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_conv_q31.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_conv_q31.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 1:fdd22bb7aa52 10 * Description: Convolution of Q31 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 Conv
emilmont 1:fdd22bb7aa52 48 * @{
emilmont 1:fdd22bb7aa52 49 */
emilmont 1:fdd22bb7aa52 50
emilmont 1:fdd22bb7aa52 51 /**
emilmont 1:fdd22bb7aa52 52 * @brief Convolution of Q31 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 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 an internal 64-bit accumulator.
emilmont 1:fdd22bb7aa52 65 * The accumulator has a 2.62 format and 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 the inputs by log2(min(srcALen, srcBLen)) (log2 is read as log to the base 2) times to avoid overflows,
emilmont 1:fdd22bb7aa52 70 * as maximum of min(srcALen, srcBLen) number of additions are carried internally.
emilmont 1:fdd22bb7aa52 71 * The 2.62 accumulator is right shifted by 31 bits and saturated to 1.31 format to yield the final result.
emilmont 1:fdd22bb7aa52 72 *
emilmont 1:fdd22bb7aa52 73 * \par
emilmont 1:fdd22bb7aa52 74 * See <code>arm_conv_fast_q31()</code> for a faster but less precise implementation of this function for Cortex-M3 and Cortex-M4.
emilmont 1:fdd22bb7aa52 75 */
emilmont 1:fdd22bb7aa52 76
emilmont 1:fdd22bb7aa52 77 void arm_conv_q31(
emilmont 1:fdd22bb7aa52 78 q31_t * pSrcA,
emilmont 1:fdd22bb7aa52 79 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 80 q31_t * pSrcB,
emilmont 1:fdd22bb7aa52 81 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 82 q31_t * pDst)
emilmont 1:fdd22bb7aa52 83 {
emilmont 1:fdd22bb7aa52 84
emilmont 1:fdd22bb7aa52 85
emilmont 1:fdd22bb7aa52 86 #ifndef ARM_MATH_CM0
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 q31_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 91 q31_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 92 q31_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 93 q31_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 94 q31_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 95 q31_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 96 q63_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 97 q63_t acc0, acc1, acc2; /* Accumulator */
emilmont 1:fdd22bb7aa52 98 q31_t x0, x1, x2, c0; /* Temporary variables to hold state and coefficient values */
emilmont 1:fdd22bb7aa52 99 uint32_t j, k, count, blkCnt, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 100
emilmont 1:fdd22bb7aa52 101 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 102 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 103 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 104 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 105 {
emilmont 1:fdd22bb7aa52 106 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 107 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 108
emilmont 1:fdd22bb7aa52 109 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 110 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 111 }
emilmont 1:fdd22bb7aa52 112 else
emilmont 1:fdd22bb7aa52 113 {
emilmont 1:fdd22bb7aa52 114 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 115 pIn1 = (q31_t *) pSrcB;
emilmont 1:fdd22bb7aa52 116
emilmont 1:fdd22bb7aa52 117 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 118 pIn2 = (q31_t *) pSrcA;
emilmont 1:fdd22bb7aa52 119
emilmont 1:fdd22bb7aa52 120 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 121 j = srcBLen;
emilmont 1:fdd22bb7aa52 122 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 123 srcALen = j;
emilmont 1:fdd22bb7aa52 124 }
emilmont 1:fdd22bb7aa52 125
emilmont 1:fdd22bb7aa52 126 /* conv(x,y) at n = x[n] * y[0] + x[n-1] * y[1] + x[n-2] * y[2] + ...+ x[n-N+1] * y[N -1] */
emilmont 1:fdd22bb7aa52 127 /* The function is internally
emilmont 1:fdd22bb7aa52 128 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 129 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 130 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 131 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 132 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 133 * for every iteration. */
emilmont 1:fdd22bb7aa52 134
emilmont 1:fdd22bb7aa52 135 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 136 The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 137 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 138 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 139 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 140
emilmont 1:fdd22bb7aa52 141 /* --------------------------
emilmont 1:fdd22bb7aa52 142 * Initializations of stage1
emilmont 1:fdd22bb7aa52 143 * -------------------------*/
emilmont 1:fdd22bb7aa52 144
emilmont 1:fdd22bb7aa52 145 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 146 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 147 * ....
