mbed official / mbed-dsp

CMSIS DSP library

Dependents:   performance_timer Surfboard_ gps2rtty Capstone ... more

Legacy Warning

This is an mbed 2 library. To learn more about mbed OS 5, visit the docs.

cmsis_dsp/FilteringFunctions/arm_correlate_q7.c@5:3762170b6d4d, 2015-11-20 (annotated)

Committer:
mbed_official
Date:
Fri Nov 20 08:45:18 2015 +0000
Revision:
5:3762170b6d4d
Parent:
3:7a284390b0ce 
Synchronized with git revision 2eb940b9a73af188d3004a2575fdfbb05febe62b



Full URL: https://github.com/mbedmicro/mbed/commit/2eb940b9a73af188d3004a2575fdfbb05febe62b/



Added option to build rpc library. closes #1426

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emilmont 1:fdd22bb7aa52 1 /* ----------------------------------------------------------------------
mbed_official 5:3762170b6d4d 2 * Copyright (C) 2010-2014 ARM Limited. All rights reserved.
emilmont 1:fdd22bb7aa52 3 *
mbed_official 5:3762170b6d4d 4 * $Date: 19. March 2015
mbed_official 5:3762170b6d4d 5 * $Revision: V.1.4.5
emilmont 1:fdd22bb7aa52 6 *
emilmont 2:da51fb522205 7 * Project: CMSIS DSP Library
emilmont 2:da51fb522205 8 * Title: arm_correlate_q7.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Correlation of Q7 sequences.
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
emilmont 1:fdd22bb7aa52 13 *
mbed_official 3:7a284390b0ce 14 * Redistribution and use in source and binary forms, with or without
mbed_official 3:7a284390b0ce 15 * modification, are permitted provided that the following conditions
mbed_official 3:7a284390b0ce 16 * are met:
mbed_official 3:7a284390b0ce 17 * - Redistributions of source code must retain the above copyright
mbed_official 3:7a284390b0ce 18 * notice, this list of conditions and the following disclaimer.
mbed_official 3:7a284390b0ce 19 * - Redistributions in binary form must reproduce the above copyright
mbed_official 3:7a284390b0ce 20 * notice, this list of conditions and the following disclaimer in
mbed_official 3:7a284390b0ce 21 * the documentation and/or other materials provided with the
mbed_official 3:7a284390b0ce 22 * distribution.
mbed_official 3:7a284390b0ce 23 * - Neither the name of ARM LIMITED nor the names of its contributors
mbed_official 3:7a284390b0ce 24 * may be used to endorse or promote products derived from this
mbed_official 3:7a284390b0ce 25 * software without specific prior written permission.
emilmont 1:fdd22bb7aa52 26 *
mbed_official 3:7a284390b0ce 27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
mbed_official 3:7a284390b0ce 28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
mbed_official 3:7a284390b0ce 29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
mbed_official 3:7a284390b0ce 30 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
mbed_official 3:7a284390b0ce 31 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
mbed_official 3:7a284390b0ce 32 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
mbed_official 3:7a284390b0ce 33 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
mbed_official 3:7a284390b0ce 34 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
mbed_official 3:7a284390b0ce 35 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
mbed_official 3:7a284390b0ce 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
mbed_official 3:7a284390b0ce 37 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
mbed_official 3:7a284390b0ce 38 * POSSIBILITY OF SUCH DAMAGE.
emilmont 1:fdd22bb7aa52 39 * -------------------------------------------------------------------- */
emilmont 1:fdd22bb7aa52 40
emilmont 1:fdd22bb7aa52 41 #include "arm_math.h"
emilmont 1:fdd22bb7aa52 42
emilmont 1:fdd22bb7aa52 43 /**
emilmont 1:fdd22bb7aa52 44 * @ingroup groupFilters
emilmont 1:fdd22bb7aa52 45 */
emilmont 1:fdd22bb7aa52 46
emilmont 1:fdd22bb7aa52 47 /**
emilmont 1:fdd22bb7aa52 48 * @addtogroup Corr
emilmont 1:fdd22bb7aa52 49 * @{
emilmont 1:fdd22bb7aa52 50 */
emilmont 1:fdd22bb7aa52 51
emilmont 1:fdd22bb7aa52 52 /**
emilmont 1:fdd22bb7aa52 53 * @brief Correlation of Q7 sequences.
emilmont 1:fdd22bb7aa52 54 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 58 * @param[out] *pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1.
emilmont 1:fdd22bb7aa52 59 * @return none.
emilmont 1:fdd22bb7aa52 60 *
emilmont 1:fdd22bb7aa52 61 * @details
emilmont 1:fdd22bb7aa52 62 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 63 *
emilmont 1:fdd22bb7aa52 64 * \par
emilmont 1:fdd22bb7aa52 65 * The function is implemented using a 32-bit internal accumulator.
