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_q15.c@2:da51fb522205, 2013-05-30 (annotated)

Committer:
emilmont
Date:
Thu May 30 17:10:11 2013 +0100
Revision:
2:da51fb522205
Parent:
1:fdd22bb7aa52
Child:
3:7a284390b0ce
Keep "cmsis-dsp" module in synch with its source

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 2:da51fb522205 5 * $Revision: V1.1.0
emilmont 1:fdd22bb7aa52 6 *
emilmont 2:da51fb522205 7 * Project: CMSIS DSP Library
emilmont 2:da51fb522205 8 * Title: arm_conv_partial_q15.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Partial convolution of Q15 sequences.
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
emilmont 1:fdd22bb7aa52 13 *
emilmont 1:fdd22bb7aa52 14 * Version 1.1.0 2012/02/15
emilmont 1:fdd22bb7aa52 15 * Updated with more optimizations, bug fixes and minor API changes.
emilmont 1:fdd22bb7aa52 16 *
emilmont 1:fdd22bb7aa52 17 * Version 1.0.11 2011/10/18
emilmont 1:fdd22bb7aa52 18 * Bug Fix in conv, correlation, partial convolution.
emilmont 1:fdd22bb7aa52 19 *
emilmont 1:fdd22bb7aa52 20 * Version 1.0.10 2011/7/15
emilmont 1:fdd22bb7aa52 21 * Big Endian support added and Merged M0 and M3/M4 Source code.
emilmont 1:fdd22bb7aa52 22 *
emilmont 1:fdd22bb7aa52 23 * Version 1.0.3 2010/11/29
emilmont 1:fdd22bb7aa52 24 * Re-organized the CMSIS folders and updated documentation.
emilmont 1:fdd22bb7aa52 25 *
emilmont 1:fdd22bb7aa52 26 * Version 1.0.2 2010/11/11
emilmont 1:fdd22bb7aa52 27 * Documentation updated.
emilmont 1:fdd22bb7aa52 28 *
emilmont 1:fdd22bb7aa52 29 * Version 1.0.1 2010/10/05
emilmont 1:fdd22bb7aa52 30 * Production release and review comments incorporated.
emilmont 1:fdd22bb7aa52 31 *
emilmont 1:fdd22bb7aa52 32 * Version 1.0.0 2010/09/20
emilmont 1:fdd22bb7aa52 33 * Production release and review comments incorporated
emilmont 1:fdd22bb7aa52 34 *
emilmont 1:fdd22bb7aa52 35 * Version 0.0.7 2010/06/10
emilmont 1:fdd22bb7aa52 36 * Misra-C changes done
emilmont 1:fdd22bb7aa52 37 *
emilmont 1:fdd22bb7aa52 38 * -------------------------------------------------------------------- */
emilmont 1:fdd22bb7aa52 39
emilmont 1:fdd22bb7aa52 40 #include "arm_math.h"
emilmont 1:fdd22bb7aa52 41
emilmont 1:fdd22bb7aa52 42 /**
emilmont 1:fdd22bb7aa52 43 * @ingroup groupFilters
emilmont 1:fdd22bb7aa52 44 */
emilmont 1:fdd22bb7aa52 45
emilmont 1:fdd22bb7aa52 46 /**
emilmont 1:fdd22bb7aa52 47 * @addtogroup 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 Q15 sequences.
emilmont 1:fdd22bb7aa52 53 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 54 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[out] *pDst points to the location where the output result is written.
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 * Refer to <code>arm_conv_partial_fast_q15()</code> for a faster but less precise version of this function for Cortex-M3 and Cortex-M4.
emilmont 1:fdd22bb7aa52 63 *
emilmont 1:fdd22bb7aa52 64 * \par
emilmont 1:fdd22bb7aa52 65 * Refer the function <code>arm_conv_partial_opt_q15()</code> for a faster implementation of this function using scratch buffers.
