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@1:fdd22bb7aa52, 2012-11-28 (annotated)

Committer:
emilmont
Date:
Wed Nov 28 12:30:09 2012 +0000
Revision:
1:fdd22bb7aa52
Child:
2:da51fb522205
DSP library code

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emilmont 1:fdd22bb7aa52 1 /* ----------------------------------------------------------------------
emilmont 1:fdd22bb7aa52 2 * Copyright (C) 2010 ARM Limited. All rights reserved.
emilmont 1:fdd22bb7aa52 3 *
emilmont 1:fdd22bb7aa52 4 * $Date: 15. February 2012
emilmont 1:fdd22bb7aa52 5 * $Revision: V1.1.0
emilmont 1:fdd22bb7aa52 6 *
emilmont 1:fdd22bb7aa52 7 * Project: CMSIS DSP Library
emilmont 1:fdd22bb7aa52 8 * Title: arm_correlate_q15.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 1:fdd22bb7aa52 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 *
emilmont 1:fdd22bb7aa52 14 * Version 1.1.0 2012/02/15
emilmont 1:fdd22bb7aa52 15 * Updated with more optimizations, bug fixes and minor API changes.
emilmont 1:fdd22bb7aa52 16 *
emilmont 1:fdd22bb7aa52 17 * Version 1.0.11 2011/10/18
emilmont 1:fdd22bb7aa52 18 * Bug Fix in conv, correlation, partial convolution.
emilmont 1:fdd22bb7aa52 19 *
emilmont 1:fdd22bb7aa52 20 * Version 1.0.10 2011/7/15
emilmont 1:fdd22bb7aa52 21 * Big Endian support added and Merged M0 and M3/M4 Source code.
emilmont 1:fdd22bb7aa52 22 *
emilmont 1:fdd22bb7aa52 23 * Version 1.0.3 2010/11/29
emilmont 1:fdd22bb7aa52 24 * Re-organized the CMSIS folders and updated documentation.
emilmont 1:fdd22bb7aa52 25 *
emilmont 1:fdd22bb7aa52 26 * Version 1.0.2 2010/11/11
emilmont 1:fdd22bb7aa52 27 * Documentation updated.
emilmont 1:fdd22bb7aa52 28 *
emilmont 1:fdd22bb7aa52 29 * Version 1.0.1 2010/10/05
emilmont 1:fdd22bb7aa52 30 * Production release and review comments incorporated.
emilmont 1:fdd22bb7aa52 31 *
emilmont 1:fdd22bb7aa52 32 * Version 1.0.0 2010/09/20
emilmont 1:fdd22bb7aa52 33 * Production release and review comments incorporated
emilmont 1:fdd22bb7aa52 34 *
emilmont 1:fdd22bb7aa52 35 * Version 0.0.7 2010/06/10
emilmont 1:fdd22bb7aa52 36 * Misra-C changes done
emilmont 1:fdd22bb7aa52 37 *
emilmont 1:fdd22bb7aa52 38 * -------------------------------------------------------------------- */
emilmont 1:fdd22bb7aa52 39
emilmont 1:fdd22bb7aa52 40 #include "arm_math.h"
emilmont 1:fdd22bb7aa52 41
emilmont 1:fdd22bb7aa52 42 /**
emilmont 1:fdd22bb7aa52 43 * @ingroup groupFilters
emilmont 1:fdd22bb7aa52 44 */
emilmont 1:fdd22bb7aa52 45
emilmont 1:fdd22bb7aa52 46 /**
emilmont 1:fdd22bb7aa52 47 * @addtogroup Corr
emilmont 1:fdd22bb7aa52 48 * @{
emilmont 1:fdd22bb7aa52 49 */
emilmont 1:fdd22bb7aa52 50
emilmont 1:fdd22bb7aa52 51 /**
emilmont 1:fdd22bb7aa52 52 * @brief Correlation of Q15 sequences.
emilmont 1:fdd22bb7aa52 53 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 54 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[out] *pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1.
emilmont 1:fdd22bb7aa52 58 * @return none.
emilmont 1:fdd22bb7aa52 59 *
emilmont 1:fdd22bb7aa52 60 * @details
emilmont 1:fdd22bb7aa52 61 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 62 *
emilmont 1:fdd22bb7aa52 63 * \par
emilmont 1:fdd22bb7aa52 64 * The function is implemented using a 64-bit internal accumulator.
