mbed official / mbed-dsp

CMSIS DSP library

Dependents:   performance_timer Surfboard_ gps2rtty Capstone ... more

Legacy Warning

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

cmsis_dsp/FilteringFunctions/arm_conv_q31.c@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_conv_q31.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Convolution of Q31 sequences.
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
emilmont 1:fdd22bb7aa52 13 *
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.
mbed_official 3:7a284390b0ce 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 Conv
emilmont 1:fdd22bb7aa52 49 * @{
emilmont 1:fdd22bb7aa52 50 */
emilmont 1:fdd22bb7aa52 51
emilmont 1:fdd22bb7aa52 52 /**
emilmont 1:fdd22bb7aa52 53 * @brief Convolution of Q31 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 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 an internal 64-bit accumulator.
emilmont 1:fdd22bb7aa52 66 * The accumulator has a 2.62 format and maintains full precision of the intermediate multiplication results but provides only a single guard bit.
emilmont 1:fdd22bb7aa52 67 * There is no saturation on intermediate additions.
emilmont 1:fdd22bb7aa52 68 * Thus, if the accumulator overflows it wraps around and distorts the result.
emilmont 1:fdd22bb7aa52 69 * The input signals should be scaled down to avoid intermediate overflows.
emilmont 1:fdd22bb7aa52 70 * Scale down the inputs by log2(min(srcALen, srcBLen)) (log2 is read as log to the base 2) times to avoid overflows,
emilmont 1:fdd22bb7aa52 71 * as maximum of min(srcALen, srcBLen) number of additions are carried internally.
emilmont 1:fdd22bb7aa52 72 * The 2.62 accumulator is right shifted by 31 bits and saturated to 1.31 format to yield the final result.
emilmont 1:fdd22bb7aa52 73 *
emilmont 1:fdd22bb7aa52 74 * \par
emilmont 1:fdd22bb7aa52 75 * See <code>arm_conv_fast_q31()</code> for a faster but less precise implementation of this function for Cortex-M3 and Cortex-M4.
emilmont 1:fdd22bb7aa52 76 */
emilmont 1:fdd22bb7aa52 77
emilmont 1:fdd22bb7aa52 78 void arm_conv_q31(
emilmont 1:fdd22bb7aa52 79 q31_t * pSrcA,
emilmont 1:fdd22bb7aa52 80 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 81 q31_t * pSrcB,
emilmont 1:fdd22bb7aa52 82 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 83 q31_t * pDst)
emilmont 1:fdd22bb7aa52 84 {
emilmont 1:fdd22bb7aa52 85
emilmont 1:fdd22bb7aa52 86
mbed_official 3:7a284390b0ce 87 #ifndef ARM_MATH_CM0_FAMILY
emilmont 1:fdd22bb7aa52 88
emilmont 1:fdd22bb7aa52 89 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 90
emilmont 1:fdd22bb7aa52 91 q31_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 92 q31_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 93 q31_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 94 q31_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 95 q31_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 96 q31_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 97 q63_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 98 q63_t acc0, acc1, acc2; /* Accumulator */
emilmont 1:fdd22bb7aa52 99 q31_t x0, x1, x2, c0; /* Temporary variables to hold state and coefficient values */
emilmont 1:fdd22bb7aa52 100 uint32_t j, k, count, blkCnt, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 101
emilmont 1:fdd22bb7aa52 102 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 103 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 104 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 105 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 106 {
emilmont 1:fdd22bb7aa52 107 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 108 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 109
emilmont 1:fdd22bb7aa52 110 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 111 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 112 }
emilmont 1:fdd22bb7aa52 113 else
emilmont 1:fdd22bb7aa52 114 {
emilmont 1:fdd22bb7aa52 115 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 116 pIn1 = (q31_t *) pSrcB;
emilmont 1:fdd22bb7aa52 117
emilmont 1:fdd22bb7aa52 118 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 119 pIn2 = (q31_t *) pSrcA;
emilmont 1:fdd22bb7aa52 120
emilmont 1:fdd22bb7aa52 121 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 122 j = srcBLen;
emilmont 1:fdd22bb7aa52 123 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 124 srcALen = j;
emilmont 1:fdd22bb7aa52 125 }
emilmont 1:fdd22bb7aa52 126
emilmont 1:fdd22bb7aa52 127 /* 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 128 /* The function is internally
emilmont 1:fdd22bb7aa52 129 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 130 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 131 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 132 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 133 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 134 * for every iteration. */
emilmont 1:fdd22bb7aa52 135
emilmont 1:fdd22bb7aa52 136 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 137 The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 138 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 139 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 140 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 141
emilmont 1:fdd22bb7aa52 142 /* --------------------------
emilmont 1:fdd22bb7aa52 143 * Initializations of stage1
emilmont 1:fdd22bb7aa52 144 * -------------------------*/
emilmont 1:fdd22bb7aa52 145
emilmont 1:fdd22bb7aa52 146 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 147 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 148 * ....