emilmont 1:fdd22bb7aa52 66 *
emilmont 1:fdd22bb7aa52 67 */
emilmont 1:fdd22bb7aa52 68
emilmont 1:fdd22bb7aa52 69
emilmont 1:fdd22bb7aa52 70 arm_status arm_conv_partial_q15(
emilmont 1:fdd22bb7aa52 71 q15_t * pSrcA,
emilmont 1:fdd22bb7aa52 72 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 73 q15_t * pSrcB,
emilmont 1:fdd22bb7aa52 74 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 75 q15_t * pDst,
emilmont 1:fdd22bb7aa52 76 uint32_t firstIndex,
emilmont 1:fdd22bb7aa52 77 uint32_t numPoints)
emilmont 1:fdd22bb7aa52 78 {
emilmont 1:fdd22bb7aa52 79
emilmont 1:fdd22bb7aa52 80 #if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE)
emilmont 1:fdd22bb7aa52 81
emilmont 1:fdd22bb7aa52 82 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 83
emilmont 1:fdd22bb7aa52 84 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 85 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 86 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 87 q63_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 88 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 89 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 90 q15_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 91 q31_t x0, x1, x2, x3, c0; /* Temporary input variables */
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; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 95
emilmont 1:fdd22bb7aa52 96 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 97 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 98 {
emilmont 1:fdd22bb7aa52 99 /* Set status as ARM_MATH_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 100 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 101 }
emilmont 1:fdd22bb7aa52 102 else
emilmont 1:fdd22bb7aa52 103 {
emilmont 1:fdd22bb7aa52 104
emilmont 1:fdd22bb7aa52 105 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 106 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 107 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 108 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 109 {
emilmont 1:fdd22bb7aa52 110 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 111 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 112
emilmont 1:fdd22bb7aa52 113 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 114 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 115 }
emilmont 1:fdd22bb7aa52 116 else
emilmont 1:fdd22bb7aa52 117 {
emilmont 1:fdd22bb7aa52 118 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 119 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 120
emilmont 1:fdd22bb7aa52 121 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 122 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 123
emilmont 1:fdd22bb7aa52 124 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 125 j = srcBLen;
emilmont 1:fdd22bb7aa52 126 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 127 srcALen = j;
emilmont 1:fdd22bb7aa52 128 }
emilmont 1:fdd22bb7aa52 129
emilmont 1:fdd22bb7aa52 130 /* Conditions to check which loopCounter holds
emilmont 1:fdd22bb7aa52 131 * the first and last indices of the output samples to be calculated. */
emilmont 1:fdd22bb7aa52 132 check = firstIndex + numPoints;
emilmont 1:fdd22bb7aa52 133 blockSize3 = ((int32_t) check - (int32_t) srcALen);
emilmont 1:fdd22bb7aa52 134 blockSize3 = (blockSize3 > 0) ? blockSize3 : 0;
emilmont 1:fdd22bb7aa52 135 blockSize1 = (((int32_t) srcBLen - 1) - (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 136 blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1u)) ? blockSize1 :
emilmont 1:fdd22bb7aa52 137 (int32_t) numPoints) : 0;
emilmont 1:fdd22bb7aa52 138 blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) +
emilmont 1:fdd22bb7aa52 139 (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 140 blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
emilmont 1:fdd22bb7aa52 141
emilmont 1:fdd22bb7aa52 142 /* 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 143 /* The function is internally
emilmont 1:fdd22bb7aa52 144 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 145 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 146 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 147 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 148 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 149 * for every iteration. */
emilmont 1:fdd22bb7aa52 150
emilmont 1:fdd22bb7aa52 151 /* Set the output pointer to point to the firstIndex
emilmont 1:fdd22bb7aa52 152 * of the output sample to be calculated. */
emilmont 1:fdd22bb7aa52 153 pOut = pDst + firstIndex;
emilmont 1:fdd22bb7aa52 154
emilmont 1:fdd22bb7aa52 155 /* --------------------------
emilmont 1:fdd22bb7aa52 156 * Initializations of stage1
emilmont 1:fdd22bb7aa52 157 * -------------------------*/
emilmont 1:fdd22bb7aa52 158
emilmont 1:fdd22bb7aa52 159 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 160 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 161 * ....
emilmont 1:fdd22bb7aa52 162 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 163 */
emilmont 1:fdd22bb7aa52 164
emilmont 1:fdd22bb7aa52 165 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 166 The count variable holds the number of MAC operations performed.
