mbed official / mbed-dsp

CMSIS DSP library

Dependents:   KL25Z_FFT_Demo Hat_Board_v5_1 KL25Z_FFT_Demo_tony KL25Z_FFT_Demo_tony ... more

Fork of mbed-dsp by mbed official

cmsis_dsp/FilteringFunctions/arm_conv_fast_q31.c@3:7a284390b0ce, 2013-11-08 (annotated)

Committer:
mbed_official
Date:
Fri Nov 08 13:45:10 2013 +0000
Revision:
3:7a284390b0ce
Parent:
2:da51fb522205 
Synchronized with git revision e69956aba2f68a2a26ac26b051f8d349deaa1ce8

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emilmont 1:fdd22bb7aa52 1 /* ----------------------------------------------------------------------
mbed_official 3:7a284390b0ce 2 * Copyright (C) 2010-2013 ARM Limited. All rights reserved.
emilmont 1:fdd22bb7aa52 3 *
mbed_official 3:7a284390b0ce 4 * $Date: 17. January 2013
mbed_official 3:7a284390b0ce 5 * $Revision: V1.4.1
emilmont 1:fdd22bb7aa52 6 *
emilmont 2:da51fb522205 7 * Project: CMSIS DSP Library
emilmont 2:da51fb522205 8 * Title: arm_conv_fast_q31.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Q31 Convolution (fast version).
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3
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 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 54 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[out] *pDst points to the location where the output result is written. Length srcALen+srcBLen-1.
emilmont 1:fdd22bb7aa52 58 * @return none.
emilmont 1:fdd22bb7aa52 59 *
emilmont 1:fdd22bb7aa52 60 * @details
emilmont 1:fdd22bb7aa52 61 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 62 *
emilmont 1:fdd22bb7aa52 63 * \par
emilmont 1:fdd22bb7aa52 64 * This function is optimized for speed at the expense of fixed-point precision and overflow protection.
emilmont 1:fdd22bb7aa52 65 * The result of each 1.31 x 1.31 multiplication is truncated to 2.30 format.
emilmont 1:fdd22bb7aa52 66 * These intermediate results are accumulated in a 32-bit register in 2.30 format.
emilmont 1:fdd22bb7aa52 67 * Finally, the accumulator is saturated and converted to a 1.31 result.
emilmont 1:fdd22bb7aa52 68 *
emilmont 1:fdd22bb7aa52 69 * \par
emilmont 1:fdd22bb7aa52 70 * The fast version has the same overflow behavior as the standard version but provides less precision since it discards the low 32 bits of each multiplication result.
emilmont 1:fdd22bb7aa52 71 * In order to avoid overflows completely the input signals must be scaled down.
emilmont 1:fdd22bb7aa52 72 * 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 73 * as maximum of min(srcALen, srcBLen) number of additions are carried internally.
emilmont 1:fdd22bb7aa52 74 *
emilmont 1:fdd22bb7aa52 75 * \par
emilmont 1:fdd22bb7aa52 76 * See <code>arm_conv_q31()</code> for a slower implementation of this function which uses 64-bit accumulation to provide higher precision.
