mbed official / mbed-dsp

CMSIS DSP library

Dependents:   performance_timer Surfboard_ gps2rtty Capstone ... more

Legacy Warning

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

cmsis_dsp/FilteringFunctions/arm_conv_partial_q7.c@5:3762170b6d4d, 2015-11-20 (annotated)

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



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



Added option to build rpc library. closes #1426

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emilmont 1:fdd22bb7aa52 1 /* ----------------------------------------------------------------------
mbed_official 5:3762170b6d4d 2 * Copyright (C) 2010-2014 ARM Limited. All rights reserved.
emilmont 1:fdd22bb7aa52 3 *
mbed_official 5:3762170b6d4d 4 * $Date: 19. March 2015
mbed_official 5:3762170b6d4d 5 * $Revision: V.1.4.5
emilmont 1:fdd22bb7aa52 6 *
emilmont 2:da51fb522205 7 * Project: CMSIS DSP Library
emilmont 2:da51fb522205 8 * Title: arm_conv_partial_q7.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Partial convolution of Q7 sequences.
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
emilmont 1:fdd22bb7aa52 13 *
mbed_official 3:7a284390b0ce 14 * Redistribution and use in source and binary forms, with or without
mbed_official 3:7a284390b0ce 15 * modification, are permitted provided that the following conditions
mbed_official 3:7a284390b0ce 16 * are met:
mbed_official 3:7a284390b0ce 17 * - Redistributions of source code must retain the above copyright
mbed_official 3:7a284390b0ce 18 * notice, this list of conditions and the following disclaimer.
mbed_official 3:7a284390b0ce 19 * - Redistributions in binary form must reproduce the above copyright
mbed_official 3:7a284390b0ce 20 * notice, this list of conditions and the following disclaimer in
mbed_official 3:7a284390b0ce 21 * the documentation and/or other materials provided with the
mbed_official 3:7a284390b0ce 22 * distribution.
mbed_official 3:7a284390b0ce 23 * - Neither the name of ARM LIMITED nor the names of its contributors
mbed_official 3:7a284390b0ce 24 * may be used to endorse or promote products derived from this
mbed_official 3:7a284390b0ce 25 * software without specific prior written permission.
emilmont 1:fdd22bb7aa52 26 *
mbed_official 3:7a284390b0ce 27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
mbed_official 3:7a284390b0ce 28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
mbed_official 3:7a284390b0ce 29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
mbed_official 3:7a284390b0ce 30 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
mbed_official 3:7a284390b0ce 31 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
mbed_official 3:7a284390b0ce 32 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
mbed_official 3:7a284390b0ce 33 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
mbed_official 3:7a284390b0ce 34 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
mbed_official 3:7a284390b0ce 35 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
mbed_official 3:7a284390b0ce 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
mbed_official 3:7a284390b0ce 37 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
mbed_official 3:7a284390b0ce 38 * POSSIBILITY OF SUCH DAMAGE.
emilmont 1:fdd22bb7aa52 39 * -------------------------------------------------------------------- */
emilmont 1:fdd22bb7aa52 40
emilmont 1:fdd22bb7aa52 41 #include "arm_math.h"
emilmont 1:fdd22bb7aa52 42
emilmont 1:fdd22bb7aa52 43 /**
emilmont 1:fdd22bb7aa52 44 * @ingroup groupFilters
emilmont 1:fdd22bb7aa52 45 */
emilmont 1:fdd22bb7aa52 46
emilmont 1:fdd22bb7aa52 47 /**
emilmont 1:fdd22bb7aa52 48 * @addtogroup PartialConv
emilmont 1:fdd22bb7aa52 49 * @{
emilmont 1:fdd22bb7aa52 50 */
emilmont 1:fdd22bb7aa52 51
emilmont 1:fdd22bb7aa52 52 /**
emilmont 1:fdd22bb7aa52 53 * @brief Partial convolution of Q7 sequences.
emilmont 1:fdd22bb7aa52 54 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 58 * @param[out] *pDst points to the location where the output result is written.
emilmont 1:fdd22bb7aa52 59 * @param[in] firstIndex is the first output sample to start with.
emilmont 1:fdd22bb7aa52 60 * @param[in] numPoints is the number of output points to be computed.
emilmont 1:fdd22bb7aa52 61 * @return Returns either ARM_MATH_SUCCESS if the function completed correctly or ARM_MATH_ARGUMENT_ERROR if the requested subset is not in the range [0 srcALen+srcBLen-2].