emilmont 1:fdd22bb7aa52 65 * Both inputs are in 1.15 format and multiplications yield a 2.30 result.
emilmont 1:fdd22bb7aa52 66 * The 2.30 intermediate results are accumulated in a 64-bit accumulator in 34.30 format.
emilmont 1:fdd22bb7aa52 67 * This approach provides 33 guard bits and there is no risk of overflow.
emilmont 1:fdd22bb7aa52 68 * 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 69 *
emilmont 1:fdd22bb7aa52 70 * \par
emilmont 1:fdd22bb7aa52 71 * 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 72 *
emilmont 1:fdd22bb7aa52 73 * \par
emilmont 1:fdd22bb7aa52 74 * Refer the function <code>arm_correlate_opt_q15()</code> for a faster implementation of this function using scratch buffers.
emilmont 1:fdd22bb7aa52 75 *
emilmont 1:fdd22bb7aa52 76 */
emilmont 1:fdd22bb7aa52 77
emilmont 1:fdd22bb7aa52 78 void arm_correlate_q15(
emilmont 1:fdd22bb7aa52 79 q15_t * pSrcA,
emilmont 1:fdd22bb7aa52 80 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 81 q15_t * pSrcB,
emilmont 1:fdd22bb7aa52 82 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 83 q15_t * pDst)
emilmont 1:fdd22bb7aa52 84 {
emilmont 1:fdd22bb7aa52 85
emilmont 1:fdd22bb7aa52 86 #if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE)
emilmont 1:fdd22bb7aa52 87
emilmont 1:fdd22bb7aa52 88 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 89
emilmont 1:fdd22bb7aa52 90 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 91 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 92 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 93 q63_t sum, acc0, acc1, acc2, acc3; /* Accumulators */
emilmont 1:fdd22bb7aa52 94 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 95 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 96 q15_t *pSrc1; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 97 q31_t x0, x1, x2, x3, c0; /* temporary variables for holding input and coefficient values */
emilmont 1:fdd22bb7aa52 98 uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 99 int32_t inc = 1; /* Destination address modifier */
emilmont 1:fdd22bb7aa52 100
emilmont 1:fdd22bb7aa52 101
emilmont 1:fdd22bb7aa52 102 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 103 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 104 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 105 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 106 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 107 /* and the destination pointer modifier, inc is set to -1 */
emilmont 1:fdd22bb7aa52 108 /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */
emilmont 1:fdd22bb7aa52 109 /* But to improve the performance,
emilmont 1:fdd22bb7aa52 110 * we include zeroes in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 111 /* If srcALen > srcBLen,
emilmont 1:fdd22bb7aa52 112 * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */
emilmont 1:fdd22bb7aa52 113 /* If srcALen < srcBLen,
emilmont 1:fdd22bb7aa52 114 * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */
emilmont 1:fdd22bb7aa52 115 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 116 {
emilmont 1:fdd22bb7aa52 117 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 118 pIn1 = (pSrcA);
emilmont 1:fdd22bb7aa52 119
emilmont 1:fdd22bb7aa52 120 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 121 pIn2 = (pSrcB);
emilmont 1:fdd22bb7aa52 122
emilmont 1:fdd22bb7aa52 123 /* Number of output samples is calculated */
emilmont 1:fdd22bb7aa52 124 outBlockSize = (2u * srcALen) - 1u;
emilmont 1:fdd22bb7aa52 125
emilmont 1:fdd22bb7aa52 126 /* When srcALen > srcBLen, zero padding is done to srcB
emilmont 1:fdd22bb7aa52 127 * to make their lengths equal.