emilmont 1:fdd22bb7aa52 167 Since the partial convolution starts from firstIndex
emilmont 1:fdd22bb7aa52 168 Number of Macs to be performed is firstIndex + 1 */
emilmont 1:fdd22bb7aa52 169 count = 1u + firstIndex;
emilmont 1:fdd22bb7aa52 170
emilmont 1:fdd22bb7aa52 171 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 172 px = pIn1;
emilmont 1:fdd22bb7aa52 173
emilmont 1:fdd22bb7aa52 174 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 175 pSrc2 = pIn2 + firstIndex;
emilmont 1:fdd22bb7aa52 176 py = pSrc2;
emilmont 1:fdd22bb7aa52 177
emilmont 1:fdd22bb7aa52 178 /* ------------------------
emilmont 1:fdd22bb7aa52 179 * Stage1 process
emilmont 1:fdd22bb7aa52 180 * ----------------------*/
emilmont 1:fdd22bb7aa52 181
emilmont 1:fdd22bb7aa52 182 /* For loop unrolling by 4, this stage is divided into two. */
emilmont 1:fdd22bb7aa52 183 /* First part of this stage computes the MAC operations less than 4 */
emilmont 1:fdd22bb7aa52 184 /* Second part of this stage computes the MAC operations greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 185
emilmont 1:fdd22bb7aa52 186 /* The first part of the stage starts here */
emilmont 1:fdd22bb7aa52 187 while((count < 4u) && (blockSize1 > 0))
emilmont 1:fdd22bb7aa52 188 {
emilmont 1:fdd22bb7aa52 189 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 190 sum = 0;
emilmont 1:fdd22bb7aa52 191
emilmont 1:fdd22bb7aa52 192 /* Loop over number of MAC operations between
emilmont 1:fdd22bb7aa52 193 * inputA samples and inputB samples */
emilmont 1:fdd22bb7aa52 194 k = count;
emilmont 1:fdd22bb7aa52 195
emilmont 1:fdd22bb7aa52 196 while(k > 0u)
emilmont 1:fdd22bb7aa52 197 {
emilmont 1:fdd22bb7aa52 198 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 199 sum = __SMLALD(*px++, *py--, sum);
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++ = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 207
emilmont 1:fdd22bb7aa52 208 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 209 py = ++pSrc2;
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 /* The second part of the stage starts here */
emilmont 1:fdd22bb7aa52 220 /* The internal loop, over count, is unrolled by 4 */
emilmont 1:fdd22bb7aa52 221 /* To, read the last two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 222 * y[srcBLen] and y[srcBLen-1] coefficients, py is decremented by 1 */
emilmont 1:fdd22bb7aa52 223 py = py - 1;
emilmont 1:fdd22bb7aa52 224
emilmont 1:fdd22bb7aa52 225 while(blockSize1 > 0)
emilmont 1:fdd22bb7aa52 226 {
emilmont 1:fdd22bb7aa52 227 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 228 sum = 0;
emilmont 1:fdd22bb7aa52 229
emilmont 1:fdd22bb7aa52 230 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 231 k = count >> 2u;
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 234 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 235 while(k > 0u)
emilmont 1:fdd22bb7aa52 236 {
emilmont 1:fdd22bb7aa52 237 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 238 /* x[0], x[1] are multiplied with y[srcBLen - 1], y[srcBLen - 2] respectively */
emilmont 1:fdd22bb7aa52 239 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 240 /* x[2], x[3] are multiplied with y[srcBLen - 3], y[srcBLen - 4] respectively */
emilmont 1:fdd22bb7aa52 241 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 242
emilmont 1:fdd22bb7aa52 243 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 244 k--;
emilmont 1:fdd22bb7aa52 245 }
emilmont 1:fdd22bb7aa52 246
emilmont 1:fdd22bb7aa52 247 /* For the next MAC operations, the pointer py is used without SIMD
emilmont 1:fdd22bb7aa52 248 * So, py is incremented by 1 */
emilmont 1:fdd22bb7aa52 249 py = py + 1u;
emilmont 1:fdd22bb7aa52 250
emilmont 1:fdd22bb7aa52 251 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 252 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 253 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 254
emilmont 1:fdd22bb7aa52 255 while(k > 0u)
emilmont 1:fdd22bb7aa52 256 {
emilmont 1:fdd22bb7aa52 257 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 258 sum = __SMLALD(*px++, *py--, sum);
emilmont 1:fdd22bb7aa52 259
emilmont 1:fdd22bb7aa52 260 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 261 k--;
emilmont 1:fdd22bb7aa52 262 }
emilmont 1:fdd22bb7aa52 263
emilmont 1:fdd22bb7aa52 264 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 265 *pOut++ = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 266
emilmont 1:fdd22bb7aa52 267 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 268 py = ++pSrc2 - 1u;
emilmont 1:fdd22bb7aa52 269 px = pIn1;
emilmont 1:fdd22bb7aa52 270
emilmont 1:fdd22bb7aa52 271 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 272 count++;
emilmont 1:fdd22bb7aa52 273
emilmont 1:fdd22bb7aa52 274 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 275 blockSize1--;
emilmont 1:fdd22bb7aa52 276 }
emilmont 1:fdd22bb7aa52 277
emilmont 1:fdd22bb7aa52 278 /* --------------------------
emilmont 1:fdd22bb7aa52 279 * Initializations of stage2
emilmont 1:fdd22bb7aa52 280 * ------------------------*/
emilmont 1:fdd22bb7aa52 281
emilmont 1:fdd22bb7aa52 282 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 283 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 284 * ....
