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_q15.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
Child:
5:3762170b6d4d
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_partial_q15.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Partial convolution of Q15 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 Q15 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 * Refer to <code>arm_conv_partial_fast_q15()</code> for a faster but less precise version of this function for Cortex-M3 and Cortex-M4.
emilmont 1:fdd22bb7aa52 64 *
emilmont 1:fdd22bb7aa52 65 * \par
emilmont 1:fdd22bb7aa52 66 * Refer the function <code>arm_conv_partial_opt_q15()</code> for a faster implementation of this function using scratch buffers.
emilmont 1:fdd22bb7aa52 67 *
emilmont 1:fdd22bb7aa52 68 */
emilmont 1:fdd22bb7aa52 69
emilmont 1:fdd22bb7aa52 70
emilmont 1:fdd22bb7aa52 71 arm_status arm_conv_partial_q15(
emilmont 1:fdd22bb7aa52 72 q15_t * pSrcA,
emilmont 1:fdd22bb7aa52 73 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 74 q15_t * pSrcB,
emilmont 1:fdd22bb7aa52 75 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 76 q15_t * pDst,
emilmont 1:fdd22bb7aa52 77 uint32_t firstIndex,
emilmont 1:fdd22bb7aa52 78 uint32_t numPoints)
emilmont 1:fdd22bb7aa52 79 {
emilmont 1:fdd22bb7aa52 80
emilmont 1:fdd22bb7aa52 81 #if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE)
emilmont 1:fdd22bb7aa52 82
emilmont 1:fdd22bb7aa52 83 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 84
emilmont 1:fdd22bb7aa52 85 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 86 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 87 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 88 q63_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 89 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 90 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 91 q15_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 92 q31_t x0, x1, x2, x3, c0; /* Temporary input variables */
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; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 96
emilmont 1:fdd22bb7aa52 97 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 98 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 99 {
emilmont 1:fdd22bb7aa52 100 /* Set status as ARM_MATH_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 101 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 102 }
emilmont 1:fdd22bb7aa52 103 else
emilmont 1:fdd22bb7aa52 104 {
emilmont 1:fdd22bb7aa52 105
emilmont 1:fdd22bb7aa52 106 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 107 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 108 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 109 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 110 {
emilmont 1:fdd22bb7aa52 111 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 112 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 113
emilmont 1:fdd22bb7aa52 114 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 115 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 116 }
emilmont 1:fdd22bb7aa52 117 else
emilmont 1:fdd22bb7aa52 118 {
emilmont 1:fdd22bb7aa52 119 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 120 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 121
emilmont 1:fdd22bb7aa52 122 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 123 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 124
emilmont 1:fdd22bb7aa52 125 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 126 j = srcBLen;
emilmont 1:fdd22bb7aa52 127 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 128 srcALen = j;
emilmont 1:fdd22bb7aa52 129 }
emilmont 1:fdd22bb7aa52 130
emilmont 1:fdd22bb7aa52 131 /* Conditions to check which loopCounter holds
emilmont 1:fdd22bb7aa52 132 * the first and last indices of the output samples to be calculated. */
emilmont 1:fdd22bb7aa52 133 check = firstIndex + numPoints;
emilmont 1:fdd22bb7aa52 134 blockSize3 = ((int32_t) check - (int32_t) srcALen);
emilmont 1:fdd22bb7aa52 135 blockSize3 = (blockSize3 > 0) ? blockSize3 : 0;
emilmont 1:fdd22bb7aa52 136 blockSize1 = (((int32_t) srcBLen - 1) - (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 137 blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1u)) ? blockSize1 :
emilmont 1:fdd22bb7aa52 138 (int32_t) numPoints) : 0;
emilmont 1:fdd22bb7aa52 139 blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) +
emilmont 1:fdd22bb7aa52 140 (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 141 blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
emilmont 1:fdd22bb7aa52 142
emilmont 1:fdd22bb7aa52 143 /* 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 144 /* The function is internally
emilmont 1:fdd22bb7aa52 145 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 146 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 147 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 148 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 149 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 150 * for every iteration. */
emilmont 1:fdd22bb7aa52 151
emilmont 1:fdd22bb7aa52 152 /* Set the output pointer to point to the firstIndex
emilmont 1:fdd22bb7aa52 153 * of the output sample to be calculated. */
emilmont 1:fdd22bb7aa52 154 pOut = pDst + firstIndex;
emilmont 1:fdd22bb7aa52 155
emilmont 1:fdd22bb7aa52 156 /* --------------------------
emilmont 1:fdd22bb7aa52 157 * Initializations of stage1
emilmont 1:fdd22bb7aa52 158 * -------------------------*/
emilmont 1:fdd22bb7aa52 159
emilmont 1:fdd22bb7aa52 160 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 161 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 162 * ....
