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_partial_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_partial_q7.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 1:fdd22bb7aa52 10 * Description: Partial 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 PartialConv
emilmont 1:fdd22bb7aa52 48 * @{
emilmont 1:fdd22bb7aa52 49 */
emilmont 1:fdd22bb7aa52 50
emilmont 1:fdd22bb7aa52 51 /**
emilmont 1:fdd22bb7aa52 52 * @brief Partial 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.
emilmont 1:fdd22bb7aa52 58 * @param[in] firstIndex is the first output sample to start with.
emilmont 1:fdd22bb7aa52 59 * @param[in] numPoints is the number of output points to be computed.
emilmont 1:fdd22bb7aa52 60 * @return Returns either ARM_MATH_SUCCESS if the function completed correctly or ARM_MATH_ARGUMENT_ERROR if the requested subset is not in the range [0 srcALen+srcBLen-2].
emilmont 1:fdd22bb7aa52 61 *
emilmont 1:fdd22bb7aa52 62 * \par
emilmont 1:fdd22bb7aa52 63 * Refer the function <code>arm_conv_partial_opt_q7()</code> for a faster implementation of this function.
emilmont 1:fdd22bb7aa52 64 *
emilmont 1:fdd22bb7aa52 65 */
emilmont 1:fdd22bb7aa52 66
emilmont 1:fdd22bb7aa52 67 arm_status arm_conv_partial_q7(
emilmont 1:fdd22bb7aa52 68 q7_t * pSrcA,
emilmont 1:fdd22bb7aa52 69 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 70 q7_t * pSrcB,
emilmont 1:fdd22bb7aa52 71 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 72 q7_t * pDst,
emilmont 1:fdd22bb7aa52 73 uint32_t firstIndex,
emilmont 1:fdd22bb7aa52 74 uint32_t numPoints)
emilmont 1:fdd22bb7aa52 75 {
emilmont 1:fdd22bb7aa52 76
emilmont 1:fdd22bb7aa52 77
emilmont 1:fdd22bb7aa52 78 #ifndef ARM_MATH_CM0
emilmont 1:fdd22bb7aa52 79
emilmont 1:fdd22bb7aa52 80 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 81
emilmont 1:fdd22bb7aa52 82 q7_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 83 q7_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 84 q7_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 85 q7_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 86 q7_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 87 q7_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 88 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 89 q31_t input1, input2;
emilmont 1:fdd22bb7aa52 90 q15_t in1, in2;
emilmont 1:fdd22bb7aa52 91 q7_t x0, x1, x2, x3, c0, c1;
emilmont 1:fdd22bb7aa52 92 uint32_t j, k, count, check, blkCnt;
emilmont 1:fdd22bb7aa52 93 int32_t blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 94 arm_status status;
emilmont 1:fdd22bb7aa52 95
emilmont 1:fdd22bb7aa52 96
emilmont 1:fdd22bb7aa52 97 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 98 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 99 {
emilmont 1:fdd22bb7aa52 100 /* Set status as ARM_MATH_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 101 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 102 }
emilmont 1:fdd22bb7aa52 103 else
emilmont 1:fdd22bb7aa52 104 {
emilmont 1:fdd22bb7aa52 105
emilmont 1:fdd22bb7aa52 106 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 107 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 108 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 109 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 110 {
emilmont 1:fdd22bb7aa52 111 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 112 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 113
emilmont 1:fdd22bb7aa52 114 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 115 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 116 }
emilmont 1:fdd22bb7aa52 117 else
emilmont 1:fdd22bb7aa52 118 {
emilmont 1:fdd22bb7aa52 119 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 120 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 121
emilmont 1:fdd22bb7aa52 122 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 123 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 124
emilmont 1:fdd22bb7aa52 125 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 126 j = srcBLen;
emilmont 1:fdd22bb7aa52 127 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 128 srcALen = j;
emilmont 1:fdd22bb7aa52 129 }
emilmont 1:fdd22bb7aa52 130
emilmont 1:fdd22bb7aa52 131 /* Conditions to check which loopCounter holds
emilmont 1:fdd22bb7aa52 132 * the first and last indices of the output samples to be calculated. */
emilmont 1:fdd22bb7aa52 133 check = firstIndex + numPoints;
emilmont 1:fdd22bb7aa52 134 blockSize3 = ((int32_t) check - (int32_t) srcALen);
emilmont 1:fdd22bb7aa52 135 blockSize3 = (blockSize3 > 0) ? blockSize3 : 0;
emilmont 1:fdd22bb7aa52 136 blockSize1 = (((int32_t) srcBLen - 1) - (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 137 blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1u)) ? blockSize1 :
