mbed official / mbed-dsp

CMSIS DSP library

Dependents:   performance_timer Surfboard_ gps2rtty Capstone ... more

Legacy Warning

This is an mbed 2 library. To learn more about mbed OS 5, visit the docs.

cmsis_dsp/FilteringFunctions/arm_correlate_q15.c@5:3762170b6d4d, 2015-11-20 (annotated)

Committer:
mbed_official
Date:
Fri Nov 20 08:45:18 2015 +0000
Revision:
5:3762170b6d4d
Parent:
3:7a284390b0ce 
Synchronized with git revision 2eb940b9a73af188d3004a2575fdfbb05febe62b



Full URL: https://github.com/mbedmicro/mbed/commit/2eb940b9a73af188d3004a2575fdfbb05febe62b/



Added option to build rpc library. closes #1426

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emilmont 1:fdd22bb7aa52 1 /* ----------------------------------------------------------------------
mbed_official 5:3762170b6d4d 2 * Copyright (C) 2010-2014 ARM Limited. All rights reserved.
emilmont 1:fdd22bb7aa52 3 *
mbed_official 5:3762170b6d4d 4 * $Date: 19. March 2015
mbed_official 5:3762170b6d4d 5 * $Revision: V.1.4.5
emilmont 1:fdd22bb7aa52 6 *
emilmont 2:da51fb522205 7 * Project: CMSIS DSP Library
emilmont 2:da51fb522205 8 * Title: arm_correlate_q15.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Correlation of Q15 sequences.
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
emilmont 1:fdd22bb7aa52 13 *
mbed_official 3:7a284390b0ce 14 * Redistribution and use in source and binary forms, with or without
mbed_official 3:7a284390b0ce 15 * modification, are permitted provided that the following conditions
mbed_official 3:7a284390b0ce 16 * are met:
mbed_official 3:7a284390b0ce 17 * - Redistributions of source code must retain the above copyright
mbed_official 3:7a284390b0ce 18 * notice, this list of conditions and the following disclaimer.
mbed_official 3:7a284390b0ce 19 * - Redistributions in binary form must reproduce the above copyright
mbed_official 3:7a284390b0ce 20 * notice, this list of conditions and the following disclaimer in
mbed_official 3:7a284390b0ce 21 * the documentation and/or other materials provided with the
mbed_official 3:7a284390b0ce 22 * distribution.
mbed_official 3:7a284390b0ce 23 * - Neither the name of ARM LIMITED nor the names of its contributors
mbed_official 3:7a284390b0ce 24 * may be used to endorse or promote products derived from this
mbed_official 3:7a284390b0ce 25 * software without specific prior written permission.
emilmont 1:fdd22bb7aa52 26 *
mbed_official 3:7a284390b0ce 27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
mbed_official 3:7a284390b0ce 28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
mbed_official 3:7a284390b0ce 29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
mbed_official 3:7a284390b0ce 30 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
mbed_official 3:7a284390b0ce 31 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
mbed_official 3:7a284390b0ce 32 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
mbed_official 3:7a284390b0ce 33 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
mbed_official 3:7a284390b0ce 34 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
mbed_official 3:7a284390b0ce 35 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
mbed_official 3:7a284390b0ce 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
mbed_official 3:7a284390b0ce 37 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
mbed_official 3:7a284390b0ce 38 * POSSIBILITY OF SUCH DAMAGE.
emilmont 1:fdd22bb7aa52 39 * -------------------------------------------------------------------- */
emilmont 1:fdd22bb7aa52 40
emilmont 1:fdd22bb7aa52 41 #include "arm_math.h"
emilmont 1:fdd22bb7aa52 42
emilmont 1:fdd22bb7aa52 43 /**
emilmont 1:fdd22bb7aa52 44 * @ingroup groupFilters
emilmont 1:fdd22bb7aa52 45 */
emilmont 1:fdd22bb7aa52 46
emilmont 1:fdd22bb7aa52 47 /**
emilmont 1:fdd22bb7aa52 48 * @addtogroup Corr
emilmont 1:fdd22bb7aa52 49 * @{
emilmont 1:fdd22bb7aa52 50 */
emilmont 1:fdd22bb7aa52 51
emilmont 1:fdd22bb7aa52 52 /**
emilmont 1:fdd22bb7aa52 53 * @brief Correlation of Q15 sequences.
emilmont 1:fdd22bb7aa52 54 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 58 * @param[out] *pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1.
emilmont 1:fdd22bb7aa52 59 * @return none.
emilmont 1:fdd22bb7aa52 60 *
emilmont 1:fdd22bb7aa52 61 * @details
emilmont 1:fdd22bb7aa52 62 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 63 *
emilmont 1:fdd22bb7aa52 64 * \par
emilmont 1:fdd22bb7aa52 65 * The function is implemented using a 64-bit internal accumulator.