emilmont 1:fdd22bb7aa52 163 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 164 */
emilmont 1:fdd22bb7aa52 165
emilmont 1:fdd22bb7aa52 166 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 167 The count variable holds the number of MAC operations performed.
emilmont 1:fdd22bb7aa52 168 Since the partial convolution starts from firstIndex
emilmont 1:fdd22bb7aa52 169 Number of Macs to be performed is firstIndex + 1 */
emilmont 1:fdd22bb7aa52 170 count = 1u + firstIndex;
emilmont 1:fdd22bb7aa52 171
emilmont 1:fdd22bb7aa52 172 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 173 px = pIn1;
emilmont 1:fdd22bb7aa52 174
emilmont 1:fdd22bb7aa52 175 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 176 pSrc2 = pIn2 + firstIndex;
emilmont 1:fdd22bb7aa52 177 py = pSrc2;
emilmont 1:fdd22bb7aa52 178
emilmont 1:fdd22bb7aa52 179 /* ------------------------
emilmont 1:fdd22bb7aa52 180 * Stage1 process
emilmont 1:fdd22bb7aa52 181 * ----------------------*/
emilmont 1:fdd22bb7aa52 182
emilmont 1:fdd22bb7aa52 183 /* For loop unrolling by 4, this stage is divided into two. */
emilmont 1:fdd22bb7aa52 184 /* First part of this stage computes the MAC operations less than 4 */
emilmont 1:fdd22bb7aa52 185 /* Second part of this stage computes the MAC operations greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 186
emilmont 1:fdd22bb7aa52 187 /* The first part of the stage starts here */
emilmont 1:fdd22bb7aa52 188 while((count < 4u) && (blockSize1 > 0))
emilmont 1:fdd22bb7aa52 189 {
emilmont 1:fdd22bb7aa52 190 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 191 sum = 0;
emilmont 1:fdd22bb7aa52 192
emilmont 1:fdd22bb7aa52 193 /* Loop over number of MAC operations between
emilmont 1:fdd22bb7aa52 194 * inputA samples and inputB samples */
emilmont 1:fdd22bb7aa52 195 k = count;
emilmont 1:fdd22bb7aa52 196
emilmont 1:fdd22bb7aa52 197 while(k > 0u)
emilmont 1:fdd22bb7aa52 198 {
emilmont 1:fdd22bb7aa52 199 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 200 sum = __SMLALD(*px++, *py--, sum);
emilmont 1:fdd22bb7aa52 201
emilmont 1:fdd22bb7aa52 202 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 203 k--;
emilmont 1:fdd22bb7aa52 204 }
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 207 *pOut++ = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 208
emilmont 1:fdd22bb7aa52 209 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 210 py = ++pSrc2;
emilmont 1:fdd22bb7aa52 211 px = pIn1;
emilmont 1:fdd22bb7aa52 212
emilmont 1:fdd22bb7aa52 213 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 214 count++;
emilmont 1:fdd22bb7aa52 215
emilmont 1:fdd22bb7aa52 216 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 217 blockSize1--;
emilmont 1:fdd22bb7aa52 218 }
emilmont 1:fdd22bb7aa52 219
emilmont 1:fdd22bb7aa52 220 /* The second part of the stage starts here */
emilmont 1:fdd22bb7aa52 221 /* The internal loop, over count, is unrolled by 4 */
emilmont 1:fdd22bb7aa52 222 /* To, read the last two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 223 * y[srcBLen] and y[srcBLen-1] coefficients, py is decremented by 1 */
emilmont 1:fdd22bb7aa52 224 py = py - 1;
emilmont 1:fdd22bb7aa52 225
emilmont 1:fdd22bb7aa52 226 while(blockSize1 > 0)
emilmont 1:fdd22bb7aa52 227 {
emilmont 1:fdd22bb7aa52 228 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 229 sum = 0;
emilmont 1:fdd22bb7aa52 230
emilmont 1:fdd22bb7aa52 231 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 232 k = count >> 2u;
emilmont 1:fdd22bb7aa52 233
emilmont 1:fdd22bb7aa52 234 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 235 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 236 while(k > 0u)
emilmont 1:fdd22bb7aa52 237 {
emilmont 1:fdd22bb7aa52 238 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 239 /* x[0], x[1] are multiplied with y[srcBLen - 1], y[srcBLen - 2] respectively */
emilmont 1:fdd22bb7aa52 240 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 241 /* x[2], x[3] are multiplied with y[srcBLen - 3], y[srcBLen - 4] respectively */
emilmont 1:fdd22bb7aa52 242 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 243