emilmont 1:fdd22bb7aa52 62 *
emilmont 1:fdd22bb7aa52 63 * \par
emilmont 1:fdd22bb7aa52 64 * Refer the function <code>arm_conv_partial_opt_q7()</code> for a faster implementation of this function.
emilmont 1:fdd22bb7aa52 65 *
emilmont 1:fdd22bb7aa52 66 */
emilmont 1:fdd22bb7aa52 67
emilmont 1:fdd22bb7aa52 68 arm_status arm_conv_partial_q7(
emilmont 1:fdd22bb7aa52 69 q7_t * pSrcA,
emilmont 1:fdd22bb7aa52 70 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 71 q7_t * pSrcB,
emilmont 1:fdd22bb7aa52 72 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 73 q7_t * pDst,
emilmont 1:fdd22bb7aa52 74 uint32_t firstIndex,
emilmont 1:fdd22bb7aa52 75 uint32_t numPoints)
emilmont 1:fdd22bb7aa52 76 {
emilmont 1:fdd22bb7aa52 77
emilmont 1:fdd22bb7aa52 78
mbed_official 3:7a284390b0ce 79 #ifndef ARM_MATH_CM0_FAMILY
emilmont 1:fdd22bb7aa52 80
emilmont 1:fdd22bb7aa52 81 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 82
emilmont 1:fdd22bb7aa52 83 q7_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 84 q7_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 85 q7_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 86 q7_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 87 q7_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 88 q7_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 89 q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 90 q31_t input1, input2;
emilmont 1:fdd22bb7aa52 91 q15_t in1, in2;
emilmont 1:fdd22bb7aa52 92 q7_t x0, x1, x2, x3, c0, c1;
emilmont 1:fdd22bb7aa52 93 uint32_t j, k, count, check, blkCnt;
emilmont 1:fdd22bb7aa52 94 int32_t blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 95 arm_status status;
emilmont 1:fdd22bb7aa52 96
emilmont 1:fdd22bb7aa52 97
emilmont 1:fdd22bb7aa52 98 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 99 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 100 {
emilmont 1:fdd22bb7aa52 101 /* Set status as ARM_MATH_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 102 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 103 }
emilmont 1:fdd22bb7aa52 104 else
emilmont 1:fdd22bb7aa52 105 {
emilmont 1:fdd22bb7aa52 106
emilmont 1:fdd22bb7aa52 107 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 108 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 109 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 110 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 111 {
emilmont 1:fdd22bb7aa52 112 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 113 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 114
emilmont 1:fdd22bb7aa52 115 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 116 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 117 }
emilmont 1:fdd22bb7aa52 118 else
emilmont 1:fdd22bb7aa52 119 {
emilmont 1:fdd22bb7aa52 120 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 121 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 122
emilmont 1:fdd22bb7aa52 123 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 124 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 125
emilmont 1:fdd22bb7aa52 126 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 127 j = srcBLen;
emilmont 1:fdd22bb7aa52 128 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 129 srcALen = j;
emilmont 1:fdd22bb7aa52 130 }
emilmont 1:fdd22bb7aa52 131
emilmont 1:fdd22bb7aa52 132 /* Conditions to check which loopCounter holds
emilmont 1:fdd22bb7aa52 133 * the first and last indices of the output samples to be calculated. */
emilmont 1:fdd22bb7aa52 134 check = firstIndex + numPoints;
mbed_official 5:3762170b6d4d 135 blockSize3 = ((int32_t)check > (int32_t)srcALen) ? (int32_t)check - (int32_t)srcALen : 0;
mbed_official 5:3762170b6d4d 136 blockSize3 = ((int32_t)firstIndex > (int32_t)srcALen - 1) ? blockSize3 - (int32_t)firstIndex + (int32_t)srcALen : blockSize3;
emilmont 1:fdd22bb7aa52 137 blockSize1 = (((int32_t) srcBLen - 1) - (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 138 blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1u)) ? blockSize1 :
emilmont 1:fdd22bb7aa52 139 (int32_t) numPoints) : 0;
emilmont 1:fdd22bb7aa52 140 blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) +
emilmont 1:fdd22bb7aa52 141 (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 142 blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
emilmont 1:fdd22bb7aa52 143
emilmont 1:fdd22bb7aa52 144 /* 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 145 /* The function is internally
emilmont 1:fdd22bb7aa52 146 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 147 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 148 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 149 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 150 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 151 * for every iteration. */
emilmont 1:fdd22bb7aa52 152
emilmont 1:fdd22bb7aa52 153 /* Set the output pointer to point to the firstIndex
emilmont 1:fdd22bb7aa52 154 * of the output sample to be calculated. */
emilmont 1:fdd22bb7aa52 155 pOut = pDst + firstIndex;
emilmont 1:fdd22bb7aa52 156
emilmont 1:fdd22bb7aa52 157 /* --------------------------
emilmont 1:fdd22bb7aa52 158 * Initializations of stage1
emilmont 1:fdd22bb7aa52 159 * -------------------------*/
emilmont 1:fdd22bb7aa52 160
emilmont 1:fdd22bb7aa52 161 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 162 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 163 * ....