emilmont 1:fdd22bb7aa52 77 */
emilmont 1:fdd22bb7aa52 78
emilmont 1:fdd22bb7aa52 79 void arm_conv_fast_q31(
emilmont 1:fdd22bb7aa52 80 q31_t * pSrcA,
emilmont 1:fdd22bb7aa52 81 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 82 q31_t * pSrcB,
emilmont 1:fdd22bb7aa52 83 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 84 q31_t * pDst)
emilmont 1:fdd22bb7aa52 85 {
emilmont 1:fdd22bb7aa52 86 q31_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 87 q31_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 88 q31_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 89 q31_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 90 q31_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 91 q31_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 92 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 93 q31_t x0, x1, x2, x3, c0; /* Temporary variables to hold state and coefficient values */
emilmont 1:fdd22bb7aa52 94 uint32_t j, k, count, blkCnt, blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 95
emilmont 1:fdd22bb7aa52 96 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 97 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 98 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 99 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 100 {
emilmont 1:fdd22bb7aa52 101 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 102 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 103
emilmont 1:fdd22bb7aa52 104 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 105 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 106 }
emilmont 1:fdd22bb7aa52 107 else
emilmont 1:fdd22bb7aa52 108 {
emilmont 1:fdd22bb7aa52 109 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 110 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 111
emilmont 1:fdd22bb7aa52 112 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 113 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 114
emilmont 1:fdd22bb7aa52 115 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 116 j = srcBLen;
emilmont 1:fdd22bb7aa52 117 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 118 srcALen = j;
emilmont 1:fdd22bb7aa52 119 }
emilmont 1:fdd22bb7aa52 120
emilmont 1:fdd22bb7aa52 121 /* 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 122 /* The function is internally
emilmont 1:fdd22bb7aa52 123 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 124 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 125 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 126 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 127 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 128 * for every iteration. */
emilmont 1:fdd22bb7aa52 129
emilmont 1:fdd22bb7aa52 130 /* The algorithm is implemented in three stages.
emilmont 1:fdd22bb7aa52 131 The loop counters of each stage is initiated here. */
emilmont 1:fdd22bb7aa52 132 blockSize1 = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 133 blockSize2 = srcALen - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 134 blockSize3 = blockSize1;
emilmont 1:fdd22bb7aa52 135
emilmont 1:fdd22bb7aa52 136 /* --------------------------
emilmont 1:fdd22bb7aa52 137 * Initializations of stage1
emilmont 1:fdd22bb7aa52 138 * -------------------------*/
emilmont 1:fdd22bb7aa52 139
emilmont 1:fdd22bb7aa52 140 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 141 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 142 * ....
emilmont 1:fdd22bb7aa52 143 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 144 */
emilmont 1:fdd22bb7aa52 145
emilmont 1:fdd22bb7aa52 146 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 147 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 148 count = 1u;
emilmont 1:fdd22bb7aa52 149
emilmont 1:fdd22bb7aa52 150 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 151 px = pIn1;
emilmont 1:fdd22bb7aa52 152
emilmont 1:fdd22bb7aa52 153 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 154 py = pIn2;
emilmont 1:fdd22bb7aa52 155
emilmont 1:fdd22bb7aa52 156
emilmont 1:fdd22bb7aa52 157 /* ------------------------
emilmont 1:fdd22bb7aa52 158 * Stage1 process
emilmont 1:fdd22bb7aa52 159 * ----------------------*/
emilmont 1:fdd22bb7aa52 160
emilmont 1:fdd22bb7aa52 161 /* The first stage starts here */
emilmont 1:fdd22bb7aa52 162 while(blockSize1 > 0u)
emilmont 1:fdd22bb7aa52 163 {
emilmont 1:fdd22bb7aa52 164 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 165 sum = 0;
emilmont 1:fdd22bb7aa52 166
emilmont 1:fdd22bb7aa52 167 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 168 k = count >> 2u;
emilmont 1:fdd22bb7aa52 169
emilmont 1:fdd22bb7aa52 170 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 171 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 172 while(k > 0u)
emilmont 1:fdd22bb7aa52 173 {
emilmont 1:fdd22bb7aa52 174 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 175 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 176 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 177
emilmont 1:fdd22bb7aa52 178 /* x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 179 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 180 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 181