emilmont 1:fdd22bb7aa52 244 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 245 k--;
emilmont 1:fdd22bb7aa52 246 }
emilmont 1:fdd22bb7aa52 247
emilmont 1:fdd22bb7aa52 248 /* For the next MAC operations, the pointer py is used without SIMD
emilmont 1:fdd22bb7aa52 249 * So, py is incremented by 1 */
emilmont 1:fdd22bb7aa52 250 py = py + 1u;
emilmont 1:fdd22bb7aa52 251
emilmont 1:fdd22bb7aa52 252 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 253 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 254 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 255
emilmont 1:fdd22bb7aa52 256 while(k > 0u)
emilmont 1:fdd22bb7aa52 257 {
emilmont 1:fdd22bb7aa52 258 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 259 sum = __SMLALD(*px++, *py--, sum);
emilmont 1:fdd22bb7aa52 260
emilmont 1:fdd22bb7aa52 261 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 262 k--;
emilmont 1:fdd22bb7aa52 263 }
emilmont 1:fdd22bb7aa52 264
emilmont 1:fdd22bb7aa52 265 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 266 *pOut++ = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 267
emilmont 1:fdd22bb7aa52 268 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 269 py = ++pSrc2 - 1u;
emilmont 1:fdd22bb7aa52 270 px = pIn1;
emilmont 1:fdd22bb7aa52 271
emilmont 1:fdd22bb7aa52 272 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 273 count++;
emilmont 1:fdd22bb7aa52 274
emilmont 1:fdd22bb7aa52 275 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 276 blockSize1--;
emilmont 1:fdd22bb7aa52 277 }
emilmont 1:fdd22bb7aa52 278
emilmont 1:fdd22bb7aa52 279 /* --------------------------
emilmont 1:fdd22bb7aa52 280 * Initializations of stage2
emilmont 1:fdd22bb7aa52 281 * ------------------------*/
emilmont 1:fdd22bb7aa52 282
emilmont 1:fdd22bb7aa52 283 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 284 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 285 * ....
emilmont 1:fdd22bb7aa52 286 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 287 */
emilmont 1:fdd22bb7aa52 288
emilmont 1:fdd22bb7aa52 289 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 290 px = pIn1;
emilmont 1:fdd22bb7aa52 291
emilmont 1:fdd22bb7aa52 292 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 293 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 294 py = pSrc2;
emilmont 1:fdd22bb7aa52 295
emilmont 1:fdd22bb7aa52 296 /* count is the index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 297 count = 0u;
emilmont 1:fdd22bb7aa52 298
emilmont 1:fdd22bb7aa52 299
emilmont 1:fdd22bb7aa52 300 /* --------------------
emilmont 1:fdd22bb7aa52 301 * Stage2 process
emilmont 1:fdd22bb7aa52 302 * -------------------*/
emilmont 1:fdd22bb7aa52 303
emilmont 1:fdd22bb7aa52 304 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 305 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 306 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 307 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 308 {
emilmont 1:fdd22bb7aa52 309 /* Loop unroll over blockSize2, by 4 */
emilmont 1:fdd22bb7aa52 310 blkCnt = blockSize2 >> 2u;
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 313 {
emilmont 1:fdd22bb7aa52 314 py = py - 1u;
emilmont 1:fdd22bb7aa52 315
emilmont 1:fdd22bb7aa52 316 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 317 acc0 = 0;
emilmont 1:fdd22bb7aa52 318 acc1 = 0;
emilmont 1:fdd22bb7aa52 319 acc2 = 0;
emilmont 1:fdd22bb7aa52 320 acc3 = 0;
emilmont 1:fdd22bb7aa52 321
emilmont 1:fdd22bb7aa52 322
emilmont 1:fdd22bb7aa52 323 /* read x[0], x[1] samples */
emilmont 1:fdd22bb7aa52 324 x0 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 325 /* read x[1], x[2] samples */
emilmont 1:fdd22bb7aa52 326 x1 = _SIMD32_OFFSET(px+1);
emilmont 2:da51fb522205 327 px+= 2u;
emilmont 1:fdd22bb7aa52 328
emilmont 1:fdd22bb7aa52 329
emilmont 1:fdd22bb7aa52 330 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 331 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 332
emilmont 1:fdd22bb7aa52 333 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 334 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 335 do
emilmont 1:fdd22bb7aa52 336 {