emilmont 1:fdd22bb7aa52 138 (int32_t) numPoints) : 0;
emilmont 1:fdd22bb7aa52 139 blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) +
emilmont 1:fdd22bb7aa52 140 (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 141 blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
emilmont 1:fdd22bb7aa52 142
emilmont 1:fdd22bb7aa52 143 /* 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 144 /* The function is internally
emilmont 1:fdd22bb7aa52 145 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 146 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 147 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 148 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 149 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 150 * for every iteration. */
emilmont 1:fdd22bb7aa52 151
emilmont 1:fdd22bb7aa52 152 /* Set the output pointer to point to the firstIndex
emilmont 1:fdd22bb7aa52 153 * of the output sample to be calculated. */
emilmont 1:fdd22bb7aa52 154 pOut = pDst + firstIndex;
emilmont 1:fdd22bb7aa52 155
emilmont 1:fdd22bb7aa52 156 /* --------------------------
emilmont 1:fdd22bb7aa52 157 * Initializations of stage1
emilmont 1:fdd22bb7aa52 158 * -------------------------*/
emilmont 1:fdd22bb7aa52 159
emilmont 1:fdd22bb7aa52 160 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 161 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 162 * ....
emilmont 1:fdd22bb7aa52 163 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 164 */
emilmont 1:fdd22bb7aa52 165
emilmont 1:fdd22bb7aa52 166 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 167 The count variable holds the number of MAC operations performed.
emilmont 1:fdd22bb7aa52 168 Since the partial convolution starts from from firstIndex
emilmont 1:fdd22bb7aa52 169 Number of Macs to be performed is firstIndex + 1 */
emilmont 1:fdd22bb7aa52 170 count = 1u + firstIndex;
emilmont 1:fdd22bb7aa52 171
emilmont 1:fdd22bb7aa52 172 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 173 px = pIn1;
emilmont 1:fdd22bb7aa52 174
emilmont 1:fdd22bb7aa52 175 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 176 pSrc2 = pIn2 + firstIndex;
emilmont 1:fdd22bb7aa52 177 py = pSrc2;
emilmont 1:fdd22bb7aa52 178
emilmont 1:fdd22bb7aa52 179 /* ------------------------
emilmont 1:fdd22bb7aa52 180 * Stage1 process
emilmont 1:fdd22bb7aa52 181 * ----------------------*/
emilmont 1:fdd22bb7aa52 182
emilmont 1:fdd22bb7aa52 183 /* The first stage starts here */
emilmont 1:fdd22bb7aa52 184 while(blockSize1 > 0)
emilmont 1:fdd22bb7aa52 185 {
emilmont 1:fdd22bb7aa52 186 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 187 sum = 0;
emilmont 1:fdd22bb7aa52 188
emilmont 1:fdd22bb7aa52 189 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 190 k = count >> 2u;
emilmont 1:fdd22bb7aa52 191
emilmont 1:fdd22bb7aa52 192 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 193 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 194 while(k > 0u)
emilmont 1:fdd22bb7aa52 195 {
emilmont 1:fdd22bb7aa52 196 /* x[0] , x[1] */
emilmont 1:fdd22bb7aa52 197 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 198 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 199 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 200
emilmont 1:fdd22bb7aa52 201 /* y[srcBLen - 1] , y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 202 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 203 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 204 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 207 /* x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 208 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 209
emilmont 1:fdd22bb7aa52 210 /* x[2] , x[3] */
emilmont 1:fdd22bb7aa52 211 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 212 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 213 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 214
emilmont 1:fdd22bb7aa52 215 /* y[srcBLen - 3] , y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 216 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 217 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 218 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 219
emilmont 1:fdd22bb7aa52 220 /* x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 221 /* x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 222 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 223
emilmont 1:fdd22bb7aa52 224 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 225 k--;
emilmont 1:fdd22bb7aa52 226 }
emilmont 1:fdd22bb7aa52 227
emilmont 1:fdd22bb7aa52 228 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 229 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 230 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 231
