Eli Hughes / CMSIS_DSP_401

V4.0.1 of the ARM CMSIS DSP libraries. Note that arm_bitreversal2.s, arm_cfft_f32.c and arm_rfft_fast_f32.c had to be removed. arm_bitreversal2.s will not assemble with the online tools. So, the fast f32 FFT functions are not yet available. All the other FFT functions are available.

Dependents:   MPU9150_Example fir_f32 fir_f32 MPU9150_nucleo_noni2cdev ... more

FilteringFunctions/arm_conv_fast_q31.c@0:3d9c67d97d6f, 2014-07-28 (annotated)

Committer:
emh203
Date:
Mon Jul 28 15:03:15 2014 +0000
Revision:
0:3d9c67d97d6f
1st working commit.   Had to remove arm_bitreversal2.s     arm_cfft_f32.c and arm_rfft_fast_f32.c.    The .s will not assemble.      For now I removed these functions so we could at least have a library for the other functions.

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emh203 0:3d9c67d97d6f 1 /* ----------------------------------------------------------------------
emh203 0:3d9c67d97d6f 2 * Copyright (C) 2010-2014 ARM Limited. All rights reserved.
emh203 0:3d9c67d97d6f 3 *
emh203 0:3d9c67d97d6f 4 * $Date: 12. March 2014
emh203 0:3d9c67d97d6f 5 * $Revision: V1.4.3
emh203 0:3d9c67d97d6f 6 *
emh203 0:3d9c67d97d6f 7 * Project: CMSIS DSP Library
emh203 0:3d9c67d97d6f 8 * Title: arm_conv_fast_q31.c
emh203 0:3d9c67d97d6f 9 *
emh203 0:3d9c67d97d6f 10 * Description: Q31 Convolution (fast version).
emh203 0:3d9c67d97d6f 11 *
emh203 0:3d9c67d97d6f 12 * Target Processor: Cortex-M4/Cortex-M3
emh203 0:3d9c67d97d6f 13 *
emh203 0:3d9c67d97d6f 14 * Redistribution and use in source and binary forms, with or without
emh203 0:3d9c67d97d6f 15 * modification, are permitted provided that the following conditions
emh203 0:3d9c67d97d6f 16 * are met:
emh203 0:3d9c67d97d6f 17 * - Redistributions of source code must retain the above copyright
emh203 0:3d9c67d97d6f 18 * notice, this list of conditions and the following disclaimer.
emh203 0:3d9c67d97d6f 19 * - Redistributions in binary form must reproduce the above copyright
emh203 0:3d9c67d97d6f 20 * notice, this list of conditions and the following disclaimer in
emh203 0:3d9c67d97d6f 21 * the documentation and/or other materials provided with the
emh203 0:3d9c67d97d6f 22 * distribution.
emh203 0:3d9c67d97d6f 23 * - Neither the name of ARM LIMITED nor the names of its contributors
emh203 0:3d9c67d97d6f 24 * may be used to endorse or promote products derived from this
emh203 0:3d9c67d97d6f 25 * software without specific prior written permission.
emh203 0:3d9c67d97d6f 26 *
emh203 0:3d9c67d97d6f 27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
emh203 0:3d9c67d97d6f 28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
emh203 0:3d9c67d97d6f 29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
emh203 0:3d9c67d97d6f 30 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
emh203 0:3d9c67d97d6f 31 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
emh203 0:3d9c67d97d6f 32 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
emh203 0:3d9c67d97d6f 33 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
emh203 0:3d9c67d97d6f 34 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
emh203 0:3d9c67d97d6f 35 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
emh203 0:3d9c67d97d6f 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
emh203 0:3d9c67d97d6f 37 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
emh203 0:3d9c67d97d6f 38 * POSSIBILITY OF SUCH DAMAGE.
emh203 0:3d9c67d97d6f 39 * -------------------------------------------------------------------- */
emh203 0:3d9c67d97d6f 40
emh203 0:3d9c67d97d6f 41 #include "arm_math.h"
emh203 0:3d9c67d97d6f 42
emh203 0:3d9c67d97d6f 43 /**
emh203 0:3d9c67d97d6f 44 * @ingroup groupFilters
emh203 0:3d9c67d97d6f 45 */
emh203 0:3d9c67d97d6f 46
emh203 0:3d9c67d97d6f 47 /**
emh203 0:3d9c67d97d6f 48 * @addtogroup Conv
emh203 0:3d9c67d97d6f 49 * @{
emh203 0:3d9c67d97d6f 50 */
emh203 0:3d9c67d97d6f 51
emh203 0:3d9c67d97d6f 52 /**
emh203 0:3d9c67d97d6f 53 * @param[in] *pSrcA points to the first input sequence.
emh203 0:3d9c67d97d6f 54 * @param[in] srcALen length of the first input sequence.
emh203 0:3d9c67d97d6f 55 * @param[in] *pSrcB points to the second input sequence.
emh203 0:3d9c67d97d6f 56 * @param[in] srcBLen length of the second input sequence.