emilmont 1:fdd22bb7aa52 337 /* Read the last two inputB samples using SIMD:
emilmont 1:fdd22bb7aa52 338 * y[srcBLen - 1] and y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 339 c0 = *__SIMD32(py)--;
emilmont 1:fdd22bb7aa52 340
emilmont 1:fdd22bb7aa52 341 /* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 342 acc0 = __SMLALDX(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 343
emilmont 1:fdd22bb7aa52 344 /* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 345 acc1 = __SMLALDX(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 346
emilmont 1:fdd22bb7aa52 347 /* Read x[2], x[3] */
emilmont 1:fdd22bb7aa52 348 x2 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 349
emilmont 1:fdd22bb7aa52 350 /* Read x[3], x[4] */
emilmont 1:fdd22bb7aa52 351 x3 = _SIMD32_OFFSET(px+1);
emilmont 1:fdd22bb7aa52 352
emilmont 1:fdd22bb7aa52 353 /* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 354 acc2 = __SMLALDX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 355
emilmont 1:fdd22bb7aa52 356 /* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 357 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 358
emilmont 1:fdd22bb7aa52 359 /* Read y[srcBLen - 3] and y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 360 c0 = *__SIMD32(py)--;
emilmont 1:fdd22bb7aa52 361
emilmont 1:fdd22bb7aa52 362 /* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 363 acc0 = __SMLALDX(x2, c0, acc0);
emilmont 1:fdd22bb7aa52 364
emilmont 1:fdd22bb7aa52 365 /* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 366 acc1 = __SMLALDX(x3, c0, acc1);
emilmont 1:fdd22bb7aa52 367
emilmont 1:fdd22bb7aa52 368 /* Read x[4], x[5] */
emilmont 1:fdd22bb7aa52 369 x0 = _SIMD32_OFFSET(px+2);
emilmont 1:fdd22bb7aa52 370
emilmont 1:fdd22bb7aa52 371 /* Read x[5], x[6] */
emilmont 1:fdd22bb7aa52 372 x1 = _SIMD32_OFFSET(px+3);
emilmont 2:da51fb522205 373 px += 4u;
emilmont 1:fdd22bb7aa52 374
emilmont 1:fdd22bb7aa52 375 /* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 376 acc2 = __SMLALDX(x0, c0, acc2);
emilmont 1:fdd22bb7aa52 377
emilmont 1:fdd22bb7aa52 378 /* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 379 acc3 = __SMLALDX(x1, c0, acc3);
emilmont 1:fdd22bb7aa52 380
emilmont 1:fdd22bb7aa52 381 } while(--k);
emilmont 1:fdd22bb7aa52 382
emilmont 1:fdd22bb7aa52 383 /* For the next MAC operations, SIMD is not used
emilmont 1:fdd22bb7aa52 384 * So, the 16 bit pointer if inputB, py is updated */
emilmont 1:fdd22bb7aa52 385
emilmont 1:fdd22bb7aa52 386 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 387 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 388 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 389
emilmont 1:fdd22bb7aa52 390 if(k == 1u)
emilmont 1:fdd22bb7aa52 391 {
emilmont 1:fdd22bb7aa52 392 /* Read y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 393 c0 = *(py+1);
emilmont 1:fdd22bb7aa52 394
emilmont 1:fdd22bb7aa52 395 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 396
emilmont 1:fdd22bb7aa52 397 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 398
emilmont 1:fdd22bb7aa52 399 #else
emilmont 1:fdd22bb7aa52 400
emilmont 1:fdd22bb7aa52 401 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 402
emilmont 1:fdd22bb7aa52 403 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 404
emilmont 1:fdd22bb7aa52 405 /* Read x[7] */
emilmont 1:fdd22bb7aa52 406 x3 = *__SIMD32(px);
emilmont 2:da51fb522205 407 px++;
emilmont 1:fdd22bb7aa52 408
emilmont 1:fdd22bb7aa52 409 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 410 acc0 = __SMLALD(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 411 acc1 = __SMLALD(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 412 acc2 = __SMLALDX(x1, c0, acc2);
emilmont 1:fdd22bb7aa52 413 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 414 }
emilmont 1:fdd22bb7aa52 415
emilmont 1:fdd22bb7aa52 416 if(k == 2u)
emilmont 1:fdd22bb7aa52 417 {
emilmont 1:fdd22bb7aa52 418 /* Read y[srcBLen - 5], y[srcBLen - 6] */
emilmont 1:fdd22bb7aa52 419 c0 = _SIMD32_OFFSET(py);
emilmont 1:fdd22bb7aa52 420
emilmont 1:fdd22bb7aa52 421 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 422 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 423
