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_q7.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_q7.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 1:fdd22bb7aa52 10 * Description: Convolution of Q7 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 Q7 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 a 32-bit internal accumulator.
emilmont 1:fdd22bb7aa52 65 * Both the inputs are represented in 1.7 format and multiplications yield a 2.14 result.
emilmont 1:fdd22bb7aa52 66 * The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format.
emilmont 1:fdd22bb7aa52 67 * This approach provides 17 guard bits and there is no risk of overflow as long as <code>max(srcALen, srcBLen)<131072</code>.
emilmont 1:fdd22bb7aa52 68 * The 18.14 result is then truncated to 18.7 format by discarding the low 7 bits and then saturated to 1.7 format.
emilmont 1:fdd22bb7aa52 69 *
emilmont 1:fdd22bb7aa52 70 * \par
emilmont 1:fdd22bb7aa52 71 * Refer the function <code>arm_conv_opt_q7()</code> for a faster implementation of this function.
emilmont 1:fdd22bb7aa52 72 *
emilmont 1:fdd22bb7aa52 73 */
emilmont 1:fdd22bb7aa52 74
emilmont 1:fdd22bb7aa52 75 void arm_conv_q7(
emilmont 1:fdd22bb7aa52 76 q7_t * pSrcA,
emilmont 1:fdd22bb7aa52 77 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 78 q7_t * pSrcB,
emilmont 1:fdd22bb7aa52 79 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 80 q7_t * pDst)
emilmont 1:fdd22bb7aa52 81 {
emilmont 1:fdd22bb7aa52 82
emilmont 1:fdd22bb7aa52 83
emilmont 1:fdd22bb7aa52 84 #ifndef ARM_MATH_CM0
emilmont 1:fdd22bb7aa52 85
emilmont 1:fdd22bb7aa52 86 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 87
emilmont 1:fdd22bb7aa52 88 q7_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 89 q7_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 90 q7_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 91 q7_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 92 q7_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 93 q7_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 94 q7_t x0, x1, x2, x3, c0, c1; /* Temporary variables to hold state and coefficient values */
emilmont 1:fdd22bb7aa52 95 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 96 q31_t input1, input2; /* Temporary input variables */
emilmont 1:fdd22bb7aa52 97 q15_t in1, in2; /* Temporary input variables */
emilmont 1:fdd22bb7aa52 98 uint32_t j, k, count, blkCnt, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 99
emilmont 1:fdd22bb7aa52 100 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 101 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 102 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 103 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 104 {
emilmont 1:fdd22bb7aa52 105 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 106 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 107
emilmont 1:fdd22bb7aa52 108 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 109 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 110 }
emilmont 1:fdd22bb7aa52 111 else
emilmont 1:fdd22bb7aa52 112 {
emilmont 1:fdd22bb7aa52 113 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 114 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 115
emilmont 1:fdd22bb7aa52 116 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 117 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 118
emilmont 1:fdd22bb7aa52 119 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 120 j = srcBLen;
emilmont 1:fdd22bb7aa52 121 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 122 srcALen = j;
emilmont 1:fdd22bb7aa52 123 }
emilmont 1:fdd22bb7aa52 124
emilmont 1:fdd22bb7aa52 125 /* 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 126 /* The function is internally
emilmont 1:fdd22bb7aa52 127 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 128 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 129 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 130 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 131 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 132 * for every iteration. */
emilmont 1:fdd22bb7aa52 133
emilmont 1:fdd22bb7aa52 134 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 135 The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 136 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 137 blockSize2 = (srcALen - srcBLen) + 1u;
emilmont 1:fdd22bb7aa52 138 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 139
emilmont 1:fdd22bb7aa52 140 /* --------------------------
emilmont 1:fdd22bb7aa52 141 * Initializations of stage1
emilmont 1:fdd22bb7aa52 142 * -------------------------*/
emilmont 1:fdd22bb7aa52 143
emilmont 1:fdd22bb7aa52 144 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 145 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 146 * ....