emilmont 1:fdd22bb7aa52 66 * Both inputs are in 1.15 format and multiplications yield a 2.30 result.
emilmont 1:fdd22bb7aa52 67 * The 2.30 intermediate results are accumulated in a 64-bit accumulator in 34.30 format.
emilmont 1:fdd22bb7aa52 68 * This approach provides 33 guard bits and there is no risk of overflow.
emilmont 1:fdd22bb7aa52 69 * The 34.30 result is then truncated to 34.15 format by discarding the low 15 bits and then saturated to 1.15 format.
emilmont 1:fdd22bb7aa52 70 *
emilmont 1:fdd22bb7aa52 71 * \par
emilmont 1:fdd22bb7aa52 72 * Refer to <code>arm_correlate_fast_q15()</code> for a faster but less precise version of this function for Cortex-M3 and Cortex-M4.
emilmont 1:fdd22bb7aa52 73 *
emilmont 1:fdd22bb7aa52 74 * \par
emilmont 1:fdd22bb7aa52 75 * Refer the function <code>arm_correlate_opt_q15()</code> for a faster implementation of this function using scratch buffers.
emilmont 1:fdd22bb7aa52 76 *
emilmont 1:fdd22bb7aa52 77 */
emilmont 1:fdd22bb7aa52 78
emilmont 1:fdd22bb7aa52 79 void arm_correlate_q15(
emilmont 1:fdd22bb7aa52 80 q15_t * pSrcA,
emilmont 1:fdd22bb7aa52 81 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 82 q15_t * pSrcB,
emilmont 1:fdd22bb7aa52 83 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 84 q15_t * pDst)
emilmont 1:fdd22bb7aa52 85 {
emilmont 1:fdd22bb7aa52 86
emilmont 1:fdd22bb7aa52 87 #if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE)
emilmont 1:fdd22bb7aa52 88
emilmont 1:fdd22bb7aa52 89 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 90
emilmont 1:fdd22bb7aa52 91 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 92 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 93 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 94 q63_t sum, acc0, acc1, acc2, acc3; /* Accumulators */
emilmont 1:fdd22bb7aa52 95 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 96 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 97 q15_t *pSrc1; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 98 q31_t x0, x1, x2, x3, c0; /* temporary variables for holding input and coefficient values */
emilmont 1:fdd22bb7aa52 99 uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 100 int32_t inc = 1; /* Destination address modifier */
emilmont 1:fdd22bb7aa52 101
emilmont 1:fdd22bb7aa52 102
emilmont 1:fdd22bb7aa52 103 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 104 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 105 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 106 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 107 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 108 /* and the destination pointer modifier, inc is set to -1 */
emilmont 1:fdd22bb7aa52 109 /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */
emilmont 1:fdd22bb7aa52 110 /* But to improve the performance,
emilmont 1:fdd22bb7aa52 111 * we include zeroes in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 112 /* If srcALen > srcBLen,
emilmont 1:fdd22bb7aa52 113 * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */
emilmont 1:fdd22bb7aa52 114 /* If srcALen < srcBLen,
emilmont 1:fdd22bb7aa52 115 * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */
emilmont 1:fdd22bb7aa52 116 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 117 {
emilmont 1:fdd22bb7aa52 118 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 119 pIn1 = (pSrcA);
emilmont 1:fdd22bb7aa52 120
emilmont 1:fdd22bb7aa52 121 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 122 pIn2 = (pSrcB);
emilmont 1:fdd22bb7aa52 123
emilmont 1:fdd22bb7aa52 124 /* Number of output samples is calculated */
emilmont 1:fdd22bb7aa52 125 outBlockSize = (2u * srcALen) - 1u;
emilmont 1:fdd22bb7aa52 126
emilmont 1:fdd22bb7aa52 127 /* When srcALen > srcBLen, zero padding is done to srcB
emilmont 1:fdd22bb7aa52 128 * to make their lengths equal.