emilmont 1:fdd22bb7aa52 424 /* Read x[9] */
emilmont 1:fdd22bb7aa52 425 x2 = _SIMD32_OFFSET(px+1);
emilmont 2:da51fb522205 426 px += 2u;
emilmont 1:fdd22bb7aa52 427
emilmont 1:fdd22bb7aa52 428 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 429 acc0 = __SMLALDX(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 430 acc1 = __SMLALDX(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 431 acc2 = __SMLALDX(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 432 acc3 = __SMLALDX(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 433 }
emilmont 1:fdd22bb7aa52 434
emilmont 1:fdd22bb7aa52 435 if(k == 3u)
emilmont 1:fdd22bb7aa52 436 {
emilmont 1:fdd22bb7aa52 437 /* Read y[srcBLen - 5], y[srcBLen - 6] */
emilmont 1:fdd22bb7aa52 438 c0 = _SIMD32_OFFSET(py);
emilmont 1:fdd22bb7aa52 439
emilmont 1:fdd22bb7aa52 440 /* Read x[7], x[8] */
emilmont 1:fdd22bb7aa52 441 x3 = *__SIMD32(px);
emilmont 1:fdd22bb7aa52 442
emilmont 1:fdd22bb7aa52 443 /* Read x[9] */
emilmont 1:fdd22bb7aa52 444 x2 = _SIMD32_OFFSET(px+1);
emilmont 1:fdd22bb7aa52 445
emilmont 1:fdd22bb7aa52 446 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 447 acc0 = __SMLALDX(x0, c0, acc0);
emilmont 1:fdd22bb7aa52 448 acc1 = __SMLALDX(x1, c0, acc1);
emilmont 1:fdd22bb7aa52 449 acc2 = __SMLALDX(x3, c0, acc2);
emilmont 1:fdd22bb7aa52 450 acc3 = __SMLALDX(x2, c0, acc3);
emilmont 1:fdd22bb7aa52 451
emilmont 2:da51fb522205 452 c0 = *(py-1);
emilmont 1:fdd22bb7aa52 453
emilmont 1:fdd22bb7aa52 454 #ifdef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 455
emilmont 1:fdd22bb7aa52 456 c0 = c0 << 16u;
emilmont 1:fdd22bb7aa52 457 #else
emilmont 1:fdd22bb7aa52 458
emilmont 1:fdd22bb7aa52 459 c0 = c0 & 0x0000FFFF;
emilmont 1:fdd22bb7aa52 460 #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 461
emilmont 1:fdd22bb7aa52 462 /* Read x[10] */
emilmont 1:fdd22bb7aa52 463 x3 = _SIMD32_OFFSET(px+2);
emilmont 2:da51fb522205 464 px += 3u;
emilmont 1:fdd22bb7aa52 465
emilmont 1:fdd22bb7aa52 466 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 467 acc0 = __SMLALDX(x1, c0, acc0);
emilmont 1:fdd22bb7aa52 468 acc1 = __SMLALD(x2, c0, acc1);
emilmont 1:fdd22bb7aa52 469 acc2 = __SMLALDX(x2, c0, acc2);
emilmont 1:fdd22bb7aa52 470 acc3 = __SMLALDX(x3, c0, acc3);
emilmont 1:fdd22bb7aa52 471 }
emilmont 1:fdd22bb7aa52 472
emilmont 1:fdd22bb7aa52 473
emilmont 1:fdd22bb7aa52 474 /* Store the results in the accumulators in the destination buffer. */
emilmont 1:fdd22bb7aa52 475
emilmont 1:fdd22bb7aa52 476 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 477
emilmont 1:fdd22bb7aa52 478 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 479 __PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 480 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 481 __PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 482
emilmont 1:fdd22bb7aa52 483 #else
emilmont 1:fdd22bb7aa52 484
emilmont 1:fdd22bb7aa52 485 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 486 __PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 487 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 488 __PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 489
emilmont 1:fdd22bb7aa52 490 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 491
emilmont 1:fdd22bb7aa52 492 /* Increment the pointer pIn1 index, count by 4 */
emilmont 1:fdd22bb7aa52 493 count += 4u;
emilmont 1:fdd22bb7aa52 494
emilmont 1:fdd22bb7aa52 495 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 496 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 497 py = pSrc2;
emilmont 1:fdd22bb7aa52 498
emilmont 1:fdd22bb7aa52 499 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 500 blkCnt--;
emilmont 1:fdd22bb7aa52 501 }
emilmont 1:fdd22bb7aa52 502
emilmont 1:fdd22bb7aa52 503 /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 504 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 505 blkCnt = (uint32_t) blockSize2 % 0x4u;
emilmont 2:da51fb522205 506
emilmont 1:fdd22bb7aa52 507 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 508 {
