1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_absmin_q7.c
4 * Description: Minimum value of absolute values of a Q7 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 /**
37 @addtogroup AbsMin
38 @{
39 */
40
41 /**
42 @brief Minimum value of absolute values of a Q7 vector.
43 @param[in] pSrc points to the input vector
44 @param[in] blockSize number of samples in input vector
45 @param[out] pResult minimum value returned here
46 @param[out] pIndex index of minimum value returned here
47 */
48 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
49
50 #include <stdint.h>
51 #include "arm_helium_utils.h"
52
53 #define MAX_BLKSZ_S8 (UINT8_MAX+1)
54
arm_small_blk_absmin_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult,uint32_t * pIndex)55 static void arm_small_blk_absmin_q7(
56 const q7_t *pSrc,
57 uint32_t blockSize,
58 q7_t *pResult,
59 uint32_t *pIndex)
60 {
61 uint16_t blkCnt; /* loop counters */
62 q7x16_t vecSrc;
63 q7_t const *pSrcVec;
64 q7x16_t curExtremValVec = vdupq_n_s8(Q7_ABSMAX);
65 q7_t minValue = Q7_ABSMAX;
66 uint16_t idx = blockSize - 1;
67 uint8x16_t indexVec;
68 uint8x16_t curExtremIdxVec;
69 uint32_t startIdx = 0;
70 mve_pred16_t p0;
71
72
73 indexVec = vidupq_wb_u8(&startIdx, 1);
74 curExtremIdxVec = vdupq_n_u8(0);
75
76 pSrcVec = (q7_t const *) pSrc;
77 blkCnt = blockSize >> 4;
78 while (blkCnt > 0U)
79 {
80 vecSrc = vld1q(pSrcVec);
81 pSrcVec += 16;
82 vecSrc = vabsq(vecSrc);
83 /*
84 * Get current min per lane and current index per lane
85 * when a min is selected
86 */
87 p0 = vcmpleq(vecSrc, curExtremValVec);
88 curExtremValVec = vpselq(vecSrc, curExtremValVec, p0);
89 curExtremIdxVec = vpselq(indexVec, curExtremIdxVec, p0);
90
91 indexVec = vidupq_wb_u8(&startIdx, 1);
92 /*
93 * Decrement the blockSize loop counter
94 */
95 blkCnt--;
96 }
97 /*
98 * tail
99 * (will be merged thru tail predication)
100 */
101 blkCnt = blockSize & 0xF;
102 if (blkCnt > 0U)
103 {
104 vecSrc = vld1q(pSrcVec);
105 pSrcVec += 16;
106 vecSrc = vabsq(vecSrc);
107
108 p0 = vctp8q(blkCnt);
109 /*
110 * Get current min per lane and current index per lane
111 * when a min is selected
112 */
113 p0 = vcmpleq_m(vecSrc, curExtremValVec, p0);
114 curExtremValVec = vpselq(vecSrc, curExtremValVec, p0);
115 curExtremIdxVec = vpselq(indexVec, curExtremIdxVec, p0);
116 }
117 /*
118 * Get min value across the vector
119 */
120 minValue = vminvq(minValue, curExtremValVec);
121 /*
122 * set index for lower values to min possible index
123 */
124 p0 = vcmpleq(curExtremValVec, minValue);
125 idx = vminvq_p_u8(idx, curExtremIdxVec, p0);
126 /*
127 * Save result
128 */
129 *pIndex = idx;
130 *pResult = minValue;
131 }
132
133
arm_absmin_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult,uint32_t * pIndex)134 ARM_DSP_ATTRIBUTE void arm_absmin_q7(
135 const q7_t * pSrc,
136 uint32_t blockSize,
137 q7_t * pResult,
138 uint32_t * pIndex)
139 {
140 int32_t totalSize = blockSize;
141
142 if (totalSize <= MAX_BLKSZ_S8)
143 {
144 arm_small_blk_absmin_q7(pSrc, blockSize, pResult, pIndex);
145 }
146 else
147 {
148 uint32_t curIdx = 0;
149 q7_t curBlkExtr = Q7_MAX;
150 uint32_t curBlkPos = 0;
151 uint32_t curBlkIdx = 0;
152 /*
153 * process blocks of 255 elts
154 */
155 while (totalSize >= MAX_BLKSZ_S8)
156 {
157 const q7_t *curSrc = pSrc;
158
159 arm_small_blk_absmin_q7(curSrc, MAX_BLKSZ_S8, pResult, pIndex);
160 if (*pResult < curBlkExtr)
161 {
162 /*
163 * update partial extrema
164 */
165 curBlkExtr = *pResult;
166 curBlkPos = *pIndex;
167 curBlkIdx = curIdx;
168 }
169 curIdx++;
170 pSrc += MAX_BLKSZ_S8;
171 totalSize -= MAX_BLKSZ_S8;
172 }
173 /*
174 * remainder
175 */
176 arm_small_blk_absmin_q7(pSrc, totalSize, pResult, pIndex);
177 if (*pResult < curBlkExtr)
178 {
179 curBlkExtr = *pResult;
180 curBlkPos = *pIndex;
181 curBlkIdx = curIdx;
182 }
183 *pIndex = curBlkIdx * MAX_BLKSZ_S8 + curBlkPos;