emilmont 1:fdd22bb7aa52 128 * Instead, (outBlockSize - (srcALen + srcBLen - 1))
emilmont 1:fdd22bb7aa52 129 * number of output samples are made zero */
emilmont 1:fdd22bb7aa52 130 j = outBlockSize - (srcALen + (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 131
emilmont 1:fdd22bb7aa52 132 /* Updating the pointer position to non zero value */
emilmont 1:fdd22bb7aa52 133 pOut += j;
emilmont 1:fdd22bb7aa52 134
emilmont 1:fdd22bb7aa52 135 }
emilmont 1:fdd22bb7aa52 136 else
emilmont 1:fdd22bb7aa52 137 {
emilmont 1:fdd22bb7aa52 138 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 139 pIn1 = (pSrcB);
emilmont 1:fdd22bb7aa52 140
emilmont 1:fdd22bb7aa52 141 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 142 pIn2 = (pSrcA);
emilmont 1:fdd22bb7aa52 143
emilmont 1:fdd22bb7aa52 144 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 145 j = srcBLen;
emilmont 1:fdd22bb7aa52 146 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 147 srcALen = j;
emilmont 1:fdd22bb7aa52 148
emilmont 1:fdd22bb7aa52 149 /* CORR(x, y) = Reverse order(CORR(y, x)) */
emilmont 1:fdd22bb7aa52 150 /* Hence set the destination pointer to point to the last output sample */
emilmont 1:fdd22bb7aa52 151 pOut = pDst + ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 152
emilmont 1:fdd22bb7aa52 153 /* Destination address modifier is set to -1 */
emilmont 1:fdd22bb7aa52 154 inc = -1;
emilmont 1:fdd22bb7aa52 155
emilmont 1:fdd22bb7aa52 156 }
emilmont 1:fdd22bb7aa52 157
emilmont 1:fdd22bb7aa52 158 /* The function is internally
emilmont 1:fdd22bb7aa52 159 * divided into three parts according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 160 * taken place between inputA samples and inputB samples. In the first part of the
emilmont 1:fdd22bb7aa52 161 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 162 * In the second part of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 163 * In the third part of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 164 * for every iteration.*/
emilmont 1:fdd22bb7aa52 165 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 166 * The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 167 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 168 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 169 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 170
emilmont 1:fdd22bb7aa52 171 /* --------------------------
emilmont 1:fdd22bb7aa52 172 * Initializations of stage1
emilmont 1:fdd22bb7aa52 173 * -------------------------*/
emilmont 1:fdd22bb7aa52 174
emilmont 1:fdd22bb7aa52 175 /* sum = x[0] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 176 * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 177 * ....
emilmont 1:fdd22bb7aa52 178 * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1]
emilmont 1:fdd22bb7aa52 179 */
emilmont 1:fdd22bb7aa52 180
emilmont 1:fdd22bb7aa52 181 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 182 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 183 count = 1u;
emilmont 1:fdd22bb7aa52 184
emilmont 1:fdd22bb7aa52 185 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 186 px = pIn1;
emilmont 1:fdd22bb7aa52 187
emilmont 1:fdd22bb7aa52 188 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 189 pSrc1 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 190 py = pSrc1;
emilmont 1:fdd22bb7aa52 191
emilmont 1:fdd22bb7aa52 192 /* ------------------------
emilmont 1:fdd22bb7aa52 193 * Stage1 process
emilmont 1:fdd22bb7aa52 194 * ----------------------*/
emilmont 1:fdd22bb7aa52 195
emilmont 1:fdd22bb7aa52 196 /* The first loop starts here */
emilmont 1:fdd22bb7aa52 197 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 198 {
emilmont 1:fdd22bb7aa52 199 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 200 sum = 0;
emilmont 1:fdd22bb7aa52 201
emilmont 1:fdd22bb7aa52 202 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 203 k = count >> 2;
emilmont 1:fdd22bb7aa52 204
emilmont 1:fdd22bb7aa52 205 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 206 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 207 while(k > 0u)
emilmont 1:fdd22bb7aa52 208 {
emilmont 1:fdd22bb7aa52 209 /* x[0] * y[srcBLen - 4] , x[1] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 210 sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 211 /* x[3] * y[srcBLen - 1] , x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 212 sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
emilmont 1:fdd22bb7aa52 213
emilmont 1:fdd22bb7aa52 214 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 215 k--;
emilmont 1:fdd22bb7aa52 216 }
emilmont 1:fdd22bb7aa52 217
emilmont 1:fdd22bb7aa52 218 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 219 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 220 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 221
emilmont 1:fdd22bb7aa52 222 while(k > 0u)
emilmont 1:fdd22bb7aa52 223 {
emilmont 1:fdd22bb7aa52 224 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 225 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 226 sum = __SMLALD(*px++, *py++, sum);
emilmont 1:fdd22bb7aa52 227
emilmont 1:fdd22bb7aa52 228 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 229 k--;
emilmont 1:fdd22bb7aa52 230 }
emilmont 1:fdd22bb7aa52 231
emilmont 1:fdd22bb7aa52 232 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 233 *pOut = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 234 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 235 pOut += inc;
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 238 py = pSrc1 - count;
emilmont 1:fdd22bb7aa52 239 px = pIn1;
emilmont 1:fdd22bb7aa52 240
emilmont 1:fdd22bb7aa52 241 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 242 count++;
emilmont 1:fdd22bb7aa52 243
emilmont 1:fdd22bb7aa52 244 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 245 blockSize1--;
emilmont 1:fdd22bb7aa52 246 }
emilmont 1:fdd22bb7aa52 247
emilmont 1:fdd22bb7aa52 248 /* --------------------------
emilmont 1:fdd22bb7aa52 249 * Initializations of stage2
emilmont 1:fdd22bb7aa52 250 * ------------------------*/
emilmont 1:fdd22bb7aa52 251
emilmont 1:fdd22bb7aa52 252 /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 253 * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 254 * ....