emh203 0:3d9c67d97d6f 57 * @param[out] *pDst points to the location where the output result is written. Length srcALen+srcBLen-1.
emh203 0:3d9c67d97d6f 58 * @return none.
emh203 0:3d9c67d97d6f 59 *
emh203 0:3d9c67d97d6f 60 * @details
emh203 0:3d9c67d97d6f 61 * <b>Scaling and Overflow Behavior:</b>
emh203 0:3d9c67d97d6f 62 *
emh203 0:3d9c67d97d6f 63 * \par
emh203 0:3d9c67d97d6f 64 * This function is optimized for speed at the expense of fixed-point precision and overflow protection.
emh203 0:3d9c67d97d6f 65 * The result of each 1.31 x 1.31 multiplication is truncated to 2.30 format.
emh203 0:3d9c67d97d6f 66 * These intermediate results are accumulated in a 32-bit register in 2.30 format.
emh203 0:3d9c67d97d6f 67 * Finally, the accumulator is saturated and converted to a 1.31 result.
emh203 0:3d9c67d97d6f 68 *
emh203 0:3d9c67d97d6f 69 * \par
emh203 0:3d9c67d97d6f 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.
emh203 0:3d9c67d97d6f 71 * In order to avoid overflows completely the input signals must be scaled down.
emh203 0:3d9c67d97d6f 72 * Scale down the inputs by log2(min(srcALen, srcBLen)) (log2 is read as log to the base 2) times to avoid overflows,
emh203 0:3d9c67d97d6f 73 * as maximum of min(srcALen, srcBLen) number of additions are carried internally.
emh203 0:3d9c67d97d6f 74 *
emh203 0:3d9c67d97d6f 75 * \par
emh203 0:3d9c67d97d6f 76 * See <code>arm_conv_q31()</code> for a slower implementation of this function which uses 64-bit accumulation to provide higher precision.
emh203 0:3d9c67d97d6f 77 */
emh203 0:3d9c67d97d6f 78
emh203 0:3d9c67d97d6f 79 void arm_conv_fast_q31(
emh203 0:3d9c67d97d6f 80 q31_t * pSrcA,
emh203 0:3d9c67d97d6f 81 uint32_t srcALen,
emh203 0:3d9c67d97d6f 82 q31_t * pSrcB,
emh203 0:3d9c67d97d6f 83 uint32_t srcBLen,
emh203 0:3d9c67d97d6f 84 q31_t * pDst)
emh203 0:3d9c67d97d6f 85 {
emh203 0:3d9c67d97d6f 86 q31_t *pIn1; /* inputA pointer */
emh203 0:3d9c67d97d6f 87 q31_t *pIn2; /* inputB pointer */
emh203 0:3d9c67d97d6f 88 q31_t *pOut = pDst; /* output pointer */
emh203 0:3d9c67d97d6f 89 q31_t *px; /* Intermediate inputA pointer */
emh203 0:3d9c67d97d6f 90 q31_t *py; /* Intermediate inputB pointer */
emh203 0:3d9c67d97d6f 91 q31_t *pSrc1, *pSrc2; /* Intermediate pointers */
emh203 0:3d9c67d97d6f 92 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
emh203 0:3d9c67d97d6f 93 q31_t x0, x1, x2, x3, c0; /* Temporary variables to hold state and coefficient values */
emh203 0:3d9c67d97d6f 94 uint32_t j, k, count, blkCnt, blockSize1, blockSize2, blockSize3; /* loop counter */
emh203 0:3d9c67d97d6f 95
emh203 0:3d9c67d97d6f 96 /* The algorithm implementation is based on the lengths of the inputs. */
emh203 0:3d9c67d97d6f 97 /* srcB is always made to slide across srcA. */
emh203 0:3d9c67d97d6f 98 /* So srcBLen is always considered as shorter or equal to srcALen */
emh203 0:3d9c67d97d6f 99 if(srcALen >= srcBLen)
emh203 0:3d9c67d97d6f 100 {
emh203 0:3d9c67d97d6f 101 /* Initialization of inputA pointer */
emh203 0:3d9c67d97d6f 102 pIn1 = pSrcA;
emh203 0:3d9c67d97d6f 103
emh203 0:3d9c67d97d6f 104 /* Initialization of inputB pointer */
emh203 0:3d9c67d97d6f 105 pIn2 = pSrcB;
emh203 0:3d9c67d97d6f 106 }
emh203 0:3d9c67d97d6f 107 else
emh203 0:3d9c67d97d6f 108 {
emh203 0:3d9c67d97d6f 109 /* Initialization of inputA pointer */
emh203 0:3d9c67d97d6f 110 pIn1 = pSrcB;
emh203 0:3d9c67d97d6f 111
emh203 0:3d9c67d97d6f 112 /* Initialization of inputB pointer */
emh203 0:3d9c67d97d6f 113 pIn2 = pSrcA;
emh203 0:3d9c67d97d6f 114
emh203 0:3d9c67d97d6f 115 /* srcBLen is always considered as shorter or equal to srcALen */
emh203 0:3d9c67d97d6f 116 j = srcBLen;
emh203 0:3d9c67d97d6f 117 srcBLen = srcALen;
emh203 0:3d9c67d97d6f 118 srcALen = j;
emh203 0:3d9c67d97d6f 119 }
emh203 0:3d9c67d97d6f 120
emh203 0:3d9c67d97d6f 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] */
emh203 0:3d9c67d97d6f 122 /* The function is internally
emh203 0:3d9c67d97d6f 123 * divided into three stages according to the number of multiplications that has to be
emh203 0:3d9c67d97d6f 124 * taken place between inputA samples and inputB samples. In the first stage of the
emh203 0:3d9c67d97d6f 125 * algorithm, the multiplications increase by one for every iteration.
