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_q31.c@5:3762170b6d4d, 2015-11-20 (annotated)

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



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



Added option to build rpc library. closes #1426

Who changed what in which revision?

UserRevisionLine numberNew contents of line
emilmont 1:fdd22bb7aa52 1 /* ----------------------------------------------------------------------
mbed_official 5:3762170b6d4d 2 * Copyright (C) 2010-2014 ARM Limited. All rights reserved.
emilmont 1:fdd22bb7aa52 3 *
mbed_official 5:3762170b6d4d 4 * $Date: 19. March 2015
mbed_official 5:3762170b6d4d 5 * $Revision: V.1.4.5
emilmont 1:fdd22bb7aa52 6 *
emilmont 2:da51fb522205 7 * Project: CMSIS DSP Library
emilmont 2:da51fb522205 8 * Title: arm_conv_partial_q31.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 2:da51fb522205 10 * Description: Partial convolution of Q31 sequences.
emilmont 1:fdd22bb7aa52 11 *
emilmont 1:fdd22bb7aa52 12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
emilmont 1:fdd22bb7aa52 13 *
mbed_official 3:7a284390b0ce 14 * Redistribution and use in source and binary forms, with or without
mbed_official 3:7a284390b0ce 15 * modification, are permitted provided that the following conditions
mbed_official 3:7a284390b0ce 16 * are met:
mbed_official 3:7a284390b0ce 17 * - Redistributions of source code must retain the above copyright
mbed_official 3:7a284390b0ce 18 * notice, this list of conditions and the following disclaimer.
mbed_official 3:7a284390b0ce 19 * - Redistributions in binary form must reproduce the above copyright
mbed_official 3:7a284390b0ce 20 * notice, this list of conditions and the following disclaimer in
mbed_official 3:7a284390b0ce 21 * the documentation and/or other materials provided with the
mbed_official 3:7a284390b0ce 22 * distribution.
mbed_official 3:7a284390b0ce 23 * - Neither the name of ARM LIMITED nor the names of its contributors
mbed_official 3:7a284390b0ce 24 * may be used to endorse or promote products derived from this
mbed_official 3:7a284390b0ce 25 * software without specific prior written permission.
mbed_official 3:7a284390b0ce 26 *
mbed_official 3:7a284390b0ce 27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
mbed_official 3:7a284390b0ce 28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
mbed_official 3:7a284390b0ce 29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
mbed_official 3:7a284390b0ce 30 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
mbed_official 3:7a284390b0ce 31 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
mbed_official 3:7a284390b0ce 32 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
mbed_official 3:7a284390b0ce 33 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
mbed_official 3:7a284390b0ce 34 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
mbed_official 3:7a284390b0ce 35 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
mbed_official 3:7a284390b0ce 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
mbed_official 3:7a284390b0ce 37 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
mbed_official 3:7a284390b0ce 38 * POSSIBILITY OF SUCH DAMAGE.
emilmont 1:fdd22bb7aa52 39 * -------------------------------------------------------------------- */
emilmont 1:fdd22bb7aa52 40
emilmont 1:fdd22bb7aa52 41 #include "arm_math.h"
emilmont 1:fdd22bb7aa52 42
emilmont 1:fdd22bb7aa52 43 /**
emilmont 1:fdd22bb7aa52 44 * @ingroup groupFilters
emilmont 1:fdd22bb7aa52 45 */
emilmont 1:fdd22bb7aa52 46
emilmont 1:fdd22bb7aa52 47 /**
emilmont 1:fdd22bb7aa52 48 * @addtogroup 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 Q31 sequences.
emilmont 1:fdd22bb7aa52 54 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 58 * @param[out] *pDst points to the location where the output result is written.
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 * See <code>arm_conv_partial_fast_q31()</code> for a faster but less precise implementation of this function for Cortex-M3 and Cortex-M4.