emilmont 1:fdd22bb7aa52 509 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 510 sum = 0;
emilmont 1:fdd22bb7aa52 511
emilmont 1:fdd22bb7aa52 512 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 513 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 514
emilmont 1:fdd22bb7aa52 515 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 516 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 517 while(k > 0u)
emilmont 1:fdd22bb7aa52 518 {
emilmont 1:fdd22bb7aa52 519 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 520 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 521 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 522 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 523 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 524
emilmont 1:fdd22bb7aa52 525 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 526 k--;
emilmont 1:fdd22bb7aa52 527 }
emilmont 1:fdd22bb7aa52 528
emilmont 1:fdd22bb7aa52 529 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 530 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 531 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 532
emilmont 1:fdd22bb7aa52 533 while(k > 0u)
emilmont 1:fdd22bb7aa52 534 {
emilmont 1:fdd22bb7aa52 535 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 536 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 537
emilmont 1:fdd22bb7aa52 538 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 539 k--;
emilmont 1:fdd22bb7aa52 540 }
emilmont 1:fdd22bb7aa52 541
emilmont 1:fdd22bb7aa52 542 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 543 *pOut++ = (q15_t) (__SSAT(sum >> 15, 16));
emilmont 1:fdd22bb7aa52 544
emilmont 1:fdd22bb7aa52 545 /* Increment the pointer pIn1 index, count by 1 */
emilmont 1:fdd22bb7aa52 546 count++;
emilmont 1:fdd22bb7aa52 547
emilmont 1:fdd22bb7aa52 548 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 549 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 550 py = pSrc2;
emilmont 1:fdd22bb7aa52 551
emilmont 1:fdd22bb7aa52 552 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 553 blkCnt--;
emilmont 1:fdd22bb7aa52 554 }
emilmont 1:fdd22bb7aa52 555 }
emilmont 1:fdd22bb7aa52 556 else
emilmont 1:fdd22bb7aa52 557 {
emilmont 1:fdd22bb7aa52 558 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 559 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 560 blkCnt = (uint32_t) blockSize2;
emilmont 1:fdd22bb7aa52 561
emilmont 1:fdd22bb7aa52 562 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 563 {
emilmont 1:fdd22bb7aa52 564 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 565 sum = 0;
emilmont 1:fdd22bb7aa52 566
emilmont 1:fdd22bb7aa52 567 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 568 k = srcBLen;
emilmont 1:fdd22bb7aa52 569
emilmont 1:fdd22bb7aa52 570 while(k > 0u)
emilmont 1:fdd22bb7aa52 571 {
emilmont 1:fdd22bb7aa52 572 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 573 sum += (q63_t) ((q31_t) * px++ * *py--);
emilmont 1:fdd22bb7aa52 574
emilmont 1:fdd22bb7aa52 575 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 576 k--;
emilmont 1:fdd22bb7aa52 577 }
emilmont 1:fdd22bb7aa52 578
emilmont 1:fdd22bb7aa52 579 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 580 *pOut++ = (q15_t) (__SSAT(sum >> 15, 16));
emilmont 1:fdd22bb7aa52 581
emilmont 1:fdd22bb7aa52 582 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 583 count++;
emilmont 1:fdd22bb7aa52 584
emilmont 1:fdd22bb7aa52 585 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 586 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 587 py = pSrc2;
emilmont 1:fdd22bb7aa52 588
emilmont 1:fdd22bb7aa52 589 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 590 blkCnt--;
emilmont 1:fdd22bb7aa52 591 }
emilmont 1:fdd22bb7aa52 592 }
emilmont 1:fdd22bb7aa52 593
emilmont 1:fdd22bb7aa52 594
emilmont 1:fdd22bb7aa52 595 /* --------------------------
emilmont 1:fdd22bb7aa52 596 * Initializations of stage3
emilmont 1:fdd22bb7aa52 597 * -------------------------*/
emilmont 1:fdd22bb7aa52 598
emilmont 1:fdd22bb7aa52 599 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 600 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 601 * ....