emh203 0:3d9c67d97d6f 126 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emh203 0:3d9c67d97d6f 127 * In the third stage of the algorithm, the multiplications decrease by one
emh203 0:3d9c67d97d6f 128 * for every iteration. */
emh203 0:3d9c67d97d6f 129
emh203 0:3d9c67d97d6f 130 /* The algorithm is implemented in three stages.
emh203 0:3d9c67d97d6f 131 The loop counters of each stage is initiated here. */
emh203 0:3d9c67d97d6f 132 blockSize1 = srcBLen - 1u;
emh203 0:3d9c67d97d6f 133 blockSize2 = srcALen - (srcBLen - 1u);
emh203 0:3d9c67d97d6f 134 blockSize3 = blockSize1;
emh203 0:3d9c67d97d6f 135
emh203 0:3d9c67d97d6f 136 /* --------------------------
emh203 0:3d9c67d97d6f 137 * Initializations of stage1
emh203 0:3d9c67d97d6f 138 * -------------------------*/
emh203 0:3d9c67d97d6f 139
emh203 0:3d9c67d97d6f 140 /* sum = x[0] * y[0]
emh203 0:3d9c67d97d6f 141 * sum = x[0] * y[1] + x[1] * y[0]
emh203 0:3d9c67d97d6f 142 * ....
emh203 0:3d9c67d97d6f 143 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emh203 0:3d9c67d97d6f 144 */
emh203 0:3d9c67d97d6f 145
emh203 0:3d9c67d97d6f 146 /* In this stage the MAC operations are increased by 1 for every iteration.
emh203 0:3d9c67d97d6f 147 The count variable holds the number of MAC operations performed */
emh203 0:3d9c67d97d6f 148 count = 1u;
emh203 0:3d9c67d97d6f 149
emh203 0:3d9c67d97d6f 150 /* Working pointer of inputA */
emh203 0:3d9c67d97d6f 151 px = pIn1;
emh203 0:3d9c67d97d6f 152
emh203 0:3d9c67d97d6f 153 /* Working pointer of inputB */
emh203 0:3d9c67d97d6f 154 py = pIn2;
emh203 0:3d9c67d97d6f 155
emh203 0:3d9c67d97d6f 156
emh203 0:3d9c67d97d6f 157 /* ------------------------
emh203 0:3d9c67d97d6f 158 * Stage1 process
emh203 0:3d9c67d97d6f 159 * ----------------------*/
emh203 0:3d9c67d97d6f 160
emh203 0:3d9c67d97d6f 161 /* The first stage starts here */
emh203 0:3d9c67d97d6f 162 while(blockSize1 > 0u)
emh203 0:3d9c67d97d6f 163 {
emh203 0:3d9c67d97d6f 164 /* Accumulator is made zero for every iteration */
emh203 0:3d9c67d97d6f 165 sum = 0;
emh203 0:3d9c67d97d6f 166
emh203 0:3d9c67d97d6f 167 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emh203 0:3d9c67d97d6f 168 k = count >> 2u;
emh203 0:3d9c67d97d6f 169
emh203 0:3d9c67d97d6f 170 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emh203 0:3d9c67d97d6f 171 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emh203 0:3d9c67d97d6f 172 while(k > 0u)
emh203 0:3d9c67d97d6f 173 {
emh203 0:3d9c67d97d6f 174 /* x[0] * y[srcBLen - 1] */
emh203 0:3d9c67d97d6f 175 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 176 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 177
emh203 0:3d9c67d97d6f 178 /* x[1] * y[srcBLen - 2] */
emh203 0:3d9c67d97d6f 179 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 180 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 181
emh203 0:3d9c67d97d6f 182 /* x[2] * y[srcBLen - 3] */
emh203 0:3d9c67d97d6f 183 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 184 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 185
emh203 0:3d9c67d97d6f 186 /* x[3] * y[srcBLen - 4] */
emh203 0:3d9c67d97d6f 187 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 188 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 189
emh203 0:3d9c67d97d6f 190 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 191 k--;
emh203 0:3d9c67d97d6f 192 }
emh203 0:3d9c67d97d6f 193
emh203 0:3d9c67d97d6f 194 /* If the count is not a multiple of 4, compute any remaining MACs here.