emilmont 1:fdd22bb7aa52 64 */
emilmont 1:fdd22bb7aa52 65
emilmont 1:fdd22bb7aa52 66 arm_status arm_conv_partial_q31(
emilmont 1:fdd22bb7aa52 67 q31_t * pSrcA,
emilmont 1:fdd22bb7aa52 68 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 69 q31_t * pSrcB,
emilmont 1:fdd22bb7aa52 70 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 71 q31_t * pDst,
emilmont 1:fdd22bb7aa52 72 uint32_t firstIndex,
emilmont 1:fdd22bb7aa52 73 uint32_t numPoints)
emilmont 1:fdd22bb7aa52 74 {
emilmont 1:fdd22bb7aa52 75
emilmont 1:fdd22bb7aa52 76
mbed_official 3:7a284390b0ce 77 #ifndef ARM_MATH_CM0_FAMILY
emilmont 1:fdd22bb7aa52 78
emilmont 1:fdd22bb7aa52 79 /* Run the below code for Cortex-M4 and Cortex-M3 */
emilmont 1:fdd22bb7aa52 80
emilmont 1:fdd22bb7aa52 81 q31_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 82 q31_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 83 q31_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 84 q31_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 85 q31_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 86 q31_t *pSrc1, *pSrc2; /* Intermediate pointers */
emilmont 1:fdd22bb7aa52 87 q63_t sum, acc0, acc1, acc2; /* Accumulator */
emilmont 1:fdd22bb7aa52 88 q31_t x0, x1, x2, c0;
emilmont 1:fdd22bb7aa52 89 uint32_t j, k, count, check, blkCnt;
emilmont 1:fdd22bb7aa52 90 int32_t blockSize1, blockSize2, blockSize3; /* loop counter */
emilmont 1:fdd22bb7aa52 91 arm_status status; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 92
emilmont 1:fdd22bb7aa52 93
emilmont 1:fdd22bb7aa52 94 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 95 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 96 {
emilmont 1:fdd22bb7aa52 97 /* Set status as ARM_MATH_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 98 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 99 }
emilmont 1:fdd22bb7aa52 100 else
emilmont 1:fdd22bb7aa52 101 {
emilmont 1:fdd22bb7aa52 102
emilmont 1:fdd22bb7aa52 103 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 104 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 105 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 106 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 107 {
emilmont 1:fdd22bb7aa52 108 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 109 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 110
emilmont 1:fdd22bb7aa52 111 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 112 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 113 }
emilmont 1:fdd22bb7aa52 114 else
emilmont 1:fdd22bb7aa52 115 {
emilmont 1:fdd22bb7aa52 116 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 117 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 118
emilmont 1:fdd22bb7aa52 119 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 120 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 121
emilmont 1:fdd22bb7aa52 122 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 123 j = srcBLen;
emilmont 1:fdd22bb7aa52 124 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 125 srcALen = j;
emilmont 1:fdd22bb7aa52 126 }
emilmont 1:fdd22bb7aa52 127
emilmont 1:fdd22bb7aa52 128 /* Conditions to check which loopCounter holds
emilmont 1:fdd22bb7aa52 129 * the first and last indices of the output samples to be calculated. */
emilmont 1:fdd22bb7aa52 130 check = firstIndex + numPoints;
mbed_official 5:3762170b6d4d 131 blockSize3 = ((int32_t)check > (int32_t)srcALen) ? (int32_t)check - (int32_t)srcALen : 0;
mbed_official 5:3762170b6d4d 132 blockSize3 = ((int32_t)firstIndex > (int32_t)srcALen - 1) ? blockSize3 - (int32_t)firstIndex + (int32_t)srcALen : blockSize3;
emilmont 1:fdd22bb7aa52 133 blockSize1 = (((int32_t) srcBLen - 1) - (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 134 blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1u)) ? blockSize1 :
emilmont 1:fdd22bb7aa52 135 (int32_t) numPoints) : 0;
emilmont 1:fdd22bb7aa52 136 blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) +
emilmont 1:fdd22bb7aa52 137 (int32_t) firstIndex);
emilmont 1:fdd22bb7aa52 138 blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
emilmont 1:fdd22bb7aa52 139
emilmont 1:fdd22bb7aa52 140 /* 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 141 /* The function is internally
emilmont 1:fdd22bb7aa52 142 * divided into three stages according to the number of multiplications that has to be
emilmont 1:fdd22bb7aa52 143 * taken place between inputA samples and inputB samples. In the first stage of the
emilmont 1:fdd22bb7aa52 144 * algorithm, the multiplications increase by one for every iteration.
emilmont 1:fdd22bb7aa52 145 * In the second stage of the algorithm, srcBLen number of multiplications are done.
