2U;
125
126 while (blkCnt > 0U)
127 {
128 /* Initialize maxVal to next consecutive values one by one */
129 maxVal = *pSrc++;
130
131 /* compare for the maximum value */
132 if (out < maxVal)
133 {
134 /* Update the maximum value and it's index */
135 out = maxVal;
136 outIndex = index + 1U;
137 }
138
139 maxVal = *pSrc++;
140 if (out < maxVal)
141 {
142 out = maxVal;
143 outIndex = index + 2U;
144 }
145
146 maxVal = *pSrc++;
147 if (out < maxVal)
148 {
149 out = maxVal;
150 outIndex = index + 3U;
151 }
152
153 maxVal = *pSrc++;
154 if (out < maxVal)
155 {
156 out = maxVal;
157 outIndex = index + 4U;
158 }
159
160 index += 4U;
161
162 /* Decrement loop counter */
163 blkCnt--;
164 }
165
166 /* Loop unrolling: Compute remaining outputs */
167 blkCnt = (blockSize - 1U) % 4U;
168
169 #else
170
171 /* Initialize blkCnt with number of samples */
172 blkCnt = (blockSize - 1U);
173
174 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
175
176 while (blkCnt > 0U)
177 {
178 /* Initialize maxVal to the next consecutive values one by one */
179 maxVal = *pSrc++;
180
181 /* compare for the maximum value */
182 if (out < maxVal)
183 {
184 /* Update the maximum value and it's index */
185 out = maxVal;
186 outIndex = blockSize - blkCnt;
187 }
188
189 /* Decrement loop counter */
190 blkCnt--;
191 }
192
193 /* Store the maximum value and it's index into destination pointers */
194 *pResult = out;
195 *pIndex = outIndex;
196 }
197 #endif /* defined(ARM_MATH_MVEI) */
198 /**
199 @} end of Max group
200 */
201