emh203 0:3d9c67d97d6f 195 ** No loop unrolling is used. */
emh203 0:3d9c67d97d6f 196 k = count % 0x4u;
emh203 0:3d9c67d97d6f 197
emh203 0:3d9c67d97d6f 198 while(k > 0u)
emh203 0:3d9c67d97d6f 199 {
emh203 0:3d9c67d97d6f 200 /* Perform the multiply-accumulate */
emh203 0:3d9c67d97d6f 201 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 202 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 203
emh203 0:3d9c67d97d6f 204 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 205 k--;
emh203 0:3d9c67d97d6f 206 }
emh203 0:3d9c67d97d6f 207
emh203 0:3d9c67d97d6f 208 /* Store the result in the accumulator in the destination buffer. */
emh203 0:3d9c67d97d6f 209 *pOut++ = sum << 1;
emh203 0:3d9c67d97d6f 210
emh203 0:3d9c67d97d6f 211 /* Update the inputA and inputB pointers for next MAC calculation */
emh203 0:3d9c67d97d6f 212 py = pIn2 + count;
emh203 0:3d9c67d97d6f 213 px = pIn1;
emh203 0:3d9c67d97d6f 214
emh203 0:3d9c67d97d6f 215 /* Increment the MAC count */
emh203 0:3d9c67d97d6f 216 count++;
emh203 0:3d9c67d97d6f 217
emh203 0:3d9c67d97d6f 218 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 219 blockSize1--;
emh203 0:3d9c67d97d6f 220 }
emh203 0:3d9c67d97d6f 221
emh203 0:3d9c67d97d6f 222 /* --------------------------
emh203 0:3d9c67d97d6f 223 * Initializations of stage2
emh203 0:3d9c67d97d6f 224 * ------------------------*/
emh203 0:3d9c67d97d6f 225
emh203 0:3d9c67d97d6f 226 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emh203 0:3d9c67d97d6f 227 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emh203 0:3d9c67d97d6f 228 * ....
emh203 0:3d9c67d97d6f 229 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emh203 0:3d9c67d97d6f 230 */
emh203 0:3d9c67d97d6f 231
emh203 0:3d9c67d97d6f 232 /* Working pointer of inputA */
emh203 0:3d9c67d97d6f 233 px = pIn1;
emh203 0:3d9c67d97d6f 234
emh203 0:3d9c67d97d6f 235 /* Working pointer of inputB */
emh203 0:3d9c67d97d6f 236 pSrc2 = pIn2 + (srcBLen - 1u);
emh203 0:3d9c67d97d6f 237 py = pSrc2;
emh203 0:3d9c67d97d6f 238
emh203 0:3d9c67d97d6f 239 /* count is index by which the pointer pIn1 to be incremented */
emh203 0:3d9c67d97d6f 240 count = 0u;
emh203 0:3d9c67d97d6f 241
emh203 0:3d9c67d97d6f 242 /* -------------------
emh203 0:3d9c67d97d6f 243 * Stage2 process
emh203 0:3d9c67d97d6f 244 * ------------------*/
emh203 0:3d9c67d97d6f 245
emh203 0:3d9c67d97d6f 246 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emh203 0:3d9c67d97d6f 247 * So, to loop unroll over blockSize2,
emh203 0:3d9c67d97d6f 248 * srcBLen should be greater than or equal to 4 */
emh203 0:3d9c67d97d6f 249 if(srcBLen >= 4u)
emh203 0:3d9c67d97d6f 250 {
emh203 0:3d9c67d97d6f 251 /* Loop unroll over blockSize2, by 4 */
emh203 0:3d9c67d97d6f 252 blkCnt = blockSize2 >> 2u;
emh203 0:3d9c67d97d6f 253
emh203 0:3d9c67d97d6f 254 while(blkCnt > 0u)
emh203 0:3d9c67d97d6f 255 {
emh203 0:3d9c67d97d6f 256 /* Set all accumulators to zero */
emh203 0:3d9c67d97d6f 257 acc0 = 0;
emh203 0:3d9c67d97d6f 258 acc1 = 0;
emh203 0:3d9c67d97d6f 259 acc2 = 0;
emh203 0:3d9c67d97d6f 260 acc3 = 0;
emh203 0:3d9c67d97d6f 261
emh203 0:3d9c67d97d6f 262 /* read x[0], x[1], x[2] samples */
emh203 0:3d9c67d97d6f 263 x0 = *(px++);
emh203 0:3d9c67d97d6f 264 x1 = *(px++);
emh203 0:3d9c67d97d6f 265 x2 = *(px++);
emh203 0:3d9c67d97d6f 266
emh203 0:3d9c67d97d6f 267 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emh203 0:3d9c67d97d6f 268 k = srcBLen >> 2u;
emh203 0:3d9c67d97d6f 269
emh203 0:3d9c67d97d6f 270 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emh203 0:3d9c67d97d6f 271 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emh203 0:3d9c67d97d6f 272 do
emh203 0:3d9c67d97d6f 273 {
emh203 0:3d9c67d97d6f 274 /* Read y[srcBLen - 1] sample */
emh203 0:3d9c67d97d6f 275 c0 = *(py--);