emilmont 1:fdd22bb7aa52 146 * In the third stage of the algorithm, the multiplications decrease by one
emilmont 1:fdd22bb7aa52 147 * for every iteration. */
emilmont 1:fdd22bb7aa52 148
emilmont 1:fdd22bb7aa52 149 /* Set the output pointer to point to the firstIndex
emilmont 1:fdd22bb7aa52 150 * of the output sample to be calculated. */
emilmont 1:fdd22bb7aa52 151 pOut = pDst + firstIndex;
emilmont 1:fdd22bb7aa52 152
emilmont 1:fdd22bb7aa52 153 /* --------------------------
emilmont 1:fdd22bb7aa52 154 * Initializations of stage1
emilmont 1:fdd22bb7aa52 155 * -------------------------*/
emilmont 1:fdd22bb7aa52 156
emilmont 1:fdd22bb7aa52 157 /* sum = x[0] * y[0]
emilmont 1:fdd22bb7aa52 158 * sum = x[0] * y[1] + x[1] * y[0]
emilmont 1:fdd22bb7aa52 159 * ....
emilmont 1:fdd22bb7aa52 160 * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
emilmont 1:fdd22bb7aa52 161 */
emilmont 1:fdd22bb7aa52 162
emilmont 1:fdd22bb7aa52 163 /* In this stage the MAC operations are increased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 164 The count variable holds the number of MAC operations performed.
emilmont 1:fdd22bb7aa52 165 Since the partial convolution starts from firstIndex
emilmont 1:fdd22bb7aa52 166 Number of Macs to be performed is firstIndex + 1 */
emilmont 1:fdd22bb7aa52 167 count = 1u + firstIndex;
emilmont 1:fdd22bb7aa52 168
emilmont 1:fdd22bb7aa52 169 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 170 px = pIn1;
emilmont 1:fdd22bb7aa52 171
emilmont 1:fdd22bb7aa52 172 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 173 pSrc2 = pIn2 + firstIndex;
emilmont 1:fdd22bb7aa52 174 py = pSrc2;
emilmont 1:fdd22bb7aa52 175
emilmont 1:fdd22bb7aa52 176 /* ------------------------
emilmont 1:fdd22bb7aa52 177 * Stage1 process
emilmont 1:fdd22bb7aa52 178 * ----------------------*/
emilmont 1:fdd22bb7aa52 179
emilmont 1:fdd22bb7aa52 180 /* The first loop starts here */
emilmont 1:fdd22bb7aa52 181 while(blockSize1 > 0)
emilmont 1:fdd22bb7aa52 182 {
emilmont 1:fdd22bb7aa52 183 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 184 sum = 0;
emilmont 1:fdd22bb7aa52 185
emilmont 1:fdd22bb7aa52 186 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 187 k = count >> 2u;
emilmont 1:fdd22bb7aa52 188
emilmont 1:fdd22bb7aa52 189 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 190 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 191 while(k > 0u)
emilmont 1:fdd22bb7aa52 192 {
emilmont 1:fdd22bb7aa52 193 /* x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 194 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 195 /* x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 196 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 197 /* x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 198 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 199 /* x[3] * y[srcBLen - 4] */
emilmont 1:fdd22bb7aa52 200 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 201
emilmont 1:fdd22bb7aa52 202 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 203 k--;
emilmont 1:fdd22bb7aa52 204 }
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* If the count is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 207 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 208 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 209
emilmont 1:fdd22bb7aa52 210 while(k > 0u)
emilmont 1:fdd22bb7aa52 211 {
emilmont 1:fdd22bb7aa52 212 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 213 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 214
emilmont 1:fdd22bb7aa52 215 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 216 k--;
emilmont 1:fdd22bb7aa52 217 }
emilmont 1:fdd22bb7aa52 218
emilmont 1:fdd22bb7aa52 219 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 220 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 221
emilmont 1:fdd22bb7aa52 222 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 223 py = ++pSrc2;
emilmont 1:fdd22bb7aa52 224 px = pIn1;
emilmont 1:fdd22bb7aa52 225
emilmont 1:fdd22bb7aa52 226 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 227 count++;
emilmont 1:fdd22bb7aa52 228
emilmont 1:fdd22bb7aa52 229 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 230 blockSize1--;
emilmont 1:fdd22bb7aa52 231 }
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* --------------------------
emilmont 1:fdd22bb7aa52 234 * Initializations of stage2
emilmont 1:fdd22bb7aa52 235 * ------------------------*/
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
emilmont 1:fdd22bb7aa52 238 * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
emilmont 1:fdd22bb7aa52 239 * ....