emilmont 1:fdd22bb7aa52 149 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 150 */
emilmont 1:fdd22bb7aa52 151
emilmont 1:fdd22bb7aa52 152 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 153 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 154 count = 1u;
emilmont 1:fdd22bb7aa52 155
emilmont 1:fdd22bb7aa52 156 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 157 px = pIn1;
emilmont 1:fdd22bb7aa52 158
emilmont 1:fdd22bb7aa52 159 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 160 py = pIn2;
emilmont 1:fdd22bb7aa52 161
emilmont 1:fdd22bb7aa52 162
emilmont 1:fdd22bb7aa52 163 /* ------------------------
emilmont 1:fdd22bb7aa52 164 * Stage1 process
emilmont 1:fdd22bb7aa52 165 * ----------------------*/
emilmont 1:fdd22bb7aa52 166
emilmont 1:fdd22bb7aa52 167 /* The first stage starts here */
emilmont 1:fdd22bb7aa52 168 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 169 {
emilmont 1:fdd22bb7aa52 170 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 171 sum = 0;
emilmont 1:fdd22bb7aa52 172
emilmont 1:fdd22bb7aa52 173 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 174 k = count >> 2u;
emilmont 1:fdd22bb7aa52 175
emilmont 1:fdd22bb7aa52 176 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 177 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 178 while(k > 0u)
emilmont 1:fdd22bb7aa52 179 {
emilmont 1:fdd22bb7aa52 180 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 181 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 182 /* x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 183 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 184 /* x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 185 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 186 /* x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 187 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 188
emilmont 1:fdd22bb7aa52 189 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 190 k--;
emilmont 1:fdd22bb7aa52 191 }
emilmont 1:fdd22bb7aa52 192
emilmont 1:fdd22bb7aa52 193 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 194 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 195 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 196
emilmont 1:fdd22bb7aa52 197 while(k > 0u)
emilmont 1:fdd22bb7aa52 198 {
emilmont 1:fdd22bb7aa52 199 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 200 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 201
emilmont 1:fdd22bb7aa52 202 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 203 k--;
emilmont 1:fdd22bb7aa52 204 }
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 207 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 208
emilmont 1:fdd22bb7aa52 209 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 210 py = pIn2 + count;
emilmont 1:fdd22bb7aa52 211 px = pIn1;
emilmont 1:fdd22bb7aa52 212
emilmont 1:fdd22bb7aa52 213 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 214 count++;
emilmont 1:fdd22bb7aa52 215
emilmont 1:fdd22bb7aa52 216 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 217 blockSize1--;
emilmont 1:fdd22bb7aa52 218 }
emilmont 1:fdd22bb7aa52 219
emilmont 1:fdd22bb7aa52 220 /* --------------------------
emilmont 1:fdd22bb7aa52 221 * Initializations of stage2
emilmont 1:fdd22bb7aa52 222 * ------------------------*/
emilmont 1:fdd22bb7aa52 223
emilmont 1:fdd22bb7aa52 224 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 225 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 226 * ....
emilmont 1:fdd22bb7aa52 227 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 228 */
emilmont 1:fdd22bb7aa52 229
emilmont 1:fdd22bb7aa52 230 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 231 px = pIn1;
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 234 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 235 py = pSrc2;
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 238 count = 0u;
emilmont 1:fdd22bb7aa52 239
emilmont 1:fdd22bb7aa52 240 /* -------------------
emilmont 1:fdd22bb7aa52 241 * Stage2 process
emilmont 1:fdd22bb7aa52 242 * ------------------*/
emilmont 1:fdd22bb7aa52 243
emilmont 1:fdd22bb7aa52 244 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 245 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 246 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 247 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 248 {
emilmont 1:fdd22bb7aa52 249 /* Loop unroll by 3 */
emilmont 1:fdd22bb7aa52 250 blkCnt = blockSize2 / 3;
emilmont 1:fdd22bb7aa52 251
emilmont 1:fdd22bb7aa52 252 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 253 {
emilmont 1:fdd22bb7aa52 254 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 255 acc0 = 0;
emilmont 1:fdd22bb7aa52 256 acc1 = 0;
emilmont 1:fdd22bb7aa52 257 acc2 = 0;
emilmont 1:fdd22bb7aa52 258
emilmont 1:fdd22bb7aa52 259 /* read x[0], x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 260 x0 = *(px++);
emilmont 1:fdd22bb7aa52 261 x1 = *(px++);
emilmont 1:fdd22bb7aa52 262
emilmont 1:fdd22bb7aa52 263 /* Apply loop unrolling and compute 3 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 264 k = srcBLen / 3;
emilmont 1:fdd22bb7aa52 265
emilmont 1:fdd22bb7aa52 266 /* First part of the processing with loop unrolling. Compute 3 MACs at a time.