emilmont 1:fdd22bb7aa52 129 * Instead, (outBlockSize - (srcALen + srcBLen - 1))
emilmont 1:fdd22bb7aa52 130 * number of output samples are made zero */
emilmont 1:fdd22bb7aa52 131 j = outBlockSize - (srcALen + (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 132
emilmont 1:fdd22bb7aa52 133 /* Updating the pointer position to non zero value */
emilmont 1:fdd22bb7aa52 134 pOut += j;
emilmont 1:fdd22bb7aa52 135
emilmont 1:fdd22bb7aa52 136 }
emilmont 1:fdd22bb7aa52 137 else
emilmont 1:fdd22bb7aa52 138 {
emilmont 1:fdd22bb7aa52 139 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 140 pIn1 = (pSrcB);
emilmont 1:fdd22bb7aa52 141
emilmont 1:fdd22bb7aa52 142 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 143 pIn2 = (pSrcA);
emilmont 1:fdd22bb7aa52 144
emilmont 1:fdd22bb7aa52 145 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 146 j = srcBLen;
emilmont 1:fdd22bb7aa52 147 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 148 srcALen = j;
emilmont 1:fdd22bb7aa52 149
emilmont 1:fdd22bb7aa52 150 /* CORR(x, y) = Reverse order(CORR(y, x)) */
emilmont 1:fdd22bb7aa52 151 /* Hence set the destination pointer to point to the last output sample */
emilmont 1:fdd22bb7aa52 152 pOut = pDst + ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 153
emilmont 1:fdd22bb7aa52 154 /* Destination address modifier is set to -1 */
emilmont 1:fdd22bb7aa52 155 inc = -1;
emilmont 1:fdd22bb7aa52 156
emilmont 1:fdd22bb7aa52 157 }
emilmont 1:fdd22bb7aa52 158
emilmont 1:fdd22bb7aa52 159 /* The function is internally
emilmont 1:fdd22bb7aa52 160 * divided into three parts according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 161 * taken place between inputA samples and inputB samples. In the first part of the
emilmont 1:fdd22bb7aa52 162 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 163 * In the second part of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 164 * In the third part of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 165 * for every iteration.*/
emilmont 1:fdd22bb7aa52 166 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 167 * The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 168 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 169 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 170 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 171
emilmont 1:fdd22bb7aa52 172 /* --------------------------
emilmont 1:fdd22bb7aa52 173 * Initializations of stage1
emilmont 1:fdd22bb7aa52 174 * -------------------------*/
emilmont 1:fdd22bb7aa52 175
emilmont 1:fdd22bb7aa52 176 /* sum = x[0] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 177 * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 178 * ....
emilmont 1:fdd22bb7aa52 179 * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1]
emilmont 1:fdd22bb7aa52 180 */
emilmont 1:fdd22bb7aa52 181
emilmont 1:fdd22bb7aa52 182 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 183 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 184 count = 1u;
emilmont 1:fdd22bb7aa52 185
emilmont 1:fdd22bb7aa52 186 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 187 px = pIn1;
emilmont 1:fdd22bb7aa52 188
emilmont 1:fdd22bb7aa52 189 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 190 pSrc1 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 191 py = pSrc1;
emilmont 1:fdd22bb7aa52 192
emilmont 1:fdd22bb7aa52 193 /* ------------------------
emilmont 1:fdd22bb7aa52 194 * Stage1 process
emilmont 1:fdd22bb7aa52 195 * ----------------------*/
emilmont 1:fdd22bb7aa52 196
emilmont 1:fdd22bb7aa52 197 /* The first loop starts here */
emilmont 1:fdd22bb7aa52 198 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 199 {
emilmont 1:fdd22bb7aa52 200 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 201 sum = 0;
emilmont 1:fdd22bb7aa52 202
emilmont 1:fdd22bb7aa52 203 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 204 k = count >> 2;
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 207 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 208 while(k > 0u)
emilmont 1:fdd22bb7aa52 209 {
emilmont 1:fdd22bb7aa52 210 /* x[0] * y[srcBLen - 4] , x[1] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 211 sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 212 /* x[3] * y[srcBLen - 1] , x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 213 sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 214
emilmont 1:fdd22bb7aa52 215 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 216 k--;
emilmont 1:fdd22bb7aa52 217 }
emilmont 1:fdd22bb7aa52 218
emilmont 1:fdd22bb7aa52 219 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 220 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 221 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 222
emilmont 1:fdd22bb7aa52 223 while(k > 0u)
emilmont 1:fdd22bb7aa52 224 {
emilmont 1:fdd22bb7aa52 225 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 226 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 227 sum = __SMLALD(*px++, *py++, sum);
emilmont 1:fdd22bb7aa52 228
emilmont 1:fdd22bb7aa52 229 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 230 k--;
emilmont 1:fdd22bb7aa52 231 }
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 234 *pOut = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 235 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 236 pOut += inc;
emilmont 1:fdd22bb7aa52 237
emilmont 1:fdd22bb7aa52 238 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 239 py = pSrc1 - count;
emilmont 1:fdd22bb7aa52 240 px = pIn1;
emilmont 1:fdd22bb7aa52 241
emilmont 1:fdd22bb7aa52 242 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 243 count++;
emilmont 1:fdd22bb7aa52 244
emilmont 1:fdd22bb7aa52 245 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 246 blockSize1--;
emilmont 1:fdd22bb7aa52 247 }
emilmont 1:fdd22bb7aa52 248
emilmont 1:fdd22bb7aa52 249 /* --------------------------
emilmont 1:fdd22bb7aa52 250 * Initializations of stage2
emilmont 1:fdd22bb7aa52 251 * ------------------------*/
emilmont 1:fdd22bb7aa52 252
emilmont 1:fdd22bb7aa52 253 /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 254 * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 255 * ....