emilmont 1:fdd22bb7aa52 66 * Both the inputs are represented in 1.7 format and multiplications yield a 2.14 result.
emilmont 1:fdd22bb7aa52 67 * The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format.
emilmont 1:fdd22bb7aa52 68 * 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 69 * The 18.14 result is then truncated to 18.7 format by discarding the low 7 bits and saturated to 1.7 format.
emilmont 1:fdd22bb7aa52 70 *
emilmont 1:fdd22bb7aa52 71 * \par
emilmont 1:fdd22bb7aa52 72 * Refer the function <code>arm_correlate_opt_q7()</code> for a faster implementation of this function.
emilmont 1:fdd22bb7aa52 73 *
emilmont 1:fdd22bb7aa52 74 */
emilmont 1:fdd22bb7aa52 75
emilmont 1:fdd22bb7aa52 76 void arm_correlate_q7(
emilmont 1:fdd22bb7aa52 77 q7_t * pSrcA,
emilmont 1:fdd22bb7aa52 78 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 79 q7_t * pSrcB,
emilmont 1:fdd22bb7aa52 80 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 81 q7_t * pDst)
emilmont 1:fdd22bb7aa52 82 {
emilmont 1:fdd22bb7aa52 83
emilmont 1:fdd22bb7aa52 84
mbed_official 3:7a284390b0ce 85 #ifndef ARM_MATH_CM0_FAMILY
emilmont 1:fdd22bb7aa52 86
emilmont 1:fdd22bb7aa52 87 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 88
emilmont 1:fdd22bb7aa52 89 q7_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 90 q7_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 91 q7_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 92 q7_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 93 q7_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 94 q7_t *pSrc1; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 95 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulators */
emilmont 1:fdd22bb7aa52 96 q31_t input1, input2; /* temporary variables */
emilmont 1:fdd22bb7aa52 97 q15_t in1, in2; /* temporary variables */
emilmont 1:fdd22bb7aa52 98 q7_t x0, x1, x2, x3, c0, c1; /* temporary variables for holding input and coefficient values */
emilmont 1:fdd22bb7aa52 99 uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 100 int32_t inc = 1;
emilmont 1:fdd22bb7aa52 101
emilmont 1:fdd22bb7aa52 102
emilmont 1:fdd22bb7aa52 103 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 104 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 105 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 106 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 107 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 108 /* and the destination pointer modifier, inc is set to -1 */
emilmont 1:fdd22bb7aa52 109 /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */
emilmont 1:fdd22bb7aa52 110 /* But to improve the performance,
emilmont 1:fdd22bb7aa52 111 * we include zeroes in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 112 /* If srcALen > srcBLen,
emilmont 1:fdd22bb7aa52 113 * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */
emilmont 1:fdd22bb7aa52 114 /* If srcALen < srcBLen,
emilmont 1:fdd22bb7aa52 115 * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */
emilmont 1:fdd22bb7aa52 116 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 117 {
emilmont 1:fdd22bb7aa52 118 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 119 pIn1 = (pSrcA);
emilmont 1:fdd22bb7aa52 120
emilmont 1:fdd22bb7aa52 121 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 122 pIn2 = (pSrcB);
emilmont 1:fdd22bb7aa52 123
emilmont 1:fdd22bb7aa52 124 /* Number of output samples is calculated */
emilmont 1:fdd22bb7aa52 125 outBlockSize = (2u * srcALen) - 1u;
emilmont 1:fdd22bb7aa52 126
emilmont 1:fdd22bb7aa52 127 /* When srcALen > srcBLen, zero padding is done to srcB
emilmont 1:fdd22bb7aa52 128 * to make their lengths equal.