emilmont 1:fdd22bb7aa52 164 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 165 */
emilmont 1:fdd22bb7aa52 166
emilmont 1:fdd22bb7aa52 167 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 168 The count variable holds the number of MAC operations performed.
emilmont 1:fdd22bb7aa52 169 Since the partial convolution starts from from firstIndex
emilmont 1:fdd22bb7aa52 170 Number of Macs to be performed is firstIndex + 1 */
emilmont 1:fdd22bb7aa52 171 count = 1u + firstIndex;
emilmont 1:fdd22bb7aa52 172
emilmont 1:fdd22bb7aa52 173 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 174 px = pIn1;
emilmont 1:fdd22bb7aa52 175
emilmont 1:fdd22bb7aa52 176 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 177 pSrc2 = pIn2 + firstIndex;
emilmont 1:fdd22bb7aa52 178 py = pSrc2;
emilmont 1:fdd22bb7aa52 179
emilmont 1:fdd22bb7aa52 180 /* ------------------------
emilmont 1:fdd22bb7aa52 181 * Stage1 process
emilmont 1:fdd22bb7aa52 182 * ----------------------*/
emilmont 1:fdd22bb7aa52 183
emilmont 1:fdd22bb7aa52 184 /* The first stage starts here */
emilmont 1:fdd22bb7aa52 185 while(blockSize1 > 0)
emilmont 1:fdd22bb7aa52 186 {
emilmont 1:fdd22bb7aa52 187 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 188 sum = 0;
emilmont 1:fdd22bb7aa52 189
emilmont 1:fdd22bb7aa52 190 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 191 k = count >> 2u;
emilmont 1:fdd22bb7aa52 192
emilmont 1:fdd22bb7aa52 193 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 194 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 195 while(k > 0u)
emilmont 1:fdd22bb7aa52 196 {
emilmont 1:fdd22bb7aa52 197 /* x[0] , x[1] */
emilmont 1:fdd22bb7aa52 198 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 199 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 200 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 201
emilmont 1:fdd22bb7aa52 202 /* y[srcBLen - 1] , y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 203 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 204 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 205 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 206
emilmont 1:fdd22bb7aa52 207 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 208 /* x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 209 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 210
emilmont 1:fdd22bb7aa52 211 /* x[2] , x[3] */
emilmont 1:fdd22bb7aa52 212 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 213 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 214 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 215
emilmont 1:fdd22bb7aa52 216 /* y[srcBLen - 3] , y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 217 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 218 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 219 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 220
emilmont 1:fdd22bb7aa52 221 /* x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 222 /* x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 223 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 224
emilmont 1:fdd22bb7aa52 225 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 226 k--;
emilmont 1:fdd22bb7aa52 227 }
emilmont 1:fdd22bb7aa52 228
emilmont 1:fdd22bb7aa52 229 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 230 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 231 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 while(k > 0u)
emilmont 1:fdd22bb7aa52 234 {
emilmont 1:fdd22bb7aa52 235 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 236 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 237
emilmont 1:fdd22bb7aa52 238 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 239 k--;
emilmont 1:fdd22bb7aa52 240 }
emilmont 1:fdd22bb7aa52 241
emilmont 1:fdd22bb7aa52 242 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 243 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 244
emilmont 1:fdd22bb7aa52 245 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 246 py = ++pSrc2;
emilmont 1:fdd22bb7aa52 247 px = pIn1;
emilmont 1:fdd22bb7aa52 248
emilmont 1:fdd22bb7aa52 249 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 250 count++;
emilmont 1:fdd22bb7aa52 251
emilmont 1:fdd22bb7aa52 252 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 253 blockSize1--;
emilmont 1:fdd22bb7aa52 254 }
emilmont 1:fdd22bb7aa52 255
emilmont 1:fdd22bb7aa52 256 /* --------------------------
emilmont 1:fdd22bb7aa52 257 * Initializations of stage2
emilmont 1:fdd22bb7aa52 258 * ------------------------*/
emilmont 1:fdd22bb7aa52 259
emilmont 1:fdd22bb7aa52 260 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 261 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 262 * ....