emilmont 1:fdd22bb7aa52 240 * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
emilmont 1:fdd22bb7aa52 241 */
emilmont 1:fdd22bb7aa52 242
emilmont 1:fdd22bb7aa52 243 /* Working pointer of inputA */
mbed_official 5:3762170b6d4d 244 if((int32_t)firstIndex - (int32_t)srcBLen + 1 > 0)
mbed_official 5:3762170b6d4d 245 {
mbed_official 5:3762170b6d4d 246 px = pIn1 + firstIndex - srcBLen + 1;
mbed_official 5:3762170b6d4d 247 }
mbed_official 5:3762170b6d4d 248 else
mbed_official 5:3762170b6d4d 249 {
mbed_official 5:3762170b6d4d 250 px = pIn1;
mbed_official 5:3762170b6d4d 251 }
emilmont 1:fdd22bb7aa52 252
emilmont 1:fdd22bb7aa52 253 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 254 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 255 py = pSrc2;
emilmont 1:fdd22bb7aa52 256
emilmont 1:fdd22bb7aa52 257 /* count is index by which the pointer pIn1 to be incremented */
emilmont 1:fdd22bb7aa52 258 count = 0u;
emilmont 1:fdd22bb7aa52 259
emilmont 1:fdd22bb7aa52 260 /* -------------------
emilmont 1:fdd22bb7aa52 261 * Stage2 process
emilmont 1:fdd22bb7aa52 262 * ------------------*/
emilmont 1:fdd22bb7aa52 263
emilmont 1:fdd22bb7aa52 264 /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
emilmont 1:fdd22bb7aa52 265 * So, to loop unroll over blockSize2,
emilmont 1:fdd22bb7aa52 266 * srcBLen should be greater than or equal to 4 */
emilmont 1:fdd22bb7aa52 267 if(srcBLen >= 4u)
emilmont 1:fdd22bb7aa52 268 {
emilmont 1:fdd22bb7aa52 269 /* Loop unroll over blkCnt */
emilmont 1:fdd22bb7aa52 270
emilmont 1:fdd22bb7aa52 271 blkCnt = blockSize2 / 3;
emilmont 1:fdd22bb7aa52 272 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 273 {
emilmont 1:fdd22bb7aa52 274 /* Set all accumulators to zero */
emilmont 1:fdd22bb7aa52 275 acc0 = 0;
emilmont 1:fdd22bb7aa52 276 acc1 = 0;
emilmont 1:fdd22bb7aa52 277 acc2 = 0;
emilmont 1:fdd22bb7aa52 278
emilmont 1:fdd22bb7aa52 279 /* read x[0], x[1] samples */
emilmont 1:fdd22bb7aa52 280 x0 = *(px++);
emilmont 1:fdd22bb7aa52 281 x1 = *(px++);
emilmont 1:fdd22bb7aa52 282
emilmont 1:fdd22bb7aa52 283 /* Apply loop unrolling and compute 3 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 284 k = srcBLen / 3;
emilmont 1:fdd22bb7aa52 285
emilmont 1:fdd22bb7aa52 286 /* First part of the processing with loop unrolling. Compute 3 MACs at a time.
emilmont 1:fdd22bb7aa52 287 ** a second loop below computes MACs for the remaining 1 to 2 samples. */
emilmont 1:fdd22bb7aa52 288 do
emilmont 1:fdd22bb7aa52 289 {
emilmont 1:fdd22bb7aa52 290 /* Read y[srcBLen - 1] sample */
emilmont 1:fdd22bb7aa52 291 c0 = *(py);
emilmont 1:fdd22bb7aa52 292
emilmont 1:fdd22bb7aa52 293 /* Read x[2] sample */
emilmont 1:fdd22bb7aa52 294 x2 = *(px);
emilmont 1:fdd22bb7aa52 295
emilmont 1:fdd22bb7aa52 296 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 297 /* acc0 += x[0] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 298 acc0 += (q63_t) x0 *c0;
emilmont 1:fdd22bb7aa52 299 /* acc1 += x[1] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 300 acc1 += (q63_t) x1 *c0;
emilmont 1:fdd22bb7aa52 301 /* acc2 += x[2] * y[srcBLen - 1] */
emilmont 1:fdd22bb7aa52 302 acc2 += (q63_t) x2 *c0;
emilmont 1:fdd22bb7aa52 303
emilmont 1:fdd22bb7aa52 304 /* Read y[srcBLen - 2] sample */
emilmont 1:fdd22bb7aa52 305 c0 = *(py - 1u);
emilmont 1:fdd22bb7aa52 306
emilmont 1:fdd22bb7aa52 307 /* Read x[3] sample */
emilmont 1:fdd22bb7aa52 308 x0 = *(px + 1u);
emilmont 1:fdd22bb7aa52 309