emilmont 1:fdd22bb7aa52 285 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 286 */
emilmont 1:fdd22bb7aa52 287
emilmont 1:fdd22bb7aa52 288 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 289 px = pIn1;
emilmont 1:fdd22bb7aa52 290
emilmont 1:fdd22bb7aa52 291 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 292 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 293 py = pSrc2;
emilmont 1:fdd22bb7aa52 294
emilmont 1:fdd22bb7aa52 295 /* count is the index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 296 count = 0u;
emilmont 1:fdd22bb7aa52 297
emilmont 1:fdd22bb7aa52 298
emilmont 1:fdd22bb7aa52 299 /* --------------------
emilmont 1:fdd22bb7aa52 300 * Stage2 process
emilmont 1:fdd22bb7aa52 301 * -------------------*/
emilmont 1:fdd22bb7aa52 302
emilmont 1:fdd22bb7aa52 303 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 304 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 305 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 306 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 307 {
emilmont 1:fdd22bb7aa52 308 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 309 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 310
emilmont 1:fdd22bb7aa52 311 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 312 {
emilmont 1:fdd22bb7aa52 313 py = py - 1u;
emilmont 1:fdd22bb7aa52 314
emilmont 1:fdd22bb7aa52 315 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 316 acc0 = 0;
emilmont 1:fdd22bb7aa52 317 acc1 = 0;
emilmont 1:fdd22bb7aa52 318 acc2 = 0;
emilmont 1:fdd22bb7aa52 319 acc3 = 0;
emilmont 1:fdd22bb7aa52 320
emilmont 1:fdd22bb7aa52 321
emilmont 1:fdd22bb7aa52 322 /* read x[0], x[1] samples */
emilmont 1:fdd22bb7aa52 323 x0 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 324 /* read x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 325 x1 = _SIMD32_OFFSET(px+1);
emilmont 2:da51fb522205 326 px+= 2u;
emilmont 1:fdd22bb7aa52 327
emilmont 1:fdd22bb7aa52 328
emilmont 1:fdd22bb7aa52 329 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 330 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 331
emilmont 1:fdd22bb7aa52 332 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 333 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 334 do
emilmont 1:fdd22bb7aa52 335 {
emilmont 1:fdd22bb7aa52 336 /* Read the last two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 337 * y[srcBLen - 1] and y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 338 c0 = *__SIMD32(py)--;
emilmont 1:fdd22bb7aa52 339
emilmont 1:fdd22bb7aa52 340 /* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 341 acc0 = __SMLALDX(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 342
emilmont 1:fdd22bb7aa52 343 /* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 344 acc1 = __SMLALDX(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 345
emilmont 1:fdd22bb7aa52 346 /* Read x[2], x[3] */
emilmont 1:fdd22bb7aa52 347 x2 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 348
emilmont 1:fdd22bb7aa52 349 /* Read x[3], x[4] */
emilmont 1:fdd22bb7aa52 350 x3 = _SIMD32_OFFSET(px+1);
emilmont 1:fdd22bb7aa52 351
emilmont 1:fdd22bb7aa52 352 /* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 353 acc2 = __SMLALDX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 354
emilmont 1:fdd22bb7aa52 355 /* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 356 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 357
emilmont 1:fdd22bb7aa52 358 /* Read y[srcBLen - 3] and y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 359 c0 = *__SIMD32(py)--;
emilmont 1:fdd22bb7aa52 360
emilmont 1:fdd22bb7aa52 361 /* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 362 acc0 = __SMLALDX(x2, c0, acc0);
emilmont 1:fdd22bb7aa52 363
emilmont 1:fdd22bb7aa52 364 /* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 365 acc1 = __SMLALDX(x3, c0, acc1);
emilmont 1:fdd22bb7aa52 366
emilmont 1:fdd22bb7aa52 367 /* Read x[4], x[5] */
emilmont 1:fdd22bb7aa52 368 x0 = _SIMD32_OFFSET(px+2);
emilmont 1:fdd22bb7aa52 369
emilmont 1:fdd22bb7aa52 370 /* Read x[5], x[6] */
emilmont 1:fdd22bb7aa52 371 x1 = _SIMD32_OFFSET(px+3);
emilmont 2:da51fb522205 372 px += 4u;
emilmont 1:fdd22bb7aa52 373
emilmont 1:fdd22bb7aa52 374 /* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 375 acc2 = __SMLALDX(x0, c0, acc2);
emilmont 1:fdd22bb7aa52 376
emilmont 1:fdd22bb7aa52 377 /* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 378 acc3 = __SMLALDX(x1, c0, acc3);
emilmont 1:fdd22bb7aa52 379
emilmont 1:fdd22bb7aa52 380 } while(--k);
emilmont 1:fdd22bb7aa52 381
emilmont 1:fdd22bb7aa52 382 /* For the next MAC operations, SIMD is not used
emilmont 1:fdd22bb7aa52 383 * So, the 16 bit pointer if inputB, py is updated */
emilmont 1:fdd22bb7aa52 384
emilmont 1:fdd22bb7aa52 385 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 386 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 387 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 388