emilmont 1:fdd22bb7aa52 232 while(k > 0u)
emilmont 1:fdd22bb7aa52 233 {
emilmont 1:fdd22bb7aa52 234 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 235 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 238 k--;
emilmont 1:fdd22bb7aa52 239 }
emilmont 1:fdd22bb7aa52 240
emilmont 1:fdd22bb7aa52 241 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 242 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 243
emilmont 1:fdd22bb7aa52 244 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 245 py = ++pSrc2;
emilmont 1:fdd22bb7aa52 246 px = pIn1;
emilmont 1:fdd22bb7aa52 247
emilmont 1:fdd22bb7aa52 248 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 249 count++;
emilmont 1:fdd22bb7aa52 250
emilmont 1:fdd22bb7aa52 251 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 252 blockSize1--;
emilmont 1:fdd22bb7aa52 253 }
emilmont 1:fdd22bb7aa52 254
emilmont 1:fdd22bb7aa52 255 /* --------------------------
emilmont 1:fdd22bb7aa52 256 * Initializations of stage2
emilmont 1:fdd22bb7aa52 257 * ------------------------*/
emilmont 1:fdd22bb7aa52 258
emilmont 1:fdd22bb7aa52 259 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 260 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 261 * ....
emilmont 1:fdd22bb7aa52 262 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 263 */
emilmont 1:fdd22bb7aa52 264
emilmont 1:fdd22bb7aa52 265 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 266 px = pIn1;
emilmont 1:fdd22bb7aa52 267
emilmont 1:fdd22bb7aa52 268 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 269 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 270 py = pSrc2;
emilmont 1:fdd22bb7aa52 271
emilmont 1:fdd22bb7aa52 272 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 273 count = 0u;
emilmont 1:fdd22bb7aa52 274
emilmont 1:fdd22bb7aa52 275 /* -------------------
emilmont 1:fdd22bb7aa52 276 * Stage2 process
emilmont 1:fdd22bb7aa52 277 * ------------------*/
emilmont 1:fdd22bb7aa52 278
emilmont 1:fdd22bb7aa52 279 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 280 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 281 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 282 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 283 {
emilmont 1:fdd22bb7aa52 284 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 285 blkCnt = ((uint32_t) blockSize2 >> 2u);
emilmont 1:fdd22bb7aa52 286
emilmont 1:fdd22bb7aa52 287 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 288 {
emilmont 1:fdd22bb7aa52 289 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 290 acc0 = 0;
emilmont 1:fdd22bb7aa52 291 acc1 = 0;
emilmont 1:fdd22bb7aa52 292 acc2 = 0;
emilmont 1:fdd22bb7aa52 293 acc3 = 0;
emilmont 1:fdd22bb7aa52 294
emilmont 1:fdd22bb7aa52 295 /* read x[0], x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 296 x0 = *(px++);
emilmont 1:fdd22bb7aa52 297 x1 = *(px++);
emilmont 1:fdd22bb7aa52 298 x2 = *(px++);
emilmont 1:fdd22bb7aa52 299
emilmont 1:fdd22bb7aa52 300 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 301 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 302
emilmont 1:fdd22bb7aa52 303 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 304 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 305 do
emilmont 1:fdd22bb7aa52 306 {
emilmont 1:fdd22bb7aa52 307 /* Read y[srcBLen - 1] sample */
emilmont 1:fdd22bb7aa52 308 c0 = *(py--);
emilmont 1:fdd22bb7aa52 309 /* Read y[srcBLen - 2] sample */
emilmont 1:fdd22bb7aa52 310 c1 = *(py--);
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 313 x3 = *(px++);
emilmont 1:fdd22bb7aa52 314
emilmont 1:fdd22bb7aa52 315 /* x[0] and x[1] are packed */
emilmont 1:fdd22bb7aa52 316 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 317 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 318
emilmont 1:fdd22bb7aa52 319 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 320
emilmont 1:fdd22bb7aa52 321 /* y[srcBLen - 1] and y[srcBLen - 2] are packed */
emilmont 1:fdd22bb7aa52 322 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 323 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 324
emilmont 1:fdd22bb7aa52 325 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 326
emilmont 1:fdd22bb7aa52 327 /* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 328 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 329
emilmont 1:fdd22bb7aa52 330 /* x[1] and x[2] are packed */
emilmont 1:fdd22bb7aa52 331 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 332 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 333
emilmont 1:fdd22bb7aa52 334 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 335