emilmont 1:fdd22bb7aa52 255 * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 256 */
emilmont 1:fdd22bb7aa52 257
emilmont 1:fdd22bb7aa52 258 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 259 px = pIn1;
emilmont 1:fdd22bb7aa52 260
emilmont 1:fdd22bb7aa52 261 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 262 py = pIn2;
emilmont 1:fdd22bb7aa52 263
emilmont 1:fdd22bb7aa52 264 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 265 count = 0u;
emilmont 1:fdd22bb7aa52 266
emilmont 1:fdd22bb7aa52 267 /* -------------------
emilmont 1:fdd22bb7aa52 268 * Stage2 process
emilmont 1:fdd22bb7aa52 269 * ------------------*/
emilmont 1:fdd22bb7aa52 270
emilmont 1:fdd22bb7aa52 271 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 272 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 273 * srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */
emilmont 1:fdd22bb7aa52 274 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 275 {
emilmont 1:fdd22bb7aa52 276 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 277 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 278
emilmont 1:fdd22bb7aa52 279 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 280 {
emilmont 1:fdd22bb7aa52 281 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 282 acc0 = 0;
emilmont 1:fdd22bb7aa52 283 acc1 = 0;
emilmont 1:fdd22bb7aa52 284 acc2 = 0;
emilmont 1:fdd22bb7aa52 285 acc3 = 0;
emilmont 1:fdd22bb7aa52 286
emilmont 1:fdd22bb7aa52 287 /* read x[0], x[1] samples */
emilmont 1:fdd22bb7aa52 288 x0 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 289 /* read x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 290 x1 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 291 px += 2u;
emilmont 1:fdd22bb7aa52 292
emilmont 1:fdd22bb7aa52 293 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 294 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 295
emilmont 1:fdd22bb7aa52 296 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 297 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 298 do
emilmont 1:fdd22bb7aa52 299 {
emilmont 1:fdd22bb7aa52 300 /* Read the first two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 301 * y[0] and y[1] */
emilmont 1:fdd22bb7aa52 302 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 303
emilmont 1:fdd22bb7aa52 304 /* acc0 += x[0] * y[0] + x[1] * y[1] */
emilmont 1:fdd22bb7aa52 305 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 306
emilmont 1:fdd22bb7aa52 307 /* acc1 += x[1] * y[0] + x[2] * y[1] */
emilmont 1:fdd22bb7aa52 308 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 309
emilmont 1:fdd22bb7aa52 310 /* Read x[2], x[3] */
emilmont 1:fdd22bb7aa52 311 x2 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 312
emilmont 1:fdd22bb7aa52 313 /* Read x[3], x[4] */
emilmont 1:fdd22bb7aa52 314 x3 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 315
emilmont 1:fdd22bb7aa52 316 /* acc2 += x[2] * y[0] + x[3] * y[1] */
emilmont 1:fdd22bb7aa52 317 acc2 = __SMLALD(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 318
emilmont 1:fdd22bb7aa52 319 /* acc3 += x[3] * y[0] + x[4] * y[1] */
emilmont 1:fdd22bb7aa52 320 acc3 = __SMLALD(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 321
emilmont 1:fdd22bb7aa52 322 /* Read y[2] and y[3] */
emilmont 1:fdd22bb7aa52 323 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 324
emilmont 1:fdd22bb7aa52 325 /* acc0 += x[2] * y[2] + x[3] * y[3] */
emilmont 1:fdd22bb7aa52 326 acc0 = __SMLALD(x2, c0, acc0);
emilmont 1:fdd22bb7aa52 327
emilmont 1:fdd22bb7aa52 328 /* acc1 += x[3] * y[2] + x[4] * y[3] */
emilmont 1:fdd22bb7aa52 329 acc1 = __SMLALD(x3, c0, acc1);
emilmont 1:fdd22bb7aa52 330
emilmont 1:fdd22bb7aa52 331 /* Read x[4], x[5] */