emilmont 1:fdd22bb7aa52 256 * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 257 */
emilmont 1:fdd22bb7aa52 258
emilmont 1:fdd22bb7aa52 259 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 260 px = pIn1;
emilmont 1:fdd22bb7aa52 261
emilmont 1:fdd22bb7aa52 262 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 263 py = pIn2;
emilmont 1:fdd22bb7aa52 264
emilmont 1:fdd22bb7aa52 265 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 266 count = 0u;
emilmont 1:fdd22bb7aa52 267
emilmont 1:fdd22bb7aa52 268 /* -------------------
emilmont 1:fdd22bb7aa52 269 * Stage2 process
emilmont 1:fdd22bb7aa52 270 * ------------------*/
emilmont 1:fdd22bb7aa52 271
emilmont 1:fdd22bb7aa52 272 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 273 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 274 * srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */
emilmont 1:fdd22bb7aa52 275 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 276 {
emilmont 1:fdd22bb7aa52 277 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 278 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 279
emilmont 1:fdd22bb7aa52 280 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 281 {
emilmont 1:fdd22bb7aa52 282 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 283 acc0 = 0;
emilmont 1:fdd22bb7aa52 284 acc1 = 0;
emilmont 1:fdd22bb7aa52 285 acc2 = 0;
emilmont 1:fdd22bb7aa52 286 acc3 = 0;
emilmont 1:fdd22bb7aa52 287
emilmont 1:fdd22bb7aa52 288 /* read x[0], x[1] samples */
emilmont 1:fdd22bb7aa52 289 x0 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 290 /* read x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 291 x1 = _SIMD32_OFFSET(px + 1);
emilmont 2:da51fb522205 292 px += 2u;
emilmont 1:fdd22bb7aa52 293
emilmont 1:fdd22bb7aa52 294 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 295 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 296
emilmont 1:fdd22bb7aa52 297 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 298 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 299 do
emilmont 1:fdd22bb7aa52 300 {
emilmont 1:fdd22bb7aa52 301 /* Read the first two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 302 * y[0] and y[1] */
emilmont 1:fdd22bb7aa52 303 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 304
emilmont 1:fdd22bb7aa52 305 /* acc0 += x[0] * y[0] + x[1] * y[1] */
emilmont 1:fdd22bb7aa52 306 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 307
emilmont 1:fdd22bb7aa52 308 /* acc1 += x[1] * y[0] + x[2] * y[1] */
emilmont 1:fdd22bb7aa52 309 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 310
emilmont 1:fdd22bb7aa52 311 /* Read x[2], x[3] */
emilmont 1:fdd22bb7aa52 312 x2 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 313
emilmont 1:fdd22bb7aa52 314 /* Read x[3], x[4] */
emilmont 1:fdd22bb7aa52 315 x3 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 316
emilmont 1:fdd22bb7aa52 317 /* acc2 += x[2] * y[0] + x[3] * y[1] */
emilmont 1:fdd22bb7aa52 318 acc2 = __SMLALD(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 319
emilmont 1:fdd22bb7aa52 320 /* acc3 += x[3] * y[0] + x[4] * y[1] */
emilmont 1:fdd22bb7aa52 321 acc3 = __SMLALD(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 322
emilmont 1:fdd22bb7aa52 323 /* Read y[2] and y[3] */
emilmont 1:fdd22bb7aa52 324 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 325
emilmont 1:fdd22bb7aa52 326 /* acc0 += x[2] * y[2] + x[3] * y[3] */
emilmont 1:fdd22bb7aa52 327 acc0 = __SMLALD(x2, c0, acc0);
emilmont 1:fdd22bb7aa52 328
emilmont 1:fdd22bb7aa52 329 /* acc1 += x[3] * y[2] + x[4] * y[3] */
emilmont 1:fdd22bb7aa52 330 acc1 = __SMLALD(x3, c0, acc1);
emilmont 1:fdd22bb7aa52 331
emilmont 1:fdd22bb7aa52 332 /* Read x[4], x[5] */
emilmont 1:fdd22bb7aa52 333 x0 = _SIMD32_OFFSET(px + 2);
emilmont 1:fdd22bb7aa52 334
emilmont 1:fdd22bb7aa52 335 /* Read x[5], x[6] */
emilmont 1:fdd22bb7aa52 336 x1 = _SIMD32_OFFSET(px + 3);
emilmont 1:fdd22bb7aa52 337
emilmont 2:da51fb522205 338 px += 4u;
emilmont 1:fdd22bb7aa52 339
emilmont 1:fdd22bb7aa52 340 /* acc2 += x[4] * y[2] + x[5] * y[3] */
emilmont 1:fdd22bb7aa52 341 acc2 = __SMLALD(x0, c0, acc2);
emilmont 1:fdd22bb7aa52 342
emilmont 1:fdd22bb7aa52 343 /* acc3 += x[5] * y[2] + x[6] * y[3] */
emilmont 1:fdd22bb7aa52 344 acc3 = __SMLALD(x1, c0, acc3);
emilmont 1:fdd22bb7aa52 345
emilmont 1:fdd22bb7aa52 346 } while(--k);
emilmont 1:fdd22bb7aa52 347
emilmont 1:fdd22bb7aa52 348 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 349 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 350 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 351
emilmont 1:fdd22bb7aa52 352 if(k == 1u)
emilmont 1:fdd22bb7aa52 353 {
emilmont 1:fdd22bb7aa52 354 /* Read y[4] */