emilmont 1:fdd22bb7aa52 602 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 603 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 604 */
emilmont 1:fdd22bb7aa52 605
emilmont 1:fdd22bb7aa52 606 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 607 The count variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 608 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 609
emilmont 1:fdd22bb7aa52 610 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 611 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 612 px = pSrc1;
emilmont 1:fdd22bb7aa52 613
emilmont 1:fdd22bb7aa52 614 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 615 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 616 pIn2 = pSrc2 - 1u;
emilmont 1:fdd22bb7aa52 617 py = pIn2;
emilmont 1:fdd22bb7aa52 618
emilmont 1:fdd22bb7aa52 619 /* -------------------
emilmont 1:fdd22bb7aa52 620 * Stage3 process
emilmont 1:fdd22bb7aa52 621 * ------------------*/
emilmont 1:fdd22bb7aa52 622
emilmont 1:fdd22bb7aa52 623 /* For loop unrolling by 4, this stage is divided into two. */
emilmont 1:fdd22bb7aa52 624 /* First part of this stage computes the MAC operations greater than 4 */
emilmont 1:fdd22bb7aa52 625 /* Second part of this stage computes the MAC operations less than or equal to 4 */
emilmont 1:fdd22bb7aa52 626
emilmont 1:fdd22bb7aa52 627 /* The first part of the stage starts here */
emilmont 1:fdd22bb7aa52 628 j = count >> 2u;
emilmont 1:fdd22bb7aa52 629
emilmont 1:fdd22bb7aa52 630 while((j > 0u) && (blockSize3 > 0))
emilmont 1:fdd22bb7aa52 631 {
emilmont 1:fdd22bb7aa52 632 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 633 sum = 0;
emilmont 1:fdd22bb7aa52 634
emilmont 1:fdd22bb7aa52 635 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 636 k = count >> 2u;
emilmont 1:fdd22bb7aa52 637
emilmont 1:fdd22bb7aa52 638 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 639 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 640 while(k > 0u)
emilmont 1:fdd22bb7aa52 641 {
emilmont 1:fdd22bb7aa52 642 /* x[srcALen - srcBLen + 1], x[srcALen - srcBLen + 2] are multiplied
emilmont 1:fdd22bb7aa52 643 * with y[srcBLen - 1], y[srcBLen - 2] respectively */
emilmont 1:fdd22bb7aa52 644 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 645 /* x[srcALen - srcBLen + 3], x[srcALen - srcBLen + 4] are multiplied
emilmont 1:fdd22bb7aa52 646 * with y[srcBLen - 3], y[srcBLen - 4] respectively */
emilmont 1:fdd22bb7aa52 647 sum = __SMLALDX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
emilmont 1:fdd22bb7aa52 648
emilmont 1:fdd22bb7aa52 649 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 650 k--;
emilmont 1:fdd22bb7aa52 651 }
emilmont 1:fdd22bb7aa52 652
emilmont 1:fdd22bb7aa52 653 /* For the next MAC operations, the pointer py is used without SIMD
emilmont 1:fdd22bb7aa52 654 * So, py is incremented by 1 */
emilmont 1:fdd22bb7aa52 655 py = py + 1u;
emilmont 1:fdd22bb7aa52 656
emilmont 1:fdd22bb7aa52 657 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 658 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 659 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 660
emilmont 1:fdd22bb7aa52 661 while(k > 0u)
emilmont 1:fdd22bb7aa52 662 {
emilmont 1:fdd22bb7aa52 663 /* sum += x[srcALen - srcBLen + 5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 664 sum = __SMLALD(*px++, *py--, sum);
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++ = (q15_t) (__SSAT((sum >> 15), 16));
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 = pIn2;
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 j--;
emilmont 1:fdd22bb7aa52 684 }
emilmont 1:fdd22bb7aa52 685
emilmont 1:fdd22bb7aa52 686 /* The second part of the stage starts here */
emilmont 1:fdd22bb7aa52 687 /* SIMD is not used for the next MAC operations,
emilmont 1:fdd22bb7aa52 688 * so pointer py is updated to read only one sample at a time */
emilmont 1:fdd22bb7aa52 689 py = py + 1u;
emilmont 1:fdd22bb7aa52 690
emilmont 1:fdd22bb7aa52 691 while(blockSize3 > 0)
emilmont 1:fdd22bb7aa52 692 {
emilmont 1:fdd22bb7aa52 693 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 694 sum = 0;