emilmont 1:fdd22bb7aa52 148 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 149 */
emilmont 1:fdd22bb7aa52 150
emilmont 1:fdd22bb7aa52 151 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 152 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 153 count = 1u;
emilmont 1:fdd22bb7aa52 154
emilmont 1:fdd22bb7aa52 155 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 156 px = pIn1;
emilmont 1:fdd22bb7aa52 157
emilmont 1:fdd22bb7aa52 158 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 159 py = pIn2;
emilmont 1:fdd22bb7aa52 160
emilmont 1:fdd22bb7aa52 161
emilmont 1:fdd22bb7aa52 162 /* ------------------------
emilmont 1:fdd22bb7aa52 163 * Stage1 process
emilmont 1:fdd22bb7aa52 164 * ----------------------*/
emilmont 1:fdd22bb7aa52 165
emilmont 1:fdd22bb7aa52 166 /* The first stage starts here */
emilmont 1:fdd22bb7aa52 167 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 168 {
emilmont 1:fdd22bb7aa52 169 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 170 sum = 0;
emilmont 1:fdd22bb7aa52 171
emilmont 1:fdd22bb7aa52 172 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 173 k = count >> 2u;
emilmont 1:fdd22bb7aa52 174
emilmont 1:fdd22bb7aa52 175 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 176 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 177 while(k > 0u)
emilmont 1:fdd22bb7aa52 178 {
emilmont 1:fdd22bb7aa52 179 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 180 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 181 /* x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 182 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 183 /* x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 184 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 185 /* x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 186 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 187
emilmont 1:fdd22bb7aa52 188 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 189 k--;
emilmont 1:fdd22bb7aa52 190 }
emilmont 1:fdd22bb7aa52 191
emilmont 1:fdd22bb7aa52 192 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 193 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 194 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 195
emilmont 1:fdd22bb7aa52 196 while(k > 0u)
emilmont 1:fdd22bb7aa52 197 {
emilmont 1:fdd22bb7aa52 198 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 199 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 200
emilmont 1:fdd22bb7aa52 201 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 202 k--;
emilmont 1:fdd22bb7aa52 203 }
emilmont 1:fdd22bb7aa52 204
emilmont 1:fdd22bb7aa52 205 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 206 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 207
emilmont 1:fdd22bb7aa52 208 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 209 py = pIn2 + count;
emilmont 1:fdd22bb7aa52 210 px = pIn1;
emilmont 1:fdd22bb7aa52 211
emilmont 1:fdd22bb7aa52 212 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 213 count++;
emilmont 1:fdd22bb7aa52 214
emilmont 1:fdd22bb7aa52 215 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 216 blockSize1--;
emilmont 1:fdd22bb7aa52 217 }
emilmont 1:fdd22bb7aa52 218
emilmont 1:fdd22bb7aa52 219 /* --------------------------
emilmont 1:fdd22bb7aa52 220 * Initializations of stage2
emilmont 1:fdd22bb7aa52 221 * ------------------------*/
emilmont 1:fdd22bb7aa52 222
emilmont 1:fdd22bb7aa52 223 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 224 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 225 * ....
emilmont 1:fdd22bb7aa52 226 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 227 */
emilmont 1:fdd22bb7aa52 228
emilmont 1:fdd22bb7aa52 229 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 230 px = pIn1;
emilmont 1:fdd22bb7aa52 231
emilmont 1:fdd22bb7aa52 232 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 233 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 234 py = pSrc2;
emilmont 1:fdd22bb7aa52 235
emilmont 1:fdd22bb7aa52 236 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 237 count = 0u;
emilmont 1:fdd22bb7aa52 238
emilmont 1:fdd22bb7aa52 239 /* -------------------
emilmont 1:fdd22bb7aa52 240 * Stage2 process
emilmont 1:fdd22bb7aa52 241 * ------------------*/
emilmont 1:fdd22bb7aa52 242
emilmont 1:fdd22bb7aa52 243 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 244 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 245 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 246 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 247 {
emilmont 1:fdd22bb7aa52 248 /* Loop unroll by 3 */
emilmont 1:fdd22bb7aa52 249 blkCnt = blockSize2 / 3;
emilmont 1:fdd22bb7aa52 250
emilmont 1:fdd22bb7aa52 251 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 252 {
emilmont 1:fdd22bb7aa52 253 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 254 acc0 = 0;
emilmont 1:fdd22bb7aa52 255 acc1 = 0;
emilmont 1:fdd22bb7aa52 256 acc2 = 0;
emilmont 1:fdd22bb7aa52 257
emilmont 1:fdd22bb7aa52 258 /* read x[0], x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 259 x0 = *(px++);
emilmont 1:fdd22bb7aa52 260 x1 = *(px++);
emilmont 1:fdd22bb7aa52 261
emilmont 1:fdd22bb7aa52 262 /* Apply loop unrolling and compute 3 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 263 k = srcBLen / 3;
emilmont 1:fdd22bb7aa52 264
emilmont 1:fdd22bb7aa52 265 /* First part of the processing with loop unrolling. Compute 3 MACs at a time.