emilmont 1:fdd22bb7aa52 129 * Instead, (outBlockSize - (srcALen + srcBLen - 1))
emilmont 1:fdd22bb7aa52 130 * number of output samples are made zero */
emilmont 1:fdd22bb7aa52 131 j = outBlockSize - (srcALen + (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 132
emilmont 1:fdd22bb7aa52 133 /* Updating the pointer position to non zero value */
emilmont 1:fdd22bb7aa52 134 pOut += j;
emilmont 1:fdd22bb7aa52 135
emilmont 1:fdd22bb7aa52 136 }
emilmont 1:fdd22bb7aa52 137 else
emilmont 1:fdd22bb7aa52 138 {
emilmont 1:fdd22bb7aa52 139 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 140 pIn1 = (pSrcB);
emilmont 1:fdd22bb7aa52 141
emilmont 1:fdd22bb7aa52 142 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 143 pIn2 = (pSrcA);
emilmont 1:fdd22bb7aa52 144
emilmont 1:fdd22bb7aa52 145 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 146 j = srcBLen;
emilmont 1:fdd22bb7aa52 147 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 148 srcALen = j;
emilmont 1:fdd22bb7aa52 149
emilmont 1:fdd22bb7aa52 150 /* CORR(x, y) = Reverse order(CORR(y, x)) */
emilmont 1:fdd22bb7aa52 151 /* Hence set the destination pointer to point to the last output sample */
emilmont 1:fdd22bb7aa52 152 pOut = pDst + ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 153
emilmont 1:fdd22bb7aa52 154 /* Destination address modifier is set to -1 */
emilmont 1:fdd22bb7aa52 155 inc = -1;
emilmont 1:fdd22bb7aa52 156
emilmont 1:fdd22bb7aa52 157 }
emilmont 1:fdd22bb7aa52 158
emilmont 1:fdd22bb7aa52 159 /* The function is internally
emilmont 1:fdd22bb7aa52 160 * divided into three parts according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 161 * taken place between inputA samples and inputB samples. In the first part of the
emilmont 1:fdd22bb7aa52 162 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 163 * In the second part of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 164 * In the third part of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 165 * for every iteration.*/
emilmont 1:fdd22bb7aa52 166 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 167 * The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 168 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 169 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 170 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 171
emilmont 1:fdd22bb7aa52 172 /* --------------------------
emilmont 1:fdd22bb7aa52 173 * Initializations of stage1
emilmont 1:fdd22bb7aa52 174 * -------------------------*/
emilmont 1:fdd22bb7aa52 175
emilmont 1:fdd22bb7aa52 176 /* sum = x[0] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 177 * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1]
emilmont 1:fdd22bb7aa52 178 * ....
emilmont 1:fdd22bb7aa52 179 * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1]
emilmont 1:fdd22bb7aa52 180 */
emilmont 1:fdd22bb7aa52 181
emilmont 1:fdd22bb7aa52 182 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 183 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 184 count = 1u;
emilmont 1:fdd22bb7aa52 185
emilmont 1:fdd22bb7aa52 186 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 187 px = pIn1;
emilmont 1:fdd22bb7aa52 188
emilmont 1:fdd22bb7aa52 189 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 190 pSrc1 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 191 py = pSrc1;
emilmont 1:fdd22bb7aa52 192
emilmont 1:fdd22bb7aa52 193 /* ------------------------
emilmont 1:fdd22bb7aa52 194 * Stage1 process
emilmont 1:fdd22bb7aa52 195 * ----------------------*/
emilmont 1:fdd22bb7aa52 196
emilmont 1:fdd22bb7aa52 197 /* The first stage starts here */
emilmont 1:fdd22bb7aa52 198 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 199 {
emilmont 1:fdd22bb7aa52 200 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 201 sum = 0;
emilmont 1:fdd22bb7aa52 202
emilmont 1:fdd22bb7aa52 203 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 204 k = count >> 2;
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 207 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 208 while(k > 0u)
emilmont 1:fdd22bb7aa52 209 {
emilmont 1:fdd22bb7aa52 210 /* x[0] , x[1] */
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 - 4] , y[srcBLen - 3] */
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[0] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 221 /* x[1] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 222 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 223
emilmont 1:fdd22bb7aa52 224 /* x[2] , x[3] */
emilmont 1:fdd22bb7aa52 225 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 226 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 227 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 228
emilmont 1:fdd22bb7aa52 229 /* y[srcBLen - 2] , y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 230 in1 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 231 in2 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 232 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 233
emilmont 1:fdd22bb7aa52 234 /* x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 235 /* x[3] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 236 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 237
emilmont 1:fdd22bb7aa52 238
emilmont 1:fdd22bb7aa52 239 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 240 k--;
emilmont 1:fdd22bb7aa52 241 }
emilmont 1:fdd22bb7aa52 242
emilmont 1:fdd22bb7aa52 243 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 244 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 245 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 246
emilmont 1:fdd22bb7aa52 247 while(k > 0u)
emilmont 1:fdd22bb7aa52 248 {
emilmont 1:fdd22bb7aa52 249 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 250 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 251 sum += (q31_t) ((q15_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 252
emilmont 1:fdd22bb7aa52 253 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 254 k--;
emilmont 1:fdd22bb7aa52 255 }
emilmont 1:fdd22bb7aa52 256
emilmont 1:fdd22bb7aa52 257 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 258 *pOut = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 259 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 260 pOut += inc;
emilmont 1:fdd22bb7aa52 261
emilmont 1:fdd22bb7aa52 262 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 263 py = pSrc1 - count;
emilmont 1:fdd22bb7aa52 264 px = pIn1;
emilmont 1:fdd22bb7aa52 265
emilmont 1:fdd22bb7aa52 266 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 267 count++;
emilmont 1:fdd22bb7aa52 268
emilmont 1:fdd22bb7aa52 269 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 270 blockSize1--;
emilmont 1:fdd22bb7aa52 271 }
emilmont 1:fdd22bb7aa52 272
emilmont 1:fdd22bb7aa52 273 /* --------------------------
emilmont 1:fdd22bb7aa52 274 * Initializations of stage2
emilmont 1:fdd22bb7aa52 275 * ------------------------*/
emilmont 1:fdd22bb7aa52 276
emilmont 1:fdd22bb7aa52 277 /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 278 * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 279 * ....