emilmont 1:fdd22bb7aa52 147 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 148 */
emilmont 1:fdd22bb7aa52 149
emilmont 1:fdd22bb7aa52 150 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 151 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 152 count = 1u;
emilmont 1:fdd22bb7aa52 153
emilmont 1:fdd22bb7aa52 154 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 155 px = pIn1;
emilmont 1:fdd22bb7aa52 156
emilmont 1:fdd22bb7aa52 157 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 158 py = pIn2;
emilmont 1:fdd22bb7aa52 159
emilmont 1:fdd22bb7aa52 160
emilmont 1:fdd22bb7aa52 161 /* ------------------------
emilmont 1:fdd22bb7aa52 162 * Stage1 process
emilmont 1:fdd22bb7aa52 163 * ----------------------*/
emilmont 1:fdd22bb7aa52 164
emilmont 1:fdd22bb7aa52 165 /* The first stage starts here */
emilmont 1:fdd22bb7aa52 166 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 167 {
emilmont 1:fdd22bb7aa52 168 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 169 sum = 0;
emilmont 1:fdd22bb7aa52 170
emilmont 1:fdd22bb7aa52 171 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 172 k = count >> 2u;
emilmont 1:fdd22bb7aa52 173
emilmont 1:fdd22bb7aa52 174 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 175 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 176 while(k > 0u)
emilmont 1:fdd22bb7aa52 177 {
emilmont 1:fdd22bb7aa52 178 /* x[0] , x[1] */
emilmont 1:fdd22bb7aa52 179 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 180 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 181 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 182
emilmont 1:fdd22bb7aa52 183 /* y[srcBLen - 1] , y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 184 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 185 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 186 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 187
emilmont 1:fdd22bb7aa52 188 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 189 /* x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 190 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 191
emilmont 1:fdd22bb7aa52 192 /* x[2] , x[3] */
emilmont 1:fdd22bb7aa52 193 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 194 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 195 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 196
emilmont 1:fdd22bb7aa52 197 /* y[srcBLen - 3] , y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 198 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 199 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 200 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 201
emilmont 1:fdd22bb7aa52 202 /* x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 203 /* x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 204 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 207 k--;
emilmont 1:fdd22bb7aa52 208 }
emilmont 1:fdd22bb7aa52 209
emilmont 1:fdd22bb7aa52 210 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 211 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 212 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 213
emilmont 1:fdd22bb7aa52 214 while(k > 0u)
emilmont 1:fdd22bb7aa52 215 {
emilmont 1:fdd22bb7aa52 216 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 217 sum += ((q15_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 218
emilmont 1:fdd22bb7aa52 219 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 220 k--;
emilmont 1:fdd22bb7aa52 221 }
emilmont 1:fdd22bb7aa52 222
emilmont 1:fdd22bb7aa52 223 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 224 *pOut++ = (q7_t) (__SSAT(sum >> 7u, 8));
emilmont 1:fdd22bb7aa52 225
emilmont 1:fdd22bb7aa52 226 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 227 py = pIn2 + count;
emilmont 1:fdd22bb7aa52 228 px = pIn1;
emilmont 1:fdd22bb7aa52 229
emilmont 1:fdd22bb7aa52 230 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 231 count++;
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 234 blockSize1--;
emilmont 1:fdd22bb7aa52 235 }
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 /* --------------------------
emilmont 1:fdd22bb7aa52 238 * Initializations of stage2
emilmont 1:fdd22bb7aa52 239 * ------------------------*/
emilmont 1:fdd22bb7aa52 240
emilmont 1:fdd22bb7aa52 241 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 242 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 243 * ....