emilmont 1:fdd22bb7aa52 182 /* x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 183 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 184 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 185
emilmont 1:fdd22bb7aa52 186 /* x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 187 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 188 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 189
emilmont 1:fdd22bb7aa52 190 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 191 k--;
emilmont 1:fdd22bb7aa52 192 }
emilmont 1:fdd22bb7aa52 193
emilmont 1:fdd22bb7aa52 194 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 195 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 196 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 197
emilmont 1:fdd22bb7aa52 198 while(k > 0u)
emilmont 1:fdd22bb7aa52 199 {
emilmont 1:fdd22bb7aa52 200 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 201 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 202 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 203
emilmont 1:fdd22bb7aa52 204 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 205 k--;
emilmont 1:fdd22bb7aa52 206 }
emilmont 1:fdd22bb7aa52 207
emilmont 1:fdd22bb7aa52 208 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 209 *pOut++ = sum << 1;
emilmont 1:fdd22bb7aa52 210
emilmont 1:fdd22bb7aa52 211 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 212 py = pIn2 + count;
emilmont 1:fdd22bb7aa52 213 px = pIn1;
emilmont 1:fdd22bb7aa52 214
emilmont 1:fdd22bb7aa52 215 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 216 count++;
emilmont 1:fdd22bb7aa52 217
emilmont 1:fdd22bb7aa52 218 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 219 blockSize1--;
emilmont 1:fdd22bb7aa52 220 }
emilmont 1:fdd22bb7aa52 221
emilmont 1:fdd22bb7aa52 222 /* --------------------------
emilmont 1:fdd22bb7aa52 223 * Initializations of stage2
emilmont 1:fdd22bb7aa52 224 * ------------------------*/
emilmont 1:fdd22bb7aa52 225
emilmont 1:fdd22bb7aa52 226 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 227 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 228 * ....
emilmont 1:fdd22bb7aa52 229 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 230 */
emilmont 1:fdd22bb7aa52 231
emilmont 1:fdd22bb7aa52 232 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 233 px = pIn1;
emilmont 1:fdd22bb7aa52 234
emilmont 1:fdd22bb7aa52 235 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 236 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 237 py = pSrc2;
emilmont 1:fdd22bb7aa52 238
emilmont 1:fdd22bb7aa52 239 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 240 count = 0u;
emilmont 1:fdd22bb7aa52 241
emilmont 1:fdd22bb7aa52 242 /* -------------------
emilmont 1:fdd22bb7aa52 243 * Stage2 process
emilmont 1:fdd22bb7aa52 244 * ------------------*/
emilmont 1:fdd22bb7aa52 245
emilmont 1:fdd22bb7aa52 246 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 247 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 248 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 249 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 250 {
emilmont 1:fdd22bb7aa52 251 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 252 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 253
emilmont 1:fdd22bb7aa52 254 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 255 {
emilmont 1:fdd22bb7aa52 256 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 257 acc0 = 0;
emilmont 1:fdd22bb7aa52 258 acc1 = 0;
emilmont 1:fdd22bb7aa52 259 acc2 = 0;
emilmont 1:fdd22bb7aa52 260 acc3 = 0;
emilmont 1:fdd22bb7aa52 261
emilmont 1:fdd22bb7aa52 262 /* read x[0], x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 263 x0 = *(px++);
emilmont 1:fdd22bb7aa52 264 x1 = *(px++);
emilmont 1:fdd22bb7aa52 265 x2 = *(px++);
emilmont 1:fdd22bb7aa52 266
emilmont 1:fdd22bb7aa52 267 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 268 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 269
emilmont 1:fdd22bb7aa52 270 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 271 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 272 do
emilmont 1:fdd22bb7aa52 273 {
emilmont 1:fdd22bb7aa52 274 /* Read y[srcBLen - 1] sample */
emilmont 1:fdd22bb7aa52 275 c0 = *(py--);
emilmont 1:fdd22bb7aa52 276
emilmont 1:fdd22bb7aa52 277 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 278 x3 = *(px++);
emilmont 1:fdd22bb7aa52 279
emilmont 1:fdd22bb7aa52 280 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 281 /* acc0 += x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 282 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x0 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 283
emilmont 1:fdd22bb7aa52 284 /* acc1 += x[1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 285 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x1 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 286
emilmont 1:fdd22bb7aa52 287 /* acc2 += x[2] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 288 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x2 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 289
emilmont 1:fdd22bb7aa52 290 /* acc3 += x[3] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 291 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x3 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 292