emilmont 1:fdd22bb7aa52 355 c0 = *py;
emilmont 1:fdd22bb7aa52 356 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 357
emilmont 1:fdd22bb7aa52 358 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 359
emilmont 1:fdd22bb7aa52 360 #else
emilmont 1:fdd22bb7aa52 361
emilmont 1:fdd22bb7aa52 362 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 363
emilmont 1:fdd22bb7aa52 364 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 365 /* Read x[7] */
emilmont 1:fdd22bb7aa52 366 x3 = *__SIMD32(px);
emilmont 2:da51fb522205 367 px++;
emilmont 1:fdd22bb7aa52 368
emilmont 1:fdd22bb7aa52 369 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 370 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 371 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 372 acc2 = __SMLALDX(x1, c0, acc2);
emilmont 1:fdd22bb7aa52 373 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 374 }
emilmont 1:fdd22bb7aa52 375
emilmont 1:fdd22bb7aa52 376 if(k == 2u)
emilmont 1:fdd22bb7aa52 377 {
emilmont 1:fdd22bb7aa52 378 /* Read y[4], y[5] */
emilmont 1:fdd22bb7aa52 379 c0 = *__SIMD32(py);
emilmont 1:fdd22bb7aa52 380
emilmont 1:fdd22bb7aa52 381 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 382 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 383
emilmont 1:fdd22bb7aa52 384 /* Read x[9] */
emilmont 1:fdd22bb7aa52 385 x2 = _SIMD32_OFFSET(px + 1);
emilmont 2:da51fb522205 386 px += 2u;
emilmont 1:fdd22bb7aa52 387
emilmont 1:fdd22bb7aa52 388 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 389 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 390 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 391 acc2 = __SMLALD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 392 acc3 = __SMLALD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 393 }
emilmont 1:fdd22bb7aa52 394
emilmont 1:fdd22bb7aa52 395 if(k == 3u)
emilmont 1:fdd22bb7aa52 396 {
emilmont 1:fdd22bb7aa52 397 /* Read y[4], y[5] */
emilmont 1:fdd22bb7aa52 398 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 399
emilmont 1:fdd22bb7aa52 400 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 401 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 402
emilmont 1:fdd22bb7aa52 403 /* Read x[9] */
emilmont 1:fdd22bb7aa52 404 x2 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 405
emilmont 1:fdd22bb7aa52 406 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 407 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 408 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 409 acc2 = __SMLALD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 410 acc3 = __SMLALD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 411
emilmont 1:fdd22bb7aa52 412 c0 = (*py);
emilmont 1:fdd22bb7aa52 413
emilmont 1:fdd22bb7aa52 414 /* Read y[6] */
emilmont 1:fdd22bb7aa52 415 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 416
emilmont 1:fdd22bb7aa52 417 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 418 #else
emilmont 1:fdd22bb7aa52 419
emilmont 1:fdd22bb7aa52 420 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 421 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 422 /* Read x[10] */
emilmont 1:fdd22bb7aa52 423 x3 = _SIMD32_OFFSET(px + 2);
emilmont 2:da51fb522205 424 px += 3u;
emilmont 1:fdd22bb7aa52 425
emilmont 1:fdd22bb7aa52 426 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 427 acc0 = __SMLALDX(x1, c0, acc0);
emilmont 1:fdd22bb7aa52 428 acc1 = __SMLALD(x2, c0, acc1);
emilmont 1:fdd22bb7aa52 429 acc2 = __SMLALDX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 430 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 431 }
emilmont 1:fdd22bb7aa52 432
emilmont 1:fdd22bb7aa52 433 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 434 *pOut = (q15_t) (__SSAT(acc0 >> 15, 16));
emilmont 1:fdd22bb7aa52 435 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 436 pOut += inc;
emilmont 1:fdd22bb7aa52 437
emilmont 1:fdd22bb7aa52 438 *pOut = (q15_t) (__SSAT(acc1 >> 15, 16));
emilmont 1:fdd22bb7aa52 439 pOut += inc;
emilmont 1:fdd22bb7aa52 440
emilmont 1:fdd22bb7aa52 441 *pOut = (q15_t) (__SSAT(acc2 >> 15, 16));
emilmont 1:fdd22bb7aa52 442 pOut += inc;
emilmont 1:fdd22bb7aa52 443
emilmont 1:fdd22bb7aa52 444 *pOut = (q15_t) (__SSAT(acc3 >> 15, 16));
emilmont 1:fdd22bb7aa52 445 pOut += inc;
emilmont 1:fdd22bb7aa52 446
emilmont 1:fdd22bb7aa52 447 /* Increment the count by 4 as 4 output values are computed */
emilmont 1:fdd22bb7aa52 448 count += 4u;
emilmont 1:fdd22bb7aa52 449
emilmont 1:fdd22bb7aa52 450 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 451 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 452 py = pIn2;
emilmont 1:fdd22bb7aa52 453
emilmont 1:fdd22bb7aa52 454 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 455 blkCnt--;