emilmont 1:fdd22bb7aa52 332 x0 = _SIMD32_OFFSET(px + 2);
emilmont 1:fdd22bb7aa52 333
emilmont 1:fdd22bb7aa52 334 /* Read x[5], x[6] */
emilmont 1:fdd22bb7aa52 335 x1 = _SIMD32_OFFSET(px + 3);
emilmont 1:fdd22bb7aa52 336
emilmont 1:fdd22bb7aa52 337 px += 4u;
emilmont 1:fdd22bb7aa52 338
emilmont 1:fdd22bb7aa52 339 /* acc2 += x[4] * y[2] + x[5] * y[3] */
emilmont 1:fdd22bb7aa52 340 acc2 = __SMLALD(x0, c0, acc2);
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 /* acc3 += x[5] * y[2] + x[6] * y[3] */
emilmont 1:fdd22bb7aa52 343 acc3 = __SMLALD(x1, c0, acc3);
emilmont 1:fdd22bb7aa52 344
emilmont 1:fdd22bb7aa52 345 } while(--k);
emilmont 1:fdd22bb7aa52 346
emilmont 1:fdd22bb7aa52 347 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 348 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 349 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 350
emilmont 1:fdd22bb7aa52 351 if(k == 1u)
emilmont 1:fdd22bb7aa52 352 {
emilmont 1:fdd22bb7aa52 353 /* Read y[4] */
emilmont 1:fdd22bb7aa52 354 c0 = *py;
emilmont 1:fdd22bb7aa52 355 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 356
emilmont 1:fdd22bb7aa52 357 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 358
emilmont 1:fdd22bb7aa52 359 #else
emilmont 1:fdd22bb7aa52 360
emilmont 1:fdd22bb7aa52 361 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 362
emilmont 1:fdd22bb7aa52 363 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 364 /* Read x[7] */
emilmont 1:fdd22bb7aa52 365 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 366 px++;
emilmont 1:fdd22bb7aa52 367
emilmont 1:fdd22bb7aa52 368 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 369 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 370 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 371 acc2 = __SMLALDX(x1, c0, acc2);
emilmont 1:fdd22bb7aa52 372 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 373 }
emilmont 1:fdd22bb7aa52 374
emilmont 1:fdd22bb7aa52 375 if(k == 2u)
emilmont 1:fdd22bb7aa52 376 {
emilmont 1:fdd22bb7aa52 377 /* Read y[4], y[5] */
emilmont 1:fdd22bb7aa52 378 c0 = *__SIMD32(py);
emilmont 1:fdd22bb7aa52 379
emilmont 1:fdd22bb7aa52 380 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 381 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 382
emilmont 1:fdd22bb7aa52 383 /* Read x[9] */
emilmont 1:fdd22bb7aa52 384 x2 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 385 px += 2u;
emilmont 1:fdd22bb7aa52 386
emilmont 1:fdd22bb7aa52 387 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 388 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 389 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 390 acc2 = __SMLALD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 391 acc3 = __SMLALD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 392 }
emilmont 1:fdd22bb7aa52 393
emilmont 1:fdd22bb7aa52 394 if(k == 3u)
emilmont 1:fdd22bb7aa52 395 {
emilmont 1:fdd22bb7aa52 396 /* Read y[4], y[5] */
emilmont 1:fdd22bb7aa52 397 c0 = *__SIMD32(py)++;
emilmont 1:fdd22bb7aa52 398
emilmont 1:fdd22bb7aa52 399 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 400 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 401
emilmont 1:fdd22bb7aa52 402 /* Read x[9] */
emilmont 1:fdd22bb7aa52 403 x2 = _SIMD32_OFFSET(px + 1);
emilmont 1:fdd22bb7aa52 404
emilmont 1:fdd22bb7aa52 405 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 406 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 407 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 408 acc2 = __SMLALD(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 409 acc3 = __SMLALD(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 410
emilmont 1:fdd22bb7aa52 411 c0 = (*py);
emilmont 1:fdd22bb7aa52 412