emilmont 1:fdd22bb7aa52 263 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 264 */
emilmont 1:fdd22bb7aa52 265
emilmont 1:fdd22bb7aa52 266 /* Working pointer of inputA */
mbed_official 5:3762170b6d4d 267 if((int32_t)firstIndex - (int32_t)srcBLen + 1 > 0)
mbed_official 5:3762170b6d4d 268 {
mbed_official 5:3762170b6d4d 269 px = pIn1 + firstIndex - srcBLen + 1;
mbed_official 5:3762170b6d4d 270 }
mbed_official 5:3762170b6d4d 271 else
mbed_official 5:3762170b6d4d 272 {
mbed_official 5:3762170b6d4d 273 px = pIn1;
mbed_official 5:3762170b6d4d 274 }
emilmont 1:fdd22bb7aa52 275
emilmont 1:fdd22bb7aa52 276 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 277 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 278 py = pSrc2;
emilmont 1:fdd22bb7aa52 279
emilmont 1:fdd22bb7aa52 280 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 281 count = 0u;
emilmont 1:fdd22bb7aa52 282
emilmont 1:fdd22bb7aa52 283 /* -------------------
emilmont 1:fdd22bb7aa52 284 * Stage2 process
emilmont 1:fdd22bb7aa52 285 * ------------------*/
emilmont 1:fdd22bb7aa52 286
emilmont 1:fdd22bb7aa52 287 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 288 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 289 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 290 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 291 {
emilmont 1:fdd22bb7aa52 292 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 293 blkCnt = ((uint32_t) blockSize2 >> 2u);
emilmont 1:fdd22bb7aa52 294
emilmont 1:fdd22bb7aa52 295 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 296 {
emilmont 1:fdd22bb7aa52 297 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 298 acc0 = 0;
emilmont 1:fdd22bb7aa52 299 acc1 = 0;
emilmont 1:fdd22bb7aa52 300 acc2 = 0;
emilmont 1:fdd22bb7aa52 301 acc3 = 0;
emilmont 1:fdd22bb7aa52 302
emilmont 1:fdd22bb7aa52 303 /* read x[0], x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 304 x0 = *(px++);
emilmont 1:fdd22bb7aa52 305 x1 = *(px++);
emilmont 1:fdd22bb7aa52 306 x2 = *(px++);
emilmont 1:fdd22bb7aa52 307
emilmont 1:fdd22bb7aa52 308 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 309 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 310
emilmont 1:fdd22bb7aa52 311 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 312 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 313 do
emilmont 1:fdd22bb7aa52 314 {
emilmont 1:fdd22bb7aa52 315 /* Read y[srcBLen - 1] sample */
emilmont 1:fdd22bb7aa52 316 c0 = *(py--);
emilmont 1:fdd22bb7aa52 317 /* Read y[srcBLen - 2] sample */
emilmont 1:fdd22bb7aa52 318 c1 = *(py--);
emilmont 1:fdd22bb7aa52 319
emilmont 1:fdd22bb7aa52 320 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 321 x3 = *(px++);
emilmont 1:fdd22bb7aa52 322
emilmont 1:fdd22bb7aa52 323 /* x[0] and x[1] are packed */
emilmont 1:fdd22bb7aa52 324 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 325 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 326
emilmont 1:fdd22bb7aa52 327 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 328
emilmont 1:fdd22bb7aa52 329 /* y[srcBLen - 1] and y[srcBLen - 2] are packed */
emilmont 1:fdd22bb7aa52 330 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 331 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 332
emilmont 1:fdd22bb7aa52 333 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 334
emilmont 1:fdd22bb7aa52 335 /* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 336 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 337
emilmont 1:fdd22bb7aa52 338 /* x[1] and x[2] are packed */
emilmont 1:fdd22bb7aa52 339 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 340 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 343
emilmont 1:fdd22bb7aa52 344 /* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 345 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 346
emilmont 1:fdd22bb7aa52 347 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 348 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 349 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 350
emilmont 1:fdd22bb7aa52 351 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 352
emilmont 1:fdd22bb7aa52 353 /* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 354 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 355
emilmont 1:fdd22bb7aa52 356 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 357 x0 = *(px++);
emilmont 1:fdd22bb7aa52 358
emilmont 1:fdd22bb7aa52 359 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 360 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 361 in2 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 362