emilmont 1:fdd22bb7aa52 456 }
emilmont 1:fdd22bb7aa52 457
emilmont 1:fdd22bb7aa52 458 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 459 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 460 blkCnt = blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 461
emilmont 1:fdd22bb7aa52 462 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 463 {
emilmont 1:fdd22bb7aa52 464 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 465 sum = 0;
emilmont 1:fdd22bb7aa52 466
emilmont 1:fdd22bb7aa52 467 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 468 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 469
emilmont 1:fdd22bb7aa52 470 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 471 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 472 while(k > 0u)
emilmont 1:fdd22bb7aa52 473 {
emilmont 1:fdd22bb7aa52 474 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 475 sum += ((q63_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 476 sum += ((q63_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 477 sum += ((q63_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 478 sum += ((q63_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 479
emilmont 1:fdd22bb7aa52 480 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 481 k--;
emilmont 1:fdd22bb7aa52 482 }
emilmont 1:fdd22bb7aa52 483
emilmont 1:fdd22bb7aa52 484 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 485 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 486 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 487
emilmont 1:fdd22bb7aa52 488 while(k > 0u)
emilmont 1:fdd22bb7aa52 489 {
emilmont 1:fdd22bb7aa52 490 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 491 sum += ((q63_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 492
emilmont 1:fdd22bb7aa52 493 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 494 k--;
emilmont 1:fdd22bb7aa52 495 }
emilmont 1:fdd22bb7aa52 496
emilmont 1:fdd22bb7aa52 497 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 498 *pOut = (q15_t) (__SSAT(sum >> 15, 16));
emilmont 1:fdd22bb7aa52 499 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 500 pOut += inc;
emilmont 1:fdd22bb7aa52 501
emilmont 1:fdd22bb7aa52 502 /* Increment count by 1, as one output value is computed */
emilmont 1:fdd22bb7aa52 503 count++;
emilmont 1:fdd22bb7aa52 504
emilmont 1:fdd22bb7aa52 505 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 506 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 507 py = pIn2;
emilmont 1:fdd22bb7aa52 508
emilmont 1:fdd22bb7aa52 509 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 510 blkCnt--;
emilmont 1:fdd22bb7aa52 511 }
emilmont 1:fdd22bb7aa52 512 }
emilmont 1:fdd22bb7aa52 513 else
emilmont 1:fdd22bb7aa52 514 {
emilmont 1:fdd22bb7aa52 515 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 516 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 517 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 518
emilmont 1:fdd22bb7aa52 519 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 520 {
emilmont 1:fdd22bb7aa52 521 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 522 sum = 0;
emilmont 1:fdd22bb7aa52 523
emilmont 1:fdd22bb7aa52 524 /* Loop over srcBLen */
emilmont 1:fdd22bb7aa52 525 k = srcBLen;
emilmont 1:fdd22bb7aa52 526
emilmont 1:fdd22bb7aa52 527 while(k > 0u)
emilmont 1:fdd22bb7aa52 528 {
emilmont 1:fdd22bb7aa52 529 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 530 sum += ((q63_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 533 k--;
emilmont 1:fdd22bb7aa52 534 }
emilmont 1:fdd22bb7aa52 535
emilmont 1:fdd22bb7aa52 536 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 537 *pOut = (q15_t) (__SSAT(sum >> 15, 16));
emilmont 1:fdd22bb7aa52 538 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 539 pOut += inc;
emilmont 1:fdd22bb7aa52 540
emilmont 1:fdd22bb7aa52 541 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 542 count++;
emilmont 1:fdd22bb7aa52 543
emilmont 1:fdd22bb7aa52 544 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 545 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 546 py = pIn2;
emilmont 1:fdd22bb7aa52 547
emilmont 1:fdd22bb7aa52 548 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 549 blkCnt--;
emilmont 1:fdd22bb7aa52 550 }
emilmont 1:fdd22bb7aa52 551 }
emilmont 1:fdd22bb7aa52 552
emilmont 1:fdd22bb7aa52 553 /* --------------------------
emilmont 1:fdd22bb7aa52 554 * Initializations of stage3
emilmont 1:fdd22bb7aa52 555 * -------------------------*/
emilmont 1:fdd22bb7aa52 556
emilmont 1:fdd22bb7aa52 557 /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 558 * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 559 * ....