emh203 0:3d9c67d97d6f 276
emh203 0:3d9c67d97d6f 277 /* Read x[3] sample */
emh203 0:3d9c67d97d6f 278 x3 = *(px++);
emh203 0:3d9c67d97d6f 279
emh203 0:3d9c67d97d6f 280 /* Perform the multiply-accumulates */
emh203 0:3d9c67d97d6f 281 /* acc0 += x[0] * y[srcBLen - 1] */
emh203 0:3d9c67d97d6f 282 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x0 * c0)) >> 32);
emh203 0:3d9c67d97d6f 283
emh203 0:3d9c67d97d6f 284 /* acc1 += x[1] * y[srcBLen - 1] */
emh203 0:3d9c67d97d6f 285 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x1 * c0)) >> 32);
emh203 0:3d9c67d97d6f 286
emh203 0:3d9c67d97d6f 287 /* acc2 += x[2] * y[srcBLen - 1] */
emh203 0:3d9c67d97d6f 288 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x2 * c0)) >> 32);
emh203 0:3d9c67d97d6f 289
emh203 0:3d9c67d97d6f 290 /* acc3 += x[3] * y[srcBLen - 1] */
emh203 0:3d9c67d97d6f 291 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x3 * c0)) >> 32);
emh203 0:3d9c67d97d6f 292
emh203 0:3d9c67d97d6f 293 /* Read y[srcBLen - 2] sample */
emh203 0:3d9c67d97d6f 294 c0 = *(py--);
emh203 0:3d9c67d97d6f 295
emh203 0:3d9c67d97d6f 296 /* Read x[4] sample */
emh203 0:3d9c67d97d6f 297 x0 = *(px++);
emh203 0:3d9c67d97d6f 298
emh203 0:3d9c67d97d6f 299 /* Perform the multiply-accumulate */
emh203 0:3d9c67d97d6f 300 /* acc0 += x[1] * y[srcBLen - 2] */
emh203 0:3d9c67d97d6f 301 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x1 * c0)) >> 32);
emh203 0:3d9c67d97d6f 302 /* acc1 += x[2] * y[srcBLen - 2] */
emh203 0:3d9c67d97d6f 303 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x2 * c0)) >> 32);
emh203 0:3d9c67d97d6f 304 /* acc2 += x[3] * y[srcBLen - 2] */
emh203 0:3d9c67d97d6f 305 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x3 * c0)) >> 32);
emh203 0:3d9c67d97d6f 306 /* acc3 += x[4] * y[srcBLen - 2] */
emh203 0:3d9c67d97d6f 307 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x0 * c0)) >> 32);
emh203 0:3d9c67d97d6f 308
emh203 0:3d9c67d97d6f 309 /* Read y[srcBLen - 3] sample */
emh203 0:3d9c67d97d6f 310 c0 = *(py--);
emh203 0:3d9c67d97d6f 311
emh203 0:3d9c67d97d6f 312 /* Read x[5] sample */
emh203 0:3d9c67d97d6f 313 x1 = *(px++);
emh203 0:3d9c67d97d6f 314
emh203 0:3d9c67d97d6f 315 /* Perform the multiply-accumulates */
emh203 0:3d9c67d97d6f 316 /* acc0 += x[2] * y[srcBLen - 3] */
emh203 0:3d9c67d97d6f 317 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x2 * c0)) >> 32);
emh203 0:3d9c67d97d6f 318 /* acc1 += x[3] * y[srcBLen - 3] */
emh203 0:3d9c67d97d6f 319 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x3 * c0)) >> 32);
emh203 0:3d9c67d97d6f 320 /* acc2 += x[4] * y[srcBLen - 3] */
emh203 0:3d9c67d97d6f 321 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x0 * c0)) >> 32);
emh203 0:3d9c67d97d6f 322 /* acc3 += x[5] * y[srcBLen - 3] */
emh203 0:3d9c67d97d6f 323 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x1 * c0)) >> 32);
emh203 0:3d9c67d97d6f 324
emh203 0:3d9c67d97d6f 325 /* Read y[srcBLen - 4] sample */
emh203 0:3d9c67d97d6f 326 c0 = *(py--);
emh203 0:3d9c67d97d6f 327
emh203 0:3d9c67d97d6f 328 /* Read x[6] sample */
emh203 0:3d9c67d97d6f 329 x2 = *(px++);
emh203 0:3d9c67d97d6f 330
emh203 0:3d9c67d97d6f 331 /* Perform the multiply-accumulates */
emh203 0:3d9c67d97d6f 332 /* acc0 += x[3] * y[srcBLen - 4] */
emh203 0:3d9c67d97d6f 333 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x3 * c0)) >> 32);
emh203 0:3d9c67d97d6f 334 /* acc1 += x[4] * y[srcBLen - 4] */
emh203 0:3d9c67d97d6f 335 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x0 * c0)) >> 32);
emh203 0:3d9c67d97d6f 336 /* acc2 += x[5] * y[srcBLen - 4] */
emh203 0:3d9c67d97d6f 337 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x1 * c0)) >> 32);
emh203 0:3d9c67d97d6f 338 /* acc3 += x[6] * y[srcBLen - 4] */