emilmont 1:fdd22bb7aa52 363 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 364
emilmont 1:fdd22bb7aa52 365 /* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 366 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 367
emilmont 1:fdd22bb7aa52 368 /* Read y[srcBLen - 3] sample */
emilmont 1:fdd22bb7aa52 369 c0 = *(py--);
emilmont 1:fdd22bb7aa52 370 /* Read y[srcBLen - 4] sample */
emilmont 1:fdd22bb7aa52 371 c1 = *(py--);
emilmont 1:fdd22bb7aa52 372
emilmont 1:fdd22bb7aa52 373 /* Read x[5] sample */
emilmont 1:fdd22bb7aa52 374 x1 = *(px++);
emilmont 1:fdd22bb7aa52 375
emilmont 1:fdd22bb7aa52 376 /* x[2] and x[3] are packed */
emilmont 1:fdd22bb7aa52 377 in1 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 378 in2 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 379
emilmont 1:fdd22bb7aa52 380 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 381
emilmont 1:fdd22bb7aa52 382 /* y[srcBLen - 3] and y[srcBLen - 4] are packed */
emilmont 1:fdd22bb7aa52 383 in1 = (q15_t) c0;
emilmont 1:fdd22bb7aa52 384 in2 = (q15_t) c1;
emilmont 1:fdd22bb7aa52 385
emilmont 1:fdd22bb7aa52 386 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 387
emilmont 1:fdd22bb7aa52 388 /* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 389 acc0 = __SMLAD(input1, input2, acc0);
emilmont 1:fdd22bb7aa52 390
emilmont 1:fdd22bb7aa52 391 /* x[3] and x[4] are packed */
emilmont 1:fdd22bb7aa52 392 in1 = (q15_t) x3;
emilmont 1:fdd22bb7aa52 393 in2 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 394
emilmont 1:fdd22bb7aa52 395 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 396
emilmont 1:fdd22bb7aa52 397 /* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 398 acc1 = __SMLAD(input1, input2, acc1);
emilmont 1:fdd22bb7aa52 399
emilmont 1:fdd22bb7aa52 400 /* x[4] and x[5] are packed */
emilmont 1:fdd22bb7aa52 401 in1 = (q15_t) x0;
emilmont 1:fdd22bb7aa52 402 in2 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 405
emilmont 1:fdd22bb7aa52 406 /* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 407 acc2 = __SMLAD(input1, input2, acc2);
emilmont 1:fdd22bb7aa52 408
emilmont 1:fdd22bb7aa52 409 /* Read x[6] sample */
emilmont 1:fdd22bb7aa52 410 x2 = *(px++);
emilmont 1:fdd22bb7aa52 411
emilmont 1:fdd22bb7aa52 412 /* x[5] and x[6] are packed */
emilmont 1:fdd22bb7aa52 413 in1 = (q15_t) x1;
emilmont 1:fdd22bb7aa52 414 in2 = (q15_t) x2;
emilmont 1:fdd22bb7aa52 415
emilmont 1:fdd22bb7aa52 416 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 417
emilmont 1:fdd22bb7aa52 418 /* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 419 acc3 = __SMLAD(input1, input2, acc3);
emilmont 1:fdd22bb7aa52 420
emilmont 1:fdd22bb7aa52 421 } while(--k);
emilmont 1:fdd22bb7aa52 422
emilmont 1:fdd22bb7aa52 423 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 424 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 425 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 426
emilmont 1:fdd22bb7aa52 427 while(k > 0u)
emilmont 1:fdd22bb7aa52 428 {
emilmont 1:fdd22bb7aa52 429 /* Read y[srcBLen - 5] sample */
emilmont 1:fdd22bb7aa52 430 c0 = *(py--);
emilmont 1:fdd22bb7aa52 431
emilmont 1:fdd22bb7aa52 432 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 433 x3 = *(px++);
emilmont 1:fdd22bb7aa52 434
emilmont 1:fdd22bb7aa52 435 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 436 /* acc0 += x[4] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 437 acc0 += ((q31_t) x0 * c0);
emilmont 1:fdd22bb7aa52 438 /* acc1 += x[5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 439 acc1 += ((q31_t) x1 * c0);
emilmont 1:fdd22bb7aa52 440 /* acc2 += x[6] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 441 acc2 += ((q31_t) x2 * c0);
emilmont 1:fdd22bb7aa52 442 /* acc3 += x[7] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 443 acc3 += ((q31_t) x3 * c0);
emilmont 1:fdd22bb7aa52 444
emilmont 1:fdd22bb7aa52 445 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 446 x0 = x1;
emilmont 1:fdd22bb7aa52 447 x1 = x2;
emilmont 1:fdd22bb7aa52 448 x2 = x3;
emilmont 1:fdd22bb7aa52 449
emilmont 1:fdd22bb7aa52 450 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 451 k--;
emilmont 1:fdd22bb7aa52 452 }
emilmont 1:fdd22bb7aa52 453
emilmont 1:fdd22bb7aa52 454 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 455 *pOut++ = (q7_t) (__SSAT(acc0 >> 7, 8));
emilmont 1:fdd22bb7aa52 456 *pOut++ = (q7_t) (__SSAT(acc1 >> 7, 8));