emilmont 1:fdd22bb7aa52 560 * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 561 * sum += x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 562 */
emilmont 1:fdd22bb7aa52 563
emilmont 1:fdd22bb7aa52 564 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 565 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 566 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 567
emilmont 1:fdd22bb7aa52 568 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 569 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 570 px = pSrc1;
emilmont 1:fdd22bb7aa52 571
emilmont 1:fdd22bb7aa52 572 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 573 py = pIn2;
emilmont 1:fdd22bb7aa52 574
emilmont 1:fdd22bb7aa52 575 /* -------------------
emilmont 1:fdd22bb7aa52 576 * Stage3 process
emilmont 1:fdd22bb7aa52 577 * ------------------*/
emilmont 1:fdd22bb7aa52 578
emilmont 1:fdd22bb7aa52 579 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 580 {
emilmont 1:fdd22bb7aa52 581 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 582 sum = 0;
emilmont 1:fdd22bb7aa52 583
emilmont 1:fdd22bb7aa52 584 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 585 k = count >> 2u;
emilmont 1:fdd22bb7aa52 586
emilmont 1:fdd22bb7aa52 587 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 588 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 589 while(k > 0u)
emilmont 1:fdd22bb7aa52 590 {
emilmont 1:fdd22bb7aa52 591 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 592 /* sum += x[srcALen - srcBLen + 4] * y[3] , sum += x[srcALen - srcBLen + 3] * y[2] */
emilmont 1:fdd22bb7aa52 593 sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 594 /* sum += x[srcALen - srcBLen + 2] * y[1] , sum += x[srcALen - srcBLen + 1] * y[0] */
emilmont 1:fdd22bb7aa52 595 sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 596
emilmont 1:fdd22bb7aa52 597 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 598 k--;
emilmont 1:fdd22bb7aa52 599 }
emilmont 1:fdd22bb7aa52 600
emilmont 1:fdd22bb7aa52 601 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 602 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 603 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 604
emilmont 1:fdd22bb7aa52 605 while(k > 0u)
emilmont 1:fdd22bb7aa52 606 {
emilmont 1:fdd22bb7aa52 607 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 608 sum = __SMLALD(*px++, *py++, sum);
emilmont 1:fdd22bb7aa52 609
emilmont 1:fdd22bb7aa52 610 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 611 k--;
emilmont 1:fdd22bb7aa52 612 }
emilmont 1:fdd22bb7aa52 613
emilmont 1:fdd22bb7aa52 614 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 615 *pOut = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 616 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 617 pOut += inc;
emilmont 1:fdd22bb7aa52 618
emilmont 1:fdd22bb7aa52 619 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 620 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 621 py = pIn2;
emilmont 1:fdd22bb7aa52 622
emilmont 1:fdd22bb7aa52 623 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 624 count--;
emilmont 1:fdd22bb7aa52 625
emilmont 1:fdd22bb7aa52 626 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 627 blockSize3--;
emilmont 1:fdd22bb7aa52 628 }
emilmont 1:fdd22bb7aa52 629
emilmont 1:fdd22bb7aa52 630 #else
emilmont 1:fdd22bb7aa52 631
emilmont 1:fdd22bb7aa52 632 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 633
emilmont 1:fdd22bb7aa52 634 q15_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 635 q15_t *pIn2 = pSrcB + (srcBLen - 1u); /* inputB pointer */
emilmont 1:fdd22bb7aa52 636 q63_t sum; /* Accumulators */
emilmont 1:fdd22bb7aa52 637 uint32_t i = 0u, j; /* loop counters */
emilmont 1:fdd22bb7aa52 638 uint32_t inv = 0u; /* Reverse order flag */
emilmont 1:fdd22bb7aa52 639 uint32_t tot = 0u; /* Length */
emilmont 1:fdd22bb7aa52 640