emilmont 1:fdd22bb7aa52 336 /* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 337 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 338
emilmont 1:fdd22bb7aa52 339 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 340 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 341 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 342
emilmont 1:fdd22bb7aa52 343 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 344
emilmont 1:fdd22bb7aa52 345 /* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 346 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 347
emilmont 1:fdd22bb7aa52 348 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 349 x0 = *(px++);
emilmont 1:fdd22bb7aa52 350
emilmont 1:fdd22bb7aa52 351 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 352 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 353 in2 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 354
emilmont 1:fdd22bb7aa52 355 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 356
emilmont 1:fdd22bb7aa52 357 /* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 358 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 359
emilmont 1:fdd22bb7aa52 360 /* Read y[srcBLen - 3] sample */
emilmont 1:fdd22bb7aa52 361 c0 = *(py--);
emilmont 1:fdd22bb7aa52 362 /* Read y[srcBLen - 4] sample */
emilmont 1:fdd22bb7aa52 363 c1 = *(py--);
emilmont 1:fdd22bb7aa52 364
emilmont 1:fdd22bb7aa52 365 /* Read x[5] sample */
emilmont 1:fdd22bb7aa52 366 x1 = *(px++);
emilmont 1:fdd22bb7aa52 367
emilmont 1:fdd22bb7aa52 368 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 369 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 370 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 371
emilmont 1:fdd22bb7aa52 372 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 373
emilmont 1:fdd22bb7aa52 374 /* y[srcBLen - 3] and y[srcBLen - 4] are packed */
emilmont 1:fdd22bb7aa52 375 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 376 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 377
emilmont 1:fdd22bb7aa52 378 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 379
emilmont 1:fdd22bb7aa52 380 /* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 381 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 382
emilmont 1:fdd22bb7aa52 383 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 384 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 385 in2 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 386
emilmont 1:fdd22bb7aa52 387 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 388
emilmont 1:fdd22bb7aa52 389 /* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 390 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 391
emilmont 1:fdd22bb7aa52 392 /* x[4] and x[5] are packed */
emilmont 1:fdd22bb7aa52 393 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 394 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 395
emilmont 1:fdd22bb7aa52 396 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 397
emilmont 1:fdd22bb7aa52 398 /* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 399 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 400
emilmont 1:fdd22bb7aa52 401 /* Read x[6] sample */
emilmont 1:fdd22bb7aa52 402 x2 = *(px++);
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 /* x[5] and x[6] are packed */
emilmont 1:fdd22bb7aa52 405 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 406 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 407
emilmont 1:fdd22bb7aa52 408 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 409
emilmont 1:fdd22bb7aa52 410 /* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 411 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 412
emilmont 1:fdd22bb7aa52 413 } while(--k);
emilmont 1:fdd22bb7aa52 414
emilmont 1:fdd22bb7aa52 415 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 416 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 417 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 418
emilmont 1:fdd22bb7aa52 419 while(k > 0u)
emilmont 1:fdd22bb7aa52 420 {
emilmont 1:fdd22bb7aa52 421 /* Read y[srcBLen - 5] sample */
emilmont 1:fdd22bb7aa52 422 c0 = *(py--);
emilmont 1:fdd22bb7aa52 423
emilmont 1:fdd22bb7aa52 424 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 425 x3 = *(px++);
emilmont 1:fdd22bb7aa52 426
emilmont 1:fdd22bb7aa52 427 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 428 /* acc0 += x[4] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 429 acc0 += ((q31_t) x0 * c0);
emilmont 1:fdd22bb7aa52 430 /* acc1 += x[5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 431 acc1 += ((q31_t) x1 * c0);
emilmont 1:fdd22bb7aa52 432 /* acc2 += x[6] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 433 acc2 += ((q31_t) x2 * c0);
