mbed official / mbed-dsp

CMSIS DSP library

Dependents:   KL25Z_FFT_Demo Hat_Board_v5_1 KL25Z_FFT_Demo_tony KL25Z_FFT_Demo_tony ... more

Fork of mbed-dsp by mbed official

cmsis_dsp/FilteringFunctions/arm_correlate_fast_q15.c@3:7a284390b0ce, 2013-11-08 (annotated)

Committer:
mbed_official
Date:
Fri Nov 08 13:45:10 2013 +0000
Revision:
3:7a284390b0ce
Parent:
2:da51fb522205 
Synchronized with git revision e69956aba2f68a2a26ac26b051f8d349deaa1ce8

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emilmont 1:fdd22bb7aa52 1 /* ----------------------------------------------------------------------
mbed_official 3:7a284390b0ce 2 * Copyright (C) 2010-2013 ARM Limited. All rights reserved.
emilmont 1:fdd22bb7aa52 3 *
mbed_official 3:7a284390b0ce 4 * $Date: 17. January 2013
mbed_official 3:7a284390b0ce 5 * $Revision: V1.4.1
emilmont 1:fdd22bb7aa52 6 *
emilmont 2:da51fb522205 7 * Project: CMSIS DSP Library
emilmont 2:da51fb522205 8 * Title: arm_correlate_fast_q15.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Fast Q15 Correlation.
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3
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 (fast version) for Cortex-M3 and Cortex-M4.
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 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 62 *
emilmont 1:fdd22bb7aa52 63 * \par
emilmont 1:fdd22bb7aa52 64 * This fast version uses a 32-bit accumulator with 2.30 format.
emilmont 1:fdd22bb7aa52 65 * The accumulator maintains full precision of the intermediate multiplication results but provides only a single guard bit.
emilmont 1:fdd22bb7aa52 66 * There is no saturation on intermediate additions.
emilmont 1:fdd22bb7aa52 67 * Thus, if the accumulator overflows it wraps around and distorts the result.
emilmont 1:fdd22bb7aa52 68 * The input signals should be scaled down to avoid intermediate overflows.
emilmont 1:fdd22bb7aa52 69 * Scale down one of the inputs by 1/min(srcALen, srcBLen) to avoid overflow since a
emilmont 1:fdd22bb7aa52 70 * maximum of min(srcALen, srcBLen) number of additions is carried internally.
emilmont 1:fdd22bb7aa52 71 * The 2.30 accumulator is right shifted by 15 bits and then saturated to 1.15 format to yield the final result.
emilmont 1:fdd22bb7aa52 72 *
emilmont 1:fdd22bb7aa52 73 * \par
emilmont 1:fdd22bb7aa52 74 * See <code>arm_correlate_q15()</code> for a slower implementation of this function which uses a 64-bit accumulator to avoid wrap around distortion.
emilmont 1:fdd22bb7aa52 75 */
emilmont 1:fdd22bb7aa52 76
emilmont 1:fdd22bb7aa52 77 void arm_correlate_fast_q15(
emilmont 1:fdd22bb7aa52 78 q15_t * pSrcA,
emilmont 1:fdd22bb7aa52 79 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 80 q15_t * pSrcB,
emilmont 1:fdd22bb7aa52 81 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 82 q15_t * pDst)
emilmont 1:fdd22bb7aa52 83 {
emilmont 1:fdd22bb7aa52 84 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 85
emilmont 1:fdd22bb7aa52 86 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 87 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 88 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 89 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulators */
emilmont 1:fdd22bb7aa52 90 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 91 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 92 q15_t *pSrc1; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 93 q31_t x0, x1, x2, x3, c0; /* temporary variables for holding input and coefficient values */
emilmont 1:fdd22bb7aa52 94 uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 95 int32_t inc = 1; /* Destination address modifier */
emilmont 1:fdd22bb7aa52 96
emilmont 1:fdd22bb7aa52 97
emilmont 1:fdd22bb7aa52 98 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 99 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 100 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 101 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 102 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 103 /* and the destination pointer modifier, inc is set to -1 */
emilmont 1:fdd22bb7aa52 104 /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */
emilmont 1:fdd22bb7aa52 105 /* But to improve the performance,
emilmont 1:fdd22bb7aa52 106 * we include zeroes in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 107 /* If srcALen > srcBLen,
emilmont 1:fdd22bb7aa52 108 * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */
emilmont 1:fdd22bb7aa52 109 /* If srcALen < srcBLen,
emilmont 1:fdd22bb7aa52 110 * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */
emilmont 1:fdd22bb7aa52 111 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 112 {
emilmont 1:fdd22bb7aa52 113 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 114 pIn1 = (pSrcA);
emilmont 1:fdd22bb7aa52 115
emilmont 1:fdd22bb7aa52 116 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 117 pIn2 = (pSrcB);
emilmont 1:fdd22bb7aa52 118
emilmont 1:fdd22bb7aa52 119 /* Number of output samples is calculated */
emilmont 1:fdd22bb7aa52 120 outBlockSize = (2u * srcALen) - 1u;
emilmont 1:fdd22bb7aa52 121
emilmont 1:fdd22bb7aa52 122 /* When srcALen > srcBLen, zero padding is done to srcB
emilmont 1:fdd22bb7aa52 123 * to make their lengths equal.
emilmont 1:fdd22bb7aa52 124 * Instead, (outBlockSize - (srcALen + srcBLen - 1))
emilmont 1:fdd22bb7aa52 125 * number of output samples are made zero */
emilmont 1:fdd22bb7aa52 126 j = outBlockSize - (srcALen + (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 127
emilmont 1:fdd22bb7aa52 128 /* Updating the pointer position to non zero value */
emilmont 1:fdd22bb7aa52 129 pOut += j;
emilmont 1:fdd22bb7aa52 130
emilmont 1:fdd22bb7aa52 131 }
emilmont 1:fdd22bb7aa52 132 else
emilmont 1:fdd22bb7aa52 133 {
emilmont 1:fdd22bb7aa52 134 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 135 pIn1 = (pSrcB);
emilmont 1:fdd22bb7aa52 136
emilmont 1:fdd22bb7aa52 137 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 138 pIn2 = (pSrcA);
emilmont 1:fdd22bb7aa52 139
emilmont 1:fdd22bb7aa52 140 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 141 j = srcBLen;
emilmont 1:fdd22bb7aa52 142 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 143 srcALen = j;
emilmont 1:fdd22bb7aa52 144
emilmont 1:fdd22bb7aa52 145 /* CORR(x, y) = Reverse order(CORR(y, x)) */
emilmont 1:fdd22bb7aa52 146 /* Hence set the destination pointer to point to the last output sample */
emilmont 1:fdd22bb7aa52 147 pOut = pDst + ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 148
emilmont 1:fdd22bb7aa52 149 /* Destination address modifier is set to -1 */
emilmont 1:fdd22bb7aa52 150 inc = -1;
emilmont 1:fdd22bb7aa52 151
emilmont 1:fdd22bb7aa52 152 }
emilmont 1:fdd22bb7aa52 153
emilmont 1:fdd22bb7aa52 154 /* The function is internally
emilmont 1:fdd22bb7aa52 155 * divided into three parts according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 156 * taken place between inputA samples and inputB samples. In the first part of the
emilmont 1:fdd22bb7aa52 157 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 158 * In the second part of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 159 * In the third part of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 160 * for every iteration.*/
emilmont 1:fdd22bb7aa52 161 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 162 * The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 163 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 164 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 165 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 166
emilmont 1:fdd22bb7aa52 167 /* --------------------------
emilmont 1:fdd22bb7aa52 168 * Initializations of stage1
emilmont 1:fdd22bb7aa52 169 * -------------------------*/
emilmont 1:fdd22bb7aa52 170
emilmont 1:fdd22bb7aa52 171 /* sum = x[0] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 172 * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 173 * ....