emilmont 1:fdd22bb7aa52 389 if(k == 1u)
emilmont 1:fdd22bb7aa52 390 {
emilmont 1:fdd22bb7aa52 391 /* Read y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 392 c0 = *(py+1);
emilmont 1:fdd22bb7aa52 393
emilmont 1:fdd22bb7aa52 394 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 395
emilmont 1:fdd22bb7aa52 396 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 397
emilmont 1:fdd22bb7aa52 398 #else
emilmont 1:fdd22bb7aa52 399
emilmont 1:fdd22bb7aa52 400 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 401
emilmont 1:fdd22bb7aa52 402 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 /* Read x[7] */
emilmont 1:fdd22bb7aa52 405 x3 = *__SIMD32(px);
emilmont 2:da51fb522205 406 px++;
emilmont 1:fdd22bb7aa52 407
emilmont 1:fdd22bb7aa52 408 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 409 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 410 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 411 acc2 = __SMLALDX(x1, c0, acc2);
emilmont 1:fdd22bb7aa52 412 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 413 }
emilmont 1:fdd22bb7aa52 414
emilmont 1:fdd22bb7aa52 415 if(k == 2u)
emilmont 1:fdd22bb7aa52 416 {
emilmont 1:fdd22bb7aa52 417 /* Read y[srcBLen - 5], y[srcBLen - 6] */
emilmont 1:fdd22bb7aa52 418 c0 = _SIMD32_OFFSET(py);
emilmont 1:fdd22bb7aa52 419
emilmont 1:fdd22bb7aa52 420 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 421 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 422
emilmont 1:fdd22bb7aa52 423 /* Read x[9] */
emilmont 1:fdd22bb7aa52 424 x2 = _SIMD32_OFFSET(px+1);
emilmont 2:da51fb522205 425 px += 2u;
emilmont 1:fdd22bb7aa52 426
emilmont 1:fdd22bb7aa52 427 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 428 acc0 = __SMLALDX(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 429 acc1 = __SMLALDX(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 430 acc2 = __SMLALDX(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 431 acc3 = __SMLALDX(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 432 }
emilmont 1:fdd22bb7aa52 433
emilmont 1:fdd22bb7aa52 434 if(k == 3u)
emilmont 1:fdd22bb7aa52 435 {
emilmont 1:fdd22bb7aa52 436 /* Read y[srcBLen - 5], y[srcBLen - 6] */
emilmont 1:fdd22bb7aa52 437 c0 = _SIMD32_OFFSET(py);
emilmont 1:fdd22bb7aa52 438
emilmont 1:fdd22bb7aa52 439 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 440 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 441
emilmont 1:fdd22bb7aa52 442 /* Read x[9] */
emilmont 1:fdd22bb7aa52 443 x2 = _SIMD32_OFFSET(px+1);
emilmont 1:fdd22bb7aa52 444
emilmont 1:fdd22bb7aa52 445 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 446 acc0 = __SMLALDX(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 447 acc1 = __SMLALDX(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 448 acc2 = __SMLALDX(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 449 acc3 = __SMLALDX(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 450
emilmont 2:da51fb522205 451 c0 = *(py-1);
emilmont 1:fdd22bb7aa52 452
emilmont 1:fdd22bb7aa52 453 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 454
emilmont 1:fdd22bb7aa52 455 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 456 #else
emilmont 1:fdd22bb7aa52 457
emilmont 1:fdd22bb7aa52 458 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 459 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 460
emilmont 1:fdd22bb7aa52 461 /* Read x[10] */
emilmont 1:fdd22bb7aa52 462 x3 = _SIMD32_OFFSET(px+2);
emilmont 2:da51fb522205 463 px += 3u;
emilmont 1:fdd22bb7aa52 464
emilmont 1:fdd22bb7aa52 465 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 466 acc0 = __SMLALDX(x1, c0, acc0);
emilmont 1:fdd22bb7aa52 467 acc1 = __SMLALD(x2, c0, acc1);
emilmont 1:fdd22bb7aa52 468 acc2 = __SMLALDX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 469 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 470 }
emilmont 1:fdd22bb7aa52 471
emilmont 1:fdd22bb7aa52 472
emilmont 1:fdd22bb7aa52 473 /* Store the results in the accumulators in the destination buffer. */
emilmont 1:fdd22bb7aa52 474
emilmont 1:fdd22bb7aa52 475 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 476
emilmont 1:fdd22bb7aa52 477 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 478 __PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 479 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 480 __PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 481
emilmont 1:fdd22bb7aa52 482 #else
emilmont 1:fdd22bb7aa52 483
emilmont 1:fdd22bb7aa52 484 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 485 __PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 486 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 487 __PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 488
emilmont 1:fdd22bb7aa52 489 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 490