emilmont 1:fdd22bb7aa52 280 * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 281 */
emilmont 1:fdd22bb7aa52 282
emilmont 1:fdd22bb7aa52 283 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 284 px = pIn1;
emilmont 1:fdd22bb7aa52 285
emilmont 1:fdd22bb7aa52 286 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 287 py = pIn2;
emilmont 1:fdd22bb7aa52 288
emilmont 1:fdd22bb7aa52 289 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 290 count = 0u;
emilmont 1:fdd22bb7aa52 291
emilmont 1:fdd22bb7aa52 292 /* -------------------
emilmont 1:fdd22bb7aa52 293 * Stage2 process
emilmont 1:fdd22bb7aa52 294 * ------------------*/
emilmont 1:fdd22bb7aa52 295
emilmont 1:fdd22bb7aa52 296 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 297 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 298 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 299 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 300 {
emilmont 1:fdd22bb7aa52 301 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 302 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 303
emilmont 1:fdd22bb7aa52 304 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 305 {
emilmont 1:fdd22bb7aa52 306 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 307 acc0 = 0;
emilmont 1:fdd22bb7aa52 308 acc1 = 0;
emilmont 1:fdd22bb7aa52 309 acc2 = 0;
emilmont 1:fdd22bb7aa52 310 acc3 = 0;
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 /* read x[0], x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 313 x0 = *px++;
emilmont 1:fdd22bb7aa52 314 x1 = *px++;
emilmont 1:fdd22bb7aa52 315 x2 = *px++;
emilmont 1:fdd22bb7aa52 316
emilmont 1:fdd22bb7aa52 317 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 318 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 319
emilmont 1:fdd22bb7aa52 320 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 321 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 322 do
emilmont 1:fdd22bb7aa52 323 {
emilmont 1:fdd22bb7aa52 324 /* Read y[0] sample */
emilmont 1:fdd22bb7aa52 325 c0 = *py++;
emilmont 1:fdd22bb7aa52 326 /* Read y[1] sample */
emilmont 1:fdd22bb7aa52 327 c1 = *py++;
emilmont 1:fdd22bb7aa52 328
emilmont 1:fdd22bb7aa52 329 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 330 x3 = *px++;
emilmont 1:fdd22bb7aa52 331
emilmont 1:fdd22bb7aa52 332 /* x[0] and x[1] are packed */
emilmont 1:fdd22bb7aa52 333 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 334 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 335
emilmont 1:fdd22bb7aa52 336 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 337
emilmont 1:fdd22bb7aa52 338 /* y[0] and y[1] are packed */
emilmont 1:fdd22bb7aa52 339 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 340 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 343
emilmont 1:fdd22bb7aa52 344 /* acc0 += x[0] * y[0] + x[1] * y[1] */
emilmont 1:fdd22bb7aa52 345 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 346
emilmont 1:fdd22bb7aa52 347 /* x[1] and x[2] are packed */
emilmont 1:fdd22bb7aa52 348 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 349 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 350
emilmont 1:fdd22bb7aa52 351 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 352
emilmont 1:fdd22bb7aa52 353 /* acc1 += x[1] * y[0] + x[2] * y[1] */
emilmont 1:fdd22bb7aa52 354 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 355
emilmont 1:fdd22bb7aa52 356 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 357 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 358 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 359
emilmont 1:fdd22bb7aa52 360 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 361
emilmont 1:fdd22bb7aa52 362 /* acc2 += x[2] * y[0] + x[3] * y[1] */
emilmont 1:fdd22bb7aa52 363 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 364
emilmont 1:fdd22bb7aa52 365 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 366 x0 = *(px++);
emilmont 1:fdd22bb7aa52 367
emilmont 1:fdd22bb7aa52 368 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 369 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 370 in2 = (q15_t) x0;
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 /* acc3 += x[3] * y[0] + x[4] * y[1] */
emilmont 1:fdd22bb7aa52 375 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 376
emilmont 1:fdd22bb7aa52 377 /* Read y[2] sample */
emilmont 1:fdd22bb7aa52 378 c0 = *py++;
emilmont 1:fdd22bb7aa52 379 /* Read y[3] sample */
emilmont 1:fdd22bb7aa52 380 c1 = *py++;
emilmont 1:fdd22bb7aa52 381
emilmont 1:fdd22bb7aa52 382 /* Read x[5] sample */
emilmont 1:fdd22bb7aa52 383 x1 = *px++;
emilmont 1:fdd22bb7aa52 384
emilmont 1:fdd22bb7aa52 385 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 386 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 387 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 388
emilmont 1:fdd22bb7aa52 389 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 390
emilmont 1:fdd22bb7aa52 391 /* y[2] and y[3] are packed */