emilmont 1:fdd22bb7aa52 174 * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1]
emilmont 1:fdd22bb7aa52 175 */
emilmont 1:fdd22bb7aa52 176
emilmont 1:fdd22bb7aa52 177 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 178 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 179 count = 1u;
emilmont 1:fdd22bb7aa52 180
emilmont 1:fdd22bb7aa52 181 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 182 px = pIn1;
emilmont 1:fdd22bb7aa52 183
emilmont 1:fdd22bb7aa52 184 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 185 pSrc1 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 186 py = pSrc1;
emilmont 1:fdd22bb7aa52 187
emilmont 1:fdd22bb7aa52 188 /* ------------------------
emilmont 1:fdd22bb7aa52 189 * Stage1 process
emilmont 1:fdd22bb7aa52 190 * ----------------------*/
emilmont 1:fdd22bb7aa52 191
emilmont 1:fdd22bb7aa52 192 /* The first loop starts here */
emilmont 1:fdd22bb7aa52 193 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 194 {
emilmont 1:fdd22bb7aa52 195 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 196 sum = 0;
emilmont 1:fdd22bb7aa52 197
emilmont 1:fdd22bb7aa52 198 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 199 k = count >> 2;
emilmont 1:fdd22bb7aa52 200
emilmont 1:fdd22bb7aa52 201 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 202 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 203 while(k > 0u)
emilmont 1:fdd22bb7aa52 204 {
emilmont 1:fdd22bb7aa52 205 /* x[0] * y[srcBLen - 4] , x[1] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 206 sum = __SMLAD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 207 /* x[3] * y[srcBLen - 1] , x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 208 sum = __SMLAD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 209
emilmont 1:fdd22bb7aa52 210 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 211 k--;
emilmont 1:fdd22bb7aa52 212 }
emilmont 1:fdd22bb7aa52 213
emilmont 1:fdd22bb7aa52 214 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 215 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 216 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 217
emilmont 1:fdd22bb7aa52 218 while(k > 0u)
emilmont 1:fdd22bb7aa52 219 {
emilmont 1:fdd22bb7aa52 220 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 221 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 222 sum = __SMLAD(*px++, *py++, sum);
emilmont 1:fdd22bb7aa52 223
emilmont 1:fdd22bb7aa52 224 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 225 k--;
emilmont 1:fdd22bb7aa52 226 }
emilmont 1:fdd22bb7aa52 227
emilmont 1:fdd22bb7aa52 228 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 229 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 230 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 231 pOut += inc;
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 234 py = pSrc1 - count;
emilmont 1:fdd22bb7aa52 235 px = pIn1;
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 238 count++;
emilmont 1:fdd22bb7aa52 239
emilmont 1:fdd22bb7aa52 240 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 241 blockSize1--;
emilmont 1:fdd22bb7aa52 242 }
emilmont 1:fdd22bb7aa52 243
emilmont 1:fdd22bb7aa52 244 /* --------------------------
emilmont 1:fdd22bb7aa52 245 * Initializations of stage2
emilmont 1:fdd22bb7aa52 246 * ------------------------*/
emilmont 1:fdd22bb7aa52 247
emilmont 1:fdd22bb7aa52 248 /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 249 * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 250 * ....
emilmont 1:fdd22bb7aa52 251 * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 252 */
emilmont 1:fdd22bb7aa52 253
emilmont 1:fdd22bb7aa52 254 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 255 px = pIn1;
emilmont 1:fdd22bb7aa52 256
emilmont 1:fdd22bb7aa52 257 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 258 py = pIn2;
emilmont 1:fdd22bb7aa52 259
emilmont 1:fdd22bb7aa52 260 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 261 count = 0u;
emilmont 1:fdd22bb7aa52 262
emilmont 1:fdd22bb7aa52 263 /* -------------------
emilmont 1:fdd22bb7aa52 264 * Stage2 process
emilmont 1:fdd22bb7aa52 265 * ------------------*/
emilmont 1:fdd22bb7aa52 266
emilmont 1:fdd22bb7aa52 267 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 268 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 269 * srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */
emilmont 1:fdd22bb7aa52 270 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 271 {
emilmont 1:fdd22bb7aa52 272 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 273 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 274
emilmont 1:fdd22bb7aa52 275 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 276 {
emilmont 1:fdd22bb7aa52 277 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 278 acc0 = 0;
emilmont 1:fdd22bb7aa52 279 acc1 = 0;
emilmont 1:fdd22bb7aa52 280 acc2 = 0;
emilmont 1:fdd22bb7aa52 281 acc3 = 0;
emilmont 1:fdd22bb7aa52 282
emilmont 1:fdd22bb7aa52 283 /* read x[0], x[1] samples */
emilmont 1:fdd22bb7aa52 284 x0 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 285 /* read x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 286 x1 = _SIMD32_OFFSET(px + 1);
emilmont 2:da51fb522205 287 px += 2u;
emilmont 1:fdd22bb7aa52 288
emilmont 1:fdd22bb7aa52 289 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 290 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 291
emilmont 1:fdd22bb7aa52 292 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 293 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 294 do
emilmont 1:fdd22bb7aa52 295 {
emilmont 1:fdd22bb7aa52 296 /* Read the first two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 297 * y[0] and y[1] */
emilmont 1:fdd22bb7aa52 298 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 299
emilmont 1:fdd22bb7aa52 300 /* acc0 += x[0] * y[0] + x[1] * y[1] */
emilmont 1:fdd22bb7aa52 301 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 302
emilmont 1:fdd22bb7aa52 303 /* acc1 += x[1] * y[0] + x[2] * y[1] */
emilmont 1:fdd22bb7aa52 304 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 305
emilmont 1:fdd22bb7aa52 306 /* Read x[2], x[3] */
emilmont 1:fdd22bb7aa52 307 x2 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 308
emilmont 1:fdd22bb7aa52 309 /* Read x[3], x[4] */
emilmont 1:fdd22bb7aa52 310 x3 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 /* acc2 += x[2] * y[0] + x[3] * y[1] */
emilmont 1:fdd22bb7aa52 313 acc2 = __SMLAD(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 314
emilmont 1:fdd22bb7aa52 315 /* acc3 += x[3] * y[0] + x[4] * y[1] */
emilmont 1:fdd22bb7aa52 316 acc3 = __SMLAD(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 317
emilmont 1:fdd22bb7aa52 318 /* Read y[2] and y[3] */
emilmont 1:fdd22bb7aa52 319 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 320
emilmont 1:fdd22bb7aa52 321 /* acc0 += x[2] * y[2] + x[3] * y[3] */
emilmont 1:fdd22bb7aa52 322 acc0 = __SMLAD(x2, c0, acc0);
emilmont 1:fdd22bb7aa52 323
emilmont 1:fdd22bb7aa52 324 /* acc1 += x[3] * y[2] + x[4] * y[3] */
emilmont 1:fdd22bb7aa52 325 acc1 = __SMLAD(x3, c0, acc1);
emilmont 1:fdd22bb7aa52 326
emilmont 1:fdd22bb7aa52 327 /* Read x[4], x[5] */
emilmont 1:fdd22bb7aa52 328 x0 = _SIMD32_OFFSET(px + 2);
emilmont 1:fdd22bb7aa52 329
emilmont 1:fdd22bb7aa52 330 /* Read x[5], x[6] */
emilmont 1:fdd22bb7aa52 331 x1 = _SIMD32_OFFSET(px + 3);
emilmont 2:da51fb522205 332 px += 4u;
emilmont 1:fdd22bb7aa52 333
emilmont 1:fdd22bb7aa52 334 /* acc2 += x[4] * y[2] + x[5] * y[3] */
emilmont 1:fdd22bb7aa52 335 acc2 = __SMLAD(x0, c0, acc2);
emilmont 1:fdd22bb7aa52 336
emilmont 1:fdd22bb7aa52 337 /* acc3 += x[5] * y[2] + x[6] * y[3] */
emilmont 1:fdd22bb7aa52 338 acc3 = __SMLAD(x1, c0, acc3);
emilmont 1:fdd22bb7aa52 339
emilmont 1:fdd22bb7aa52 340 } while(--k);