emilmont 1:fdd22bb7aa52 244 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 245 */
emilmont 1:fdd22bb7aa52 246
emilmont 1:fdd22bb7aa52 247 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 248 px = pIn1;
emilmont 1:fdd22bb7aa52 249
emilmont 1:fdd22bb7aa52 250 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 251 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 252 py = pSrc2;
emilmont 1:fdd22bb7aa52 253
emilmont 1:fdd22bb7aa52 254 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 255 count = 0u;
emilmont 1:fdd22bb7aa52 256
emilmont 1:fdd22bb7aa52 257 /* -------------------
emilmont 1:fdd22bb7aa52 258 * Stage2 process
emilmont 1:fdd22bb7aa52 259 * ------------------*/
emilmont 1:fdd22bb7aa52 260
emilmont 1:fdd22bb7aa52 261 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 262 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 263 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 264 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 265 {
emilmont 1:fdd22bb7aa52 266 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 267 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 268
emilmont 1:fdd22bb7aa52 269 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 270 {
emilmont 1:fdd22bb7aa52 271 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 272 acc0 = 0;
emilmont 1:fdd22bb7aa52 273 acc1 = 0;
emilmont 1:fdd22bb7aa52 274 acc2 = 0;
emilmont 1:fdd22bb7aa52 275 acc3 = 0;
emilmont 1:fdd22bb7aa52 276
emilmont 1:fdd22bb7aa52 277 /* read x[0], x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 278 x0 = *(px++);
emilmont 1:fdd22bb7aa52 279 x1 = *(px++);
emilmont 1:fdd22bb7aa52 280 x2 = *(px++);
emilmont 1:fdd22bb7aa52 281
emilmont 1:fdd22bb7aa52 282 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 283 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 284
emilmont 1:fdd22bb7aa52 285 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 286 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 287 do
emilmont 1:fdd22bb7aa52 288 {
emilmont 1:fdd22bb7aa52 289 /* Read y[srcBLen - 1] sample */
emilmont 1:fdd22bb7aa52 290 c0 = *(py--);
emilmont 1:fdd22bb7aa52 291 /* Read y[srcBLen - 2] sample */
emilmont 1:fdd22bb7aa52 292 c1 = *(py--);
emilmont 1:fdd22bb7aa52 293
emilmont 1:fdd22bb7aa52 294 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 295 x3 = *(px++);
emilmont 1:fdd22bb7aa52 296
emilmont 1:fdd22bb7aa52 297 /* x[0] and x[1] are packed */
emilmont 1:fdd22bb7aa52 298 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 299 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 300
emilmont 1:fdd22bb7aa52 301 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 302
emilmont 1:fdd22bb7aa52 303 /* y[srcBLen - 1] and y[srcBLen - 2] are packed */
emilmont 1:fdd22bb7aa52 304 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 305 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 306
emilmont 1:fdd22bb7aa52 307 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 308
emilmont 1:fdd22bb7aa52 309 /* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 310 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 /* x[1] and x[2] are packed */
emilmont 1:fdd22bb7aa52 313 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 314 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 315
emilmont 1:fdd22bb7aa52 316 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 317
emilmont 1:fdd22bb7aa52 318 /* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 319 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 320
emilmont 1:fdd22bb7aa52 321 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 322 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 323 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 324
emilmont 1:fdd22bb7aa52 325 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 326
emilmont 1:fdd22bb7aa52 327 /* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 328 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 329
emilmont 1:fdd22bb7aa52 330 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 331 x0 = *(px++);
emilmont 1:fdd22bb7aa52 332
emilmont 1:fdd22bb7aa52 333 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 334 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 335 in2 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 336
emilmont 1:fdd22bb7aa52 337 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 338
emilmont 1:fdd22bb7aa52 339 /* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 340 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 /* Read y[srcBLen - 3] sample */
emilmont 1:fdd22bb7aa52 343 c0 = *(py--);
emilmont 1:fdd22bb7aa52 344 /* Read y[srcBLen - 4] sample */
emilmont 1:fdd22bb7aa52 345 c1 = *(py--);
emilmont 1:fdd22bb7aa52 346
emilmont 1:fdd22bb7aa52 347 /* Read x[5] sample */
emilmont 1:fdd22bb7aa52 348 x1 = *(px++);
emilmont 1:fdd22bb7aa52 349
emilmont 1:fdd22bb7aa52 350 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 351 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 352 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 353