emilmont 1:fdd22bb7aa52 695
emilmont 1:fdd22bb7aa52 696 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 697 k = count;
emilmont 1:fdd22bb7aa52 698
emilmont 1:fdd22bb7aa52 699 while(k > 0u)
emilmont 1:fdd22bb7aa52 700 {
emilmont 1:fdd22bb7aa52 701 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 702 /* sum += x[srcALen-1] * y[srcBLen-1] */
emilmont 1:fdd22bb7aa52 703 sum = __SMLALD(*px++, *py--, sum);
emilmont 1:fdd22bb7aa52 704
emilmont 1:fdd22bb7aa52 705 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 706 k--;
emilmont 1:fdd22bb7aa52 707 }
emilmont 1:fdd22bb7aa52 708
emilmont 1:fdd22bb7aa52 709 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 710 *pOut++ = (q15_t) (__SSAT((sum >> 15), 16));
emilmont 1:fdd22bb7aa52 711
emilmont 1:fdd22bb7aa52 712 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 713 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 714 py = pSrc2;
emilmont 1:fdd22bb7aa52 715
emilmont 1:fdd22bb7aa52 716 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 717 count--;
emilmont 1:fdd22bb7aa52 718
emilmont 1:fdd22bb7aa52 719 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 720 blockSize3--;
emilmont 1:fdd22bb7aa52 721 }
emilmont 1:fdd22bb7aa52 722
emilmont 1:fdd22bb7aa52 723 /* set status as ARM_MATH_SUCCESS */
emilmont 1:fdd22bb7aa52 724 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 725 }
emilmont 1:fdd22bb7aa52 726
emilmont 1:fdd22bb7aa52 727 /* Return to application */
emilmont 1:fdd22bb7aa52 728 return (status);
emilmont 1:fdd22bb7aa52 729
emilmont 1:fdd22bb7aa52 730 #else
emilmont 1:fdd22bb7aa52 731
emilmont 1:fdd22bb7aa52 732 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 733
emilmont 1:fdd22bb7aa52 734 q15_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 735 q15_t *pIn2 = pSrcB; /* inputB pointer */
emilmont 1:fdd22bb7aa52 736 q63_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 737 uint32_t i, j; /* loop counters */
emilmont 1:fdd22bb7aa52 738 arm_status status; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 739
emilmont 1:fdd22bb7aa52 740 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 741 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 742 {
emilmont 1:fdd22bb7aa52 743 /* Set status as ARM_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 744 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 745 }
emilmont 1:fdd22bb7aa52 746 else
emilmont 1:fdd22bb7aa52 747 {
emilmont 1:fdd22bb7aa52 748 /* Loop to calculate convolution for output length number of values */
emilmont 1:fdd22bb7aa52 749 for (i = firstIndex; i <= (firstIndex + numPoints - 1); i++)
emilmont 1:fdd22bb7aa52 750 {
emilmont 1:fdd22bb7aa52 751 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 752 sum = 0;
emilmont 1:fdd22bb7aa52 753
emilmont 1:fdd22bb7aa52 754 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 755 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 756 {
emilmont 1:fdd22bb7aa52 757 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 758 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 759 {
emilmont 1:fdd22bb7aa52 760 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 761 sum += ((q31_t) pIn1[j] * (pIn2[i - j]));
emilmont 1:fdd22bb7aa52 762 }
emilmont 1:fdd22bb7aa52 763 }
emilmont 1:fdd22bb7aa52 764
emilmont 1:fdd22bb7aa52 765 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 766 pDst[i] = (q15_t) __SSAT((sum >> 15u), 16u);
emilmont 1:fdd22bb7aa52 767 }
emilmont 1:fdd22bb7aa52 768 /* set status as ARM_SUCCESS as there are no argument errors */
emilmont 1:fdd22bb7aa52 769 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 770 }
emilmont 1:fdd22bb7aa52 771 return (status);
emilmont 1:fdd22bb7aa52 772
emilmont 1:fdd22bb7aa52 773 #endif /* #if (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM3)) && !defined(UNALIGNED_SUPPORT_DISABLE) */
emilmont 1:fdd22bb7aa52 774
emilmont 1:fdd22bb7aa52 775 }
emilmont 1:fdd22bb7aa52 776
emilmont 1:fdd22bb7aa52 777 /**
emilmont 1:fdd22bb7aa52 778 * @} end of PartialConv group
emilmont 1:fdd22bb7aa52 779 */