emilmont 1:fdd22bb7aa52 266 ** a second loop below computes MACs for the remaining 1 to 2 samples. */
emilmont 1:fdd22bb7aa52 267 do
emilmont 1:fdd22bb7aa52 268 {
emilmont 1:fdd22bb7aa52 269 /* Read y[srcBLen - 1] sample */
emilmont 1:fdd22bb7aa52 270 c0 = *(py);
emilmont 1:fdd22bb7aa52 271
emilmont 1:fdd22bb7aa52 272 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 273 x2 = *(px);
emilmont 1:fdd22bb7aa52 274
emilmont 1:fdd22bb7aa52 275 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 276 /* acc0 += x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 277 acc0 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 278 /* acc1 += x[1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 279 acc1 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 280 /* acc2 += x[2] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 281 acc2 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 282
emilmont 1:fdd22bb7aa52 283 /* Read y[srcBLen - 2] sample */
emilmont 1:fdd22bb7aa52 284 c0 = *(py - 1u);
emilmont 1:fdd22bb7aa52 285
emilmont 1:fdd22bb7aa52 286 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 287 x0 = *(px + 1u);
emilmont 1:fdd22bb7aa52 288
emilmont 1:fdd22bb7aa52 289 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 290 /* acc0 += x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 291 acc0 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 292 /* acc1 += x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 293 acc1 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 294 /* acc2 += x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 295 acc2 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 296
emilmont 1:fdd22bb7aa52 297 /* Read y[srcBLen - 3] sample */
emilmont 1:fdd22bb7aa52 298 c0 = *(py - 2u);
emilmont 1:fdd22bb7aa52 299
emilmont 1:fdd22bb7aa52 300 /* Read x[5] sample */
emilmont 1:fdd22bb7aa52 301 x1 = *(px + 2u);
emilmont 1:fdd22bb7aa52 302
emilmont 1:fdd22bb7aa52 303 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 304 /* acc0 += x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 305 acc0 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 306 /* acc1 += x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 307 acc1 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 308 /* acc2 += x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 309 acc2 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 310
emilmont 1:fdd22bb7aa52 311 /* update scratch pointers */
emilmont 1:fdd22bb7aa52 312 px += 3u;
emilmont 1:fdd22bb7aa52 313 py -= 3u;
emilmont 1:fdd22bb7aa52 314
emilmont 1:fdd22bb7aa52 315 } while(--k);
emilmont 1:fdd22bb7aa52 316
emilmont 1:fdd22bb7aa52 317 /* If the srcBLen is not a multiple of 3, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 318 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 319 k = srcBLen - (3 * (srcBLen / 3));
emilmont 1:fdd22bb7aa52 320
emilmont 1:fdd22bb7aa52 321 while(k > 0u)
emilmont 1:fdd22bb7aa52 322 {
emilmont 1:fdd22bb7aa52 323 /* Read y[srcBLen - 5] sample */
emilmont 1:fdd22bb7aa52 324 c0 = *(py--);
emilmont 1:fdd22bb7aa52 325
emilmont 1:fdd22bb7aa52 326 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 327 x2 = *(px++);
emilmont 1:fdd22bb7aa52 328
emilmont 1:fdd22bb7aa52 329 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 330 /* acc0 += x[4] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 331 acc0 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 332 /* acc1 += x[5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 333 acc1 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 334 /* acc2 += x[6] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 335 acc2 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 336
emilmont 1:fdd22bb7aa52 337 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 338 x0 = x1;
emilmont 1:fdd22bb7aa52 339 x1 = x2;
emilmont 1:fdd22bb7aa52 340
emilmont 1:fdd22bb7aa52 341 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 342 k--;
emilmont 1:fdd22bb7aa52 343 }
emilmont 1:fdd22bb7aa52 344
emilmont 1:fdd22bb7aa52 345 /* Store the results in the accumulators in the destination buffer. */
emilmont 1:fdd22bb7aa52 346 *pOut++ = (q31_t) (acc0 >> 31);
emilmont 1:fdd22bb7aa52 347 *pOut++ = (q31_t) (acc1 >> 31);
emilmont 1:fdd22bb7aa52 348 *pOut++ = (q31_t) (acc2 >> 31);
emilmont 1:fdd22bb7aa52 349
emilmont 1:fdd22bb7aa52 350 /* Increment the pointer pIn1 index, count by 3 */
emilmont 1:fdd22bb7aa52 351 count += 3u;
emilmont 1:fdd22bb7aa52 352
emilmont 1:fdd22bb7aa52 353 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 354 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 355 py = pSrc2;