emilmont 1:fdd22bb7aa52 267 ** a second loop below computes MACs for the remaining 1 to 2 samples. */
emilmont 1:fdd22bb7aa52 268 do
emilmont 1:fdd22bb7aa52 269 {
emilmont 1:fdd22bb7aa52 270 /* Read y[srcBLen - 1] sample */
emilmont 1:fdd22bb7aa52 271 c0 = *(py);
emilmont 1:fdd22bb7aa52 272
emilmont 1:fdd22bb7aa52 273 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 274 x2 = *(px);
emilmont 1:fdd22bb7aa52 275
emilmont 1:fdd22bb7aa52 276 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 277 /* acc0 += x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 278 acc0 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 279 /* acc1 += x[1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 280 acc1 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 281 /* acc2 += x[2] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 282 acc2 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 283
emilmont 1:fdd22bb7aa52 284 /* Read y[srcBLen - 2] sample */
emilmont 1:fdd22bb7aa52 285 c0 = *(py - 1u);
emilmont 1:fdd22bb7aa52 286
emilmont 1:fdd22bb7aa52 287 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 288 x0 = *(px + 1u);
emilmont 1:fdd22bb7aa52 289
emilmont 1:fdd22bb7aa52 290 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 291 /* acc0 += x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 292 acc0 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 293 /* acc1 += x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 294 acc1 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 295 /* acc2 += x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 296 acc2 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 297
emilmont 1:fdd22bb7aa52 298 /* Read y[srcBLen - 3] sample */
emilmont 1:fdd22bb7aa52 299 c0 = *(py - 2u);
emilmont 1:fdd22bb7aa52 300
emilmont 1:fdd22bb7aa52 301 /* Read x[5] sample */
emilmont 1:fdd22bb7aa52 302 x1 = *(px + 2u);
emilmont 1:fdd22bb7aa52 303
emilmont 1:fdd22bb7aa52 304 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 305 /* acc0 += x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 306 acc0 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 307 /* acc1 += x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 308 acc1 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 309 /* acc2 += x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 310 acc2 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 /* update scratch pointers */
emilmont 1:fdd22bb7aa52 313 px += 3u;
emilmont 1:fdd22bb7aa52 314 py -= 3u;
emilmont 1:fdd22bb7aa52 315
emilmont 1:fdd22bb7aa52 316 } while(--k);
emilmont 1:fdd22bb7aa52 317
emilmont 1:fdd22bb7aa52 318 /* If the srcBLen is not a multiple of 3, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 319 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 320 k = srcBLen - (3 * (srcBLen / 3));
emilmont 1:fdd22bb7aa52 321
emilmont 1:fdd22bb7aa52 322 while(k > 0u)
emilmont 1:fdd22bb7aa52 323 {
emilmont 1:fdd22bb7aa52 324 /* Read y[srcBLen - 5] sample */
emilmont 1:fdd22bb7aa52 325 c0 = *(py--);
emilmont 1:fdd22bb7aa52 326
emilmont 1:fdd22bb7aa52 327 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 328 x2 = *(px++);
emilmont 1:fdd22bb7aa52 329
emilmont 1:fdd22bb7aa52 330 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 331 /* acc0 += x[4] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 332 acc0 += ((q63_t) x0 * c0);
emilmont 1:fdd22bb7aa52 333 /* acc1 += x[5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 334 acc1 += ((q63_t) x1 * c0);
emilmont 1:fdd22bb7aa52 335 /* acc2 += x[6] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 336 acc2 += ((q63_t) x2 * c0);
emilmont 1:fdd22bb7aa52 337
emilmont 1:fdd22bb7aa52 338 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 339 x0 = x1;
emilmont 1:fdd22bb7aa52 340 x1 = x2;
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 343 k--;
emilmont 1:fdd22bb7aa52 344 }
emilmont 1:fdd22bb7aa52 345
emilmont 1:fdd22bb7aa52 346 /* Store the results in the accumulators in the destination buffer. */
emilmont 1:fdd22bb7aa52 347 *pOut++ = (q31_t) (acc0 >> 31);
emilmont 1:fdd22bb7aa52 348 *pOut++ = (q31_t) (acc1 >> 31);
emilmont 1:fdd22bb7aa52 349 *pOut++ = (q31_t) (acc2 >> 31);
emilmont 1:fdd22bb7aa52 350
emilmont 1:fdd22bb7aa52 351 /* Increment the pointer pIn1 index, count by 3 */
emilmont 1:fdd22bb7aa52 352 count += 3u;
emilmont 1:fdd22bb7aa52 353
emilmont 1:fdd22bb7aa52 354 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 355 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 356 py = pSrc2;