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 /* For the next MAC operations, SIMD is not used
emilmont 1:fdd22bb7aa52 343 * So, the 16 bit pointer if inputB, py is updated */
emilmont 1:fdd22bb7aa52 344
emilmont 1:fdd22bb7aa52 345 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 346 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 347 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 348
emilmont 1:fdd22bb7aa52 349 if(k == 1u)
emilmont 1:fdd22bb7aa52 350 {
emilmont 1:fdd22bb7aa52 351 /* Read y[4] */
emilmont 1:fdd22bb7aa52 352 c0 = *py;
emilmont 1:fdd22bb7aa52 353 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 354
emilmont 1:fdd22bb7aa52 355 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 356
emilmont 1:fdd22bb7aa52 357 #else
emilmont 1:fdd22bb7aa52 358
emilmont 1:fdd22bb7aa52 359 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 360
emilmont 1:fdd22bb7aa52 361 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 362
emilmont 1:fdd22bb7aa52 363 /* Read x[7] */
emilmont 1:fdd22bb7aa52 364 x3 = *__SIMD32(px);
emilmont 2:da51fb522205 365 px++;
emilmont 1:fdd22bb7aa52 366
emilmont 1:fdd22bb7aa52 367 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 368 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 369 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 370 acc2 = __SMLADX(x1, c0, acc2);
emilmont 1:fdd22bb7aa52 371 acc3 = __SMLADX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 372 }
emilmont 1:fdd22bb7aa52 373
emilmont 1:fdd22bb7aa52 374 if(k == 2u)
emilmont 1:fdd22bb7aa52 375 {
emilmont 1:fdd22bb7aa52 376 /* Read y[4], y[5] */
emilmont 1:fdd22bb7aa52 377 c0 = *__SIMD32(py);
emilmont 1:fdd22bb7aa52 378
emilmont 1:fdd22bb7aa52 379 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 380 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 381
emilmont 1:fdd22bb7aa52 382 /* Read x[9] */
emilmont 1:fdd22bb7aa52 383 x2 = _SIMD32_OFFSET(px + 1);
emilmont 2:da51fb522205 384 px += 2u;
emilmont 1:fdd22bb7aa52 385
emilmont 1:fdd22bb7aa52 386 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 387 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 388 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 389 acc2 = __SMLAD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 390 acc3 = __SMLAD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 391 }
emilmont 1:fdd22bb7aa52 392
emilmont 1:fdd22bb7aa52 393 if(k == 3u)
emilmont 1:fdd22bb7aa52 394 {
emilmont 1:fdd22bb7aa52 395 /* Read y[4], y[5] */
emilmont 1:fdd22bb7aa52 396 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 397
emilmont 1:fdd22bb7aa52 398 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 399 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 400
emilmont 1:fdd22bb7aa52 401 /* Read x[9] */
emilmont 1:fdd22bb7aa52 402 x2 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 405 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 406 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 407 acc2 = __SMLAD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 408 acc3 = __SMLAD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 409
emilmont 1:fdd22bb7aa52 410 c0 = (*py);
emilmont 1:fdd22bb7aa52 411 /* Read y[6] */
emilmont 1:fdd22bb7aa52 412 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 413
emilmont 1:fdd22bb7aa52 414 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 415 #else
emilmont 1:fdd22bb7aa52 416
emilmont 1:fdd22bb7aa52 417 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 418 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 419
emilmont 1:fdd22bb7aa52 420 /* Read x[10] */
emilmont 1:fdd22bb7aa52 421 x3 = _SIMD32_OFFSET(px + 2);
emilmont 2:da51fb522205 422 px += 3u;
emilmont 1:fdd22bb7aa52 423
emilmont 1:fdd22bb7aa52 424 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 425 acc0 = __SMLADX(x1, c0, acc0);
emilmont 1:fdd22bb7aa52 426 acc1 = __SMLAD(x2, c0, acc1);
emilmont 1:fdd22bb7aa52 427 acc2 = __SMLADX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 428 acc3 = __SMLADX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 429 }
emilmont 1:fdd22bb7aa52 430
emilmont 1:fdd22bb7aa52 431 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 432 *pOut = (q15_t) (acc0 >> 15);
emilmont 1:fdd22bb7aa52 433 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 434 pOut += inc;
emilmont 1:fdd22bb7aa52 435
emilmont 1:fdd22bb7aa52 436 *pOut = (q15_t) (acc1 >> 15);
emilmont 1:fdd22bb7aa52 437 pOut += inc;
emilmont 1:fdd22bb7aa52 438
emilmont 1:fdd22bb7aa52 439 *pOut = (q15_t) (acc2 >> 15);
emilmont 1:fdd22bb7aa52 440 pOut += inc;
emilmont 1:fdd22bb7aa52 441
emilmont 1:fdd22bb7aa52 442 *pOut = (q15_t) (acc3 >> 15);
emilmont 1:fdd22bb7aa52 443 pOut += inc;
emilmont 1:fdd22bb7aa52 444
emilmont 1:fdd22bb7aa52 445 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 446 count += 4u;
emilmont 1:fdd22bb7aa52 447
emilmont 1:fdd22bb7aa52 448 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 449 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 450 py = pIn2;
emilmont 1:fdd22bb7aa52 451
emilmont 1:fdd22bb7aa52 452
emilmont 1:fdd22bb7aa52 453 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 454 blkCnt--;
emilmont 1:fdd22bb7aa52 455 }
emilmont 1:fdd22bb7aa52 456
emilmont 1:fdd22bb7aa52 457 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 458 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 459 blkCnt = blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 460
emilmont 1:fdd22bb7aa52 461 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 462 {
emilmont 1:fdd22bb7aa52 463 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 464 sum = 0;
emilmont 1:fdd22bb7aa52 465
emilmont 1:fdd22bb7aa52 466 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 467 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 468
emilmont 1:fdd22bb7aa52 469 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 470 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 471 while(k > 0u)
emilmont 1:fdd22bb7aa52 472 {
emilmont 1:fdd22bb7aa52 473 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 474 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 475 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 476 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 477 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 478
emilmont 1:fdd22bb7aa52 479 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 480 k--;
emilmont 1:fdd22bb7aa52 481 }
emilmont 1:fdd22bb7aa52 482
emilmont 1:fdd22bb7aa52 483 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 484 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 485 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 486
emilmont 1:fdd22bb7aa52 487 while(k > 0u)
emilmont 1:fdd22bb7aa52 488 {
emilmont 1:fdd22bb7aa52 489 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 490 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 491
emilmont 1:fdd22bb7aa52 492 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 493 k--;
emilmont 1:fdd22bb7aa52 494 }
emilmont 1:fdd22bb7aa52 495
emilmont 1:fdd22bb7aa52 496 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 497 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 498 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 499 pOut += inc;
emilmont 1:fdd22bb7aa52 500
emilmont 1:fdd22bb7aa52 501 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 502 count++;
emilmont 1:fdd22bb7aa52 503
emilmont 1:fdd22bb7aa52 504 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 505 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 506 py = pIn2;
emilmont 1:fdd22bb7aa52 507
emilmont 1:fdd22bb7aa52 508 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 509 blkCnt--;
emilmont 1:fdd22bb7aa52 510 }
emilmont 1:fdd22bb7aa52 511 }