emilmont 1:fdd22bb7aa52 356
emilmont 1:fdd22bb7aa52 357 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 358 blkCnt--;
emilmont 1:fdd22bb7aa52 359 }
emilmont 1:fdd22bb7aa52 360
emilmont 1:fdd22bb7aa52 361 /* If the blockSize2 is not a multiple of 3, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 362 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 363 blkCnt = blockSize2 - 3 * (blockSize2 / 3);
emilmont 1:fdd22bb7aa52 364
emilmont 1:fdd22bb7aa52 365 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 366 {
emilmont 1:fdd22bb7aa52 367 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 368 sum = 0;
emilmont 1:fdd22bb7aa52 369
emilmont 1:fdd22bb7aa52 370 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 371 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 372
emilmont 1:fdd22bb7aa52 373 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 374 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 375 while(k > 0u)
emilmont 1:fdd22bb7aa52 376 {
emilmont 1:fdd22bb7aa52 377 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 378 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 379 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 380 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 381 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 382
emilmont 1:fdd22bb7aa52 383 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 384 k--;
emilmont 1:fdd22bb7aa52 385 }
emilmont 1:fdd22bb7aa52 386
emilmont 1:fdd22bb7aa52 387 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 388 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 389 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 390
emilmont 1:fdd22bb7aa52 391 while(k > 0u)
emilmont 1:fdd22bb7aa52 392 {
emilmont 1:fdd22bb7aa52 393 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 394 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 395
emilmont 1:fdd22bb7aa52 396 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 397 k--;
emilmont 1:fdd22bb7aa52 398 }
emilmont 1:fdd22bb7aa52 399
emilmont 1:fdd22bb7aa52 400 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 401 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 402
emilmont 1:fdd22bb7aa52 403 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 404 count++;
emilmont 1:fdd22bb7aa52 405
emilmont 1:fdd22bb7aa52 406 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 407 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 408 py = pSrc2;
emilmont 1:fdd22bb7aa52 409
emilmont 1:fdd22bb7aa52 410 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 411 blkCnt--;
emilmont 1:fdd22bb7aa52 412 }
emilmont 1:fdd22bb7aa52 413 }
emilmont 1:fdd22bb7aa52 414 else
emilmont 1:fdd22bb7aa52 415 {
emilmont 1:fdd22bb7aa52 416 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 417 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 418 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 419
emilmont 1:fdd22bb7aa52 420 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 421 {
emilmont 1:fdd22bb7aa52 422 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 423 sum = 0;
emilmont 1:fdd22bb7aa52 424
emilmont 1:fdd22bb7aa52 425 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 426 k = srcBLen;
emilmont 1:fdd22bb7aa52 427
emilmont 1:fdd22bb7aa52 428 while(k > 0u)
emilmont 1:fdd22bb7aa52 429 {
emilmont 1:fdd22bb7aa52 430 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 431 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 432
emilmont 1:fdd22bb7aa52 433 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 434 k--;
emilmont 1:fdd22bb7aa52 435 }
emilmont 1:fdd22bb7aa52 436
emilmont 1:fdd22bb7aa52 437 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 438 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 439
emilmont 1:fdd22bb7aa52 440 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 441 count++;
emilmont 1:fdd22bb7aa52 442
emilmont 1:fdd22bb7aa52 443 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 444 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 445 py = pSrc2;
emilmont 1:fdd22bb7aa52 446
emilmont 1:fdd22bb7aa52 447 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 448 blkCnt--;
emilmont 1:fdd22bb7aa52 449 }
emilmont 1:fdd22bb7aa52 450 }
emilmont 1:fdd22bb7aa52 451
emilmont 1:fdd22bb7aa52 452
emilmont 1:fdd22bb7aa52 453 /* --------------------------
emilmont 1:fdd22bb7aa52 454 * Initializations of stage3
emilmont 1:fdd22bb7aa52 455 * -------------------------*/
emilmont 1:fdd22bb7aa52 456
emilmont 1:fdd22bb7aa52 457 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 458 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 459 * ....