emilmont 1:fdd22bb7aa52 392 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 393 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 394
emilmont 1:fdd22bb7aa52 395 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 396
emilmont 1:fdd22bb7aa52 397 /* acc0 += x[2] * y[2] + x[3] * y[3] */
emilmont 1:fdd22bb7aa52 398 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 399
emilmont 1:fdd22bb7aa52 400 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 401 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 402 in2 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 405
emilmont 1:fdd22bb7aa52 406 /* acc1 += x[3] * y[2] + x[4] * y[3] */
emilmont 1:fdd22bb7aa52 407 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 408
emilmont 1:fdd22bb7aa52 409 /* x[4] and x[5] are packed */
emilmont 1:fdd22bb7aa52 410 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 411 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 412
emilmont 1:fdd22bb7aa52 413 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 414
emilmont 1:fdd22bb7aa52 415 /* acc2 += x[4] * y[2] + x[5] * y[3] */
emilmont 1:fdd22bb7aa52 416 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 417
emilmont 1:fdd22bb7aa52 418 /* Read x[6] sample */
emilmont 1:fdd22bb7aa52 419 x2 = *px++;
emilmont 1:fdd22bb7aa52 420
emilmont 1:fdd22bb7aa52 421 /* x[5] and x[6] are packed */
emilmont 1:fdd22bb7aa52 422 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 423 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 424
emilmont 1:fdd22bb7aa52 425 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 426
emilmont 1:fdd22bb7aa52 427 /* acc3 += x[5] * y[2] + x[6] * y[3] */
emilmont 1:fdd22bb7aa52 428 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 429
emilmont 1:fdd22bb7aa52 430 } while(--k);
emilmont 1:fdd22bb7aa52 431
emilmont 1:fdd22bb7aa52 432 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 433 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 434 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 435
emilmont 1:fdd22bb7aa52 436 while(k > 0u)
emilmont 1:fdd22bb7aa52 437 {
emilmont 1:fdd22bb7aa52 438 /* Read y[4] sample */
emilmont 1:fdd22bb7aa52 439 c0 = *py++;
emilmont 1:fdd22bb7aa52 440
emilmont 1:fdd22bb7aa52 441 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 442 x3 = *px++;
emilmont 1:fdd22bb7aa52 443
emilmont 1:fdd22bb7aa52 444 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 445 /* acc0 += x[4] * y[4] */
emilmont 1:fdd22bb7aa52 446 acc0 += ((q15_t) x0 * c0);
emilmont 1:fdd22bb7aa52 447 /* acc1 += x[5] * y[4] */
emilmont 1:fdd22bb7aa52 448 acc1 += ((q15_t) x1 * c0);
emilmont 1:fdd22bb7aa52 449 /* acc2 += x[6] * y[4] */
emilmont 1:fdd22bb7aa52 450 acc2 += ((q15_t) x2 * c0);
emilmont 1:fdd22bb7aa52 451 /* acc3 += x[7] * y[4] */
emilmont 1:fdd22bb7aa52 452 acc3 += ((q15_t) x3 * c0);
emilmont 1:fdd22bb7aa52 453
emilmont 1:fdd22bb7aa52 454 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 455 x0 = x1;
emilmont 1:fdd22bb7aa52 456 x1 = x2;
emilmont 1:fdd22bb7aa52 457 x2 = x3;
emilmont 1:fdd22bb7aa52 458
emilmont 1:fdd22bb7aa52 459 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 460 k--;
emilmont 1:fdd22bb7aa52 461 }
emilmont 1:fdd22bb7aa52 462
emilmont 1:fdd22bb7aa52 463 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 464 *pOut = (q7_t) (__SSAT(acc0 >> 7, 8));
emilmont 1:fdd22bb7aa52 465 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 466 pOut += inc;
emilmont 1:fdd22bb7aa52 467
emilmont 1:fdd22bb7aa52 468 *pOut = (q7_t) (__SSAT(acc1 >> 7, 8));
emilmont 1:fdd22bb7aa52 469 pOut += inc;
emilmont 1:fdd22bb7aa52 470
emilmont 1:fdd22bb7aa52 471 *pOut = (q7_t) (__SSAT(acc2 >> 7, 8));
emilmont 1:fdd22bb7aa52 472 pOut += inc;
emilmont 1:fdd22bb7aa52 473
emilmont 1:fdd22bb7aa52 474 *pOut = (q7_t) (__SSAT(acc3 >> 7, 8));
emilmont 1:fdd22bb7aa52 475 pOut += inc;
emilmont 1:fdd22bb7aa52 476
emilmont 2:da51fb522205 477 count += 4u;
emilmont 1:fdd22bb7aa52 478 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 479 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 480 py = pIn2;
emilmont 1:fdd22bb7aa52 481
emilmont 1:fdd22bb7aa52 482 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 483 blkCnt--;
emilmont 1:fdd22bb7aa52 484 }
emilmont 1:fdd22bb7aa52 485
emilmont 1:fdd22bb7aa52 486 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 487 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 488 blkCnt = blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 489
emilmont 1:fdd22bb7aa52 490 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 491 {
emilmont 1:fdd22bb7aa52 492 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 493 sum = 0;
emilmont 1:fdd22bb7aa52 494
emilmont 1:fdd22bb7aa52 495 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 496 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 497
emilmont 1:fdd22bb7aa52 498 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 499 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 500 while(k > 0u)