emilmont 1:fdd22bb7aa52 512 else
emilmont 1:fdd22bb7aa52 513 {
emilmont 1:fdd22bb7aa52 514 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 515 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 516 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 517
emilmont 1:fdd22bb7aa52 518 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 519 {
emilmont 1:fdd22bb7aa52 520 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 521 sum = 0;
emilmont 1:fdd22bb7aa52 522
emilmont 1:fdd22bb7aa52 523 /* Loop over srcBLen */
emilmont 1:fdd22bb7aa52 524 k = srcBLen;
emilmont 1:fdd22bb7aa52 525
emilmont 1:fdd22bb7aa52 526 while(k > 0u)
emilmont 1:fdd22bb7aa52 527 {
emilmont 1:fdd22bb7aa52 528 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 529 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 530
emilmont 1:fdd22bb7aa52 531 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 532 k--;
emilmont 1:fdd22bb7aa52 533 }
emilmont 1:fdd22bb7aa52 534
emilmont 1:fdd22bb7aa52 535 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 536 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 537 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 538 pOut += inc;
emilmont 1:fdd22bb7aa52 539
emilmont 1:fdd22bb7aa52 540 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 541 count++;
emilmont 1:fdd22bb7aa52 542
emilmont 1:fdd22bb7aa52 543 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 544 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 545 py = pIn2;
emilmont 1:fdd22bb7aa52 546
emilmont 1:fdd22bb7aa52 547 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 548 blkCnt--;
emilmont 1:fdd22bb7aa52 549 }
emilmont 1:fdd22bb7aa52 550 }
emilmont 1:fdd22bb7aa52 551
emilmont 1:fdd22bb7aa52 552 /* --------------------------
emilmont 1:fdd22bb7aa52 553 * Initializations of stage3
emilmont 1:fdd22bb7aa52 554 * -------------------------*/
emilmont 1:fdd22bb7aa52 555
emilmont 1:fdd22bb7aa52 556 /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 557 * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 558 * ....
emilmont 1:fdd22bb7aa52 559 * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 560 * sum += x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 561 */
emilmont 1:fdd22bb7aa52 562
emilmont 1:fdd22bb7aa52 563 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 564 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 565 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 566
emilmont 1:fdd22bb7aa52 567 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 568 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 569 px = pSrc1;
emilmont 1:fdd22bb7aa52 570
emilmont 1:fdd22bb7aa52 571 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 572 py = pIn2;
emilmont 1:fdd22bb7aa52 573
emilmont 1:fdd22bb7aa52 574 /* -------------------
emilmont 1:fdd22bb7aa52 575 * Stage3 process
emilmont 1:fdd22bb7aa52 576 * ------------------*/
emilmont 1:fdd22bb7aa52 577
emilmont 1:fdd22bb7aa52 578 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 579 {
emilmont 1:fdd22bb7aa52 580 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 581 sum = 0;
emilmont 1:fdd22bb7aa52 582
emilmont 1:fdd22bb7aa52 583 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 584 k = count >> 2u;
emilmont 1:fdd22bb7aa52 585
emilmont 1:fdd22bb7aa52 586 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 587 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 588 while(k > 0u)
emilmont 1:fdd22bb7aa52 589 {
emilmont 1:fdd22bb7aa52 590 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 591 /* sum += x[srcALen - srcBLen + 4] * y[3] , sum += x[srcALen - srcBLen + 3] * y[2] */
emilmont 1:fdd22bb7aa52 592 sum = __SMLAD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 593 /* sum += x[srcALen - srcBLen + 2] * y[1] , sum += x[srcALen - srcBLen + 1] * y[0] */
emilmont 1:fdd22bb7aa52 594 sum = __SMLAD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 595
emilmont 1:fdd22bb7aa52 596 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 597 k--;
emilmont 1:fdd22bb7aa52 598 }
emilmont 1:fdd22bb7aa52 599
emilmont 1:fdd22bb7aa52 600 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 601 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 602 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 603
emilmont 1:fdd22bb7aa52 604 while(k > 0u)
emilmont 1:fdd22bb7aa52 605 {
emilmont 1:fdd22bb7aa52 606 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 607 sum = __SMLAD(*px++, *py++, sum);
emilmont 1:fdd22bb7aa52 608
emilmont 1:fdd22bb7aa52 609 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 610 k--;
emilmont 1:fdd22bb7aa52 611 }
emilmont 1:fdd22bb7aa52 612
emilmont 1:fdd22bb7aa52 613 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 614 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 615 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 616 pOut += inc;
emilmont 1:fdd22bb7aa52 617
emilmont 1:fdd22bb7aa52 618 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 619 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 620 py = pIn2;
emilmont 1:fdd22bb7aa52 621
emilmont 1:fdd22bb7aa52 622 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 623 count--;
emilmont 1:fdd22bb7aa52 624
emilmont 1:fdd22bb7aa52 625 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 626 blockSize3--;
emilmont 1:fdd22bb7aa52 627 }
emilmont 1:fdd22bb7aa52 628
emilmont 1:fdd22bb7aa52 629 #else
emilmont 1:fdd22bb7aa52 630
emilmont 1:fdd22bb7aa52 631 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 632 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 633 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 634 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulators */
emilmont 1:fdd22bb7aa52 635 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 636 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 637 q15_t *pSrc1; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 638 q31_t x0, x1, x2, x3, c0; /* temporary variables for holding input and coefficient values */
emilmont 1:fdd22bb7aa52 639 uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 640 int32_t inc = 1; /* Destination address modifier */
emilmont 1:fdd22bb7aa52 641 q15_t a, b;
emilmont 1:fdd22bb7aa52 642
emilmont 1:fdd22bb7aa52 643
emilmont 1:fdd22bb7aa52 644 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 645 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 646 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 647 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 648 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 649 /* and the destination pointer modifier, inc is set to -1 */
emilmont 1:fdd22bb7aa52 650 /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */
emilmont 1:fdd22bb7aa52 651 /* But to improve the performance,
emilmont 1:fdd22bb7aa52 652 * we include zeroes in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 653 /* If srcALen > srcBLen,
emilmont 1:fdd22bb7aa52 654 * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */
emilmont 1:fdd22bb7aa52 655 /* If srcALen < srcBLen,
emilmont 1:fdd22bb7aa52 656 * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */
emilmont 1:fdd22bb7aa52 657 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 658 {
emilmont 1:fdd22bb7aa52 659 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 660 pIn1 = (pSrcA);
emilmont 1:fdd22bb7aa52 661
emilmont 1:fdd22bb7aa52 662 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 663 pIn2 = (pSrcB);
emilmont 1:fdd22bb7aa52 664
emilmont 1:fdd22bb7aa52 665 /* Number of output samples is calculated */
emilmont 1:fdd22bb7aa52 666 outBlockSize = (2u * srcALen) - 1u;
emilmont 1:fdd22bb7aa52 667
emilmont 1:fdd22bb7aa52 668 /* When srcALen > srcBLen, zero padding is done to srcB
emilmont 1:fdd22bb7aa52 669 * to make their lengths equal.
