1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmax_no_idx_f32.c
4 * Description: Maximum value of absolute values of a floating-point 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 #if (defined(ARM_MATH_NEON) || defined(ARM_MATH_MVEF)) && !defined(ARM_MATH_AUTOVECTORIZE)
31 #include <limits.h>
32 #endif
33
34 /**
35 @ingroup groupStats
36 */
37
38
39 /**
40 @addtogroup AbsMax
41 @{
42 */
43
44 /**
45 @brief Maximum value of absolute values of a floating-point vector.
46 @param[in] pSrc points to the input vector
47 @param[in] blockSize number of samples in input vector
48 @param[out] pResult maximum value returned here
49 */
50 #if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
51
52 #include "arm_helium_utils.h"
53
arm_absmax_no_idx_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)54 ARM_DSP_ATTRIBUTE void arm_absmax_no_idx_f32(
55 const float32_t * pSrc,
56 uint32_t blockSize,
57 float32_t * pResult)
58 {
59 int32_t blkCnt; /* loop counters */
60 f32x4_t vecSrc;
61 float32_t const *pSrcVec;
62 f32x4_t curExtremValVec = vdupq_n_f32(F32_ABSMIN);
63 float32_t maxValue = F32_ABSMIN;
64 mve_pred16_t p0;
65
66
67 pSrcVec = (float32_t const *) pSrc;
68 blkCnt = blockSize >> 2;
69 while (blkCnt > 0)
70 {
71 vecSrc = vldrwq_f32(pSrcVec);
72 pSrcVec += 4;
73 /*
74 * update per-lane max.
75 */
76 curExtremValVec = vmaxnmaq(vecSrc, curExtremValVec);
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 & 3;
87 if (blkCnt > 0)
88 {
89 vecSrc = vldrwq_f32(pSrcVec);
90 pSrcVec += 4;
91 p0 = vctp32q(blkCnt);
92 /*
93 * Get current max per lane and current index per lane
94 * when a max is selected
95 */
96 curExtremValVec = vmaxnmaq_m(curExtremValVec, vecSrc, p0);
97 }
98 /*
99 * Get max value across the vector
100 */
101 maxValue = vmaxnmavq(maxValue, curExtremValVec);
102 *pResult = maxValue;
103 }
104
105
106 #else
107 #if defined(ARM_MATH_LOOPUNROLL)
arm_absmax_no_idx_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)108 ARM_DSP_ATTRIBUTE void arm_absmax_no_idx_f32(
109 const float32_t * pSrc,
110 uint32_t blockSize,
111 float32_t * pResult)
112 {
113 float32_t cur_absmax, out; /* Temporary variables to store the output value. */\
114 uint32_t blkCnt; /* Loop counter */ \
115 \
116 /* Load first input value that act as reference value for comparision */ \
117 out = *pSrc++; \
118 out = (out > 0.0f) ? out : -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_absmax to next consecutive values one by one */ \
126 cur_absmax = *pSrc++; \
127 cur_absmax = (cur_absmax > 0.0f) ? cur_absmax : -cur_absmax; \
128 /* compare for the extrema value */ \
129 if (cur_absmax > out) \
130 { \
131 /* Update the extrema value and it's index */ \
132 out = cur_absmax; \
133 } \
134 \
135 cur_absmax = *pSrc++; \
136 cur_absmax = (cur_absmax > 0.0f) ? cur_absmax : -cur_absmax; \
137 if (cur_absmax > out) \
138 { \
139 out = cur_absmax; \
140 } \
141 \
142 cur_absmax = *pSrc++; \
143 cur_absmax = (cur_absmax > 0.0f) ? cur_absmax : -cur_absmax; \
144 if (cur_absmax > out) \
145 { \
146 out = cur_absmax; \
147 } \
148 \
149 cur_absmax = *pSrc++; \
150 cur_absmax = (cur_absmax > 0.0f) ? cur_absmax : -cur_absmax; \
151 if (cur_absmax > out) \
152 { \
153 out = cur_absmax; \
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_absmax = *pSrc++; \
168 cur_absmax = (cur_absmax > 0.0f) ? cur_absmax : -cur_absmax; \
169 if (cur_absmax > out) \
170 { \
171 out = cur_absmax; \
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_absmax_no_idx_f32(const float32_t * pSrc,uint32_t blockSize,float32_t * pResult)182 ARM_DSP_ATTRIBUTE void arm_absmax_no_idx_f32(
183 const float32_t * pSrc,
184 uint32_t blockSize,
185 float32_t * pResult)
186 {
187 float32_t maxVal, out; /* Temporary variables to store the output value. */
188 uint32_t blkCnt; /* Loop counter */
189
190
191
192
193
194 /* Load first input value that act as reference value for comparision */
195 out = fabsf(*pSrc++);
196
197 /* Initialize blkCnt with number of samples */
198 blkCnt = (blockSize - 1U);
199
200
201 while (blkCnt > 0U)
202 {
203 /* Initialize maxVal to the next consecutive values one by one */
204 maxVal = fabsf(*pSrc++);
205
206 /* compare for the maximum value */
207 if (out < maxVal)
208 {
209 /* Update the maximum value and it's index */
210 out = maxVal;
211 }
212
213 /* Decrement loop counter */
214 blkCnt--;
215 }
216
217 /* Store the maximum value and it's index into destination pointers */
218 *pResult = out;
219 }
220 #endif /* defined(ARM_MATH_LOOPUNROLL) */
221 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
222 /**
223 @} end of AbsMax group
224 */
225