emilmont 1:fdd22bb7aa52 501 {
emilmont 1:fdd22bb7aa52 502 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 503 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 504 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 505 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 506
emilmont 1:fdd22bb7aa52 507 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 508 in1 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 509 in2 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 510 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 511
emilmont 1:fdd22bb7aa52 512 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 513 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 514
emilmont 1:fdd22bb7aa52 515 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 516 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 517 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 518 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 519
emilmont 1:fdd22bb7aa52 520 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 521 in1 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 522 in2 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 523 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 524
emilmont 1:fdd22bb7aa52 525 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 526 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 527
emilmont 1:fdd22bb7aa52 528 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 529 k--;
emilmont 1:fdd22bb7aa52 530 }
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 533 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 534 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 535
emilmont 1:fdd22bb7aa52 536 while(k > 0u)
emilmont 1:fdd22bb7aa52 537 {
emilmont 1:fdd22bb7aa52 538 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 539 sum += ((q15_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 540
emilmont 1:fdd22bb7aa52 541 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 542 k--;
emilmont 1:fdd22bb7aa52 543 }
emilmont 1:fdd22bb7aa52 544
emilmont 1:fdd22bb7aa52 545 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 546 *pOut = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 547 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 548 pOut += inc;
emilmont 1:fdd22bb7aa52 549
emilmont 1:fdd22bb7aa52 550 /* Increment the pointer pIn1 index, count by 1 */
emilmont 2:da51fb522205 551 count++;
emilmont 1:fdd22bb7aa52 552
emilmont 1:fdd22bb7aa52 553 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 554 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 555 py = pIn2;
emilmont 1:fdd22bb7aa52 556
emilmont 1:fdd22bb7aa52 557 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 558 blkCnt--;
emilmont 1:fdd22bb7aa52 559 }
emilmont 1:fdd22bb7aa52 560 }
emilmont 1:fdd22bb7aa52 561 else
emilmont 1:fdd22bb7aa52 562 {
emilmont 1:fdd22bb7aa52 563 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 564 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 565 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 566
emilmont 1:fdd22bb7aa52 567 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 568 {
emilmont 1:fdd22bb7aa52 569 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 570 sum = 0;
emilmont 1:fdd22bb7aa52 571
emilmont 1:fdd22bb7aa52 572 /* Loop over srcBLen */
emilmont 1:fdd22bb7aa52 573 k = srcBLen;
emilmont 1:fdd22bb7aa52 574
emilmont 1:fdd22bb7aa52 575 while(k > 0u)
emilmont 1:fdd22bb7aa52 576 {
emilmont 1:fdd22bb7aa52 577 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 578 sum += ((q15_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 579
emilmont 1:fdd22bb7aa52 580 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 581 k--;
emilmont 1:fdd22bb7aa52 582 }
emilmont 1:fdd22bb7aa52 583
emilmont 1:fdd22bb7aa52 584 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 585 *pOut = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 586 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 587 pOut += inc;
emilmont 1:fdd22bb7aa52 588
emilmont 1:fdd22bb7aa52 589 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 590 count++;
emilmont 1:fdd22bb7aa52 591
emilmont 1:fdd22bb7aa52 592 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 593 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 594 py = pIn2;
emilmont 1:fdd22bb7aa52 595
emilmont 1:fdd22bb7aa52 596
emilmont 1:fdd22bb7aa52 597 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 598 blkCnt--;
emilmont 1:fdd22bb7aa52 599 }
emilmont 1:fdd22bb7aa52 600 }
emilmont 1:fdd22bb7aa52 601
emilmont 1:fdd22bb7aa52 602 /* --------------------------
emilmont 1:fdd22bb7aa52 603 * Initializations of stage3
emilmont 1:fdd22bb7aa52 604 * -------------------------*/
emilmont 1:fdd22bb7aa52 605
emilmont 1:fdd22bb7aa52 606 /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 607 * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 608 * ....