emilmont 1:fdd22bb7aa52 293 /* Read y[srcBLen - 2] sample */
emilmont 1:fdd22bb7aa52 294 c0 = *(py--);
emilmont 1:fdd22bb7aa52 295
emilmont 1:fdd22bb7aa52 296 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 297 x0 = *(px++);
emilmont 1:fdd22bb7aa52 298
emilmont 1:fdd22bb7aa52 299 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 300 /* acc0 += x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 301 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x1 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 302 /* acc1 += x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 303 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x2 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 304 /* acc2 += x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 305 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x3 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 306 /* acc3 += x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 307 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x0 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 308
emilmont 1:fdd22bb7aa52 309 /* Read y[srcBLen - 3] sample */
emilmont 1:fdd22bb7aa52 310 c0 = *(py--);
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 /* Read x[5] sample */
emilmont 1:fdd22bb7aa52 313 x1 = *(px++);
emilmont 1:fdd22bb7aa52 314
emilmont 1:fdd22bb7aa52 315 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 316 /* acc0 += x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 317 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x2 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 318 /* acc1 += x[3] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 319 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x3 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 320 /* acc2 += x[4] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 321 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x0 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 322 /* acc3 += x[5] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 323 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x1 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 324
emilmont 1:fdd22bb7aa52 325 /* Read y[srcBLen - 4] sample */
emilmont 1:fdd22bb7aa52 326 c0 = *(py--);
emilmont 1:fdd22bb7aa52 327
emilmont 1:fdd22bb7aa52 328 /* Read x[6] sample */
emilmont 1:fdd22bb7aa52 329 x2 = *(px++);
emilmont 1:fdd22bb7aa52 330
emilmont 1:fdd22bb7aa52 331 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 332 /* acc0 += x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 333 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x3 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 334 /* acc1 += x[4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 335 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x0 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 336 /* acc2 += x[5] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 337 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x1 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 338 /* acc3 += x[6] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 339 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x2 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 340
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 } while(--k);
emilmont 1:fdd22bb7aa52 343
emilmont 1:fdd22bb7aa52 344 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 345 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 346 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 347
emilmont 1:fdd22bb7aa52 348 while(k > 0u)
emilmont 1:fdd22bb7aa52 349 {
emilmont 1:fdd22bb7aa52 350 /* Read y[srcBLen - 5] sample */
emilmont 1:fdd22bb7aa52 351 c0 = *(py--);
emilmont 1:fdd22bb7aa52 352
emilmont 1:fdd22bb7aa52 353 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 354 x3 = *(px++);
emilmont 1:fdd22bb7aa52 355
emilmont 1:fdd22bb7aa52 356 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 357 /* acc0 += x[4] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 358 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x0 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 359 /* acc1 += x[5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 360 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x1 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 361 /* acc2 += x[6] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 362 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x2 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 363 /* acc3 += x[7] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 364 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x3 * c0)) >> 32);
emilmont 1:fdd22bb7aa52 365
emilmont 1:fdd22bb7aa52 366 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 367 x0 = x1;
emilmont 1:fdd22bb7aa52 368 x1 = x2;
emilmont 1:fdd22bb7aa52 369 x2 = x3;
emilmont 1:fdd22bb7aa52 370
emilmont 1:fdd22bb7aa52 371 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 372 k--;
emilmont 1:fdd22bb7aa52 373 }
emilmont 1:fdd22bb7aa52 374
emilmont 1:fdd22bb7aa52 375 /* Store the results in the accumulators in the destination buffer. */
emilmont 1:fdd22bb7aa52 376 *pOut++ = (q31_t) (acc0 << 1);
emilmont 1:fdd22bb7aa52 377 *pOut++ = (q31_t) (acc1 << 1);