emilmont 1:fdd22bb7aa52 357
emilmont 1:fdd22bb7aa52 358 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 359 blkCnt--;
emilmont 1:fdd22bb7aa52 360 }
emilmont 1:fdd22bb7aa52 361
emilmont 1:fdd22bb7aa52 362 /* If the blockSize2 is not a multiple of 3, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 363 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 364 blkCnt = blockSize2 - 3 * (blockSize2 / 3);
emilmont 1:fdd22bb7aa52 365
emilmont 1:fdd22bb7aa52 366 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 367 {
emilmont 1:fdd22bb7aa52 368 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 369 sum = 0;
emilmont 1:fdd22bb7aa52 370
emilmont 1:fdd22bb7aa52 371 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 372 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 373
emilmont 1:fdd22bb7aa52 374 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 375 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 376 while(k > 0u)
emilmont 1:fdd22bb7aa52 377 {
emilmont 1:fdd22bb7aa52 378 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 379 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 380 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 381 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 382 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 383
emilmont 1:fdd22bb7aa52 384 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 385 k--;
emilmont 1:fdd22bb7aa52 386 }
emilmont 1:fdd22bb7aa52 387
emilmont 1:fdd22bb7aa52 388 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 389 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 390 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 391
emilmont 1:fdd22bb7aa52 392 while(k > 0u)
emilmont 1:fdd22bb7aa52 393 {
emilmont 1:fdd22bb7aa52 394 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 395 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 396
emilmont 1:fdd22bb7aa52 397 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 398 k--;
emilmont 1:fdd22bb7aa52 399 }
emilmont 1:fdd22bb7aa52 400
emilmont 1:fdd22bb7aa52 401 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 402 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 405 count++;
emilmont 1:fdd22bb7aa52 406
emilmont 1:fdd22bb7aa52 407 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 408 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 409 py = pSrc2;
emilmont 1:fdd22bb7aa52 410
emilmont 1:fdd22bb7aa52 411 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 412 blkCnt--;
emilmont 1:fdd22bb7aa52 413 }
emilmont 1:fdd22bb7aa52 414 }
emilmont 1:fdd22bb7aa52 415 else
emilmont 1:fdd22bb7aa52 416 {
emilmont 1:fdd22bb7aa52 417 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 418 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 419 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 420
emilmont 1:fdd22bb7aa52 421 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 422 {
emilmont 1:fdd22bb7aa52 423 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 424 sum = 0;
emilmont 1:fdd22bb7aa52 425
emilmont 1:fdd22bb7aa52 426 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 427 k = srcBLen;
emilmont 1:fdd22bb7aa52 428
emilmont 1:fdd22bb7aa52 429 while(k > 0u)
emilmont 1:fdd22bb7aa52 430 {
emilmont 1:fdd22bb7aa52 431 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 432 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 433
emilmont 1:fdd22bb7aa52 434 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 435 k--;
emilmont 1:fdd22bb7aa52 436 }
emilmont 1:fdd22bb7aa52 437
emilmont 1:fdd22bb7aa52 438 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 439 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 440
emilmont 1:fdd22bb7aa52 441 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 442 count++;
emilmont 1:fdd22bb7aa52 443
emilmont 1:fdd22bb7aa52 444 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 445 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 446 py = pSrc2;
emilmont 1:fdd22bb7aa52 447
emilmont 1:fdd22bb7aa52 448 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 449 blkCnt--;
emilmont 1:fdd22bb7aa52 450 }
emilmont 1:fdd22bb7aa52 451 }
emilmont 1:fdd22bb7aa52 452
emilmont 1:fdd22bb7aa52 453
emilmont 1:fdd22bb7aa52 454 /* --------------------------
emilmont 1:fdd22bb7aa52 455 * Initializations of stage3
emilmont 1:fdd22bb7aa52 456 * -------------------------*/
emilmont 1:fdd22bb7aa52 457
emilmont 1:fdd22bb7aa52 458 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 459 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 460 * ....