emilmont 1:fdd22bb7aa52 354 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 355
emilmont 1:fdd22bb7aa52 356 /* y[srcBLen - 3] and y[srcBLen - 4] are packed */
emilmont 1:fdd22bb7aa52 357 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 358 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 359
emilmont 1:fdd22bb7aa52 360 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 361
emilmont 1:fdd22bb7aa52 362 /* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 363 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 364
emilmont 1:fdd22bb7aa52 365 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 366 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 367 in2 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 368
emilmont 1:fdd22bb7aa52 369 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 370
emilmont 1:fdd22bb7aa52 371 /* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 372 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 373
emilmont 1:fdd22bb7aa52 374 /* x[4] and x[5] are packed */
emilmont 1:fdd22bb7aa52 375 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 376 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 377
emilmont 1:fdd22bb7aa52 378 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 379
emilmont 1:fdd22bb7aa52 380 /* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 381 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 382
emilmont 1:fdd22bb7aa52 383 /* Read x[6] sample */
emilmont 1:fdd22bb7aa52 384 x2 = *(px++);
emilmont 1:fdd22bb7aa52 385
emilmont 1:fdd22bb7aa52 386 /* x[5] and x[6] are packed */
emilmont 1:fdd22bb7aa52 387 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 388 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 389
emilmont 1:fdd22bb7aa52 390 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 391
emilmont 1:fdd22bb7aa52 392 /* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 393 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 394
emilmont 1:fdd22bb7aa52 395 } while(--k);
emilmont 1:fdd22bb7aa52 396
emilmont 1:fdd22bb7aa52 397 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 398 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 399 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 400
emilmont 1:fdd22bb7aa52 401 while(k > 0u)
emilmont 1:fdd22bb7aa52 402 {
emilmont 1:fdd22bb7aa52 403 /* Read y[srcBLen - 5] sample */
emilmont 1:fdd22bb7aa52 404 c0 = *(py--);
emilmont 1:fdd22bb7aa52 405
emilmont 1:fdd22bb7aa52 406 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 407 x3 = *(px++);
emilmont 1:fdd22bb7aa52 408
emilmont 1:fdd22bb7aa52 409 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 410 /* acc0 += x[4] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 411 acc0 += ((q15_t) x0 * c0);
emilmont 1:fdd22bb7aa52 412 /* acc1 += x[5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 413 acc1 += ((q15_t) x1 * c0);
emilmont 1:fdd22bb7aa52 414 /* acc2 += x[6] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 415 acc2 += ((q15_t) x2 * c0);
emilmont 1:fdd22bb7aa52 416 /* acc3 += x[7] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 417 acc3 += ((q15_t) x3 * c0);
emilmont 1:fdd22bb7aa52 418
emilmont 1:fdd22bb7aa52 419 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 420 x0 = x1;
emilmont 1:fdd22bb7aa52 421 x1 = x2;
emilmont 1:fdd22bb7aa52 422 x2 = x3;
emilmont 1:fdd22bb7aa52 423
emilmont 1:fdd22bb7aa52 424 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 425 k--;
emilmont 1:fdd22bb7aa52 426 }
emilmont 1:fdd22bb7aa52 427
emilmont 1:fdd22bb7aa52 428
emilmont 1:fdd22bb7aa52 429 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 430 *pOut++ = (q7_t) (__SSAT(acc0 >> 7u, 8));
emilmont 1:fdd22bb7aa52 431 *pOut++ = (q7_t) (__SSAT(acc1 >> 7u, 8));
emilmont 1:fdd22bb7aa52 432 *pOut++ = (q7_t) (__SSAT(acc2 >> 7u, 8));
emilmont 1:fdd22bb7aa52 433 *pOut++ = (q7_t) (__SSAT(acc3 >> 7u, 8));
emilmont 1:fdd22bb7aa52 434
emilmont 1:fdd22bb7aa52 435 /* Increment the pointer pIn1 index, count by 4 */
emilmont 1:fdd22bb7aa52 436 count += 4u;
emilmont 1:fdd22bb7aa52 437
emilmont 1:fdd22bb7aa52 438 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 439 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 440 py = pSrc2;
emilmont 1:fdd22bb7aa52 441
emilmont 1:fdd22bb7aa52 442 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 443 blkCnt--;
emilmont 1:fdd22bb7aa52 444 }
emilmont 1:fdd22bb7aa52 445
emilmont 1:fdd22bb7aa52 446 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 447 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 448 blkCnt = blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 449
emilmont 1:fdd22bb7aa52 450 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 451 {
emilmont 1:fdd22bb7aa52 452 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 453 sum = 0;
emilmont 1:fdd22bb7aa52 454
emilmont 1:fdd22bb7aa52 455 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 456 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 457