emilmont 1:fdd22bb7aa52 491 /* Increment the pointer pIn1 index, count by 4 */
emilmont 1:fdd22bb7aa52 492 count += 4u;
emilmont 1:fdd22bb7aa52 493
emilmont 1:fdd22bb7aa52 494 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 495 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 496 py = pSrc2;
emilmont 1:fdd22bb7aa52 497
emilmont 1:fdd22bb7aa52 498 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 499 blkCnt--;
emilmont 1:fdd22bb7aa52 500 }
emilmont 1:fdd22bb7aa52 501
emilmont 1:fdd22bb7aa52 502 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 503 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 504 blkCnt = (uint32_t) blockSize2 % 0x4u;
emilmont 2:da51fb522205 505
emilmont 1:fdd22bb7aa52 506 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 507 {
emilmont 1:fdd22bb7aa52 508 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 509 sum = 0;
emilmont 1:fdd22bb7aa52 510
emilmont 1:fdd22bb7aa52 511 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 512 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 513
emilmont 1:fdd22bb7aa52 514 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 515 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 516 while(k > 0u)
emilmont 1:fdd22bb7aa52 517 {
emilmont 1:fdd22bb7aa52 518 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 519 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 520 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 521 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 522 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 523
emilmont 1:fdd22bb7aa52 524 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 525 k--;
emilmont 1:fdd22bb7aa52 526 }
emilmont 1:fdd22bb7aa52 527
emilmont 1:fdd22bb7aa52 528 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 529 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 530 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 while(k > 0u)
emilmont 1:fdd22bb7aa52 533 {
emilmont 1:fdd22bb7aa52 534 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 535 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 536
emilmont 1:fdd22bb7aa52 537 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 538 k--;
emilmont 1:fdd22bb7aa52 539 }
emilmont 1:fdd22bb7aa52 540
emilmont 1:fdd22bb7aa52 541 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 542 *pOut++ = (q15_t) (__SSAT(sum >> 15, 16));
emilmont 1:fdd22bb7aa52 543
emilmont 1:fdd22bb7aa52 544 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 545 count++;
emilmont 1:fdd22bb7aa52 546
emilmont 1:fdd22bb7aa52 547 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 548 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 549 py = pSrc2;
emilmont 1:fdd22bb7aa52 550
emilmont 1:fdd22bb7aa52 551 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 552 blkCnt--;
emilmont 1:fdd22bb7aa52 553 }
emilmont 1:fdd22bb7aa52 554 }
emilmont 1:fdd22bb7aa52 555 else
emilmont 1:fdd22bb7aa52 556 {
emilmont 1:fdd22bb7aa52 557 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 558 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 559 blkCnt = (uint32_t) blockSize2;
emilmont 1:fdd22bb7aa52 560
emilmont 1:fdd22bb7aa52 561 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 562 {
emilmont 1:fdd22bb7aa52 563 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 564 sum = 0;
emilmont 1:fdd22bb7aa52 565
emilmont 1:fdd22bb7aa52 566 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 567 k = srcBLen;
emilmont 1:fdd22bb7aa52 568
emilmont 1:fdd22bb7aa52 569 while(k > 0u)
emilmont 1:fdd22bb7aa52 570 {
emilmont 1:fdd22bb7aa52 571 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 572 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 573
emilmont 1:fdd22bb7aa52 574 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 575 k--;
emilmont 1:fdd22bb7aa52 576 }
emilmont 1:fdd22bb7aa52 577
emilmont 1:fdd22bb7aa52 578 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 579 *pOut++ = (q15_t) (__SSAT(sum >> 15, 16));
emilmont 1:fdd22bb7aa52 580
emilmont 1:fdd22bb7aa52 581 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 582 count++;
emilmont 1:fdd22bb7aa52 583
emilmont 1:fdd22bb7aa52 584 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 585 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 586 py = pSrc2;
emilmont 1:fdd22bb7aa52 587
emilmont 1:fdd22bb7aa52 588 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 589 blkCnt--;
emilmont 1:fdd22bb7aa52 590 }
emilmont 1:fdd22bb7aa52 591 }
emilmont 1:fdd22bb7aa52 592
emilmont 1:fdd22bb7aa52 593
emilmont 1:fdd22bb7aa52 594 /* --------------------------
emilmont 1:fdd22bb7aa52 595 * Initializations of stage3
emilmont 1:fdd22bb7aa52 596 * -------------------------*/
emilmont 1:fdd22bb7aa52 597
emilmont 1:fdd22bb7aa52 598 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 599 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 600 * ....