emilmont 1:fdd22bb7aa52 434 /* acc3 += x[7] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 435 acc3 += ((q31_t) x3 * c0);
emilmont 1:fdd22bb7aa52 436
emilmont 1:fdd22bb7aa52 437 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 438 x0 = x1;
emilmont 1:fdd22bb7aa52 439 x1 = x2;
emilmont 1:fdd22bb7aa52 440 x2 = x3;
emilmont 1:fdd22bb7aa52 441
emilmont 1:fdd22bb7aa52 442 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 443 k--;
emilmont 1:fdd22bb7aa52 444 }
emilmont 1:fdd22bb7aa52 445
emilmont 1:fdd22bb7aa52 446 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 447 *pOut++ = (q7_t) (__SSAT(acc0 >> 7, 8));
emilmont 1:fdd22bb7aa52 448 *pOut++ = (q7_t) (__SSAT(acc1 >> 7, 8));
emilmont 1:fdd22bb7aa52 449 *pOut++ = (q7_t) (__SSAT(acc2 >> 7, 8));
emilmont 1:fdd22bb7aa52 450 *pOut++ = (q7_t) (__SSAT(acc3 >> 7, 8));
emilmont 1:fdd22bb7aa52 451
emilmont 1:fdd22bb7aa52 452 /* Increment the pointer pIn1 index, count by 4 */
emilmont 1:fdd22bb7aa52 453 count += 4u;
emilmont 1:fdd22bb7aa52 454
emilmont 1:fdd22bb7aa52 455 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 456 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 457 py = pSrc2;
emilmont 1:fdd22bb7aa52 458
emilmont 1:fdd22bb7aa52 459
emilmont 1:fdd22bb7aa52 460 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 461 blkCnt--;
emilmont 1:fdd22bb7aa52 462 }
emilmont 1:fdd22bb7aa52 463
emilmont 1:fdd22bb7aa52 464 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 465 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 466 blkCnt = (uint32_t) blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 467
emilmont 1:fdd22bb7aa52 468 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 469 {
emilmont 1:fdd22bb7aa52 470 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 471 sum = 0;
emilmont 1:fdd22bb7aa52 472
emilmont 1:fdd22bb7aa52 473 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 474 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 475
emilmont 1:fdd22bb7aa52 476 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 477 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 478 while(k > 0u)
emilmont 1:fdd22bb7aa52 479 {
emilmont 1:fdd22bb7aa52 480
emilmont 1:fdd22bb7aa52 481 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 482 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 483 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 484 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 485
emilmont 1:fdd22bb7aa52 486 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 487 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 488 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 489 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 490
emilmont 1:fdd22bb7aa52 491 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 492 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 493
emilmont 1:fdd22bb7aa52 494 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 495 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 496 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 497 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 498
emilmont 1:fdd22bb7aa52 499 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 500 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 501 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 502 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 503
emilmont 1:fdd22bb7aa52 504 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 505 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 506
emilmont 1:fdd22bb7aa52 507 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 508 k--;
emilmont 1:fdd22bb7aa52 509 }
emilmont 1:fdd22bb7aa52 510
emilmont 1:fdd22bb7aa52 511 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 512 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 513 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 514
emilmont 1:fdd22bb7aa52 515 while(k > 0u)
emilmont 1:fdd22bb7aa52 516 {
emilmont 1:fdd22bb7aa52 517 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 518 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 519
emilmont 1:fdd22bb7aa52 520 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 521 k--;
emilmont 1:fdd22bb7aa52 522 }
emilmont 1:fdd22bb7aa52 523
emilmont 1:fdd22bb7aa52 524 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 525 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 526
emilmont 1:fdd22bb7aa52 527 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 528 count++;
emilmont 1:fdd22bb7aa52 529
emilmont 1:fdd22bb7aa52 530 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 531 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 532 py = pSrc2;
emilmont 1:fdd22bb7aa52 533
emilmont 1:fdd22bb7aa52 534 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 535 blkCnt--;