emilmont 1:fdd22bb7aa52 461 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 462 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 463 */
emilmont 1:fdd22bb7aa52 464
emilmont 1:fdd22bb7aa52 465 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 466 The blockSize3 variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 467
emilmont 1:fdd22bb7aa52 468 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 469 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 470 px = pSrc1;
emilmont 1:fdd22bb7aa52 471
emilmont 1:fdd22bb7aa52 472 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 473 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 474 py = pSrc2;
emilmont 1:fdd22bb7aa52 475
emilmont 1:fdd22bb7aa52 476 /* -------------------
emilmont 1:fdd22bb7aa52 477 * Stage3 process
emilmont 1:fdd22bb7aa52 478 * ------------------*/
emilmont 1:fdd22bb7aa52 479
emilmont 1:fdd22bb7aa52 480 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 481 {
emilmont 1:fdd22bb7aa52 482 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 483 sum = 0;
emilmont 1:fdd22bb7aa52 484
emilmont 1:fdd22bb7aa52 485 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 486 k = blockSize3 >> 2u;
emilmont 1:fdd22bb7aa52 487
emilmont 1:fdd22bb7aa52 488 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 489 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 490 while(k > 0u)
emilmont 1:fdd22bb7aa52 491 {
emilmont 1:fdd22bb7aa52 492 /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 493 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 494 /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 495 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 496 /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 497 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 498 /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 499 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 500
emilmont 1:fdd22bb7aa52 501 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 502 k--;
emilmont 1:fdd22bb7aa52 503 }
emilmont 1:fdd22bb7aa52 504
emilmont 1:fdd22bb7aa52 505 /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 506 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 507 k = blockSize3 % 0x4u;
emilmont 1:fdd22bb7aa52 508
emilmont 1:fdd22bb7aa52 509 while(k > 0u)
emilmont 1:fdd22bb7aa52 510 {
emilmont 1:fdd22bb7aa52 511 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 512 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 513
emilmont 1:fdd22bb7aa52 514 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 515 k--;
emilmont 1:fdd22bb7aa52 516 }
emilmont 1:fdd22bb7aa52 517
emilmont 1:fdd22bb7aa52 518 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 519 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 520
emilmont 1:fdd22bb7aa52 521 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 522 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 523 py = pSrc2;
emilmont 1:fdd22bb7aa52 524
emilmont 1:fdd22bb7aa52 525 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 526 blockSize3--;
emilmont 1:fdd22bb7aa52 527 }
emilmont 1:fdd22bb7aa52 528
emilmont 1:fdd22bb7aa52 529 #else
emilmont 1:fdd22bb7aa52 530
emilmont 1:fdd22bb7aa52 531 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 532
emilmont 1:fdd22bb7aa52 533 q31_t *pIn1 = pSrcA; /* input pointer */
emilmont 1:fdd22bb7aa52 534 q31_t *pIn2 = pSrcB; /* coefficient pointer */
emilmont 1:fdd22bb7aa52 535 q63_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 536 uint32_t i, j; /* loop counter */
emilmont 1:fdd22bb7aa52 537
emilmont 1:fdd22bb7aa52 538 /* Loop to calculate output of convolution for output length number of times */
emilmont 1:fdd22bb7aa52 539 for (i = 0; i < (srcALen + srcBLen - 1); i++)
emilmont 1:fdd22bb7aa52 540 {
emilmont 1:fdd22bb7aa52 541 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 542 sum = 0;
emilmont 1:fdd22bb7aa52 543
emilmont 1:fdd22bb7aa52 544 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 545 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 546 {
emilmont 1:fdd22bb7aa52 547 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 548 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 549 {
emilmont 1:fdd22bb7aa52 550 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 551 sum += ((q63_t) pIn1[j] * (pIn2[i - j]));
emilmont 1:fdd22bb7aa52 552 }
emilmont 1:fdd22bb7aa52 553 }
emilmont 1:fdd22bb7aa52 554
emilmont 1:fdd22bb7aa52 555 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 556 pDst[i] = (q31_t) (sum >> 31u);
emilmont 1:fdd22bb7aa52 557 }
emilmont 1:fdd22bb7aa52 558
mbed_official 3:7a284390b0ce 559 #endif /* #ifndef ARM_MATH_CM0_FAMILY */
emilmont 1:fdd22bb7aa52 560
emilmont 1:fdd22bb7aa52 561 }
emilmont 1:fdd22bb7aa52 562
emilmont 1:fdd22bb7aa52 563 /**
emilmont 1:fdd22bb7aa52 564 * @} end of Conv group
emilmont 1:fdd22bb7aa52 565 */