emilmont 1:fdd22bb7aa52 601 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 602 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 603 */
emilmont 1:fdd22bb7aa52 604
emilmont 1:fdd22bb7aa52 605 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 606 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 607 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 608
emilmont 1:fdd22bb7aa52 609 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 610 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 611 px = pSrc1;
emilmont 1:fdd22bb7aa52 612
emilmont 1:fdd22bb7aa52 613 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 614 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 615 pIn2 = pSrc2 - 1u;
emilmont 1:fdd22bb7aa52 616 py = pIn2;
emilmont 1:fdd22bb7aa52 617
emilmont 1:fdd22bb7aa52 618 /* -------------------
emilmont 1:fdd22bb7aa52 619 * Stage3 process
emilmont 1:fdd22bb7aa52 620 * ------------------*/
emilmont 1:fdd22bb7aa52 621
emilmont 1:fdd22bb7aa52 622 /* For loop unrolling by 4, this stage is divided into two. */
emilmont 1:fdd22bb7aa52 623 /* First part of this stage computes the MAC operations greater than 4 */
emilmont 1:fdd22bb7aa52 624 /* Second part of this stage computes the MAC operations less than or equal to 4 */
emilmont 1:fdd22bb7aa52 625
emilmont 1:fdd22bb7aa52 626 /* The first part of the stage starts here */
emilmont 1:fdd22bb7aa52 627 j = count >> 2u;
emilmont 1:fdd22bb7aa52 628
emilmont 1:fdd22bb7aa52 629 while((j > 0u) && (blockSize3 > 0))
emilmont 1:fdd22bb7aa52 630 {
emilmont 1:fdd22bb7aa52 631 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 632 sum = 0;
emilmont 1:fdd22bb7aa52 633
emilmont 1:fdd22bb7aa52 634 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 635 k = count >> 2u;
emilmont 1:fdd22bb7aa52 636
emilmont 1:fdd22bb7aa52 637 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 638 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 639 while(k > 0u)
emilmont 1:fdd22bb7aa52 640 {
emilmont 1:fdd22bb7aa52 641 /* x[srcALen - srcBLen + 1], x[srcALen - srcBLen + 2] are multiplied
emilmont 1:fdd22bb7aa52 642 * with y[srcBLen - 1], y[srcBLen - 2] respectively */
emilmont 1:fdd22bb7aa52 643 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 644 /* x[srcALen - srcBLen + 3], x[srcALen - srcBLen + 4] are multiplied
emilmont 1:fdd22bb7aa52 645 * with y[srcBLen - 3], y[srcBLen - 4] respectively */
emilmont 1:fdd22bb7aa52 646 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 647
emilmont 1:fdd22bb7aa52 648 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 649 k--;
emilmont 1:fdd22bb7aa52 650 }
emilmont 1:fdd22bb7aa52 651
emilmont 1:fdd22bb7aa52 652 /* For the next MAC operations, the pointer py is used without SIMD
emilmont 1:fdd22bb7aa52 653 * So, py is incremented by 1 */
emilmont 1:fdd22bb7aa52 654 py = py + 1u;
emilmont 1:fdd22bb7aa52 655
emilmont 1:fdd22bb7aa52 656 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 657 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 658 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 659
emilmont 1:fdd22bb7aa52 660 while(k > 0u)
emilmont 1:fdd22bb7aa52 661 {
emilmont 1:fdd22bb7aa52 662 /* sum += x[srcALen - srcBLen + 5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 663 sum = __SMLALD(*px++, *py--, sum);
emilmont 1:fdd22bb7aa52 664
emilmont 1:fdd22bb7aa52 665 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 666 k--;
emilmont 1:fdd22bb7aa52 667 }
emilmont 1:fdd22bb7aa52 668
emilmont 1:fdd22bb7aa52 669 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 670 *pOut++ = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 671
emilmont 1:fdd22bb7aa52 672 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 673 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 674 py = pIn2;
emilmont 1:fdd22bb7aa52 675
emilmont 1:fdd22bb7aa52 676 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 677 count--;
emilmont 1:fdd22bb7aa52 678
emilmont 1:fdd22bb7aa52 679 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 680 blockSize3--;
emilmont 1:fdd22bb7aa52 681
emilmont 1:fdd22bb7aa52 682 j--;
emilmont 1:fdd22bb7aa52 683 }
emilmont 1:fdd22bb7aa52 684
emilmont 1:fdd22bb7aa52 685 /* The second part of the stage starts here */
emilmont 1:fdd22bb7aa52 686 /* SIMD is not used for the next MAC operations,