emilmont 1:fdd22bb7aa52 670 * Instead, (outBlockSize - (srcALen + srcBLen - 1))
emilmont 1:fdd22bb7aa52 671 * number of output samples are made zero */
emilmont 1:fdd22bb7aa52 672 j = outBlockSize - (srcALen + (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 673
emilmont 1:fdd22bb7aa52 674 /* Updating the pointer position to non zero value */
emilmont 1:fdd22bb7aa52 675 pOut += j;
emilmont 1:fdd22bb7aa52 676
emilmont 1:fdd22bb7aa52 677 }
emilmont 1:fdd22bb7aa52 678 else
emilmont 1:fdd22bb7aa52 679 {
emilmont 1:fdd22bb7aa52 680 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 681 pIn1 = (pSrcB);
emilmont 1:fdd22bb7aa52 682
emilmont 1:fdd22bb7aa52 683 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 684 pIn2 = (pSrcA);
emilmont 1:fdd22bb7aa52 685
emilmont 1:fdd22bb7aa52 686 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 687 j = srcBLen;
emilmont 1:fdd22bb7aa52 688 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 689 srcALen = j;
emilmont 1:fdd22bb7aa52 690
emilmont 1:fdd22bb7aa52 691 /* CORR(x, y) = Reverse order(CORR(y, x)) */
emilmont 1:fdd22bb7aa52 692 /* Hence set the destination pointer to point to the last output sample */
emilmont 1:fdd22bb7aa52 693 pOut = pDst + ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 694
emilmont 1:fdd22bb7aa52 695 /* Destination address modifier is set to -1 */
emilmont 1:fdd22bb7aa52 696 inc = -1;
emilmont 1:fdd22bb7aa52 697
emilmont 1:fdd22bb7aa52 698 }
emilmont 1:fdd22bb7aa52 699
emilmont 1:fdd22bb7aa52 700 /* The function is internally
emilmont 1:fdd22bb7aa52 701 * divided into three parts according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 702 * taken place between inputA samples and inputB samples. In the first part of the
emilmont 1:fdd22bb7aa52 703 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 704 * In the second part of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 705 * In the third part of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 706 * for every iteration.*/
emilmont 1:fdd22bb7aa52 707 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 708 * The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 709 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 710 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 711 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 712
emilmont 1:fdd22bb7aa52 713 /* --------------------------
emilmont 1:fdd22bb7aa52 714 * Initializations of stage1
emilmont 1:fdd22bb7aa52 715 * -------------------------*/
emilmont 1:fdd22bb7aa52 716
emilmont 1:fdd22bb7aa52 717 /* sum = x[0] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 718 * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 719 * ....
emilmont 1:fdd22bb7aa52 720 * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1]
emilmont 1:fdd22bb7aa52 721 */
emilmont 1:fdd22bb7aa52 722
emilmont 1:fdd22bb7aa52 723 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 724 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 725 count = 1u;
emilmont 1:fdd22bb7aa52 726
emilmont 1:fdd22bb7aa52 727 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 728 px = pIn1;
emilmont 1:fdd22bb7aa52 729
emilmont 1:fdd22bb7aa52 730 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 731 pSrc1 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 732 py = pSrc1;
emilmont 1:fdd22bb7aa52 733
emilmont 1:fdd22bb7aa52 734 /* ------------------------
emilmont 1:fdd22bb7aa52 735 * Stage1 process
emilmont 1:fdd22bb7aa52 736 * ----------------------*/
emilmont 1:fdd22bb7aa52 737
emilmont 1:fdd22bb7aa52 738 /* The first loop starts here */
emilmont 1:fdd22bb7aa52 739 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 740 {
emilmont 1:fdd22bb7aa52 741 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 742 sum = 0;
emilmont 1:fdd22bb7aa52 743
emilmont 1:fdd22bb7aa52 744 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 745 k = count >> 2;
emilmont 1:fdd22bb7aa52 746
emilmont 1:fdd22bb7aa52 747 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 748 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 749 while(k > 0u)
emilmont 1:fdd22bb7aa52 750 {
emilmont 1:fdd22bb7aa52 751 /* x[0] * y[srcBLen - 4] , x[1] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 752 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 753 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 754 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 755 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 756
emilmont 1:fdd22bb7aa52 757 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 758 k--;
emilmont 1:fdd22bb7aa52 759 }
emilmont 1:fdd22bb7aa52 760
emilmont 1:fdd22bb7aa52 761 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 762 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 763 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 764
emilmont 1:fdd22bb7aa52 765 while(k > 0u)
emilmont 1:fdd22bb7aa52 766 {
emilmont 1:fdd22bb7aa52 767 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 768 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 769 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 770
emilmont 1:fdd22bb7aa52 771 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 772 k--;
emilmont 1:fdd22bb7aa52 773 }
emilmont 1:fdd22bb7aa52 774
emilmont 1:fdd22bb7aa52 775 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 776 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 777 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 778 pOut += inc;
emilmont 1:fdd22bb7aa52 779
emilmont 1:fdd22bb7aa52 780 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 781 py = pSrc1 - count;
emilmont 1:fdd22bb7aa52 782 px = pIn1;
emilmont 1:fdd22bb7aa52 783
emilmont 1:fdd22bb7aa52 784 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 785 count++;
emilmont 1:fdd22bb7aa52 786
emilmont 1:fdd22bb7aa52 787 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 788 blockSize1--;
emilmont 1:fdd22bb7aa52 789 }
emilmont 1:fdd22bb7aa52 790
emilmont 1:fdd22bb7aa52 791 /* --------------------------
emilmont 1:fdd22bb7aa52 792 * Initializations of stage2
emilmont 1:fdd22bb7aa52 793 * ------------------------*/
emilmont 1:fdd22bb7aa52 794
emilmont 1:fdd22bb7aa52 795 /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 796 * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 797 * ....