emilmont 1:fdd22bb7aa52 310 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 311 /* acc0 += x[1] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 312 acc0 += (q63_t) x1 *c0;
emilmont 1:fdd22bb7aa52 313 /* acc1 += x[2] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 314 acc1 += (q63_t) x2 *c0;
emilmont 1:fdd22bb7aa52 315 /* acc2 += x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 316 acc2 += (q63_t) x0 *c0;
emilmont 1:fdd22bb7aa52 317
emilmont 1:fdd22bb7aa52 318 /* Read y[srcBLen - 3] sample */
emilmont 1:fdd22bb7aa52 319 c0 = *(py - 2u);
emilmont 1:fdd22bb7aa52 320
emilmont 1:fdd22bb7aa52 321 /* Read x[4] sample */
emilmont 1:fdd22bb7aa52 322 x1 = *(px + 2u);
emilmont 1:fdd22bb7aa52 323
emilmont 1:fdd22bb7aa52 324 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 325 /* acc0 += x[2] * y[srcBLen - 3] */
emilmont 1:fdd22bb7aa52 326 acc0 += (q63_t) x2 *c0;
emilmont 1:fdd22bb7aa52 327 /* acc1 += x[3] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 328 acc1 += (q63_t) x0 *c0;
emilmont 1:fdd22bb7aa52 329 /* acc2 += x[4] * y[srcBLen - 2] */
emilmont 1:fdd22bb7aa52 330 acc2 += (q63_t) x1 *c0;
emilmont 1:fdd22bb7aa52 331
emilmont 1:fdd22bb7aa52 332
emilmont 1:fdd22bb7aa52 333 px += 3u;
emilmont 1:fdd22bb7aa52 334
emilmont 1:fdd22bb7aa52 335 py -= 3u;
emilmont 1:fdd22bb7aa52 336
emilmont 1:fdd22bb7aa52 337 } while(--k);
emilmont 1:fdd22bb7aa52 338
emilmont 1:fdd22bb7aa52 339 /* If the srcBLen is not a multiple of 3, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 340 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 341 k = srcBLen - (3 * (srcBLen / 3));
emilmont 1:fdd22bb7aa52 342
emilmont 1:fdd22bb7aa52 343 while(k > 0u)
emilmont 1:fdd22bb7aa52 344 {
emilmont 1:fdd22bb7aa52 345 /* Read y[srcBLen - 5] sample */
emilmont 1:fdd22bb7aa52 346 c0 = *(py--);
emilmont 1:fdd22bb7aa52 347
emilmont 1:fdd22bb7aa52 348 /* Read x[7] sample */
emilmont 1:fdd22bb7aa52 349 x2 = *(px++);
emilmont 1:fdd22bb7aa52 350
emilmont 1:fdd22bb7aa52 351 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 352 /* acc0 += x[4] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 353 acc0 += (q63_t) x0 *c0;
emilmont 1:fdd22bb7aa52 354 /* acc1 += x[5] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 355 acc1 += (q63_t) x1 *c0;
emilmont 1:fdd22bb7aa52 356 /* acc2 += x[6] * y[srcBLen - 5] */
emilmont 1:fdd22bb7aa52 357 acc2 += (q63_t) x2 *c0;
emilmont 1:fdd22bb7aa52 358
emilmont 1:fdd22bb7aa52 359 /* Reuse the present samples for the next MAC */
emilmont 1:fdd22bb7aa52 360 x0 = x1;
emilmont 1:fdd22bb7aa52 361 x1 = x2;
emilmont 1:fdd22bb7aa52 362
emilmont 1:fdd22bb7aa52 363 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 364 k--;
emilmont 1:fdd22bb7aa52 365 }
emilmont 1:fdd22bb7aa52 366
emilmont 1:fdd22bb7aa52 367 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 368 *pOut++ = (q31_t) (acc0 >> 31);
emilmont 1:fdd22bb7aa52 369 *pOut++ = (q31_t) (acc1 >> 31);
emilmont 1:fdd22bb7aa52 370 *pOut++ = (q31_t) (acc2 >> 31);
emilmont 1:fdd22bb7aa52 371
emilmont 1:fdd22bb7aa52 372 /* Increment the pointer pIn1 index, count by 3 */
emilmont 1:fdd22bb7aa52 373 count += 3u;
emilmont 1:fdd22bb7aa52 374
emilmont 1:fdd22bb7aa52 375 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 376 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 377 py = pSrc2;
emilmont 1:fdd22bb7aa52 378
emilmont 1:fdd22bb7aa52 379 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 380 blkCnt--;
emilmont 1:fdd22bb7aa52 381 }
emilmont 1:fdd22bb7aa52 382
emilmont 1:fdd22bb7aa52 383 /* If the blockSize2 is not a multiple of 3, compute any remaining output samples here.