emilmont 1:fdd22bb7aa52 457 *pOut++ = (q7_t) (__SSAT(acc2 >> 7, 8));
emilmont 1:fdd22bb7aa52 458 *pOut++ = (q7_t) (__SSAT(acc3 >> 7, 8));
emilmont 1:fdd22bb7aa52 459
emilmont 1:fdd22bb7aa52 460 /* Increment the pointer pIn1 index, count by 4 */
emilmont 1:fdd22bb7aa52 461 count += 4u;
emilmont 1:fdd22bb7aa52 462
emilmont 1:fdd22bb7aa52 463 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 464 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 465 py = pSrc2;
emilmont 1:fdd22bb7aa52 466
emilmont 1:fdd22bb7aa52 467
emilmont 1:fdd22bb7aa52 468 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 469 blkCnt--;
emilmont 1:fdd22bb7aa52 470 }
emilmont 1:fdd22bb7aa52 471
emilmont 1:fdd22bb7aa52 472 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 473 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 474 blkCnt = (uint32_t) blockSize2 % 0x4u;
emilmont 1:fdd22bb7aa52 475
emilmont 1:fdd22bb7aa52 476 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 477 {
emilmont 1:fdd22bb7aa52 478 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 479 sum = 0;
emilmont 1:fdd22bb7aa52 480
emilmont 1:fdd22bb7aa52 481 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 482 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 483
emilmont 1:fdd22bb7aa52 484 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 485 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 486 while(k > 0u)
emilmont 1:fdd22bb7aa52 487 {
emilmont 1:fdd22bb7aa52 488
emilmont 1:fdd22bb7aa52 489 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 490 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 491 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 492 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 493
emilmont 1:fdd22bb7aa52 494 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 495 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 496 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 497 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 498
emilmont 1:fdd22bb7aa52 499 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 500 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 501
emilmont 1:fdd22bb7aa52 502 /* Reading two inputs of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 503 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 504 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 505 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 506
emilmont 1:fdd22bb7aa52 507 /* Reading two inputs of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 508 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 509 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 510 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 511
emilmont 1:fdd22bb7aa52 512 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 513 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 514
emilmont 1:fdd22bb7aa52 515 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 516 k--;
emilmont 1:fdd22bb7aa52 517 }
emilmont 1:fdd22bb7aa52 518
emilmont 1:fdd22bb7aa52 519 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 520 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 521 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 522
emilmont 1:fdd22bb7aa52 523 while(k > 0u)
emilmont 1:fdd22bb7aa52 524 {
emilmont 1:fdd22bb7aa52 525 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 526 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 527
emilmont 1:fdd22bb7aa52 528 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 529 k--;
emilmont 1:fdd22bb7aa52 530 }
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 533 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 534
emilmont 1:fdd22bb7aa52 535 /* Increment the pointer pIn1 index, count by 1 */
emilmont 2:da51fb522205 536 count++;
emilmont 1:fdd22bb7aa52 537
emilmont 1:fdd22bb7aa52 538 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 2:da51fb522205 539 px = pIn1 + count;
emilmont 2:da51fb522205 540 py = pSrc2;
emilmont 1:fdd22bb7aa52 541
emilmont 1:fdd22bb7aa52 542 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 543 blkCnt--;
emilmont 1:fdd22bb7aa52 544 }
emilmont 1:fdd22bb7aa52 545 }
emilmont 1:fdd22bb7aa52 546 else
emilmont 1:fdd22bb7aa52 547 {
emilmont 1:fdd22bb7aa52 548 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 549 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 550 blkCnt = (uint32_t) blockSize2;
emilmont 1:fdd22bb7aa52 551
emilmont 1:fdd22bb7aa52 552 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 553 {