emilmont 1:fdd22bb7aa52 687 * so pointer py is updated to read only one sample at a time */
emilmont 1:fdd22bb7aa52 688 py = py + 1u;
emilmont 1:fdd22bb7aa52 689
emilmont 1:fdd22bb7aa52 690 while(blockSize3 > 0)
emilmont 1:fdd22bb7aa52 691 {
emilmont 1:fdd22bb7aa52 692 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 693 sum = 0;
emilmont 1:fdd22bb7aa52 694
emilmont 1:fdd22bb7aa52 695 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 696 k = count;
emilmont 1:fdd22bb7aa52 697
emilmont 1:fdd22bb7aa52 698 while(k > 0u)
emilmont 1:fdd22bb7aa52 699 {
emilmont 1:fdd22bb7aa52 700 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 701 /* sum += x[srcALen-1] * y[srcBLen-1] */
emilmont 1:fdd22bb7aa52 702 sum = __SMLALD(*px++, *py--, sum);
emilmont 1:fdd22bb7aa52 703
emilmont 1:fdd22bb7aa52 704 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 705 k--;
emilmont 1:fdd22bb7aa52 706 }
emilmont 1:fdd22bb7aa52 707
emilmont 1:fdd22bb7aa52 708 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 709 *pOut++ = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 710
emilmont 1:fdd22bb7aa52 711 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 712 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 713 py = pSrc2;
emilmont 1:fdd22bb7aa52 714
emilmont 1:fdd22bb7aa52 715 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 716 count--;
emilmont 1:fdd22bb7aa52 717
emilmont 1:fdd22bb7aa52 718 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 719 blockSize3--;
emilmont 1:fdd22bb7aa52 720 }
emilmont 1:fdd22bb7aa52 721
emilmont 1:fdd22bb7aa52 722 /* set status as ARM_MATH_SUCCESS */
emilmont 1:fdd22bb7aa52 723 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 724 }
emilmont 1:fdd22bb7aa52 725
emilmont 1:fdd22bb7aa52 726 /* Return to application */
emilmont 1:fdd22bb7aa52 727 return (status);
emilmont 1:fdd22bb7aa52 728
emilmont 1:fdd22bb7aa52 729 #else
emilmont 1:fdd22bb7aa52 730
emilmont 1:fdd22bb7aa52 731 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 732
emilmont 1:fdd22bb7aa52 733 q15_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 734 q15_t *pIn2 = pSrcB; /* inputB pointer */
emilmont 1:fdd22bb7aa52 735 q63_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 736 uint32_t i, j; /* loop counters */
emilmont 1:fdd22bb7aa52 737 arm_status status; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 738
emilmont 1:fdd22bb7aa52 739 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 740 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 741 {
emilmont 1:fdd22bb7aa52 742 /* Set status as ARM_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 743 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 744 }
emilmont 1:fdd22bb7aa52 745 else
emilmont 1:fdd22bb7aa52 746 {
emilmont 1:fdd22bb7aa52 747 /* Loop to calculate convolution for output length number of values */
emilmont 1:fdd22bb7aa52 748 for (i = firstIndex; i <= (firstIndex + numPoints - 1); i++)
emilmont 1:fdd22bb7aa52 749 {
emilmont 1:fdd22bb7aa52 750 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 751 sum = 0;
emilmont 1:fdd22bb7aa52 752
emilmont 1:fdd22bb7aa52 753 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 754 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 755 {
emilmont 1:fdd22bb7aa52 756 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 757 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 758 {
emilmont 1:fdd22bb7aa52 759 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 760 sum += ((q31_t) pIn1[j] * (pIn2[i - j]));
emilmont 1:fdd22bb7aa52 761 }
emilmont 1:fdd22bb7aa52 762 }
emilmont 1:fdd22bb7aa52 763
emilmont 1:fdd22bb7aa52 764 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 765 pDst[i] = (q15_t) __SSAT((sum >> 15u), 16u);
emilmont 1:fdd22bb7aa52 766 }
emilmont 1:fdd22bb7aa52 767 /* set status as ARM_SUCCESS as there are no argument errors */
emilmont 1:fdd22bb7aa52 768 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 769 }
emilmont 1:fdd22bb7aa52 770 return (status);
emilmont 1:fdd22bb7aa52 771
emilmont 1:fdd22bb7aa52 772 #endif /* #if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE) */
emilmont 1:fdd22bb7aa52 773
emilmont 1:fdd22bb7aa52 774 }
emilmont 1:fdd22bb7aa52 775
emilmont 1:fdd22bb7aa52 776 /**
emilmont 1:fdd22bb7aa52 777 * @} end of PartialConv group
emilmont 1:fdd22bb7aa52 778 */