emilmont 1:fdd22bb7aa52 413 /* Read y[6] */
emilmont 1:fdd22bb7aa52 414 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 415
emilmont 1:fdd22bb7aa52 416 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 417 #else
emilmont 1:fdd22bb7aa52 418
emilmont 1:fdd22bb7aa52 419 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 420 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 421 /* Read x[10] */
emilmont 1:fdd22bb7aa52 422 x3 = _SIMD32_OFFSET(px + 2);
emilmont 1:fdd22bb7aa52 423 px += 3u;
emilmont 1:fdd22bb7aa52 424
emilmont 1:fdd22bb7aa52 425 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 426 acc0 = __SMLALDX(x1, c0, acc0);
emilmont 1:fdd22bb7aa52 427 acc1 = __SMLALD(x2, c0, acc1);
emilmont 1:fdd22bb7aa52 428 acc2 = __SMLALDX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 429 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 430 }
emilmont 1:fdd22bb7aa52 431
emilmont 1:fdd22bb7aa52 432 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 433 *pOut = (q15_t) (__SSAT(acc0 >> 15, 16));
emilmont 1:fdd22bb7aa52 434 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 435 pOut += inc;
emilmont 1:fdd22bb7aa52 436
emilmont 1:fdd22bb7aa52 437 *pOut = (q15_t) (__SSAT(acc1 >> 15, 16));
emilmont 1:fdd22bb7aa52 438 pOut += inc;
emilmont 1:fdd22bb7aa52 439
emilmont 1:fdd22bb7aa52 440 *pOut = (q15_t) (__SSAT(acc2 >> 15, 16));
emilmont 1:fdd22bb7aa52 441 pOut += inc;
emilmont 1:fdd22bb7aa52 442
emilmont 1:fdd22bb7aa52 443 *pOut = (q15_t) (__SSAT(acc3 >> 15, 16));
emilmont 1:fdd22bb7aa52 444 pOut += inc;
emilmont 1:fdd22bb7aa52 445
emilmont 1:fdd22bb7aa52 446 /* Increment the count by 4 as 4 output values are computed */
emilmont 1:fdd22bb7aa52 447 count += 4u;
emilmont 1:fdd22bb7aa52 448
emilmont 1:fdd22bb7aa52 449 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 450 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 451 py = pIn2;
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 += ((q63_t) * px++ * *py++);
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
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 += ((q63_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) (__SSAT(sum >> 15, 16));
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 count by 1, as one output value is computed */
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 += ((q63_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) (__SSAT(sum >> 15, 16));
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 = __SMLALD(*__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 = __SMLALD(*__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 = __SMLALD(*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) (__SSAT((sum >> 15), 16));
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 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 632
emilmont 1:fdd22bb7aa52 633 q15_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 634 q15_t *pIn2 = pSrcB + (srcBLen - 1u); /* inputB pointer */
emilmont 1:fdd22bb7aa52 635 q63_t sum; /* Accumulators */
emilmont 1:fdd22bb7aa52 636 uint32_t i = 0u, j; /* loop counters */
emilmont 1:fdd22bb7aa52 637 uint32_t inv = 0u; /* Reverse order flag */
emilmont 1:fdd22bb7aa52 638 uint32_t tot = 0u; /* Length */
emilmont 1:fdd22bb7aa52 639
emilmont 1:fdd22bb7aa52 640 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 641 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 642 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 643 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 644 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 645 /* and a varaible, inv is set to 1 */
emilmont 1:fdd22bb7aa52 646 /* If lengths are not equal then zero pad has to be done to make the two