emilmont 1:fdd22bb7aa52 384 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 385 blkCnt = blockSize2 - 3 * (blockSize2 / 3);
emilmont 1:fdd22bb7aa52 386
emilmont 1:fdd22bb7aa52 387 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 388 {
emilmont 1:fdd22bb7aa52 389 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 390 sum = 0;
emilmont 1:fdd22bb7aa52 391
emilmont 1:fdd22bb7aa52 392 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 393 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 394
emilmont 1:fdd22bb7aa52 395 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 396 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 397 while(k > 0u)
emilmont 1:fdd22bb7aa52 398 {
emilmont 1:fdd22bb7aa52 399 /* Perform the multiply-accumulates */
emilmont 1:fdd22bb7aa52 400 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 401 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 402 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 403 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 404
emilmont 1:fdd22bb7aa52 405 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 406 k--;
emilmont 1:fdd22bb7aa52 407 }
emilmont 1:fdd22bb7aa52 408
emilmont 1:fdd22bb7aa52 409 /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 410 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 411 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 412
emilmont 1:fdd22bb7aa52 413 while(k > 0u)
emilmont 1:fdd22bb7aa52 414 {
emilmont 1:fdd22bb7aa52 415 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 416 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 417
emilmont 1:fdd22bb7aa52 418 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 419 k--;
emilmont 1:fdd22bb7aa52 420 }
emilmont 1:fdd22bb7aa52 421
emilmont 1:fdd22bb7aa52 422 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 423 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 424
emilmont 1:fdd22bb7aa52 425 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 426 count++;
emilmont 1:fdd22bb7aa52 427
emilmont 1:fdd22bb7aa52 428 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 429 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 430 py = pSrc2;
emilmont 1:fdd22bb7aa52 431
emilmont 1:fdd22bb7aa52 432 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 433 blkCnt--;
emilmont 1:fdd22bb7aa52 434 }
emilmont 1:fdd22bb7aa52 435 }
emilmont 1:fdd22bb7aa52 436 else
emilmont 1:fdd22bb7aa52 437 {
emilmont 1:fdd22bb7aa52 438 /* If the srcBLen is not a multiple of 4,
emilmont 1:fdd22bb7aa52 439 * the blockSize2 loop cannot be unrolled by 4 */
emilmont 1:fdd22bb7aa52 440 blkCnt = (uint32_t) blockSize2;
emilmont 1:fdd22bb7aa52 441
emilmont 1:fdd22bb7aa52 442 while(blkCnt > 0u)
emilmont 1:fdd22bb7aa52 443 {
emilmont 1:fdd22bb7aa52 444 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 445 sum = 0;
emilmont 1:fdd22bb7aa52 446
emilmont 1:fdd22bb7aa52 447 /* srcBLen number of MACS should be performed */
emilmont 1:fdd22bb7aa52 448 k = srcBLen;
emilmont 1:fdd22bb7aa52 449
emilmont 1:fdd22bb7aa52 450 while(k > 0u)
emilmont 1:fdd22bb7aa52 451 {
emilmont 1:fdd22bb7aa52 452 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 453 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 454
emilmont 1:fdd22bb7aa52 455 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 456 k--;
emilmont 1:fdd22bb7aa52 457 }
emilmont 1:fdd22bb7aa52 458
emilmont 1:fdd22bb7aa52 459 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 460 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 461
emilmont 1:fdd22bb7aa52 462 /* Increment the MAC count */
emilmont 1:fdd22bb7aa52 463 count++;
emilmont 1:fdd22bb7aa52 464
emilmont 1:fdd22bb7aa52 465 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 466 px = pIn1 + count;
emilmont 1:fdd22bb7aa52 467 py = pSrc2;
emilmont 1:fdd22bb7aa52 468
emilmont 1:fdd22bb7aa52 469 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 470 blkCnt--;
emilmont 1:fdd22bb7aa52 471 }
emilmont 1:fdd22bb7aa52 472 }
emilmont 1:fdd22bb7aa52 473
emilmont 1:fdd22bb7aa52 474
emilmont 1:fdd22bb7aa52 475 /* --------------------------
emilmont 1:fdd22bb7aa52 476 * Initializations of stage3
emilmont 1:fdd22bb7aa52 477 * -------------------------*/
emilmont 1:fdd22bb7aa52 478
emilmont 1:fdd22bb7aa52 479 /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
emilmont 1:fdd22bb7aa52 480 * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
emilmont 1:fdd22bb7aa52 481 * ....