emilmont 1:fdd22bb7aa52 458 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 459 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 460 while(k > 0u)
emilmont 1:fdd22bb7aa52 461 {
emilmont 1:fdd22bb7aa52 462
emilmont 1:fdd22bb7aa52 463 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 464 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 465 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 466 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 467
emilmont 1:fdd22bb7aa52 468 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 469 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 470 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 471 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 472
emilmont 1:fdd22bb7aa52 473 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 474 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 475
emilmont 1:fdd22bb7aa52 476 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 477 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 478 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 479 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 480
emilmont 1:fdd22bb7aa52 481 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 482 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 483 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 484 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 485
emilmont 1:fdd22bb7aa52 486 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 487 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 488
emilmont 1:fdd22bb7aa52 489 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 490 k--;
emilmont 1:fdd22bb7aa52 491 }
emilmont 1:fdd22bb7aa52 492
emilmont 1:fdd22bb7aa52 493 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 494 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 495 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 496
emilmont 1:fdd22bb7aa52 497 while(k > 0u)
emilmont 1:fdd22bb7aa52 498 {
emilmont 1:fdd22bb7aa52 499 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 500 sum += ((q15_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 501
emilmont 1:fdd22bb7aa52 502 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 503 k--;
emilmont 1:fdd22bb7aa52 504 }
emilmont 1:fdd22bb7aa52 505
emilmont 1:fdd22bb7aa52 506 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 507 *pOut++ = (q7_t) (__SSAT(sum >> 7u, 8));
emilmont 1:fdd22bb7aa52 508
emilmont 1:fdd22bb7aa52 509 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 510 count++;
emilmont 1:fdd22bb7aa52 511
emilmont 1:fdd22bb7aa52 512 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 513 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 514 py = pSrc2;
emilmont 1:fdd22bb7aa52 515
emilmont 1:fdd22bb7aa52 516 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 517 blkCnt--;
emilmont 1:fdd22bb7aa52 518 }
emilmont 1:fdd22bb7aa52 519 }
emilmont 1:fdd22bb7aa52 520 else
emilmont 1:fdd22bb7aa52 521 {
emilmont 1:fdd22bb7aa52 522 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 523 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 524 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 525
emilmont 1:fdd22bb7aa52 526 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 527 {
emilmont 1:fdd22bb7aa52 528 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 529 sum = 0;
emilmont 1:fdd22bb7aa52 530
emilmont 1:fdd22bb7aa52 531 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 532 k = srcBLen;
emilmont 1:fdd22bb7aa52 533
emilmont 1:fdd22bb7aa52 534 while(k > 0u)
emilmont 1:fdd22bb7aa52 535 {
emilmont 1:fdd22bb7aa52 536 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 537 sum += ((q15_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 538
emilmont 1:fdd22bb7aa52 539 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 540 k--;
emilmont 1:fdd22bb7aa52 541 }
emilmont 1:fdd22bb7aa52 542
emilmont 1:fdd22bb7aa52 543 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 544 *pOut++ = (q7_t) (__SSAT(sum >> 7u, 8));
emilmont 1:fdd22bb7aa52 545
emilmont 1:fdd22bb7aa52 546 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 547 count++;
emilmont 1:fdd22bb7aa52 548
emilmont 1:fdd22bb7aa52 549 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 550 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 551 py = pSrc2;
emilmont 1:fdd22bb7aa52 552
emilmont 1:fdd22bb7aa52 553 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 554 blkCnt--;
emilmont 1:fdd22bb7aa52 555 }
emilmont 1:fdd22bb7aa52 556 }
emilmont 1:fdd22bb7aa52 557
emilmont 1:fdd22bb7aa52 558
emilmont 1:fdd22bb7aa52 559 /* --------------------------
emilmont 1:fdd22bb7aa52 560 * Initializations of stage3
emilmont 1:fdd22bb7aa52 561 * -------------------------*/
emilmont 1:fdd22bb7aa52 562
emilmont 1:fdd22bb7aa52 563 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 564 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 565 * ....