emilmont 1:fdd22bb7aa52 647 * inputs of same length. But to improve the performance, we include zeroes
emilmont 1:fdd22bb7aa52 648 * in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 649 /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the
emilmont 1:fdd22bb7aa52 650 * starting of the output buffer */
emilmont 1:fdd22bb7aa52 651 /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the
emilmont 1:fdd22bb7aa52 652 * ending of the output buffer */
emilmont 1:fdd22bb7aa52 653 /* Once the zero padding is done the remaining of the output is calcualted
emilmont 1:fdd22bb7aa52 654 * using convolution but with the shorter signal time shifted. */
emilmont 1:fdd22bb7aa52 655
emilmont 1:fdd22bb7aa52 656 /* Calculate the length of the remaining sequence */
emilmont 1:fdd22bb7aa52 657 tot = ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 658
emilmont 1:fdd22bb7aa52 659 if(srcALen > srcBLen)
emilmont 1:fdd22bb7aa52 660 {
emilmont 1:fdd22bb7aa52 661 /* Calculating the number of zeros to be padded to the output */
emilmont 1:fdd22bb7aa52 662 j = srcALen - srcBLen;
emilmont 1:fdd22bb7aa52 663
emilmont 1:fdd22bb7aa52 664 /* Initialise the pointer after zero padding */
emilmont 1:fdd22bb7aa52 665 pDst += j;
emilmont 1:fdd22bb7aa52 666 }
emilmont 1:fdd22bb7aa52 667
emilmont 1:fdd22bb7aa52 668 else if(srcALen < srcBLen)
emilmont 1:fdd22bb7aa52 669 {
emilmont 1:fdd22bb7aa52 670 /* Initialization to inputB pointer */
emilmont 1:fdd22bb7aa52 671 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 672
emilmont 1:fdd22bb7aa52 673 /* Initialization to the end of inputA pointer */
emilmont 1:fdd22bb7aa52 674 pIn2 = pSrcA + (srcALen - 1u);
emilmont 1:fdd22bb7aa52 675
emilmont 1:fdd22bb7aa52 676 /* Initialisation of the pointer after zero padding */
emilmont 1:fdd22bb7aa52 677 pDst = pDst + tot;
emilmont 1:fdd22bb7aa52 678
emilmont 1:fdd22bb7aa52 679 /* Swapping the lengths */
emilmont 1:fdd22bb7aa52 680 j = srcALen;
emilmont 1:fdd22bb7aa52 681 srcALen = srcBLen;
emilmont 1:fdd22bb7aa52 682 srcBLen = j;
emilmont 1:fdd22bb7aa52 683
emilmont 1:fdd22bb7aa52 684 /* Setting the reverse flag */
emilmont 1:fdd22bb7aa52 685 inv = 1;
emilmont 1:fdd22bb7aa52 686
emilmont 1:fdd22bb7aa52 687 }
emilmont 1:fdd22bb7aa52 688
emilmont 1:fdd22bb7aa52 689 /* Loop to calculate convolution for output length number of times */
emilmont 1:fdd22bb7aa52 690 for (i = 0u; i <= tot; i++)
emilmont 1:fdd22bb7aa52 691 {
emilmont 1:fdd22bb7aa52 692 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 693 sum = 0;
emilmont 1:fdd22bb7aa52 694
emilmont 1:fdd22bb7aa52 695 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 696 for (j = 0u; j <= i; j++)
emilmont 1:fdd22bb7aa52 697 {
emilmont 1:fdd22bb7aa52 698 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 699 if((((i - j) < srcBLen) && (j < srcALen)))
emilmont 1:fdd22bb7aa52 700 {
emilmont 1:fdd22bb7aa52 701 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 702 sum += ((q31_t) pIn1[j] * pIn2[-((int32_t) i - j)]);
emilmont 1:fdd22bb7aa52 703 }
emilmont 1:fdd22bb7aa52 704 }
emilmont 1:fdd22bb7aa52 705 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 706 if(inv == 1)
emilmont 1:fdd22bb7aa52 707 *pDst-- = (q15_t) __SSAT((sum >> 15u), 16u);
emilmont 1:fdd22bb7aa52 708 else
emilmont 1:fdd22bb7aa52 709 *pDst++ = (q15_t) __SSAT((sum >> 15u), 16u);
emilmont 1:fdd22bb7aa52 710 }
emilmont 1:fdd22bb7aa52 711
emilmont 1:fdd22bb7aa52 712 #endif /*#if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE) */
emilmont 1:fdd22bb7aa52 713
emilmont 1:fdd22bb7aa52 714 }
emilmont 1:fdd22bb7aa52 715
emilmont 1:fdd22bb7aa52 716 /**
emilmont 1:fdd22bb7aa52 717 * @} end of Corr group
emilmont 1:fdd22bb7aa52 718 */