emilmont 1:fdd22bb7aa52 460 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 461 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 462 */
emilmont 1:fdd22bb7aa52 463
emilmont 1:fdd22bb7aa52 464 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 465 The blockSize3 variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 466
emilmont 1:fdd22bb7aa52 467 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 468 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 469 px = pSrc1;
emilmont 1:fdd22bb7aa52 470
emilmont 1:fdd22bb7aa52 471 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 472 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 473 py = pSrc2;
emilmont 1:fdd22bb7aa52 474
emilmont 1:fdd22bb7aa52 475 /* -------------------
emilmont 1:fdd22bb7aa52 476 * Stage3 process
emilmont 1:fdd22bb7aa52 477 * ------------------*/
emilmont 1:fdd22bb7aa52 478
emilmont 1:fdd22bb7aa52 479 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 480 {
emilmont 1:fdd22bb7aa52 481 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 482 sum = 0;
emilmont 1:fdd22bb7aa52 483
emilmont 1:fdd22bb7aa52 484 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 485 k = blockSize3 >> 2u;
emilmont 1:fdd22bb7aa52 486
emilmont 1:fdd22bb7aa52 487 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 488 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 489 while(k > 0u)
emilmont 1:fdd22bb7aa52 490 {
emilmont 1:fdd22bb7aa52 491 /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 492 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 493 /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 494 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 495 /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 496 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 497 /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 498 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 499
emilmont 1:fdd22bb7aa52 500 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 501 k--;
emilmont 1:fdd22bb7aa52 502 }
emilmont 1:fdd22bb7aa52 503
emilmont 1:fdd22bb7aa52 504 /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 505 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 506 k = blockSize3 % 0x4u;
emilmont 1:fdd22bb7aa52 507
emilmont 1:fdd22bb7aa52 508 while(k > 0u)
emilmont 1:fdd22bb7aa52 509 {
emilmont 1:fdd22bb7aa52 510 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 511 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 512
emilmont 1:fdd22bb7aa52 513 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 514 k--;
emilmont 1:fdd22bb7aa52 515 }
emilmont 1:fdd22bb7aa52 516
emilmont 1:fdd22bb7aa52 517 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 518 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 519
emilmont 1:fdd22bb7aa52 520 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 521 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 522 py = pSrc2;
emilmont 1:fdd22bb7aa52 523
emilmont 1:fdd22bb7aa52 524 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 525 blockSize3--;
emilmont 1:fdd22bb7aa52 526 }
emilmont 1:fdd22bb7aa52 527
emilmont 1:fdd22bb7aa52 528 #else
emilmont 1:fdd22bb7aa52 529
emilmont 1:fdd22bb7aa52 530 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 q31_t *pIn1 = pSrcA; /* input pointer */
emilmont 1:fdd22bb7aa52 533 q31_t *pIn2 = pSrcB; /* coefficient pointer */
emilmont 1:fdd22bb7aa52 534 q63_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 535 uint32_t i, j; /* loop counter */
emilmont 1:fdd22bb7aa52 536
emilmont 1:fdd22bb7aa52 537 /* Loop to calculate output of convolution for output length number of times */
emilmont 1:fdd22bb7aa52 538 for (i = 0; i < (srcALen + srcBLen - 1); i++)
emilmont 1:fdd22bb7aa52 539 {
emilmont 1:fdd22bb7aa52 540 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 541 sum = 0;
emilmont 1:fdd22bb7aa52 542
emilmont 1:fdd22bb7aa52 543 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 544 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 545 {
emilmont 1:fdd22bb7aa52 546 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 547 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 548 {
emilmont 1:fdd22bb7aa52 549 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 550 sum += ((q63_t) pIn1[j] * (pIn2[i - j]));
emilmont 1:fdd22bb7aa52 551 }
emilmont 1:fdd22bb7aa52 552 }
emilmont 1:fdd22bb7aa52 553
emilmont 1:fdd22bb7aa52 554 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 555 pDst[i] = (q31_t) (sum >> 31u);
emilmont 1:fdd22bb7aa52 556 }
emilmont 1:fdd22bb7aa52 557
emilmont 1:fdd22bb7aa52 558 #endif /* #ifndef ARM_MATH_CM0 */
emilmont 1:fdd22bb7aa52 559
emilmont 1:fdd22bb7aa52 560 }
emilmont 1:fdd22bb7aa52 561
emilmont 1:fdd22bb7aa52 562 /**
emilmont 1:fdd22bb7aa52 563 * @} end of Conv group
emilmont 1:fdd22bb7aa52 564 */