emh203 0:3d9c67d97d6f 339 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x2 * c0)) >> 32);
emh203 0:3d9c67d97d6f 340
emh203 0:3d9c67d97d6f 341
emh203 0:3d9c67d97d6f 342 } while(--k);
emh203 0:3d9c67d97d6f 343
emh203 0:3d9c67d97d6f 344 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emh203 0:3d9c67d97d6f 345 ** No loop unrolling is used. */
emh203 0:3d9c67d97d6f 346 k = srcBLen % 0x4u;
emh203 0:3d9c67d97d6f 347
emh203 0:3d9c67d97d6f 348 while(k > 0u)
emh203 0:3d9c67d97d6f 349 {
emh203 0:3d9c67d97d6f 350 /* Read y[srcBLen - 5] sample */
emh203 0:3d9c67d97d6f 351 c0 = *(py--);
emh203 0:3d9c67d97d6f 352
emh203 0:3d9c67d97d6f 353 /* Read x[7] sample */
emh203 0:3d9c67d97d6f 354 x3 = *(px++);
emh203 0:3d9c67d97d6f 355
emh203 0:3d9c67d97d6f 356 /* Perform the multiply-accumulates */
emh203 0:3d9c67d97d6f 357 /* acc0 += x[4] * y[srcBLen - 5] */
emh203 0:3d9c67d97d6f 358 acc0 = (q31_t) ((((q63_t) acc0 << 32) + ((q63_t) x0 * c0)) >> 32);
emh203 0:3d9c67d97d6f 359 /* acc1 += x[5] * y[srcBLen - 5] */
emh203 0:3d9c67d97d6f 360 acc1 = (q31_t) ((((q63_t) acc1 << 32) + ((q63_t) x1 * c0)) >> 32);
emh203 0:3d9c67d97d6f 361 /* acc2 += x[6] * y[srcBLen - 5] */
emh203 0:3d9c67d97d6f 362 acc2 = (q31_t) ((((q63_t) acc2 << 32) + ((q63_t) x2 * c0)) >> 32);
emh203 0:3d9c67d97d6f 363 /* acc3 += x[7] * y[srcBLen - 5] */
emh203 0:3d9c67d97d6f 364 acc3 = (q31_t) ((((q63_t) acc3 << 32) + ((q63_t) x3 * c0)) >> 32);
emh203 0:3d9c67d97d6f 365
emh203 0:3d9c67d97d6f 366 /* Reuse the present samples for the next MAC */
emh203 0:3d9c67d97d6f 367 x0 = x1;
emh203 0:3d9c67d97d6f 368 x1 = x2;
emh203 0:3d9c67d97d6f 369 x2 = x3;
emh203 0:3d9c67d97d6f 370
emh203 0:3d9c67d97d6f 371 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 372 k--;
emh203 0:3d9c67d97d6f 373 }
emh203 0:3d9c67d97d6f 374
emh203 0:3d9c67d97d6f 375 /* Store the results in the accumulators in the destination buffer. */
emh203 0:3d9c67d97d6f 376 *pOut++ = (q31_t) (acc0 << 1);
emh203 0:3d9c67d97d6f 377 *pOut++ = (q31_t) (acc1 << 1);
emh203 0:3d9c67d97d6f 378 *pOut++ = (q31_t) (acc2 << 1);
emh203 0:3d9c67d97d6f 379 *pOut++ = (q31_t) (acc3 << 1);
emh203 0:3d9c67d97d6f 380
emh203 0:3d9c67d97d6f 381 /* Increment the pointer pIn1 index, count by 4 */
emh203 0:3d9c67d97d6f 382 count += 4u;
emh203 0:3d9c67d97d6f 383
emh203 0:3d9c67d97d6f 384 /* Update the inputA and inputB pointers for next MAC calculation */
emh203 0:3d9c67d97d6f 385 px = pIn1 + count;
emh203 0:3d9c67d97d6f 386 py = pSrc2;
emh203 0:3d9c67d97d6f 387
emh203 0:3d9c67d97d6f 388 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 389 blkCnt--;
emh203 0:3d9c67d97d6f 390 }
emh203 0:3d9c67d97d6f 391
emh203 0:3d9c67d97d6f 392 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emh203 0:3d9c67d97d6f 393 ** No loop unrolling is used. */
emh203 0:3d9c67d97d6f 394 blkCnt = blockSize2 % 0x4u;
emh203 0:3d9c67d97d6f 395
emh203 0:3d9c67d97d6f 396 while(blkCnt > 0u)
emh203 0:3d9c67d97d6f 397 {
emh203 0:3d9c67d97d6f 398 /* Accumulator is made zero for every iteration */
emh203 0:3d9c67d97d6f 399 sum = 0;
emh203 0:3d9c67d97d6f 400
emh203 0:3d9c67d97d6f 401 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emh203 0:3d9c67d97d6f 402 k = srcBLen >> 2u;
emh203 0:3d9c67d97d6f 403
emh203 0:3d9c67d97d6f 404 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emh203 0:3d9c67d97d6f 405 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emh203 0:3d9c67d97d6f 406 while(k > 0u)
emh203 0:3d9c67d97d6f 407 {
emh203 0:3d9c67d97d6f 408 /* Perform the multiply-accumulates */
emh203 0:3d9c67d97d6f 409 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 410 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 411 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 412 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 413 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 414 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 415 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 416 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 417