emilmont 1:fdd22bb7aa52 482 * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
emilmont 1:fdd22bb7aa52 483 * sum += x[srcALen-1] * y[srcBLen-1]
emilmont 1:fdd22bb7aa52 484 */
emilmont 1:fdd22bb7aa52 485
emilmont 1:fdd22bb7aa52 486 /* In this stage the MAC operations are decreased by 1 for every iteration.
emilmont 1:fdd22bb7aa52 487 The blockSize3 variable holds the number of MAC operations performed */
emilmont 1:fdd22bb7aa52 488 count = srcBLen - 1u;
emilmont 1:fdd22bb7aa52 489
emilmont 1:fdd22bb7aa52 490 /* Working pointer of inputA */
emilmont 1:fdd22bb7aa52 491 pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 492 px = pSrc1;
emilmont 1:fdd22bb7aa52 493
emilmont 1:fdd22bb7aa52 494 /* Working pointer of inputB */
emilmont 1:fdd22bb7aa52 495 pSrc2 = pIn2 + (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 496 py = pSrc2;
emilmont 1:fdd22bb7aa52 497
emilmont 1:fdd22bb7aa52 498 /* -------------------
emilmont 1:fdd22bb7aa52 499 * Stage3 process
emilmont 1:fdd22bb7aa52 500 * ------------------*/
emilmont 1:fdd22bb7aa52 501
emilmont 1:fdd22bb7aa52 502 while(blockSize3 > 0)
emilmont 1:fdd22bb7aa52 503 {
emilmont 1:fdd22bb7aa52 504 /* Accumulator is made zero for every iteration */
emilmont 1:fdd22bb7aa52 505 sum = 0;
emilmont 1:fdd22bb7aa52 506
emilmont 1:fdd22bb7aa52 507 /* Apply loop unrolling and compute 4 MACs simultaneously. */
emilmont 1:fdd22bb7aa52 508 k = count >> 2u;
emilmont 1:fdd22bb7aa52 509
emilmont 1:fdd22bb7aa52 510 /* First part of the processing with loop unrolling. Compute 4 MACs at a time.
emilmont 1:fdd22bb7aa52 511 ** a second loop below computes MACs for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 512 while(k > 0u)
emilmont 1:fdd22bb7aa52 513 {
emilmont 1:fdd22bb7aa52 514 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 515 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 516 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 517 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 518
emilmont 1:fdd22bb7aa52 519 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 520 k--;
emilmont 1:fdd22bb7aa52 521 }
emilmont 1:fdd22bb7aa52 522
emilmont 1:fdd22bb7aa52 523 /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
emilmont 1:fdd22bb7aa52 524 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 525 k = count % 0x4u;
emilmont 1:fdd22bb7aa52 526
emilmont 1:fdd22bb7aa52 527 while(k > 0u)
emilmont 1:fdd22bb7aa52 528 {
emilmont 1:fdd22bb7aa52 529 /* Perform the multiply-accumulate */
emilmont 1:fdd22bb7aa52 530 sum += (q63_t) * px++ * (*py--);
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 533 k--;
emilmont 1:fdd22bb7aa52 534 }
emilmont 1:fdd22bb7aa52 535
emilmont 1:fdd22bb7aa52 536 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 537 *pOut++ = (q31_t) (sum >> 31);
emilmont 1:fdd22bb7aa52 538
emilmont 1:fdd22bb7aa52 539 /* Update the inputA and inputB pointers for next MAC calculation */
emilmont 1:fdd22bb7aa52 540 px = ++pSrc1;
emilmont 1:fdd22bb7aa52 541 py = pSrc2;
emilmont 1:fdd22bb7aa52 542
emilmont 1:fdd22bb7aa52 543 /* Decrement the MAC count */
emilmont 1:fdd22bb7aa52 544 count--;