emilmont 1:fdd22bb7aa52 536 }
emilmont 1:fdd22bb7aa52 537 }
emilmont 1:fdd22bb7aa52 538 else
emilmont 1:fdd22bb7aa52 539 {
emilmont 1:fdd22bb7aa52 540 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 541 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 542 blkCnt = (uint32_t) blockSize2;
emilmont 1:fdd22bb7aa52 543
emilmont 1:fdd22bb7aa52 544 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 545 {
emilmont 1:fdd22bb7aa52 546 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 547 sum = 0;
emilmont 1:fdd22bb7aa52 548
emilmont 1:fdd22bb7aa52 549 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 550 k = srcBLen;
emilmont 1:fdd22bb7aa52 551
emilmont 1:fdd22bb7aa52 552 while(k > 0u)
emilmont 1:fdd22bb7aa52 553 {
emilmont 1:fdd22bb7aa52 554 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 555 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 556
emilmont 1:fdd22bb7aa52 557 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 558 k--;
emilmont 1:fdd22bb7aa52 559 }
emilmont 1:fdd22bb7aa52 560
emilmont 1:fdd22bb7aa52 561 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 562 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 563
emilmont 1:fdd22bb7aa52 564 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 565 count++;
emilmont 1:fdd22bb7aa52 566
emilmont 1:fdd22bb7aa52 567 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 568 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 569 py = pSrc2;
emilmont 1:fdd22bb7aa52 570
emilmont 1:fdd22bb7aa52 571 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 572 blkCnt--;
emilmont 1:fdd22bb7aa52 573 }
emilmont 1:fdd22bb7aa52 574 }
emilmont 1:fdd22bb7aa52 575
emilmont 1:fdd22bb7aa52 576
emilmont 1:fdd22bb7aa52 577 /* --------------------------
emilmont 1:fdd22bb7aa52 578 * Initializations of stage3
emilmont 1:fdd22bb7aa52 579 * -------------------------*/
emilmont 1:fdd22bb7aa52 580
emilmont 1:fdd22bb7aa52 581 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 582 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 583 * ....
emilmont 1:fdd22bb7aa52 584 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 585 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 586 */
emilmont 1:fdd22bb7aa52 587
emilmont 1:fdd22bb7aa52 588 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 589 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 590 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 591
emilmont 1:fdd22bb7aa52 592 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 593 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 594 px = pSrc1;
emilmont 1:fdd22bb7aa52 595
emilmont 1:fdd22bb7aa52 596 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 597 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 598 py = pSrc2;
emilmont 1:fdd22bb7aa52 599
emilmont 1:fdd22bb7aa52 600 /* -------------------
emilmont 1:fdd22bb7aa52 601 * Stage3 process
emilmont 1:fdd22bb7aa52 602 * ------------------*/
emilmont 1:fdd22bb7aa52 603
emilmont 1:fdd22bb7aa52 604 while(blockSize3 > 0)
emilmont 1:fdd22bb7aa52 605 {
emilmont 1:fdd22bb7aa52 606 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 607 sum = 0;
emilmont 1:fdd22bb7aa52 608
emilmont 1:fdd22bb7aa52 609 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 610 k = count >> 2u;
emilmont 1:fdd22bb7aa52 611
emilmont 1:fdd22bb7aa52 612 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 613 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 614 while(k > 0u)
emilmont 1:fdd22bb7aa52 615 {
emilmont 1:fdd22bb7aa52 616 /* Reading two inputs, x[srcALen - srcBLen + 1] and x[srcALen - srcBLen + 2] of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 617 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 618 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 619 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 620
emilmont 1:fdd22bb7aa52 621 /* Reading two inputs, y[srcBLen - 1] and y[srcBLen - 2] of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 622 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 623 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 624 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 625
emilmont 1:fdd22bb7aa52 626 /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 627 /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 628 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 629
emilmont 1:fdd22bb7aa52 630 /* Reading two inputs, x[srcALen - srcBLen + 3] and x[srcALen - srcBLen + 4] of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 631 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 632 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 633 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 634