emilmont 1:fdd22bb7aa52 641 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 642 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 643 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 644 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 645 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 646 /* and a varaible, inv is set to 1 */
emilmont 1:fdd22bb7aa52 647 /* If lengths are not equal then zero pad has to be done to make the two
emilmont 1:fdd22bb7aa52 648 * inputs of same length. But to improve the performance, we include zeroes
emilmont 1:fdd22bb7aa52 649 * in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 650 /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the
emilmont 1:fdd22bb7aa52 651 * starting of the output buffer */
emilmont 1:fdd22bb7aa52 652 /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the
emilmont 1:fdd22bb7aa52 653 * ending of the output buffer */
emilmont 1:fdd22bb7aa52 654 /* Once the zero padding is done the remaining of the output is calcualted
emilmont 1:fdd22bb7aa52 655 * using convolution but with the shorter signal time shifted. */
emilmont 1:fdd22bb7aa52 656
emilmont 1:fdd22bb7aa52 657 /* Calculate the length of the remaining sequence */
emilmont 1:fdd22bb7aa52 658 tot = ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 659
emilmont 1:fdd22bb7aa52 660 if(srcALen > srcBLen)
emilmont 1:fdd22bb7aa52 661 {
emilmont 1:fdd22bb7aa52 662 /* Calculating the number of zeros to be padded to the output */
emilmont 1:fdd22bb7aa52 663 j = srcALen - srcBLen;
emilmont 1:fdd22bb7aa52 664
emilmont 1:fdd22bb7aa52 665 /* Initialise the pointer after zero padding */
emilmont 1:fdd22bb7aa52 666 pDst += j;
emilmont 1:fdd22bb7aa52 667 }
emilmont 1:fdd22bb7aa52 668
emilmont 1:fdd22bb7aa52 669 else if(srcALen < srcBLen)
emilmont 1:fdd22bb7aa52 670 {
emilmont 1:fdd22bb7aa52 671 /* Initialization to inputB pointer */
emilmont 1:fdd22bb7aa52 672 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 673
emilmont 1:fdd22bb7aa52 674 /* Initialization to the end of inputA pointer */
emilmont 1:fdd22bb7aa52 675 pIn2 = pSrcA + (srcALen - 1u);
emilmont 1:fdd22bb7aa52 676
emilmont 1:fdd22bb7aa52 677 /* Initialisation of the pointer after zero padding */
emilmont 1:fdd22bb7aa52 678 pDst = pDst + tot;
emilmont 1:fdd22bb7aa52 679
emilmont 1:fdd22bb7aa52 680 /* Swapping the lengths */
emilmont 1:fdd22bb7aa52 681 j = srcALen;
emilmont 1:fdd22bb7aa52 682 srcALen = srcBLen;
emilmont 1:fdd22bb7aa52 683 srcBLen = j;
emilmont 1:fdd22bb7aa52 684
emilmont 1:fdd22bb7aa52 685 /* Setting the reverse flag */
emilmont 1:fdd22bb7aa52 686 inv = 1;
emilmont 1:fdd22bb7aa52 687
emilmont 1:fdd22bb7aa52 688 }
emilmont 1:fdd22bb7aa52 689
emilmont 1:fdd22bb7aa52 690 /* Loop to calculate convolution for output length number of times */
emilmont 1:fdd22bb7aa52 691 for (i = 0u; i <= tot; i++)
emilmont 1:fdd22bb7aa52 692 {
emilmont 1:fdd22bb7aa52 693 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 694 sum = 0;
emilmont 1:fdd22bb7aa52 695
emilmont 1:fdd22bb7aa52 696 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 697 for (j = 0u; j <= i; j++)
emilmont 1:fdd22bb7aa52 698 {
emilmont 1:fdd22bb7aa52 699 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 700 if((((i - j) < srcBLen) && (j < srcALen)))
emilmont 1:fdd22bb7aa52 701 {
emilmont 1:fdd22bb7aa52 702 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 703 sum += ((q31_t) pIn1[j] * pIn2[-((int32_t) i - j)]);
emilmont 1:fdd22bb7aa52 704 }
emilmont 1:fdd22bb7aa52 705 }
emilmont 1:fdd22bb7aa52 706 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 707 if(inv == 1)
emilmont 1:fdd22bb7aa52 708 *pDst-- = (q15_t) __SSAT((sum >> 15u), 16u);
emilmont 1:fdd22bb7aa52 709 else
emilmont 1:fdd22bb7aa52 710 *pDst++ = (q15_t) __SSAT((sum >> 15u), 16u);
emilmont 1:fdd22bb7aa52 711 }
emilmont 1:fdd22bb7aa52 712
emilmont 1:fdd22bb7aa52 713 #endif /*#if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE) */
emilmont 1:fdd22bb7aa52 714
emilmont 1:fdd22bb7aa52 715 }
emilmont 1:fdd22bb7aa52 716
emilmont 1:fdd22bb7aa52 717 /**
emilmont 1:fdd22bb7aa52 718 * @} end of Corr group
emilmont 1:fdd22bb7aa52 719 */