emilmont 1:fdd22bb7aa52 554 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 555 sum = 0;
emilmont 1:fdd22bb7aa52 556
emilmont 1:fdd22bb7aa52 557 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 558 k = srcBLen;
emilmont 1:fdd22bb7aa52 559
emilmont 1:fdd22bb7aa52 560 while(k > 0u)
emilmont 1:fdd22bb7aa52 561 {
emilmont 1:fdd22bb7aa52 562 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 563 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 564
emilmont 1:fdd22bb7aa52 565 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 566 k--;
emilmont 1:fdd22bb7aa52 567 }
emilmont 1:fdd22bb7aa52 568
emilmont 1:fdd22bb7aa52 569 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 570 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 571
emilmont 1:fdd22bb7aa52 572 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 573 count++;
emilmont 1:fdd22bb7aa52 574
emilmont 1:fdd22bb7aa52 575 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 576 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 577 py = pSrc2;
emilmont 1:fdd22bb7aa52 578
emilmont 1:fdd22bb7aa52 579 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 580 blkCnt--;
emilmont 1:fdd22bb7aa52 581 }
emilmont 1:fdd22bb7aa52 582 }
emilmont 1:fdd22bb7aa52 583
emilmont 1:fdd22bb7aa52 584
emilmont 1:fdd22bb7aa52 585 /* --------------------------
emilmont 1:fdd22bb7aa52 586 * Initializations of stage3
emilmont 1:fdd22bb7aa52 587 * -------------------------*/
emilmont 1:fdd22bb7aa52 588
emilmont 1:fdd22bb7aa52 589 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 590 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 591 * ....
emilmont 1:fdd22bb7aa52 592 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 593 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 594 */
emilmont 1:fdd22bb7aa52 595
emilmont 1:fdd22bb7aa52 596 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 597 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 598 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 599
emilmont 1:fdd22bb7aa52 600 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 601 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 602 px = pSrc1;
emilmont 1:fdd22bb7aa52 603
emilmont 1:fdd22bb7aa52 604 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 605 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 606 py = pSrc2;
emilmont 1:fdd22bb7aa52 607
emilmont 1:fdd22bb7aa52 608 /* -------------------
emilmont 1:fdd22bb7aa52 609 * Stage3 process
emilmont 1:fdd22bb7aa52 610 * ------------------*/
emilmont 1:fdd22bb7aa52 611
emilmont 1:fdd22bb7aa52 612 while(blockSize3 > 0)
emilmont 1:fdd22bb7aa52 613 {
emilmont 1:fdd22bb7aa52 614 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 615 sum = 0;
emilmont 1:fdd22bb7aa52 616
emilmont 1:fdd22bb7aa52 617 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 618 k = count >> 2u;
emilmont 1:fdd22bb7aa52 619
emilmont 1:fdd22bb7aa52 620 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 621 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 622 while(k > 0u)
emilmont 1:fdd22bb7aa52 623 {
emilmont 1:fdd22bb7aa52 624 /* Reading two inputs, x[srcALen - srcBLen + 1] and x[srcALen - srcBLen + 2] of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 625 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 626 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 627 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 628
emilmont 1:fdd22bb7aa52 629 /* Reading two inputs, y[srcBLen - 1] and y[srcBLen - 2] of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 630 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 631 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 632 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 633
emilmont 1:fdd22bb7aa52 634 /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 635 /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 636 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 637
emilmont 1:fdd22bb7aa52 638 /* Reading two inputs, x[srcALen - srcBLen + 3] and x[srcALen - srcBLen + 4] of SrcA buffer and packing */
emilmont 1:fdd22bb7aa52 639 in1 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 640 in2 = (q15_t) * px++;
emilmont 1:fdd22bb7aa52 641 input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 642
emilmont 1:fdd22bb7aa52 643 /* Reading two inputs, y[srcBLen - 3] and y[srcBLen - 4] of SrcB buffer and packing */
emilmont 1:fdd22bb7aa52 644 in1 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 645 in2 = (q15_t) * py--;