emh203 0:3d9c67d97d6f 418 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 419 k--;
emh203 0:3d9c67d97d6f 420 }
emh203 0:3d9c67d97d6f 421
emh203 0:3d9c67d97d6f 422 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emh203 0:3d9c67d97d6f 423 ** No loop unrolling is used. */
emh203 0:3d9c67d97d6f 424 k = srcBLen % 0x4u;
emh203 0:3d9c67d97d6f 425
emh203 0:3d9c67d97d6f 426 while(k > 0u)
emh203 0:3d9c67d97d6f 427 {
emh203 0:3d9c67d97d6f 428 /* Perform the multiply-accumulate */
emh203 0:3d9c67d97d6f 429 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 430 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 431
emh203 0:3d9c67d97d6f 432 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 433 k--;
emh203 0:3d9c67d97d6f 434 }
emh203 0:3d9c67d97d6f 435
emh203 0:3d9c67d97d6f 436 /* Store the result in the accumulator in the destination buffer. */
emh203 0:3d9c67d97d6f 437 *pOut++ = sum << 1;
emh203 0:3d9c67d97d6f 438
emh203 0:3d9c67d97d6f 439 /* Increment the MAC count */
emh203 0:3d9c67d97d6f 440 count++;
emh203 0:3d9c67d97d6f 441
emh203 0:3d9c67d97d6f 442 /* Update the inputA and inputB pointers for next MAC calculation */
emh203 0:3d9c67d97d6f 443 px = pIn1 + count;
emh203 0:3d9c67d97d6f 444 py = pSrc2;
emh203 0:3d9c67d97d6f 445
emh203 0:3d9c67d97d6f 446 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 447 blkCnt--;
emh203 0:3d9c67d97d6f 448 }
emh203 0:3d9c67d97d6f 449 }
emh203 0:3d9c67d97d6f 450 else
emh203 0:3d9c67d97d6f 451 {
emh203 0:3d9c67d97d6f 452 /* If the srcBLen is not a multiple of 4,
emh203 0:3d9c67d97d6f 453 * the blockSize2 loop cannot be unrolled by 4 */
emh203 0:3d9c67d97d6f 454 blkCnt = blockSize2;
emh203 0:3d9c67d97d6f 455
emh203 0:3d9c67d97d6f 456 while(blkCnt > 0u)
emh203 0:3d9c67d97d6f 457 {
emh203 0:3d9c67d97d6f 458 /* Accumulator is made zero for every iteration */
emh203 0:3d9c67d97d6f 459 sum = 0;
emh203 0:3d9c67d97d6f 460
emh203 0:3d9c67d97d6f 461 /* srcBLen number of MACS should be performed */
emh203 0:3d9c67d97d6f 462 k = srcBLen;
emh203 0:3d9c67d97d6f 463
emh203 0:3d9c67d97d6f 464 while(k > 0u)
emh203 0:3d9c67d97d6f 465 {
emh203 0:3d9c67d97d6f 466 /* Perform the multiply-accumulate */
emh203 0:3d9c67d97d6f 467 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 468 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 469
emh203 0:3d9c67d97d6f 470 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 471 k--;
emh203 0:3d9c67d97d6f 472 }
emh203 0:3d9c67d97d6f 473
emh203 0:3d9c67d97d6f 474 /* Store the result in the accumulator in the destination buffer. */
emh203 0:3d9c67d97d6f 475 *pOut++ = sum << 1;
emh203 0:3d9c67d97d6f 476
emh203 0:3d9c67d97d6f 477 /* Increment the MAC count */
emh203 0:3d9c67d97d6f 478 count++;
emh203 0:3d9c67d97d6f 479
emh203 0:3d9c67d97d6f 480 /* Update the inputA and inputB pointers for next MAC calculation */
emh203 0:3d9c67d97d6f 481 px = pIn1 + count;
emh203 0:3d9c67d97d6f 482 py = pSrc2;
emh203 0:3d9c67d97d6f 483
emh203 0:3d9c67d97d6f 484 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 485 blkCnt--;
emh203 0:3d9c67d97d6f 486 }
emh203 0:3d9c67d97d6f 487 }
emh203 0:3d9c67d97d6f 488
emh203 0:3d9c67d97d6f 489
emh203 0:3d9c67d97d6f 490 /* --------------------------
emh203 0:3d9c67d97d6f 491 * Initializations of stage3
emh203 0:3d9c67d97d6f 492 * -------------------------*/
emh203 0:3d9c67d97d6f 493
emh203 0:3d9c67d97d6f 494 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emh203 0:3d9c67d97d6f 495 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emh203 0:3d9c67d97d6f 496 * ....