emilmont 1:fdd22bb7aa52 378 *pOut++ = (q31_t) (acc2 << 1);
emilmont 1:fdd22bb7aa52 379 *pOut++ = (q31_t) (acc3 << 1);
emilmont 1:fdd22bb7aa52 380
emilmont 1:fdd22bb7aa52 381 /* Increment the pointer pIn1 index, count by 4 */
emilmont 1:fdd22bb7aa52 382 count += 4u;
emilmont 1:fdd22bb7aa52 383
emilmont 1:fdd22bb7aa52 384 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 385 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 386 py = pSrc2;
emilmont 1:fdd22bb7aa52 387
emilmont 1:fdd22bb7aa52 388 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 389 blkCnt--;
emilmont 1:fdd22bb7aa52 390 }
emilmont 1:fdd22bb7aa52 391
emilmont 1:fdd22bb7aa52 392 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 393 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 394 blkCnt = blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 395
emilmont 1:fdd22bb7aa52 396 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 397 {
emilmont 1:fdd22bb7aa52 398 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 399 sum = 0;
emilmont 1:fdd22bb7aa52 400
emilmont 1:fdd22bb7aa52 401 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 402 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 405 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 406 while(k > 0u)
emilmont 1:fdd22bb7aa52 407 {
emilmont 1:fdd22bb7aa52 408 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 409 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 410 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 411 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 412 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 413 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 414 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 415 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 416 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 417
emilmont 1:fdd22bb7aa52 418 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 419 k--;
emilmont 1:fdd22bb7aa52 420 }
emilmont 1:fdd22bb7aa52 421
emilmont 1:fdd22bb7aa52 422 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 423 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 424 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 425
emilmont 1:fdd22bb7aa52 426 while(k > 0u)
emilmont 1:fdd22bb7aa52 427 {
emilmont 1:fdd22bb7aa52 428 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 429 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 430 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 431
emilmont 1:fdd22bb7aa52 432 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 433 k--;
emilmont 1:fdd22bb7aa52 434 }
emilmont 1:fdd22bb7aa52 435
emilmont 1:fdd22bb7aa52 436 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 437 *pOut++ = sum << 1;
emilmont 1:fdd22bb7aa52 438
emilmont 1:fdd22bb7aa52 439 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 440 count++;
emilmont 1:fdd22bb7aa52 441
emilmont 1:fdd22bb7aa52 442 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 443 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 444 py = pSrc2;
emilmont 1:fdd22bb7aa52 445
emilmont 1:fdd22bb7aa52 446 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 447 blkCnt--;
emilmont 1:fdd22bb7aa52 448 }
emilmont 1:fdd22bb7aa52 449 }
emilmont 1:fdd22bb7aa52 450 else
emilmont 1:fdd22bb7aa52 451 {
emilmont 1:fdd22bb7aa52 452 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 453 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 454 blkCnt = blockSize2;
emilmont 1:fdd22bb7aa52 455
emilmont 1:fdd22bb7aa52 456 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 457 {
emilmont 1:fdd22bb7aa52 458 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 459 sum = 0;
emilmont 1:fdd22bb7aa52 460
emilmont 1:fdd22bb7aa52 461 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 462 k = srcBLen;
emilmont 1:fdd22bb7aa52 463
emilmont 1:fdd22bb7aa52 464 while(k > 0u)
emilmont 1:fdd22bb7aa52 465 {
emilmont 1:fdd22bb7aa52 466 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 467 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 468 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 469
emilmont 1:fdd22bb7aa52 470 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 471 k--;
emilmont 1:fdd22bb7aa52 472 }
emilmont 1:fdd22bb7aa52 473
emilmont 1:fdd22bb7aa52 474 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 475 *pOut++ = sum << 1;
emilmont 1:fdd22bb7aa52 476
emilmont 1:fdd22bb7aa52 477 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 478 count++;
emilmont 1:fdd22bb7aa52 479
emilmont 1:fdd22bb7aa52 480 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 481 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 482 py = pSrc2;
emilmont 1:fdd22bb7aa52 483
emilmont 1:fdd22bb7aa52 484 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 485 blkCnt--;
emilmont 1:fdd22bb7aa52 486 }
emilmont 1:fdd22bb7aa52 487 }
emilmont 1:fdd22bb7aa52 488
emilmont 1:fdd22bb7aa52 489
emilmont 1:fdd22bb7aa52 490 /* --------------------------
emilmont 1:fdd22bb7aa52 491 * Initializations of stage3
emilmont 1:fdd22bb7aa52 492 * -------------------------*/
emilmont 1:fdd22bb7aa52 493
emilmont 1:fdd22bb7aa52 494 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 495 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 496 * ....