emilmont 1:fdd22bb7aa52 609 * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 610 * sum += x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 611 */
emilmont 1:fdd22bb7aa52 612
emilmont 1:fdd22bb7aa52 613 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 614 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 615 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 616
emilmont 1:fdd22bb7aa52 617 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 618 pSrc1 = pIn1 + (srcALen - (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 619 px = pSrc1;
emilmont 1:fdd22bb7aa52 620
emilmont 1:fdd22bb7aa52 621 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 622 py = pIn2;
emilmont 1:fdd22bb7aa52 623
emilmont 1:fdd22bb7aa52 624 /* -------------------
emilmont 1:fdd22bb7aa52 625 * Stage3 process
emilmont 1:fdd22bb7aa52 626 * ------------------*/
emilmont 1:fdd22bb7aa52 627
emilmont 1:fdd22bb7aa52 628 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 629 {
emilmont 1:fdd22bb7aa52 630 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 631 sum = 0;
emilmont 1:fdd22bb7aa52 632
emilmont 1:fdd22bb7aa52 633 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 634 k = count >> 2u;
emilmont 1:fdd22bb7aa52 635
emilmont 1:fdd22bb7aa52 636 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 637 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 638 while(k > 0u)
emilmont 1:fdd22bb7aa52 639 {
emilmont 1:fdd22bb7aa52 640 /* x[srcALen - srcBLen + 1] , x[srcALen - srcBLen + 2] */
emilmont 1:fdd22bb7aa52 641 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 642 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 643 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 644
emilmont 1:fdd22bb7aa52 645 /* y[0] , y[1] */
emilmont 1:fdd22bb7aa52 646 in1 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 647 in2 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 648 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 649
emilmont 1:fdd22bb7aa52 650 /* sum += x[srcALen - srcBLen + 1] * y[0] */
emilmont 1:fdd22bb7aa52 651 /* sum += x[srcALen - srcBLen + 2] * y[1] */
emilmont 1:fdd22bb7aa52 652 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 653
emilmont 1:fdd22bb7aa52 654 /* x[srcALen - srcBLen + 3] , x[srcALen - srcBLen + 4] */
emilmont 1:fdd22bb7aa52 655 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 656 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 657 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 658
emilmont 1:fdd22bb7aa52 659 /* y[2] , y[3] */
emilmont 1:fdd22bb7aa52 660 in1 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 661 in2 = (q15_t) * py++;
emilmont 1:fdd22bb7aa52 662 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 663
emilmont 1:fdd22bb7aa52 664 /* sum += x[srcALen - srcBLen + 3] * y[2] */
emilmont 1:fdd22bb7aa52 665 /* sum += x[srcALen - srcBLen + 4] * y[3] */
emilmont 1:fdd22bb7aa52 666 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 667
emilmont 1:fdd22bb7aa52 668 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 669 k--;
emilmont 1:fdd22bb7aa52 670 }
emilmont 1:fdd22bb7aa52 671
emilmont 1:fdd22bb7aa52 672 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 673 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 674 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 675
emilmont 1:fdd22bb7aa52 676 while(k > 0u)
emilmont 1:fdd22bb7aa52 677 {
emilmont 1:fdd22bb7aa52 678 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 679 sum += ((q15_t) * px++ * *py++);
emilmont 1:fdd22bb7aa52 680
emilmont 1:fdd22bb7aa52 681 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 682 k--;
emilmont 1:fdd22bb7aa52 683 }
emilmont 1:fdd22bb7aa52 684
emilmont 1:fdd22bb7aa52 685 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 686 *pOut = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 687 /* Destination pointer is updated according to the address modifier, inc */
emilmont 1:fdd22bb7aa52 688 pOut += inc;
emilmont 1:fdd22bb7aa52 689
emilmont 1:fdd22bb7aa52 690 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 691 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 692 py = pIn2;
emilmont 1:fdd22bb7aa52 693
emilmont 1:fdd22bb7aa52 694 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 695 count--;
emilmont 1:fdd22bb7aa52 696
emilmont 1:fdd22bb7aa52 697 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 698 blockSize3--;
emilmont 1:fdd22bb7aa52 699 }
emilmont 1:fdd22bb7aa52 700
emilmont 1:fdd22bb7aa52 701 #else
emilmont 1:fdd22bb7aa52 702
emilmont 1:fdd22bb7aa52 703 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 704