emilmont 1:fdd22bb7aa52 646 input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16);
emilmont 1:fdd22bb7aa52 647
emilmont 1:fdd22bb7aa52 648 /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 649 /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 650 sum = __SMLAD(input1, input2, sum);
emilmont 1:fdd22bb7aa52 651
emilmont 1:fdd22bb7aa52 652 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 653 k--;
emilmont 1:fdd22bb7aa52 654 }
emilmont 1:fdd22bb7aa52 655
emilmont 1:fdd22bb7aa52 656 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 657 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 658 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 659
emilmont 1:fdd22bb7aa52 660 while(k > 0u)
emilmont 1:fdd22bb7aa52 661 {
emilmont 1:fdd22bb7aa52 662 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 663 /* sum += x[srcALen-1] * y[srcBLen-1] */
emilmont 1:fdd22bb7aa52 664 sum += ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 665
emilmont 1:fdd22bb7aa52 666 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 667 k--;
emilmont 1:fdd22bb7aa52 668 }
emilmont 1:fdd22bb7aa52 669
emilmont 1:fdd22bb7aa52 670 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 671 *pOut++ = (q7_t) (__SSAT(sum >> 7, 8));
emilmont 1:fdd22bb7aa52 672
emilmont 1:fdd22bb7aa52 673 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 674 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 675 py = pSrc2;
emilmont 1:fdd22bb7aa52 676
emilmont 1:fdd22bb7aa52 677 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 678 count--;
emilmont 1:fdd22bb7aa52 679
emilmont 1:fdd22bb7aa52 680 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 681 blockSize3--;
emilmont 1:fdd22bb7aa52 682
emilmont 1:fdd22bb7aa52 683 }
emilmont 1:fdd22bb7aa52 684
emilmont 1:fdd22bb7aa52 685 /* set status as ARM_MATH_SUCCESS */
emilmont 1:fdd22bb7aa52 686 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 687 }
emilmont 1:fdd22bb7aa52 688
emilmont 1:fdd22bb7aa52 689 /* Return to application */
emilmont 1:fdd22bb7aa52 690 return (status);
emilmont 1:fdd22bb7aa52 691
emilmont 1:fdd22bb7aa52 692 #else
emilmont 1:fdd22bb7aa52 693
emilmont 1:fdd22bb7aa52 694 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 695
emilmont 1:fdd22bb7aa52 696 q7_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 697 q7_t *pIn2 = pSrcB; /* inputB pointer */
emilmont 1:fdd22bb7aa52 698 q31_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 699 uint32_t i, j; /* loop counters */
emilmont 1:fdd22bb7aa52 700 arm_status status; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 701
emilmont 1:fdd22bb7aa52 702 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 703 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 704 {
emilmont 1:fdd22bb7aa52 705 /* Set status as ARM_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 706 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 707 }
emilmont 1:fdd22bb7aa52 708 else
emilmont 1:fdd22bb7aa52 709 {
emilmont 1:fdd22bb7aa52 710 /* Loop to calculate convolution for output length number of values */
emilmont 1:fdd22bb7aa52 711 for (i = firstIndex; i <= (firstIndex + numPoints - 1); i++)
emilmont 1:fdd22bb7aa52 712 {
emilmont 1:fdd22bb7aa52 713 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 714 sum = 0;
emilmont 1:fdd22bb7aa52 715
emilmont 1:fdd22bb7aa52 716 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 717 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 718 {
emilmont 1:fdd22bb7aa52 719 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 720 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 721 {
emilmont 1:fdd22bb7aa52 722 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 723 sum += ((q15_t) pIn1[j] * (pIn2[i - j]));
emilmont 1:fdd22bb7aa52 724 }
emilmont 1:fdd22bb7aa52 725 }
emilmont 1:fdd22bb7aa52 726
emilmont 1:fdd22bb7aa52 727 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 728 pDst[i] = (q7_t) __SSAT((sum >> 7u), 8u);
emilmont 1:fdd22bb7aa52 729 }
emilmont 1:fdd22bb7aa52 730 /* set status as ARM_SUCCESS as there are no argument errors */
emilmont 1:fdd22bb7aa52 731 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 732 }
emilmont 1:fdd22bb7aa52 733 return (status);
emilmont 1:fdd22bb7aa52 734
mbed_official 3:7a284390b0ce 735 #endif /* #ifndef ARM_MATH_CM0_FAMILY */
emilmont 1:fdd22bb7aa52 736
emilmont 1:fdd22bb7aa52 737 }
emilmont 1:fdd22bb7aa52 738
emilmont 1:fdd22bb7aa52 739 /**
emilmont 1:fdd22bb7aa52 740 * @} end of PartialConv group
emilmont 1:fdd22bb7aa52 741 */