emilmont 1:fdd22bb7aa52 497 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 498 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 499 */
emilmont 1:fdd22bb7aa52 500
emilmont 1:fdd22bb7aa52 501 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 502 The blockSize3 variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 503
emilmont 1:fdd22bb7aa52 504 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 505 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 506 px = pSrc1;
emilmont 1:fdd22bb7aa52 507
emilmont 1:fdd22bb7aa52 508 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 509 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 510 py = pSrc2;
emilmont 1:fdd22bb7aa52 511
emilmont 1:fdd22bb7aa52 512 /* -------------------
emilmont 1:fdd22bb7aa52 513 * Stage3 process
emilmont 1:fdd22bb7aa52 514 * ------------------*/
emilmont 1:fdd22bb7aa52 515
emilmont 1:fdd22bb7aa52 516 while(blockSize3 > 0u)
emilmont 1:fdd22bb7aa52 517 {
emilmont 1:fdd22bb7aa52 518 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 519 sum = 0;
emilmont 1:fdd22bb7aa52 520
emilmont 1:fdd22bb7aa52 521 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 522 k = blockSize3 >> 2u;
emilmont 1:fdd22bb7aa52 523
emilmont 1:fdd22bb7aa52 524 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 525 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 526 while(k > 0u)
emilmont 1:fdd22bb7aa52 527 {
emilmont 1:fdd22bb7aa52 528 /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 529 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 530 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 533 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 534 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 535
emilmont 1:fdd22bb7aa52 536 /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 537 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 538 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 539
emilmont 1:fdd22bb7aa52 540 /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 541 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 542 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 543
emilmont 1:fdd22bb7aa52 544 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 545 k--;
emilmont 1:fdd22bb7aa52 546 }
emilmont 1:fdd22bb7aa52 547
emilmont 1:fdd22bb7aa52 548 /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 549 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 550 k = blockSize3 % 0x4u;
emilmont 1:fdd22bb7aa52 551
emilmont 1:fdd22bb7aa52 552 while(k > 0u)
emilmont 1:fdd22bb7aa52 553 {
emilmont 1:fdd22bb7aa52 554 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 555 sum = (q31_t) ((((q63_t) sum << 32) +
emilmont 1:fdd22bb7aa52 556 ((q63_t) * px++ * (*py--))) >> 32);
emilmont 1:fdd22bb7aa52 557
emilmont 1:fdd22bb7aa52 558 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 559 k--;
emilmont 1:fdd22bb7aa52 560 }
emilmont 1:fdd22bb7aa52 561
emilmont 1:fdd22bb7aa52 562 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 563 *pOut++ = sum << 1;
emilmont 1:fdd22bb7aa52 564
emilmont 1:fdd22bb7aa52 565 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 566 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 567 py = pSrc2;
emilmont 1:fdd22bb7aa52 568
emilmont 1:fdd22bb7aa52 569 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 570 blockSize3--;
emilmont 1:fdd22bb7aa52 571 }
emilmont 1:fdd22bb7aa52 572
emilmont 1:fdd22bb7aa52 573 }
emilmont 1:fdd22bb7aa52 574
emilmont 1:fdd22bb7aa52 575 /**
emilmont 1:fdd22bb7aa52 576 * @} end of Conv group
emilmont 1:fdd22bb7aa52 577 */