emh203 0:3d9c67d97d6f 497 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emh203 0:3d9c67d97d6f 498 * sum += x[srcALen-1] * y[srcBLen-1]
emh203 0:3d9c67d97d6f 499 */
emh203 0:3d9c67d97d6f 500
emh203 0:3d9c67d97d6f 501 /* In this stage the MAC operations are decreased by 1 for every iteration.
emh203 0:3d9c67d97d6f 502 The blockSize3 variable holds the number of MAC operations performed */
emh203 0:3d9c67d97d6f 503
emh203 0:3d9c67d97d6f 504 /* Working pointer of inputA */
emh203 0:3d9c67d97d6f 505 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emh203 0:3d9c67d97d6f 506 px = pSrc1;
emh203 0:3d9c67d97d6f 507
emh203 0:3d9c67d97d6f 508 /* Working pointer of inputB */
emh203 0:3d9c67d97d6f 509 pSrc2 = pIn2 + (srcBLen - 1u);
emh203 0:3d9c67d97d6f 510 py = pSrc2;
emh203 0:3d9c67d97d6f 511
emh203 0:3d9c67d97d6f 512 /* -------------------
emh203 0:3d9c67d97d6f 513 * Stage3 process
emh203 0:3d9c67d97d6f 514 * ------------------*/
emh203 0:3d9c67d97d6f 515
emh203 0:3d9c67d97d6f 516 while(blockSize3 > 0u)
emh203 0:3d9c67d97d6f 517 {
emh203 0:3d9c67d97d6f 518 /* Accumulator is made zero for every iteration */
emh203 0:3d9c67d97d6f 519 sum = 0;
emh203 0:3d9c67d97d6f 520
emh203 0:3d9c67d97d6f 521 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emh203 0:3d9c67d97d6f 522 k = blockSize3 >> 2u;
emh203 0:3d9c67d97d6f 523
emh203 0:3d9c67d97d6f 524 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emh203 0:3d9c67d97d6f 525 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emh203 0:3d9c67d97d6f 526 while(k > 0u)
emh203 0:3d9c67d97d6f 527 {
emh203 0:3d9c67d97d6f 528 /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */
emh203 0:3d9c67d97d6f 529 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 530 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 531
emh203 0:3d9c67d97d6f 532 /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */
emh203 0:3d9c67d97d6f 533 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 534 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 535
emh203 0:3d9c67d97d6f 536 /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */
emh203 0:3d9c67d97d6f 537 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 538 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 539
emh203 0:3d9c67d97d6f 540 /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */
emh203 0:3d9c67d97d6f 541 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 542 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 543
emh203 0:3d9c67d97d6f 544 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 545 k--;
emh203 0:3d9c67d97d6f 546 }
emh203 0:3d9c67d97d6f 547
emh203 0:3d9c67d97d6f 548 /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
emh203 0:3d9c67d97d6f 549 ** No loop unrolling is used. */
emh203 0:3d9c67d97d6f 550 k = blockSize3 % 0x4u;
emh203 0:3d9c67d97d6f 551
emh203 0:3d9c67d97d6f 552 while(k > 0u)
emh203 0:3d9c67d97d6f 553 {
emh203 0:3d9c67d97d6f 554 /* Perform the multiply-accumulate */
emh203 0:3d9c67d97d6f 555 sum = (q31_t) ((((q63_t) sum << 32) +
emh203 0:3d9c67d97d6f 556 ((q63_t) * px++ * (*py--))) >> 32);
emh203 0:3d9c67d97d6f 557
emh203 0:3d9c67d97d6f 558 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 559 k--;
emh203 0:3d9c67d97d6f 560 }
emh203 0:3d9c67d97d6f 561
emh203 0:3d9c67d97d6f 562 /* Store the result in the accumulator in the destination buffer. */
emh203 0:3d9c67d97d6f 563 *pOut++ = sum << 1;
emh203 0:3d9c67d97d6f 564
emh203 0:3d9c67d97d6f 565 /* Update the inputA and inputB pointers for next MAC calculation */
emh203 0:3d9c67d97d6f 566 px = ++pSrc1;
emh203 0:3d9c67d97d6f 567 py = pSrc2;
emh203 0:3d9c67d97d6f 568
emh203 0:3d9c67d97d6f 569 /* Decrement the loop counter */
emh203 0:3d9c67d97d6f 570 blockSize3--;
emh203 0:3d9c67d97d6f 571 }
emh203 0:3d9c67d97d6f 572
emh203 0:3d9c67d97d6f 573 }
emh203 0:3d9c67d97d6f 574
emh203 0:3d9c67d97d6f 575 /**
emh203 0:3d9c67d97d6f 576 * @} end of Conv group
emh203 0:3d9c67d97d6f 577 */