emilmont 1:fdd22bb7aa52 798 * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 799 */
emilmont 1:fdd22bb7aa52 800
emilmont 1:fdd22bb7aa52 801 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 802 px = pIn1;
emilmont 1:fdd22bb7aa52 803
emilmont 1:fdd22bb7aa52 804 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 805 py = pIn2;
emilmont 1:fdd22bb7aa52 806
emilmont 1:fdd22bb7aa52 807 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 808 count = 0u;
emilmont 1:fdd22bb7aa52 809
emilmont 1:fdd22bb7aa52 810 /* -------------------
emilmont 1:fdd22bb7aa52 811 * Stage2 process
emilmont 1:fdd22bb7aa52 812 * ------------------*/
emilmont 1:fdd22bb7aa52 813
emilmont 1:fdd22bb7aa52 814 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 815 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 816 * srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */
emilmont 1:fdd22bb7aa52 817 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 818 {
emilmont 1:fdd22bb7aa52 819 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 820 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 821
emilmont 1:fdd22bb7aa52 822 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 823 {
emilmont 1:fdd22bb7aa52 824 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 825 acc0 = 0;
emilmont 1:fdd22bb7aa52 826 acc1 = 0;
emilmont 1:fdd22bb7aa52 827 acc2 = 0;
emilmont 1:fdd22bb7aa52 828 acc3 = 0;
emilmont 1:fdd22bb7aa52 829
emilmont 1:fdd22bb7aa52 830 /* read x[0], x[1], x[2] samples */
emilmont 2:da51fb522205 831 a = *px;
emilmont 2:da51fb522205 832 b = *(px + 1);
emilmont 1:fdd22bb7aa52 833
emilmont 1:fdd22bb7aa52 834 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 835
emilmont 2:da51fb522205 836 x0 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 837 a = *(px + 2);
emilmont 2:da51fb522205 838 x1 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 839
emilmont 1:fdd22bb7aa52 840 #else
emilmont 1:fdd22bb7aa52 841
emilmont 2:da51fb522205 842 x0 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 843 a = *(px + 2);
emilmont 2:da51fb522205 844 x1 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 845
emilmont 2:da51fb522205 846 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 847
emilmont 2:da51fb522205 848 px += 2u;
emilmont 1:fdd22bb7aa52 849
emilmont 1:fdd22bb7aa52 850 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 851 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 852
emilmont 1:fdd22bb7aa52 853 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 854 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 855 do
emilmont 1:fdd22bb7aa52 856 {
emilmont 1:fdd22bb7aa52 857 /* Read the first two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 858 * y[0] and y[1] */
emilmont 2:da51fb522205 859 a = *py;
emilmont 2:da51fb522205 860 b = *(py + 1);
emilmont 2:da51fb522205 861
emilmont 1:fdd22bb7aa52 862 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 2:da51fb522205 863
emilmont 2:da51fb522205 864 c0 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 865
emilmont 1:fdd22bb7aa52 866 #else
emilmont 2:da51fb522205 867
emilmont 2:da51fb522205 868 c0 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 869
emilmont 2:da51fb522205 870 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 871
emilmont 1:fdd22bb7aa52 872 /* acc0 += x[0] * y[0] + x[1] * y[1] */
emilmont 1:fdd22bb7aa52 873 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 874
emilmont 1:fdd22bb7aa52 875 /* acc1 += x[1] * y[0] + x[2] * y[1] */
emilmont 1:fdd22bb7aa52 876 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 877
emilmont 1:fdd22bb7aa52 878 /* Read x[2], x[3], x[4] */
emilmont 2:da51fb522205 879 a = *px;
emilmont 2:da51fb522205 880 b = *(px + 1);
emilmont 1:fdd22bb7aa52 881
emilmont 1:fdd22bb7aa52 882 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 883
emilmont 2:da51fb522205 884 x2 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 885 a = *(px + 2);
emilmont 2:da51fb522205 886 x3 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 887
emilmont 1:fdd22bb7aa52 888 #else
emilmont 1:fdd22bb7aa52 889
emilmont 2:da51fb522205 890 x2 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 891 a = *(px + 2);
emilmont 2:da51fb522205 892 x3 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 893
emilmont 2:da51fb522205 894 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 895
emilmont 1:fdd22bb7aa52 896 /* acc2 += x[2] * y[0] + x[3] * y[1] */
emilmont 1:fdd22bb7aa52 897 acc2 = __SMLAD(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 898
emilmont 1:fdd22bb7aa52 899 /* acc3 += x[3] * y[0] + x[4] * y[1] */
emilmont 1:fdd22bb7aa52 900 acc3 = __SMLAD(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 901
emilmont 1:fdd22bb7aa52 902 /* Read y[2] and y[3] */
emilmont 2:da51fb522205 903 a = *(py + 2);
emilmont 2:da51fb522205 904 b = *(py + 3);
emilmont 1:fdd22bb7aa52 905
emilmont 2:da51fb522205 906 py += 4u;
emilmont 2:da51fb522205 907
emilmont 1:fdd22bb7aa52 908 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 2:da51fb522205 909
emilmont 2:da51fb522205 910 c0 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 911
emilmont 1:fdd22bb7aa52 912 #else
emilmont 2:da51fb522205 913
emilmont 2:da51fb522205 914 c0 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 915
emilmont 2:da51fb522205 916 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 917
emilmont 1:fdd22bb7aa52 918 /* acc0 += x[2] * y[2] + x[3] * y[3] */
emilmont 1:fdd22bb7aa52 919 acc0 = __SMLAD(x2, c0, acc0);
emilmont 1:fdd22bb7aa52 920
emilmont 1:fdd22bb7aa52 921 /* acc1 += x[3] * y[2] + x[4] * y[3] */
emilmont 1:fdd22bb7aa52 922 acc1 = __SMLAD(x3, c0, acc1);
emilmont 1:fdd22bb7aa52 923
emilmont 1:fdd22bb7aa52 924 /* Read x[4], x[5], x[6] */
emilmont 2:da51fb522205 925 a = *(px + 2);
emilmont 2:da51fb522205 926 b = *(px + 3);
emilmont 1:fdd22bb7aa52 927
emilmont 1:fdd22bb7aa52 928 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 929
emilmont 2:da51fb522205 930 x0 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 931 a = *(px + 4);
emilmont 2:da51fb522205 932 x1 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 933
emilmont 1:fdd22bb7aa52 934 #else
emilmont 1:fdd22bb7aa52 935
emilmont 2:da51fb522205 936 x0 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 937 a = *(px + 4);
emilmont 2:da51fb522205 938 x1 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 939
emilmont 2:da51fb522205 940 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 941
emilmont 2:da51fb522205 942 px += 4u;
emilmont 1:fdd22bb7aa52 943
emilmont 1:fdd22bb7aa52 944 /* acc2 += x[4] * y[2] + x[5] * y[3] */
emilmont 1:fdd22bb7aa52 945 acc2 = __SMLAD(x0, c0, acc2);
emilmont 1:fdd22bb7aa52 946
emilmont 1:fdd22bb7aa52 947 /* acc3 += x[5] * y[2] + x[6] * y[3] */
emilmont 1:fdd22bb7aa52 948 acc3 = __SMLAD(x1, c0, acc3);
emilmont 1:fdd22bb7aa52 949
emilmont 1:fdd22bb7aa52 950 } while(--k);
emilmont 1:fdd22bb7aa52 951
emilmont 1:fdd22bb7aa52 952 /* For the next MAC operations, SIMD is not used
emilmont 1:fdd22bb7aa52 953 * So, the 16 bit pointer if inputB, py is updated */
emilmont 1:fdd22bb7aa52 954
emilmont 1:fdd22bb7aa52 955 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 956 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 957 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 958
emilmont 1:fdd22bb7aa52 959 if(k == 1u)
emilmont 1:fdd22bb7aa52 960 {
emilmont 1:fdd22bb7aa52 961 /* Read y[4] */
emilmont 1:fdd22bb7aa52 962 c0 = *py;
emilmont 1:fdd22bb7aa52 963 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 964
emilmont 1:fdd22bb7aa52 965 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 966
emilmont 1:fdd22bb7aa52 967 #else
emilmont 1:fdd22bb7aa52 968
emilmont 1:fdd22bb7aa52 969 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 970
emilmont 1:fdd22bb7aa52 971 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 972