emilmont 1:fdd22bb7aa52 705 q7_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 706 q7_t *pIn2 = pSrcB + (srcBLen - 1u); /* inputB pointer */
emilmont 1:fdd22bb7aa52 707 q31_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 708 uint32_t i = 0u, j; /* loop counters */
emilmont 1:fdd22bb7aa52 709 uint32_t inv = 0u; /* Reverse order flag */
emilmont 1:fdd22bb7aa52 710 uint32_t tot = 0u; /* Length */
emilmont 1:fdd22bb7aa52 711
emilmont 1:fdd22bb7aa52 712 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 713 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 714 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 715 /* But CORR(x, y) is reverse of CORR(y, x) */
emilmont 1:fdd22bb7aa52 716 /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
emilmont 1:fdd22bb7aa52 717 /* and a varaible, inv is set to 1 */
emilmont 1:fdd22bb7aa52 718 /* If lengths are not equal then zero pad has to be done to make the two
emilmont 1:fdd22bb7aa52 719 * inputs of same length. But to improve the performance, we include zeroes
emilmont 1:fdd22bb7aa52 720 * in the output instead of zero padding either of the the inputs*/
emilmont 1:fdd22bb7aa52 721 /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the
emilmont 1:fdd22bb7aa52 722 * starting of the output buffer */
emilmont 1:fdd22bb7aa52 723 /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the
emilmont 1:fdd22bb7aa52 724 * ending of the output buffer */
emilmont 1:fdd22bb7aa52 725 /* Once the zero padding is done the remaining of the output is calcualted
emilmont 1:fdd22bb7aa52 726 * using convolution but with the shorter signal time shifted. */
emilmont 1:fdd22bb7aa52 727
emilmont 1:fdd22bb7aa52 728 /* Calculate the length of the remaining sequence */
emilmont 1:fdd22bb7aa52 729 tot = ((srcALen + srcBLen) - 2u);
emilmont 1:fdd22bb7aa52 730
emilmont 1:fdd22bb7aa52 731 if(srcALen > srcBLen)
emilmont 1:fdd22bb7aa52 732 {
emilmont 1:fdd22bb7aa52 733 /* Calculating the number of zeros to be padded to the output */
emilmont 1:fdd22bb7aa52 734 j = srcALen - srcBLen;
emilmont 1:fdd22bb7aa52 735
emilmont 1:fdd22bb7aa52 736 /* Initialise the pointer after zero padding */
emilmont 1:fdd22bb7aa52 737 pDst += j;
emilmont 1:fdd22bb7aa52 738 }
emilmont 1:fdd22bb7aa52 739
emilmont 1:fdd22bb7aa52 740 else if(srcALen < srcBLen)
emilmont 1:fdd22bb7aa52 741 {
emilmont 1:fdd22bb7aa52 742 /* Initialization to inputB pointer */
emilmont 1:fdd22bb7aa52 743 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 744
emilmont 1:fdd22bb7aa52 745 /* Initialization to the end of inputA pointer */
emilmont 1:fdd22bb7aa52 746 pIn2 = pSrcA + (srcALen - 1u);
emilmont 1:fdd22bb7aa52 747
emilmont 1:fdd22bb7aa52 748 /* Initialisation of the pointer after zero padding */
emilmont 1:fdd22bb7aa52 749 pDst = pDst + tot;
emilmont 1:fdd22bb7aa52 750
emilmont 1:fdd22bb7aa52 751 /* Swapping the lengths */
emilmont 1:fdd22bb7aa52 752 j = srcALen;
emilmont 1:fdd22bb7aa52 753 srcALen = srcBLen;
emilmont 1:fdd22bb7aa52 754 srcBLen = j;
emilmont 1:fdd22bb7aa52 755
emilmont 1:fdd22bb7aa52 756 /* Setting the reverse flag */
emilmont 1:fdd22bb7aa52 757 inv = 1;
emilmont 1:fdd22bb7aa52 758
emilmont 1:fdd22bb7aa52 759 }
emilmont 1:fdd22bb7aa52 760
emilmont 1:fdd22bb7aa52 761 /* Loop to calculate convolution for output length number of times */
emilmont 1:fdd22bb7aa52 762 for (i = 0u; i <= tot; i++)
emilmont 1:fdd22bb7aa52 763 {
emilmont 1:fdd22bb7aa52 764 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 765 sum = 0;
emilmont 1:fdd22bb7aa52 766
emilmont 1:fdd22bb7aa52 767 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 768 for (j = 0u; j <= i; j++)
emilmont 1:fdd22bb7aa52 769 {
emilmont 1:fdd22bb7aa52 770 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 771 if((((i - j) < srcBLen) && (j < srcALen)))
emilmont 1:fdd22bb7aa52 772 {
emilmont 1:fdd22bb7aa52 773 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 774 sum += ((q15_t) pIn1[j] * pIn2[-((int32_t) i - j)]);
emilmont 1:fdd22bb7aa52 775 }
emilmont 1:fdd22bb7aa52 776 }
emilmont 1:fdd22bb7aa52 777 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 778 if(inv == 1)
emilmont 1:fdd22bb7aa52 779 *pDst-- = (q7_t) __SSAT((sum >> 7u), 8u);
emilmont 1:fdd22bb7aa52 780 else
emilmont 1:fdd22bb7aa52 781 *pDst++ = (q7_t) __SSAT((sum >> 7u), 8u);
emilmont 1:fdd22bb7aa52 782 }
emilmont 1:fdd22bb7aa52 783
mbed_official 3:7a284390b0ce 784 #endif /* #ifndef ARM_MATH_CM0_FAMILY */
emilmont 1:fdd22bb7aa52 785
emilmont 1:fdd22bb7aa52 786 }
emilmont 1:fdd22bb7aa52 787
emilmont 1:fdd22bb7aa52 788 /**
emilmont 1:fdd22bb7aa52 789 * @} end of Corr group
emilmont 1:fdd22bb7aa52 790 */