emilmont 1:fdd22bb7aa52 545
emilmont 1:fdd22bb7aa52 546 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 547 blockSize3--;
emilmont 1:fdd22bb7aa52 548
emilmont 1:fdd22bb7aa52 549 }
emilmont 1:fdd22bb7aa52 550
emilmont 1:fdd22bb7aa52 551 /* set status as ARM_MATH_SUCCESS */
emilmont 1:fdd22bb7aa52 552 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 553 }
emilmont 1:fdd22bb7aa52 554
emilmont 1:fdd22bb7aa52 555 /* Return to application */
emilmont 1:fdd22bb7aa52 556 return (status);
emilmont 1:fdd22bb7aa52 557
emilmont 1:fdd22bb7aa52 558 #else
emilmont 1:fdd22bb7aa52 559
emilmont 1:fdd22bb7aa52 560 /* Run the below code for Cortex-M0 */
emilmont 1:fdd22bb7aa52 561
emilmont 1:fdd22bb7aa52 562 q31_t *pIn1 = pSrcA; /* inputA pointer */
emilmont 1:fdd22bb7aa52 563 q31_t *pIn2 = pSrcB; /* inputB pointer */
emilmont 1:fdd22bb7aa52 564 q63_t sum; /* Accumulator */
emilmont 1:fdd22bb7aa52 565 uint32_t i, j; /* loop counters */
emilmont 1:fdd22bb7aa52 566 arm_status status; /* status of Partial convolution */
emilmont 1:fdd22bb7aa52 567
emilmont 1:fdd22bb7aa52 568 /* Check for range of output samples to be calculated */
emilmont 1:fdd22bb7aa52 569 if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
emilmont 1:fdd22bb7aa52 570 {
emilmont 1:fdd22bb7aa52 571 /* Set status as ARM_ARGUMENT_ERROR */
emilmont 1:fdd22bb7aa52 572 status = ARM_MATH_ARGUMENT_ERROR;
emilmont 1:fdd22bb7aa52 573 }
emilmont 1:fdd22bb7aa52 574 else
emilmont 1:fdd22bb7aa52 575 {
emilmont 1:fdd22bb7aa52 576 /* Loop to calculate convolution for output length number of values */
emilmont 1:fdd22bb7aa52 577 for (i = firstIndex; i <= (firstIndex + numPoints - 1); i++)
emilmont 1:fdd22bb7aa52 578 {
emilmont 1:fdd22bb7aa52 579 /* Initialize sum with zero to carry on MAC operations */
emilmont 1:fdd22bb7aa52 580 sum = 0;
emilmont 1:fdd22bb7aa52 581
emilmont 1:fdd22bb7aa52 582 /* Loop to perform MAC operations according to convolution equation */
emilmont 1:fdd22bb7aa52 583 for (j = 0; j <= i; j++)
emilmont 1:fdd22bb7aa52 584 {
emilmont 1:fdd22bb7aa52 585 /* Check the array limitations */
emilmont 1:fdd22bb7aa52 586 if(((i - j) < srcBLen) && (j < srcALen))
emilmont 1:fdd22bb7aa52 587 {
emilmont 1:fdd22bb7aa52 588 /* z[i] += x[i-j] * y[j] */
emilmont 1:fdd22bb7aa52 589 sum += ((q63_t) pIn1[j] * (pIn2[i - j]));
emilmont 1:fdd22bb7aa52 590 }
emilmont 1:fdd22bb7aa52 591 }
emilmont 1:fdd22bb7aa52 592
emilmont 1:fdd22bb7aa52 593 /* Store the output in the destination buffer */
emilmont 1:fdd22bb7aa52 594 pDst[i] = (q31_t) (sum >> 31u);
emilmont 1:fdd22bb7aa52 595 }
emilmont 1:fdd22bb7aa52 596 /* set status as ARM_SUCCESS as there are no argument errors */
emilmont 1:fdd22bb7aa52 597 status = ARM_MATH_SUCCESS;
emilmont 1:fdd22bb7aa52 598 }
emilmont 1:fdd22bb7aa52 599 return (status);
emilmont 1:fdd22bb7aa52 600
mbed_official 3:7a284390b0ce 601 #endif /* #ifndef ARM_MATH_CM0_FAMILY */
emilmont 1:fdd22bb7aa52 602
emilmont 1:fdd22bb7aa52 603 }
emilmont 1:fdd22bb7aa52 604
emilmont 1:fdd22bb7aa52 605 /**
emilmont 1:fdd22bb7aa52 606 * @} end of PartialConv group
emilmont 1:fdd22bb7aa52 607 */