emilmont 1:fdd22bb7aa52 973 /* Read x[7] */
emilmont 2:da51fb522205 974 a = *px;
emilmont 2:da51fb522205 975 b = *(px + 1);
emilmont 1:fdd22bb7aa52 976
emilmont 2:da51fb522205 977 px++;;
emilmont 2:da51fb522205 978
emilmont 1:fdd22bb7aa52 979 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 2:da51fb522205 980
emilmont 2:da51fb522205 981 x3 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 982
emilmont 1:fdd22bb7aa52 983 #else
emilmont 2:da51fb522205 984
emilmont 2:da51fb522205 985 x3 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 986
emilmont 2:da51fb522205 987 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 988
emilmont 2:da51fb522205 989 px++;
emilmont 1:fdd22bb7aa52 990
emilmont 1:fdd22bb7aa52 991 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 992 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 993 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 994 acc2 = __SMLADX(x1, c0, acc2);
emilmont 1:fdd22bb7aa52 995 acc3 = __SMLADX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 996 }
emilmont 1:fdd22bb7aa52 997
emilmont 1:fdd22bb7aa52 998 if(k == 2u)
emilmont 1:fdd22bb7aa52 999 {
emilmont 1:fdd22bb7aa52 1000 /* Read y[4], y[5] */
emilmont 2:da51fb522205 1001 a = *py;
emilmont 2:da51fb522205 1002 b = *(py + 1);
emilmont 2:da51fb522205 1003
emilmont 1:fdd22bb7aa52 1004 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 2:da51fb522205 1005
emilmont 2:da51fb522205 1006 c0 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 1007
emilmont 1:fdd22bb7aa52 1008 #else
emilmont 2:da51fb522205 1009
emilmont 2:da51fb522205 1010 c0 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 1011
emilmont 2:da51fb522205 1012 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 1013
emilmont 1:fdd22bb7aa52 1014 /* Read x[7], x[8], x[9] */
emilmont 2:da51fb522205 1015 a = *px;
emilmont 2:da51fb522205 1016 b = *(px + 1);
emilmont 1:fdd22bb7aa52 1017
emilmont 1:fdd22bb7aa52 1018 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 1019
emilmont 2:da51fb522205 1020 x3 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 1021 a = *(px + 2);
emilmont 2:da51fb522205 1022 x2 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 1023
emilmont 1:fdd22bb7aa52 1024 #else
emilmont 1:fdd22bb7aa52 1025
emilmont 2:da51fb522205 1026 x3 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 1027 a = *(px + 2);
emilmont 2:da51fb522205 1028 x2 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 1029
emilmont 2:da51fb522205 1030 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 1031
emilmont 2:da51fb522205 1032 px += 2u;
emilmont 1:fdd22bb7aa52 1033
emilmont 1:fdd22bb7aa52 1034 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 1035 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 1036 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 1037 acc2 = __SMLAD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 1038 acc3 = __SMLAD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 1039 }
emilmont 1:fdd22bb7aa52 1040
emilmont 1:fdd22bb7aa52 1041 if(k == 3u)
emilmont 1:fdd22bb7aa52 1042 {
emilmont 1:fdd22bb7aa52 1043 /* Read y[4], y[5] */
emilmont 2:da51fb522205 1044 a = *py;
emilmont 2:da51fb522205 1045 b = *(py + 1);
emilmont 2:da51fb522205 1046
emilmont 1:fdd22bb7aa52 1047 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 2:da51fb522205 1048
emilmont 2:da51fb522205 1049 c0 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 1050
emilmont 1:fdd22bb7aa52 1051 #else
emilmont 2:da51fb522205 1052
emilmont 2:da51fb522205 1053 c0 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 1054
emilmont 2:da51fb522205 1055 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 1056
emilmont 2:da51fb522205 1057 py += 2u;
emilmont 1:fdd22bb7aa52 1058
emilmont 1:fdd22bb7aa52 1059 /* Read x[7], x[8], x[9] */
emilmont 2:da51fb522205 1060 a = *px;
emilmont 2:da51fb522205 1061 b = *(px + 1);
emilmont 1:fdd22bb7aa52 1062
emilmont 1:fdd22bb7aa52 1063 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 1064
emilmont 2:da51fb522205 1065 x3 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 1066 a = *(px + 2);
emilmont 2:da51fb522205 1067 x2 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 1068
emilmont 1:fdd22bb7aa52 1069 #else
emilmont 1:fdd22bb7aa52 1070
emilmont 2:da51fb522205 1071 x3 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 1072 a = *(px + 2);
emilmont 2:da51fb522205 1073 x2 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 1074
emilmont 2:da51fb522205 1075 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 1076
emilmont 1:fdd22bb7aa52 1077 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 1078 acc0 = __SMLAD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 1079 acc1 = __SMLAD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 1080 acc2 = __SMLAD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 1081 acc3 = __SMLAD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 1082
emilmont 1:fdd22bb7aa52 1083 c0 = (*py);
emilmont 1:fdd22bb7aa52 1084 /* Read y[6] */
emilmont 1:fdd22bb7aa52 1085 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 1086
emilmont 1:fdd22bb7aa52 1087 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 1088 #else
emilmont 1:fdd22bb7aa52 1089
emilmont 1:fdd22bb7aa52 1090 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 1091 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 1092
emilmont 1:fdd22bb7aa52 1093 /* Read x[10] */
emilmont 2:da51fb522205 1094 b = *(px + 3);
emilmont 2:da51fb522205 1095
emilmont 1:fdd22bb7aa52 1096 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 2:da51fb522205 1097
emilmont 2:da51fb522205 1098 x3 = __PKHBT(a, b, 16);
emilmont 2:da51fb522205 1099
emilmont 1:fdd22bb7aa52 1100 #else
emilmont 2:da51fb522205 1101
emilmont 2:da51fb522205 1102 x3 = __PKHBT(b, a, 16);
emilmont 2:da51fb522205 1103
emilmont 2:da51fb522205 1104 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 1105
emilmont 2:da51fb522205 1106 px += 3u;
emilmont 1:fdd22bb7aa52 1107
emilmont 1:fdd22bb7aa52 1108 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 1109 acc0 = __SMLADX(x1, c0, acc0);
emilmont 1:fdd22bb7aa52 1110 acc1 = __SMLAD(x2, c0, acc1);
emilmont 1:fdd22bb7aa52 1111 acc2 = __SMLADX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 1112 acc3 = __SMLADX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 1113 }
emilmont 1:fdd22bb7aa52 1114
emilmont 1:fdd22bb7aa52 1115 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 1116 *pOut = (q15_t) (acc0 >> 15);
emilmont 1:fdd22bb7aa52 1117 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 1118 pOut += inc;
emilmont 1:fdd22bb7aa52 1119
emilmont 1:fdd22bb7aa52 1120 *pOut = (q15_t) (acc1 >> 15);
emilmont 1:fdd22bb7aa52 1121 pOut += inc;
emilmont 1:fdd22bb7aa52 1122
emilmont 1:fdd22bb7aa52 1123 *pOut = (q15_t) (acc2 >> 15);
emilmont 1:fdd22bb7aa52 1124 pOut += inc;
emilmont 1:fdd22bb7aa52 1125
emilmont 1:fdd22bb7aa52 1126 *pOut = (q15_t) (acc3 >> 15);
emilmont 1:fdd22bb7aa52 1127 pOut += inc;
emilmont 1:fdd22bb7aa52 1128
emilmont 1:fdd22bb7aa52 1129 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 1130 count += 4u;
emilmont 1:fdd22bb7aa52 1131
emilmont 1:fdd22bb7aa52 1132 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 1133 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 1134 py = pIn2;
emilmont 1:fdd22bb7aa52 1135
emilmont 1:fdd22bb7aa52 1136
emilmont 1:fdd22bb7aa52 1137 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1138 blkCnt--;
emilmont 1:fdd22bb7aa52 1139 }
emilmont 1:fdd22bb7aa52 1140
emilmont 1:fdd22bb7aa52 1141 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 1142 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 1143 blkCnt = blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 1144
emilmont 1:fdd22bb7aa52 1145 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 1146 {