emilmont 1:fdd22bb7aa52 566 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 567 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 568 */
emilmont 1:fdd22bb7aa52 569
emilmont 1:fdd22bb7aa52 570 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 571 The blockSize3 variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 572
emilmont 1:fdd22bb7aa52 573 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 574 pSrc1 = pIn1 + (srcALen - (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 575 px = pSrc1;
emilmont 1:fdd22bb7aa52 576
emilmont 1:fdd22bb7aa52 577 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 578 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 579 py = pSrc2;
emilmont 1:fdd22bb7aa52 580
emilmont 1:fdd22bb7aa52 581 /* -------------------
emilmont 1:fdd22bb7aa52 582 * Stage3 process
emilmont 1:fdd22bb7aa52 583 * ------------------*/
emilmont 1:fdd22bb7aa52 584
emilmont 1:fdd22bb7aa52 585 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 586 {
emilmont 1:fdd22bb7aa52 587 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 588 sum = 0;
emilmont 1:fdd22bb7aa52 589
emilmont 1:fdd22bb7aa52 590 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 591 k = blockSize3 >> 2u;
emilmont 1:fdd22bb7aa52 592
emilmont 1:fdd22bb7aa52 593 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 594 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 595 while(k > 0u)
emilmont 1:fdd22bb7aa52 596 {
emilmont 1:fdd22bb7aa52 597 /* Reading two inputs, x[srcALen - srcBLen + 1] and x[srcALen - srcBLen + 2] of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 598 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 599 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 600 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 601
emilmont 1:fdd22bb7aa52 602 /* Reading two inputs, y[srcBLen - 1] and y[srcBLen - 2] of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 603 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 604 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 605 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 606
emilmont 1:fdd22bb7aa52 607 /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 608 /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 609 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 610
emilmont 1:fdd22bb7aa52 611 /* Reading two inputs, x[srcALen - srcBLen + 3] and x[srcALen - srcBLen + 4] of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 612 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 613 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 614 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 615
emilmont 1:fdd22bb7aa52 616 /* Reading two inputs, y[srcBLen - 3] and y[srcBLen - 4] of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 617 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 618 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 619 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16u);
emilmont 1:fdd22bb7aa52 620
emilmont 1:fdd22bb7aa52 621 /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 622 /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 623 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 624
emilmont 1:fdd22bb7aa52 625 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 626 k--;
emilmont 1:fdd22bb7aa52 627 }
emilmont 1:fdd22bb7aa52 628
emilmont 1:fdd22bb7aa52 629 /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 630 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 631 k = blockSize3 % 0x4u;
emilmont 1:fdd22bb7aa52 632
emilmont 1:fdd22bb7aa52 633 while(k > 0u)
emilmont 1:fdd22bb7aa52 634 {
emilmont 1:fdd22bb7aa52 635 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 636 sum += ((q15_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 637
emilmont 1:fdd22bb7aa52 638 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 639 k--;
emilmont 1:fdd22bb7aa52 640 }
emilmont 1:fdd22bb7aa52 641
emilmont 1:fdd22bb7aa52 642 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 643 *pOut++ = (q7_t) (__SSAT(sum >> 7u, 8));
emilmont 1:fdd22bb7aa52 644
emilmont 1:fdd22bb7aa52 645 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 646 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 647 py = pSrc2;
emilmont 1:fdd22bb7aa52 648
emilmont 1:fdd22bb7aa52 649 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 650 blockSize3--;
emilmont 1:fdd22bb7aa52 651 }
emilmont 1:fdd22bb7aa52 652
emilmont 1:fdd22bb7aa52 653 #else
emilmont 1:fdd22bb7aa52 654
emilmont 1:fdd22bb7aa52 655 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 656
emilmont 1:fdd22bb7aa52 657 q7_t *pIn1 = pSrcA; /* input pointer */
emilmont 1:fdd22bb7aa52 658 q7_t *pIn2 = pSrcB; /* coefficient pointer */
emilmont 1:fdd22bb7aa52 659 q31_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 660 uint32_t i, j; /* loop counter */
emilmont 1:fdd22bb7aa52 661
emilmont 1:fdd22bb7aa52 662 /* Loop to calculate output of convolution for output length number of times */
emilmont 1:fdd22bb7aa52 663 for (i = 0; i < (srcALen + srcBLen - 1); i++)
emilmont 1:fdd22bb7aa52 664 {
emilmont 1:fdd22bb7aa52 665 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 666 sum = 0;
emilmont 1:fdd22bb7aa52 667
emilmont 1:fdd22bb7aa52 668 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 669 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 670 {
emilmont 1:fdd22bb7aa52 671 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 672 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 673 {
emilmont 1:fdd22bb7aa52 674 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 675 sum += (q15_t) pIn1[j] * (pIn2[i - j]);
emilmont 1:fdd22bb7aa52 676 }
emilmont 1:fdd22bb7aa52 677 }
emilmont 1:fdd22bb7aa52 678
emilmont 1:fdd22bb7aa52 679 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 680 pDst[i] = (q7_t) __SSAT((sum >> 7u), 8u);
emilmont 1:fdd22bb7aa52 681 }
emilmont 1:fdd22bb7aa52 682
emilmont 1:fdd22bb7aa52 683 #endif /* #ifndef ARM_MATH_CM0 */
emilmont 1:fdd22bb7aa52 684
emilmont 1:fdd22bb7aa52 685 }
emilmont 1:fdd22bb7aa52 686
emilmont 1:fdd22bb7aa52 687 /**
emilmont 1:fdd22bb7aa52 688 * @} end of Conv group
emilmont 1:fdd22bb7aa52 689 */