184 *pResult = curBlkExtr;
185 }
186 }
187
188 #else
189 #if defined(ARM_MATH_DSP)
arm_absmin_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult,uint32_t * pIndex)190 ARM_DSP_ATTRIBUTE void arm_absmin_q7(
191 const q7_t * pSrc,
192 uint32_t blockSize,
193 q7_t * pResult,
194 uint32_t * pIndex)
195 {
196 q7_t cur_absmin, out; /* Temporary variables to store the output value. */\
197 uint32_t blkCnt, outIndex; /* Loop counter */ \
198 uint32_t index; /* index of maximum value */ \
199 \
200 /* Initialize index value to zero. */ \
201 outIndex = 0U; \
202 /* Load first input value that act as reference value for comparision */ \
203 out = *pSrc++; \
204 out = (out > 0) ? out : (q7_t)__QSUB8(0, out); \
205 /* Initialize index of extrema value. */ \
206 index = 0U; \
207 \
208 /* Loop unrolling: Compute 4 outputs at a time */ \
209 blkCnt = (blockSize - 1U) >> 2U; \
210 \
211 while (blkCnt > 0U) \
212 { \
213 /* Initialize cur_absmin to next consecutive values one by one */ \
214 cur_absmin = *pSrc++; \
215 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
216 /* compare for the extrema value */ \
217 if (cur_absmin < out) \
218 { \
219 /* Update the extrema value and it's index */ \
220 out = cur_absmin; \
221 outIndex = index + 1U; \
222 } \
223 \
224 cur_absmin = *pSrc++; \
225 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
226 if (cur_absmin < out) \
227 { \
228 out = cur_absmin; \
229 outIndex = index + 2U; \
230 } \
231 \
232 cur_absmin = *pSrc++; \
233 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
234 if (cur_absmin < out) \
235 { \
236 out = cur_absmin; \
237 outIndex = index + 3U; \
238 } \
239 \
240 cur_absmin = *pSrc++; \
241 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
242 if (cur_absmin < out) \
243 { \
244 out = cur_absmin; \
245 outIndex = index + 4U; \
246 } \
247 \
248 index += 4U; \
249 \
250 /* Decrement loop counter */ \
251 blkCnt--; \
252 } \
253 \
254 /* Loop unrolling: Compute remaining outputs */ \
255 blkCnt = (blockSize - 1U) % 4U; \
256 \
257 \
258 while (blkCnt > 0U) \
259 { \
260 cur_absmin = *pSrc++; \
261 cur_absmin = (cur_absmin > 0) ? cur_absmin : (q7_t)__QSUB8(0, cur_absmin); \
262 if (cur_absmin < out) \
263 { \
264 out = cur_absmin; \
265 outIndex = blockSize - blkCnt; \
266 } \
267 \
268 /* Decrement loop counter */ \
269 blkCnt--; \
270 } \
271 \
272 /* Store the extrema value and it's index into destination pointers */ \
273 *pResult = out; \
274 *pIndex = outIndex;
275 }
276 #else
arm_absmin_q7(const q7_t * pSrc,uint32_t blockSize,q7_t * pResult,uint32_t * pIndex)277 ARM_DSP_ATTRIBUTE void arm_absmin_q7(
278 const q7_t * pSrc,
279 uint32_t blockSize,
280 q7_t * pResult,
281 uint32_t * pIndex)
282 {
283 q7_t minVal, out; /* Temporary variables to store the output value. */
284 uint32_t blkCnt, outIndex; /* Loop counter */
285
286 /* Initialise index value to zero. */
287 outIndex = 0U;
288 /* Load first input value that act as reference value for comparision */
289 out = (*pSrc > 0) ? *pSrc : ((*pSrc == (q7_t) 0x80) ? (q7_t) 0x7f : -*pSrc);
290 pSrc++;
291
292 /* Initialize blkCnt with number of samples */
293 blkCnt = (blockSize - 1U);
294
295 while (blkCnt > 0U)
296 {
297 /* Initialize minVal to the next consecutive values one by one */
298 minVal = (*pSrc > 0) ? *pSrc : ((*pSrc == (q7_t) 0x80) ? (q7_t) 0x7f : -*pSrc);
299 pSrc++;
300
301 /* compare for the minimum value */
302 if (out > minVal)
303 {
304 /* Update the minimum value and it's index */
305 out = minVal;
306 outIndex = blockSize - blkCnt;
307 }
308
309 /* Decrement loop counter */
310 blkCnt--;
311 }
312
313 /* Store the minimum value and it's index into destination pointers */
314 *pResult = out;
315 *pIndex = outIndex;
316 }
317 #endif /* defined(ARM_MATH_DSP) */
318 #endif /* defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) */
319 /**
320 @} end of AbsMin group
321 */
322