emilmont 1:fdd22bb7aa52 1147 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 1148 sum = 0;
emilmont 1:fdd22bb7aa52 1149
emilmont 1:fdd22bb7aa52 1150 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 1151 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 1152
emilmont 1:fdd22bb7aa52 1153 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 1154 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 1155 while(k > 0u)
emilmont 1:fdd22bb7aa52 1156 {
emilmont 1:fdd22bb7aa52 1157 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 1158 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1159 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1160 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1161 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1162
emilmont 1:fdd22bb7aa52 1163 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1164 k--;
emilmont 1:fdd22bb7aa52 1165 }
emilmont 1:fdd22bb7aa52 1166
emilmont 1:fdd22bb7aa52 1167 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 1168 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 1169 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 1170
emilmont 1:fdd22bb7aa52 1171 while(k > 0u)
emilmont 1:fdd22bb7aa52 1172 {
emilmont 1:fdd22bb7aa52 1173 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 1174 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1175
emilmont 1:fdd22bb7aa52 1176 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1177 k--;
emilmont 1:fdd22bb7aa52 1178 }
emilmont 1:fdd22bb7aa52 1179
emilmont 1:fdd22bb7aa52 1180 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 1181 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 1182 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 1183 pOut += inc;
emilmont 1:fdd22bb7aa52 1184
emilmont 1:fdd22bb7aa52 1185 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 1186 count++;
emilmont 1:fdd22bb7aa52 1187
emilmont 1:fdd22bb7aa52 1188 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 1189 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 1190 py = pIn2;
emilmont 1:fdd22bb7aa52 1191
emilmont 1:fdd22bb7aa52 1192 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1193 blkCnt--;
emilmont 1:fdd22bb7aa52 1194 }
emilmont 1:fdd22bb7aa52 1195 }
emilmont 1:fdd22bb7aa52 1196 else
emilmont 1:fdd22bb7aa52 1197 {
emilmont 1:fdd22bb7aa52 1198 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 1199 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 1200 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 1201
emilmont 1:fdd22bb7aa52 1202 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 1203 {
emilmont 1:fdd22bb7aa52 1204 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 1205 sum = 0;
emilmont 1:fdd22bb7aa52 1206
emilmont 1:fdd22bb7aa52 1207 /* Loop over srcBLen */
emilmont 1:fdd22bb7aa52 1208 k = srcBLen;
emilmont 1:fdd22bb7aa52 1209
emilmont 1:fdd22bb7aa52 1210 while(k > 0u)
emilmont 1:fdd22bb7aa52 1211 {
emilmont 1:fdd22bb7aa52 1212 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 1213 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1214
emilmont 1:fdd22bb7aa52 1215 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1216 k--;
emilmont 1:fdd22bb7aa52 1217 }
emilmont 1:fdd22bb7aa52 1218
emilmont 1:fdd22bb7aa52 1219 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 1220 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 1221 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 1222 pOut += inc;
emilmont 1:fdd22bb7aa52 1223
emilmont 1:fdd22bb7aa52 1224 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 1225 count++;
emilmont 1:fdd22bb7aa52 1226
emilmont 1:fdd22bb7aa52 1227 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 1228 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 1229 py = pIn2;
emilmont 1:fdd22bb7aa52 1230
emilmont 1:fdd22bb7aa52 1231 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1232 blkCnt--;
emilmont 1:fdd22bb7aa52 1233 }
emilmont 1:fdd22bb7aa52 1234 }
emilmont 1:fdd22bb7aa52 1235
emilmont 1:fdd22bb7aa52 1236 /* --------------------------
emilmont 1:fdd22bb7aa52 1237 * Initializations of stage3
emilmont 1:fdd22bb7aa52 1238 * -------------------------*/
emilmont 1:fdd22bb7aa52 1239
emilmont 1:fdd22bb7aa52 1240 /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 1241 * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 1242 * ....
emilmont 1:fdd22bb7aa52 1243 * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 1244 * sum += x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 1245 */
emilmont 1:fdd22bb7aa52 1246
emilmont 1:fdd22bb7aa52 1247 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 1248 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 1249 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 1250
emilmont 1:fdd22bb7aa52 1251 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 1252 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 1253 px = pSrc1;
emilmont 1:fdd22bb7aa52 1254
emilmont 1:fdd22bb7aa52 1255 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 1256 py = pIn2;
emilmont 1:fdd22bb7aa52 1257
emilmont 1:fdd22bb7aa52 1258 /* -------------------
emilmont 1:fdd22bb7aa52 1259 * Stage3 process
emilmont 1:fdd22bb7aa52 1260 * ------------------*/
emilmont 1:fdd22bb7aa52 1261
emilmont 1:fdd22bb7aa52 1262 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 1263 {
emilmont 1:fdd22bb7aa52 1264 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 1265 sum = 0;
emilmont 1:fdd22bb7aa52 1266
emilmont 1:fdd22bb7aa52 1267 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 1268 k = count >> 2u;
emilmont 1:fdd22bb7aa52 1269
emilmont 1:fdd22bb7aa52 1270 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 1271 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 1272 while(k > 0u)
emilmont 1:fdd22bb7aa52 1273 {
emilmont 1:fdd22bb7aa52 1274 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 1275 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1276 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1277 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1278 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1279
emilmont 1:fdd22bb7aa52 1280 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1281 k--;
emilmont 1:fdd22bb7aa52 1282 }
emilmont 1:fdd22bb7aa52 1283
emilmont 1:fdd22bb7aa52 1284 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 1285 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 1286 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 1287
emilmont 1:fdd22bb7aa52 1288 while(k > 0u)
emilmont 1:fdd22bb7aa52 1289 {
emilmont 1:fdd22bb7aa52 1290 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 1291 sum += ((q31_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 1292
emilmont 1:fdd22bb7aa52 1293 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1294 k--;
emilmont 1:fdd22bb7aa52 1295 }
emilmont 1:fdd22bb7aa52 1296
emilmont 1:fdd22bb7aa52 1297 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 1298 *pOut = (q15_t) (sum >> 15);
emilmont 1:fdd22bb7aa52 1299 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 1300 pOut += inc;
emilmont 1:fdd22bb7aa52 1301
emilmont 1:fdd22bb7aa52 1302 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 1303 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 1304 py = pIn2;
emilmont 1:fdd22bb7aa52 1305
emilmont 1:fdd22bb7aa52 1306 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 1307 count--;
emilmont 1:fdd22bb7aa52 1308
emilmont 1:fdd22bb7aa52 1309 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 1310 blockSize3--;
emilmont 1:fdd22bb7aa52 1311 }
emilmont 1:fdd22bb7aa52 1312
emilmont 1:fdd22bb7aa52 1313 #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
emilmont 1:fdd22bb7aa52 1314
emilmont 1:fdd22bb7aa52 1315 }
emilmont 1:fdd22bb7aa52 1316
emilmont 1:fdd22bb7aa52 1317 /**
emilmont 1:fdd22bb7aa52 1318 * @} end of Corr group
emilmont 1:fdd22bb7aa52 1319 */