emilmont 1:fdd22bb7aa52 635 /* Reading two inputs, y[srcBLen - 3] and y[srcBLen - 4] of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 636 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 637 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 638 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 639
emilmont 1:fdd22bb7aa52 640 /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 641 /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 642 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 643
emilmont 1:fdd22bb7aa52 644 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 645 k--;
emilmont 1:fdd22bb7aa52 646 }
emilmont 1:fdd22bb7aa52 647
emilmont 1:fdd22bb7aa52 648 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 649 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 650 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 651
emilmont 1:fdd22bb7aa52 652 while(k > 0u)
emilmont 1:fdd22bb7aa52 653 {
emilmont 1:fdd22bb7aa52 654 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 655 /* sum += x[srcALen-1] * y[srcBLen-1] */
emilmont 1:fdd22bb7aa52 656 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 657
emilmont 1:fdd22bb7aa52 658 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 659 k--;
emilmont 1:fdd22bb7aa52 660 }
emilmont 1:fdd22bb7aa52 661
emilmont 1:fdd22bb7aa52 662 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 663 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 664
emilmont 1:fdd22bb7aa52 665 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 666 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 667 py = pSrc2;
emilmont 1:fdd22bb7aa52 668
emilmont 1:fdd22bb7aa52 669 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 670 count--;
emilmont 1:fdd22bb7aa52 671
emilmont 1:fdd22bb7aa52 672 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 673 blockSize3--;
emilmont 1:fdd22bb7aa52 674
emilmont 1:fdd22bb7aa52 675 }
emilmont 1:fdd22bb7aa52 676
emilmont 1:fdd22bb7aa52 677 /* set status as ARM_MATH_SUCCESS */
emilmont 1:fdd22bb7aa52 678 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 679 }
emilmont 1:fdd22bb7aa52 680
emilmont 1:fdd22bb7aa52 681 /* Return to application */
emilmont 1:fdd22bb7aa52 682 return (status);
emilmont 1:fdd22bb7aa52 683
emilmont 1:fdd22bb7aa52 684 #else
emilmont 1:fdd22bb7aa52 685
emilmont 1:fdd22bb7aa52 686 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 687
emilmont 1:fdd22bb7aa52 688 q7_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 689 q7_t *pIn2 = pSrcB; /* inputB pointer */
emilmont 1:fdd22bb7aa52 690 q31_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 691 uint32_t i, j; /* loop counters */
emilmont 1:fdd22bb7aa52 692 arm_status status; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 693
emilmont 1:fdd22bb7aa52 694 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 695 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 696 {
emilmont 1:fdd22bb7aa52 697 /* Set status as ARM_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 698 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 699 }
emilmont 1:fdd22bb7aa52 700 else
emilmont 1:fdd22bb7aa52 701 {
emilmont 1:fdd22bb7aa52 702 /* Loop to calculate convolution for output length number of values */
emilmont 1:fdd22bb7aa52 703 for (i = firstIndex; i <= (firstIndex + numPoints - 1); i++)
emilmont 1:fdd22bb7aa52 704 {
emilmont 1:fdd22bb7aa52 705 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 706 sum = 0;
emilmont 1:fdd22bb7aa52 707
emilmont 1:fdd22bb7aa52 708 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 709 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 710 {
emilmont 1:fdd22bb7aa52 711 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 712 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 713 {
emilmont 1:fdd22bb7aa52 714 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 715 sum += ((q15_t) pIn1[j] * (pIn2[i - j]));
emilmont 1:fdd22bb7aa52 716 }
emilmont 1:fdd22bb7aa52 717 }
emilmont 1:fdd22bb7aa52 718
emilmont 1:fdd22bb7aa52 719 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 720 pDst[i] = (q7_t) __SSAT((sum >> 7u), 8u);
emilmont 1:fdd22bb7aa52 721 }
emilmont 1:fdd22bb7aa52 722 /* set status as ARM_SUCCESS as there are no argument errors */
emilmont 1:fdd22bb7aa52 723 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 724 }
emilmont 1:fdd22bb7aa52 725 return (status);
emilmont 1:fdd22bb7aa52 726
emilmont 1:fdd22bb7aa52 727 #endif /* #ifndef ARM_MATH_CM0 */
emilmont 1:fdd22bb7aa52 728
emilmont 1:fdd22bb7aa52 729 }
emilmont 1:fdd22bb7aa52 730
emilmont 1:fdd22bb7aa52 731 /**
emilmont 1:fdd22bb7aa52 732 * @} end of PartialConv group
emilmont 1:fdd22bb7aa52 733 */